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 <scsi/fc/fc_fcoe.h>
53 #include <net/udp_tunnel.h>
54 #include <net/pkt_cls.h>
55 #include <net/tc_act/tc_gact.h>
56 #include <net/tc_act/tc_mirred.h>
57 #include <net/vxlan.h>
60 #include "ixgbe_common.h"
61 #include "ixgbe_dcb_82599.h"
62 #include "ixgbe_sriov.h"
63 #include "ixgbe_model.h"
65 char ixgbe_driver_name[] = "ixgbe";
66 static const char ixgbe_driver_string[] =
67 "Intel(R) 10 Gigabit PCI Express Network Driver";
69 char ixgbe_default_device_descr[] =
70 "Intel(R) 10 Gigabit Network Connection";
72 static char ixgbe_default_device_descr[] =
73 "Intel(R) 10 Gigabit Network Connection";
75 #define DRV_VERSION "4.4.0-k"
76 const char ixgbe_driver_version[] = DRV_VERSION;
77 static const char ixgbe_copyright[] =
78 "Copyright (c) 1999-2016 Intel Corporation.";
80 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";
82 static const struct ixgbe_info *ixgbe_info_tbl[] = {
83 [board_82598] = &ixgbe_82598_info,
84 [board_82599] = &ixgbe_82599_info,
85 [board_X540] = &ixgbe_X540_info,
86 [board_X550] = &ixgbe_X550_info,
87 [board_X550EM_x] = &ixgbe_X550EM_x_info,
88 [board_x550em_a] = &ixgbe_x550em_a_info,
89 [board_x550em_a_fw] = &ixgbe_x550em_a_fw_info,
92 /* ixgbe_pci_tbl - PCI Device ID Table
94 * Wildcard entries (PCI_ANY_ID) should come last
95 * Last entry must be all 0s
97 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
98 * Class, Class Mask, private data (not used) }
100 static const struct pci_device_id ixgbe_pci_tbl[] = {
101 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
102 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
103 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
104 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
105 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
106 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
107 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
108 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
109 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
110 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
111 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
112 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
113 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
114 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
115 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
116 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
117 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
118 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
119 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
120 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
121 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
122 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
123 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
124 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
125 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
126 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
127 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
128 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
129 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
130 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
131 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
132 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
133 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
134 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
135 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
136 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
137 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
138 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
139 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
140 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
141 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
142 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
143 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
144 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
145 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
146 /* required last entry */
149 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
151 #ifdef CONFIG_IXGBE_DCA
152 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
154 static struct notifier_block dca_notifier = {
155 .notifier_call = ixgbe_notify_dca,
161 #ifdef CONFIG_PCI_IOV
162 static unsigned int max_vfs;
163 module_param(max_vfs, uint, 0);
164 MODULE_PARM_DESC(max_vfs,
165 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
166 #endif /* CONFIG_PCI_IOV */
168 static unsigned int allow_unsupported_sfp;
169 module_param(allow_unsupported_sfp, uint, 0);
170 MODULE_PARM_DESC(allow_unsupported_sfp,
171 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
173 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
174 static int debug = -1;
175 module_param(debug, int, 0);
176 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
178 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
179 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
180 MODULE_LICENSE("GPL");
181 MODULE_VERSION(DRV_VERSION);
183 static struct workqueue_struct *ixgbe_wq;
185 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
186 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
188 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
191 struct pci_dev *parent_dev;
192 struct pci_bus *parent_bus;
194 parent_bus = adapter->pdev->bus->parent;
198 parent_dev = parent_bus->self;
202 if (!pci_is_pcie(parent_dev))
205 pcie_capability_read_word(parent_dev, reg, value);
206 if (*value == IXGBE_FAILED_READ_CFG_WORD &&
207 ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
212 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
214 struct ixgbe_hw *hw = &adapter->hw;
218 hw->bus.type = ixgbe_bus_type_pci_express;
220 /* Get the negotiated link width and speed from PCI config space of the
221 * parent, as this device is behind a switch
223 err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
225 /* assume caller will handle error case */
229 hw->bus.width = ixgbe_convert_bus_width(link_status);
230 hw->bus.speed = ixgbe_convert_bus_speed(link_status);
236 * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
237 * @hw: hw specific details
239 * This function is used by probe to determine whether a device's PCI-Express
240 * bandwidth details should be gathered from the parent bus instead of from the
241 * device. Used to ensure that various locations all have the correct device ID
244 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
246 switch (hw->device_id) {
247 case IXGBE_DEV_ID_82599_SFP_SF_QP:
248 case IXGBE_DEV_ID_82599_QSFP_SF_QP:
255 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
258 struct ixgbe_hw *hw = &adapter->hw;
260 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
261 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
262 struct pci_dev *pdev;
264 /* Some devices are not connected over PCIe and thus do not negotiate
265 * speed. These devices do not have valid bus info, and thus any report
266 * we generate may not be correct.
268 if (hw->bus.type == ixgbe_bus_type_internal)
271 /* determine whether to use the parent device */
272 if (ixgbe_pcie_from_parent(&adapter->hw))
273 pdev = adapter->pdev->bus->parent->self;
275 pdev = adapter->pdev;
277 if (pcie_get_minimum_link(pdev, &speed, &width) ||
278 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
279 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
284 case PCIE_SPEED_2_5GT:
285 /* 8b/10b encoding reduces max throughput by 20% */
288 case PCIE_SPEED_5_0GT:
289 /* 8b/10b encoding reduces max throughput by 20% */
292 case PCIE_SPEED_8_0GT:
293 /* 128b/130b encoding reduces throughput by less than 2% */
297 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
301 e_dev_info("PCI Express bandwidth of %dGT/s available\n",
303 e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
304 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
305 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
306 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
309 (speed == PCIE_SPEED_2_5GT ? "20%" :
310 speed == PCIE_SPEED_5_0GT ? "20%" :
311 speed == PCIE_SPEED_8_0GT ? "<2%" :
314 if (max_gts < expected_gts) {
315 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
316 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
318 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
322 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
324 if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
325 !test_bit(__IXGBE_REMOVING, &adapter->state) &&
326 !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
327 queue_work(ixgbe_wq, &adapter->service_task);
330 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
332 struct ixgbe_adapter *adapter = hw->back;
337 e_dev_err("Adapter removed\n");
338 if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
339 ixgbe_service_event_schedule(adapter);
342 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
346 /* The following check not only optimizes a bit by not
347 * performing a read on the status register when the
348 * register just read was a status register read that
349 * returned IXGBE_FAILED_READ_REG. It also blocks any
350 * potential recursion.
352 if (reg == IXGBE_STATUS) {
353 ixgbe_remove_adapter(hw);
356 value = ixgbe_read_reg(hw, IXGBE_STATUS);
357 if (value == IXGBE_FAILED_READ_REG)
358 ixgbe_remove_adapter(hw);
362 * ixgbe_read_reg - Read from device register
363 * @hw: hw specific details
364 * @reg: offset of register to read
366 * Returns : value read or IXGBE_FAILED_READ_REG if removed
368 * This function is used to read device registers. It checks for device
369 * removal by confirming any read that returns all ones by checking the
370 * status register value for all ones. This function avoids reading from
371 * the hardware if a removal was previously detected in which case it
372 * returns IXGBE_FAILED_READ_REG (all ones).
374 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
376 u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
379 if (ixgbe_removed(reg_addr))
380 return IXGBE_FAILED_READ_REG;
381 if (unlikely(hw->phy.nw_mng_if_sel &
382 IXGBE_NW_MNG_IF_SEL_ENABLE_10_100M)) {
383 struct ixgbe_adapter *adapter;
386 for (i = 0; i < 200; ++i) {
387 value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
389 goto writes_completed;
390 if (value == IXGBE_FAILED_READ_REG) {
391 ixgbe_remove_adapter(hw);
392 return IXGBE_FAILED_READ_REG;
398 e_warn(hw, "register writes incomplete %08x\n", value);
402 value = readl(reg_addr + reg);
403 if (unlikely(value == IXGBE_FAILED_READ_REG))
404 ixgbe_check_remove(hw, reg);
408 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
412 pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
413 if (value == IXGBE_FAILED_READ_CFG_WORD) {
414 ixgbe_remove_adapter(hw);
420 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
422 struct ixgbe_adapter *adapter = hw->back;
425 if (ixgbe_removed(hw->hw_addr))
426 return IXGBE_FAILED_READ_CFG_WORD;
427 pci_read_config_word(adapter->pdev, reg, &value);
428 if (value == IXGBE_FAILED_READ_CFG_WORD &&
429 ixgbe_check_cfg_remove(hw, adapter->pdev))
430 return IXGBE_FAILED_READ_CFG_WORD;
434 #ifdef CONFIG_PCI_IOV
435 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
437 struct ixgbe_adapter *adapter = hw->back;
440 if (ixgbe_removed(hw->hw_addr))
441 return IXGBE_FAILED_READ_CFG_DWORD;
442 pci_read_config_dword(adapter->pdev, reg, &value);
443 if (value == IXGBE_FAILED_READ_CFG_DWORD &&
444 ixgbe_check_cfg_remove(hw, adapter->pdev))
445 return IXGBE_FAILED_READ_CFG_DWORD;
448 #endif /* CONFIG_PCI_IOV */
450 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
452 struct ixgbe_adapter *adapter = hw->back;
454 if (ixgbe_removed(hw->hw_addr))
456 pci_write_config_word(adapter->pdev, reg, value);
459 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
461 BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
463 /* flush memory to make sure state is correct before next watchdog */
464 smp_mb__before_atomic();
465 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
468 struct ixgbe_reg_info {
473 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
475 /* General Registers */
476 {IXGBE_CTRL, "CTRL"},
477 {IXGBE_STATUS, "STATUS"},
478 {IXGBE_CTRL_EXT, "CTRL_EXT"},
480 /* Interrupt Registers */
481 {IXGBE_EICR, "EICR"},
484 {IXGBE_SRRCTL(0), "SRRCTL"},
485 {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
486 {IXGBE_RDLEN(0), "RDLEN"},
487 {IXGBE_RDH(0), "RDH"},
488 {IXGBE_RDT(0), "RDT"},
489 {IXGBE_RXDCTL(0), "RXDCTL"},
490 {IXGBE_RDBAL(0), "RDBAL"},
491 {IXGBE_RDBAH(0), "RDBAH"},
494 {IXGBE_TDBAL(0), "TDBAL"},
495 {IXGBE_TDBAH(0), "TDBAH"},
496 {IXGBE_TDLEN(0), "TDLEN"},
497 {IXGBE_TDH(0), "TDH"},
498 {IXGBE_TDT(0), "TDT"},
499 {IXGBE_TXDCTL(0), "TXDCTL"},
501 /* List Terminator */
507 * ixgbe_regdump - register printout routine
509 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
515 switch (reginfo->ofs) {
516 case IXGBE_SRRCTL(0):
517 for (i = 0; i < 64; i++)
518 regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
520 case IXGBE_DCA_RXCTRL(0):
521 for (i = 0; i < 64; i++)
522 regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
525 for (i = 0; i < 64; i++)
526 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
529 for (i = 0; i < 64; i++)
530 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
533 for (i = 0; i < 64; i++)
534 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
536 case IXGBE_RXDCTL(0):
537 for (i = 0; i < 64; i++)
538 regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
541 for (i = 0; i < 64; i++)
542 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
545 for (i = 0; i < 64; i++)
546 regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
549 for (i = 0; i < 64; i++)
550 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
553 for (i = 0; i < 64; i++)
554 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
557 for (i = 0; i < 64; i++)
558 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
561 for (i = 0; i < 64; i++)
562 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
565 for (i = 0; i < 64; i++)
566 regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
568 case IXGBE_TXDCTL(0):
569 for (i = 0; i < 64; i++)
570 regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
573 pr_info("%-15s %08x\n", reginfo->name,
574 IXGBE_READ_REG(hw, reginfo->ofs));
578 for (i = 0; i < 8; i++) {
579 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
580 pr_err("%-15s", rname);
581 for (j = 0; j < 8; j++)
582 pr_cont(" %08x", regs[i*8+j]);
589 * ixgbe_dump - Print registers, tx-rings and rx-rings
591 static void ixgbe_dump(struct ixgbe_adapter *adapter)
593 struct net_device *netdev = adapter->netdev;
594 struct ixgbe_hw *hw = &adapter->hw;
595 struct ixgbe_reg_info *reginfo;
597 struct ixgbe_ring *tx_ring;
598 struct ixgbe_tx_buffer *tx_buffer;
599 union ixgbe_adv_tx_desc *tx_desc;
600 struct my_u0 { u64 a; u64 b; } *u0;
601 struct ixgbe_ring *rx_ring;
602 union ixgbe_adv_rx_desc *rx_desc;
603 struct ixgbe_rx_buffer *rx_buffer_info;
607 if (!netif_msg_hw(adapter))
610 /* Print netdevice Info */
612 dev_info(&adapter->pdev->dev, "Net device Info\n");
613 pr_info("Device Name state "
615 pr_info("%-15s %016lX %016lX\n",
618 dev_trans_start(netdev));
621 /* Print Registers */
622 dev_info(&adapter->pdev->dev, "Register Dump\n");
623 pr_info(" Register Name Value\n");
624 for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
625 reginfo->name; reginfo++) {
626 ixgbe_regdump(hw, reginfo);
629 /* Print TX Ring Summary */
630 if (!netdev || !netif_running(netdev))
633 dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
634 pr_info(" %s %s %s %s\n",
635 "Queue [NTU] [NTC] [bi(ntc)->dma ]",
636 "leng", "ntw", "timestamp");
637 for (n = 0; n < adapter->num_tx_queues; n++) {
638 tx_ring = adapter->tx_ring[n];
639 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
640 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
641 n, tx_ring->next_to_use, tx_ring->next_to_clean,
642 (u64)dma_unmap_addr(tx_buffer, dma),
643 dma_unmap_len(tx_buffer, len),
644 tx_buffer->next_to_watch,
645 (u64)tx_buffer->time_stamp);
649 if (!netif_msg_tx_done(adapter))
650 goto rx_ring_summary;
652 dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
654 /* Transmit Descriptor Formats
656 * 82598 Advanced Transmit Descriptor
657 * +--------------------------------------------------------------+
658 * 0 | Buffer Address [63:0] |
659 * +--------------------------------------------------------------+
660 * 8 | PAYLEN | POPTS | IDX | STA | DCMD |DTYP | RSV | DTALEN |
661 * +--------------------------------------------------------------+
662 * 63 46 45 40 39 36 35 32 31 24 23 20 19 0
664 * 82598 Advanced Transmit Descriptor (Write-Back Format)
665 * +--------------------------------------------------------------+
667 * +--------------------------------------------------------------+
668 * 8 | RSV | STA | NXTSEQ |
669 * +--------------------------------------------------------------+
672 * 82599+ Advanced Transmit Descriptor
673 * +--------------------------------------------------------------+
674 * 0 | Buffer Address [63:0] |
675 * +--------------------------------------------------------------+
676 * 8 |PAYLEN |POPTS|CC|IDX |STA |DCMD |DTYP |MAC |RSV |DTALEN |
677 * +--------------------------------------------------------------+
678 * 63 46 45 40 39 38 36 35 32 31 24 23 20 19 18 17 16 15 0
680 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
681 * +--------------------------------------------------------------+
683 * +--------------------------------------------------------------+
684 * 8 | RSV | STA | RSV |
685 * +--------------------------------------------------------------+
689 for (n = 0; n < adapter->num_tx_queues; n++) {
690 tx_ring = adapter->tx_ring[n];
691 pr_info("------------------------------------\n");
692 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
693 pr_info("------------------------------------\n");
694 pr_info("%s%s %s %s %s %s\n",
695 "T [desc] [address 63:0 ] ",
696 "[PlPOIdStDDt Ln] [bi->dma ] ",
697 "leng", "ntw", "timestamp", "bi->skb");
699 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
700 tx_desc = IXGBE_TX_DESC(tx_ring, i);
701 tx_buffer = &tx_ring->tx_buffer_info[i];
702 u0 = (struct my_u0 *)tx_desc;
703 if (dma_unmap_len(tx_buffer, len) > 0) {
704 pr_info("T [0x%03X] %016llX %016llX %016llX %08X %p %016llX %p",
708 (u64)dma_unmap_addr(tx_buffer, dma),
709 dma_unmap_len(tx_buffer, len),
710 tx_buffer->next_to_watch,
711 (u64)tx_buffer->time_stamp,
713 if (i == tx_ring->next_to_use &&
714 i == tx_ring->next_to_clean)
716 else if (i == tx_ring->next_to_use)
718 else if (i == tx_ring->next_to_clean)
723 if (netif_msg_pktdata(adapter) &&
725 print_hex_dump(KERN_INFO, "",
726 DUMP_PREFIX_ADDRESS, 16, 1,
727 tx_buffer->skb->data,
728 dma_unmap_len(tx_buffer, len),
734 /* Print RX Rings Summary */
736 dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
737 pr_info("Queue [NTU] [NTC]\n");
738 for (n = 0; n < adapter->num_rx_queues; n++) {
739 rx_ring = adapter->rx_ring[n];
740 pr_info("%5d %5X %5X\n",
741 n, rx_ring->next_to_use, rx_ring->next_to_clean);
745 if (!netif_msg_rx_status(adapter))
748 dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
750 /* Receive Descriptor Formats
752 * 82598 Advanced Receive Descriptor (Read) Format
754 * +-----------------------------------------------------+
755 * 0 | Packet Buffer Address [63:1] |A0/NSE|
756 * +----------------------------------------------+------+
757 * 8 | Header Buffer Address [63:1] | DD |
758 * +-----------------------------------------------------+
761 * 82598 Advanced Receive Descriptor (Write-Back) Format
763 * 63 48 47 32 31 30 21 20 16 15 4 3 0
764 * +------------------------------------------------------+
765 * 0 | RSS Hash / |SPH| HDR_LEN | RSV |Packet| RSS |
766 * | Packet | IP | | | | Type | Type |
767 * | Checksum | Ident | | | | | |
768 * +------------------------------------------------------+
769 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
770 * +------------------------------------------------------+
771 * 63 48 47 32 31 20 19 0
773 * 82599+ Advanced Receive Descriptor (Read) Format
775 * +-----------------------------------------------------+
776 * 0 | Packet Buffer Address [63:1] |A0/NSE|
777 * +----------------------------------------------+------+
778 * 8 | Header Buffer Address [63:1] | DD |
779 * +-----------------------------------------------------+
782 * 82599+ Advanced Receive Descriptor (Write-Back) Format
784 * 63 48 47 32 31 30 21 20 17 16 4 3 0
785 * +------------------------------------------------------+
786 * 0 |RSS / Frag Checksum|SPH| HDR_LEN |RSC- |Packet| RSS |
787 * |/ RTT / PCoE_PARAM | | | CNT | Type | Type |
788 * |/ Flow Dir Flt ID | | | | | |
789 * +------------------------------------------------------+
790 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
791 * +------------------------------------------------------+
792 * 63 48 47 32 31 20 19 0
795 for (n = 0; n < adapter->num_rx_queues; n++) {
796 rx_ring = adapter->rx_ring[n];
797 pr_info("------------------------------------\n");
798 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
799 pr_info("------------------------------------\n");
801 "R [desc] [ PktBuf A0] ",
802 "[ HeadBuf DD] [bi->dma ] [bi->skb ] ",
803 "<-- Adv Rx Read format\n");
805 "RWB[desc] [PcsmIpSHl PtRs] ",
806 "[vl er S cks ln] ---------------- [bi->skb ] ",
807 "<-- Adv Rx Write-Back format\n");
809 for (i = 0; i < rx_ring->count; i++) {
810 rx_buffer_info = &rx_ring->rx_buffer_info[i];
811 rx_desc = IXGBE_RX_DESC(rx_ring, i);
812 u0 = (struct my_u0 *)rx_desc;
813 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
814 if (staterr & IXGBE_RXD_STAT_DD) {
815 /* Descriptor Done */
816 pr_info("RWB[0x%03X] %016llX "
817 "%016llX ---------------- %p", i,
820 rx_buffer_info->skb);
822 pr_info("R [0x%03X] %016llX "
823 "%016llX %016llX %p", i,
826 (u64)rx_buffer_info->dma,
827 rx_buffer_info->skb);
829 if (netif_msg_pktdata(adapter) &&
830 rx_buffer_info->dma) {
831 print_hex_dump(KERN_INFO, "",
832 DUMP_PREFIX_ADDRESS, 16, 1,
833 page_address(rx_buffer_info->page) +
834 rx_buffer_info->page_offset,
835 ixgbe_rx_bufsz(rx_ring), true);
839 if (i == rx_ring->next_to_use)
841 else if (i == rx_ring->next_to_clean)
850 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
854 /* Let firmware take over control of h/w */
855 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
856 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
857 ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
860 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
864 /* Let firmware know the driver has taken over */
865 ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
866 IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
867 ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
871 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
872 * @adapter: pointer to adapter struct
873 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
874 * @queue: queue to map the corresponding interrupt to
875 * @msix_vector: the vector to map to the corresponding queue
878 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
879 u8 queue, u8 msix_vector)
882 struct ixgbe_hw *hw = &adapter->hw;
883 switch (hw->mac.type) {
884 case ixgbe_mac_82598EB:
885 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
888 index = (((direction * 64) + queue) >> 2) & 0x1F;
889 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
890 ivar &= ~(0xFF << (8 * (queue & 0x3)));
891 ivar |= (msix_vector << (8 * (queue & 0x3)));
892 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
894 case ixgbe_mac_82599EB:
897 case ixgbe_mac_X550EM_x:
898 case ixgbe_mac_x550em_a:
899 if (direction == -1) {
901 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
902 index = ((queue & 1) * 8);
903 ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
904 ivar &= ~(0xFF << index);
905 ivar |= (msix_vector << index);
906 IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
909 /* tx or rx causes */
910 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
911 index = ((16 * (queue & 1)) + (8 * direction));
912 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
913 ivar &= ~(0xFF << index);
914 ivar |= (msix_vector << index);
915 IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
923 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
928 switch (adapter->hw.mac.type) {
929 case ixgbe_mac_82598EB:
930 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
931 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
933 case ixgbe_mac_82599EB:
936 case ixgbe_mac_X550EM_x:
937 case ixgbe_mac_x550em_a:
938 mask = (qmask & 0xFFFFFFFF);
939 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
940 mask = (qmask >> 32);
941 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
948 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
949 struct ixgbe_tx_buffer *tx_buffer)
951 if (tx_buffer->skb) {
952 dev_kfree_skb_any(tx_buffer->skb);
953 if (dma_unmap_len(tx_buffer, len))
954 dma_unmap_single(ring->dev,
955 dma_unmap_addr(tx_buffer, dma),
956 dma_unmap_len(tx_buffer, len),
958 } else if (dma_unmap_len(tx_buffer, len)) {
959 dma_unmap_page(ring->dev,
960 dma_unmap_addr(tx_buffer, dma),
961 dma_unmap_len(tx_buffer, len),
964 tx_buffer->next_to_watch = NULL;
965 tx_buffer->skb = NULL;
966 dma_unmap_len_set(tx_buffer, len, 0);
967 /* tx_buffer must be completely set up in the transmit path */
970 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
972 struct ixgbe_hw *hw = &adapter->hw;
973 struct ixgbe_hw_stats *hwstats = &adapter->stats;
977 if ((hw->fc.current_mode != ixgbe_fc_full) &&
978 (hw->fc.current_mode != ixgbe_fc_rx_pause))
981 switch (hw->mac.type) {
982 case ixgbe_mac_82598EB:
983 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
986 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
988 hwstats->lxoffrxc += data;
990 /* refill credits (no tx hang) if we received xoff */
994 for (i = 0; i < adapter->num_tx_queues; i++)
995 clear_bit(__IXGBE_HANG_CHECK_ARMED,
996 &adapter->tx_ring[i]->state);
999 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
1001 struct ixgbe_hw *hw = &adapter->hw;
1002 struct ixgbe_hw_stats *hwstats = &adapter->stats;
1006 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
1008 if (adapter->ixgbe_ieee_pfc)
1009 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
1011 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
1012 ixgbe_update_xoff_rx_lfc(adapter);
1016 /* update stats for each tc, only valid with PFC enabled */
1017 for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
1020 switch (hw->mac.type) {
1021 case ixgbe_mac_82598EB:
1022 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
1025 pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
1027 hwstats->pxoffrxc[i] += pxoffrxc;
1028 /* Get the TC for given UP */
1029 tc = netdev_get_prio_tc_map(adapter->netdev, i);
1030 xoff[tc] += pxoffrxc;
1033 /* disarm tx queues that have received xoff frames */
1034 for (i = 0; i < adapter->num_tx_queues; i++) {
1035 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
1037 tc = tx_ring->dcb_tc;
1039 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1043 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1045 return ring->stats.packets;
1048 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1050 struct ixgbe_adapter *adapter;
1051 struct ixgbe_hw *hw;
1054 if (ring->l2_accel_priv)
1055 adapter = ring->l2_accel_priv->real_adapter;
1057 adapter = netdev_priv(ring->netdev);
1060 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1061 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1064 return (head < tail) ?
1065 tail - head : (tail + ring->count - head);
1070 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1072 u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1073 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1074 u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1076 clear_check_for_tx_hang(tx_ring);
1079 * Check for a hung queue, but be thorough. This verifies
1080 * that a transmit has been completed since the previous
1081 * check AND there is at least one packet pending. The
1082 * ARMED bit is set to indicate a potential hang. The
1083 * bit is cleared if a pause frame is received to remove
1084 * false hang detection due to PFC or 802.3x frames. By
1085 * requiring this to fail twice we avoid races with
1086 * pfc clearing the ARMED bit and conditions where we
1087 * run the check_tx_hang logic with a transmit completion
1088 * pending but without time to complete it yet.
1090 if (tx_done_old == tx_done && tx_pending)
1091 /* make sure it is true for two checks in a row */
1092 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1094 /* update completed stats and continue */
1095 tx_ring->tx_stats.tx_done_old = tx_done;
1096 /* reset the countdown */
1097 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1103 * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1104 * @adapter: driver private struct
1106 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1109 /* Do the reset outside of interrupt context */
1110 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1111 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1112 e_warn(drv, "initiating reset due to tx timeout\n");
1113 ixgbe_service_event_schedule(adapter);
1118 * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1120 static int ixgbe_tx_maxrate(struct net_device *netdev,
1121 int queue_index, u32 maxrate)
1123 struct ixgbe_adapter *adapter = netdev_priv(netdev);
1124 struct ixgbe_hw *hw = &adapter->hw;
1125 u32 bcnrc_val = ixgbe_link_mbps(adapter);
1130 /* Calculate the rate factor values to set */
1131 bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1132 bcnrc_val /= maxrate;
1134 /* clear everything but the rate factor */
1135 bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1136 IXGBE_RTTBCNRC_RF_DEC_MASK;
1138 /* enable the rate scheduler */
1139 bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1141 IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1142 IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1148 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1149 * @q_vector: structure containing interrupt and ring information
1150 * @tx_ring: tx ring to clean
1151 * @napi_budget: Used to determine if we are in netpoll
1153 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1154 struct ixgbe_ring *tx_ring, int napi_budget)
1156 struct ixgbe_adapter *adapter = q_vector->adapter;
1157 struct ixgbe_tx_buffer *tx_buffer;
1158 union ixgbe_adv_tx_desc *tx_desc;
1159 unsigned int total_bytes = 0, total_packets = 0;
1160 unsigned int budget = q_vector->tx.work_limit;
1161 unsigned int i = tx_ring->next_to_clean;
1163 if (test_bit(__IXGBE_DOWN, &adapter->state))
1166 tx_buffer = &tx_ring->tx_buffer_info[i];
1167 tx_desc = IXGBE_TX_DESC(tx_ring, i);
1168 i -= tx_ring->count;
1171 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1173 /* if next_to_watch is not set then there is no work pending */
1177 /* prevent any other reads prior to eop_desc */
1178 read_barrier_depends();
1180 /* if DD is not set pending work has not been completed */
1181 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1184 /* clear next_to_watch to prevent false hangs */
1185 tx_buffer->next_to_watch = NULL;
1187 /* update the statistics for this packet */
1188 total_bytes += tx_buffer->bytecount;
1189 total_packets += tx_buffer->gso_segs;
1192 napi_consume_skb(tx_buffer->skb, napi_budget);
1194 /* unmap skb header data */
1195 dma_unmap_single(tx_ring->dev,
1196 dma_unmap_addr(tx_buffer, dma),
1197 dma_unmap_len(tx_buffer, len),
1200 /* clear tx_buffer data */
1201 tx_buffer->skb = NULL;
1202 dma_unmap_len_set(tx_buffer, len, 0);
1204 /* unmap remaining buffers */
1205 while (tx_desc != eop_desc) {
1210 i -= tx_ring->count;
1211 tx_buffer = tx_ring->tx_buffer_info;
1212 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1215 /* unmap any remaining paged data */
1216 if (dma_unmap_len(tx_buffer, len)) {
1217 dma_unmap_page(tx_ring->dev,
1218 dma_unmap_addr(tx_buffer, dma),
1219 dma_unmap_len(tx_buffer, len),
1221 dma_unmap_len_set(tx_buffer, len, 0);
1225 /* move us one more past the eop_desc for start of next pkt */
1230 i -= tx_ring->count;
1231 tx_buffer = tx_ring->tx_buffer_info;
1232 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1235 /* issue prefetch for next Tx descriptor */
1238 /* update budget accounting */
1240 } while (likely(budget));
1242 i += tx_ring->count;
1243 tx_ring->next_to_clean = i;
1244 u64_stats_update_begin(&tx_ring->syncp);
1245 tx_ring->stats.bytes += total_bytes;
1246 tx_ring->stats.packets += total_packets;
1247 u64_stats_update_end(&tx_ring->syncp);
1248 q_vector->tx.total_bytes += total_bytes;
1249 q_vector->tx.total_packets += total_packets;
1251 if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1252 /* schedule immediate reset if we believe we hung */
1253 struct ixgbe_hw *hw = &adapter->hw;
1254 e_err(drv, "Detected Tx Unit Hang\n"
1256 " TDH, TDT <%x>, <%x>\n"
1257 " next_to_use <%x>\n"
1258 " next_to_clean <%x>\n"
1259 "tx_buffer_info[next_to_clean]\n"
1260 " time_stamp <%lx>\n"
1262 tx_ring->queue_index,
1263 IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1264 IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1265 tx_ring->next_to_use, i,
1266 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1268 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1271 "tx hang %d detected on queue %d, resetting adapter\n",
1272 adapter->tx_timeout_count + 1, tx_ring->queue_index);
1274 /* schedule immediate reset if we believe we hung */
1275 ixgbe_tx_timeout_reset(adapter);
1277 /* the adapter is about to reset, no point in enabling stuff */
1281 netdev_tx_completed_queue(txring_txq(tx_ring),
1282 total_packets, total_bytes);
1284 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1285 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1286 (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1287 /* Make sure that anybody stopping the queue after this
1288 * sees the new next_to_clean.
