1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
3 
4 #include <linux/types.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/netdevice.h>
8 #include <linux/vmalloc.h>
9 #include <linux/string.h>
10 #include <linux/in.h>
11 #include <linux/interrupt.h>
12 #include <linux/ip.h>
13 #include <linux/tcp.h>
14 #include <linux/sctp.h>
15 #include <linux/pkt_sched.h>
16 #include <linux/ipv6.h>
17 #include <linux/slab.h>
18 #include <net/checksum.h>
19 #include <net/ip6_checksum.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/if.h>
23 #include <linux/if_vlan.h>
24 #include <linux/if_macvlan.h>
25 #include <linux/if_bridge.h>
26 #include <linux/prefetch.h>
27 #include <linux/bpf.h>
28 #include <linux/bpf_trace.h>
29 #include <linux/atomic.h>
30 #include <scsi/fc/fc_fcoe.h>
31 #include <net/udp_tunnel.h>
32 #include <net/pkt_cls.h>
33 #include <net/tc_act/tc_gact.h>
34 #include <net/tc_act/tc_mirred.h>
35 #include <net/vxlan.h>
36 #include <net/mpls.h>
37 #include <net/xdp_sock.h>
38 
39 #include "ixgbe.h"
40 #include "ixgbe_common.h"
41 #include "ixgbe_dcb_82599.h"
42 #include "ixgbe_sriov.h"
43 #include "ixgbe_model.h"
44 #include "ixgbe_txrx_common.h"
45 
46 char ixgbe_driver_name[] = "ixgbe";
47 static const char ixgbe_driver_string[] =
48 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
49 #ifdef IXGBE_FCOE
50 char ixgbe_default_device_descr[] =
51 			      "Intel(R) 10 Gigabit Network Connection";
52 #else
53 static char ixgbe_default_device_descr[] =
54 			      "Intel(R) 10 Gigabit Network Connection";
55 #endif
56 #define DRV_VERSION "5.1.0-k"
57 const char ixgbe_driver_version[] = DRV_VERSION;
58 static const char ixgbe_copyright[] =
59 				"Copyright (c) 1999-2016 Intel Corporation.";
60 
61 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";
62 
63 static const struct ixgbe_info *ixgbe_info_tbl[] = {
64 	[board_82598]		= &ixgbe_82598_info,
65 	[board_82599]		= &ixgbe_82599_info,
66 	[board_X540]		= &ixgbe_X540_info,
67 	[board_X550]		= &ixgbe_X550_info,
68 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
69 	[board_x550em_x_fw]	= &ixgbe_x550em_x_fw_info,
70 	[board_x550em_a]	= &ixgbe_x550em_a_info,
71 	[board_x550em_a_fw]	= &ixgbe_x550em_a_fw_info,
72 };
73 
74 /* ixgbe_pci_tbl - PCI Device ID Table
75  *
76  * Wildcard entries (PCI_ANY_ID) should come last
77  * Last entry must be all 0s
78  *
79  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
80  *   Class, Class Mask, private data (not used) }
81  */
82 static const struct pci_device_id ixgbe_pci_tbl[] = {
83 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
84 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
130 	/* required last entry */
131 	{0, }
132 };
133 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
134 
135 #ifdef CONFIG_IXGBE_DCA
136 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
137 			    void *p);
138 static struct notifier_block dca_notifier = {
139 	.notifier_call = ixgbe_notify_dca,
140 	.next          = NULL,
141 	.priority      = 0
142 };
143 #endif
144 
145 #ifdef CONFIG_PCI_IOV
146 static unsigned int max_vfs;
147 module_param(max_vfs, uint, 0);
148 MODULE_PARM_DESC(max_vfs,
149 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
150 #endif /* CONFIG_PCI_IOV */
151 
152 static unsigned int allow_unsupported_sfp;
153 module_param(allow_unsupported_sfp, uint, 0);
154 MODULE_PARM_DESC(allow_unsupported_sfp,
155 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
156 
157 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
158 static int debug = -1;
159 module_param(debug, int, 0);
160 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
161 
162 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
163 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
164 MODULE_LICENSE("GPL v2");
165 MODULE_VERSION(DRV_VERSION);
166 
167 static struct workqueue_struct *ixgbe_wq;
168 
169 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
170 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
171 
172 static const struct net_device_ops ixgbe_netdev_ops;
173 
174 static bool netif_is_ixgbe(struct net_device *dev)
175 {
176 	return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
177 }
178 
179 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
180 					  u32 reg, u16 *value)
181 {
182 	struct pci_dev *parent_dev;
183 	struct pci_bus *parent_bus;
184 
185 	parent_bus = adapter->pdev->bus->parent;
186 	if (!parent_bus)
187 		return -1;
188 
189 	parent_dev = parent_bus->self;
190 	if (!parent_dev)
191 		return -1;
192 
193 	if (!pci_is_pcie(parent_dev))
194 		return -1;
195 
196 	pcie_capability_read_word(parent_dev, reg, value);
197 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
198 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
199 		return -1;
200 	return 0;
201 }
202 
203 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
204 {
205 	struct ixgbe_hw *hw = &adapter->hw;
206 	u16 link_status = 0;
207 	int err;
208 
209 	hw->bus.type = ixgbe_bus_type_pci_express;
210 
211 	/* Get the negotiated link width and speed from PCI config space of the
212 	 * parent, as this device is behind a switch
213 	 */
214 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
215 
216 	/* assume caller will handle error case */
217 	if (err)
218 		return err;
219 
220 	hw->bus.width = ixgbe_convert_bus_width(link_status);
221 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
222 
223 	return 0;
224 }
225 
226 /**
227  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
228  * @hw: hw specific details
229  *
230  * This function is used by probe to determine whether a device's PCI-Express
231  * bandwidth details should be gathered from the parent bus instead of from the
232  * device. Used to ensure that various locations all have the correct device ID
233  * checks.
234  */
235 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
236 {
237 	switch (hw->device_id) {
238 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
239 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
240 		return true;
241 	default:
242 		return false;
243 	}
244 }
245 
246 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
247 				     int expected_gts)
248 {
249 	struct ixgbe_hw *hw = &adapter->hw;
250 	struct pci_dev *pdev;
251 
252 	/* Some devices are not connected over PCIe and thus do not negotiate
253 	 * speed. These devices do not have valid bus info, and thus any report
254 	 * we generate may not be correct.
255 	 */
256 	if (hw->bus.type == ixgbe_bus_type_internal)
257 		return;
258 
259 	/* determine whether to use the parent device */
260 	if (ixgbe_pcie_from_parent(&adapter->hw))
261 		pdev = adapter->pdev->bus->parent->self;
262 	else
263 		pdev = adapter->pdev;
264 
265 	pcie_print_link_status(pdev);
266 }
267 
268 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
269 {
270 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
271 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
272 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
273 		queue_work(ixgbe_wq, &adapter->service_task);
274 }
275 
276 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
277 {
278 	struct ixgbe_adapter *adapter = hw->back;
279 
280 	if (!hw->hw_addr)
281 		return;
282 	hw->hw_addr = NULL;
283 	e_dev_err("Adapter removed\n");
284 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
285 		ixgbe_service_event_schedule(adapter);
286 }
287 
288 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
289 {
290 	u8 __iomem *reg_addr;
291 	u32 value;
292 	int i;
293 
294 	reg_addr = READ_ONCE(hw->hw_addr);
295 	if (ixgbe_removed(reg_addr))
296 		return IXGBE_FAILED_READ_REG;
297 
298 	/* Register read of 0xFFFFFFF can indicate the adapter has been removed,
299 	 * so perform several status register reads to determine if the adapter
300 	 * has been removed.
301 	 */
302 	for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
303 		value = readl(reg_addr + IXGBE_STATUS);
304 		if (value != IXGBE_FAILED_READ_REG)
305 			break;
306 		mdelay(3);
307 	}
308 
309 	if (value == IXGBE_FAILED_READ_REG)
310 		ixgbe_remove_adapter(hw);
311 	else
312 		value = readl(reg_addr + reg);
313 	return value;
314 }
315 
316 /**
317  * ixgbe_read_reg - Read from device register
318  * @hw: hw specific details
319  * @reg: offset of register to read
320  *
321  * Returns : value read or IXGBE_FAILED_READ_REG if removed
322  *
323  * This function is used to read device registers. It checks for device
324  * removal by confirming any read that returns all ones by checking the
325  * status register value for all ones. This function avoids reading from
326  * the hardware if a removal was previously detected in which case it
327  * returns IXGBE_FAILED_READ_REG (all ones).
328  */
329 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
330 {
331 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
332 	u32 value;
333 
334 	if (ixgbe_removed(reg_addr))
335 		return IXGBE_FAILED_READ_REG;
336 	if (unlikely(hw->phy.nw_mng_if_sel &
337 		     IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
338 		struct ixgbe_adapter *adapter;
339 		int i;
340 
341 		for (i = 0; i < 200; ++i) {
342 			value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
343 			if (likely(!value))
344 				goto writes_completed;
345 			if (value == IXGBE_FAILED_READ_REG) {
346 				ixgbe_remove_adapter(hw);
347 				return IXGBE_FAILED_READ_REG;
348 			}
349 			udelay(5);
350 		}
351 
352 		adapter = hw->back;
353 		e_warn(hw, "register writes incomplete %08x\n", value);
354 	}
355 
356 writes_completed:
357 	value = readl(reg_addr + reg);
358 	if (unlikely(value == IXGBE_FAILED_READ_REG))
359 		value = ixgbe_check_remove(hw, reg);
360 	return value;
361 }
362 
363 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
364 {
365 	u16 value;
366 
367 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
368 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
369 		ixgbe_remove_adapter(hw);
370 		return true;
371 	}
372 	return false;
373 }
374 
375 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
376 {
377 	struct ixgbe_adapter *adapter = hw->back;
378 	u16 value;
379 
380 	if (ixgbe_removed(hw->hw_addr))
381 		return IXGBE_FAILED_READ_CFG_WORD;
382 	pci_read_config_word(adapter->pdev, reg, &value);
383 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
384 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
385 		return IXGBE_FAILED_READ_CFG_WORD;
386 	return value;
387 }
388 
389 #ifdef CONFIG_PCI_IOV
390 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
391 {
392 	struct ixgbe_adapter *adapter = hw->back;
393 	u32 value;
394 
395 	if (ixgbe_removed(hw->hw_addr))
396 		return IXGBE_FAILED_READ_CFG_DWORD;
397 	pci_read_config_dword(adapter->pdev, reg, &value);
398 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
399 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
400 		return IXGBE_FAILED_READ_CFG_DWORD;
401 	return value;
402 }
403 #endif /* CONFIG_PCI_IOV */
404 
405 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
406 {
407 	struct ixgbe_adapter *adapter = hw->back;
408 
409 	if (ixgbe_removed(hw->hw_addr))
410 		return;
411 	pci_write_config_word(adapter->pdev, reg, value);
412 }
413 
414 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
415 {
416 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
417 
418 	/* flush memory to make sure state is correct before next watchdog */
419 	smp_mb__before_atomic();
420 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
421 }
422 
423 struct ixgbe_reg_info {
424 	u32 ofs;
425 	char *name;
426 };
427 
428 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
429 
430 	/* General Registers */
431 	{IXGBE_CTRL, "CTRL"},
432 	{IXGBE_STATUS, "STATUS"},
433 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
434 
435 	/* Interrupt Registers */
436 	{IXGBE_EICR, "EICR"},
437 
438 	/* RX Registers */
439 	{IXGBE_SRRCTL(0), "SRRCTL"},
440 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
441 	{IXGBE_RDLEN(0), "RDLEN"},
442 	{IXGBE_RDH(0), "RDH"},
443 	{IXGBE_RDT(0), "RDT"},
444 	{IXGBE_RXDCTL(0), "RXDCTL"},
445 	{IXGBE_RDBAL(0), "RDBAL"},
446 	{IXGBE_RDBAH(0), "RDBAH"},
447 
448 	/* TX Registers */
449 	{IXGBE_TDBAL(0), "TDBAL"},
450 	{IXGBE_TDBAH(0), "TDBAH"},
451 	{IXGBE_TDLEN(0), "TDLEN"},
452 	{IXGBE_TDH(0), "TDH"},
453 	{IXGBE_TDT(0), "TDT"},
454 	{IXGBE_TXDCTL(0), "TXDCTL"},
455 
456 	/* List Terminator */
457 	{ .name = NULL }
458 };
459 
460 
461 /*
462  * ixgbe_regdump - register printout routine
463  */
464 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
465 {
466 	int i;
467 	char rname[16];
468 	u32 regs[64];
469 
470 	switch (reginfo->ofs) {
471 	case IXGBE_SRRCTL(0):
472 		for (i = 0; i < 64; i++)
473 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
474 		break;
475 	case IXGBE_DCA_RXCTRL(0):
476 		for (i = 0; i < 64; i++)
477 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
478 		break;
479 	case IXGBE_RDLEN(0):
480 		for (i = 0; i < 64; i++)
481 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
482 		break;
483 	case IXGBE_RDH(0):
484 		for (i = 0; i < 64; i++)
485 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
486 		break;
487 	case IXGBE_RDT(0):
488 		for (i = 0; i < 64; i++)
489 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
490 		break;
491 	case IXGBE_RXDCTL(0):
492 		for (i = 0; i < 64; i++)
493 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
494 		break;
495 	case IXGBE_RDBAL(0):
496 		for (i = 0; i < 64; i++)
497 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
498 		break;
499 	case IXGBE_RDBAH(0):
500 		for (i = 0; i < 64; i++)
501 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
502 		break;
503 	case IXGBE_TDBAL(0):
504 		for (i = 0; i < 64; i++)
505 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
506 		break;
507 	case IXGBE_TDBAH(0):
508 		for (i = 0; i < 64; i++)
509 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
510 		break;
511 	case IXGBE_TDLEN(0):
512 		for (i = 0; i < 64; i++)
513 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
514 		break;
515 	case IXGBE_TDH(0):
516 		for (i = 0; i < 64; i++)
517 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
518 		break;
519 	case IXGBE_TDT(0):
520 		for (i = 0; i < 64; i++)
521 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
522 		break;
523 	case IXGBE_TXDCTL(0):
524 		for (i = 0; i < 64; i++)
525 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
526 		break;
527 	default:
528 		pr_info("%-15s %08x\n",
529 			reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
530 		return;
531 	}
532 
533 	i = 0;
534 	while (i < 64) {
535 		int j;
536 		char buf[9 * 8 + 1];
537 		char *p = buf;
538 
539 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
540 		for (j = 0; j < 8; j++)
541 			p += sprintf(p, " %08x", regs[i++]);
542 		pr_err("%-15s%s\n", rname, buf);
543 	}
544 
545 }
546 
547 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
548 {
549 	struct ixgbe_tx_buffer *tx_buffer;
550 
551 	tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
552 	pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
553 		n, ring->next_to_use, ring->next_to_clean,
554 		(u64)dma_unmap_addr(tx_buffer, dma),
555 		dma_unmap_len(tx_buffer, len),
556 		tx_buffer->next_to_watch,
557 		(u64)tx_buffer->time_stamp);
558 }
559 
560 /*
561  * ixgbe_dump - Print registers, tx-rings and rx-rings
562  */
563 static void ixgbe_dump(struct ixgbe_adapter *adapter)
564 {
565 	struct net_device *netdev = adapter->netdev;
566 	struct ixgbe_hw *hw = &adapter->hw;
567 	struct ixgbe_reg_info *reginfo;
568 	int n = 0;
569 	struct ixgbe_ring *ring;
570 	struct ixgbe_tx_buffer *tx_buffer;
571 	union ixgbe_adv_tx_desc *tx_desc;
572 	struct my_u0 { u64 a; u64 b; } *u0;
573 	struct ixgbe_ring *rx_ring;
574 	union ixgbe_adv_rx_desc *rx_desc;
575 	struct ixgbe_rx_buffer *rx_buffer_info;
576 	int i = 0;
577 
578 	if (!netif_msg_hw(adapter))
579 		return;
580 
581 	/* Print netdevice Info */
582 	if (netdev) {
583 		dev_info(&adapter->pdev->dev, "Net device Info\n");
584 		pr_info("Device Name     state            "
585 			"trans_start\n");
586 		pr_info("%-15s %016lX %016lX\n",
587 			netdev->name,
588 			netdev->state,
589 			dev_trans_start(netdev));
590 	}
591 
592 	/* Print Registers */