1291 if (__netif_subqueue_stopped(tx_ring->netdev,
1292 tx_ring->queue_index)
1293 && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1294 netif_wake_subqueue(tx_ring->netdev,
1295 tx_ring->queue_index);
1296 ++tx_ring->tx_stats.restart_queue;
1303 #ifdef CONFIG_IXGBE_DCA
1304 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1305 struct ixgbe_ring *tx_ring,
1308 struct ixgbe_hw *hw = &adapter->hw;
1312 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1313 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1315 switch (hw->mac.type) {
1316 case ixgbe_mac_82598EB:
1317 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1319 case ixgbe_mac_82599EB:
1320 case ixgbe_mac_X540:
1321 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1322 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1325 /* for unknown hardware do not write register */
1330 * We can enable relaxed ordering for reads, but not writes when
1331 * DCA is enabled. This is due to a known issue in some chipsets
1332 * which will cause the DCA tag to be cleared.
1334 txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1335 IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1336 IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1338 IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1341 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1342 struct ixgbe_ring *rx_ring,
1345 struct ixgbe_hw *hw = &adapter->hw;
1347 u8 reg_idx = rx_ring->reg_idx;
1349 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1350 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1352 switch (hw->mac.type) {
1353 case ixgbe_mac_82599EB:
1354 case ixgbe_mac_X540:
1355 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1362 * We can enable relaxed ordering for reads, but not writes when
1363 * DCA is enabled. This is due to a known issue in some chipsets
1364 * which will cause the DCA tag to be cleared.
1366 rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1367 IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1368 IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1370 IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1373 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1375 struct ixgbe_adapter *adapter = q_vector->adapter;
1376 struct ixgbe_ring *ring;
1377 int cpu = get_cpu();
1379 if (q_vector->cpu == cpu)
1382 ixgbe_for_each_ring(ring, q_vector->tx)
1383 ixgbe_update_tx_dca(adapter, ring, cpu);
1385 ixgbe_for_each_ring(ring, q_vector->rx)
1386 ixgbe_update_rx_dca(adapter, ring, cpu);
1388 q_vector->cpu = cpu;
1393 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1397 /* always use CB2 mode, difference is masked in the CB driver */
1398 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1399 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1400 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1402 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1403 IXGBE_DCA_CTRL_DCA_DISABLE);
1405 for (i = 0; i < adapter->num_q_vectors; i++) {
1406 adapter->q_vector[i]->cpu = -1;
1407 ixgbe_update_dca(adapter->q_vector[i]);
1411 static int __ixgbe_notify_dca(struct device *dev, void *data)
1413 struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1414 unsigned long event = *(unsigned long *)data;
1416 if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1420 case DCA_PROVIDER_ADD:
1421 /* if we're already enabled, don't do it again */
1422 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1424 if (dca_add_requester(dev) == 0) {
1425 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1426 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1427 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1430 /* Fall Through since DCA is disabled. */
1431 case DCA_PROVIDER_REMOVE:
1432 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1433 dca_remove_requester(dev);
1434 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1435 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1436 IXGBE_DCA_CTRL_DCA_DISABLE);
1444 #endif /* CONFIG_IXGBE_DCA */
1446 #define IXGBE_RSS_L4_TYPES_MASK \
1447 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1448 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1449 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1450 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1452 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1453 union ixgbe_adv_rx_desc *rx_desc,
1454 struct sk_buff *skb)
1458 if (!(ring->netdev->features & NETIF_F_RXHASH))
1461 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1462 IXGBE_RXDADV_RSSTYPE_MASK;
1467 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1468 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1469 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1474 * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1475 * @ring: structure containing ring specific data
1476 * @rx_desc: advanced rx descriptor
1478 * Returns : true if it is FCoE pkt
1480 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1481 union ixgbe_adv_rx_desc *rx_desc)
1483 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1485 return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1486 ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1487 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1488 IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1491 #endif /* IXGBE_FCOE */
1493 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1494 * @ring: structure containing ring specific data
1495 * @rx_desc: current Rx descriptor being processed
1496 * @skb: skb currently being received and modified
1498 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1499 union ixgbe_adv_rx_desc *rx_desc,
1500 struct sk_buff *skb)
1502 __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1503 bool encap_pkt = false;
1505 skb_checksum_none_assert(skb);
1507 /* Rx csum disabled */
1508 if (!(ring->netdev->features & NETIF_F_RXCSUM))
1511 /* check for VXLAN and Geneve packets */
1512 if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1514 skb->encapsulation = 1;
1517 /* if IP and error */
1518 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1519 ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1520 ring->rx_stats.csum_err++;
1524 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1527 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1529 * 82599 errata, UDP frames with a 0 checksum can be marked as
1532 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1533 test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1536 ring->rx_stats.csum_err++;
1540 /* It must be a TCP or UDP packet with a valid checksum */
1541 skb->ip_summed = CHECKSUM_UNNECESSARY;
1543 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1546 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1547 skb->ip_summed = CHECKSUM_NONE;
1550 /* If we checked the outer header let the stack know */
1551 skb->csum_level = 1;
1555 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1556 struct ixgbe_rx_buffer *bi)
1558 struct page *page = bi->page;
1561 /* since we are recycling buffers we should seldom need to alloc */
1565 /* alloc new page for storage */
1566 page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1567 if (unlikely(!page)) {
1568 rx_ring->rx_stats.alloc_rx_page_failed++;
1572 /* map page for use */
1573 dma = dma_map_page(rx_ring->dev, page, 0,
1574 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1577 * if mapping failed free memory back to system since
1578 * there isn't much point in holding memory we can't use
1580 if (dma_mapping_error(rx_ring->dev, dma)) {
1581 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1583 rx_ring->rx_stats.alloc_rx_page_failed++;
1589 bi->page_offset = 0;
1595 * ixgbe_alloc_rx_buffers - Replace used receive buffers
1596 * @rx_ring: ring to place buffers on
1597 * @cleaned_count: number of buffers to replace
1599 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1601 union ixgbe_adv_rx_desc *rx_desc;
1602 struct ixgbe_rx_buffer *bi;
1603 u16 i = rx_ring->next_to_use;
1609 rx_desc = IXGBE_RX_DESC(rx_ring, i);
1610 bi = &rx_ring->rx_buffer_info[i];
1611 i -= rx_ring->count;
1614 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1618 * Refresh the desc even if buffer_addrs didn't change
1619 * because each write-back erases this info.
1621 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1627 rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1628 bi = rx_ring->rx_buffer_info;
1629 i -= rx_ring->count;
1632 /* clear the status bits for the next_to_use descriptor */
1633 rx_desc->wb.upper.status_error = 0;
1636 } while (cleaned_count);
1638 i += rx_ring->count;
1640 if (rx_ring->next_to_use != i) {
1641 rx_ring->next_to_use = i;
1643 /* update next to alloc since we have filled the ring */
1644 rx_ring->next_to_alloc = i;
1646 /* Force memory writes to complete before letting h/w
1647 * know there are new descriptors to fetch. (Only
1648 * applicable for weak-ordered memory model archs,
1652 writel(i, rx_ring->tail);
1656 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1657 struct sk_buff *skb)
1659 u16 hdr_len = skb_headlen(skb);
1661 /* set gso_size to avoid messing up TCP MSS */
1662 skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1663 IXGBE_CB(skb)->append_cnt);
1664 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1667 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1668 struct sk_buff *skb)
1670 /* if append_cnt is 0 then frame is not RSC */
1671 if (!IXGBE_CB(skb)->append_cnt)
1674 rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1675 rx_ring->rx_stats.rsc_flush++;
1677 ixgbe_set_rsc_gso_size(rx_ring, skb);
1679 /* gso_size is computed using append_cnt so always clear it last */
1680 IXGBE_CB(skb)->append_cnt = 0;
1684 * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1685 * @rx_ring: rx descriptor ring packet is being transacted on
1686 * @rx_desc: pointer to the EOP Rx descriptor
1687 * @skb: pointer to current skb being populated
1689 * This function checks the ring, descriptor, and packet information in
1690 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1691 * other fields within the skb.
1693 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1694 union ixgbe_adv_rx_desc *rx_desc,
1695 struct sk_buff *skb)
1697 struct net_device *dev = rx_ring->netdev;
1698 u32 flags = rx_ring->q_vector->adapter->flags;
1700 ixgbe_update_rsc_stats(rx_ring, skb);
1702 ixgbe_rx_hash(rx_ring, rx_desc, skb);
1704 ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1706 if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1707 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1709 if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1710 ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1711 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1712 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1715 skb_record_rx_queue(skb, rx_ring->queue_index);
1717 skb->protocol = eth_type_trans(skb, dev);
1720 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1721 struct sk_buff *skb)
1723 napi_gro_receive(&q_vector->napi, skb);
1727 * ixgbe_is_non_eop - process handling of non-EOP buffers
1728 * @rx_ring: Rx ring being processed
1729 * @rx_desc: Rx descriptor for current buffer
1730 * @skb: Current socket buffer containing buffer in progress
1732 * This function updates next to clean. If the buffer is an EOP buffer
1733 * this function exits returning false, otherwise it will place the
1734 * sk_buff in the next buffer to be chained and return true indicating
1735 * that this is in fact a non-EOP buffer.
1737 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1738 union ixgbe_adv_rx_desc *rx_desc,
1739 struct sk_buff *skb)
1741 u32 ntc = rx_ring->next_to_clean + 1;
1743 /* fetch, update, and store next to clean */
1744 ntc = (ntc < rx_ring->count) ? ntc : 0;
1745 rx_ring->next_to_clean = ntc;
1747 prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1749 /* update RSC append count if present */
1750 if (ring_is_rsc_enabled(rx_ring)) {
1751 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1752 cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1754 if (unlikely(rsc_enabled)) {
1755 u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1757 rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1758 IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1760 /* update ntc based on RSC value */
1761 ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1762 ntc &= IXGBE_RXDADV_NEXTP_MASK;
1763 ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1767 /* if we are the last buffer then there is nothing else to do */
1768 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1771 /* place skb in next buffer to be received */
1772 rx_ring->rx_buffer_info[ntc].skb = skb;
1773 rx_ring->rx_stats.non_eop_descs++;
1779 * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1780 * @rx_ring: rx descriptor ring packet is being transacted on
1781 * @skb: pointer to current skb being adjusted
1783 * This function is an ixgbe specific version of __pskb_pull_tail. The
1784 * main difference between this version and the original function is that
1785 * this function can make several assumptions about the state of things
1786 * that allow for significant optimizations versus the standard function.
1787 * As a result we can do things like drop a frag and maintain an accurate
1788 * truesize for the skb.
1790 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1791 struct sk_buff *skb)
1793 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1795 unsigned int pull_len;
1798 * it is valid to use page_address instead of kmap since we are
1799 * working with pages allocated out of the lomem pool per
1800 * alloc_page(GFP_ATOMIC)
1802 va = skb_frag_address(frag);
1805 * we need the header to contain the greater of either ETH_HLEN or
1806 * 60 bytes if the skb->len is less than 60 for skb_pad.
1808 pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1810 /* align pull length to size of long to optimize memcpy performance */
1811 skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1813 /* update all of the pointers */
1814 skb_frag_size_sub(frag, pull_len);
1815 frag->page_offset += pull_len;
1816 skb->data_len -= pull_len;
1817 skb->tail += pull_len;
1821 * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1822 * @rx_ring: rx descriptor ring packet is being transacted on
1823 * @skb: pointer to current skb being updated
1825 * This function provides a basic DMA sync up for the first fragment of an
1826 * skb. The reason for doing this is that the first fragment cannot be
1827 * unmapped until we have reached the end of packet descriptor for a buffer
1830 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1831 struct sk_buff *skb)
1833 /* if the page was released unmap it, else just sync our portion */
1834 if (unlikely(IXGBE_CB(skb)->page_released)) {
1835 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1836 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1837 IXGBE_CB(skb)->page_released = false;
1839 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1841 dma_sync_single_range_for_cpu(rx_ring->dev,
1844 ixgbe_rx_bufsz(rx_ring),
1847 IXGBE_CB(skb)->dma = 0;
1851 * ixgbe_cleanup_headers - Correct corrupted or empty headers
1852 * @rx_ring: rx descriptor ring packet is being transacted on
1853 * @rx_desc: pointer to the EOP Rx descriptor
1854 * @skb: pointer to current skb being fixed
1856 * Check for corrupted packet headers caused by senders on the local L2
1857 * embedded NIC switch not setting up their Tx Descriptors right. These
1858 * should be very rare.
1860 * Also address the case where we are pulling data in on pages only
1861 * and as such no data is present in the skb header.
1863 * In addition if skb is not at least 60 bytes we need to pad it so that
1864 * it is large enough to qualify as a valid Ethernet frame.
1866 * Returns true if an error was encountered and skb was freed.
1868 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1869 union ixgbe_adv_rx_desc *rx_desc,
1870 struct sk_buff *skb)
1872 struct net_device *netdev = rx_ring->netdev;
1874 /* verify that the packet does not have any known errors */
1875 if (unlikely(ixgbe_test_staterr(rx_desc,
1876 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1877 !(netdev->features & NETIF_F_RXALL))) {
1878 dev_kfree_skb_any(skb);
1882 /* place header in linear portion of buffer */
1883 if (skb_is_nonlinear(skb))
1884 ixgbe_pull_tail(rx_ring, skb);
1887 /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1888 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1892 /* if eth_skb_pad returns an error the skb was freed */
1893 if (eth_skb_pad(skb))
1900 * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1901 * @rx_ring: rx descriptor ring to store buffers on
1902 * @old_buff: donor buffer to have page reused
1904 * Synchronizes page for reuse by the adapter
1906 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1907 struct ixgbe_rx_buffer *old_buff)
1909 struct ixgbe_rx_buffer *new_buff;
1910 u16 nta = rx_ring->next_to_alloc;
1912 new_buff = &rx_ring->rx_buffer_info[nta];
1914 /* update, and store next to alloc */
1916 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1918 /* transfer page from old buffer to new buffer */
1919 *new_buff = *old_buff;
1921 /* sync the buffer for use by the device */
1922 dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1923 new_buff->page_offset,
1924 ixgbe_rx_bufsz(rx_ring),
1928 static inline bool ixgbe_page_is_reserved(struct page *page)
1930 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1934 * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1935 * @rx_ring: rx descriptor ring to transact packets on
1936 * @rx_buffer: buffer containing page to add
1937 * @rx_desc: descriptor containing length of buffer written by hardware
1938 * @skb: sk_buff to place the data into
1940 * This function will add the data contained in rx_buffer->page to the skb.
1941 * This is done either through a direct copy if the data in the buffer is
1942 * less than the skb header size, otherwise it will just attach the page as
1943 * a frag to the skb.
1945 * The function will then update the page offset if necessary and return
1946 * true if the buffer can be reused by the adapter.
1948 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1949 struct ixgbe_rx_buffer *rx_buffer,
1950 union ixgbe_adv_rx_desc *rx_desc,
1951 struct sk_buff *skb)
1953 struct page *page = rx_buffer->page;
1954 unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1955 #if (PAGE_SIZE < 8192)
1956 unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1958 unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1959 unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1960 ixgbe_rx_bufsz(rx_ring);
1963 if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1964 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1966 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1968 /* page is not reserved, we can reuse buffer as-is */
1969 if (likely(!ixgbe_page_is_reserved(page)))
1972 /* this page cannot be reused so discard it */
1973 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1977 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1978 rx_buffer->page_offset, size, truesize);
1980 /* avoid re-using remote pages */
1981 if (unlikely(ixgbe_page_is_reserved(page)))
1984 #if (PAGE_SIZE < 8192)
1985 /* if we are only owner of page we can reuse it */
1986 if (unlikely(page_count(page) != 1))
1989 /* flip page offset to other buffer */
1990 rx_buffer->page_offset ^= truesize;
1992 /* move offset up to the next cache line */
1993 rx_buffer->page_offset += truesize;
1995 if (rx_buffer->page_offset > last_offset)
1999 /* Even if we own the page, we are not allowed to use atomic_set()
2000 * This would break get_page_unless_zero() users.
2007 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
2008 union ixgbe_adv_rx_desc *rx_desc)
2010 struct ixgbe_rx_buffer *rx_buffer;
2011 struct sk_buff *skb;
2014 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2015 page = rx_buffer->page;
2018 skb = rx_buffer->skb;
2021 void *page_addr = page_address(page) +
2022 rx_buffer->page_offset;
2024 /* prefetch first cache line of first page */
2025 prefetch(page_addr);
2026 #if L1_CACHE_BYTES < 128
2027 prefetch(page_addr + L1_CACHE_BYTES);
2030 /* allocate a skb to store the frags */
2031 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
2033 if (unlikely(!skb)) {
2034 rx_ring->rx_stats.alloc_rx_buff_failed++;
2039 * we will be copying header into skb->data in
2040 * pskb_may_pull so it is in our interest to prefetch
2041 * it now to avoid a possible cache miss
2043 prefetchw(skb->data);
2046 * Delay unmapping of the first packet. It carries the
2047 * header information, HW may still access the header
2048 * after the writeback. Only unmap it when EOP is
2051 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
2054 IXGBE_CB(skb)->dma = rx_buffer->dma;
2056 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2057 ixgbe_dma_sync_frag(rx_ring, skb);
2060 /* we are reusing so sync this buffer for CPU use */
2061 dma_sync_single_range_for_cpu(rx_ring->dev,
2063 rx_buffer->page_offset,
2064 ixgbe_rx_bufsz(rx_ring),
2067 rx_buffer->skb = NULL;
2070 /* pull page into skb */
2071 if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2072 /* hand second half of page back to the ring */
2073 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2074 } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2075 /* the page has been released from the ring */
2076 IXGBE_CB(skb)->page_released = true;
2078 /* we are not reusing the buffer so unmap it */
2079 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2080 ixgbe_rx_pg_size(rx_ring),
2084 /* clear contents of buffer_info */
2085 rx_buffer->page = NULL;
2091 * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2092 * @q_vector: structure containing interrupt and ring information
2093 * @rx_ring: rx descriptor ring to transact packets on
2094 * @budget: Total limit on number of packets to process
2096 * This function provides a "bounce buffer" approach to Rx interrupt
2097 * processing. The advantage to this is that on systems that have
2098 * expensive overhead for IOMMU access this provides a means of avoiding
2099 * it by maintaining the mapping of the page to the syste.
2101 * Returns amount of work completed
2103 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2104 struct ixgbe_ring *rx_ring,
2107 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2109 struct ixgbe_adapter *adapter = q_vector->adapter;
2111 unsigned int mss = 0;
2112 #endif /* IXGBE_FCOE */
2113 u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2115 while (likely(total_rx_packets < budget)) {
2116 union ixgbe_adv_rx_desc *rx_desc;
2117 struct sk_buff *skb;
2119 /* return some buffers to hardware, one at a time is too slow */
2120 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2121 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2125 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2127 if (!rx_desc->wb.upper.status_error)
2130 /* This memory barrier is needed to keep us from reading
2131 * any other fields out of the rx_desc until we know the
2132 * descriptor has been written back
2136 /* retrieve a buffer from the ring */
2137 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2139 /* exit if we failed to retrieve a buffer */
2145 /* place incomplete frames back on ring for completion */
2146 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2149 /* verify the packet layout is correct */
2150 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2153 /* probably a little skewed due to removing CRC */
2154 total_rx_bytes += skb->len;
2156 /* populate checksum, timestamp, VLAN, and protocol */
2157 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2160 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2161 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2162 ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2163 /* include DDPed FCoE data */
2164 if (ddp_bytes > 0) {
2166 mss = rx_ring->netdev->mtu -
2167 sizeof(struct fcoe_hdr) -
2168 sizeof(struct fc_frame_header) -
2169 sizeof(struct fcoe_crc_eof);
2173 total_rx_bytes += ddp_bytes;
2174 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2178 dev_kfree_skb_any(skb);
2183 #endif /* IXGBE_FCOE */
2184 ixgbe_rx_skb(q_vector, skb);
2186 /* update budget accounting */
2190 u64_stats_update_begin(&rx_ring->syncp);
2191 rx_ring->stats.packets += total_rx_packets;
2192 rx_ring->stats.bytes += total_rx_bytes;
2193 u64_stats_update_end(&rx_ring->syncp);
2194 q_vector->rx.total_packets += total_rx_packets;
2195 q_vector->rx.total_bytes += total_rx_bytes;
2197 return total_rx_packets;
2201 * ixgbe_configure_msix - Configure MSI-X hardware
2202 * @adapter: board private structure
2204 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2207 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2209 struct ixgbe_q_vector *q_vector;
2213 /* Populate MSIX to EITR Select */
2214 if (adapter->num_vfs > 32) {
2215 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2216 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2220 * Populate the IVAR table and set the ITR values to the
2221 * corresponding register.
2223 for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2224 struct ixgbe_ring *ring;
2225 q_vector = adapter->q_vector[v_idx];
2227 ixgbe_for_each_ring(ring, q_vector->rx)
2228 ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2230 ixgbe_for_each_ring(ring, q_vector->tx)
2231 ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2233 ixgbe_write_eitr(q_vector);
2236 switch (adapter->hw.mac.type) {
2237 case ixgbe_mac_82598EB:
2238 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2241 case ixgbe_mac_82599EB:
2242 case ixgbe_mac_X540:
2243 case ixgbe_mac_X550:
2244 case ixgbe_mac_X550EM_x:
2245 case ixgbe_mac_x550em_a:
2246 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2251 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2253 /* set up to autoclear timer, and the vectors */
2254 mask = IXGBE_EIMS_ENABLE_MASK;
2255 mask &= ~(IXGBE_EIMS_OTHER |
2256 IXGBE_EIMS_MAILBOX |
2259 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2262 enum latency_range {
2266 latency_invalid = 255
2270 * ixgbe_update_itr - update the dynamic ITR value based on statistics
2271 * @q_vector: structure containing interrupt and ring information
2272 * @ring_container: structure containing ring performance data
2274 * Stores a new ITR value based on packets and byte
2275 * counts during the last interrupt. The advantage of per interrupt
2276 * computation is faster updates and more accurate ITR for the current
2277 * traffic pattern. Constants in this function were computed
2278 * based on theoretical maximum wire speed and thresholds were set based
2279 * on testing data as well as attempting to minimize response time
2280 * while increasing bulk throughput.
2281 * this functionality is controlled by the InterruptThrottleRate module
2282 * parameter (see ixgbe_param.c)
2284 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2285 struct ixgbe_ring_container *ring_container)
2287 int bytes = ring_container->total_bytes;
2288 int packets = ring_container->total_packets;
2291 u8 itr_setting = ring_container->itr;
2296 /* simple throttlerate management
2297 * 0-10MB/s lowest (100000 ints/s)
2298 * 10-20MB/s low (20000 ints/s)
2299 * 20-1249MB/s bulk (12000 ints/s)
2301 /* what was last interrupt timeslice? */
2302 timepassed_us = q_vector->itr >> 2;
2303 if (timepassed_us == 0)
2306 bytes_perint = bytes / timepassed_us; /* bytes/usec */
2308 switch (itr_setting) {
2309 case lowest_latency:
2310 if (bytes_perint > 10)
2311 itr_setting = low_latency;
2314 if (bytes_perint > 20)
2315 itr_setting = bulk_latency;
2316 else if (bytes_perint <= 10)
2317 itr_setting = lowest_latency;
2320 if (bytes_perint <= 20)
2321 itr_setting = low_latency;
2325 /* clear work counters since we have the values we need */
2326 ring_container->total_bytes = 0;
2327 ring_container->total_packets = 0;
2329 /* write updated itr to ring container */
2330 ring_container->itr = itr_setting;
2334 * ixgbe_write_eitr - write EITR register in hardware specific way
2335 * @q_vector: structure containing interrupt and ring information
2337 * This function is made to be called by ethtool and by the driver
2338 * when it needs to update EITR registers at runtime. Hardware
2339 * specific quirks/differences are taken care of here.
2341 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2343 struct ixgbe_adapter *adapter = q_vector->adapter;
2344 struct ixgbe_hw *hw = &adapter->hw;
2345 int v_idx = q_vector->v_idx;
2346 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2348 switch (adapter->hw.mac.type) {
2349 case ixgbe_mac_82598EB:
2350 /* must write high and low 16 bits to reset counter */
2351 itr_reg |= (itr_reg << 16);
2353 case ixgbe_mac_82599EB:
2354 case ixgbe_mac_X540:
2355 case ixgbe_mac_X550:
2356 case ixgbe_mac_X550EM_x:
2357 case ixgbe_mac_x550em_a:
2359 * set the WDIS bit to not clear the timer bits and cause an
2360 * immediate assertion of the interrupt
2362 itr_reg |= IXGBE_EITR_CNT_WDIS;
2367 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2370 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2372 u32 new_itr = q_vector->itr;
2375 ixgbe_update_itr(q_vector, &q_vector->tx);
2376 ixgbe_update_itr(q_vector, &q_vector->rx);
2378 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2380 switch (current_itr) {
2381 /* counts and packets in update_itr are dependent on these numbers */
2382 case lowest_latency:
2383 new_itr = IXGBE_100K_ITR;
2386 new_itr = IXGBE_20K_ITR;
2389 new_itr = IXGBE_12K_ITR;
2395 if (new_itr != q_vector->itr) {
2396 /* do an exponential smoothing */
2397 new_itr = (10 * new_itr * q_vector->itr) /
2398 ((9 * new_itr) + q_vector->itr);
2400 /* save the algorithm value here */
2401 q_vector->itr = new_itr;
2403 ixgbe_write_eitr(q_vector);
2408 * ixgbe_check_overtemp_subtask - check for over temperature
2409 * @adapter: pointer to adapter
2411 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2413 struct ixgbe_hw *hw = &adapter->hw;
2414 u32 eicr = adapter->interrupt_event;
2417 if (test_bit(__IXGBE_DOWN, &adapter->state))
2420 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2421 !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2424 adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2426 switch (hw->device_id) {
2427 case IXGBE_DEV_ID_82599_T3_LOM:
2429 * Since the warning interrupt is for both ports
2430 * we don't have to check if:
2431 * - This interrupt wasn't for our port.
2432 * - We may have missed the interrupt so always have to
2433 * check if we got a LSC
2435 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2436 !(eicr & IXGBE_EICR_LSC))
2439 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2441 bool link_up = false;
2443 hw->mac.ops.check_link(hw, &speed, &link_up, false);
2449 /* Check if this is not due to overtemp */
2450 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2454 case IXGBE_DEV_ID_X550EM_A_1G_T:
2455 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2456 rc = hw->phy.ops.check_overtemp(hw);
2457 if (rc != IXGBE_ERR_OVERTEMP)
2461 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2463 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2467 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2469 adapter->interrupt_event = 0;
2472 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2474 struct ixgbe_hw *hw = &adapter->hw;
2476 if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2477 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2478 e_crit(probe, "Fan has stopped, replace the adapter\n");
2479 /* write to clear the interrupt */
2480 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2484 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2486 struct ixgbe_hw *hw = &adapter->hw;
2488 if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2491 switch (adapter->hw.mac.type) {
2492 case ixgbe_mac_82599EB:
2494 * Need to check link state so complete overtemp check
2497 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2498 (eicr & IXGBE_EICR_LSC)) &&
2499 (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2500 adapter->interrupt_event = eicr;
2501 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2502 ixgbe_service_event_schedule(adapter);
2506 case ixgbe_mac_x550em_a:
2507 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2508 adapter->interrupt_event = eicr;
2509 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2510 ixgbe_service_event_schedule(adapter);
2511 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2512 IXGBE_EICR_GPI_SDP0_X550EM_a);
2513 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2514 IXGBE_EICR_GPI_SDP0_X550EM_a);
2517 case ixgbe_mac_X550:
2518 case ixgbe_mac_X540:
2519 if (!(eicr & IXGBE_EICR_TS))
2526 e_crit(drv, "%s\n", ixgbe_overheat_msg);
2529 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2531 switch (hw->mac.type) {
2532 case ixgbe_mac_82598EB:
2533 if (hw->phy.type == ixgbe_phy_nl)
2536 case ixgbe_mac_82599EB:
2537 case ixgbe_mac_X550EM_x:
2538 case ixgbe_mac_x550em_a:
2539 switch (hw->mac.ops.get_media_type(hw)) {
2540 case ixgbe_media_type_fiber:
2541 case ixgbe_media_type_fiber_qsfp:
2551 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2553 struct ixgbe_hw *hw = &adapter->hw;
2554 u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2556 if (!ixgbe_is_sfp(hw))
2559 /* Later MAC's use different SDP */
2560 if (hw->mac.type >= ixgbe_mac_X540)
2561 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2563 if (eicr & eicr_mask) {
2564 /* Clear the interrupt */
2565 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2566 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2567 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2568 adapter->sfp_poll_time = 0;
2569 ixgbe_service_event_schedule(adapter);
2573 if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2574 (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2575 /* Clear the interrupt */
2576 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2577 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2578 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2579 ixgbe_service_event_schedule(adapter);
2584 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2586 struct ixgbe_hw *hw = &adapter->hw;
2589 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2590 adapter->link_check_timeout = jiffies;
2591 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2592 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2593 IXGBE_WRITE_FLUSH(hw);
2594 ixgbe_service_event_schedule(adapter);
2598 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2602 struct ixgbe_hw *hw = &adapter->hw;
2604 switch (hw->mac.type) {
2605 case ixgbe_mac_82598EB:
2606 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2607 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2609 case ixgbe_mac_82599EB:
2610 case ixgbe_mac_X540:
2611 case ixgbe_mac_X550:
2612 case ixgbe_mac_X550EM_x:
2613 case ixgbe_mac_x550em_a:
2614 mask = (qmask & 0xFFFFFFFF);
2616 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2617 mask = (qmask >> 32);
2619 IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2624 /* skip the flush */
2627 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2631 struct ixgbe_hw *hw = &adapter->hw;
2633 switch (hw->mac.type) {
2634 case ixgbe_mac_82598EB:
2635 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2636 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2638 case ixgbe_mac_82599EB:
2639 case ixgbe_mac_X540:
2640 case ixgbe_mac_X550:
2641 case ixgbe_mac_X550EM_x:
2642 case ixgbe_mac_x550em_a:
2643 mask = (qmask & 0xFFFFFFFF);
2645 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2646 mask = (qmask >> 32);
2648 IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2653 /* skip the flush */
2657 * ixgbe_irq_enable - Enable default interrupt generation settings
2658 * @adapter: board private structure
2660 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2663 struct ixgbe_hw *hw = &adapter->hw;
2664 u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2666 /* don't reenable LSC while waiting for link */
2667 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2668 mask &= ~IXGBE_EIMS_LSC;
2670 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2671 switch (adapter->hw.mac.type) {
2672 case ixgbe_mac_82599EB:
2673 mask |= IXGBE_EIMS_GPI_SDP0(hw);
2675 case ixgbe_mac_X540:
2676 case ixgbe_mac_X550:
2677 case ixgbe_mac_X550EM_x:
2678 case ixgbe_mac_x550em_a:
2679 mask |= IXGBE_EIMS_TS;
2684 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2685 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2686 switch (adapter->hw.mac.type) {
2687 case ixgbe_mac_82599EB:
2688 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2689 mask |= IXGBE_EIMS_GPI_SDP2(hw);
2691 case ixgbe_mac_X540:
2692 case ixgbe_mac_X550:
2693 case ixgbe_mac_X550EM_x:
2694 case ixgbe_mac_x550em_a:
2695 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
2696 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
2697 adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
2698 mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
2699 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
2700 mask |= IXGBE_EICR_GPI_SDP0_X540;
2701 mask |= IXGBE_EIMS_ECC;
2702 mask |= IXGBE_EIMS_MAILBOX;
2708 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2709 !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2710 mask |= IXGBE_EIMS_FLOW_DIR;
2712 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2714 ixgbe_irq_enable_queues(adapter, ~0);
2716 IXGBE_WRITE_FLUSH(&adapter->hw);
2719 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2721 struct ixgbe_adapter *adapter = data;
2722 struct ixgbe_hw *hw = &adapter->hw;
2726 * Workaround for Silicon errata. Use clear-by-write instead
2727 * of clear-by-read. Reading with EICS will return the
2728 * interrupt causes without clearing, which later be done
2729 * with the write to EICR.
2731 eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2733 /* The lower 16bits of the EICR register are for the queue interrupts
2734 * which should be masked here in order to not accidentally clear them if
2735 * the bits are high when ixgbe_msix_other is called. There is a race
2736 * condition otherwise which results in possible performance loss
2737 * especially if the ixgbe_msix_other interrupt is triggering
2738 * consistently (as it would when PPS is turned on for the X540 device)
2742 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2744 if (eicr & IXGBE_EICR_LSC)
2745 ixgbe_check_lsc(adapter);
2747 if (eicr & IXGBE_EICR_MAILBOX)
2748 ixgbe_msg_task(adapter);
2750 switch (hw->mac.type) {
2751 case ixgbe_mac_82599EB:
2752 case ixgbe_mac_X540:
2753 case ixgbe_mac_X550:
2754 case ixgbe_mac_X550EM_x:
2755 case ixgbe_mac_x550em_a:
2756 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
2757 (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
2758 adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
2759 ixgbe_service_event_schedule(adapter);
2760 IXGBE_WRITE_REG(hw, IXGBE_EICR,
2761 IXGBE_EICR_GPI_SDP0_X540);
2763 if (eicr & IXGBE_EICR_ECC) {
2764 e_info(link, "Received ECC Err, initiating reset\n");
2765 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
2766 ixgbe_service_event_schedule(adapter);
2767 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2769 /* Handle Flow Director Full threshold interrupt */
2770 if (eicr & IXGBE_EICR_FLOW_DIR) {
2771 int reinit_count = 0;
2773 for (i = 0; i < adapter->num_tx_queues; i++) {
2774 struct ixgbe_ring *ring = adapter->tx_ring[i];
2775 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2780 /* no more flow director interrupts until after init */
2781 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2782 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2783 ixgbe_service_event_schedule(adapter);
2786 ixgbe_check_sfp_event(adapter, eicr);
2787 ixgbe_check_overtemp_event(adapter, eicr);
2793 ixgbe_check_fan_failure(adapter, eicr);
2795 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2796 ixgbe_ptp_check_pps_event(adapter);
2798 /* re-enable the original interrupt state, no lsc, no queues */
2799 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2800 ixgbe_irq_enable(adapter, false, false);
2805 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2807 struct ixgbe_q_vector *q_vector = data;
2809 /* EIAM disabled interrupts (on this vector) for us */
2811 if (q_vector->rx.ring || q_vector->tx.ring)
2812 napi_schedule_irqoff(&q_vector->napi);
2818 * ixgbe_poll - NAPI Rx polling callback
2819 * @napi: structure for representing this polling device
2820 * @budget: how many packets driver is allowed to clean
2822 * This function is used for legacy and MSI, NAPI mode
2824 int ixgbe_poll(struct napi_struct *napi, int budget)
2826 struct ixgbe_q_vector *q_vector =
2827 container_of(napi, struct ixgbe_q_vector, napi);
2828 struct ixgbe_adapter *adapter = q_vector->adapter;
2829 struct ixgbe_ring *ring;
2830 int per_ring_budget, work_done = 0;
2831 bool clean_complete = true;
2833 #ifdef CONFIG_IXGBE_DCA
2834 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2835 ixgbe_update_dca(q_vector);
2838 ixgbe_for_each_ring(ring, q_vector->tx) {
2839 if (!ixgbe_clean_tx_irq(q_vector, ring, budget))
2840 clean_complete = false;
2843 /* Exit if we are called by netpoll */
2847 /* attempt to distribute budget to each queue fairly, but don't allow
2848 * the budget to go below 1 because we'll exit polling */
2849 if (q_vector->rx.count > 1)
2850 per_ring_budget = max(budget/q_vector->rx.count, 1);
2852 per_ring_budget = budget;
2854 ixgbe_for_each_ring(ring, q_vector->rx) {
2855 int cleaned = ixgbe_clean_rx_irq(q_vector, ring,
2858 work_done += cleaned;
2859 if (cleaned >= per_ring_budget)
2860 clean_complete = false;
2863 /* If all work not completed, return budget and keep polling */
2864 if (!clean_complete)
2867 /* all work done, exit the polling mode */
2868 napi_complete_done(napi, work_done);
2869 if (adapter->rx_itr_setting & 1)
2870 ixgbe_set_itr(q_vector);
2871 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2872 ixgbe_irq_enable_queues(adapter, BIT_ULL(q_vector->v_idx));
2874 return min(work_done, budget - 1);
2878 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2879 * @adapter: board private structure
2881 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2882 * interrupts from the kernel.
2884 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2886 struct net_device *netdev = adapter->netdev;
2890 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2891 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2892 struct msix_entry *entry = &adapter->msix_entries[vector];
2894 if (q_vector->tx.ring && q_vector->rx.ring) {
2895 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2896 "%s-%s-%d", netdev->name, "TxRx", ri++);
2898 } else if (q_vector->rx.ring) {
2899 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2900 "%s-%s-%d", netdev->name, "rx", ri++);
2901 } else if (q_vector->tx.ring) {
2902 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2903 "%s-%s-%d", netdev->name, "tx", ti++);
2905 /* skip this unused q_vector */
2908 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2909 q_vector->name, q_vector);
2911 e_err(probe, "request_irq failed for MSIX interrupt "
2912 "Error: %d\n", err);
2913 goto free_queue_irqs;
2915 /* If Flow Director is enabled, set interrupt affinity */
2916 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2917 /* assign the mask for this irq */
2918 irq_set_affinity_hint(entry->vector,
2919 &q_vector->affinity_mask);
2923 err = request_irq(adapter->msix_entries[vector].vector,
2924 ixgbe_msix_other, 0, netdev->name, adapter);
2926 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2927 goto free_queue_irqs;
2935 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2937 free_irq(adapter->msix_entries[vector].vector,
2938 adapter->q_vector[vector]);
2940 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2941 pci_disable_msix(adapter->pdev);
2942 kfree(adapter->msix_entries);
2943 adapter->msix_entries = NULL;
2948 * ixgbe_intr - legacy mode Interrupt Handler
2949 * @irq: interrupt number
2950 * @data: pointer to a network interface device structure
2952 static irqreturn_t ixgbe_intr(int irq, void *data)
2954 struct ixgbe_adapter *adapter = data;
2955 struct ixgbe_hw *hw = &adapter->hw;
2956 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2960 * Workaround for silicon errata #26 on 82598. Mask the interrupt
2961 * before the read of EICR.
2963 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2965 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2966 * therefore no explicit interrupt disable is necessary */
2967 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2970 * shared interrupt alert!
2971 * make sure interrupts are enabled because the read will
2972 * have disabled interrupts due to EIAM
2973 * finish the workaround of silicon errata on 82598. Unmask
2974 * the interrupt that we masked before the EICR read.
2976 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2977 ixgbe_irq_enable(adapter, true, true);
2978 return IRQ_NONE; /* Not our interrupt */
2981 if (eicr & IXGBE_EICR_LSC)
2982 ixgbe_check_lsc(adapter);
2984 switch (hw->mac.type) {
2985 case ixgbe_mac_82599EB:
2986 ixgbe_check_sfp_event(adapter, eicr);
2988 case ixgbe_mac_X540:
2989 case ixgbe_mac_X550:
2990 case ixgbe_mac_X550EM_x:
2991 case ixgbe_mac_x550em_a:
2992 if (eicr & IXGBE_EICR_ECC) {
2993 e_info(link, "Received ECC Err, initiating reset\n");
2994 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
2995 ixgbe_service_event_schedule(adapter);
2996 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2998 ixgbe_check_overtemp_event(adapter, eicr);
3004 ixgbe_check_fan_failure(adapter, eicr);
3005 if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3006 ixgbe_ptp_check_pps_event(adapter);
3008 /* would disable interrupts here but EIAM disabled it */
3009 napi_schedule_irqoff(&q_vector->napi);
3012 * re-enable link(maybe) and non-queue interrupts, no flush.
3013 * ixgbe_poll will re-enable the queue interrupts
3015 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3016 ixgbe_irq_enable(adapter, false, false);
3022 * ixgbe_request_irq - initialize interrupts
3023 * @adapter: board private structure
3025 * Attempts to configure interrupts using the best available
3026 * capabilities of the hardware and kernel.
3028 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3030 struct net_device *netdev = adapter->netdev;
3033 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3034 err = ixgbe_request_msix_irqs(adapter);
3035 else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3036 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3037 netdev->name, adapter);
3039 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3040 netdev->name, adapter);
3043 e_err(probe, "request_irq failed, Error %d\n", err);
3048 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3052 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3053 free_irq(adapter->pdev->irq, adapter);
3057 if (!adapter->msix_entries)
3060 for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3061 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3062 struct msix_entry *entry = &adapter->msix_entries[vector];
3064 /* free only the irqs that were actually requested */
3065 if (!q_vector->rx.ring && !q_vector->tx.ring)
3068 /* clear the affinity_mask in the IRQ descriptor */
3069 irq_set_affinity_hint(entry->vector, NULL);
3071 free_irq(entry->vector, q_vector);
3074 free_irq(adapter->msix_entries[vector].vector, adapter);
3078 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3079 * @adapter: board private structure
3081 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3083 switch (adapter->hw.mac.type) {
3084 case ixgbe_mac_82598EB:
3085 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3087 case ixgbe_mac_82599EB:
3088 case ixgbe_mac_X540:
3089 case ixgbe_mac_X550:
3090 case ixgbe_mac_X550EM_x:
3091 case ixgbe_mac_x550em_a:
3092 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3093 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3094 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3099 IXGBE_WRITE_FLUSH(&adapter->hw);
3100 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3103 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3104 synchronize_irq(adapter->msix_entries[vector].vector);
3106 synchronize_irq(adapter->msix_entries[vector++].vector);
3108 synchronize_irq(adapter->pdev->irq);
3113 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3116 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3118 struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3120 ixgbe_write_eitr(q_vector);
3122 ixgbe_set_ivar(adapter, 0, 0, 0);
3123 ixgbe_set_ivar(adapter, 1, 0, 0);
3125 e_info(hw, "Legacy interrupt IVAR setup done\n");
3129 * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3130 * @adapter: board private structure
3131 * @ring: structure containing ring specific data
3133 * Configure the Tx descriptor ring after a reset.
3135 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3136 struct ixgbe_ring *ring)
3138 struct ixgbe_hw *hw = &adapter->hw;
3139 u64 tdba = ring->dma;
3141 u32 txdctl = IXGBE_TXDCTL_ENABLE;
3142 u8 reg_idx = ring->reg_idx;
3144 /* disable queue to avoid issues while updating state */
3145 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3146 IXGBE_WRITE_FLUSH(hw);
3148 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3149 (tdba & DMA_BIT_MASK(32)));
3150 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3151 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3152 ring->count * sizeof(union ixgbe_adv_tx_desc));
3153 IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3154 IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3155 ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3158 * set WTHRESH to encourage burst writeback, it should not be set
3159 * higher than 1 when:
3160 * - ITR is 0 as it could cause false TX hangs
3161 * - ITR is set to > 100k int/sec and BQL is enabled
3163 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3164 * to or less than the number of on chip descriptors, which is
3167 if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3168 txdctl |= 1u << 16; /* WTHRESH = 1 */
3170 txdctl |= 8u << 16; /* WTHRESH = 8 */
3173 * Setting PTHRESH to 32 both improves performance
3174 * and avoids a TX hang with DFP enabled
3176 txdctl |= (1u << 8) | /* HTHRESH = 1 */
3177 32; /* PTHRESH = 32 */
3179 /* reinitialize flowdirector state */
3180 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3181 ring->atr_sample_rate = adapter->atr_sample_rate;
3182 ring->atr_count = 0;
3183 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3185 ring->atr_sample_rate = 0;
3188 /* initialize XPS */
3189 if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3190 struct ixgbe_q_vector *q_vector = ring->q_vector;
3193 netif_set_xps_queue(ring->netdev,
3194 &q_vector->affinity_mask,
3198 clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3201 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3203 /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3204 if (hw->mac.type == ixgbe_mac_82598EB &&
3205 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3208 /* poll to verify queue is enabled */
3210 usleep_range(1000, 2000);
3211 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3212 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3214 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3217 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3219 struct ixgbe_hw *hw = &adapter->hw;
3221 u8 tcs = netdev_get_num_tc(adapter->netdev);
3223 if (hw->mac.type == ixgbe_mac_82598EB)
3226 /* disable the arbiter while setting MTQC */
3227 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3228 rttdcs |= IXGBE_RTTDCS_ARBDIS;
3229 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3231 /* set transmit pool layout */
3232 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3233 mtqc = IXGBE_MTQC_VT_ENA;
3235 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3237 mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3238 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3239 IXGBE_82599_VMDQ_4Q_MASK)
3240 mtqc |= IXGBE_MTQC_32VF;
3242 mtqc |= IXGBE_MTQC_64VF;
3245 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3247 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3249 mtqc = IXGBE_MTQC_64Q_1PB;
3252 IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3254 /* Enable Security TX Buffer IFG for multiple pb */
3256 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3257 sectx |= IXGBE_SECTX_DCB;
3258 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3261 /* re-enable the arbiter */
3262 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3263 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3267 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3268 * @adapter: board private structure
3270 * Configure the Tx unit of the MAC after a reset.
3272 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3274 struct ixgbe_hw *hw = &adapter->hw;
3278 ixgbe_setup_mtqc(adapter);
3280 if (hw->mac.type != ixgbe_mac_82598EB) {
3281 /* DMATXCTL.EN must be before Tx queues are enabled */
3282 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3283 dmatxctl |= IXGBE_DMATXCTL_TE;
3284 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3287 /* Setup the HW Tx Head and Tail descriptor pointers */
3288 for (i = 0; i < adapter->num_tx_queues; i++)
3289 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3292 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3293 struct ixgbe_ring *ring)
3295 struct ixgbe_hw *hw = &adapter->hw;
3296 u8 reg_idx = ring->reg_idx;
3297 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3299 srrctl |= IXGBE_SRRCTL_DROP_EN;
3301 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3304 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3305 struct ixgbe_ring *ring)
3307 struct ixgbe_hw *hw = &adapter->hw;
3308 u8 reg_idx = ring->reg_idx;
3309 u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3311 srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3313 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3316 #ifdef CONFIG_IXGBE_DCB
3317 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3319 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3323 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3325 if (adapter->ixgbe_ieee_pfc)
3326 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3329 * We should set the drop enable bit if:
3332 * Number of Rx queues > 1 and flow control is disabled
3334 * This allows us to avoid head of line blocking for security
3335 * and performance reasons.
3337 if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3338 !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3339 for (i = 0; i < adapter->num_rx_queues; i++)
3340 ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3342 for (i = 0; i < adapter->num_rx_queues; i++)
3343 ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3347 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3349 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3350 struct ixgbe_ring *rx_ring)
3352 struct ixgbe_hw *hw = &adapter->hw;
3354 u8 reg_idx = rx_ring->reg_idx;
3356 if (hw->mac.type == ixgbe_mac_82598EB) {
3357 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3360 * if VMDq is not active we must program one srrctl register
3361 * per RSS queue since we have enabled RDRXCTL.MVMEN
3366 /* configure header buffer length, needed for RSC */
3367 srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3369 /* configure the packet buffer length */
3370 srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3372 /* configure descriptor type */
3373 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3375 IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3379 * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3380 * @adapter: device handle
3382 * - 82598/82599/X540: 128
3383 * - X550(non-SRIOV mode): 512
3384 * - X550(SRIOV mode): 64
3386 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3388 if (adapter->hw.mac.type < ixgbe_mac_X550)
3390 else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3397 * ixgbe_store_reta - Write the RETA table to HW
3398 * @adapter: device handle
3400 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3402 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3404 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3405 struct ixgbe_hw *hw = &adapter->hw;
3408 u8 *indir_tbl = adapter->rss_indir_tbl;
3410 /* Fill out the redirection table as follows:
3411 * - 82598: 8 bit wide entries containing pair of 4 bit RSS
3413 * - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3414 * - X550: 8 bit wide entries containing 6 bit RSS index
3416 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3417 indices_multi = 0x11;
3419 indices_multi = 0x1;
3421 /* Write redirection table to HW */
3422 for (i = 0; i < reta_entries; i++) {
3423 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3426 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3428 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3436 * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3437 * @adapter: device handle
3439 * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3441 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3443 u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3444 struct ixgbe_hw *hw = &adapter->hw;
3446 unsigned int pf_pool = adapter->num_vfs;
3448 /* Write redirection table to HW */
3449 for (i = 0; i < reta_entries; i++) {
3450 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3452 IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3459 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3461 struct ixgbe_hw *hw = &adapter->hw;
3463 u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3464 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3466 /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3467 * make full use of any rings they may have. We will use the
3468 * PSRTYPE register to control how many rings we use within the PF.
3470 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3473 /* Fill out hash function seeds */
3474 for (i = 0; i < 10; i++)
3475 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3477 /* Fill out redirection table */
3478 memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3480 for (i = 0, j = 0; i < reta_entries; i++, j++) {
3484 adapter->rss_indir_tbl[i] = j;
3487 ixgbe_store_reta(adapter);
3490 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3492 struct ixgbe_hw *hw = &adapter->hw;
3493 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3494 unsigned int pf_pool = adapter->num_vfs;
3497 /* Fill out hash function seeds */
3498 for (i = 0; i < 10; i++)
3499 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3500 adapter->rss_key[i]);
3502 /* Fill out the redirection table */
3503 for (i = 0, j = 0; i < 64; i++, j++) {
3507 adapter->rss_indir_tbl[i] = j;
3510 ixgbe_store_vfreta(adapter);
3513 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3515 struct ixgbe_hw *hw = &adapter->hw;
3516 u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3519 /* Disable indicating checksum in descriptor, enables RSS hash */
3520 rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3521 rxcsum |= IXGBE_RXCSUM_PCSD;
3522 IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3524 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3525 if (adapter->ring_feature[RING_F_RSS].mask)
3526 mrqc = IXGBE_MRQC_RSSEN;
3528 u8 tcs = netdev_get_num_tc(adapter->netdev);
3530 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3532 mrqc = IXGBE_MRQC_VMDQRT8TCEN; /* 8 TCs */
3534 mrqc = IXGBE_MRQC_VMDQRT4TCEN; /* 4 TCs */
3535 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3536 IXGBE_82599_VMDQ_4Q_MASK)
3537 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3539 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3542 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3544 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3546 mrqc = IXGBE_MRQC_RSSEN;
3550 /* Perform hash on these packet types */
3551 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3552 IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3553 IXGBE_MRQC_RSS_FIELD_IPV6 |
3554 IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3556 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3557 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3558 if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3559 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3561 netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3562 if ((hw->mac.type >= ixgbe_mac_X550) &&
3563 (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3564 unsigned int pf_pool = adapter->num_vfs;
3566 /* Enable VF RSS mode */
3567 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3568 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3570 /* Setup RSS through the VF registers */
3571 ixgbe_setup_vfreta(adapter);
3572 vfmrqc = IXGBE_MRQC_RSSEN;
3573 vfmrqc |= rss_field;
3574 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3576 ixgbe_setup_reta(adapter);
3578 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3583 * ixgbe_configure_rscctl - enable RSC for the indicated ring
3584 * @adapter: address of board private structure
3585 * @index: index of ring to set
3587 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3588 struct ixgbe_ring *ring)
3590 struct ixgbe_hw *hw = &adapter->hw;
3592 u8 reg_idx = ring->reg_idx;
3594 if (!ring_is_rsc_enabled(ring))
3597 rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3598 rscctrl |= IXGBE_RSCCTL_RSCEN;
3600 * we must limit the number of descriptors so that the
3601 * total size of max desc * buf_len is not greater
3604 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3605 IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3608 #define IXGBE_MAX_RX_DESC_POLL 10
3609 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3610 struct ixgbe_ring *ring)
3612 struct ixgbe_hw *hw = &adapter->hw;
3613 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3615 u8 reg_idx = ring->reg_idx;
3617 if (ixgbe_removed(hw->hw_addr))
3619 /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3620 if (hw->mac.type == ixgbe_mac_82598EB &&
3621 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3625 usleep_range(1000, 2000);
3626 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3627 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3630 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3631 "the polling period\n", reg_idx);
3635 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3636 struct ixgbe_ring *ring)
3638 struct ixgbe_hw *hw = &adapter->hw;
3639 int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3641 u8 reg_idx = ring->reg_idx;
3643 if (ixgbe_removed(hw->hw_addr))
3645 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3646 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3648 /* write value back with RXDCTL.ENABLE bit cleared */
3649 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3651 if (hw->mac.type == ixgbe_mac_82598EB &&
3652 !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3655 /* the hardware may take up to 100us to really disable the rx queue */
3658 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3659 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3662 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3663 "the polling period\n", reg_idx);
3667 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3668 struct ixgbe_ring *ring)
3670 struct ixgbe_hw *hw = &adapter->hw;
3671 u64 rdba = ring->dma;
3673 u8 reg_idx = ring->reg_idx;
3675 /* disable queue to avoid issues while updating state */
3676 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3677 ixgbe_disable_rx_queue(adapter, ring);
3679 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3680 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3681 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3682 ring->count * sizeof(union ixgbe_adv_rx_desc));
3683 /* Force flushing of IXGBE_RDLEN to prevent MDD */
3684 IXGBE_WRITE_FLUSH(hw);
3686 IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3687 IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3688 ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3690 ixgbe_configure_srrctl(adapter, ring);
3691 ixgbe_configure_rscctl(adapter, ring);
3693 if (hw->mac.type == ixgbe_mac_82598EB) {
3695 * enable cache line friendly hardware writes:
3696 * PTHRESH=32 descriptors (half the internal cache),
3697 * this also removes ugly rx_no_buffer_count increment
3698 * HTHRESH=4 descriptors (to minimize latency on fetch)
3699 * WTHRESH=8 burst writeback up to two cache lines
3701 rxdctl &= ~0x3FFFFF;
3705 /* enable receive descriptor ring */
3706 rxdctl |= IXGBE_RXDCTL_ENABLE;
3707 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3709 ixgbe_rx_desc_queue_enable(adapter, ring);
3710 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3713 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3715 struct ixgbe_hw *hw = &adapter->hw;
3716 int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3719 /* PSRTYPE must be initialized in non 82598 adapters */
3720 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3721 IXGBE_PSRTYPE_UDPHDR |
3722 IXGBE_PSRTYPE_IPV4HDR |
3723 IXGBE_PSRTYPE_L2HDR |
3724 IXGBE_PSRTYPE_IPV6HDR;
3726 if (hw->mac.type == ixgbe_mac_82598EB)
3730 psrtype |= 2u << 29;
3732 psrtype |= 1u << 29;
3734 for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3735 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3738 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3740 struct ixgbe_hw *hw = &adapter->hw;
3741 u32 reg_offset, vf_shift;
3742 u32 gcr_ext, vmdctl;
3745 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3748 vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3749 vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3750 vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3751 vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3752 vmdctl |= IXGBE_VT_CTL_REPLEN;
3753 IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3755 vf_shift = VMDQ_P(0) % 32;
3756 reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3758 /* Enable only the PF's pool for Tx/Rx */
3759 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
3760 IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3761 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
3762 IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3763 if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3764 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3766 /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3767 hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3769 /* clear VLAN promisc flag so VFTA will be updated if necessary */
3770 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
3773 * Set up VF register offsets for selected VT Mode,
3774 * i.e. 32 or 64 VFs for SR-IOV
3776 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3777 case IXGBE_82599_VMDQ_8Q_MASK:
3778 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3780 case IXGBE_82599_VMDQ_4Q_MASK:
3781 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3784 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3788 IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3790 for (i = 0; i < adapter->num_vfs; i++) {
3791 /* configure spoof checking */
3792 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
3793 adapter->vfinfo[i].spoofchk_enabled);
3795 /* Enable/Disable RSS query feature */
3796 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3797 adapter->vfinfo[i].rss_query_enabled);
3801 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3803 struct ixgbe_hw *hw = &adapter->hw;
3804 struct net_device *netdev = adapter->netdev;
3805 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3806 struct ixgbe_ring *rx_ring;
3811 /* adjust max frame to be able to do baby jumbo for FCoE */
3812 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3813 (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3814 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3816 #endif /* IXGBE_FCOE */
3818 /* adjust max frame to be at least the size of a standard frame */
3819 if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3820 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3822 mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3823 if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3824 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3825 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3827 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3830 hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3831 /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3832 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3833 IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3836 * Setup the HW Rx Head and Tail Descriptor Pointers and
3837 * the Base and Length of the Rx Descriptor Ring
3839 for (i = 0; i < adapter->num_rx_queues; i++) {
3840 rx_ring = adapter->rx_ring[i];
3841 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3842 set_ring_rsc_enabled(rx_ring);
3844 clear_ring_rsc_enabled(rx_ring);
3848 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3850 struct ixgbe_hw *hw = &adapter->hw;
3851 u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3853 switch (hw->mac.type) {
3854 case ixgbe_mac_82598EB:
3856 * For VMDq support of different descriptor types or
3857 * buffer sizes through the use of multiple SRRCTL
3858 * registers, RDRXCTL.MVMEN must be set to 1
3860 * also, the manual doesn't mention it clearly but DCA hints
3861 * will only use queue 0's tags unless this bit is set. Side
3862 * effects of setting this bit are only that SRRCTL must be
3863 * fully programmed [0..15]
3865 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3867 case ixgbe_mac_X550:
3868 case ixgbe_mac_X550EM_x:
3869 case ixgbe_mac_x550em_a:
3870 if (adapter->num_vfs)
3871 rdrxctl |= IXGBE_RDRXCTL_PSP;
3872 /* fall through for older HW */
3873 case ixgbe_mac_82599EB:
3874 case ixgbe_mac_X540:
3875 /* Disable RSC for ACK packets */
3876 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3877 (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3878 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3879 /* hardware requires some bits to be set by default */
3880 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3881 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3884 /* We should do nothing since we don't know this hardware */
3888 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3892 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3893 * @adapter: board private structure
3895 * Configure the Rx unit of the MAC after a reset.