593 	dev_info(&adapter->pdev->dev, "Register Dump\n");
594 	pr_info(" Register Name   Value\n");
595 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
596 	     reginfo->name; reginfo++) {
597 		ixgbe_regdump(hw, reginfo);
598 	}
599 
600 	/* Print TX Ring Summary */
601 	if (!netdev || !netif_running(netdev))
602 		return;
603 
604 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
605 	pr_info(" %s     %s              %s        %s\n",
606 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
607 		"leng", "ntw", "timestamp");
608 	for (n = 0; n < adapter->num_tx_queues; n++) {
609 		ring = adapter->tx_ring[n];
610 		ixgbe_print_buffer(ring, n);
611 	}
612 
613 	for (n = 0; n < adapter->num_xdp_queues; n++) {
614 		ring = adapter->xdp_ring[n];
615 		ixgbe_print_buffer(ring, n);
616 	}
617 
618 	/* Print TX Rings */
619 	if (!netif_msg_tx_done(adapter))
620 		goto rx_ring_summary;
621 
622 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
623 
624 	/* Transmit Descriptor Formats
625 	 *
626 	 * 82598 Advanced Transmit Descriptor
627 	 *   +--------------------------------------------------------------+
628 	 * 0 |         Buffer Address [63:0]                                |
629 	 *   +--------------------------------------------------------------+
630 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
631 	 *   +--------------------------------------------------------------+
632 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
633 	 *
634 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
635 	 *   +--------------------------------------------------------------+
636 	 * 0 |                          RSV [63:0]                          |
637 	 *   +--------------------------------------------------------------+
638 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
639 	 *   +--------------------------------------------------------------+
640 	 *   63                       36 35   32 31                         0
641 	 *
642 	 * 82599+ Advanced Transmit Descriptor
643 	 *   +--------------------------------------------------------------+
644 	 * 0 |         Buffer Address [63:0]                                |
645 	 *   +--------------------------------------------------------------+
646 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
647 	 *   +--------------------------------------------------------------+
648 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
649 	 *
650 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
651 	 *   +--------------------------------------------------------------+
652 	 * 0 |                          RSV [63:0]                          |
653 	 *   +--------------------------------------------------------------+
654 	 * 8 |            RSV           |  STA  |           RSV             |
655 	 *   +--------------------------------------------------------------+
656 	 *   63                       36 35   32 31                         0
657 	 */
658 
659 	for (n = 0; n < adapter->num_tx_queues; n++) {
660 		ring = adapter->tx_ring[n];
661 		pr_info("------------------------------------\n");
662 		pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
663 		pr_info("------------------------------------\n");
664 		pr_info("%s%s    %s              %s        %s          %s\n",
665 			"T [desc]     [address 63:0  ] ",
666 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
667 			"leng", "ntw", "timestamp", "bi->skb");
668 
669 		for (i = 0; ring->desc && (i < ring->count); i++) {
670 			tx_desc = IXGBE_TX_DESC(ring, i);
671 			tx_buffer = &ring->tx_buffer_info[i];
672 			u0 = (struct my_u0 *)tx_desc;
673 			if (dma_unmap_len(tx_buffer, len) > 0) {
674 				const char *ring_desc;
675 
676 				if (i == ring->next_to_use &&
677 				    i == ring->next_to_clean)
678 					ring_desc = " NTC/U";
679 				else if (i == ring->next_to_use)
680 					ring_desc = " NTU";
681 				else if (i == ring->next_to_clean)
682 					ring_desc = " NTC";
683 				else
684 					ring_desc = "";
685 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
686 					i,
687 					le64_to_cpu((__force __le64)u0->a),
688 					le64_to_cpu((__force __le64)u0->b),
689 					(u64)dma_unmap_addr(tx_buffer, dma),
690 					dma_unmap_len(tx_buffer, len),
691 					tx_buffer->next_to_watch,
692 					(u64)tx_buffer->time_stamp,
693 					tx_buffer->skb,
694 					ring_desc);
695 
696 				if (netif_msg_pktdata(adapter) &&
697 				    tx_buffer->skb)
698 					print_hex_dump(KERN_INFO, "",
699 						DUMP_PREFIX_ADDRESS, 16, 1,
700 						tx_buffer->skb->data,
701 						dma_unmap_len(tx_buffer, len),
702 						true);
703 			}
704 		}
705 	}
706 
707 	/* Print RX Rings Summary */
708 rx_ring_summary:
709 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
710 	pr_info("Queue [NTU] [NTC]\n");
711 	for (n = 0; n < adapter->num_rx_queues; n++) {
712 		rx_ring = adapter->rx_ring[n];
713 		pr_info("%5d %5X %5X\n",
714 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
715 	}
716 
717 	/* Print RX Rings */
718 	if (!netif_msg_rx_status(adapter))
719 		return;
720 
721 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
722 
723 	/* Receive Descriptor Formats
724 	 *
725 	 * 82598 Advanced Receive Descriptor (Read) Format
726 	 *    63                                           1        0
727 	 *    +-----------------------------------------------------+
728 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
729 	 *    +----------------------------------------------+------+
730 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
731 	 *    +-----------------------------------------------------+
732 	 *
733 	 *
734 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
735 	 *
736 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
737 	 *   +------------------------------------------------------+
738 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
739 	 *   | Packet   | IP     |   |          |     | Type | Type |
740 	 *   | Checksum | Ident  |   |          |     |      |      |
741 	 *   +------------------------------------------------------+
742 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
743 	 *   +------------------------------------------------------+
744 	 *   63       48 47    32 31            20 19               0
745 	 *
746 	 * 82599+ Advanced Receive Descriptor (Read) Format
747 	 *    63                                           1        0
748 	 *    +-----------------------------------------------------+
749 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
750 	 *    +----------------------------------------------+------+
751 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
752 	 *    +-----------------------------------------------------+
753 	 *
754 	 *
755 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
756 	 *
757 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
758 	 *   +------------------------------------------------------+
759 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
760 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
761 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
762 	 *   +------------------------------------------------------+
763 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
764 	 *   +------------------------------------------------------+
765 	 *   63       48 47    32 31          20 19                 0
766 	 */
767 
768 	for (n = 0; n < adapter->num_rx_queues; n++) {
769 		rx_ring = adapter->rx_ring[n];
770 		pr_info("------------------------------------\n");
771 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
772 		pr_info("------------------------------------\n");
773 		pr_info("%s%s%s\n",
774 			"R  [desc]      [ PktBuf     A0] ",
775 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
776 			"<-- Adv Rx Read format");
777 		pr_info("%s%s%s\n",
778 			"RWB[desc]      [PcsmIpSHl PtRs] ",
779 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
780 			"<-- Adv Rx Write-Back format");
781 
782 		for (i = 0; i < rx_ring->count; i++) {
783 			const char *ring_desc;
784 
785 			if (i == rx_ring->next_to_use)
786 				ring_desc = " NTU";
787 			else if (i == rx_ring->next_to_clean)
788 				ring_desc = " NTC";
789 			else
790 				ring_desc = "";
791 
792 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
793 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
794 			u0 = (struct my_u0 *)rx_desc;
795 			if (rx_desc->wb.upper.length) {
796 				/* Descriptor Done */
797 				pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
798 					i,
799 					le64_to_cpu((__force __le64)u0->a),
800 					le64_to_cpu((__force __le64)u0->b),
801 					rx_buffer_info->skb,
802 					ring_desc);
803 			} else {
804 				pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
805 					i,
806 					le64_to_cpu((__force __le64)u0->a),
807 					le64_to_cpu((__force __le64)u0->b),
808 					(u64)rx_buffer_info->dma,
809 					rx_buffer_info->skb,
810 					ring_desc);
811 
812 				if (netif_msg_pktdata(adapter) &&
813 				    rx_buffer_info->dma) {
814 					print_hex_dump(KERN_INFO, "",
815 					   DUMP_PREFIX_ADDRESS, 16, 1,
816 					   page_address(rx_buffer_info->page) +
817 						    rx_buffer_info->page_offset,
818 					   ixgbe_rx_bufsz(rx_ring), true);
819 				}
820 			}
821 		}
822 	}
823 }
824 
825 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
826 {
827 	u32 ctrl_ext;
828 
829 	/* Let firmware take over control of h/w */
830 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
831 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
832 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
833 }
834 
835 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
836 {
837 	u32 ctrl_ext;
838 
839 	/* Let firmware know the driver has taken over */
840 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
841 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
842 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
843 }
844 
845 /**
846  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
847  * @adapter: pointer to adapter struct
848  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
849  * @queue: queue to map the corresponding interrupt to
850  * @msix_vector: the vector to map to the corresponding queue
851  *
852  */
853 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
854 			   u8 queue, u8 msix_vector)
855 {
856 	u32 ivar, index;
857 	struct ixgbe_hw *hw = &adapter->hw;
858 	switch (hw->mac.type) {
859 	case ixgbe_mac_82598EB:
860 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
861 		if (direction == -1)
862 			direction = 0;
863 		index = (((direction * 64) + queue) >> 2) & 0x1F;
864 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
865 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
866 		ivar |= (msix_vector << (8 * (queue & 0x3)));
867 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
868 		break;
869 	case ixgbe_mac_82599EB:
870 	case ixgbe_mac_X540:
871 	case ixgbe_mac_X550:
872 	case ixgbe_mac_X550EM_x:
873 	case ixgbe_mac_x550em_a:
874 		if (direction == -1) {
875 			/* other causes */
876 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
877 			index = ((queue & 1) * 8);
878 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
879 			ivar &= ~(0xFF << index);
880 			ivar |= (msix_vector << index);
881 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
882 			break;
883 		} else {
884 			/* tx or rx causes */
885 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
886 			index = ((16 * (queue & 1)) + (8 * direction));
887 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
888 			ivar &= ~(0xFF << index);
889 			ivar |= (msix_vector << index);
890 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
891 			break;
892 		}
893 	default:
894 		break;
895 	}
896 }
897 
898 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
899 			    u64 qmask)
900 {
901 	u32 mask;
902 
903 	switch (adapter->hw.mac.type) {
904 	case ixgbe_mac_82598EB:
905 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
906 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
907 		break;
908 	case ixgbe_mac_82599EB:
909 	case ixgbe_mac_X540:
910 	case ixgbe_mac_X550:
911 	case ixgbe_mac_X550EM_x:
912 	case ixgbe_mac_x550em_a:
913 		mask = (qmask & 0xFFFFFFFF);
914 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
915 		mask = (qmask >> 32);
916 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
917 		break;
918 	default:
919 		break;
920 	}
921 }
922 
923 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
924 {
925 	struct ixgbe_hw *hw = &adapter->hw;
926 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
927 	int i;
928 	u32 data;
929 
930 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
931 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
932 		return;
933 
934 	switch (hw->mac.type) {
935 	case ixgbe_mac_82598EB:
936 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
937 		break;
938 	default:
939 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
940 	}
941 	hwstats->lxoffrxc += data;
942 
943 	/* refill credits (no tx hang) if we received xoff */
944 	if (!data)
945 		return;
946 
947 	for (i = 0; i < adapter->num_tx_queues; i++)
948 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
949 			  &adapter->tx_ring[i]->state);
950 
951 	for (i = 0; i < adapter->num_xdp_queues; i++)
952 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
953 			  &adapter->xdp_ring[i]->state);
954 }
955 
956 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
957 {
958 	struct ixgbe_hw *hw = &adapter->hw;
959 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
960 	u32 xoff[8] = {0};
961 	u8 tc;
962 	int i;
963 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
964 
965 	if (adapter->ixgbe_ieee_pfc)
966 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
967 
968 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
969 		ixgbe_update_xoff_rx_lfc(adapter);
970 		return;
971 	}
972 
973 	/* update stats for each tc, only valid with PFC enabled */
974 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
975 		u32 pxoffrxc;
976 
977 		switch (hw->mac.type) {
978 		case ixgbe_mac_82598EB:
979 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
980 			break;
981 		default:
982 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
983 		}
984 		hwstats->pxoffrxc[i] += pxoffrxc;
985 		/* Get the TC for given UP */
986 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
987 		xoff[tc] += pxoffrxc;
988 	}
989 
990 	/* disarm tx queues that have received xoff frames */
991 	for (i = 0; i < adapter->num_tx_queues; i++) {
992 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
993 
994 		tc = tx_ring->dcb_tc;
995 		if (xoff[tc])
996 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
997 	}
998 
999 	for (i = 0; i < adapter->num_xdp_queues; i++) {
1000 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1001 
1002 		tc = xdp_ring->dcb_tc;
1003 		if (xoff[tc])
1004 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1005 	}
1006 }
1007 
1008 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1009 {
1010 	return ring->stats.packets;
1011 }
1012 
1013 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1014 {
1015 	unsigned int head, tail;
1016 
1017 	head = ring->next_to_clean;
1018 	tail = ring->next_to_use;
1019 
1020 	return ((head <= tail) ? tail : tail + ring->count) - head;
1021 }
1022 
1023 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1024 {
1025 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1026 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1027 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1028 
1029 	clear_check_for_tx_hang(tx_ring);
1030 
1031 	/*
1032 	 * Check for a hung queue, but be thorough. This verifies
1033 	 * that a transmit has been completed since the previous
1034 	 * check AND there is at least one packet pending. The
1035 	 * ARMED bit is set to indicate a potential hang. The
1036 	 * bit is cleared if a pause frame is received to remove
1037 	 * false hang detection due to PFC or 802.3x frames. By
1038 	 * requiring this to fail twice we avoid races with
1039 	 * pfc clearing the ARMED bit and conditions where we
1040 	 * run the check_tx_hang logic with a transmit completion
1041 	 * pending but without time to complete it yet.