3897 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3899 struct ixgbe_hw *hw = &adapter->hw;
3903 /* disable receives while setting up the descriptors */
3904 hw->mac.ops.disable_rx(hw);
3906 ixgbe_setup_psrtype(adapter);
3907 ixgbe_setup_rdrxctl(adapter);
3910 rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3911 rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3912 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3913 rfctl |= IXGBE_RFCTL_RSC_DIS;
3915 /* disable NFS filtering */
3916 rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
3917 IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3919 /* Program registers for the distribution of queues */
3920 ixgbe_setup_mrqc(adapter);
3922 /* set_rx_buffer_len must be called before ring initialization */
3923 ixgbe_set_rx_buffer_len(adapter);
3926 * Setup the HW Rx Head and Tail Descriptor Pointers and
3927 * the Base and Length of the Rx Descriptor Ring
3929 for (i = 0; i < adapter->num_rx_queues; i++)
3930 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3932 rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3933 /* disable drop enable for 82598 parts */
3934 if (hw->mac.type == ixgbe_mac_82598EB)
3935 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3937 /* enable all receives */
3938 rxctrl |= IXGBE_RXCTRL_RXEN;
3939 hw->mac.ops.enable_rx_dma(hw, rxctrl);
3942 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3943 __be16 proto, u16 vid)
3945 struct ixgbe_adapter *adapter = netdev_priv(netdev);
3946 struct ixgbe_hw *hw = &adapter->hw;
3948 /* add VID to filter table */
3949 if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
3950 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
3952 set_bit(vid, adapter->active_vlans);
3957 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
3962 /* short cut the special case */
3966 /* Search for the vlan id in the VLVF entries */
3967 for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
3968 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
3969 if ((vlvf & VLAN_VID_MASK) == vlan)
3976 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
3978 struct ixgbe_hw *hw = &adapter->hw;
3982 idx = ixgbe_find_vlvf_entry(hw, vid);
3986 /* See if any other pools are set for this VLAN filter
3987 * entry other than the PF.
3989 word = idx * 2 + (VMDQ_P(0) / 32);
3990 bits = ~BIT(VMDQ_P(0) % 32);
3991 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
3993 /* Disable the filter so this falls into the default pool. */
3994 if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
3995 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
3996 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
3997 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4001 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4002 __be16 proto, u16 vid)
4004 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4005 struct ixgbe_hw *hw = &adapter->hw;
4007 /* remove VID from filter table */
4008 if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4009 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4011 clear_bit(vid, adapter->active_vlans);
4017 * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4018 * @adapter: driver data
4020 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4022 struct ixgbe_hw *hw = &adapter->hw;
4026 switch (hw->mac.type) {
4027 case ixgbe_mac_82598EB:
4028 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4029 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4030 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4032 case ixgbe_mac_82599EB:
4033 case ixgbe_mac_X540:
4034 case ixgbe_mac_X550:
4035 case ixgbe_mac_X550EM_x:
4036 case ixgbe_mac_x550em_a:
4037 for (i = 0; i < adapter->num_rx_queues; i++) {
4038 struct ixgbe_ring *ring = adapter->rx_ring[i];
4040 if (ring->l2_accel_priv)
4043 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4044 vlnctrl &= ~IXGBE_RXDCTL_VME;
4045 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4054 * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4055 * @adapter: driver data
4057 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4059 struct ixgbe_hw *hw = &adapter->hw;
4063 switch (hw->mac.type) {
4064 case ixgbe_mac_82598EB:
4065 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4066 vlnctrl |= IXGBE_VLNCTRL_VME;
4067 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4069 case ixgbe_mac_82599EB:
4070 case ixgbe_mac_X540:
4071 case ixgbe_mac_X550:
4072 case ixgbe_mac_X550EM_x:
4073 case ixgbe_mac_x550em_a:
4074 for (i = 0; i < adapter->num_rx_queues; i++) {
4075 struct ixgbe_ring *ring = adapter->rx_ring[i];
4077 if (ring->l2_accel_priv)
4080 vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4081 vlnctrl |= IXGBE_RXDCTL_VME;
4082 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4090 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4092 struct ixgbe_hw *hw = &adapter->hw;
4095 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4097 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4098 /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4099 vlnctrl |= IXGBE_VLNCTRL_VFE;
4100 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4102 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4103 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4107 /* Nothing to do for 82598 */
4108 if (hw->mac.type == ixgbe_mac_82598EB)
4111 /* We are already in VLAN promisc, nothing to do */
4112 if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4115 /* Set flag so we don't redo unnecessary work */
4116 adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4118 /* Add PF to all active pools */
4119 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4120 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4121 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4123 vlvfb |= BIT(VMDQ_P(0) % 32);
4124 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4127 /* Set all bits in the VLAN filter table array */
4128 for (i = hw->mac.vft_size; i--;)
4129 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4132 #define VFTA_BLOCK_SIZE 8
4133 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4135 struct ixgbe_hw *hw = &adapter->hw;
4136 u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4137 u32 vid_start = vfta_offset * 32;
4138 u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4139 u32 i, vid, word, bits;
4141 for (i = IXGBE_VLVF_ENTRIES; --i;) {
4142 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4144 /* pull VLAN ID from VLVF */
4145 vid = vlvf & VLAN_VID_MASK;
4147 /* only concern outselves with a certain range */
4148 if (vid < vid_start || vid >= vid_end)
4152 /* record VLAN ID in VFTA */
4153 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4155 /* if PF is part of this then continue */
4156 if (test_bit(vid, adapter->active_vlans))
4160 /* remove PF from the pool */
4161 word = i * 2 + VMDQ_P(0) / 32;
4162 bits = ~BIT(VMDQ_P(0) % 32);
4163 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4164 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4167 /* extract values from active_vlans and write back to VFTA */
4168 for (i = VFTA_BLOCK_SIZE; i--;) {
4169 vid = (vfta_offset + i) * 32;
4170 word = vid / BITS_PER_LONG;
4171 bits = vid % BITS_PER_LONG;
4173 vfta[i] |= adapter->active_vlans[word] >> bits;
4175 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4179 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4181 struct ixgbe_hw *hw = &adapter->hw;
4184 /* Set VLAN filtering to enabled */
4185 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4186 vlnctrl |= IXGBE_VLNCTRL_VFE;
4187 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4189 if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4190 hw->mac.type == ixgbe_mac_82598EB)
4193 /* We are not in VLAN promisc, nothing to do */
4194 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4197 /* Set flag so we don't redo unnecessary work */
4198 adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4200 for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4201 ixgbe_scrub_vfta(adapter, i);
4204 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4208 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4210 for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4211 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4215 * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4216 * @netdev: network interface device structure
4218 * Writes multicast address list to the MTA hash table.
4219 * Returns: -ENOMEM on failure
4220 * 0 on no addresses written
4221 * X on writing X addresses to MTA
4223 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4225 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4226 struct ixgbe_hw *hw = &adapter->hw;
4228 if (!netif_running(netdev))
4231 if (hw->mac.ops.update_mc_addr_list)
4232 hw->mac.ops.update_mc_addr_list(hw, netdev);
4236 #ifdef CONFIG_PCI_IOV
4237 ixgbe_restore_vf_multicasts(adapter);
4240 return netdev_mc_count(netdev);
4243 #ifdef CONFIG_PCI_IOV
4244 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4246 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4247 struct ixgbe_hw *hw = &adapter->hw;
4250 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4251 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4253 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4254 hw->mac.ops.set_rar(hw, i,
4259 hw->mac.ops.clear_rar(hw, i);
4264 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4266 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4267 struct ixgbe_hw *hw = &adapter->hw;
4270 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4271 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4274 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4276 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4277 hw->mac.ops.set_rar(hw, i,
4282 hw->mac.ops.clear_rar(hw, i);
4286 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4288 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4289 struct ixgbe_hw *hw = &adapter->hw;
4292 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4293 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4294 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4297 ixgbe_sync_mac_table(adapter);
4300 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4302 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4303 struct ixgbe_hw *hw = &adapter->hw;
4306 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4307 /* do not count default RAR as available */
4308 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4311 /* only count unused and addresses that belong to us */
4312 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4313 if (mac_table->pool != pool)
4323 /* this function destroys the first RAR entry */
4324 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4326 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4327 struct ixgbe_hw *hw = &adapter->hw;
4329 memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4330 mac_table->pool = VMDQ_P(0);
4332 mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4334 hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4338 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4339 const u8 *addr, u16 pool)
4341 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4342 struct ixgbe_hw *hw = &adapter->hw;
4345 if (is_zero_ether_addr(addr))
4348 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4349 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4352 ether_addr_copy(mac_table->addr, addr);
4353 mac_table->pool = pool;
4355 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4356 IXGBE_MAC_STATE_IN_USE;
4358 ixgbe_sync_mac_table(adapter);
4366 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4367 const u8 *addr, u16 pool)
4369 struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4370 struct ixgbe_hw *hw = &adapter->hw;
4373 if (is_zero_ether_addr(addr))
4376 /* search table for addr, if found clear IN_USE flag and sync */
4377 for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4378 /* we can only delete an entry if it is in use */
4379 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4381 /* we only care about entries that belong to the given pool */
4382 if (mac_table->pool != pool)
4384 /* we only care about a specific MAC address */
4385 if (!ether_addr_equal(addr, mac_table->addr))
4388 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4389 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4391 ixgbe_sync_mac_table(adapter);
4399 * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4400 * @netdev: network interface device structure
4402 * Writes unicast address list to the RAR table.
4403 * Returns: -ENOMEM on failure/insufficient address space
4404 * 0 on no addresses written
4405 * X on writing X addresses to the RAR table
4407 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4409 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4412 /* return ENOMEM indicating insufficient memory for addresses */
4413 if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter, vfn))
4416 if (!netdev_uc_empty(netdev)) {
4417 struct netdev_hw_addr *ha;
4418 netdev_for_each_uc_addr(ha, netdev) {
4419 ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4420 ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4427 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4429 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4432 ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4434 return min_t(int, ret, 0);
4437 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4439 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4441 ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4447 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4448 * @netdev: network interface device structure
4450 * The set_rx_method entry point is called whenever the unicast/multicast
4451 * address list or the network interface flags are updated. This routine is
4452 * responsible for configuring the hardware for proper unicast, multicast and
4455 void ixgbe_set_rx_mode(struct net_device *netdev)
4457 struct ixgbe_adapter *adapter = netdev_priv(netdev);
4458 struct ixgbe_hw *hw = &adapter->hw;
4459 u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4460 netdev_features_t features = netdev->features;
4463 /* Check for Promiscuous and All Multicast modes */
4464 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4466 /* set all bits that we expect to always be set */
4467 fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4468 fctrl |= IXGBE_FCTRL_BAM;
4469 fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4470 fctrl |= IXGBE_FCTRL_PMCF;
4472 /* clear the bits we are changing the status of */
4473 fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4474 if (netdev->flags & IFF_PROMISC) {
4475 hw->addr_ctrl.user_set_promisc = true;
4476 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4477 vmolr |= IXGBE_VMOLR_MPE;
4478 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4480 if (netdev->flags & IFF_ALLMULTI) {
4481 fctrl |= IXGBE_FCTRL_MPE;
4482 vmolr |= IXGBE_VMOLR_MPE;
4484 hw->addr_ctrl.user_set_promisc = false;
4488 * Write addresses to available RAR registers, if there is not
4489 * sufficient space to store all the addresses then enable
4490 * unicast promiscuous mode
4492 if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4493 fctrl |= IXGBE_FCTRL_UPE;
4494 vmolr |= IXGBE_VMOLR_ROPE;
4497 /* Write addresses to the MTA, if the attempt fails
4498 * then we should just turn on promiscuous mode so
4499 * that we can at least receive multicast traffic
4501 count = ixgbe_write_mc_addr_list(netdev);
4503 fctrl |= IXGBE_FCTRL_MPE;
4504 vmolr |= IXGBE_VMOLR_MPE;
4506 vmolr |= IXGBE_VMOLR_ROMPE;
4509 if (hw->mac.type != ixgbe_mac_82598EB) {
4510 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4511 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4513 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4516 /* This is useful for sniffing bad packets. */
4517 if (features & NETIF_F_RXALL) {
4518 /* UPE and MPE will be handled by normal PROMISC logic
4519 * in e1000e_set_rx_mode */
4520 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4521 IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4522 IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4524 fctrl &= ~(IXGBE_FCTRL_DPF);
4525 /* NOTE: VLAN filtering is disabled by setting PROMISC */
4528 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4530 if (features & NETIF_F_HW_VLAN_CTAG_RX)
4531 ixgbe_vlan_strip_enable(adapter);
4533 ixgbe_vlan_strip_disable(adapter);
4535 if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4536 ixgbe_vlan_promisc_disable(adapter);
4538 ixgbe_vlan_promisc_enable(adapter);
4541 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4545 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4546 napi_enable(&adapter->q_vector[q_idx]->napi);
4549 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4553 for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4554 napi_disable(&adapter->q_vector[q_idx]->napi);
4557 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
4559 struct ixgbe_hw *hw = &adapter->hw;
4562 if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
4563 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
4566 vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) && ~mask;
4567 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
4569 if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
4570 adapter->vxlan_port = 0;
4572 if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
4573 adapter->geneve_port = 0;
4576 #ifdef CONFIG_IXGBE_DCB
4578 * ixgbe_configure_dcb - Configure DCB hardware
4579 * @adapter: ixgbe adapter struct
4581 * This is called by the driver on open to configure the DCB hardware.
4582 * This is also called by the gennetlink interface when reconfiguring
4585 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4587 struct ixgbe_hw *hw = &adapter->hw;
4588 int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4590 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4591 if (hw->mac.type == ixgbe_mac_82598EB)
4592 netif_set_gso_max_size(adapter->netdev, 65536);
4596 if (hw->mac.type == ixgbe_mac_82598EB)
4597 netif_set_gso_max_size(adapter->netdev, 32768);
4600 if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4601 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4604 /* reconfigure the hardware */
4605 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4606 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4608 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4610 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4611 } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4612 ixgbe_dcb_hw_ets(&adapter->hw,
4613 adapter->ixgbe_ieee_ets,
4615 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4616 adapter->ixgbe_ieee_pfc->pfc_en,
4617 adapter->ixgbe_ieee_ets->prio_tc);
4620 /* Enable RSS Hash per TC */
4621 if (hw->mac.type != ixgbe_mac_82598EB) {
4623 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4630 /* write msb to all 8 TCs in one write */
4631 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4636 /* Additional bittime to account for IXGBE framing */
4637 #define IXGBE_ETH_FRAMING 20
4640 * ixgbe_hpbthresh - calculate high water mark for flow control
4642 * @adapter: board private structure to calculate for
4643 * @pb: packet buffer to calculate
4645 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4647 struct ixgbe_hw *hw = &adapter->hw;
4648 struct net_device *dev = adapter->netdev;
4649 int link, tc, kb, marker;
4652 /* Calculate max LAN frame size */
4653 tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4656 /* FCoE traffic class uses FCOE jumbo frames */
4657 if ((dev->features & NETIF_F_FCOE_MTU) &&
4658 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4659 (pb == ixgbe_fcoe_get_tc(adapter)))
4660 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4663 /* Calculate delay value for device */
4664 switch (hw->mac.type) {
4665 case ixgbe_mac_X540:
4666 case ixgbe_mac_X550:
4667 case ixgbe_mac_X550EM_x:
4668 case ixgbe_mac_x550em_a:
4669 dv_id = IXGBE_DV_X540(link, tc);
4672 dv_id = IXGBE_DV(link, tc);
4676 /* Loopback switch introduces additional latency */
4677 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4678 dv_id += IXGBE_B2BT(tc);
4680 /* Delay value is calculated in bit times convert to KB */
4681 kb = IXGBE_BT2KB(dv_id);
4682 rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4684 marker = rx_pba - kb;
4686 /* It is possible that the packet buffer is not large enough
4687 * to provide required headroom. In this case throw an error
4688 * to user and a do the best we can.
4691 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4692 "headroom to support flow control."
4693 "Decrease MTU or number of traffic classes\n", pb);
4701 * ixgbe_lpbthresh - calculate low water mark for for flow control
4703 * @adapter: board private structure to calculate for
4704 * @pb: packet buffer to calculate
4706 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4708 struct ixgbe_hw *hw = &adapter->hw;
4709 struct net_device *dev = adapter->netdev;
4713 /* Calculate max LAN frame size */
4714 tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4717 /* FCoE traffic class uses FCOE jumbo frames */
4718 if ((dev->features & NETIF_F_FCOE_MTU) &&
4719 (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4720 (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4721 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4724 /* Calculate delay value for device */
4725 switch (hw->mac.type) {
4726 case ixgbe_mac_X540:
4727 case ixgbe_mac_X550:
4728 case ixgbe_mac_X550EM_x:
4729 case ixgbe_mac_x550em_a:
4730 dv_id = IXGBE_LOW_DV_X540(tc);
4733 dv_id = IXGBE_LOW_DV(tc);
4737 /* Delay value is calculated in bit times convert to KB */
4738 return IXGBE_BT2KB(dv_id);
4742 * ixgbe_pbthresh_setup - calculate and setup high low water marks
4744 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4746 struct ixgbe_hw *hw = &adapter->hw;
4747 int num_tc = netdev_get_num_tc(adapter->netdev);
4753 for (i = 0; i < num_tc; i++) {
4754 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4755 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4757 /* Low water marks must not be larger than high water marks */
4758 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4759 hw->fc.low_water[i] = 0;
4762 for (; i < MAX_TRAFFIC_CLASS; i++)
4763 hw->fc.high_water[i] = 0;
4766 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4768 struct ixgbe_hw *hw = &adapter->hw;
4770 u8 tc = netdev_get_num_tc(adapter->netdev);
4772 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4773 adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4774 hdrm = 32 << adapter->fdir_pballoc;
4778 hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4779 ixgbe_pbthresh_setup(adapter);
4782 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4784 struct ixgbe_hw *hw = &adapter->hw;
4785 struct hlist_node *node2;
4786 struct ixgbe_fdir_filter *filter;
4788 spin_lock(&adapter->fdir_perfect_lock);
4790 if (!hlist_empty(&adapter->fdir_filter_list))
4791 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4793 hlist_for_each_entry_safe(filter, node2,
4794 &adapter->fdir_filter_list, fdir_node) {
4795 ixgbe_fdir_write_perfect_filter_82599(hw,
4798 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4799 IXGBE_FDIR_DROP_QUEUE :
4800 adapter->rx_ring[filter->action]->reg_idx);
4803 spin_unlock(&adapter->fdir_perfect_lock);
4806 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4807 struct ixgbe_adapter *adapter)
4809 struct ixgbe_hw *hw = &adapter->hw;
4812 /* No unicast promiscuous support for VMDQ devices. */
4813 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4814 vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4816 /* clear the affected bit */
4817 vmolr &= ~IXGBE_VMOLR_MPE;
4819 if (dev->flags & IFF_ALLMULTI) {
4820 vmolr |= IXGBE_VMOLR_MPE;
4822 vmolr |= IXGBE_VMOLR_ROMPE;
4823 hw->mac.ops.update_mc_addr_list(hw, dev);
4825 ixgbe_write_uc_addr_list(adapter->netdev, pool);
4826 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4829 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4831 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4832 int rss_i = adapter->num_rx_queues_per_pool;
4833 struct ixgbe_hw *hw = &adapter->hw;
4834 u16 pool = vadapter->pool;
4835 u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4836 IXGBE_PSRTYPE_UDPHDR |
4837 IXGBE_PSRTYPE_IPV4HDR |
4838 IXGBE_PSRTYPE_L2HDR |
4839 IXGBE_PSRTYPE_IPV6HDR;
4841 if (hw->mac.type == ixgbe_mac_82598EB)
4845 psrtype |= 2u << 29;
4847 psrtype |= 1u << 29;
4849 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4853 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4854 * @rx_ring: ring to free buffers from
4856 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4858 struct device *dev = rx_ring->dev;
4862 /* ring already cleared, nothing to do */
4863 if (!rx_ring->rx_buffer_info)
4866 /* Free all the Rx ring sk_buffs */
4867 for (i = 0; i < rx_ring->count; i++) {
4868 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4870 if (rx_buffer->skb) {
4871 struct sk_buff *skb = rx_buffer->skb;
4872 if (IXGBE_CB(skb)->page_released)
4875 ixgbe_rx_bufsz(rx_ring),
4878 rx_buffer->skb = NULL;
4881 if (!rx_buffer->page)
4884 dma_unmap_page(dev, rx_buffer->dma,
4885 ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4886 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4888 rx_buffer->page = NULL;
4891 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4892 memset(rx_ring->rx_buffer_info, 0, size);
4894 /* Zero out the descriptor ring */
4895 memset(rx_ring->desc, 0, rx_ring->size);
4897 rx_ring->next_to_alloc = 0;
4898 rx_ring->next_to_clean = 0;
4899 rx_ring->next_to_use = 0;
4902 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4903 struct ixgbe_ring *rx_ring)
4905 struct ixgbe_adapter *adapter = vadapter->real_adapter;
4906 int index = rx_ring->queue_index + vadapter->rx_base_queue;
4908 /* shutdown specific queue receive and wait for dma to settle */
4909 ixgbe_disable_rx_queue(adapter, rx_ring);
4910 usleep_range(10000, 20000);
4911 ixgbe_irq_disable_queues(adapter, BIT_ULL(index));
4912 ixgbe_clean_rx_ring(rx_ring);
4913 rx_ring->l2_accel_priv = NULL;
4916 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4917 struct ixgbe_fwd_adapter *accel)
4919 struct ixgbe_adapter *adapter = accel->real_adapter;
4920 unsigned int rxbase = accel->rx_base_queue;
4921 unsigned int txbase = accel->tx_base_queue;
4924 netif_tx_stop_all_queues(vdev);
4926 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4927 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4928 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4931 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4932 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4933 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4940 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4941 struct ixgbe_fwd_adapter *accel)
4943 struct ixgbe_adapter *adapter = accel->real_adapter;
4944 unsigned int rxbase, txbase, queues;
4945 int i, baseq, err = 0;
4947 if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4950 baseq = accel->pool * adapter->num_rx_queues_per_pool;
4951 netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4952 accel->pool, adapter->num_rx_pools,
4953 baseq, baseq + adapter->num_rx_queues_per_pool,
4954 adapter->fwd_bitmask);
4956 accel->netdev = vdev;
4957 accel->rx_base_queue = rxbase = baseq;
4958 accel->tx_base_queue = txbase = baseq;
4960 for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4961 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4963 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4964 adapter->rx_ring[rxbase + i]->netdev = vdev;
4965 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4966 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4969 for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4970 adapter->tx_ring[txbase + i]->netdev = vdev;
4971 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4974 queues = min_t(unsigned int,
4975 adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4976 err = netif_set_real_num_tx_queues(vdev, queues);
4980 err = netif_set_real_num_rx_queues(vdev, queues);
4984 if (is_valid_ether_addr(vdev->dev_addr))
4985 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4987 ixgbe_fwd_psrtype(accel);
4988 ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4991 ixgbe_fwd_ring_down(vdev, accel);
4995 static int ixgbe_upper_dev_walk(struct net_device *upper, void *data)
4997 if (netif_is_macvlan(upper)) {
4998 struct macvlan_dev *dfwd = netdev_priv(upper);
4999 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
5002 ixgbe_fwd_ring_up(upper, vadapter);
5008 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5010 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5011 ixgbe_upper_dev_walk, NULL);
5014 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5016 struct ixgbe_hw *hw = &adapter->hw;
5018 ixgbe_configure_pb(adapter);
5019 #ifdef CONFIG_IXGBE_DCB
5020 ixgbe_configure_dcb(adapter);
5023 * We must restore virtualization before VLANs or else
5024 * the VLVF registers will not be populated
5026 ixgbe_configure_virtualization(adapter);
5028 ixgbe_set_rx_mode(adapter->netdev);
5029 ixgbe_restore_vlan(adapter);
5031 switch (hw->mac.type) {
5032 case ixgbe_mac_82599EB:
5033 case ixgbe_mac_X540:
5034 hw->mac.ops.disable_rx_buff(hw);
5040 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5041 ixgbe_init_fdir_signature_82599(&adapter->hw,
5042 adapter->fdir_pballoc);
5043 } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5044 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5045 adapter->fdir_pballoc);
5046 ixgbe_fdir_filter_restore(adapter);
5049 switch (hw->mac.type) {
5050 case ixgbe_mac_82599EB:
5051 case ixgbe_mac_X540:
5052 hw->mac.ops.enable_rx_buff(hw);
5058 #ifdef CONFIG_IXGBE_DCA
5060 if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5061 ixgbe_setup_dca(adapter);
5062 #endif /* CONFIG_IXGBE_DCA */
5065 /* configure FCoE L2 filters, redirection table, and Rx control */
5066 ixgbe_configure_fcoe(adapter);
5068 #endif /* IXGBE_FCOE */
5069 ixgbe_configure_tx(adapter);
5070 ixgbe_configure_rx(adapter);
5071 ixgbe_configure_dfwd(adapter);
5075 * ixgbe_sfp_link_config - set up SFP+ link
5076 * @adapter: pointer to private adapter struct
5078 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5081 * We are assuming the worst case scenario here, and that
5082 * is that an SFP was inserted/removed after the reset
5083 * but before SFP detection was enabled. As such the best
5084 * solution is to just start searching as soon as we start
5086 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5087 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5089 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5090 adapter->sfp_poll_time = 0;
5094 * ixgbe_non_sfp_link_config - set up non-SFP+ link
5095 * @hw: pointer to private hardware struct
5097 * Returns 0 on success, negative on failure
5099 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5102 bool autoneg, link_up = false;
5103 int ret = IXGBE_ERR_LINK_SETUP;
5105 if (hw->mac.ops.check_link)
5106 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5111 speed = hw->phy.autoneg_advertised;
5112 if ((!speed) && (hw->mac.ops.get_link_capabilities))
5113 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5118 if (hw->mac.ops.setup_link)
5119 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5124 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5126 struct ixgbe_hw *hw = &adapter->hw;
5129 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5130 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5132 gpie |= IXGBE_GPIE_EIAME;
5134 * use EIAM to auto-mask when MSI-X interrupt is asserted
5135 * this saves a register write for every interrupt
5137 switch (hw->mac.type) {
5138 case ixgbe_mac_82598EB:
5139 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5141 case ixgbe_mac_82599EB:
5142 case ixgbe_mac_X540:
5143 case ixgbe_mac_X550:
5144 case ixgbe_mac_X550EM_x:
5145 case ixgbe_mac_x550em_a:
5147 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5148 IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5152 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5153 * specifically only auto mask tx and rx interrupts */
5154 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5157 /* XXX: to interrupt immediately for EICS writes, enable this */
5158 /* gpie |= IXGBE_GPIE_EIMEN; */
5160 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5161 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5163 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5164 case IXGBE_82599_VMDQ_8Q_MASK:
5165 gpie |= IXGBE_GPIE_VTMODE_16;
5167 case IXGBE_82599_VMDQ_4Q_MASK:
5168 gpie |= IXGBE_GPIE_VTMODE_32;
5171 gpie |= IXGBE_GPIE_VTMODE_64;
5176 /* Enable Thermal over heat sensor interrupt */
5177 if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5178 switch (adapter->hw.mac.type) {
5179 case ixgbe_mac_82599EB:
5180 gpie |= IXGBE_SDP0_GPIEN_8259X;
5187 /* Enable fan failure interrupt */
5188 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5189 gpie |= IXGBE_SDP1_GPIEN(hw);
5191 switch (hw->mac.type) {
5192 case ixgbe_mac_82599EB:
5193 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5195 case ixgbe_mac_X550EM_x:
5196 case ixgbe_mac_x550em_a:
5197 gpie |= IXGBE_SDP0_GPIEN_X540;
5203 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5206 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5208 struct ixgbe_hw *hw = &adapter->hw;
5212 ixgbe_get_hw_control(adapter);
5213 ixgbe_setup_gpie(adapter);
5215 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5216 ixgbe_configure_msix(adapter);
5218 ixgbe_configure_msi_and_legacy(adapter);
5220 /* enable the optics for 82599 SFP+ fiber */
5221 if (hw->mac.ops.enable_tx_laser)
5222 hw->mac.ops.enable_tx_laser(hw);
5224 if (hw->phy.ops.set_phy_power)
5225 hw->phy.ops.set_phy_power(hw, true);
5227 smp_mb__before_atomic();
5228 clear_bit(__IXGBE_DOWN, &adapter->state);
5229 ixgbe_napi_enable_all(adapter);
5231 if (ixgbe_is_sfp(hw)) {
5232 ixgbe_sfp_link_config(adapter);
5234 err = ixgbe_non_sfp_link_config(hw);
5236 e_err(probe, "link_config FAILED %d\n", err);
5239 /* clear any pending interrupts, may auto mask */
5240 IXGBE_READ_REG(hw, IXGBE_EICR);
5241 ixgbe_irq_enable(adapter, true, true);
5244 * If this adapter has a fan, check to see if we had a failure
5245 * before we enabled the interrupt.