1042 	 */
1043 	if (tx_done_old == tx_done && tx_pending)
1044 		/* make sure it is true for two checks in a row */
1045 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1046 					&tx_ring->state);
1047 	/* update completed stats and continue */
1048 	tx_ring->tx_stats.tx_done_old = tx_done;
1049 	/* reset the countdown */
1050 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1051 
1052 	return false;
1053 }
1054 
1055 /**
1056  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1057  * @adapter: driver private struct
1058  **/
1059 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1060 {
1061 
1062 	/* Do the reset outside of interrupt context */
1063 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1064 		set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1065 		e_warn(drv, "initiating reset due to tx timeout\n");
1066 		ixgbe_service_event_schedule(adapter);
1067 	}
1068 }
1069 
1070 /**
1071  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1072  * @netdev: network interface device structure
1073  * @queue_index: Tx queue to set
1074  * @maxrate: desired maximum transmit bitrate
1075  **/
1076 static int ixgbe_tx_maxrate(struct net_device *netdev,
1077 			    int queue_index, u32 maxrate)
1078 {
1079 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
1080 	struct ixgbe_hw *hw = &adapter->hw;
1081 	u32 bcnrc_val = ixgbe_link_mbps(adapter);
1082 
1083 	if (!maxrate)
1084 		return 0;
1085 
1086 	/* Calculate the rate factor values to set */
1087 	bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1088 	bcnrc_val /= maxrate;
1089 
1090 	/* clear everything but the rate factor */
1091 	bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1092 	IXGBE_RTTBCNRC_RF_DEC_MASK;
1093 
1094 	/* enable the rate scheduler */
1095 	bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1096 
1097 	IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1098 	IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1099 
1100 	return 0;
1101 }
1102 
1103 /**
1104  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1105  * @q_vector: structure containing interrupt and ring information
1106  * @tx_ring: tx ring to clean
1107  * @napi_budget: Used to determine if we are in netpoll
1108  **/
1109 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1110 			       struct ixgbe_ring *tx_ring, int napi_budget)
1111 {
1112 	struct ixgbe_adapter *adapter = q_vector->adapter;
1113 	struct ixgbe_tx_buffer *tx_buffer;
1114 	union ixgbe_adv_tx_desc *tx_desc;
1115 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1116 	unsigned int budget = q_vector->tx.work_limit;
1117 	unsigned int i = tx_ring->next_to_clean;
1118 
1119 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1120 		return true;
1121 
1122 	tx_buffer = &tx_ring->tx_buffer_info[i];
1123 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1124 	i -= tx_ring->count;
1125 
1126 	do {
1127 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1128 
1129 		/* if next_to_watch is not set then there is no work pending */
1130 		if (!eop_desc)
1131 			break;
1132 
1133 		/* prevent any other reads prior to eop_desc */
1134 		smp_rmb();
1135 
1136 		/* if DD is not set pending work has not been completed */
1137 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1138 			break;
1139 
1140 		/* clear next_to_watch to prevent false hangs */
1141 		tx_buffer->next_to_watch = NULL;
1142 
1143 		/* update the statistics for this packet */
1144 		total_bytes += tx_buffer->bytecount;
1145 		total_packets += tx_buffer->gso_segs;
1146 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1147 			total_ipsec++;
1148 
1149 		/* free the skb */
1150 		if (ring_is_xdp(tx_ring))
1151 			xdp_return_frame(tx_buffer->xdpf);
1152 		else
1153 			napi_consume_skb(tx_buffer->skb, napi_budget);
1154 
1155 		/* unmap skb header data */
1156 		dma_unmap_single(tx_ring->dev,
1157 				 dma_unmap_addr(tx_buffer, dma),
1158 				 dma_unmap_len(tx_buffer, len),
1159 				 DMA_TO_DEVICE);
1160 
1161 		/* clear tx_buffer data */
1162 		dma_unmap_len_set(tx_buffer, len, 0);
1163 
1164 		/* unmap remaining buffers */
1165 		while (tx_desc != eop_desc) {
1166 			tx_buffer++;
1167 			tx_desc++;
1168 			i++;
1169 			if (unlikely(!i)) {
1170 				i -= tx_ring->count;
1171 				tx_buffer = tx_ring->tx_buffer_info;
1172 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1173 			}
1174 
1175 			/* unmap any remaining paged data */
1176 			if (dma_unmap_len(tx_buffer, len)) {
1177 				dma_unmap_page(tx_ring->dev,
1178 					       dma_unmap_addr(tx_buffer, dma),
1179 					       dma_unmap_len(tx_buffer, len),
1180 					       DMA_TO_DEVICE);
1181 				dma_unmap_len_set(tx_buffer, len, 0);
1182 			}
1183 		}
1184 
1185 		/* move us one more past the eop_desc for start of next pkt */
1186 		tx_buffer++;
1187 		tx_desc++;
1188 		i++;
1189 		if (unlikely(!i)) {
1190 			i -= tx_ring->count;
1191 			tx_buffer = tx_ring->tx_buffer_info;
1192 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1193 		}
1194 
1195 		/* issue prefetch for next Tx descriptor */
1196 		prefetch(tx_desc);
1197 
1198 		/* update budget accounting */
1199 		budget--;
1200 	} while (likely(budget));
1201 
1202 	i += tx_ring->count;
1203 	tx_ring->next_to_clean = i;
1204 	u64_stats_update_begin(&tx_ring->syncp);
1205 	tx_ring->stats.bytes += total_bytes;
1206 	tx_ring->stats.packets += total_packets;
1207 	u64_stats_update_end(&tx_ring->syncp);
1208 	q_vector->tx.total_bytes += total_bytes;
1209 	q_vector->tx.total_packets += total_packets;
1210 	adapter->tx_ipsec += total_ipsec;
1211 
1212 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1213 		/* schedule immediate reset if we believe we hung */
1214 		struct ixgbe_hw *hw = &adapter->hw;
1215 		e_err(drv, "Detected Tx Unit Hang %s\n"
1216 			"  Tx Queue             <%d>\n"
1217 			"  TDH, TDT             <%x>, <%x>\n"
1218 			"  next_to_use          <%x>\n"
1219 			"  next_to_clean        <%x>\n"
1220 			"tx_buffer_info[next_to_clean]\n"
1221 			"  time_stamp           <%lx>\n"
1222 			"  jiffies              <%lx>\n",
1223 			ring_is_xdp(tx_ring) ? "(XDP)" : "",
1224 			tx_ring->queue_index,
1225 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1226 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1227 			tx_ring->next_to_use, i,
1228 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1229 
1230 		if (!ring_is_xdp(tx_ring))
1231 			netif_stop_subqueue(tx_ring->netdev,
1232 					    tx_ring->queue_index);
1233 
1234 		e_info(probe,
1235 		       "tx hang %d detected on queue %d, resetting adapter\n",
1236 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1237 
1238 		/* schedule immediate reset if we believe we hung */
1239 		ixgbe_tx_timeout_reset(adapter);
1240 
1241 		/* the adapter is about to reset, no point in enabling stuff */
1242 		return true;
1243 	}
1244 
1245 	if (ring_is_xdp(tx_ring))
1246 		return !!budget;
1247 
1248 	netdev_tx_completed_queue(txring_txq(tx_ring),
1249 				  total_packets, total_bytes);
1250 
1251 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1252 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1253 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1254 		/* Make sure that anybody stopping the queue after this
1255 		 * sees the new next_to_clean.
1256 		 */
1257 		smp_mb();
1258 		if (__netif_subqueue_stopped(tx_ring->netdev,
1259 					     tx_ring->queue_index)
1260 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1261 			netif_wake_subqueue(tx_ring->netdev,
1262 					    tx_ring->queue_index);
1263 			++tx_ring->tx_stats.restart_queue;
1264 		}
1265 	}
1266 
1267 	return !!budget;
1268 }
1269 
1270 #ifdef CONFIG_IXGBE_DCA
1271 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1272 				struct ixgbe_ring *tx_ring,
1273 				int cpu)
1274 {
1275 	struct ixgbe_hw *hw = &adapter->hw;
1276 	u32 txctrl = 0;
1277 	u16 reg_offset;
1278 
1279 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1280 		txctrl = dca3_get_tag(tx_ring->dev, cpu);
1281 
1282 	switch (hw->mac.type) {
1283 	case ixgbe_mac_82598EB:
1284 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1285 		break;
1286 	case ixgbe_mac_82599EB:
1287 	case ixgbe_mac_X540:
1288 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1289 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1290 		break;
1291 	default:
1292 		/* for unknown hardware do not write register */
1293 		return;
1294 	}
1295 
1296 	/*
1297 	 * We can enable relaxed ordering for reads, but not writes when
1298 	 * DCA is enabled.  This is due to a known issue in some chipsets
1299 	 * which will cause the DCA tag to be cleared.
1300 	 */
1301 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1302 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1303 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1304 
1305 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1306 }
1307 
1308 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1309 				struct ixgbe_ring *rx_ring,
1310 				int cpu)
1311 {
1312 	struct ixgbe_hw *hw = &adapter->hw;
1313 	u32 rxctrl = 0;
1314 	u8 reg_idx = rx_ring->reg_idx;
1315 
1316 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1317 		rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1318 
1319 	switch (hw->mac.type) {
1320 	case ixgbe_mac_82599EB:
1321 	case ixgbe_mac_X540:
1322 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1323 		break;
1324 	default:
1325 		break;
1326 	}
1327 
1328 	/*
1329 	 * We can enable relaxed ordering for reads, but not writes when
1330 	 * DCA is enabled.  This is due to a known issue in some chipsets
1331 	 * which will cause the DCA tag to be cleared.
1332 	 */
1333 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1334 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1335 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1336 
1337 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1338 }
1339 
1340 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1341 {
1342 	struct ixgbe_adapter *adapter = q_vector->adapter;
1343 	struct ixgbe_ring *ring;
1344 	int cpu = get_cpu();
1345 
1346 	if (q_vector->cpu == cpu)
1347 		goto out_no_update;
1348 
1349 	ixgbe_for_each_ring(ring, q_vector->tx)
1350 		ixgbe_update_tx_dca(adapter, ring, cpu);
1351 
1352 	ixgbe_for_each_ring(ring, q_vector->rx)
1353 		ixgbe_update_rx_dca(adapter, ring, cpu);
1354 
1355 	q_vector->cpu = cpu;
1356 out_no_update:
1357 	put_cpu();
1358 }
1359 
1360 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1361 {
1362 	int i;
1363 
1364 	/* always use CB2 mode, difference is masked in the CB driver */
1365 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1366 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1367 				IXGBE_DCA_CTRL_DCA_MODE_CB2);
1368 	else
1369 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1370 				IXGBE_DCA_CTRL_DCA_DISABLE);
1371 
1372 	for (i = 0; i < adapter->num_q_vectors; i++) {
1373 		adapter->q_vector[i]->cpu = -1;
1374 		ixgbe_update_dca(adapter->q_vector[i]);
1375 	}
1376 }
1377 
1378 static int __ixgbe_notify_dca(struct device *dev, void *data)
1379 {
1380 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1381 	unsigned long event = *(unsigned long *)data;
1382 
1383 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1384 		return 0;
1385 
1386 	switch (event) {
1387 	case DCA_PROVIDER_ADD:
1388 		/* if we're already enabled, don't do it again */
1389 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1390 			break;
1391 		if (dca_add_requester(dev) == 0) {
1392 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1393 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1394 					IXGBE_DCA_CTRL_DCA_MODE_CB2);
1395 			break;
1396 		}
1397 		/* fall through - DCA is disabled. */
1398 	case DCA_PROVIDER_REMOVE:
1399 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1400 			dca_remove_requester(dev);
1401 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1402 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1403 					IXGBE_DCA_CTRL_DCA_DISABLE);
1404 		}
1405 		break;
1406 	}
1407 
1408 	return 0;
1409 }
1410 
1411 #endif /* CONFIG_IXGBE_DCA */
1412 
1413 #define IXGBE_RSS_L4_TYPES_MASK \
1414 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1415 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1416 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1417 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1418 
1419 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1420 				 union ixgbe_adv_rx_desc *rx_desc,
1421 				 struct sk_buff *skb)
1422 {
1423 	u16 rss_type;
1424 
1425 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1426 		return;
1427 
1428 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1429 		   IXGBE_RXDADV_RSSTYPE_MASK;
1430 
1431 	if (!rss_type)
1432 		return;
1433 
1434 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1435 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1436 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1437 }
1438 
1439 #ifdef IXGBE_FCOE
1440 /**
1441  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1442  * @ring: structure containing ring specific data
1443  * @rx_desc: advanced rx descriptor
1444  *
1445  * Returns : true if it is FCoE pkt
1446  */
1447 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1448 				    union ixgbe_adv_rx_desc *rx_desc)
1449 {
1450 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1451 
1452 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1453 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1454 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1455 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1456 }
1457 
1458 #endif /* IXGBE_FCOE */
1459 /**
1460  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1461  * @ring: structure containing ring specific data
1462  * @rx_desc: current Rx descriptor being processed
1463  * @skb: skb currently being received and modified
1464  **/
1465 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1466 				     union ixgbe_adv_rx_desc *rx_desc,
1467 				     struct sk_buff *skb)
1468 {
1469 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1470 	bool encap_pkt = false;
1471 
1472 	skb_checksum_none_assert(skb);
1473 
1474 	/* Rx csum disabled */
1475 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1476 		return;
1477 
1478 	/* check for VXLAN and Geneve packets */
1479 	if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1480 		encap_pkt = true;
1481 		skb->encapsulation = 1;
1482 	}
1483 
1484 	/* if IP and error */
1485 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1486 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1487 		ring->rx_stats.csum_err++;
1488 		return;
1489 	}
1490 
1491 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1492 		return;
1493 
1494 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1495 		/*
1496 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1497 		 * checksum errors.
1498 		 */
1499 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1500 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1501 			return;
1502 
1503 		ring->rx_stats.csum_err++;
1504 		return;
1505 	}
1506 
1507 	/* It must be a TCP or UDP packet with a valid checksum */
1508 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1509 	if (encap_pkt) {
1510 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1511 			return;
1512 
1513 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1514 			skb->ip_summed = CHECKSUM_NONE;
1515 			return;
1516 		}
1517 		/* If we checked the outer header let the stack know */
1518 		skb->csum_level = 1;
1519 	}
1520 }
1521 
1522 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1523 {
1524 	return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1525 }
1526 
1527 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1528 				    struct ixgbe_rx_buffer *bi)
1529 {
1530 	struct page *page = bi->page;
1531 	dma_addr_t dma;
1532 
1533 	/* since we are recycling buffers we should seldom need to alloc */
1534 	if (likely(page))
1535 		return true;
1536 
1537 	/* alloc new page for storage */
1538 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1539 	if (unlikely(!page)) {
1540 		rx_ring->rx_stats.alloc_rx_page_failed++;
1541 		return false;
1542 	}
1543 
1544 	/* map page for use */
1545 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1546 				 ixgbe_rx_pg_size(rx_ring),
1547 				 DMA_FROM_DEVICE,
1548 				 IXGBE_RX_DMA_ATTR);
1549 
1550 	/*
1551 	 * if mapping failed free memory back to system since
1552 	 * there isn't much point in holding memory we can't use
1553 	 */
1554 	if (dma_mapping_error(rx_ring->dev, dma)) {
1555 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1556 
1557 		rx_ring->rx_stats.alloc_rx_page_failed++;
1558 		return false;
1559 	}
1560 
1561 	bi->dma = dma;
1562 	bi->page = page;
1563 	bi->page_offset = ixgbe_rx_offset(rx_ring);
1564 	page_ref_add(page, USHRT_MAX - 1);
1565 	bi->pagecnt_bias = USHRT_MAX;
1566 	rx_ring->rx_stats.alloc_rx_page++;
1567 
1568 	return true;
1569 }
1570 
1571 /**
1572  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1573  * @rx_ring: ring to place buffers on
1574  * @cleaned_count: number of buffers to replace
1575  **/
1576 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1577 {
1578 	union ixgbe_adv_rx_desc *rx_desc;
1579 	struct ixgbe_rx_buffer *bi;
1580 	u16 i = rx_ring->next_to_use;
1581 	u16 bufsz;
1582 
1583 	/* nothing to do */
1584 	if (!cleaned_count)
1585 		return;
1586 
1587 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1588 	bi = &rx_ring->rx_buffer_info[i];
1589 	i -= rx_ring->count;
1590 
1591 	bufsz = ixgbe_rx_bufsz(rx_ring);
1592 
1593 	do {
1594 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1595 			break;
1596 
1597 		/* sync the buffer for use by the device */
1598 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1599 						 bi->page_offset, bufsz,
1600 						 DMA_FROM_DEVICE);
1601 
1602 		/*
1603 		 * Refresh the desc even if buffer_addrs didn't change
1604 		 * because each write-back erases this info.
1605 		 */
1606 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1607 
1608 		rx_desc++;
1609 		bi++;
1610 		i++;
1611 		if (unlikely(!i)) {
1612 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1613 			bi = rx_ring->rx_buffer_info;
1614 			i -= rx_ring->count;
1615 		}
1616 
1617 		/* clear the length for the next_to_use descriptor */
1618 		rx_desc->wb.upper.length = 0;
1619 
1620 		cleaned_count--;
1621 	} while (cleaned_count);
1622 
1623 	i += rx_ring->count;
1624 
1625 	if (rx_ring->next_to_use != i) {
1626 		rx_ring->next_to_use = i;
1627 
1628 		/* update next to alloc since we have filled the ring */
1629 		rx_ring->next_to_alloc = i;
1630 
1631 		/* Force memory writes to complete before letting h/w
1632 		 * know there are new descriptors to fetch.  (Only
1633 		 * applicable for weak-ordered memory model archs,
1634 		 * such as IA-64).
1635 		 */
1636 		wmb();
1637 		writel(i, rx_ring->tail);
1638 	}
1639 }
1640 
1641 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1642 				   struct sk_buff *skb)
1643 {
1644 	u16 hdr_len = skb_headlen(skb);
1645 
1646 	/* set gso_size to avoid messing up TCP MSS */
1647 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1648 						 IXGBE_CB(skb)->append_cnt);
1649 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1650 }
1651 
1652 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1653 				   struct sk_buff *skb)
1654 {
1655 	/* if append_cnt is 0 then frame is not RSC */
1656 	if (!IXGBE_CB(skb)->append_cnt)
1657 		return;
1658 
1659 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1660 	rx_ring->rx_stats.rsc_flush++;
1661 
1662 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1663 
1664 	/* gso_size is computed using append_cnt so always clear it last */
1665 	IXGBE_CB(skb)->append_cnt = 0;
1666 }
1667 
1668 /**
1669  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1670  * @rx_ring: rx descriptor ring packet is being transacted on
1671  * @rx_desc: pointer to the EOP Rx descriptor
1672  * @skb: pointer to current skb being populated
1673  *
1674  * This function checks the ring, descriptor, and packet information in
1675  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1676  * other fields within the skb.
1677  **/
1678 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1679 			      union ixgbe_adv_rx_desc *rx_desc,
1680 			      struct sk_buff *skb)
1681 {
1682 	struct net_device *dev = rx_ring->netdev;
1683 	u32 flags = rx_ring->q_vector->adapter->flags;
1684 
1685 	ixgbe_update_rsc_stats(rx_ring, skb);
1686 
1687 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1688 
1689 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1690 
1691 	if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1692 		ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1693 
1694 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1695 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1696 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1697 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1698 	}
1699 
1700 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1701 		ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1702 
1703 	/* record Rx queue, or update MACVLAN statistics */
1704 	if (netif_is_ixgbe(dev))
1705 		skb_record_rx_queue(skb, rx_ring->queue_index);
1706 	else
1707 		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1708 				 false);
1709 
1710 	skb->protocol = eth_type_trans(skb, dev);
1711 }
1712 
1713 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1714 		  struct sk_buff *skb)
1715 {
1716 	napi_gro_receive(&q_vector->napi, skb);
1717 }
1718 
1719 /**
1720  * ixgbe_is_non_eop - process handling of non-EOP buffers
1721  * @rx_ring: Rx ring being processed
1722  * @rx_desc: Rx descriptor for current buffer
1723  * @skb: Current socket buffer containing buffer in progress
1724  *
1725  * This function updates next to clean.  If the buffer is an EOP buffer
1726  * this function exits returning false, otherwise it will place the
1727  * sk_buff in the next buffer to be chained and return true indicating
1728  * that this is in fact a non-EOP buffer.
1729  **/
1730 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1731 			     union ixgbe_adv_rx_desc *rx_desc,
1732 			     struct sk_buff *skb)
1733 {
1734 	u32 ntc = rx_ring->next_to_clean + 1;
1735 
1736 	/* fetch, update, and store next to clean */
1737 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1738 	rx_ring->next_to_clean = ntc;
1739 
1740 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1741 
1742 	/* update RSC append count if present */
1743 	if (ring_is_rsc_enabled(rx_ring)) {
1744 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1745 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1746 
1747 		if (unlikely(rsc_enabled)) {
1748 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1749 
1750 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1751 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1752 
1753 			/* update ntc based on RSC value */
1754 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1755 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1756 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1757 		}
1758 	}
1759 
1760 	/* if we are the last buffer then there is nothing else to do */
1761 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1762 		return false;
1763 
1764 	/* place skb in next buffer to be received */
1765 	rx_ring->rx_buffer_info[ntc].skb = skb;
1766 	rx_ring->rx_stats.non_eop_descs++;
1767 
1768 	return true;
1769 }
1770 
1771 /**
1772  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1773  * @rx_ring: rx descriptor ring packet is being transacted on
1774  * @skb: pointer to current skb being adjusted
1775  *
1776  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1777  * main difference between this version and the original function is that
1778  * this function can make several assumptions about the state of things
1779  * that allow for significant optimizations versus the standard function.
1780  * As a result we can do things like drop a frag and maintain an accurate
1781  * truesize for the skb.
1782  */
1783 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1784 			    struct sk_buff *skb)
1785 {
1786 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1787 	unsigned char *va;
1788 	unsigned int pull_len;
1789 
1790 	/*
1791 	 * it is valid to use page_address instead of kmap since we are
1792 	 * working with pages allocated out of the lomem pool per
1793 	 * alloc_page(GFP_ATOMIC)
1794 	 */
1795 	va = skb_frag_address(frag);
1796 
1797 	/*
1798 	 * we need the header to contain the greater of either ETH_HLEN or
1799 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1800 	 */
1801 	pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1802 
1803 	/* align pull length to size of long to optimize memcpy performance */
1804 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1805 
1806 	/* update all of the pointers */
1807 	skb_frag_size_sub(frag, pull_len);
1808 	frag->page_offset += pull_len;
1809 	skb->data_len -= pull_len;
1810 	skb->tail += pull_len;
1811 }
1812 
1813 /**
1814  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1815  * @rx_ring: rx descriptor ring packet is being transacted on
1816  * @skb: pointer to current skb being updated
1817  *
1818  * This function provides a basic DMA sync up for the first fragment of an
1819  * skb.  The reason for doing this is that the first fragment cannot be
1820  * unmapped until we have reached the end of packet descriptor for a buffer
1821  * chain.
1822  */
1823 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1824 				struct sk_buff *skb)
1825 {
1826 	/* if the page was released unmap it, else just sync our portion */
1827 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1828 		dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1829 				     ixgbe_rx_pg_size(rx_ring),
1830 				     DMA_FROM_DEVICE,
1831 				     IXGBE_RX_DMA_ATTR);
1832 	} else if (ring_uses_build_skb(rx_ring)) {
1833 		unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1834 
1835 		dma_sync_single_range_for_cpu(rx_ring->dev,
1836 					      IXGBE_CB(skb)->dma,
1837 					      offset,
1838 					      skb_headlen(skb),
1839 					      DMA_FROM_DEVICE);
1840 	} else {
1841 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1842 
1843 		dma_sync_single_range_for_cpu(rx_ring->dev,
1844 					      IXGBE_CB(skb)->dma,
1845 					      frag->page_offset,
1846 					      skb_frag_size(frag),
1847 					      DMA_FROM_DEVICE);
1848 	}
1849 }
1850 
1851 /**
1852  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1853  * @rx_ring: rx descriptor ring packet is being transacted on
1854  * @rx_desc: pointer to the EOP Rx descriptor
1855  * @skb: pointer to current skb being fixed
1856  *
1857  * Check if the skb is valid in the XDP case it will be an error pointer.
1858  * Return true in this case to abort processing and advance to next
1859  * descriptor.
1860  *
1861  * Check for corrupted packet headers caused by senders on the local L2
1862  * embedded NIC switch not setting up their Tx Descriptors right.  These
1863  * should be very rare.
1864  *
1865  * Also address the case where we are pulling data in on pages only
1866  * and as such no data is present in the skb header.
1867  *
1868  * In addition if skb is not at least 60 bytes we need to pad it so that
1869  * it is large enough to qualify as a valid Ethernet frame.
1870  *
1871  * Returns true if an error was encountered and skb was freed.
1872  **/
1873 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1874 			   union ixgbe_adv_rx_desc *rx_desc,
1875 			   struct sk_buff *skb)
1876 {
1877 	struct net_device *netdev = rx_ring->netdev;
1878 
1879 	/* XDP packets use error pointer so abort at this point */
1880 	if (IS_ERR(skb))
1881 		return true;
1882 
1883 	/* Verify netdev is present, and that packet does not have any
1884 	 * errors that would be unacceptable to the netdev.
1885 	 */
1886 	if (!netdev ||
1887 	    (unlikely(ixgbe_test_staterr(rx_desc,
1888 					 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1889 	     !(netdev->features & NETIF_F_RXALL)))) {
1890 		dev_kfree_skb_any(skb);
1891 		return true;
1892 	}
1893 
1894 	/* place header in linear portion of buffer */
1895 	if (!skb_headlen(skb))
1896 		ixgbe_pull_tail(rx_ring, skb);
1897 
1898 #ifdef IXGBE_FCOE
1899 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1900 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1901 		return false;
1902 
1903 #endif
1904 	/* if eth_skb_pad returns an error the skb was freed */
1905 	if (eth_skb_pad(skb))
1906 		return true;
1907 
1908 	return false;
1909 }
1910 
1911 /**
1912  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1913  * @rx_ring: rx descriptor ring to store buffers on
1914  * @old_buff: donor buffer to have page reused
1915  *
1916  * Synchronizes page for reuse by the adapter
1917  **/
1918 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1919 				struct ixgbe_rx_buffer *old_buff)
1920 {
1921 	struct ixgbe_rx_buffer *new_buff;
1922 	u16 nta = rx_ring->next_to_alloc;
1923 
1924 	new_buff = &rx_ring->rx_buffer_info[nta];
1925 
1926 	/* update, and store next to alloc */
1927 	nta++;
1928 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1929 
1930 	/* Transfer page from old buffer to new buffer.
1931 	 * Move each member individually to avoid possible store
1932 	 * forwarding stalls and unnecessary copy of skb.
1933 	 */
1934 	new_buff->dma		= old_buff->dma;
1935 	new_buff->page		= old_buff->page;
1936 	new_buff->page_offset	= old_buff->page_offset;
1937 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1938 }
1939 
1940 static inline bool ixgbe_page_is_reserved(struct page *page)
1941 {
1942 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1943 }
1944 
1945 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1946 {
1947 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1948 	struct page *page = rx_buffer->page;
1949 
1950 	/* avoid re-using remote pages */
1951 	if (unlikely(ixgbe_page_is_reserved(page)))
1952 		return false;
1953 
1954 #if (PAGE_SIZE < 8192)
1955 	/* if we are only owner of page we can reuse it */
1956 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1957 		return false;
1958 #else
1959 	/* The last offset is a bit aggressive in that we assume the
1960 	 * worst case of FCoE being enabled and using a 3K buffer.
1961 	 * However this should have minimal impact as the 1K extra is
1962 	 * still less than one buffer in size.
1963 	 */
1964 #define IXGBE_LAST_OFFSET \
1965 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1966 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1967 		return false;
1968 #endif
1969 
1970 	/* If we have drained the page fragment pool we need to update
1971 	 * the pagecnt_bias and page count so that we fully restock the
1972 	 * number of references the driver holds.
1973 	 */
1974 	if (unlikely(pagecnt_bias == 1)) {
1975 		page_ref_add(page, USHRT_MAX - 1);
1976 		rx_buffer->pagecnt_bias = USHRT_MAX;
1977 	}
1978 
1979 	return true;
1980 }
1981 
1982 /**
1983  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1984  * @rx_ring: rx descriptor ring to transact packets on
1985  * @rx_buffer: buffer containing page to add
1986  * @skb: sk_buff to place the data into
1987  * @size: size of data in rx_buffer
1988  *
1989  * This function will add the data contained in rx_buffer->page to the skb.
1990  * This is done either through a direct copy if the data in the buffer is
1991  * less than the skb header size, otherwise it will just attach the page as
1992  * a frag to the skb.
1993  *
1994  * The function will then update the page offset if necessary and return
1995  * true if the buffer can be reused by the adapter.
1996  **/
1997 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1998 			      struct ixgbe_rx_buffer *rx_buffer,
1999 			      struct sk_buff *skb,
2000 			      unsigned int size)
2001 {
2002 #if (PAGE_SIZE < 8192)
2003 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2004 #else
2005 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2006 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2007 				SKB_DATA_ALIGN(size);
2008 #endif
2009 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2010 			rx_buffer->page_offset, size, truesize);
2011 #if (PAGE_SIZE < 8192)
2012 	rx_buffer->page_offset ^= truesize;
2013 #else
2014 	rx_buffer->page_offset += truesize;
2015 #endif
2016 }
2017 
2018 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2019 						   union ixgbe_adv_rx_desc *rx_desc,
2020 						   struct sk_buff **skb,
2021 						   const unsigned int size)
2022 {
2023 	struct ixgbe_rx_buffer *rx_buffer;
2024 
2025 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2026 	prefetchw(rx_buffer->page);
2027 	*skb = rx_buffer->skb;
2028 
2029 	/* Delay unmapping of the first packet. It carries the header
2030 	 * information, HW may still access the header after the writeback.
2031 	 * Only unmap it when EOP is reached
2032 	 */
2033 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2034 		if (!*skb)
2035 			goto skip_sync;
2036 	} else {
2037 		if (*skb)
2038 			ixgbe_dma_sync_frag(rx_ring, *skb);
2039 	}
2040 
2041 	/* we are reusing so sync this buffer for CPU use */
2042 	dma_sync_single_range_for_cpu(rx_ring->dev,
2043 				      rx_buffer->dma,
2044 				      rx_buffer->page_offset,
2045 				      size,
2046 				      DMA_FROM_DEVICE);
2047 skip_sync:
2048 	rx_buffer->pagecnt_bias--;
2049 
2050 	return rx_buffer;
2051 }
2052 
2053 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2054 				struct ixgbe_rx_buffer *rx_buffer,
2055 				struct sk_buff *skb)
2056 {
2057 	if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2058 		/* hand second half of page back to the ring */
2059 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2060 	} else {
2061 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2062 			/* the page has been released from the ring */
2063 			IXGBE_CB(skb)->page_released = true;
2064 		} else {
2065 			/* we are not reusing the buffer so unmap it */
2066 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2067 					     ixgbe_rx_pg_size(rx_ring),
2068 					     DMA_FROM_DEVICE,
2069 					     IXGBE_RX_DMA_ATTR);
2070 		}
2071 		__page_frag_cache_drain(rx_buffer->page,
2072 					rx_buffer->pagecnt_bias);
2073 	}
2074 
2075 	/* clear contents of rx_buffer */
2076 	rx_buffer->page = NULL;
2077 	rx_buffer->skb = NULL;
2078 }
2079 
2080 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2081 					   struct ixgbe_rx_buffer *rx_buffer,
2082 					   struct xdp_buff *xdp,
2083 					   union ixgbe_adv_rx_desc *rx_desc)
2084 {
2085 	unsigned int size = xdp->data_end - xdp->data;
2086 #if (PAGE_SIZE < 8192)
2087 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2088 #else
2089 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2090 					       xdp->data_hard_start);
2091 #endif
2092 	struct sk_buff *skb;
2093 
2094 	/* prefetch first cache line of first page */
2095 	prefetch(xdp->data);
2096 #if L1_CACHE_BYTES < 128
2097 	prefetch(xdp->data + L1_CACHE_BYTES);
2098 #endif
2099 	/* Note, we get here by enabling legacy-rx via:
2100 	 *
2101 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2102 	 *
2103 	 * In this mode, we currently get 0 extra XDP headroom as
2104 	 * opposed to having legacy-rx off, where we process XDP
2105 	 * packets going to stack via ixgbe_build_skb(). The latter
2106 	 * provides us currently with 192 bytes of headroom.