5247 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5248 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5249 if (esdp & IXGBE_ESDP_SDP1)
5250 e_crit(drv, "Fan has stopped, replace the adapter\n");
5253 /* bring the link up in the watchdog, this could race with our first
5254 * link up interrupt but shouldn't be a problem */
5255 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5256 adapter->link_check_timeout = jiffies;
5257 mod_timer(&adapter->service_timer, jiffies);
5259 /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5260 ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5261 ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5262 IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5265 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5267 WARN_ON(in_interrupt());
5268 /* put off any impending NetWatchDogTimeout */
5269 netif_trans_update(adapter->netdev);
5271 while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5272 usleep_range(1000, 2000);
5273 if (adapter->hw.phy.type == ixgbe_phy_fw)
5274 ixgbe_watchdog_link_is_down(adapter);
5275 ixgbe_down(adapter);
5277 * If SR-IOV enabled then wait a bit before bringing the adapter
5278 * back up to give the VFs time to respond to the reset. The
5279 * two second wait is based upon the watchdog timer cycle in
5282 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5285 clear_bit(__IXGBE_RESETTING, &adapter->state);
5288 void ixgbe_up(struct ixgbe_adapter *adapter)
5290 /* hardware has been reset, we need to reload some things */
5291 ixgbe_configure(adapter);
5293 ixgbe_up_complete(adapter);
5296 void ixgbe_reset(struct ixgbe_adapter *adapter)
5298 struct ixgbe_hw *hw = &adapter->hw;
5299 struct net_device *netdev = adapter->netdev;
5302 if (ixgbe_removed(hw->hw_addr))
5304 /* lock SFP init bit to prevent race conditions with the watchdog */
5305 while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5306 usleep_range(1000, 2000);
5308 /* clear all SFP and link config related flags while holding SFP_INIT */
5309 adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5310 IXGBE_FLAG2_SFP_NEEDS_RESET);
5311 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5313 err = hw->mac.ops.init_hw(hw);
5316 case IXGBE_ERR_SFP_NOT_PRESENT:
5317 case IXGBE_ERR_SFP_NOT_SUPPORTED:
5319 case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5320 e_dev_err("master disable timed out\n");
5322 case IXGBE_ERR_EEPROM_VERSION:
5323 /* We are running on a pre-production device, log a warning */
5324 e_dev_warn("This device is a pre-production adapter/LOM. "
5325 "Please be aware there may be issues associated with "
5326 "your hardware. If you are experiencing problems "
5327 "please contact your Intel or hardware "
5328 "representative who provided you with this "
5332 e_dev_err("Hardware Error: %d\n", err);
5335 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5337 /* flush entries out of MAC table */
5338 ixgbe_flush_sw_mac_table(adapter);
5339 __dev_uc_unsync(netdev, NULL);
5341 /* do not flush user set addresses */
5342 ixgbe_mac_set_default_filter(adapter);
5344 /* update SAN MAC vmdq pool selection */
5345 if (hw->mac.san_mac_rar_index)
5346 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5348 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5349 ixgbe_ptp_reset(adapter);
5351 if (hw->phy.ops.set_phy_power) {
5352 if (!netif_running(adapter->netdev) && !adapter->wol)
5353 hw->phy.ops.set_phy_power(hw, false);
5355 hw->phy.ops.set_phy_power(hw, true);
5360 * ixgbe_clean_tx_ring - Free Tx Buffers
5361 * @tx_ring: ring to be cleaned
5363 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5365 struct ixgbe_tx_buffer *tx_buffer_info;
5369 /* ring already cleared, nothing to do */
5370 if (!tx_ring->tx_buffer_info)
5373 /* Free all the Tx ring sk_buffs */
5374 for (i = 0; i < tx_ring->count; i++) {
5375 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5376 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5379 netdev_tx_reset_queue(txring_txq(tx_ring));
5381 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5382 memset(tx_ring->tx_buffer_info, 0, size);
5384 /* Zero out the descriptor ring */
5385 memset(tx_ring->desc, 0, tx_ring->size);
5387 tx_ring->next_to_use = 0;
5388 tx_ring->next_to_clean = 0;
5392 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5393 * @adapter: board private structure
5395 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5399 for (i = 0; i < adapter->num_rx_queues; i++)
5400 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5404 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5405 * @adapter: board private structure
5407 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5411 for (i = 0; i < adapter->num_tx_queues; i++)
5412 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5415 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5417 struct hlist_node *node2;
5418 struct ixgbe_fdir_filter *filter;
5420 spin_lock(&adapter->fdir_perfect_lock);
5422 hlist_for_each_entry_safe(filter, node2,
5423 &adapter->fdir_filter_list, fdir_node) {
5424 hlist_del(&filter->fdir_node);
5427 adapter->fdir_filter_count = 0;
5429 spin_unlock(&adapter->fdir_perfect_lock);
5432 static int ixgbe_disable_macvlan(struct net_device *upper, void *data)
5434 if (netif_is_macvlan(upper)) {
5435 struct macvlan_dev *vlan = netdev_priv(upper);
5437 if (vlan->fwd_priv) {
5438 netif_tx_stop_all_queues(upper);
5439 netif_carrier_off(upper);
5440 netif_tx_disable(upper);
5447 void ixgbe_down(struct ixgbe_adapter *adapter)
5449 struct net_device *netdev = adapter->netdev;
5450 struct ixgbe_hw *hw = &adapter->hw;
5453 /* signal that we are down to the interrupt handler */
5454 if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5455 return; /* do nothing if already down */
5457 /* disable receives */
5458 hw->mac.ops.disable_rx(hw);
5460 /* disable all enabled rx queues */
5461 for (i = 0; i < adapter->num_rx_queues; i++)
5462 /* this call also flushes the previous write */
5463 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5465 usleep_range(10000, 20000);
5467 netif_tx_stop_all_queues(netdev);
5469 /* call carrier off first to avoid false dev_watchdog timeouts */
5470 netif_carrier_off(netdev);
5471 netif_tx_disable(netdev);
5473 /* disable any upper devices */
5474 netdev_walk_all_upper_dev_rcu(adapter->netdev,
5475 ixgbe_disable_macvlan, NULL);
5477 ixgbe_irq_disable(adapter);
5479 ixgbe_napi_disable_all(adapter);
5481 clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
5482 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
5483 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5485 del_timer_sync(&adapter->service_timer);
5487 if (adapter->num_vfs) {
5488 /* Clear EITR Select mapping */
5489 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5491 /* Mark all the VFs as inactive */
5492 for (i = 0 ; i < adapter->num_vfs; i++)
5493 adapter->vfinfo[i].clear_to_send = false;
5495 /* ping all the active vfs to let them know we are going down */
5496 ixgbe_ping_all_vfs(adapter);
5498 /* Disable all VFTE/VFRE TX/RX */
5499 ixgbe_disable_tx_rx(adapter);
5502 /* disable transmits in the hardware now that interrupts are off */
5503 for (i = 0; i < adapter->num_tx_queues; i++) {
5504 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5505 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5508 /* Disable the Tx DMA engine on 82599 and later MAC */
5509 switch (hw->mac.type) {
5510 case ixgbe_mac_82599EB:
5511 case ixgbe_mac_X540:
5512 case ixgbe_mac_X550:
5513 case ixgbe_mac_X550EM_x:
5514 case ixgbe_mac_x550em_a:
5515 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5516 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5517 ~IXGBE_DMATXCTL_TE));
5523 if (!pci_channel_offline(adapter->pdev))
5524 ixgbe_reset(adapter);
5526 /* power down the optics for 82599 SFP+ fiber */
5527 if (hw->mac.ops.disable_tx_laser)
5528 hw->mac.ops.disable_tx_laser(hw);
5530 ixgbe_clean_all_tx_rings(adapter);
5531 ixgbe_clean_all_rx_rings(adapter);
5535 * ixgbe_eee_capable - helper function to determine EEE support on X550
5536 * @adapter: board private structure
5538 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
5540 struct ixgbe_hw *hw = &adapter->hw;
5542 switch (hw->device_id) {
5543 case IXGBE_DEV_ID_X550EM_A_1G_T:
5544 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
5545 if (!hw->phy.eee_speeds_supported)
5547 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
5548 if (!hw->phy.eee_speeds_advertised)
5550 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
5553 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
5554 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
5560 * ixgbe_tx_timeout - Respond to a Tx Hang
5561 * @netdev: network interface device structure
5563 static void ixgbe_tx_timeout(struct net_device *netdev)
5565 struct ixgbe_adapter *adapter = netdev_priv(netdev);
5567 /* Do the reset outside of interrupt context */
5568 ixgbe_tx_timeout_reset(adapter);
5571 #ifdef CONFIG_IXGBE_DCB
5572 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
5574 struct ixgbe_hw *hw = &adapter->hw;
5575 struct tc_configuration *tc;
5578 switch (hw->mac.type) {
5579 case ixgbe_mac_82598EB:
5580 case ixgbe_mac_82599EB:
5581 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5582 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5584 case ixgbe_mac_X540:
5585 case ixgbe_mac_X550:
5586 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5587 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5589 case ixgbe_mac_X550EM_x:
5590 case ixgbe_mac_x550em_a:
5592 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
5593 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
5597 /* Configure DCB traffic classes */
5598 for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5599 tc = &adapter->dcb_cfg.tc_config[j];
5600 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5601 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5602 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5603 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5604 tc->dcb_pfc = pfc_disabled;
5607 /* Initialize default user to priority mapping, UPx->TC0 */
5608 tc = &adapter->dcb_cfg.tc_config[0];
5609 tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5610 tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5612 adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5613 adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5614 adapter->dcb_cfg.pfc_mode_enable = false;
5615 adapter->dcb_set_bitmap = 0x00;
5616 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
5617 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5618 memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5619 sizeof(adapter->temp_dcb_cfg));
5624 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5625 * @adapter: board private structure to initialize
5627 * ixgbe_sw_init initializes the Adapter private data structure.
5628 * Fields are initialized based on PCI device information and
5629 * OS network device settings (MTU size).
5631 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
5632 const struct ixgbe_info *ii)
5634 struct ixgbe_hw *hw = &adapter->hw;
5635 struct pci_dev *pdev = adapter->pdev;
5636 unsigned int rss, fdir;
5640 /* PCI config space info */
5642 hw->vendor_id = pdev->vendor;
5643 hw->device_id = pdev->device;
5644 hw->revision_id = pdev->revision;
5645 hw->subsystem_vendor_id = pdev->subsystem_vendor;
5646 hw->subsystem_device_id = pdev->subsystem_device;
5648 /* get_invariants needs the device IDs */
5649 ii->get_invariants(hw);
5651 /* Set common capability flags and settings */
5652 rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5653 adapter->ring_feature[RING_F_RSS].limit = rss;
5654 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5655 adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5656 adapter->atr_sample_rate = 20;
5657 fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5658 adapter->ring_feature[RING_F_FDIR].limit = fdir;
5659 adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5660 #ifdef CONFIG_IXGBE_DCA
5661 adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5663 #ifdef CONFIG_IXGBE_DCB
5664 adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
5665 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
5668 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5669 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5670 #ifdef CONFIG_IXGBE_DCB
5671 /* Default traffic class to use for FCoE */
5672 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5673 #endif /* CONFIG_IXGBE_DCB */
5674 #endif /* IXGBE_FCOE */
5676 /* initialize static ixgbe jump table entries */
5677 adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
5679 if (!adapter->jump_tables[0])
5681 adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
5683 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
5684 adapter->jump_tables[i] = NULL;
5686 adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5687 hw->mac.num_rar_entries,
5689 if (!adapter->mac_table)
5692 /* Set MAC specific capability flags and exceptions */
5693 switch (hw->mac.type) {
5694 case ixgbe_mac_82598EB:
5695 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5697 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5698 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5700 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5701 adapter->ring_feature[RING_F_FDIR].limit = 0;
5702 adapter->atr_sample_rate = 0;
5703 adapter->fdir_pballoc = 0;
5705 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5706 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5707 #ifdef CONFIG_IXGBE_DCB
5708 adapter->fcoe.up = 0;
5709 #endif /* IXGBE_DCB */
5710 #endif /* IXGBE_FCOE */
5712 case ixgbe_mac_82599EB:
5713 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5714 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5716 case ixgbe_mac_X540:
5717 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
5718 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5719 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5721 case ixgbe_mac_x550em_a:
5722 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
5723 switch (hw->device_id) {
5724 case IXGBE_DEV_ID_X550EM_A_1G_T:
5725 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
5726 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5732 case ixgbe_mac_X550EM_x:
5733 #ifdef CONFIG_IXGBE_DCB
5734 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
5737 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5738 #ifdef CONFIG_IXGBE_DCB
5739 adapter->fcoe.up = 0;
5740 #endif /* IXGBE_DCB */
5741 #endif /* IXGBE_FCOE */
5743 case ixgbe_mac_X550:
5744 if (hw->mac.type == ixgbe_mac_X550)
5745 adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5746 #ifdef CONFIG_IXGBE_DCA
5747 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5749 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
5756 /* FCoE support exists, always init the FCoE lock */
5757 spin_lock_init(&adapter->fcoe.lock);
5760 /* n-tuple support exists, always init our spinlock */
5761 spin_lock_init(&adapter->fdir_perfect_lock);
5763 #ifdef CONFIG_IXGBE_DCB
5764 ixgbe_init_dcb(adapter);
5767 /* default flow control settings */
5768 hw->fc.requested_mode = ixgbe_fc_full;
5769 hw->fc.current_mode = ixgbe_fc_full; /* init for ethtool output */
5770 ixgbe_pbthresh_setup(adapter);
5771 hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5772 hw->fc.send_xon = true;
5773 hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5775 #ifdef CONFIG_PCI_IOV
5777 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5779 /* assign number of SR-IOV VFs */
5780 if (hw->mac.type != ixgbe_mac_82598EB) {
5781 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5782 adapter->num_vfs = 0;
5783 e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5785 adapter->num_vfs = max_vfs;
5788 #endif /* CONFIG_PCI_IOV */
5790 /* enable itr by default in dynamic mode */
5791 adapter->rx_itr_setting = 1;
5792 adapter->tx_itr_setting = 1;
5794 /* set default ring sizes */
5795 adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5796 adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5798 /* set default work limits */
5799 adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5801 /* initialize eeprom parameters */
5802 if (ixgbe_init_eeprom_params_generic(hw)) {
5803 e_dev_err("EEPROM initialization failed\n");
5807 /* PF holds first pool slot */
5808 set_bit(0, &adapter->fwd_bitmask);
5809 set_bit(__IXGBE_DOWN, &adapter->state);
5815 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5816 * @tx_ring: tx descriptor ring (for a specific queue) to setup
5818 * Return 0 on success, negative on failure
5820 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5822 struct device *dev = tx_ring->dev;
5823 int orig_node = dev_to_node(dev);
5827 size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5829 if (tx_ring->q_vector)
5830 ring_node = tx_ring->q_vector->numa_node;
5832 tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5833 if (!tx_ring->tx_buffer_info)
5834 tx_ring->tx_buffer_info = vzalloc(size);
5835 if (!tx_ring->tx_buffer_info)
5838 u64_stats_init(&tx_ring->syncp);
5840 /* round up to nearest 4K */
5841 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5842 tx_ring->size = ALIGN(tx_ring->size, 4096);
5844 set_dev_node(dev, ring_node);
5845 tx_ring->desc = dma_alloc_coherent(dev,
5849 set_dev_node(dev, orig_node);
5851 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5852 &tx_ring->dma, GFP_KERNEL);
5856 tx_ring->next_to_use = 0;
5857 tx_ring->next_to_clean = 0;
5861 vfree(tx_ring->tx_buffer_info);
5862 tx_ring->tx_buffer_info = NULL;
5863 dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5868 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5869 * @adapter: board private structure
5871 * If this function returns with an error, then it's possible one or
5872 * more of the rings is populated (while the rest are not). It is the
5873 * callers duty to clean those orphaned rings.
5875 * Return 0 on success, negative on failure
5877 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5881 for (i = 0; i < adapter->num_tx_queues; i++) {
5882 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5886 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5892 /* rewind the index freeing the rings as we go */
5894 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5899 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5900 * @rx_ring: rx descriptor ring (for a specific queue) to setup
5902 * Returns 0 on success, negative on failure
5904 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5906 struct device *dev = rx_ring->dev;
5907 int orig_node = dev_to_node(dev);
5911 size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5913 if (rx_ring->q_vector)
5914 ring_node = rx_ring->q_vector->numa_node;
5916 rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5917 if (!rx_ring->rx_buffer_info)
5918 rx_ring->rx_buffer_info = vzalloc(size);
5919 if (!rx_ring->rx_buffer_info)
5922 u64_stats_init(&rx_ring->syncp);
5924 /* Round up to nearest 4K */
5925 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5926 rx_ring->size = ALIGN(rx_ring->size, 4096);
5928 set_dev_node(dev, ring_node);
5929 rx_ring->desc = dma_alloc_coherent(dev,
5933 set_dev_node(dev, orig_node);
5935 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5936 &rx_ring->dma, GFP_KERNEL);
5940 rx_ring->next_to_clean = 0;
5941 rx_ring->next_to_use = 0;
5945 vfree(rx_ring->rx_buffer_info);
5946 rx_ring->rx_buffer_info = NULL;
5947 dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5952 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5953 * @adapter: board private structure
5955 * If this function returns with an error, then it's possible one or
5956 * more of the rings is populated (while the rest are not). It is the
5957 * callers duty to clean those orphaned rings.
5959 * Return 0 on success, negative on failure
5961 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5965 for (i = 0; i < adapter->num_rx_queues; i++) {
5966 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5970 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5975 err = ixgbe_setup_fcoe_ddp_resources(adapter);
5980 /* rewind the index freeing the rings as we go */
5982 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5987 * ixgbe_free_tx_resources - Free Tx Resources per Queue
5988 * @tx_ring: Tx descriptor ring for a specific queue
5990 * Free all transmit software resources
5992 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5994 ixgbe_clean_tx_ring(tx_ring);
5996 vfree(tx_ring->tx_buffer_info);
5997 tx_ring->tx_buffer_info = NULL;
5999 /* if not set, then don't free */
6003 dma_free_coherent(tx_ring->dev, tx_ring->size,
6004 tx_ring->desc, tx_ring->dma);
6006 tx_ring->desc = NULL;
6010 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6011 * @adapter: board private structure
6013 * Free all transmit software resources
6015 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6019 for (i = 0; i < adapter->num_tx_queues; i++)
6020 if (adapter->tx_ring[i]->desc)
6021 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6025 * ixgbe_free_rx_resources - Free Rx Resources
6026 * @rx_ring: ring to clean the resources from
6028 * Free all receive software resources
6030 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6032 ixgbe_clean_rx_ring(rx_ring);
6034 vfree(rx_ring->rx_buffer_info);
6035 rx_ring->rx_buffer_info = NULL;
6037 /* if not set, then don't free */
6041 dma_free_coherent(rx_ring->dev, rx_ring->size,
6042 rx_ring->desc, rx_ring->dma);
6044 rx_ring->desc = NULL;
6048 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6049 * @adapter: board private structure
6051 * Free all receive software resources
6053 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6058 ixgbe_free_fcoe_ddp_resources(adapter);
6061 for (i = 0; i < adapter->num_rx_queues; i++)
6062 if (adapter->rx_ring[i]->desc)
6063 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6067 * ixgbe_change_mtu - Change the Maximum Transfer Unit
6068 * @netdev: network interface device structure
6069 * @new_mtu: new value for maximum frame size
6071 * Returns 0 on success, negative on failure
6073 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6075 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6078 * For 82599EB we cannot allow legacy VFs to enable their receive
6079 * paths when MTU greater than 1500 is configured. So display a
6080 * warning that legacy VFs will be disabled.
6082 if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6083 (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6084 (new_mtu > ETH_DATA_LEN))
6085 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6087 e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6089 /* must set new MTU before calling down or up */
6090 netdev->mtu = new_mtu;
6092 if (netif_running(netdev))
6093 ixgbe_reinit_locked(adapter);
6099 * ixgbe_open - Called when a network interface is made active
6100 * @netdev: network interface device structure
6102 * Returns 0 on success, negative value on failure
6104 * The open entry point is called when a network interface is made
6105 * active by the system (IFF_UP). At this point all resources needed
6106 * for transmit and receive operations are allocated, the interrupt
6107 * handler is registered with the OS, the watchdog timer is started,
6108 * and the stack is notified that the interface is ready.
6110 int ixgbe_open(struct net_device *netdev)
6112 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6113 struct ixgbe_hw *hw = &adapter->hw;
6116 /* disallow open during test */
6117 if (test_bit(__IXGBE_TESTING, &adapter->state))
6120 netif_carrier_off(netdev);
6122 /* allocate transmit descriptors */
6123 err = ixgbe_setup_all_tx_resources(adapter);
6127 /* allocate receive descriptors */
6128 err = ixgbe_setup_all_rx_resources(adapter);
6132 ixgbe_configure(adapter);
6134 err = ixgbe_request_irq(adapter);
6138 /* Notify the stack of the actual queue counts. */
6139 if (adapter->num_rx_pools > 1)
6140 queues = adapter->num_rx_queues_per_pool;
6142 queues = adapter->num_tx_queues;
6144 err = netif_set_real_num_tx_queues(netdev, queues);
6146 goto err_set_queues;
6148 if (adapter->num_rx_pools > 1 &&
6149 adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
6150 queues = IXGBE_MAX_L2A_QUEUES;
6152 queues = adapter->num_rx_queues;
6153 err = netif_set_real_num_rx_queues(netdev, queues);
6155 goto err_set_queues;
6157 ixgbe_ptp_init(adapter);
6159 ixgbe_up_complete(adapter);
6161 ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6162 udp_tunnel_get_rx_info(netdev);
6167 ixgbe_free_irq(adapter);
6169 ixgbe_free_all_rx_resources(adapter);
6170 if (hw->phy.ops.set_phy_power && !adapter->wol)
6171 hw->phy.ops.set_phy_power(&adapter->hw, false);
6173 ixgbe_free_all_tx_resources(adapter);
6175 ixgbe_reset(adapter);
6180 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6182 ixgbe_ptp_suspend(adapter);
6184 if (adapter->hw.phy.ops.enter_lplu) {
6185 adapter->hw.phy.reset_disable = true;
6186 ixgbe_down(adapter);
6187 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6188 adapter->hw.phy.reset_disable = false;
6190 ixgbe_down(adapter);
6193 ixgbe_free_irq(adapter);
6195 ixgbe_free_all_tx_resources(adapter);
6196 ixgbe_free_all_rx_resources(adapter);
6200 * ixgbe_close - Disables a network interface
6201 * @netdev: network interface device structure
6203 * Returns 0, this is not allowed to fail
6205 * The close entry point is called when an interface is de-activated
6206 * by the OS. The hardware is still under the drivers control, but
6207 * needs to be disabled. A global MAC reset is issued to stop the
6208 * hardware, and all transmit and receive resources are freed.
6210 int ixgbe_close(struct net_device *netdev)
6212 struct ixgbe_adapter *adapter = netdev_priv(netdev);
6214 ixgbe_ptp_stop(adapter);
6216 if (netif_device_present(netdev))
6217 ixgbe_close_suspend(adapter);
6219 ixgbe_fdir_filter_exit(adapter);
6221 ixgbe_release_hw_control(adapter);
6227 static int ixgbe_resume(struct pci_dev *pdev)
6229 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6230 struct net_device *netdev = adapter->netdev;
6233 adapter->hw.hw_addr = adapter->io_addr;
6234 pci_set_power_state(pdev, PCI_D0);
6235 pci_restore_state(pdev);
6237 * pci_restore_state clears dev->state_saved so call
6238 * pci_save_state to restore it.
6240 pci_save_state(pdev);
6242 err = pci_enable_device_mem(pdev);
6244 e_dev_err("Cannot enable PCI device from suspend\n");
6247 smp_mb__before_atomic();
6248 clear_bit(__IXGBE_DISABLED, &adapter->state);
6249 pci_set_master(pdev);
6251 pci_wake_from_d3(pdev, false);
6253 ixgbe_reset(adapter);
6255 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6258 err = ixgbe_init_interrupt_scheme(adapter);
6259 if (!err && netif_running(netdev))
6260 err = ixgbe_open(netdev);
6264 netif_device_attach(netdev);
6269 #endif /* CONFIG_PM */
6271 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6273 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6274 struct net_device *netdev = adapter->netdev;
6275 struct ixgbe_hw *hw = &adapter->hw;
6277 u32 wufc = adapter->wol;
6283 netif_device_detach(netdev);
6285 if (netif_running(netdev))
6286 ixgbe_close_suspend(adapter);
6288 ixgbe_clear_interrupt_scheme(adapter);
6292 retval = pci_save_state(pdev);
6297 if (hw->mac.ops.stop_link_on_d3)
6298 hw->mac.ops.stop_link_on_d3(hw);
6301 ixgbe_set_rx_mode(netdev);
6303 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6304 if (hw->mac.ops.enable_tx_laser)
6305 hw->mac.ops.enable_tx_laser(hw);
6307 /* turn on all-multi mode if wake on multicast is enabled */
6308 if (wufc & IXGBE_WUFC_MC) {
6309 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6310 fctrl |= IXGBE_FCTRL_MPE;
6311 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6314 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6315 ctrl |= IXGBE_CTRL_GIO_DIS;
6316 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6318 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6320 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6321 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6324 switch (hw->mac.type) {
6325 case ixgbe_mac_82598EB:
6326 pci_wake_from_d3(pdev, false);
6328 case ixgbe_mac_82599EB:
6329 case ixgbe_mac_X540:
6330 case ixgbe_mac_X550:
6331 case ixgbe_mac_X550EM_x:
6332 case ixgbe_mac_x550em_a:
6333 pci_wake_from_d3(pdev, !!wufc);
6339 *enable_wake = !!wufc;
6340 if (hw->phy.ops.set_phy_power && !*enable_wake)
6341 hw->phy.ops.set_phy_power(hw, false);
6343 ixgbe_release_hw_control(adapter);
6345 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6346 pci_disable_device(pdev);
6352 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6357 retval = __ixgbe_shutdown(pdev, &wake);
6362 pci_prepare_to_sleep(pdev);
6364 pci_wake_from_d3(pdev, false);
6365 pci_set_power_state(pdev, PCI_D3hot);
6370 #endif /* CONFIG_PM */
6372 static void ixgbe_shutdown(struct pci_dev *pdev)
6376 __ixgbe_shutdown(pdev, &wake);
6378 if (system_state == SYSTEM_POWER_OFF) {
6379 pci_wake_from_d3(pdev, wake);
6380 pci_set_power_state(pdev, PCI_D3hot);
6385 * ixgbe_update_stats - Update the board statistics counters.