2107 	 *
2108 	 * For ixgbe_construct_skb() mode it means that the
2109 	 * xdp->data_meta will always point to xdp->data, since
2110 	 * the helper cannot expand the head. Should this ever
2111 	 * change in future for legacy-rx mode on, then lets also
2112 	 * add xdp->data_meta handling here.
2113 	 */
2114 
2115 	/* allocate a skb to store the frags */
2116 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2117 	if (unlikely(!skb))
2118 		return NULL;
2119 
2120 	if (size > IXGBE_RX_HDR_SIZE) {
2121 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2122 			IXGBE_CB(skb)->dma = rx_buffer->dma;
2123 
2124 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2125 				xdp->data - page_address(rx_buffer->page),
2126 				size, truesize);
2127 #if (PAGE_SIZE < 8192)
2128 		rx_buffer->page_offset ^= truesize;
2129 #else
2130 		rx_buffer->page_offset += truesize;
2131 #endif
2132 	} else {
2133 		memcpy(__skb_put(skb, size),
2134 		       xdp->data, ALIGN(size, sizeof(long)));
2135 		rx_buffer->pagecnt_bias++;
2136 	}
2137 
2138 	return skb;
2139 }
2140 
2141 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2142 				       struct ixgbe_rx_buffer *rx_buffer,
2143 				       struct xdp_buff *xdp,
2144 				       union ixgbe_adv_rx_desc *rx_desc)
2145 {
2146 	unsigned int metasize = xdp->data - xdp->data_meta;
2147 #if (PAGE_SIZE < 8192)
2148 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2149 #else
2150 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2151 				SKB_DATA_ALIGN(xdp->data_end -
2152 					       xdp->data_hard_start);
2153 #endif
2154 	struct sk_buff *skb;
2155 
2156 	/* Prefetch first cache line of first page. If xdp->data_meta
2157 	 * is unused, this points extactly as xdp->data, otherwise we
2158 	 * likely have a consumer accessing first few bytes of meta
2159 	 * data, and then actual data.
2160 	 */
2161 	prefetch(xdp->data_meta);
2162 #if L1_CACHE_BYTES < 128
2163 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2164 #endif
2165 
2166 	/* build an skb to around the page buffer */
2167 	skb = build_skb(xdp->data_hard_start, truesize);
2168 	if (unlikely(!skb))
2169 		return NULL;
2170 
2171 	/* update pointers within the skb to store the data */
2172 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2173 	__skb_put(skb, xdp->data_end - xdp->data);
2174 	if (metasize)
2175 		skb_metadata_set(skb, metasize);
2176 
2177 	/* record DMA address if this is the start of a chain of buffers */
2178 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2179 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2180 
2181 	/* update buffer offset */
2182 #if (PAGE_SIZE < 8192)
2183 	rx_buffer->page_offset ^= truesize;
2184 #else
2185 	rx_buffer->page_offset += truesize;
2186 #endif
2187 
2188 	return skb;
2189 }
2190 
2191 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2192 				     struct ixgbe_ring *rx_ring,
2193 				     struct xdp_buff *xdp)
2194 {
2195 	int err, result = IXGBE_XDP_PASS;
2196 	struct bpf_prog *xdp_prog;
2197 	struct xdp_frame *xdpf;
2198 	u32 act;
2199 
2200 	rcu_read_lock();
2201 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2202 
2203 	if (!xdp_prog)
2204 		goto xdp_out;
2205 
2206 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2207 
2208 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2209 	switch (act) {
2210 	case XDP_PASS:
2211 		break;
2212 	case XDP_TX:
2213 		xdpf = convert_to_xdp_frame(xdp);
2214 		if (unlikely(!xdpf)) {
2215 			result = IXGBE_XDP_CONSUMED;
2216 			break;
2217 		}
2218 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2219 		break;
2220 	case XDP_REDIRECT:
2221 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2222 		if (!err)
2223 			result = IXGBE_XDP_REDIR;
2224 		else
2225 			result = IXGBE_XDP_CONSUMED;
2226 		break;
2227 	default:
2228 		bpf_warn_invalid_xdp_action(act);
2229 		/* fallthrough */
2230 	case XDP_ABORTED:
2231 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2232 		/* fallthrough -- handle aborts by dropping packet */
2233 	case XDP_DROP:
2234 		result = IXGBE_XDP_CONSUMED;
2235 		break;
2236 	}
2237 xdp_out:
2238 	rcu_read_unlock();
2239 	return ERR_PTR(-result);
2240 }
2241 
2242 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2243 				 struct ixgbe_rx_buffer *rx_buffer,
2244 				 unsigned int size)
2245 {
2246 #if (PAGE_SIZE < 8192)
2247 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2248 
2249 	rx_buffer->page_offset ^= truesize;
2250 #else
2251 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2252 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2253 				SKB_DATA_ALIGN(size);
2254 
2255 	rx_buffer->page_offset += truesize;
2256 #endif
2257 }
2258 
2259 /**
2260  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2261  * @q_vector: structure containing interrupt and ring information
2262  * @rx_ring: rx descriptor ring to transact packets on
2263  * @budget: Total limit on number of packets to process
2264  *
2265  * This function provides a "bounce buffer" approach to Rx interrupt
2266  * processing.  The advantage to this is that on systems that have
2267  * expensive overhead for IOMMU access this provides a means of avoiding
2268  * it by maintaining the mapping of the page to the syste.
2269  *
2270  * Returns amount of work completed
2271  **/
2272 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2273 			       struct ixgbe_ring *rx_ring,
2274 			       const int budget)
2275 {
2276 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2277 	struct ixgbe_adapter *adapter = q_vector->adapter;
2278 #ifdef IXGBE_FCOE
2279 	int ddp_bytes;
2280 	unsigned int mss = 0;
2281 #endif /* IXGBE_FCOE */
2282 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2283 	unsigned int xdp_xmit = 0;
2284 	struct xdp_buff xdp;
2285 
2286 	xdp.rxq = &rx_ring->xdp_rxq;
2287 
2288 	while (likely(total_rx_packets < budget)) {
2289 		union ixgbe_adv_rx_desc *rx_desc;
2290 		struct ixgbe_rx_buffer *rx_buffer;
2291 		struct sk_buff *skb;
2292 		unsigned int size;
2293 
2294 		/* return some buffers to hardware, one at a time is too slow */
2295 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2296 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2297 			cleaned_count = 0;
2298 		}
2299 
2300 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2301 		size = le16_to_cpu(rx_desc->wb.upper.length);
2302 		if (!size)
2303 			break;
2304 
2305 		/* This memory barrier is needed to keep us from reading
2306 		 * any other fields out of the rx_desc until we know the
2307 		 * descriptor has been written back
2308 		 */
2309 		dma_rmb();
2310 
2311 		rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2312 
2313 		/* retrieve a buffer from the ring */
2314 		if (!skb) {
2315 			xdp.data = page_address(rx_buffer->page) +
2316 				   rx_buffer->page_offset;
2317 			xdp.data_meta = xdp.data;
2318 			xdp.data_hard_start = xdp.data -
2319 					      ixgbe_rx_offset(rx_ring);
2320 			xdp.data_end = xdp.data + size;
2321 
2322 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2323 		}
2324 
2325 		if (IS_ERR(skb)) {
2326 			unsigned int xdp_res = -PTR_ERR(skb);
2327 
2328 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2329 				xdp_xmit |= xdp_res;
2330 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2331 			} else {
2332 				rx_buffer->pagecnt_bias++;
2333 			}
2334 			total_rx_packets++;
2335 			total_rx_bytes += size;
2336 		} else if (skb) {
2337 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2338 		} else if (ring_uses_build_skb(rx_ring)) {
2339 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2340 					      &xdp, rx_desc);
2341 		} else {
2342 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2343 						  &xdp, rx_desc);
2344 		}
2345 
2346 		/* exit if we failed to retrieve a buffer */
2347 		if (!skb) {
2348 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2349 			rx_buffer->pagecnt_bias++;
2350 			break;
2351 		}
2352 
2353 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2354 		cleaned_count++;
2355 
2356 		/* place incomplete frames back on ring for completion */
2357 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2358 			continue;
2359 
2360 		/* verify the packet layout is correct */
2361 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2362 			continue;
2363 
2364 		/* probably a little skewed due to removing CRC */
2365 		total_rx_bytes += skb->len;
2366 
2367 		/* populate checksum, timestamp, VLAN, and protocol */
2368 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2369 
2370 #ifdef IXGBE_FCOE
2371 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2372 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2373 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2374 			/* include DDPed FCoE data */
2375 			if (ddp_bytes > 0) {
2376 				if (!mss) {
2377 					mss = rx_ring->netdev->mtu -
2378 						sizeof(struct fcoe_hdr) -
2379 						sizeof(struct fc_frame_header) -
2380 						sizeof(struct fcoe_crc_eof);
2381 					if (mss > 512)
2382 						mss &= ~511;
2383 				}
2384 				total_rx_bytes += ddp_bytes;
2385 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2386 								 mss);
2387 			}
2388 			if (!ddp_bytes) {
2389 				dev_kfree_skb_any(skb);
2390 				continue;
2391 			}
2392 		}
2393 
2394 #endif /* IXGBE_FCOE */
2395 		ixgbe_rx_skb(q_vector, skb);
2396 
2397 		/* update budget accounting */
2398 		total_rx_packets++;
2399 	}
2400 
2401 	if (xdp_xmit & IXGBE_XDP_REDIR)
2402 		xdp_do_flush_map();
2403 
2404 	if (xdp_xmit & IXGBE_XDP_TX) {
2405 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2406 
2407 		/* Force memory writes to complete before letting h/w
2408 		 * know there are new descriptors to fetch.
2409 		 */
2410 		wmb();
2411 		writel(ring->next_to_use, ring->tail);
2412 	}
2413 
2414 	u64_stats_update_begin(&rx_ring->syncp);
2415 	rx_ring->stats.packets += total_rx_packets;
2416 	rx_ring->stats.bytes += total_rx_bytes;
2417 	u64_stats_update_end(&rx_ring->syncp);
2418 	q_vector->rx.total_packets += total_rx_packets;
2419 	q_vector->rx.total_bytes += total_rx_bytes;
2420 
2421 	return total_rx_packets;
2422 }
2423 
2424 /**
2425  * ixgbe_configure_msix - Configure MSI-X hardware
2426  * @adapter: board private structure
2427  *
2428  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2429  * interrupts.
2430  **/
2431 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2432 {
2433 	struct ixgbe_q_vector *q_vector;
2434 	int v_idx;
2435 	u32 mask;
2436 
2437 	/* Populate MSIX to EITR Select */
2438 	if (adapter->num_vfs > 32) {
2439 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2440 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2441 	}
2442 
2443 	/*
2444 	 * Populate the IVAR table and set the ITR values to the
2445 	 * corresponding register.
2446 	 */
2447 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2448 		struct ixgbe_ring *ring;
2449 		q_vector = adapter->q_vector[v_idx];
2450 
2451 		ixgbe_for_each_ring(ring, q_vector->rx)
2452 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2453 
2454 		ixgbe_for_each_ring(ring, q_vector->tx)
2455 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2456 
2457 		ixgbe_write_eitr(q_vector);
2458 	}
2459 
2460 	switch (adapter->hw.mac.type) {
2461 	case ixgbe_mac_82598EB:
2462 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2463 			       v_idx);
2464 		break;
2465 	case ixgbe_mac_82599EB:
2466 	case ixgbe_mac_X540:
2467 	case ixgbe_mac_X550:
2468 	case ixgbe_mac_X550EM_x:
2469 	case ixgbe_mac_x550em_a:
2470 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2471 		break;
2472 	default:
2473 		break;
2474 	}
2475 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2476 
2477 	/* set up to autoclear timer, and the vectors */
2478 	mask = IXGBE_EIMS_ENABLE_MASK;
2479 	mask &= ~(IXGBE_EIMS_OTHER |
2480 		  IXGBE_EIMS_MAILBOX |
2481 		  IXGBE_EIMS_LSC);
2482 
2483 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2484 }
2485 
2486 /**
2487  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2488  * @q_vector: structure containing interrupt and ring information
2489  * @ring_container: structure containing ring performance data
2490  *
2491  *      Stores a new ITR value based on packets and byte
2492  *      counts during the last interrupt.  The advantage of per interrupt
2493  *      computation is faster updates and more accurate ITR for the current
2494  *      traffic pattern.  Constants in this function were computed
2495  *      based on theoretical maximum wire speed and thresholds were set based
2496  *      on testing data as well as attempting to minimize response time
2497  *      while increasing bulk throughput.
2498  **/
2499 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2500 			     struct ixgbe_ring_container *ring_container)
2501 {
2502 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2503 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2504 	unsigned int avg_wire_size, packets, bytes;
2505 	unsigned long next_update = jiffies;
2506 
2507 	/* If we don't have any rings just leave ourselves set for maximum
2508 	 * possible latency so we take ourselves out of the equation.
2509 	 */
2510 	if (!ring_container->ring)
2511 		return;
2512 
2513 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2514 	 * that either packets are coming in so slow there hasn't been
2515 	 * any work, or that there is so much work that NAPI is dealing
2516 	 * with interrupt moderation and we don't need to do anything.
2517 	 */
2518 	if (time_after(next_update, ring_container->next_update))
2519 		goto clear_counts;
2520 
2521 	packets = ring_container->total_packets;
2522 
2523 	/* We have no packets to actually measure against. This means
2524 	 * either one of the other queues on this vector is active or
2525 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2526 	 *
2527 	 * When this occurs just tick up our delay by the minimum value
2528 	 * and hope that this extra delay will prevent us from being called
2529 	 * without any work on our queue.
2530 	 */
2531 	if (!packets) {
2532 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2533 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2534 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2535 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2536 		goto clear_counts;
2537 	}
2538 
2539 	bytes = ring_container->total_bytes;
2540 
2541 	/* If packets are less than 4 or bytes are less than 9000 assume
2542 	 * insufficient data to use bulk rate limiting approach. We are
2543 	 * likely latency driven.
2544 	 */
2545 	if (packets < 4 && bytes < 9000) {
2546 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2547 		goto adjust_by_size;
2548 	}
2549 
2550 	/* Between 4 and 48 we can assume that our current interrupt delay
2551 	 * is only slightly too low. As such we should increase it by a small
2552 	 * fixed amount.
2553 	 */
2554 	if (packets < 48) {
2555 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2556 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2557 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2558 		goto clear_counts;
2559 	}
2560 
2561 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2562 	 * out "just right". Just report that our current ITR is good for us.
2563 	 */
2564 	if (packets < 96) {
2565 		itr = q_vector->itr >> 2;
2566 		goto clear_counts;
2567 	}
2568 
2569 	/* If packet count is 96 or greater we are likely looking at a slight
2570 	 * overrun of the delay we want. Try halving our delay to see if that
2571 	 * will cut the number of packets in half per interrupt.
2572 	 */
2573 	if (packets < 256) {
2574 		itr = q_vector->itr >> 3;
2575 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2576 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2577 		goto clear_counts;
2578 	}
2579 
2580 	/* The paths below assume we are dealing with a bulk ITR since number
2581 	 * of packets is 256 or greater. We are just going to have to compute
2582 	 * a value and try to bring the count under control, though for smaller
2583 	 * packet sizes there isn't much we can do as NAPI polling will likely
2584 	 * be kicking in sooner rather than later.
2585 	 */
2586 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2587 
2588 adjust_by_size:
2589 	/* If packet counts are 256 or greater we can assume we have a gross
2590 	 * overestimation of what the rate should be. Instead of trying to fine
2591 	 * tune it just use the formula below to try and dial in an exact value
2592 	 * give the current packet size of the frame.
2593 	 */
2594 	avg_wire_size = bytes / packets;
2595 
2596 	/* The following is a crude approximation of:
2597 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2598 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2599 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2600 	 *
2601 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2602 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2603 	 * formula down to
2604 	 *
2605 	 *  (170 * (size + 24)) / (size + 640) = ITR
2606 	 *
2607 	 * We first do some math on the packet size and then finally bitshift
2608 	 * by 8 after rounding up. We also have to account for PCIe link speed
2609 	 * difference as ITR scales based on this.
2610 	 */
2611 	if (avg_wire_size <= 60) {
2612 		/* Start at 50k ints/sec */
2613 		avg_wire_size = 5120;
2614 	} else if (avg_wire_size <= 316) {
2615 		/* 50K ints/sec to 16K ints/sec */
2616 		avg_wire_size *= 40;
2617 		avg_wire_size += 2720;
2618 	} else if (avg_wire_size <= 1084) {
2619 		/* 16K ints/sec to 9.2K ints/sec */
2620 		avg_wire_size *= 15;
2621 		avg_wire_size += 11452;
2622 	} else if (avg_wire_size <= 1980) {
2623 		/* 9.2K ints/sec to 8K ints/sec */
2624 		avg_wire_size *= 5;
2625 		avg_wire_size += 22420;
2626 	} else {
2627 		/* plateau at a limit of 8K ints/sec */
2628 		avg_wire_size = 32256;
2629 	}
2630 
2631 	/* If we are in low latency mode half our delay which doubles the rate
2632 	 * to somewhere between 100K to 16K ints/sec
2633 	 */
2634 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2635 		avg_wire_size >>= 1;
2636 
2637 	/* Resultant value is 256 times larger than it needs to be. This
2638 	 * gives us room to adjust the value as needed to either increase
2639 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2640 	 *
2641 	 * Use addition as we have already recorded the new latency flag
2642 	 * for the ITR value.
2643 	 */
2644 	switch (q_vector->adapter->link_speed) {
2645 	case IXGBE_LINK_SPEED_10GB_FULL:
2646 	case IXGBE_LINK_SPEED_100_FULL:
2647 	default:
2648 		itr += DIV_ROUND_UP(avg_wire_size,
2649 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2650 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2651 		break;
2652 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2653 	case IXGBE_LINK_SPEED_1GB_FULL:
2654 	case IXGBE_LINK_SPEED_10_FULL:
2655 		itr += DIV_ROUND_UP(avg_wire_size,
2656 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2657 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2658 		break;
2659 	}
2660 
2661 clear_counts:
2662 	/* write back value */
2663 	ring_container->itr = itr;
2664 
2665 	/* next update should occur within next jiffy */
2666 	ring_container->next_update = next_update + 1;
2667 
2668 	ring_container->total_bytes = 0;
2669 	ring_container->total_packets = 0;
2670 }
2671 
2672 /**
2673  * ixgbe_write_eitr - write EITR register in hardware specific way
2674  * @q_vector: structure containing interrupt and ring information
2675  *
2676  * This function is made to be called by ethtool and by the driver
2677  * when it needs to update EITR registers at runtime.  Hardware
2678  * specific quirks/differences are taken care of here.
2679  */
2680 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2681 {
2682 	struct ixgbe_adapter *adapter = q_vector->adapter;
2683 	struct ixgbe_hw *hw = &adapter->hw;
2684 	int v_idx = q_vector->v_idx;
2685 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2686 
2687 	switch (adapter->hw.mac.type) {
2688 	case ixgbe_mac_82598EB:
2689 		/* must write high and low 16 bits to reset counter */
2690 		itr_reg |= (itr_reg << 16);
2691 		break;
2692 	case ixgbe_mac_82599EB:
2693 	case ixgbe_mac_X540:
2694 	case ixgbe_mac_X550:
2695 	case ixgbe_mac_X550EM_x:
2696 	case ixgbe_mac_x550em_a:
2697 		/*
2698 		 * set the WDIS bit to not clear the timer bits and cause an
2699 		 * immediate assertion of the interrupt
2700 		 */
2701 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2702 		break;
2703 	default:
2704 		break;
2705 	}
2706 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2707 }
2708 
2709 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2710 {
2711 	u32 new_itr;
2712 
2713 	ixgbe_update_itr(q_vector, &q_vector->tx);
2714 	ixgbe_update_itr(q_vector, &q_vector->rx);
2715 
2716 	/* use the smallest value of new ITR delay calculations */
2717 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2718 
2719 	/* Clear latency flag if set, shift into correct position */
2720 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2721 	new_itr <<= 2;
2722 
2723 	if (new_itr != q_vector->itr) {
2724 		/* save the algorithm value here */
2725 		q_vector->itr = new_itr;
2726 
2727 		ixgbe_write_eitr(q_vector);
2728 	}
2729 }
2730 
2731 /**
2732  * ixgbe_check_overtemp_subtask - check for over temperature
2733  * @adapter: pointer to adapter
2734  **/
2735 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2736 {
2737 	struct ixgbe_hw *hw = &adapter->hw;
2738 	u32 eicr = adapter->interrupt_event;
2739 	s32 rc;
2740 
2741 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2742 		return;
2743 
2744 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2745 		return;
2746 
2747 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2748 
2749 	switch (hw->device_id) {
2750 	case IXGBE_DEV_ID_82599_T3_LOM:
2751 		/*
2752 		 * Since the warning interrupt is for both ports
2753 		 * we don't have to check if:
2754 		 *  - This interrupt wasn't for our port.
2755 		 *  - We may have missed the interrupt so always have to
2756 		 *    check if we  got a LSC
2757 		 */
2758 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2759 		    !(eicr & IXGBE_EICR_LSC))
2760 			return;
2761 
2762 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2763 			u32 speed;
2764 			bool link_up = false;
2765 
2766 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2767 
2768 			if (link_up)
2769 				return;
2770 		}
2771 
2772 		/* Check if this is not due to overtemp */
2773 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2774 			return;
2775 
2776 		break;
2777 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2778 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2779 		rc = hw->phy.ops.check_overtemp(hw);
2780 		if (rc != IXGBE_ERR_OVERTEMP)
2781 			return;
2782 		break;
2783 	default:
2784 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2785 			return;
2786 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2787 			return;
2788 		break;
2789 	}
2790 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2791 
2792 	adapter->interrupt_event = 0;
2793 }
2794 
2795 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2796 {
2797 	struct ixgbe_hw *hw = &adapter->hw;
2798 
2799 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2800 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2801 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2802 		/* write to clear the interrupt */
2803 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2804 	}
2805 }
2806 
2807 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2808 {
2809 	struct ixgbe_hw *hw = &adapter->hw;
2810 
2811 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2812 		return;
2813 
2814 	switch (adapter->hw.mac.type) {
2815 	case ixgbe_mac_82599EB:
2816 		/*
2817 		 * Need to check link state so complete overtemp check
2818 		 * on service task
2819 		 */
2820 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2821 		     (eicr & IXGBE_EICR_LSC)) &&
2822 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2823 			adapter->interrupt_event = eicr;
2824 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2825 			ixgbe_service_event_schedule(adapter);
2826 			return;
2827 		}
2828 		return;
2829 	case ixgbe_mac_x550em_a:
2830 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2831 			adapter->interrupt_event = eicr;
2832 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2833 			ixgbe_service_event_schedule(adapter);
2834 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2835 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2836 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2837 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2838 		}
2839 		return;
2840 	case ixgbe_mac_X550:
2841 	case ixgbe_mac_X540:
2842 		if (!(eicr & IXGBE_EICR_TS))
2843 			return;
2844 		break;
2845 	default:
2846 		return;
2847 	}
2848 
2849 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2850 }
2851 
2852 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2853 {
2854 	switch (hw->mac.type) {
2855 	case ixgbe_mac_82598EB:
2856 		if (hw->phy.type == ixgbe_phy_nl)
2857 			return true;
2858 		return false;
2859 	case ixgbe_mac_82599EB:
2860 	case ixgbe_mac_X550EM_x:
2861 	case ixgbe_mac_x550em_a:
2862 		switch (hw->mac.ops.get_media_type(hw)) {
2863 		case ixgbe_media_type_fiber:
2864 		case ixgbe_media_type_fiber_qsfp:
2865 			return true;
2866 		default:
2867 			return false;
2868 		}
2869 	default:
2870 		return false;
2871 	}
2872 }
2873 
2874 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2875 {
2876 	struct ixgbe_hw *hw = &adapter->hw;
2877 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2878 
2879 	if (!ixgbe_is_sfp(hw))
2880 		return;
2881 
2882 	/* Later MAC's use different SDP */
2883 	if (hw->mac.type >= ixgbe_mac_X540)
2884 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2885 
2886 	if (eicr & eicr_mask) {
2887 		/* Clear the interrupt */
2888 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2889 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2890 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2891 			adapter->sfp_poll_time = 0;
2892 			ixgbe_service_event_schedule(adapter);
2893 		}
2894 	}
2895 
2896 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2897 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2898 		/* Clear the interrupt */
2899 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2900 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2901 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2902 			ixgbe_service_event_schedule(adapter);
2903 		}
2904 	}
2905 }
2906 
2907 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2908 {
2909 	struct ixgbe_hw *hw = &adapter->hw;
2910 
2911 	adapter->lsc_int++;
2912 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2913 	adapter->link_check_timeout = jiffies;
2914 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2915 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2916 		IXGBE_WRITE_FLUSH(hw);
2917 		ixgbe_service_event_schedule(adapter);
2918 	}
2919 }
2920 
2921 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2922 					   u64 qmask)
2923 {
2924 	u32 mask;
2925 	struct ixgbe_hw *hw = &adapter->hw;
2926 
2927 	switch (hw->mac.type) {
2928 	case ixgbe_mac_82598EB:
2929 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2930 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2931 		break;
2932 	case ixgbe_mac_82599EB:
2933 	case ixgbe_mac_X540:
2934 	case ixgbe_mac_X550:
2935 	case ixgbe_mac_X550EM_x:
2936 	case ixgbe_mac_x550em_a:
2937 		mask = (qmask & 0xFFFFFFFF);
2938 		if (mask)
2939 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2940 		mask = (qmask >> 32);
2941 		if (mask)
2942 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2943 		break;
2944 	default:
2945 		break;
2946 	}
2947 	/* skip the flush */
2948 }
2949 
2950 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2951 					    u64 qmask)
2952 {
2953 	u32 mask;
2954 	struct ixgbe_hw *hw = &adapter->hw;
2955 
2956 	switch (hw->mac.type) {
2957 	case ixgbe_mac_82598EB:
2958 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2959 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2960 		break;
2961 	case ixgbe_mac_82599EB:
2962 	case ixgbe_mac_X540:
2963 	case ixgbe_mac_X550:
2964 	case ixgbe_mac_X550EM_x:
2965 	case ixgbe_mac_x550em_a:
2966 		mask = (qmask & 0xFFFFFFFF);
2967 		if (mask)
2968 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2969 		mask = (qmask >> 32);
2970 		if (mask)
2971 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2972 		break;
2973 	default:
2974 		break;
2975 	}
2976 	/* skip the flush */
2977 }
2978 
2979 /**
2980  * ixgbe_irq_enable - Enable default interrupt generation settings
2981  * @adapter: board private structure
2982  * @queues: enable irqs for queues
2983  * @flush: flush register write
2984  **/
2985 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2986 				    bool flush)
2987 {
2988 	struct ixgbe_hw *hw = &adapter->hw;
2989 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2990 
2991 	/* don't reenable LSC while waiting for link */
2992 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2993 		mask &= ~IXGBE_EIMS_LSC;
2994 
2995 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2996 		switch (adapter->hw.mac.type) {
2997 		case ixgbe_mac_82599EB:
2998 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
2999 			break;
3000 		case ixgbe_mac_X540:
3001 		case ixgbe_mac_X550:
3002 		case ixgbe_mac_X550EM_x:
3003 		case ixgbe_mac_x550em_a:
3004 			mask |= IXGBE_EIMS_TS;
3005 			break;
3006 		default:
3007 			break;
3008 		}
3009 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3010 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3011 	switch (adapter->hw.mac.type) {
3012 	case ixgbe_mac_82599EB:
3013 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3014 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3015 		/* fall through */
3016 	case ixgbe_mac_X540:
3017 	case ixgbe_mac_X550:
3018 	case ixgbe_mac_X550EM_x:
3019 	case ixgbe_mac_x550em_a:
3020 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3021 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3022 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3023 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3024 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3025 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3026 		mask |= IXGBE_EIMS_ECC;
3027 		mask |= IXGBE_EIMS_MAILBOX;
3028 		break;
3029 	default:
3030 		break;
3031 	}
3032 
3033 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3034 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3035 		mask |= IXGBE_EIMS_FLOW_DIR;
3036 
3037 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3038 	if (queues)
3039 		ixgbe_irq_enable_queues(adapter, ~0);
3040 	if (flush)
3041 		IXGBE_WRITE_FLUSH(&adapter->hw);
3042 }
3043 
3044 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3045 {
3046 	struct ixgbe_adapter *adapter = data;
3047 	struct ixgbe_hw *hw = &adapter->hw;
3048 	u32 eicr;
3049 
3050 	/*
3051 	 * Workaround for Silicon errata.  Use clear-by-write instead
3052 	 * of clear-by-read.  Reading with EICS will return the
3053 	 * interrupt causes without clearing, which later be done
3054 	 * with the write to EICR.