6386 * @adapter: board private structure
6388 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6390 struct net_device *netdev = adapter->netdev;
6391 struct ixgbe_hw *hw = &adapter->hw;
6392 struct ixgbe_hw_stats *hwstats = &adapter->stats;
6394 u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6395 u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6396 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6397 u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6399 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6400 test_bit(__IXGBE_RESETTING, &adapter->state))
6403 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6406 for (i = 0; i < adapter->num_rx_queues; i++) {
6407 rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6408 rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6410 adapter->rsc_total_count = rsc_count;
6411 adapter->rsc_total_flush = rsc_flush;
6414 for (i = 0; i < adapter->num_rx_queues; i++) {
6415 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6416 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6417 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6418 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6419 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6420 bytes += rx_ring->stats.bytes;
6421 packets += rx_ring->stats.packets;
6423 adapter->non_eop_descs = non_eop_descs;
6424 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6425 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6426 adapter->hw_csum_rx_error = hw_csum_rx_error;
6427 netdev->stats.rx_bytes = bytes;
6428 netdev->stats.rx_packets = packets;
6432 /* gather some stats to the adapter struct that are per queue */
6433 for (i = 0; i < adapter->num_tx_queues; i++) {
6434 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6435 restart_queue += tx_ring->tx_stats.restart_queue;
6436 tx_busy += tx_ring->tx_stats.tx_busy;
6437 bytes += tx_ring->stats.bytes;
6438 packets += tx_ring->stats.packets;
6440 adapter->restart_queue = restart_queue;
6441 adapter->tx_busy = tx_busy;
6442 netdev->stats.tx_bytes = bytes;
6443 netdev->stats.tx_packets = packets;
6445 hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6447 /* 8 register reads */
6448 for (i = 0; i < 8; i++) {
6449 /* for packet buffers not used, the register should read 0 */
6450 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6452 hwstats->mpc[i] += mpc;
6453 total_mpc += hwstats->mpc[i];
6454 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6455 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6456 switch (hw->mac.type) {
6457 case ixgbe_mac_82598EB:
6458 hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6459 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6460 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6461 hwstats->pxonrxc[i] +=
6462 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6464 case ixgbe_mac_82599EB:
6465 case ixgbe_mac_X540:
6466 case ixgbe_mac_X550:
6467 case ixgbe_mac_X550EM_x:
6468 case ixgbe_mac_x550em_a:
6469 hwstats->pxonrxc[i] +=
6470 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6477 /*16 register reads */
6478 for (i = 0; i < 16; i++) {
6479 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6480 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6481 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6482 (hw->mac.type == ixgbe_mac_X540) ||
6483 (hw->mac.type == ixgbe_mac_X550) ||
6484 (hw->mac.type == ixgbe_mac_X550EM_x) ||
6485 (hw->mac.type == ixgbe_mac_x550em_a)) {
6486 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6487 IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6488 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6489 IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6493 hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6494 /* work around hardware counting issue */
6495 hwstats->gprc -= missed_rx;
6497 ixgbe_update_xoff_received(adapter);
6499 /* 82598 hardware only has a 32 bit counter in the high register */
6500 switch (hw->mac.type) {
6501 case ixgbe_mac_82598EB:
6502 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6503 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6504 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6505 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6507 case ixgbe_mac_X540:
6508 case ixgbe_mac_X550:
6509 case ixgbe_mac_X550EM_x:
6510 case ixgbe_mac_x550em_a:
6511 /* OS2BMC stats are X540 and later */
6512 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6513 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6514 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6515 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6516 case ixgbe_mac_82599EB:
6517 for (i = 0; i < 16; i++)
6518 adapter->hw_rx_no_dma_resources +=
6519 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6520 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6521 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6522 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6523 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6524 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6525 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6526 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6527 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6528 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6530 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6531 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6532 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6533 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6534 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6535 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6536 /* Add up per cpu counters for total ddp aloc fail */
6537 if (adapter->fcoe.ddp_pool) {
6538 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6539 struct ixgbe_fcoe_ddp_pool *ddp_pool;
6541 u64 noddp = 0, noddp_ext_buff = 0;
6542 for_each_possible_cpu(cpu) {
6543 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6544 noddp += ddp_pool->noddp;
6545 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6547 hwstats->fcoe_noddp = noddp;
6548 hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6550 #endif /* IXGBE_FCOE */
6555 bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6556 hwstats->bprc += bprc;
6557 hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6558 if (hw->mac.type == ixgbe_mac_82598EB)
6559 hwstats->mprc -= bprc;
6560 hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6561 hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6562 hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6563 hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6564 hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6565 hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6566 hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6567 hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6568 lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6569 hwstats->lxontxc += lxon;
6570 lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6571 hwstats->lxofftxc += lxoff;
6572 hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6573 hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6575 * 82598 errata - tx of flow control packets is included in tx counters
6577 xon_off_tot = lxon + lxoff;
6578 hwstats->gptc -= xon_off_tot;
6579 hwstats->mptc -= xon_off_tot;
6580 hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6581 hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6582 hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6583 hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6584 hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6585 hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6586 hwstats->ptc64 -= xon_off_tot;
6587 hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6588 hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6589 hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6590 hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6591 hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6592 hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6594 /* Fill out the OS statistics structure */
6595 netdev->stats.multicast = hwstats->mprc;
6598 netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6599 netdev->stats.rx_dropped = 0;
6600 netdev->stats.rx_length_errors = hwstats->rlec;
6601 netdev->stats.rx_crc_errors = hwstats->crcerrs;
6602 netdev->stats.rx_missed_errors = total_mpc;
6606 * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6607 * @adapter: pointer to the device adapter structure
6609 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6611 struct ixgbe_hw *hw = &adapter->hw;
6614 if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6617 adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6619 /* if interface is down do nothing */
6620 if (test_bit(__IXGBE_DOWN, &adapter->state))
6623 /* do nothing if we are not using signature filters */
6624 if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6627 adapter->fdir_overflow++;
6629 if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6630 for (i = 0; i < adapter->num_tx_queues; i++)
6631 set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6632 &(adapter->tx_ring[i]->state));
6633 /* re-enable flow director interrupts */
6634 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6636 e_err(probe, "failed to finish FDIR re-initialization, "
6637 "ignored adding FDIR ATR filters\n");
6642 * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6643 * @adapter: pointer to the device adapter structure
6645 * This function serves two purposes. First it strobes the interrupt lines
6646 * in order to make certain interrupts are occurring. Secondly it sets the
6647 * bits needed to check for TX hangs. As a result we should immediately
6648 * determine if a hang has occurred.
6650 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6652 struct ixgbe_hw *hw = &adapter->hw;
6656 /* If we're down, removing or resetting, just bail */
6657 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6658 test_bit(__IXGBE_REMOVING, &adapter->state) ||
6659 test_bit(__IXGBE_RESETTING, &adapter->state))
6662 /* Force detection of hung controller */
6663 if (netif_carrier_ok(adapter->netdev)) {
6664 for (i = 0; i < adapter->num_tx_queues; i++)
6665 set_check_for_tx_hang(adapter->tx_ring[i]);
6668 if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6670 * for legacy and MSI interrupts don't set any bits
6671 * that are enabled for EIAM, because this operation
6672 * would set *both* EIMS and EICS for any bit in EIAM
6674 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6675 (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6677 /* get one bit for every active tx/rx interrupt vector */
6678 for (i = 0; i < adapter->num_q_vectors; i++) {
6679 struct ixgbe_q_vector *qv = adapter->q_vector[i];
6680 if (qv->rx.ring || qv->tx.ring)
6685 /* Cause software interrupt to ensure rings are cleaned */
6686 ixgbe_irq_rearm_queues(adapter, eics);
6690 * ixgbe_watchdog_update_link - update the link status
6691 * @adapter: pointer to the device adapter structure
6692 * @link_speed: pointer to a u32 to store the link_speed
6694 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6696 struct ixgbe_hw *hw = &adapter->hw;
6697 u32 link_speed = adapter->link_speed;
6698 bool link_up = adapter->link_up;
6699 bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6701 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6704 if (hw->mac.ops.check_link) {
6705 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6707 /* always assume link is up, if no check link function */
6708 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6712 if (adapter->ixgbe_ieee_pfc)
6713 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6715 if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6716 hw->mac.ops.fc_enable(hw);
6717 ixgbe_set_rx_drop_en(adapter);
6721 time_after(jiffies, (adapter->link_check_timeout +
6722 IXGBE_TRY_LINK_TIMEOUT))) {
6723 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6724 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6725 IXGBE_WRITE_FLUSH(hw);
6728 adapter->link_up = link_up;
6729 adapter->link_speed = link_speed;
6732 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6734 #ifdef CONFIG_IXGBE_DCB
6735 struct net_device *netdev = adapter->netdev;
6736 struct dcb_app app = {
6737 .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6742 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6743 up = dcb_ieee_getapp_mask(netdev, &app);
6745 adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6749 static int ixgbe_enable_macvlan(struct net_device *upper, void *data)
6751 if (netif_is_macvlan(upper)) {
6752 struct macvlan_dev *vlan = netdev_priv(upper);
6755 netif_tx_wake_all_queues(upper);
6762 * ixgbe_watchdog_link_is_up - update netif_carrier status and
6763 * print link up message
6764 * @adapter: pointer to the device adapter structure
6766 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6768 struct net_device *netdev = adapter->netdev;
6769 struct ixgbe_hw *hw = &adapter->hw;
6770 u32 link_speed = adapter->link_speed;
6771 const char *speed_str;
6772 bool flow_rx, flow_tx;
6774 /* only continue if link was previously down */
6775 if (netif_carrier_ok(netdev))
6778 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6780 switch (hw->mac.type) {
6781 case ixgbe_mac_82598EB: {
6782 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6783 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6784 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6785 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6788 case ixgbe_mac_X540:
6789 case ixgbe_mac_X550:
6790 case ixgbe_mac_X550EM_x:
6791 case ixgbe_mac_x550em_a:
6792 case ixgbe_mac_82599EB: {
6793 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6794 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6795 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6796 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6805 adapter->last_rx_ptp_check = jiffies;
6807 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6808 ixgbe_ptp_start_cyclecounter(adapter);
6810 switch (link_speed) {
6811 case IXGBE_LINK_SPEED_10GB_FULL:
6812 speed_str = "10 Gbps";
6814 case IXGBE_LINK_SPEED_2_5GB_FULL:
6815 speed_str = "2.5 Gbps";
6817 case IXGBE_LINK_SPEED_1GB_FULL:
6818 speed_str = "1 Gbps";
6820 case IXGBE_LINK_SPEED_100_FULL:
6821 speed_str = "100 Mbps";
6823 case IXGBE_LINK_SPEED_10_FULL:
6824 speed_str = "10 Mbps";
6827 speed_str = "unknown speed";
6830 e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
6831 ((flow_rx && flow_tx) ? "RX/TX" :
6833 (flow_tx ? "TX" : "None"))));
6835 netif_carrier_on(netdev);
6836 ixgbe_check_vf_rate_limit(adapter);
6838 /* enable transmits */
6839 netif_tx_wake_all_queues(adapter->netdev);
6841 /* enable any upper devices */
6843 netdev_walk_all_upper_dev_rcu(adapter->netdev,
6844 ixgbe_enable_macvlan, NULL);
6847 /* update the default user priority for VFs */
6848 ixgbe_update_default_up(adapter);
6850 /* ping all the active vfs to let them know link has changed */
6851 ixgbe_ping_all_vfs(adapter);
6855 * ixgbe_watchdog_link_is_down - update netif_carrier status and
6856 * print link down message
6857 * @adapter: pointer to the adapter structure
6859 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6861 struct net_device *netdev = adapter->netdev;
6862 struct ixgbe_hw *hw = &adapter->hw;
6864 adapter->link_up = false;
6865 adapter->link_speed = 0;
6867 /* only continue if link was up previously */
6868 if (!netif_carrier_ok(netdev))
6871 /* poll for SFP+ cable when link is down */
6872 if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6873 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6875 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6876 ixgbe_ptp_start_cyclecounter(adapter);
6878 e_info(drv, "NIC Link is Down\n");
6879 netif_carrier_off(netdev);
6881 /* ping all the active vfs to let them know link has changed */
6882 ixgbe_ping_all_vfs(adapter);
6885 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6889 for (i = 0; i < adapter->num_tx_queues; i++) {
6890 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6892 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6899 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6901 struct ixgbe_hw *hw = &adapter->hw;
6902 struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6903 u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6907 if (!adapter->num_vfs)
6910 /* resetting the PF is only needed for MAC before X550 */
6911 if (hw->mac.type >= ixgbe_mac_X550)
6914 for (i = 0; i < adapter->num_vfs; i++) {
6915 for (j = 0; j < q_per_pool; j++) {
6918 h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6919 t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6930 * ixgbe_watchdog_flush_tx - flush queues on link down
6931 * @adapter: pointer to the device adapter structure
6933 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6935 if (!netif_carrier_ok(adapter->netdev)) {
6936 if (ixgbe_ring_tx_pending(adapter) ||
6937 ixgbe_vf_tx_pending(adapter)) {
6938 /* We've lost link, so the controller stops DMA,
6939 * but we've got queued Tx work that's never going
6940 * to get done, so reset controller to flush Tx.
6941 * (Do the reset outside of interrupt context).
6943 e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6944 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6949 #ifdef CONFIG_PCI_IOV
6950 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6951 struct pci_dev *vfdev)
6953 if (!pci_wait_for_pending_transaction(vfdev))
6954 e_dev_warn("Issuing VFLR with pending transactions\n");
6956 e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6957 pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6962 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6964 struct ixgbe_hw *hw = &adapter->hw;
6965 struct pci_dev *pdev = adapter->pdev;
6969 if (!(netif_carrier_ok(adapter->netdev)))
6972 gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6973 if (gpc) /* If incrementing then no need for the check below */
6975 /* Check to see if a bad DMA write target from an errant or
6976 * malicious VF has caused a PCIe error. If so then we can
6977 * issue a VFLR to the offending VF(s) and then resume without
6978 * requesting a full slot reset.
6984 /* check status reg for all VFs owned by this PF */
6985 for (vf = 0; vf < adapter->num_vfs; ++vf) {
6986 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
6991 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6992 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
6993 status_reg & PCI_STATUS_REC_MASTER_ABORT)
6994 ixgbe_issue_vf_flr(adapter, vfdev);
6998 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7002 /* Do not perform spoof check for 82598 or if not in IOV mode */
7003 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7004 adapter->num_vfs == 0)
7007 ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7010 * ssvpc register is cleared on read, if zero then no
7011 * spoofed packets in the last interval.
7016 e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7019 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7024 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7027 #endif /* CONFIG_PCI_IOV */
7031 * ixgbe_watchdog_subtask - check and bring link up
7032 * @adapter: pointer to the device adapter structure
7034 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7036 /* if interface is down, removing or resetting, do nothing */
7037 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7038 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7039 test_bit(__IXGBE_RESETTING, &adapter->state))
7042 ixgbe_watchdog_update_link(adapter);
7044 if (adapter->link_up)
7045 ixgbe_watchdog_link_is_up(adapter);
7047 ixgbe_watchdog_link_is_down(adapter);
7049 ixgbe_check_for_bad_vf(adapter);
7050 ixgbe_spoof_check(adapter);
7051 ixgbe_update_stats(adapter);
7053 ixgbe_watchdog_flush_tx(adapter);
7057 * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7058 * @adapter: the ixgbe adapter structure
7060 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7062 struct ixgbe_hw *hw = &adapter->hw;
7065 /* not searching for SFP so there is nothing to do here */
7066 if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7067 !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7070 if (adapter->sfp_poll_time &&
7071 time_after(adapter->sfp_poll_time, jiffies))
7072 return; /* If not yet time to poll for SFP */
7074 /* someone else is in init, wait until next service event */
7075 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7078 adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7080 err = hw->phy.ops.identify_sfp(hw);
7081 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7084 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7085 /* If no cable is present, then we need to reset
7086 * the next time we find a good cable. */
7087 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7094 /* exit if reset not needed */
7095 if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7098 adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7101 * A module may be identified correctly, but the EEPROM may not have
7102 * support for that module. setup_sfp() will fail in that case, so
7103 * we should not allow that module to load.
7105 if (hw->mac.type == ixgbe_mac_82598EB)
7106 err = hw->phy.ops.reset(hw);
7108 err = hw->mac.ops.setup_sfp(hw);
7110 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7113 adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7114 e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7117 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7119 if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7120 (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7121 e_dev_err("failed to initialize because an unsupported "
7122 "SFP+ module type was detected.\n");
7123 e_dev_err("Reload the driver after installing a "
7124 "supported module.\n");
7125 unregister_netdev(adapter->netdev);
7130 * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7131 * @adapter: the ixgbe adapter structure
7133 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7135 struct ixgbe_hw *hw = &adapter->hw;
7137 bool autoneg = false;
7139 if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7142 /* someone else is in init, wait until next service event */
7143 if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7146 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7148 speed = hw->phy.autoneg_advertised;
7149 if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
7150 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
7152 /* setup the highest link when no autoneg */
7154 if (speed & IXGBE_LINK_SPEED_10GB_FULL)
7155 speed = IXGBE_LINK_SPEED_10GB_FULL;
7159 if (hw->mac.ops.setup_link)
7160 hw->mac.ops.setup_link(hw, speed, true);
7162 adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7163 adapter->link_check_timeout = jiffies;
7164 clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7168 * ixgbe_service_timer - Timer Call-back
7169 * @data: pointer to adapter cast into an unsigned long
7171 static void ixgbe_service_timer(unsigned long data)
7173 struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
7174 unsigned long next_event_offset;
7176 /* poll faster when waiting for link */
7177 if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7178 next_event_offset = HZ / 10;
7180 next_event_offset = HZ * 2;
7182 /* Reset the timer */
7183 mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7185 ixgbe_service_event_schedule(adapter);
7188 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7190 struct ixgbe_hw *hw = &adapter->hw;
7193 if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7196 adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7198 if (!hw->phy.ops.handle_lasi)
7201 status = hw->phy.ops.handle_lasi(&adapter->hw);
7202 if (status != IXGBE_ERR_OVERTEMP)
7205 e_crit(drv, "%s\n", ixgbe_overheat_msg);
7208 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7210 if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7213 /* If we're already down, removing or resetting, just bail */
7214 if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7215 test_bit(__IXGBE_REMOVING, &adapter->state) ||
7216 test_bit(__IXGBE_RESETTING, &adapter->state))
7219 ixgbe_dump(adapter);
7220 netdev_err(adapter->netdev, "Reset adapter\n");
7221 adapter->tx_timeout_count++;
7224 ixgbe_reinit_locked(adapter);
7229 * ixgbe_service_task - manages and runs subtasks
7230 * @work: pointer to work_struct containing our data
7232 static void ixgbe_service_task(struct work_struct *work)
7234 struct ixgbe_adapter *adapter = container_of(work,
7235 struct ixgbe_adapter,
7237 if (ixgbe_removed(adapter->hw.hw_addr)) {
7238 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7240 ixgbe_down(adapter);
7243 ixgbe_service_event_complete(adapter);
7246 if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7248 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7249 udp_tunnel_get_rx_info(adapter->netdev);
7252 ixgbe_reset_subtask(adapter);
7253 ixgbe_phy_interrupt_subtask(adapter);
7254 ixgbe_sfp_detection_subtask(adapter);
7255 ixgbe_sfp_link_config_subtask(adapter);
7256 ixgbe_check_overtemp_subtask(adapter);
7257 ixgbe_watchdog_subtask(adapter);
7258 ixgbe_fdir_reinit_subtask(adapter);
7259 ixgbe_check_hang_subtask(adapter);
7261 if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7262 ixgbe_ptp_overflow_check(adapter);
7263 ixgbe_ptp_rx_hang(adapter);
7266 ixgbe_service_event_complete(adapter);
7269 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7270 struct ixgbe_tx_buffer *first,
7273 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7274 struct sk_buff *skb = first->skb;
7284 u32 paylen, l4_offset;
7287 if (skb->ip_summed != CHECKSUM_PARTIAL)
7290 if (!skb_is_gso(skb))
7293 err = skb_cow_head(skb, 0);
7297 ip.hdr = skb_network_header(skb);
7298 l4.hdr = skb_checksum_start(skb);
7300 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7301 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7303 /* initialize outer IP header fields */
7304 if (ip.v4->version == 4) {
7305 unsigned char *csum_start = skb_checksum_start(skb);
7306 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7308 /* IP header will have to cancel out any data that
7309 * is not a part of the outer IP header
7311 ip.v4->check = csum_fold(csum_partial(trans_start,
7312 csum_start - trans_start,
7314 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7317 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7318 IXGBE_TX_FLAGS_CSUM |
7319 IXGBE_TX_FLAGS_IPV4;
7321 ip.v6->payload_len = 0;
7322 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7323 IXGBE_TX_FLAGS_CSUM;
7326 /* determine offset of inner transport header */
7327 l4_offset = l4.hdr - skb->data;
7329 /* compute length of segmentation header */
7330 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
7332 /* remove payload length from inner checksum */
7333 paylen = skb->len - l4_offset;
7334 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
7336 /* update gso size and bytecount with header size */
7337 first->gso_segs = skb_shinfo(skb)->gso_segs;
7338 first->bytecount += (first->gso_segs - 1) * *hdr_len;
7340 /* mss_l4len_id: use 0 as index for TSO */
7341 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
7342 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
7344 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
7345 vlan_macip_lens = l4.hdr - ip.hdr;
7346 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
7347 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7349 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
7355 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
7357 unsigned int offset = 0;
7359 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
7361 return offset == skb_checksum_start_offset(skb);
7364 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
7365 struct ixgbe_tx_buffer *first)
7367 struct sk_buff *skb = first->skb;
7368 u32 vlan_macip_lens = 0;
7371 if (skb->ip_summed != CHECKSUM_PARTIAL) {
7373 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
7374 IXGBE_TX_FLAGS_CC)))
7379 switch (skb->csum_offset) {
7380 case offsetof(struct tcphdr, check):
7381 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7383 case offsetof(struct udphdr, check):
7385 case offsetof(struct sctphdr, checksum):
7386 /* validate that this is actually an SCTP request */
7387 if (((first->protocol == htons(ETH_P_IP)) &&
7388 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
7389 ((first->protocol == htons(ETH_P_IPV6)) &&
7390 ixgbe_ipv6_csum_is_sctp(skb))) {
7391 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7396 skb_checksum_help(skb);
7400 /* update TX checksum flag */
7401 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7402 vlan_macip_lens = skb_checksum_start_offset(skb) -
7403 skb_network_offset(skb);
7405 /* vlan_macip_lens: MACLEN, VLAN tag */
7406 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
7407 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7409 ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd, 0);
7412 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7413 ((_flag <= _result) ? \
7414 ((u32)(_input & _flag) * (_result / _flag)) : \
7415 ((u32)(_input & _flag) / (_flag / _result)))
7417 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7419 /* set type for advanced descriptor with frame checksum insertion */
7420 u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7421 IXGBE_ADVTXD_DCMD_DEXT |
7422 IXGBE_ADVTXD_DCMD_IFCS;
7424 /* set HW vlan bit if vlan is present */
7425 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7426 IXGBE_ADVTXD_DCMD_VLE);
7428 /* set segmentation enable bits for TSO/FSO */
7429 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7430 IXGBE_ADVTXD_DCMD_TSE);
7432 /* set timestamp bit if present */
7433 cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7434 IXGBE_ADVTXD_MAC_TSTAMP);
7436 /* insert frame checksum */
7437 cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7442 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7443 u32 tx_flags, unsigned int paylen)
7445 u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7447 /* enable L4 checksum for TSO and TX checksum offload */
7448 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7449 IXGBE_TX_FLAGS_CSUM,
7450 IXGBE_ADVTXD_POPTS_TXSM);
7452 /* enble IPv4 checksum for TSO */
7453 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7454 IXGBE_TX_FLAGS_IPV4,
7455 IXGBE_ADVTXD_POPTS_IXSM);
7458 * Check Context must be set if Tx switch is enabled, which it
7459 * always is for case where virtual functions are running
7461 olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7465 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7468 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7470 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7472 /* Herbert's original patch had:
7473 * smp_mb__after_netif_stop_queue();
7474 * but since that doesn't exist yet, just open code it.
7478 /* We need to check again in a case another CPU has just
7479 * made room available.
7481 if (likely(ixgbe_desc_unused(tx_ring) < size))
7484 /* A reprieve! - use start_queue because it doesn't call schedule */
7485 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7486 ++tx_ring->tx_stats.restart_queue;
7490 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7492 if (likely(ixgbe_desc_unused(tx_ring) >= size))
7495 return __ixgbe_maybe_stop_tx(tx_ring, size);
7498 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7501 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7502 struct ixgbe_tx_buffer *first,
7505 struct sk_buff *skb = first->skb;
7506 struct ixgbe_tx_buffer *tx_buffer;
7507 union ixgbe_adv_tx_desc *tx_desc;
7508 struct skb_frag_struct *frag;
7510 unsigned int data_len, size;
7511 u32 tx_flags = first->tx_flags;
7512 u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7513 u16 i = tx_ring->next_to_use;
7515 tx_desc = IXGBE_TX_DESC(tx_ring, i);
7517 ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7519 size = skb_headlen(skb);
7520 data_len = skb->data_len;
7523 if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7524 if (data_len < sizeof(struct fcoe_crc_eof)) {
7525 size -= sizeof(struct fcoe_crc_eof) - data_len;
7528 data_len -= sizeof(struct fcoe_crc_eof);
7533 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7537 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7538 if (dma_mapping_error(tx_ring->dev, dma))
7541 /* record length, and DMA address */
7542 dma_unmap_len_set(tx_buffer, len, size);
7543 dma_unmap_addr_set(tx_buffer, dma, dma);
7545 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7547 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7548 tx_desc->read.cmd_type_len =
7549 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7553 if (i == tx_ring->count) {
7554 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7557 tx_desc->read.olinfo_status = 0;
7559 dma += IXGBE_MAX_DATA_PER_TXD;
7560 size -= IXGBE_MAX_DATA_PER_TXD;
7562 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7565 if (likely(!data_len))
7568 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7572 if (i == tx_ring->count) {
7573 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7576 tx_desc->read.olinfo_status = 0;
7579 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7581 size = skb_frag_size(frag);
7585 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7588 tx_buffer = &tx_ring->tx_buffer_info[i];
7591 /* write last descriptor with RS and EOP bits */
7592 cmd_type |= size | IXGBE_TXD_CMD;
7593 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7595 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7597 /* set the timestamp */
7598 first->time_stamp = jiffies;
7601 * Force memory writes to complete before letting h/w know there
7602 * are new descriptors to fetch. (Only applicable for weak-ordered
7603 * memory model archs, such as IA-64).
7605 * We also need this memory barrier to make certain all of the
7606 * status bits have been updated before next_to_watch is written.
7610 /* set next_to_watch value indicating a packet is present */
7611 first->next_to_watch = tx_desc;
7614 if (i == tx_ring->count)
7617 tx_ring->next_to_use = i;
7619 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7621 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7622 writel(i, tx_ring->tail);
7624 /* we need this if more than one processor can write to our tail
7625 * at a time, it synchronizes IO on IA64/Altix systems
7632 dev_err(tx_ring->dev, "TX DMA map failed\n");
7634 /* clear dma mappings for failed tx_buffer_info map */
7636 tx_buffer = &tx_ring->tx_buffer_info[i];
7637 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7638 if (tx_buffer == first)
7645 tx_ring->next_to_use = i;
7648 static void ixgbe_atr(struct ixgbe_ring *ring,
7649 struct ixgbe_tx_buffer *first)
7651 struct ixgbe_q_vector *q_vector = ring->q_vector;
7652 union ixgbe_atr_hash_dword input = { .dword = 0 };
7653 union ixgbe_atr_hash_dword common = { .dword = 0 };
7655 unsigned char *network;
7657 struct ipv6hdr *ipv6;
7661 struct sk_buff *skb;
7665 /* if ring doesn't have a interrupt vector, cannot perform ATR */
7669 /* do nothing if sampling is disabled */
7670 if (!ring->atr_sample_rate)
7675 /* currently only IPv4/IPv6 with TCP is supported */
7676 if ((first->protocol != htons(ETH_P_IP)) &&
7677 (first->protocol != htons(ETH_P_IPV6)))
7680 /* snag network header to get L4 type and address */
7682 hdr.network = skb_network_header(skb);
7683 if (unlikely(hdr.network <= skb->data))
7685 if (skb->encapsulation &&
7686 first->protocol == htons(ETH_P_IP) &&
7687 hdr.ipv4->protocol == IPPROTO_UDP) {
7688 struct ixgbe_adapter *adapter = q_vector->adapter;
7690 if (unlikely(skb_tail_pointer(skb) < hdr.network +
7694 /* verify the port is recognized as VXLAN */
7695 if (adapter->vxlan_port &&
7696 udp_hdr(skb)->dest == adapter->vxlan_port)
7697 hdr.network = skb_inner_network_header(skb);
7699 if (adapter->geneve_port &&
7700 udp_hdr(skb)->dest == adapter->geneve_port)
7701 hdr.network = skb_inner_network_header(skb);
7704 /* Make sure we have at least [minimum IPv4 header + TCP]
7705 * or [IPv6 header] bytes
7707 if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
7710 /* Currently only IPv4/IPv6 with TCP is supported */
7711 switch (hdr.ipv4->version) {
7713 /* access ihl as u8 to avoid unaligned access on ia64 */
7714 hlen = (hdr.network[0] & 0x0F) << 2;
7715 l4_proto = hdr.ipv4->protocol;
7718 hlen = hdr.network - skb->data;
7719 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
7720 hlen -= hdr.network - skb->data;
7726 if (l4_proto != IPPROTO_TCP)
7729 if (unlikely(skb_tail_pointer(skb) < hdr.network +
7730 hlen + sizeof(struct tcphdr)))
7733 th = (struct tcphdr *)(hdr.network + hlen);
7735 /* skip this packet since the socket is closing */
7739 /* sample on all syn packets or once every atr sample count */
7740 if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7743 /* reset sample count */
7744 ring->atr_count = 0;
7746 vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7749 * src and dst are inverted, think how the receiver sees them
7751 * The input is broken into two sections, a non-compressed section
7752 * containing vm_pool, vlan_id, and flow_type. The rest of the data
7753 * is XORed together and stored in the compressed dword.