3055 	 */
3056 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3057 
3058 	/* The lower 16bits of the EICR register are for the queue interrupts
3059 	 * which should be masked here in order to not accidentally clear them if
3060 	 * the bits are high when ixgbe_msix_other is called. There is a race
3061 	 * condition otherwise which results in possible performance loss
3062 	 * especially if the ixgbe_msix_other interrupt is triggering
3063 	 * consistently (as it would when PPS is turned on for the X540 device)
3064 	 */
3065 	eicr &= 0xFFFF0000;
3066 
3067 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3068 
3069 	if (eicr & IXGBE_EICR_LSC)
3070 		ixgbe_check_lsc(adapter);
3071 
3072 	if (eicr & IXGBE_EICR_MAILBOX)
3073 		ixgbe_msg_task(adapter);
3074 
3075 	switch (hw->mac.type) {
3076 	case ixgbe_mac_82599EB:
3077 	case ixgbe_mac_X540:
3078 	case ixgbe_mac_X550:
3079 	case ixgbe_mac_X550EM_x:
3080 	case ixgbe_mac_x550em_a:
3081 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3082 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3083 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3084 			ixgbe_service_event_schedule(adapter);
3085 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3086 					IXGBE_EICR_GPI_SDP0_X540);
3087 		}
3088 		if (eicr & IXGBE_EICR_ECC) {
3089 			e_info(link, "Received ECC Err, initiating reset\n");
3090 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3091 			ixgbe_service_event_schedule(adapter);
3092 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3093 		}
3094 		/* Handle Flow Director Full threshold interrupt */
3095 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3096 			int reinit_count = 0;
3097 			int i;
3098 			for (i = 0; i < adapter->num_tx_queues; i++) {
3099 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3100 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3101 						       &ring->state))
3102 					reinit_count++;
3103 			}
3104 			if (reinit_count) {
3105 				/* no more flow director interrupts until after init */
3106 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3107 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3108 				ixgbe_service_event_schedule(adapter);
3109 			}
3110 		}
3111 		ixgbe_check_sfp_event(adapter, eicr);
3112 		ixgbe_check_overtemp_event(adapter, eicr);
3113 		break;
3114 	default:
3115 		break;
3116 	}
3117 
3118 	ixgbe_check_fan_failure(adapter, eicr);
3119 
3120 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3121 		ixgbe_ptp_check_pps_event(adapter);
3122 
3123 	/* re-enable the original interrupt state, no lsc, no queues */
3124 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3125 		ixgbe_irq_enable(adapter, false, false);
3126 
3127 	return IRQ_HANDLED;
3128 }
3129 
3130 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3131 {
3132 	struct ixgbe_q_vector *q_vector = data;
3133 
3134 	/* EIAM disabled interrupts (on this vector) for us */
3135 
3136 	if (q_vector->rx.ring || q_vector->tx.ring)
3137 		napi_schedule_irqoff(&q_vector->napi);
3138 
3139 	return IRQ_HANDLED;
3140 }
3141 
3142 /**
3143  * ixgbe_poll - NAPI Rx polling callback
3144  * @napi: structure for representing this polling device
3145  * @budget: how many packets driver is allowed to clean
3146  *
3147  * This function is used for legacy and MSI, NAPI mode
3148  **/
3149 int ixgbe_poll(struct napi_struct *napi, int budget)
3150 {
3151 	struct ixgbe_q_vector *q_vector =
3152 				container_of(napi, struct ixgbe_q_vector, napi);
3153 	struct ixgbe_adapter *adapter = q_vector->adapter;
3154 	struct ixgbe_ring *ring;
3155 	int per_ring_budget, work_done = 0;
3156 	bool clean_complete = true;
3157 
3158 #ifdef CONFIG_IXGBE_DCA
3159 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3160 		ixgbe_update_dca(q_vector);
3161 #endif
3162 
3163 	ixgbe_for_each_ring(ring, q_vector->tx) {
3164 		bool wd = ring->xsk_umem ?
3165 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3166 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3167 
3168 		if (!wd)
3169 			clean_complete = false;
3170 	}
3171 
3172 	/* Exit if we are called by netpoll */
3173 	if (budget <= 0)
3174 		return budget;
3175 
3176 	/* attempt to distribute budget to each queue fairly, but don't allow
3177 	 * the budget to go below 1 because we'll exit polling */
3178 	if (q_vector->rx.count > 1)
3179 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3180 	else
3181 		per_ring_budget = budget;
3182 
3183 	ixgbe_for_each_ring(ring, q_vector->rx) {
3184 		int cleaned = ring->xsk_umem ?
3185 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3186 						    per_ring_budget) :
3187 			      ixgbe_clean_rx_irq(q_vector, ring,
3188 						 per_ring_budget);
3189 
3190 		work_done += cleaned;
3191 		if (cleaned >= per_ring_budget)
3192 			clean_complete = false;
3193 	}
3194 
3195 	/* If all work not completed, return budget and keep polling */
3196 	if (!clean_complete)
3197 		return budget;
3198 
3199 	/* all work done, exit the polling mode */
3200 	if (likely(napi_complete_done(napi, work_done))) {
3201 		if (adapter->rx_itr_setting & 1)
3202 			ixgbe_set_itr(q_vector);
3203 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3204 			ixgbe_irq_enable_queues(adapter,
3205 						BIT_ULL(q_vector->v_idx));
3206 	}
3207 
3208 	return min(work_done, budget - 1);
3209 }
3210 
3211 /**
3212  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3213  * @adapter: board private structure
3214  *
3215  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3216  * interrupts from the kernel.
3217  **/
3218 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3219 {
3220 	struct net_device *netdev = adapter->netdev;
3221 	unsigned int ri = 0, ti = 0;
3222 	int vector, err;
3223 
3224 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3225 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3226 		struct msix_entry *entry = &adapter->msix_entries[vector];
3227 
3228 		if (q_vector->tx.ring && q_vector->rx.ring) {
3229 			snprintf(q_vector->name, sizeof(q_vector->name),
3230 				 "%s-TxRx-%u", netdev->name, ri++);
3231 			ti++;
3232 		} else if (q_vector->rx.ring) {
3233 			snprintf(q_vector->name, sizeof(q_vector->name),
3234 				 "%s-rx-%u", netdev->name, ri++);
3235 		} else if (q_vector->tx.ring) {
3236 			snprintf(q_vector->name, sizeof(q_vector->name),
3237 				 "%s-tx-%u", netdev->name, ti++);
3238 		} else {
3239 			/* skip this unused q_vector */
3240 			continue;
3241 		}
3242 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3243 				  q_vector->name, q_vector);
3244 		if (err) {
3245 			e_err(probe, "request_irq failed for MSIX interrupt "
3246 			      "Error: %d\n", err);
3247 			goto free_queue_irqs;
3248 		}
3249 		/* If Flow Director is enabled, set interrupt affinity */
3250 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3251 			/* assign the mask for this irq */
3252 			irq_set_affinity_hint(entry->vector,
3253 					      &q_vector->affinity_mask);
3254 		}
3255 	}
3256 
3257 	err = request_irq(adapter->msix_entries[vector].vector,
3258 			  ixgbe_msix_other, 0, netdev->name, adapter);
3259 	if (err) {
3260 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3261 		goto free_queue_irqs;
3262 	}
3263 
3264 	return 0;
3265 
3266 free_queue_irqs:
3267 	while (vector) {
3268 		vector--;
3269 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3270 				      NULL);
3271 		free_irq(adapter->msix_entries[vector].vector,
3272 			 adapter->q_vector[vector]);
3273 	}
3274 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3275 	pci_disable_msix(adapter->pdev);
3276 	kfree(adapter->msix_entries);
3277 	adapter->msix_entries = NULL;
3278 	return err;
3279 }
3280 
3281 /**
3282  * ixgbe_intr - legacy mode Interrupt Handler
3283  * @irq: interrupt number
3284  * @data: pointer to a network interface device structure
3285  **/
3286 static irqreturn_t ixgbe_intr(int irq, void *data)
3287 {
3288 	struct ixgbe_adapter *adapter = data;
3289 	struct ixgbe_hw *hw = &adapter->hw;
3290 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3291 	u32 eicr;
3292 
3293 	/*
3294 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3295 	 * before the read of EICR.
3296 	 */
3297 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3298 
3299 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3300 	 * therefore no explicit interrupt disable is necessary */
3301 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3302 	if (!eicr) {
3303 		/*
3304 		 * shared interrupt alert!
3305 		 * make sure interrupts are enabled because the read will
3306 		 * have disabled interrupts due to EIAM
3307 		 * finish the workaround of silicon errata on 82598.  Unmask
3308 		 * the interrupt that we masked before the EICR read.
3309 		 */
3310 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3311 			ixgbe_irq_enable(adapter, true, true);
3312 		return IRQ_NONE;	/* Not our interrupt */
3313 	}
3314 
3315 	if (eicr & IXGBE_EICR_LSC)
3316 		ixgbe_check_lsc(adapter);
3317 
3318 	switch (hw->mac.type) {
3319 	case ixgbe_mac_82599EB:
3320 		ixgbe_check_sfp_event(adapter, eicr);
3321 		/* Fall through */
3322 	case ixgbe_mac_X540:
3323 	case ixgbe_mac_X550:
3324 	case ixgbe_mac_X550EM_x:
3325 	case ixgbe_mac_x550em_a:
3326 		if (eicr & IXGBE_EICR_ECC) {
3327 			e_info(link, "Received ECC Err, initiating reset\n");
3328 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3329 			ixgbe_service_event_schedule(adapter);
3330 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3331 		}
3332 		ixgbe_check_overtemp_event(adapter, eicr);
3333 		break;
3334 	default:
3335 		break;
3336 	}
3337 
3338 	ixgbe_check_fan_failure(adapter, eicr);
3339 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3340 		ixgbe_ptp_check_pps_event(adapter);
3341 
3342 	/* would disable interrupts here but EIAM disabled it */
3343 	napi_schedule_irqoff(&q_vector->napi);
3344 
3345 	/*
3346 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3347 	 * ixgbe_poll will re-enable the queue interrupts
3348 	 */
3349 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3350 		ixgbe_irq_enable(adapter, false, false);
3351 
3352 	return IRQ_HANDLED;
3353 }
3354 
3355 /**
3356  * ixgbe_request_irq - initialize interrupts
3357  * @adapter: board private structure
3358  *
3359  * Attempts to configure interrupts using the best available
3360  * capabilities of the hardware and kernel.
3361  **/
3362 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3363 {
3364 	struct net_device *netdev = adapter->netdev;
3365 	int err;
3366 
3367 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3368 		err = ixgbe_request_msix_irqs(adapter);
3369 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3370 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3371 				  netdev->name, adapter);
3372 	else
3373 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3374 				  netdev->name, adapter);
3375 
3376 	if (err)
3377 		e_err(probe, "request_irq failed, Error %d\n", err);
3378 
3379 	return err;
3380 }
3381 
3382 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3383 {
3384 	int vector;
3385 
3386 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3387 		free_irq(adapter->pdev->irq, adapter);
3388 		return;
3389 	}
3390 
3391 	if (!adapter->msix_entries)
3392 		return;
3393 
3394 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3395 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3396 		struct msix_entry *entry = &adapter->msix_entries[vector];
3397 
3398 		/* free only the irqs that were actually requested */
3399 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3400 			continue;
3401 
3402 		/* clear the affinity_mask in the IRQ descriptor */
3403 		irq_set_affinity_hint(entry->vector, NULL);
3404 
3405 		free_irq(entry->vector, q_vector);
3406 	}
3407 
3408 	free_irq(adapter->msix_entries[vector].vector, adapter);
3409 }
3410 
3411 /**
3412  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3413  * @adapter: board private structure
3414  **/
3415 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3416 {
3417 	switch (adapter->hw.mac.type) {
3418 	case ixgbe_mac_82598EB:
3419 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3420 		break;
3421 	case ixgbe_mac_82599EB:
3422 	case ixgbe_mac_X540:
3423 	case ixgbe_mac_X550:
3424 	case ixgbe_mac_X550EM_x:
3425 	case ixgbe_mac_x550em_a:
3426 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3427 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3428 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3429 		break;
3430 	default:
3431 		break;
3432 	}
3433 	IXGBE_WRITE_FLUSH(&adapter->hw);
3434 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3435 		int vector;
3436 
3437 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3438 			synchronize_irq(adapter->msix_entries[vector].vector);
3439 
3440 		synchronize_irq(adapter->msix_entries[vector++].vector);
3441 	} else {
3442 		synchronize_irq(adapter->pdev->irq);
3443 	}
3444 }
3445 
3446 /**
3447  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3448  * @adapter: board private structure
3449  *
3450  **/
3451 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3452 {
3453 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3454 
3455 	ixgbe_write_eitr(q_vector);
3456 
3457 	ixgbe_set_ivar(adapter, 0, 0, 0);
3458 	ixgbe_set_ivar(adapter, 1, 0, 0);
3459 
3460 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3461 }
3462 
3463 /**
3464  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3465  * @adapter: board private structure
3466  * @ring: structure containing ring specific data
3467  *
3468  * Configure the Tx descriptor ring after a reset.
3469  **/
3470 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3471 			     struct ixgbe_ring *ring)
3472 {
3473 	struct ixgbe_hw *hw = &adapter->hw;
3474 	u64 tdba = ring->dma;
3475 	int wait_loop = 10;
3476 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3477 	u8 reg_idx = ring->reg_idx;
3478 
3479 	ring->xsk_umem = NULL;
3480 	if (ring_is_xdp(ring))
3481 		ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3482 
3483 	/* disable queue to avoid issues while updating state */
3484 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3485 	IXGBE_WRITE_FLUSH(hw);
3486 
3487 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3488 			(tdba & DMA_BIT_MASK(32)));
3489 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3490 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3491 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3492 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3493 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3494 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3495 
3496 	/*
3497 	 * set WTHRESH to encourage burst writeback, it should not be set
3498 	 * higher than 1 when:
3499 	 * - ITR is 0 as it could cause false TX hangs
3500 	 * - ITR is set to > 100k int/sec and BQL is enabled
3501 	 *
3502 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3503 	 * to or less than the number of on chip descriptors, which is
3504 	 * currently 40.
3505 	 */
3506 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3507 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3508 	else
3509 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3510 
3511 	/*
3512 	 * Setting PTHRESH to 32 both improves performance
3513 	 * and avoids a TX hang with DFP enabled
3514 	 */
3515 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3516 		   32;		/* PTHRESH = 32 */
3517 
3518 	/* reinitialize flowdirector state */
3519 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3520 		ring->atr_sample_rate = adapter->atr_sample_rate;
3521 		ring->atr_count = 0;
3522 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3523 	} else {
3524 		ring->atr_sample_rate = 0;
3525 	}
3526 
3527 	/* initialize XPS */
3528 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3529 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3530 
3531 		if (q_vector)
3532 			netif_set_xps_queue(ring->netdev,
3533 					    &q_vector->affinity_mask,
3534 					    ring->queue_index);
3535 	}
3536 
3537 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3538 
3539 	/* reinitialize tx_buffer_info */
3540 	memset(ring->tx_buffer_info, 0,
3541 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3542 
3543 	/* enable queue */
3544 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3545 
3546 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3547 	if (hw->mac.type == ixgbe_mac_82598EB &&
3548 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3549 		return;
3550 
3551 	/* poll to verify queue is enabled */
3552 	do {
3553 		usleep_range(1000, 2000);
3554 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3555 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3556 	if (!wait_loop)
3557 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3558 }
3559 
3560 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3561 {
3562 	struct ixgbe_hw *hw = &adapter->hw;
3563 	u32 rttdcs, mtqc;
3564 	u8 tcs = adapter->hw_tcs;
3565 
3566 	if (hw->mac.type == ixgbe_mac_82598EB)
3567 		return;
3568 
3569 	/* disable the arbiter while setting MTQC */
3570 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3571 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3572 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3573 
3574 	/* set transmit pool layout */
3575 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3576 		mtqc = IXGBE_MTQC_VT_ENA;
3577 		if (tcs > 4)
3578 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3579 		else if (tcs > 1)
3580 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3581 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3582 			 IXGBE_82599_VMDQ_4Q_MASK)
3583 			mtqc |= IXGBE_MTQC_32VF;
3584 		else
3585 			mtqc |= IXGBE_MTQC_64VF;
3586 	} else {
3587 		if (tcs > 4) {
3588 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3589 		} else if (tcs > 1) {
3590 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3591 		} else {
3592 			u8 max_txq = adapter->num_tx_queues +
3593 				adapter->num_xdp_queues;
3594 			if (max_txq > 63)
3595 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3596 			else
3597 				mtqc = IXGBE_MTQC_64Q_1PB;
3598 		}
3599 	}
3600 
3601 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3602 
3603 	/* Enable Security TX Buffer IFG for multiple pb */
3604 	if (tcs) {
3605 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3606 		sectx |= IXGBE_SECTX_DCB;
3607 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3608 	}
3609 
3610 	/* re-enable the arbiter */
3611 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3612 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3613 }
3614 
3615 /**
3616  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3617  * @adapter: board private structure
3618  *
3619  * Configure the Tx unit of the MAC after a reset.