7755 input.formatted.vlan_id = vlan_id;
7758 * since src port and flex bytes occupy the same word XOR them together
7759 * and write the value to source port portion of compressed dword
7761 if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7762 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7764 common.port.src ^= th->dest ^ first->protocol;
7765 common.port.dst ^= th->source;
7767 switch (hdr.ipv4->version) {
7769 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7770 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7773 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7774 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7775 hdr.ipv6->saddr.s6_addr32[1] ^
7776 hdr.ipv6->saddr.s6_addr32[2] ^
7777 hdr.ipv6->saddr.s6_addr32[3] ^
7778 hdr.ipv6->daddr.s6_addr32[0] ^
7779 hdr.ipv6->daddr.s6_addr32[1] ^
7780 hdr.ipv6->daddr.s6_addr32[2] ^
7781 hdr.ipv6->daddr.s6_addr32[3];
7787 if (hdr.network != skb_network_header(skb))
7788 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
7790 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7791 ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7792 input, common, ring->queue_index);
7795 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7796 void *accel_priv, select_queue_fallback_t fallback)
7798 struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7800 struct ixgbe_adapter *adapter;
7801 struct ixgbe_ring_feature *f;
7806 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7811 * only execute the code below if protocol is FCoE
7812 * or FIP and we have FCoE enabled on the adapter
7814 switch (vlan_get_protocol(skb)) {
7815 case htons(ETH_P_FCOE):
7816 case htons(ETH_P_FIP):
7817 adapter = netdev_priv(dev);
7819 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7822 return fallback(dev, skb);
7825 f = &adapter->ring_feature[RING_F_FCOE];
7827 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7830 while (txq >= f->indices)
7833 return txq + f->offset;
7835 return fallback(dev, skb);
7839 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7840 struct ixgbe_adapter *adapter,
7841 struct ixgbe_ring *tx_ring)
7843 struct ixgbe_tx_buffer *first;
7847 u16 count = TXD_USE_COUNT(skb_headlen(skb));
7848 __be16 protocol = skb->protocol;
7852 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7853 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7854 * + 2 desc gap to keep tail from touching head,
7855 * + 1 desc for context descriptor,
7856 * otherwise try next time
7858 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7859 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7861 if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7862 tx_ring->tx_stats.tx_busy++;
7863 return NETDEV_TX_BUSY;
7866 /* record the location of the first descriptor for this packet */
7867 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7869 first->bytecount = skb->len;
7870 first->gso_segs = 1;
7872 /* if we have a HW VLAN tag being added default to the HW one */
7873 if (skb_vlan_tag_present(skb)) {
7874 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7875 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7876 /* else if it is a SW VLAN check the next protocol and store the tag */
7877 } else if (protocol == htons(ETH_P_8021Q)) {
7878 struct vlan_hdr *vhdr, _vhdr;
7879 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7883 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7884 IXGBE_TX_FLAGS_VLAN_SHIFT;
7885 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7887 protocol = vlan_get_protocol(skb);
7889 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7890 adapter->ptp_clock &&
7891 !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7893 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7894 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7896 /* schedule check for Tx timestamp */
7897 adapter->ptp_tx_skb = skb_get(skb);
7898 adapter->ptp_tx_start = jiffies;
7899 schedule_work(&adapter->ptp_tx_work);
7902 skb_tx_timestamp(skb);
7904 #ifdef CONFIG_PCI_IOV
7906 * Use the l2switch_enable flag - would be false if the DMA
7907 * Tx switch had been disabled.
7909 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7910 tx_flags |= IXGBE_TX_FLAGS_CC;
7913 /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7914 if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7915 ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7916 (skb->priority != TC_PRIO_CONTROL))) {
7917 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7918 tx_flags |= (skb->priority & 0x7) <<
7919 IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7920 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7921 struct vlan_ethhdr *vhdr;
7923 if (skb_cow_head(skb, 0))
7925 vhdr = (struct vlan_ethhdr *)skb->data;
7926 vhdr->h_vlan_TCI = htons(tx_flags >>
7927 IXGBE_TX_FLAGS_VLAN_SHIFT);
7929 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7933 /* record initial flags and protocol */
7934 first->tx_flags = tx_flags;
7935 first->protocol = protocol;
7938 /* setup tx offload for FCoE */
7939 if ((protocol == htons(ETH_P_FCOE)) &&
7940 (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7941 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7948 #endif /* IXGBE_FCOE */
7949 tso = ixgbe_tso(tx_ring, first, &hdr_len);
7953 ixgbe_tx_csum(tx_ring, first);
7955 /* add the ATR filter if ATR is on */
7956 if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7957 ixgbe_atr(tx_ring, first);
7961 #endif /* IXGBE_FCOE */
7962 ixgbe_tx_map(tx_ring, first, hdr_len);
7964 return NETDEV_TX_OK;
7967 dev_kfree_skb_any(first->skb);
7970 return NETDEV_TX_OK;
7973 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7974 struct net_device *netdev,
7975 struct ixgbe_ring *ring)
7977 struct ixgbe_adapter *adapter = netdev_priv(netdev);
7978 struct ixgbe_ring *tx_ring;
7981 * The minimum packet size for olinfo paylen is 17 so pad the skb
7982 * in order to meet this minimum size requirement.
7984 if (skb_put_padto(skb, 17))
7985 return NETDEV_TX_OK;
7987 tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7989 return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7992 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7993 struct net_device *netdev)
7995 return __ixgbe_xmit_frame(skb, netdev, NULL);
7999 * ixgbe_set_mac - Change the Ethernet Address of the NIC
8000 * @netdev: network interface device structure
8001 * @p: pointer to an address structure
8003 * Returns 0 on success, negative on failure
8005 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8007 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8008 struct ixgbe_hw *hw = &adapter->hw;
8009 struct sockaddr *addr = p;
8011 if (!is_valid_ether_addr(addr->sa_data))
8012 return -EADDRNOTAVAIL;
8014 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8015 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8017 ixgbe_mac_set_default_filter(adapter);
8023 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8025 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8026 struct ixgbe_hw *hw = &adapter->hw;
8030 if (prtad != hw->phy.mdio.prtad)
8032 rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8038 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8039 u16 addr, u16 value)
8041 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8042 struct ixgbe_hw *hw = &adapter->hw;
8044 if (prtad != hw->phy.mdio.prtad)
8046 return hw->phy.ops.write_reg(hw, addr, devad, value);
8049 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8051 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8055 return ixgbe_ptp_set_ts_config(adapter, req);
8057 return ixgbe_ptp_get_ts_config(adapter, req);
8059 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8064 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8066 * @netdev: network interface device structure
8068 * Returns non-zero on failure
8070 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8073 struct ixgbe_adapter *adapter = netdev_priv(dev);
8074 struct ixgbe_hw *hw = &adapter->hw;
8076 if (is_valid_ether_addr(hw->mac.san_addr)) {
8078 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8081 /* update SAN MAC vmdq pool selection */
8082 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8088 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8090 * @netdev: network interface device structure
8092 * Returns non-zero on failure
8094 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8097 struct ixgbe_adapter *adapter = netdev_priv(dev);
8098 struct ixgbe_mac_info *mac = &adapter->hw.mac;
8100 if (is_valid_ether_addr(mac->san_addr)) {
8102 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8108 #ifdef CONFIG_NET_POLL_CONTROLLER
8110 * Polling 'interrupt' - used by things like netconsole to send skbs
8111 * without having to re-enable interrupts. It's not called while
8112 * the interrupt routine is executing.
8114 static void ixgbe_netpoll(struct net_device *netdev)
8116 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8119 /* if interface is down do nothing */
8120 if (test_bit(__IXGBE_DOWN, &adapter->state))
8123 /* loop through and schedule all active queues */
8124 for (i = 0; i < adapter->num_q_vectors; i++)
8125 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
8130 static void ixgbe_get_stats64(struct net_device *netdev,
8131 struct rtnl_link_stats64 *stats)
8133 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8137 for (i = 0; i < adapter->num_rx_queues; i++) {
8138 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
8144 start = u64_stats_fetch_begin_irq(&ring->syncp);
8145 packets = ring->stats.packets;
8146 bytes = ring->stats.bytes;
8147 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8148 stats->rx_packets += packets;
8149 stats->rx_bytes += bytes;
8153 for (i = 0; i < adapter->num_tx_queues; i++) {
8154 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
8160 start = u64_stats_fetch_begin_irq(&ring->syncp);
8161 packets = ring->stats.packets;
8162 bytes = ring->stats.bytes;
8163 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8164 stats->tx_packets += packets;
8165 stats->tx_bytes += bytes;
8170 /* following stats updated by ixgbe_watchdog_task() */
8171 stats->multicast = netdev->stats.multicast;
8172 stats->rx_errors = netdev->stats.rx_errors;
8173 stats->rx_length_errors = netdev->stats.rx_length_errors;
8174 stats->rx_crc_errors = netdev->stats.rx_crc_errors;
8175 stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8178 #ifdef CONFIG_IXGBE_DCB
8180 * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8181 * @adapter: pointer to ixgbe_adapter
8182 * @tc: number of traffic classes currently enabled
8184 * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8185 * 802.1Q priority maps to a packet buffer that exists.
8187 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8189 struct ixgbe_hw *hw = &adapter->hw;
8193 /* 82598 have a static priority to TC mapping that can not
8194 * be changed so no validation is needed.
8196 if (hw->mac.type == ixgbe_mac_82598EB)
8199 reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8202 for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8203 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8205 /* If up2tc is out of bounds default to zero */
8207 reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8211 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8217 * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8218 * @adapter: Pointer to adapter struct
8220 * Populate the netdev user priority to tc map
8222 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
8224 struct net_device *dev = adapter->netdev;
8225 struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
8226 struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
8229 for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
8232 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
8233 tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
8235 tc = ets->prio_tc[prio];
8237 netdev_set_prio_tc_map(dev, prio, tc);
8241 #endif /* CONFIG_IXGBE_DCB */
8243 * ixgbe_setup_tc - configure net_device for multiple traffic classes
8245 * @netdev: net device to configure
8246 * @tc: number of traffic classes to enable
8248 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
8250 struct ixgbe_adapter *adapter = netdev_priv(dev);
8251 struct ixgbe_hw *hw = &adapter->hw;
8254 /* Hardware supports up to 8 traffic classes */
8255 if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
8258 if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
8261 pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
8262 if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
8265 /* Hardware has to reinitialize queues and interrupts to
8266 * match packet buffer alignment. Unfortunately, the
8267 * hardware is not flexible enough to do this dynamically.
8269 if (netif_running(dev))
8272 ixgbe_reset(adapter);
8274 ixgbe_clear_interrupt_scheme(adapter);
8276 #ifdef CONFIG_IXGBE_DCB
8278 netdev_set_num_tc(dev, tc);
8279 ixgbe_set_prio_tc_map(adapter);
8281 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
8283 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
8284 adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
8285 adapter->hw.fc.requested_mode = ixgbe_fc_none;
8288 netdev_reset_tc(dev);
8290 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8291 adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
8293 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
8295 adapter->temp_dcb_cfg.pfc_mode_enable = false;
8296 adapter->dcb_cfg.pfc_mode_enable = false;
8299 ixgbe_validate_rtr(adapter, tc);
8301 #endif /* CONFIG_IXGBE_DCB */
8302 ixgbe_init_interrupt_scheme(adapter);
8304 if (netif_running(dev))
8305 return ixgbe_open(dev);
8310 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
8311 struct tc_cls_u32_offload *cls)
8313 u32 hdl = cls->knode.handle;
8314 u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
8315 u32 loc = cls->knode.handle & 0xfffff;
8317 struct ixgbe_jump_table *jump = NULL;
8319 if (loc > IXGBE_MAX_HW_ENTRIES)
8322 if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
8325 /* Clear this filter in the link data it is associated with */
8326 if (uhtid != 0x800) {
8327 jump = adapter->jump_tables[uhtid];
8330 if (!test_bit(loc - 1, jump->child_loc_map))
8332 clear_bit(loc - 1, jump->child_loc_map);
8335 /* Check if the filter being deleted is a link */
8336 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
8337 jump = adapter->jump_tables[i];
8338 if (jump && jump->link_hdl == hdl) {
8339 /* Delete filters in the hardware in the child hash
8340 * table associated with this link
8342 for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
8343 if (!test_bit(j, jump->child_loc_map))
8345 spin_lock(&adapter->fdir_perfect_lock);
8346 err = ixgbe_update_ethtool_fdir_entry(adapter,
8349 spin_unlock(&adapter->fdir_perfect_lock);
8350 clear_bit(j, jump->child_loc_map);
8352 /* Remove resources for this link */
8356 adapter->jump_tables[i] = NULL;
8361 spin_lock(&adapter->fdir_perfect_lock);
8362 err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
8363 spin_unlock(&adapter->fdir_perfect_lock);
8367 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
8369 struct tc_cls_u32_offload *cls)
8371 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
8373 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
8376 /* This ixgbe devices do not support hash tables at the moment
8377 * so abort when given hash tables.
8379 if (cls->hnode.divisor > 0)
8382 set_bit(uhtid - 1, &adapter->tables);
8386 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
8387 struct tc_cls_u32_offload *cls)
8389 u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
8391 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
8394 clear_bit(uhtid - 1, &adapter->tables);
8398 #ifdef CONFIG_NET_CLS_ACT
8399 struct upper_walk_data {
8400 struct ixgbe_adapter *adapter;
8406 static int get_macvlan_queue(struct net_device *upper, void *_data)
8408 if (netif_is_macvlan(upper)) {
8409 struct macvlan_dev *dfwd = netdev_priv(upper);
8410 struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
8411 struct upper_walk_data *data = _data;
8412 struct ixgbe_adapter *adapter = data->adapter;
8413 int ifindex = data->ifindex;
8415 if (vadapter && vadapter->netdev->ifindex == ifindex) {
8416 data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
8417 data->action = data->queue;
8425 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
8426 u8 *queue, u64 *action)
8428 unsigned int num_vfs = adapter->num_vfs, vf;
8429 struct upper_walk_data data;
8430 struct net_device *upper;
8432 /* redirect to a SRIOV VF */
8433 for (vf = 0; vf < num_vfs; ++vf) {
8434 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
8435 if (upper->ifindex == ifindex) {
8436 if (adapter->num_rx_pools > 1)
8439 *queue = vf * adapter->num_rx_queues_per_pool;
8442 *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
8447 /* redirect to a offloaded macvlan netdev */
8448 data.adapter = adapter;
8449 data.ifindex = ifindex;
8452 if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
8453 get_macvlan_queue, &data)) {
8454 *action = data.action;
8455 *queue = data.queue;
8463 static int parse_tc_actions(struct ixgbe_adapter *adapter,
8464 struct tcf_exts *exts, u64 *action, u8 *queue)
8466 const struct tc_action *a;
8470 if (tc_no_actions(exts))
8473 tcf_exts_to_list(exts, &actions);
8474 list_for_each_entry(a, &actions, list) {
8477 if (is_tcf_gact_shot(a)) {
8478 *action = IXGBE_FDIR_DROP_QUEUE;
8479 *queue = IXGBE_FDIR_DROP_QUEUE;
8483 /* Redirect to a VF or a offloaded macvlan */
8484 if (is_tcf_mirred_egress_redirect(a)) {
8485 int ifindex = tcf_mirred_ifindex(a);
8487 err = handle_redirect_action(adapter, ifindex, queue,
8497 static int parse_tc_actions(struct ixgbe_adapter *adapter,
8498 struct tcf_exts *exts, u64 *action, u8 *queue)
8502 #endif /* CONFIG_NET_CLS_ACT */
8504 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
8505 union ixgbe_atr_input *mask,
8506 struct tc_cls_u32_offload *cls,
8507 struct ixgbe_mat_field *field_ptr,
8508 struct ixgbe_nexthdr *nexthdr)
8512 bool found_entry = false, found_jump_field = false;
8514 for (i = 0; i < cls->knode.sel->nkeys; i++) {
8515 off = cls->knode.sel->keys[i].off;
8516 val = cls->knode.sel->keys[i].val;
8517 m = cls->knode.sel->keys[i].mask;
8519 for (j = 0; field_ptr[j].val; j++) {
8520 if (field_ptr[j].off == off) {
8521 field_ptr[j].val(input, mask, val, m);
8522 input->filter.formatted.flow_type |=
8529 if (nexthdr->off == cls->knode.sel->keys[i].off &&
8530 nexthdr->val == cls->knode.sel->keys[i].val &&
8531 nexthdr->mask == cls->knode.sel->keys[i].mask)
8532 found_jump_field = true;
8538 if (nexthdr && !found_jump_field)
8544 mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
8545 IXGBE_ATR_L4TYPE_MASK;
8547 if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
8548 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
8553 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
8555 struct tc_cls_u32_offload *cls)
8557 u32 loc = cls->knode.handle & 0xfffff;
8558 struct ixgbe_hw *hw = &adapter->hw;
8559 struct ixgbe_mat_field *field_ptr;
8560 struct ixgbe_fdir_filter *input = NULL;
8561 union ixgbe_atr_input *mask = NULL;
8562 struct ixgbe_jump_table *jump = NULL;
8563 int i, err = -EINVAL;
8565 u32 uhtid, link_uhtid;
8567 uhtid = TC_U32_USERHTID(cls->knode.handle);
8568 link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
8570 /* At the moment cls_u32 jumps to network layer and skips past
8571 * L2 headers. The canonical method to match L2 frames is to use
8572 * negative values. However this is error prone at best but really
8573 * just broken because there is no way to "know" what sort of hdr
8574 * is in front of the network layer. Fix cls_u32 to support L2
8575 * headers when needed.
8577 if (protocol != htons(ETH_P_IP))
8580 if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
8581 e_err(drv, "Location out of range\n");
8585 /* cls u32 is a graph starting at root node 0x800. The driver tracks
8586 * links and also the fields used to advance the parser across each
8587 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
8588 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
8589 * To add support for new nodes update ixgbe_model.h parse structures
8590 * this function _should_ be generic try not to hardcode values here.
8592 if (uhtid == 0x800) {
8593 field_ptr = (adapter->jump_tables[0])->mat;
8595 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
8597 if (!adapter->jump_tables[uhtid])
8599 field_ptr = (adapter->jump_tables[uhtid])->mat;
8605 /* At this point we know the field_ptr is valid and need to either
8606 * build cls_u32 link or attach filter. Because adding a link to
8607 * a handle that does not exist is invalid and the same for adding
8608 * rules to handles that don't exist.
8612 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
8614 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
8617 if (!test_bit(link_uhtid - 1, &adapter->tables))
8620 /* Multiple filters as links to the same hash table are not
8621 * supported. To add a new filter with the same next header
8622 * but different match/jump conditions, create a new hash table
8625 if (adapter->jump_tables[link_uhtid] &&
8626 (adapter->jump_tables[link_uhtid])->link_hdl) {
8627 e_err(drv, "Link filter exists for link: %x\n",
8632 for (i = 0; nexthdr[i].jump; i++) {
8633 if (nexthdr[i].o != cls->knode.sel->offoff ||
8634 nexthdr[i].s != cls->knode.sel->offshift ||
8635 nexthdr[i].m != cls->knode.sel->offmask)
8638 jump = kzalloc(sizeof(*jump), GFP_KERNEL);
8641 input = kzalloc(sizeof(*input), GFP_KERNEL);
8646 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
8651 jump->input = input;
8653 jump->link_hdl = cls->knode.handle;
8655 err = ixgbe_clsu32_build_input(input, mask, cls,
8656 field_ptr, &nexthdr[i]);
8658 jump->mat = nexthdr[i].jump;
8659 adapter->jump_tables[link_uhtid] = jump;
8666 input = kzalloc(sizeof(*input), GFP_KERNEL);
8669 mask = kzalloc(sizeof(*mask), GFP_KERNEL);
8675 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
8676 if ((adapter->jump_tables[uhtid])->input)
8677 memcpy(input, (adapter->jump_tables[uhtid])->input,
8679 if ((adapter->jump_tables[uhtid])->mask)
8680 memcpy(mask, (adapter->jump_tables[uhtid])->mask,
8683 /* Lookup in all child hash tables if this location is already
8684 * filled with a filter
8686 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
8687 struct ixgbe_jump_table *link = adapter->jump_tables[i];
8689 if (link && (test_bit(loc - 1, link->child_loc_map))) {
8690 e_err(drv, "Filter exists in location: %x\n",
8697 err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
8701 err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
8706 input->sw_idx = loc;
8708 spin_lock(&adapter->fdir_perfect_lock);
8710 if (hlist_empty(&adapter->fdir_filter_list)) {
8711 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
8712 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
8714 goto err_out_w_lock;
8715 } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
8717 goto err_out_w_lock;
8720 ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
8721 err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
8722 input->sw_idx, queue);
8724 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
8725 spin_unlock(&adapter->fdir_perfect_lock);
8727 if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
8728 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
8733 spin_unlock(&adapter->fdir_perfect_lock);
8743 static int __ixgbe_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
8744 struct tc_to_netdev *tc)
8746 struct ixgbe_adapter *adapter = netdev_priv(dev);
8748 if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
8749 tc->type == TC_SETUP_CLSU32) {
8750 switch (tc->cls_u32->command) {
8751 case TC_CLSU32_NEW_KNODE:
8752 case TC_CLSU32_REPLACE_KNODE:
8753 return ixgbe_configure_clsu32(adapter,
8754 proto, tc->cls_u32);
8755 case TC_CLSU32_DELETE_KNODE:
8756 return ixgbe_delete_clsu32(adapter, tc->cls_u32);
8757 case TC_CLSU32_NEW_HNODE:
8758 case TC_CLSU32_REPLACE_HNODE:
8759 return ixgbe_configure_clsu32_add_hnode(adapter, proto,
8761 case TC_CLSU32_DELETE_HNODE:
8762 return ixgbe_configure_clsu32_del_hnode(adapter,
8769 if (tc->type != TC_SETUP_MQPRIO)
8772 return ixgbe_setup_tc(dev, tc->tc);
8775 #ifdef CONFIG_PCI_IOV
8776 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
8778 struct net_device *netdev = adapter->netdev;
8781 ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
8786 void ixgbe_do_reset(struct net_device *netdev)
8788 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8790 if (netif_running(netdev))
8791 ixgbe_reinit_locked(adapter);
8793 ixgbe_reset(adapter);
8796 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
8797 netdev_features_t features)
8799 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8801 /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
8802 if (!(features & NETIF_F_RXCSUM))
8803 features &= ~NETIF_F_LRO;
8805 /* Turn off LRO if not RSC capable */
8806 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
8807 features &= ~NETIF_F_LRO;
8812 static int ixgbe_set_features(struct net_device *netdev,
8813 netdev_features_t features)
8815 struct ixgbe_adapter *adapter = netdev_priv(netdev);
8816 netdev_features_t changed = netdev->features ^ features;
8817 bool need_reset = false;
8819 /* Make sure RSC matches LRO, reset if change */
8820 if (!(features & NETIF_F_LRO)) {
8821 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8823 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
8824 } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
8825 !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
8826 if (adapter->rx_itr_setting == 1 ||
8827 adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
8828 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
8830 } else if ((changed ^ features) & NETIF_F_LRO) {
8831 e_info(probe, "rx-usecs set too low, "
8837 * Check if Flow Director n-tuple support or hw_tc support was
8838 * enabled or disabled. If the state changed, we need to reset.
8840 if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
8841 /* turn off ATR, enable perfect filters and reset */
8842 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
8845 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
8846 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8848 /* turn off perfect filters, enable ATR and reset */
8849 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
8852 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8854 /* We cannot enable ATR if SR-IOV is enabled */
8855 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
8856 /* We cannot enable ATR if we have 2 or more tcs */
8857 (netdev_get_num_tc(netdev) > 1) ||
8858 /* We cannot enable ATR if RSS is disabled */
8859 (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
8860 /* A sample rate of 0 indicates ATR disabled */
8861 (!adapter->atr_sample_rate))
8862 ; /* do nothing not supported */
8863 else /* otherwise supported and set the flag */
8864 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
8867 if (changed & NETIF_F_RXALL)
8870 netdev->features = features;
8872 if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
8873 if (features & NETIF_F_RXCSUM) {
8874 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
8876 u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
8878 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
8882 if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
8883 if (features & NETIF_F_RXCSUM) {
8884 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
8886 u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
8888 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
8893 ixgbe_do_reset(netdev);
8894 else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
8895 NETIF_F_HW_VLAN_CTAG_FILTER))
8896 ixgbe_set_rx_mode(netdev);
8902 * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
8903 * @dev: The port's netdev
8904 * @ti: Tunnel endpoint information
8906 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
8907 struct udp_tunnel_info *ti)
8909 struct ixgbe_adapter *adapter = netdev_priv(dev);
8910 struct ixgbe_hw *hw = &adapter->hw;
8911 __be16 port = ti->port;
8915 if (ti->sa_family != AF_INET)
8919 case UDP_TUNNEL_TYPE_VXLAN:
8920 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8923 if (adapter->vxlan_port == port)
8926 if (adapter->vxlan_port) {
8928 "VXLAN port %d set, not adding port %d\n",
8929 ntohs(adapter->vxlan_port),
8934 adapter->vxlan_port = port;
8936 case UDP_TUNNEL_TYPE_GENEVE:
8937 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
8940 if (adapter->geneve_port == port)
8943 if (adapter->geneve_port) {
8945 "GENEVE port %d set, not adding port %d\n",
8946 ntohs(adapter->geneve_port),
8951 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
8952 adapter->geneve_port = port;
8958 reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
8959 IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
8963 * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
8964 * @dev: The port's netdev
8965 * @ti: Tunnel endpoint information
8967 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
8968 struct udp_tunnel_info *ti)
8970 struct ixgbe_adapter *adapter = netdev_priv(dev);
8973 if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
8974 ti->type != UDP_TUNNEL_TYPE_GENEVE)
8977 if (ti->sa_family != AF_INET)
8981 case UDP_TUNNEL_TYPE_VXLAN:
8982 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8985 if (adapter->vxlan_port != ti->port) {
8986 netdev_info(dev, "VXLAN port %d not found\n",
8991 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
8993 case UDP_TUNNEL_TYPE_GENEVE:
8994 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
8997 if (adapter->geneve_port != ti->port) {
8998 netdev_info(dev, "GENEVE port %d not found\n",
9003 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9009 ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9010 adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9013 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9014 struct net_device *dev,
9015 const unsigned char *addr, u16 vid,
9018 /* guarantee we can provide a unique filter for the unicast address */
9019 if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9020 struct ixgbe_adapter *adapter = netdev_priv(dev);
9021 u16 pool = VMDQ_P(0);
9023 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9027 return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9031 * ixgbe_configure_bridge_mode - set various bridge modes
9032 * @adapter - the private structure
9033 * @mode - requested bridge mode
9035 * Configure some settings require for various bridge modes.
9037 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9040 struct ixgbe_hw *hw = &adapter->hw;
9041 unsigned int p, num_pools;
9045 case BRIDGE_MODE_VEPA:
9046 /* disable Tx loopback, rely on switch hairpin mode */
9047 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9049 /* must enable Rx switching replication to allow multicast
9050 * packet reception on all VFs, and to enable source address
9053 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9054 vmdctl |= IXGBE_VT_CTL_REPLEN;
9055 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9057 /* enable Rx source address pruning. Note, this requires
9058 * replication to be enabled or else it does nothing.
9060 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9061 for (p = 0; p < num_pools; p++) {
9062 if (hw->mac.ops.set_source_address_pruning)
9063 hw->mac.ops.set_source_address_pruning(hw,
9068 case BRIDGE_MODE_VEB:
9069 /* enable Tx loopback for internal VF/PF communication */
9070 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9071 IXGBE_PFDTXGSWC_VT_LBEN);
9073 /* disable Rx switching replication unless we have SR-IOV
9076 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9077 if (!adapter->num_vfs)
9078 vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9079 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9081 /* disable Rx source address pruning, since we don't expect to
9082 * be receiving external loopback of our transmitted frames.
9084 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9085 for (p = 0; p < num_pools; p++) {
9086 if (hw->mac.ops.set_source_address_pruning)
9087 hw->mac.ops.set_source_address_pruning(hw,
9096 adapter->bridge_mode = mode;
9098 e_info(drv, "enabling bridge mode: %s\n",
9099 mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9104 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9105 struct nlmsghdr *nlh, u16 flags)
9107 struct ixgbe_adapter *adapter = netdev_priv(dev);
9108 struct nlattr *attr, *br_spec;
9111 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9114 br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9118 nla_for_each_nested(attr, br_spec, rem) {
9122 if (nla_type(attr) != IFLA_BRIDGE_MODE)
9125 if (nla_len(attr) < sizeof(mode))
9128 mode = nla_get_u16(attr);
9129 status = ixgbe_configure_bridge_mode(adapter, mode);
9139 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
9140 struct net_device *dev,
9141 u32 filter_mask, int nlflags)
9143 struct ixgbe_adapter *adapter = netdev_priv(dev);
9145 if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9148 return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
9149 adapter->bridge_mode, 0, 0, nlflags,
9153 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
9155 struct ixgbe_fwd_adapter *fwd_adapter = NULL;
9156 struct ixgbe_adapter *adapter = netdev_priv(pdev);
9157 int used_pools = adapter->num_vfs + adapter->num_rx_pools;
9161 /* Hardware has a limited number of available pools. Each VF, and the
9162 * PF require a pool. Check to ensure we don't attempt to use more
9163 * then the available number of pools.
9165 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
9166 return ERR_PTR(-EINVAL);
9169 if (vdev->num_rx_queues != vdev->num_tx_queues) {
9170 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
9172 return ERR_PTR(-EINVAL);
9175 /* Check for hardware restriction on number of rx/tx queues */
9176 if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
9177 vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
9179 "%s: Supports RX/TX Queue counts 1,2, and 4\n",
9181 return ERR_PTR(-EINVAL);
9184 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
9185 adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
9186 (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
9187 return ERR_PTR(-EBUSY);
9189 fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
9191 return ERR_PTR(-ENOMEM);
9193 pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
9194 adapter->num_rx_pools++;
9195 set_bit(pool, &adapter->fwd_bitmask);
9196 limit = find_last_bit(&adapter->fwd_bitmask, 32);
9198 /* Enable VMDq flag so device will be set in VM mode */
9199 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
9200 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
9201 adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
9203 /* Force reinit of ring allocation with VMDQ enabled */
9204 err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
9207 fwd_adapter->pool = pool;
9208 fwd_adapter->real_adapter = adapter;
9210 if (netif_running(pdev)) {
9211 err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
9214 netif_tx_start_all_queues(vdev);
9219 /* unwind counter and free adapter struct */
9221 "%s: dfwd hardware acceleration failed\n", vdev->name);
9222 clear_bit(pool, &adapter->fwd_bitmask);
9223 adapter->num_rx_pools--;
9225 return ERR_PTR(err);
9228 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
9230 struct ixgbe_fwd_adapter *fwd_adapter = priv;
9231 struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
9234 clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
9235 adapter->num_rx_pools--;
9237 limit = find_last_bit(&adapter->fwd_bitmask, 32);
9238 adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
9239 ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
9240 ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
9241 netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
9242 fwd_adapter->pool, adapter->num_rx_pools,
9243 fwd_adapter->rx_base_queue,
9244 fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
9245 adapter->fwd_bitmask);
9249 #define IXGBE_MAX_MAC_HDR_LEN 127
9250 #define IXGBE_MAX_NETWORK_HDR_LEN 511
9252 static netdev_features_t
9253 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
9254 netdev_features_t features)
9256 unsigned int network_hdr_len, mac_hdr_len;
9258 /* Make certain the headers can be described by a context descriptor */
9259 mac_hdr_len = skb_network_header(skb) - skb->data;
9260 if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
9261 return features & ~(NETIF_F_HW_CSUM |
9263 NETIF_F_HW_VLAN_CTAG_TX |
9267 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
9268 if (unlikely(network_hdr_len > IXGBE_MAX_NETWORK_HDR_LEN))
9269 return features & ~(NETIF_F_HW_CSUM |
9274 /* We can only support IPV4 TSO in tunnels if we can mangle the
9275 * inner IP ID field, so strip TSO if MANGLEID is not supported.
9277 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
9278 features &= ~NETIF_F_TSO;
9283 static const struct net_device_ops ixgbe_netdev_ops = {
9284 .ndo_open = ixgbe_open,
9285 .ndo_stop = ixgbe_close,
9286 .ndo_start_xmit = ixgbe_xmit_frame,
9287 .ndo_select_queue = ixgbe_select_queue,
9288 .ndo_set_rx_mode = ixgbe_set_rx_mode,
9289 .ndo_validate_addr = eth_validate_addr,
9290 .ndo_set_mac_address = ixgbe_set_mac,
9291 .ndo_change_mtu = ixgbe_change_mtu,
9292 .ndo_tx_timeout = ixgbe_tx_timeout,
9293 .ndo_set_tx_maxrate = ixgbe_tx_maxrate,
9294 .ndo_vlan_rx_add_vid = ixgbe_vlan_rx_add_vid,
9295 .ndo_vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid,
9296 .ndo_do_ioctl = ixgbe_ioctl,
9297 .ndo_set_vf_mac = ixgbe_ndo_set_vf_mac,
9298 .ndo_set_vf_vlan = ixgbe_ndo_set_vf_vlan,
9299 .ndo_set_vf_rate = ixgbe_ndo_set_vf_bw,
9300 .ndo_set_vf_spoofchk = ixgbe_ndo_set_vf_spoofchk,
9301 .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
9302 .ndo_set_vf_trust = ixgbe_ndo_set_vf_trust,
9303 .ndo_get_vf_config = ixgbe_ndo_get_vf_config,
9304 .ndo_get_stats64 = ixgbe_get_stats64,
9305 .ndo_setup_tc = __ixgbe_setup_tc,
9306 #ifdef CONFIG_NET_POLL_CONTROLLER
9307 .ndo_poll_controller = ixgbe_netpoll,
9310 .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
9311 .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
9312 .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
9313 .ndo_fcoe_enable = ixgbe_fcoe_enable,
9314 .ndo_fcoe_disable = ixgbe_fcoe_disable,
9315 .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
9316 .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
9317 #endif /* IXGBE_FCOE */
9318 .ndo_set_features = ixgbe_set_features,
9319 .ndo_fix_features = ixgbe_fix_features,
9320 .ndo_fdb_add = ixgbe_ndo_fdb_add,
9321 .ndo_bridge_setlink = ixgbe_ndo_bridge_setlink,
9322 .ndo_bridge_getlink = ixgbe_ndo_bridge_getlink,
9323 .ndo_dfwd_add_station = ixgbe_fwd_add,
9324 .ndo_dfwd_del_station = ixgbe_fwd_del,
9325 .ndo_udp_tunnel_add = ixgbe_add_udp_tunnel_port,
9326 .ndo_udp_tunnel_del = ixgbe_del_udp_tunnel_port,
9327 .ndo_features_check = ixgbe_features_check,
9331 * ixgbe_enumerate_functions - Get the number of ports this device has
9332 * @adapter: adapter structure
9334 * This function enumerates the phsyical functions co-located on a single slot,
9335 * in order to determine how many ports a device has. This is most useful in
9336 * determining the required GT/s of PCIe bandwidth necessary for optimal
9339 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
9341 struct pci_dev *entry, *pdev = adapter->pdev;
9344 /* Some cards can not use the generic count PCIe functions method,
9345 * because they are behind a parent switch, so we hardcode these with
9346 * the correct number of functions.
9348 if (ixgbe_pcie_from_parent(&adapter->hw))
9351 list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
9352 /* don't count virtual functions */
9353 if (entry->is_virtfn)
9356 /* When the devices on the bus don't all match our device ID,
9357 * we can't reliably determine the correct number of
9358 * functions. This can occur if a function has been direct
9359 * attached to a virtual machine using VT-d, for example. In
9360 * this case, simply return -1 to indicate this.
9362 if ((entry->vendor != pdev->vendor) ||
9363 (entry->device != pdev->device))
9373 * ixgbe_wol_supported - Check whether device supports WoL
9374 * @adapter: the adapter private structure
9375 * @device_id: the device ID
9376 * @subdev_id: the subsystem device ID
9378 * This function is used by probe and ethtool to determine
9379 * which devices have WoL support
9382 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
9385 struct ixgbe_hw *hw = &adapter->hw;
9386 u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
9388 /* WOL not supported on 82598 */
9389 if (hw->mac.type == ixgbe_mac_82598EB)
9392 /* check eeprom to see if WOL is enabled for X540 and newer */
9393 if (hw->mac.type >= ixgbe_mac_X540) {
9394 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
9395 ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
9396 (hw->bus.func == 0)))
9400 /* WOL is determined based on device IDs for 82599 MACs */
9401 switch (device_id) {
9402 case IXGBE_DEV_ID_82599_SFP:
9403 /* Only these subdevices could supports WOL */
9404 switch (subdevice_id) {
9405 case IXGBE_SUBDEV_ID_82599_560FLR:
9406 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
9407 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
9408 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
9409 /* only support first port */
9410 if (hw->bus.func != 0)
9412 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
9413 case IXGBE_SUBDEV_ID_82599_SFP:
9414 case IXGBE_SUBDEV_ID_82599_RNDC:
9415 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
9416 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
9417 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
9418 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
9422 case IXGBE_DEV_ID_82599EN_SFP:
9423 /* Only these subdevices support WOL */
9424 switch (subdevice_id) {
9425 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
9429 case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
9430 /* All except this subdevice support WOL */
9431 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
9434 case IXGBE_DEV_ID_82599_KX4:
9444 * ixgbe_probe - Device Initialization Routine
9445 * @pdev: PCI device information struct
9446 * @ent: entry in ixgbe_pci_tbl
9448 * Returns 0 on success, negative on failure
9450 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
9451 * The OS initialization, configuring of the adapter private structure,
9452 * and a hardware reset occur.
9454 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
9456 struct net_device *netdev;
9457 struct ixgbe_adapter *adapter = NULL;
9458 struct ixgbe_hw *hw;
9459 const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
9460 int i, err, pci_using_dac, expected_gts;
9461 unsigned int indices = MAX_TX_QUEUES;
9462 u8 part_str[IXGBE_PBANUM_LENGTH];
9463 bool disable_dev = false;
9469 /* Catch broken hardware that put the wrong VF device ID in
9470 * the PCIe SR-IOV capability.
9472 if (pdev->is_virtfn) {
9473 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
9474 pci_name(pdev), pdev->vendor, pdev->device);
9478 err = pci_enable_device_mem(pdev);
9482 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
9485 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
9488 "No usable DMA configuration, aborting\n");
9494 err = pci_request_mem_regions(pdev, ixgbe_driver_name);
9497 "pci_request_selected_regions failed 0x%x\n", err);
9501 pci_enable_pcie_error_reporting(pdev);
9503 pci_set_master(pdev);
9504 pci_save_state(pdev);
9506 if (ii->mac == ixgbe_mac_82598EB) {
9507 #ifdef CONFIG_IXGBE_DCB
9508 /* 8 TC w/ 4 queues per TC */
9509 indices = 4 * MAX_TRAFFIC_CLASS;
9511 indices = IXGBE_MAX_RSS_INDICES;
9515 netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
9518 goto err_alloc_etherdev;
9521 SET_NETDEV_DEV(netdev, &pdev->dev);
9523 adapter = netdev_priv(netdev);
9525 adapter->netdev = netdev;
9526 adapter->pdev = pdev;
9529 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
9531 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
9532 pci_resource_len(pdev, 0));
9533 adapter->io_addr = hw->hw_addr;
9539 netdev->netdev_ops = &ixgbe_netdev_ops;
9540 ixgbe_set_ethtool_ops(netdev);
9541 netdev->watchdog_timeo = 5 * HZ;
9542 strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
9545 hw->mac.ops = *ii->mac_ops;
9546 hw->mac.type = ii->mac;
9547 hw->mvals = ii->mvals;
9549 hw->link.ops = *ii->link_ops;
9552 hw->eeprom.ops = *ii->eeprom_ops;
9553 eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
9554 if (ixgbe_removed(hw->hw_addr)) {
9558 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
9559 if (!(eec & BIT(8)))
9560 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
9563 hw->phy.ops = *ii->phy_ops;
9564 hw->phy.sfp_type = ixgbe_sfp_type_unknown;
9565 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
9566 hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
9567 hw->phy.mdio.mmds = 0;
9568 hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
9569 hw->phy.mdio.dev = netdev;
9570 hw->phy.mdio.mdio_read = ixgbe_mdio_read;
9571 hw->phy.mdio.mdio_write = ixgbe_mdio_write;
9573 /* setup the private structure */
9574 err = ixgbe_sw_init(adapter, ii);
9578 /* Make sure the SWFW semaphore is in a valid state */
9579 if (hw->mac.ops.init_swfw_sync)
9580 hw->mac.ops.init_swfw_sync(hw);
9582 /* Make it possible the adapter to be woken up via WOL */
9583 switch (adapter->hw.mac.type) {
9584 case ixgbe_mac_82599EB:
9585 case ixgbe_mac_X540:
9586 case ixgbe_mac_X550:
9587 case ixgbe_mac_X550EM_x:
9588 case ixgbe_mac_x550em_a:
9589 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9596 * If there is a fan on this device and it has failed log the
9599 if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
9600 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
9601 if (esdp & IXGBE_ESDP_SDP1)
9602 e_crit(probe, "Fan has stopped, replace the adapter\n");
9605 if (allow_unsupported_sfp)
9606 hw->allow_unsupported_sfp = allow_unsupported_sfp;
9608 /* reset_hw fills in the perm_addr as well */
9609 hw->phy.reset_if_overtemp = true;
9610 err = hw->mac.ops.reset_hw(hw);
9611 hw->phy.reset_if_overtemp = false;
9612 ixgbe_set_eee_capable(adapter);
9613 if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
9615 } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
9616 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
9617 e_dev_err("Reload the driver after installing a supported module.\n");
9620 e_dev_err("HW Init failed: %d\n", err);
9624 #ifdef CONFIG_PCI_IOV
9625 /* SR-IOV not supported on the 82598 */
9626 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9629 ixgbe_init_mbx_params_pf(hw);
9630 hw->mbx.ops = ii->mbx_ops;
9631 pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
9632 ixgbe_enable_sriov(adapter);
9636 netdev->features = NETIF_F_SG |
9643 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
9644 NETIF_F_GSO_GRE_CSUM | \
9645 NETIF_F_GSO_IPXIP4 | \
9646 NETIF_F_GSO_IPXIP6 | \
9647 NETIF_F_GSO_UDP_TUNNEL | \
9648 NETIF_F_GSO_UDP_TUNNEL_CSUM)
9650 netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
9651 netdev->features |= NETIF_F_GSO_PARTIAL |
9652 IXGBE_GSO_PARTIAL_FEATURES;
9654 if (hw->mac.type >= ixgbe_mac_82599EB)
9655 netdev->features |= NETIF_F_SCTP_CRC;
9657 /* copy netdev features into list of user selectable features */
9658 netdev->hw_features |= netdev->features |
9659 NETIF_F_HW_VLAN_CTAG_FILTER |
9660 NETIF_F_HW_VLAN_CTAG_RX |
9661 NETIF_F_HW_VLAN_CTAG_TX |
9663 NETIF_F_HW_L2FW_DOFFLOAD;
9665 if (hw->mac.type >= ixgbe_mac_82599EB)
9666 netdev->hw_features |= NETIF_F_NTUPLE |
9670 netdev->features |= NETIF_F_HIGHDMA;
9672 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
9673 netdev->hw_enc_features |= netdev->vlan_features;
9674 netdev->mpls_features |= NETIF_F_HW_CSUM;
9676 /* set this bit last since it cannot be part of vlan_features */
9677 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
9678 NETIF_F_HW_VLAN_CTAG_RX |
9679 NETIF_F_HW_VLAN_CTAG_TX;
9681 netdev->priv_flags |= IFF_UNICAST_FLT;
9682 netdev->priv_flags |= IFF_SUPP_NOFCS;
9684 /* MTU range: 68 - 9710 */
9685 netdev->min_mtu = ETH_MIN_MTU;
9686 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
9688 #ifdef CONFIG_IXGBE_DCB
9689 if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
9690 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
9694 if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
9695 unsigned int fcoe_l;
9697 if (hw->mac.ops.get_device_caps) {
9698 hw->mac.ops.get_device_caps(hw, &device_caps);
9699 if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
9700 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
9704 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
9705 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
9707 netdev->features |= NETIF_F_FSO |
9710 netdev->vlan_features |= NETIF_F_FSO |
9714 #endif /* IXGBE_FCOE */
9716 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
9717 netdev->hw_features |= NETIF_F_LRO;
9718 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9719 netdev->features |= NETIF_F_LRO;
9721 /* make sure the EEPROM is good */
9722 if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
9723 e_dev_err("The EEPROM Checksum Is Not Valid\n");
9728 eth_platform_get_mac_address(&adapter->pdev->dev,
9729 adapter->hw.mac.perm_addr);
9731 memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
9733 if (!is_valid_ether_addr(netdev->dev_addr)) {
9734 e_dev_err("invalid MAC address\n");
9739 /* Set hw->mac.addr to permanent MAC address */
9740 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
9741 ixgbe_mac_set_default_filter(adapter);
9743 setup_timer(&adapter->service_timer, &ixgbe_service_timer,
9744 (unsigned long) adapter);
9746 if (ixgbe_removed(hw->hw_addr)) {
9750 INIT_WORK(&adapter->service_task, ixgbe_service_task);
9751 set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
9752 clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
9754 err = ixgbe_init_interrupt_scheme(adapter);
9758 /* WOL not supported for all devices */
9760 hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
9761 hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
9762 pdev->subsystem_device);
9763 if (hw->wol_enabled)
9764 adapter->wol = IXGBE_WUFC_MAG;
9766 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
9768 /* save off EEPROM version number */
9769 hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
9770 hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
9772 /* pick up the PCI bus settings for reporting later */
9773 if (ixgbe_pcie_from_parent(hw))
9774 ixgbe_get_parent_bus_info(adapter);
9776 hw->mac.ops.get_bus_info(hw);
9778 /* calculate the expected PCIe bandwidth required for optimal
9779 * performance. Note that some older parts will never have enough
9780 * bandwidth due to being older generation PCIe parts. We clamp these
9781 * parts to ensure no warning is displayed if it can't be fixed.
9783 switch (hw->mac.type) {
9784 case ixgbe_mac_82598EB:
9785 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
9788 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
9792 /* don't check link if we failed to enumerate functions */
9793 if (expected_gts > 0)
9794 ixgbe_check_minimum_link(adapter, expected_gts);
9796 err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
9798 strlcpy(part_str, "Unknown", sizeof(part_str));
9799 if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
9800 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
9801 hw->mac.type, hw->phy.type, hw->phy.sfp_type,
9804 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
9805 hw->mac.type, hw->phy.type, part_str);
9807 e_dev_info("%pM\n", netdev->dev_addr);
9809 /* reset the hardware with the new settings */
9810 err = hw->mac.ops.start_hw(hw);
9811 if (err == IXGBE_ERR_EEPROM_VERSION) {
9812 /* We are running on a pre-production device, log a warning */
9813 e_dev_warn("This device is a pre-production adapter/LOM. "
9814 "Please be aware there may be issues associated "
9815 "with your hardware. If you are experiencing "
9816 "problems please contact your Intel or hardware "
9817 "representative who provided you with this "
9820 strcpy(netdev->name, "eth%d");
9821 err = register_netdev(netdev);
9825 pci_set_drvdata(pdev, adapter);
9827 /* power down the optics for 82599 SFP+ fiber */
9828 if (hw->mac.ops.disable_tx_laser)
9829 hw->mac.ops.disable_tx_laser(hw);
9831 /* carrier off reporting is important to ethtool even BEFORE open */
9832 netif_carrier_off(netdev);
9834 #ifdef CONFIG_IXGBE_DCA
9835 if (dca_add_requester(&pdev->dev) == 0) {
9836 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
9837 ixgbe_setup_dca(adapter);
9840 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
9841 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
9842 for (i = 0; i < adapter->num_vfs; i++)
9843 ixgbe_vf_configuration(pdev, (i | 0x10000000));
9846 /* firmware requires driver version to be 0xFFFFFFFF
9847 * since os does not support feature
9849 if (hw->mac.ops.set_fw_drv_ver)
9850 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
9851 sizeof(ixgbe_driver_version) - 1,
9852 ixgbe_driver_version);
9854 /* add san mac addr to netdev */
9855 ixgbe_add_sanmac_netdev(netdev);
9857 e_dev_info("%s\n", ixgbe_default_device_descr);
9859 #ifdef CONFIG_IXGBE_HWMON
9860 if (ixgbe_sysfs_init(adapter))
9861 e_err(probe, "failed to allocate sysfs resources\n");
9862 #endif /* CONFIG_IXGBE_HWMON */
9864 ixgbe_dbg_adapter_init(adapter);
9866 /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
9867 if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
9868 hw->mac.ops.setup_link(hw,
9869 IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
9875 ixgbe_release_hw_control(adapter);
9876 ixgbe_clear_interrupt_scheme(adapter);
9878 ixgbe_disable_sriov(adapter);
9879 adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
9880 iounmap(adapter->io_addr);
9881 kfree(adapter->jump_tables[0]);
9882 kfree(adapter->mac_table);
9884 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9885 free_netdev(netdev);
9887 pci_release_mem_regions(pdev);
9890 if (!adapter || disable_dev)
9891 pci_disable_device(pdev);
9896 * ixgbe_remove - Device Removal Routine
9897 * @pdev: PCI device information struct
9899 * ixgbe_remove is called by the PCI subsystem to alert the driver
9900 * that it should release a PCI device. The could be caused by a
9901 * Hot-Plug event, or because the driver is going to be removed from
9904 static void ixgbe_remove(struct pci_dev *pdev)
9906 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9907 struct net_device *netdev;
9911 /* if !adapter then we already cleaned up in probe */
9915 netdev = adapter->netdev;
9916 ixgbe_dbg_adapter_exit(adapter);
9918 set_bit(__IXGBE_REMOVING, &adapter->state);
9919 cancel_work_sync(&adapter->service_task);
9922 #ifdef CONFIG_IXGBE_DCA
9923 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
9924 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
9925 dca_remove_requester(&pdev->dev);
9926 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
9927 IXGBE_DCA_CTRL_DCA_DISABLE);
9931 #ifdef CONFIG_IXGBE_HWMON
9932 ixgbe_sysfs_exit(adapter);
9933 #endif /* CONFIG_IXGBE_HWMON */
9935 /* remove the added san mac */
9936 ixgbe_del_sanmac_netdev(netdev);
9938 #ifdef CONFIG_PCI_IOV
9939 ixgbe_disable_sriov(adapter);
9941 if (netdev->reg_state == NETREG_REGISTERED)
9942 unregister_netdev(netdev);
9944 ixgbe_clear_interrupt_scheme(adapter);
9946 ixgbe_release_hw_control(adapter);
9949 kfree(adapter->ixgbe_ieee_pfc);
9950 kfree(adapter->ixgbe_ieee_ets);
9953 iounmap(adapter->io_addr);
9954 pci_release_mem_regions(pdev);
9956 e_dev_info("complete\n");
9958 for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
9959 if (adapter->jump_tables[i]) {
9960 kfree(adapter->jump_tables[i]->input);
9961 kfree(adapter->jump_tables[i]->mask);
9963 kfree(adapter->jump_tables[i]);
9966 kfree(adapter->mac_table);
9967 disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9968 free_netdev(netdev);
9970 pci_disable_pcie_error_reporting(pdev);
9973 pci_disable_device(pdev);
9977 * ixgbe_io_error_detected - called when PCI error is detected
9978 * @pdev: Pointer to PCI device
9979 * @state: The current pci connection state
9981 * This function is called after a PCI bus error affecting
9982 * this device has been detected.
9984 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
9985 pci_channel_state_t state)
9987 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9988 struct net_device *netdev = adapter->netdev;
9990 #ifdef CONFIG_PCI_IOV
9991 struct ixgbe_hw *hw = &adapter->hw;
9992 struct pci_dev *bdev, *vfdev;
9993 u32 dw0, dw1, dw2, dw3;
9995 u16 req_id, pf_func;
9997 if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
9998 adapter->num_vfs == 0)
9999 goto skip_bad_vf_detection;
10001 bdev = pdev->bus->self;
10002 while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
10003 bdev = bdev->bus->self;
10006 goto skip_bad_vf_detection;
10008 pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
10010 goto skip_bad_vf_detection;
10012 dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
10013 dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
10014 dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
10015 dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
10016 if (ixgbe_removed(hw->hw_addr))
10017 goto skip_bad_vf_detection;
10019 req_id = dw1 >> 16;
10020 /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
10021 if (!(req_id & 0x0080))
10022 goto skip_bad_vf_detection;
10024 pf_func = req_id & 0x01;
10025 if ((pf_func & 1) == (pdev->devfn & 1)) {
10026 unsigned int device_id;
10028 vf = (req_id & 0x7F) >> 1;
10029 e_dev_err("VF %d has caused a PCIe error\n", vf);
10030 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
10031 "%8.8x\tdw3: %8.8x\n",
10032 dw0, dw1, dw2, dw3);
10033 switch (adapter->hw.mac.type) {
10034 case ixgbe_mac_82599EB:
10035 device_id = IXGBE_82599_VF_DEVICE_ID;
10037 case ixgbe_mac_X540:
10038 device_id = IXGBE_X540_VF_DEVICE_ID;
10040 case ixgbe_mac_X550:
10041 device_id = IXGBE_DEV_ID_X550_VF;
10043 case ixgbe_mac_X550EM_x:
10044 device_id = IXGBE_DEV_ID_X550EM_X_VF;
10046 case ixgbe_mac_x550em_a:
10047 device_id = IXGBE_DEV_ID_X550EM_A_VF;
10054 /* Find the pci device of the offending VF */
10055 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
10057 if (vfdev->devfn == (req_id & 0xFF))
10059 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
10063 * There's a slim chance the VF could have been hot plugged,
10064 * so if it is no longer present we don't need to issue the
10065 * VFLR. Just clean up the AER in that case.
10068 ixgbe_issue_vf_flr(adapter, vfdev);
10069 /* Free device reference count */
10070 pci_dev_put(vfdev);
10073 pci_cleanup_aer_uncorrect_error_status(pdev);
10077 * Even though the error may have occurred on the other port
10078 * we still need to increment the vf error reference count for
10079 * both ports because the I/O resume function will be called
10080 * for both of them.
10082 adapter->vferr_refcount++;
10084 return PCI_ERS_RESULT_RECOVERED;
10086 skip_bad_vf_detection:
10087 #endif /* CONFIG_PCI_IOV */
10088 if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
10089 return PCI_ERS_RESULT_DISCONNECT;
10092 netif_device_detach(netdev);
10094 if (state == pci_channel_io_perm_failure) {
10096 return PCI_ERS_RESULT_DISCONNECT;
10099 if (netif_running(netdev))
10100 ixgbe_close_suspend(adapter);
10102 if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
10103 pci_disable_device(pdev);
10106 /* Request a slot reset. */
10107 return PCI_ERS_RESULT_NEED_RESET;
10111 * ixgbe_io_slot_reset - called after the pci bus has been reset.
10112 * @pdev: Pointer to PCI device
10114 * Restart the card from scratch, as if from a cold-boot.
10116 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
10118 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
10119 pci_ers_result_t result;
10122 if (pci_enable_device_mem(pdev)) {
10123 e_err(probe, "Cannot re-enable PCI device after reset.\n");
10124 result = PCI_ERS_RESULT_DISCONNECT;
10126 smp_mb__before_atomic();
10127 clear_bit(__IXGBE_DISABLED, &adapter->state);
10128 adapter->hw.hw_addr = adapter->io_addr;
10129 pci_set_master(pdev);
10130 pci_restore_state(pdev);
10131 pci_save_state(pdev);
10133 pci_wake_from_d3(pdev, false);
10135 ixgbe_reset(adapter);
10136 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10137 result = PCI_ERS_RESULT_RECOVERED;
10140 err = pci_cleanup_aer_uncorrect_error_status(pdev);
10142 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
10143 "failed 0x%0x\n", err);
10144 /* non-fatal, continue */
10151 * ixgbe_io_resume - called when traffic can start flowing again.
10152 * @pdev: Pointer to PCI device
10154 * This callback is called when the error recovery driver tells us that
10155 * its OK to resume normal operation.
10157 static void ixgbe_io_resume(struct pci_dev *pdev)
10159 struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
10160 struct net_device *netdev = adapter->netdev;
10162 #ifdef CONFIG_PCI_IOV
10163 if (adapter->vferr_refcount) {
10164 e_info(drv, "Resuming after VF err\n");
10165 adapter->vferr_refcount--;
10171 if (netif_running(netdev))
10172 ixgbe_open(netdev);
10174 netif_device_attach(netdev);
10178 static const struct pci_error_handlers ixgbe_err_handler = {
10179 .error_detected = ixgbe_io_error_detected,
10180 .slot_reset = ixgbe_io_slot_reset,
10181 .resume = ixgbe_io_resume,
10184 static struct pci_driver ixgbe_driver = {
10185 .name = ixgbe_driver_name,
10186 .id_table = ixgbe_pci_tbl,
10187 .probe = ixgbe_probe,
10188 .remove = ixgbe_remove,
10190 .suspend = ixgbe_suspend,
10191 .resume = ixgbe_resume,
10193 .shutdown = ixgbe_shutdown,
10194 .sriov_configure = ixgbe_pci_sriov_configure,
10195 .err_handler = &ixgbe_err_handler
10199 * ixgbe_init_module - Driver Registration Routine
10201 * ixgbe_init_module is the first routine called when the driver is
10202 * loaded. All it does is register with the PCI subsystem.
10204 static int __init ixgbe_init_module(void)
10207 pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
10208 pr_info("%s\n", ixgbe_copyright);
10210 ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
10212 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
10218 ret = pci_register_driver(&ixgbe_driver);
10220 destroy_workqueue(ixgbe_wq);
10225 #ifdef CONFIG_IXGBE_DCA
10226 dca_register_notify(&dca_notifier);
10232 module_init(ixgbe_init_module);
10235 * ixgbe_exit_module - Driver Exit Cleanup Routine
10237 * ixgbe_exit_module is called just before the driver is removed
10240 static void __exit ixgbe_exit_module(void)
10242 #ifdef CONFIG_IXGBE_DCA
10243 dca_unregister_notify(&dca_notifier);
10245 pci_unregister_driver(&ixgbe_driver);
10249 destroy_workqueue(ixgbe_wq);
10254 #ifdef CONFIG_IXGBE_DCA
10255 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
10260 ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
10261 __ixgbe_notify_dca);
10263 return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
10266 #endif /* CONFIG_IXGBE_DCA */
10268 module_exit(ixgbe_exit_module);