3620  **/
3621 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3622 {
3623 	struct ixgbe_hw *hw = &adapter->hw;
3624 	u32 dmatxctl;
3625 	u32 i;
3626 
3627 	ixgbe_setup_mtqc(adapter);
3628 
3629 	if (hw->mac.type != ixgbe_mac_82598EB) {
3630 		/* DMATXCTL.EN must be before Tx queues are enabled */
3631 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3632 		dmatxctl |= IXGBE_DMATXCTL_TE;
3633 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3634 	}
3635 
3636 	/* Setup the HW Tx Head and Tail descriptor pointers */
3637 	for (i = 0; i < adapter->num_tx_queues; i++)
3638 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3639 	for (i = 0; i < adapter->num_xdp_queues; i++)
3640 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3641 }
3642 
3643 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3644 				 struct ixgbe_ring *ring)
3645 {
3646 	struct ixgbe_hw *hw = &adapter->hw;
3647 	u8 reg_idx = ring->reg_idx;
3648 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3649 
3650 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3651 
3652 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3653 }
3654 
3655 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3656 				  struct ixgbe_ring *ring)
3657 {
3658 	struct ixgbe_hw *hw = &adapter->hw;
3659 	u8 reg_idx = ring->reg_idx;
3660 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3661 
3662 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3663 
3664 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3665 }
3666 
3667 #ifdef CONFIG_IXGBE_DCB
3668 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3669 #else
3670 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3671 #endif
3672 {
3673 	int i;
3674 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3675 
3676 	if (adapter->ixgbe_ieee_pfc)
3677 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3678 
3679 	/*
3680 	 * We should set the drop enable bit if:
3681 	 *  SR-IOV is enabled
3682 	 *   or
3683 	 *  Number of Rx queues > 1 and flow control is disabled
3684 	 *
3685 	 *  This allows us to avoid head of line blocking for security
3686 	 *  and performance reasons.
3687 	 */
3688 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3689 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3690 		for (i = 0; i < adapter->num_rx_queues; i++)
3691 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3692 	} else {
3693 		for (i = 0; i < adapter->num_rx_queues; i++)
3694 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3695 	}
3696 }
3697 
3698 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3699 
3700 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3701 				   struct ixgbe_ring *rx_ring)
3702 {
3703 	struct ixgbe_hw *hw = &adapter->hw;
3704 	u32 srrctl;
3705 	u8 reg_idx = rx_ring->reg_idx;
3706 
3707 	if (hw->mac.type == ixgbe_mac_82598EB) {
3708 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3709 
3710 		/*
3711 		 * if VMDq is not active we must program one srrctl register
3712 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3713 		 */
3714 		reg_idx &= mask;
3715 	}
3716 
3717 	/* configure header buffer length, needed for RSC */
3718 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3719 
3720 	/* configure the packet buffer length */
3721 	if (rx_ring->xsk_umem) {
3722 		u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3723 				  XDP_PACKET_HEADROOM;
3724 
3725 		/* If the MAC support setting RXDCTL.RLPML, the
3726 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3727 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3728 		 * size. If not, then we are stuck with a 1k buffer
3729 		 * size resolution. In this case frames larger than
3730 		 * the UMEM buffer size viewed in a 1k resolution will
3731 		 * be dropped.
3732 		 */
3733 		if (hw->mac.type != ixgbe_mac_82599EB)
3734 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3735 		else
3736 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3737 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3738 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3739 	} else {
3740 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3741 	}
3742 
3743 	/* configure descriptor type */
3744 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3745 
3746 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3747 }
3748 
3749 /**
3750  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3751  * @adapter: device handle
3752  *
3753  *  - 82598/82599/X540:     128
3754  *  - X550(non-SRIOV mode): 512
3755  *  - X550(SRIOV mode):     64
3756  */
3757 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3758 {
3759 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3760 		return 128;
3761 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3762 		return 64;
3763 	else
3764 		return 512;
3765 }
3766 
3767 /**
3768  * ixgbe_store_key - Write the RSS key to HW
3769  * @adapter: device handle
3770  *
3771  * Write the RSS key stored in adapter.rss_key to HW.
3772  */
3773 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3774 {
3775 	struct ixgbe_hw *hw = &adapter->hw;
3776 	int i;
3777 
3778 	for (i = 0; i < 10; i++)
3779 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3780 }
3781 
3782 /**
3783  * ixgbe_init_rss_key - Initialize adapter RSS key
3784  * @adapter: device handle
3785  *
3786  * Allocates and initializes the RSS key if it is not allocated.
3787  **/
3788 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3789 {
3790 	u32 *rss_key;
3791 
3792 	if (!adapter->rss_key) {
3793 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3794 		if (unlikely(!rss_key))
3795 			return -ENOMEM;
3796 
3797 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3798 		adapter->rss_key = rss_key;
3799 	}
3800 
3801 	return 0;
3802 }
3803 
3804 /**
3805  * ixgbe_store_reta - Write the RETA table to HW
3806  * @adapter: device handle
3807  *
3808  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3809  */
3810 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3811 {
3812 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3813 	struct ixgbe_hw *hw = &adapter->hw;
3814 	u32 reta = 0;
3815 	u32 indices_multi;
3816 	u8 *indir_tbl = adapter->rss_indir_tbl;
3817 
3818 	/* Fill out the redirection table as follows:
3819 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3820 	 *    indices.
3821 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3822 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3823 	 */
3824 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3825 		indices_multi = 0x11;
3826 	else
3827 		indices_multi = 0x1;
3828 
3829 	/* Write redirection table to HW */
3830 	for (i = 0; i < reta_entries; i++) {
3831 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3832 		if ((i & 3) == 3) {
3833 			if (i < 128)
3834 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3835 			else
3836 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3837 						reta);
3838 			reta = 0;
3839 		}
3840 	}
3841 }
3842 
3843 /**
3844  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3845  * @adapter: device handle
3846  *
3847  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3848  */
3849 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3850 {
3851 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3852 	struct ixgbe_hw *hw = &adapter->hw;
3853 	u32 vfreta = 0;
3854 
3855 	/* Write redirection table to HW */
3856 	for (i = 0; i < reta_entries; i++) {
3857 		u16 pool = adapter->num_rx_pools;
3858 
3859 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3860 		if ((i & 3) != 3)
3861 			continue;
3862 
3863 		while (pool--)
3864 			IXGBE_WRITE_REG(hw,
3865 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3866 					vfreta);
3867 		vfreta = 0;
3868 	}
3869 }
3870 
3871 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3872 {
3873 	u32 i, j;
3874 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3875 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3876 
3877 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3878 	 * make full use of any rings they may have.  We will use the
3879 	 * PSRTYPE register to control how many rings we use within the PF.
3880 	 */
3881 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3882 		rss_i = 4;
3883 
3884 	/* Fill out hash function seeds */
3885 	ixgbe_store_key(adapter);
3886 
3887 	/* Fill out redirection table */
3888 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3889 
3890 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3891 		if (j == rss_i)
3892 			j = 0;
3893 
3894 		adapter->rss_indir_tbl[i] = j;
3895 	}
3896 
3897 	ixgbe_store_reta(adapter);
3898 }
3899 
3900 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3901 {
3902 	struct ixgbe_hw *hw = &adapter->hw;
3903 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3904 	int i, j;
3905 
3906 	/* Fill out hash function seeds */
3907 	for (i = 0; i < 10; i++) {
3908 		u16 pool = adapter->num_rx_pools;
3909 
3910 		while (pool--)
3911 			IXGBE_WRITE_REG(hw,
3912 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3913 					*(adapter->rss_key + i));
3914 	}
3915 
3916 	/* Fill out the redirection table */
3917 	for (i = 0, j = 0; i < 64; i++, j++) {
3918 		if (j == rss_i)
3919 			j = 0;
3920 
3921 		adapter->rss_indir_tbl[i] = j;
3922 	}
3923 
3924 	ixgbe_store_vfreta(adapter);
3925 }
3926 
3927 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3928 {
3929 	struct ixgbe_hw *hw = &adapter->hw;
3930 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3931 	u32 rxcsum;
3932 
3933 	/* Disable indicating checksum in descriptor, enables RSS hash */
3934 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3935 	rxcsum |= IXGBE_RXCSUM_PCSD;
3936 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3937 
3938 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3939 		if (adapter->ring_feature[RING_F_RSS].mask)
3940 			mrqc = IXGBE_MRQC_RSSEN;
3941 	} else {
3942 		u8 tcs = adapter->hw_tcs;
3943 
3944 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3945 			if (tcs > 4)
3946 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3947 			else if (tcs > 1)
3948 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3949 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3950 				 IXGBE_82599_VMDQ_4Q_MASK)
3951 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3952 			else
3953 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3954 
3955 			/* Enable L3/L4 for Tx Switched packets */
3956 			mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3957 		} else {
3958 			if (tcs > 4)
3959 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3960 			else if (tcs > 1)
3961 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3962 			else
3963 				mrqc = IXGBE_MRQC_RSSEN;
3964 		}
3965 	}
3966 
3967 	/* Perform hash on these packet types */
3968 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3969 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3970 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
3971 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3972 
3973 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3974 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3975 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3976 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3977 
3978 	if ((hw->mac.type >= ixgbe_mac_X550) &&
3979 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3980 		u16 pool = adapter->num_rx_pools;
3981 
3982 		/* Enable VF RSS mode */
3983 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3984 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3985 
3986 		/* Setup RSS through the VF registers */
3987 		ixgbe_setup_vfreta(adapter);
3988 		vfmrqc = IXGBE_MRQC_RSSEN;
3989 		vfmrqc |= rss_field;
3990 
3991 		while (pool--)
3992 			IXGBE_WRITE_REG(hw,
3993 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
3994 					vfmrqc);
3995 	} else {
3996 		ixgbe_setup_reta(adapter);
3997 		mrqc |= rss_field;
3998 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3999 	}
4000 }
4001 
4002 /**
4003  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4004  * @adapter: address of board private structure
4005  * @ring: structure containing ring specific data
4006  **/
4007 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4008 				   struct ixgbe_ring *ring)
4009 {
4010 	struct ixgbe_hw *hw = &adapter->hw;
4011 	u32 rscctrl;
4012 	u8 reg_idx = ring->reg_idx;
4013 
4014 	if (!ring_is_rsc_enabled(ring))
4015 		return;
4016 
4017 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4018 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4019 	/*
4020 	 * we must limit the number of descriptors so that the
4021 	 * total size of max desc * buf_len is not greater
4022 	 * than 65536
4023 	 */
4024 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4025 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4026 }
4027 
4028 #define IXGBE_MAX_RX_DESC_POLL 10
4029 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4030 				       struct ixgbe_ring *ring)
4031 {
4032 	struct ixgbe_hw *hw = &adapter->hw;
4033 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4034 	u32 rxdctl;
4035 	u8 reg_idx = ring->reg_idx;
4036 
4037 	if (ixgbe_removed(hw->hw_addr))
4038 		return;
4039 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4040 	if (hw->mac.type == ixgbe_mac_82598EB &&
4041 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4042 		return;
4043 
4044 	do {
4045 		usleep_range(1000, 2000);
4046 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4047 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4048 
4049 	if (!wait_loop) {
4050 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4051 		      "the polling period\n", reg_idx);
4052 	}
4053 }
4054 
4055 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4056 			     struct ixgbe_ring *ring)
4057 {
4058 	struct ixgbe_hw *hw = &adapter->hw;
4059 	union ixgbe_adv_rx_desc *rx_desc;
4060 	u64 rdba = ring->dma;
4061 	u32 rxdctl;
4062 	u8 reg_idx = ring->reg_idx;
4063 
4064 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4065 	ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4066 	if (ring->xsk_umem) {
4067 		ring->zca.free = ixgbe_zca_free;
4068 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4069 						   MEM_TYPE_ZERO_COPY,
4070 						   &ring->zca));
4071 
4072 	} else {
4073 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4074 						   MEM_TYPE_PAGE_SHARED, NULL));
4075 	}
4076 
4077 	/* disable queue to avoid use of these values while updating state */
4078 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4079 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4080 
4081 	/* write value back with RXDCTL.ENABLE bit cleared */
4082 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4083 	IXGBE_WRITE_FLUSH(hw);
4084 
4085 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4086 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4087 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4088 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4089 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4090 	IXGBE_WRITE_FLUSH(hw);
4091 
4092 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4093 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4094 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4095 
4096 	ixgbe_configure_srrctl(adapter, ring);
4097 	ixgbe_configure_rscctl(adapter, ring);
4098 
4099 	if (hw->mac.type == ixgbe_mac_82598EB) {
4100 		/*
4101 		 * enable cache line friendly hardware writes:
4102 		 * PTHRESH=32 descriptors (half the internal cache),
4103 		 * this also removes ugly rx_no_buffer_count increment
4104 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4105 		 * WTHRESH=8 burst writeback up to two cache lines
4106 		 */
4107 		rxdctl &= ~0x3FFFFF;
4108 		rxdctl |=  0x080420;
4109 #if (PAGE_SIZE < 8192)
4110 	/* RXDCTL.RLPML does not work on 82599 */
4111 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4112 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4113 			    IXGBE_RXDCTL_RLPML_EN);
4114 
4115 		/* Limit the maximum frame size so we don't overrun the skb.
4116 		 * This can happen in SRIOV mode when the MTU of the VF is
4117 		 * higher than the MTU of the PF.
4118 		 */
4119 		if (ring_uses_build_skb(ring) &&
4120 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4121 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4122 				  IXGBE_RXDCTL_RLPML_EN;
4123 #endif
4124 	}
4125 
4126 	if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4127 		u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4128 				  XDP_PACKET_HEADROOM;
4129 
4130 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4131 			    IXGBE_RXDCTL_RLPML_EN);
4132 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4133 
4134 		ring->rx_buf_len = xsk_buf_len;
4135 	}
4136 
4137 	/* initialize rx_buffer_info */
4138 	memset(ring->rx_buffer_info, 0,
4139 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4140 
4141 	/* initialize Rx descriptor 0 */
4142 	rx_desc = IXGBE_RX_DESC(ring, 0);
4143 	rx_desc->wb.upper.length = 0;
4144 
4145 	/* enable receive descriptor ring */
4146 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4147 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4148 
4149 	ixgbe_rx_desc_queue_enable(adapter, ring);
4150 	if (ring->xsk_umem)
4151 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4152 	else
4153 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4154 }
4155 
4156 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4157 {
4158 	struct ixgbe_hw *hw = &adapter->hw;
4159 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4160 	u16 pool = adapter->num_rx_pools;
4161 
4162 	/* PSRTYPE must be initialized in non 82598 adapters */
4163 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4164 		      IXGBE_PSRTYPE_UDPHDR |
4165 		      IXGBE_PSRTYPE_IPV4HDR |
4166 		      IXGBE_PSRTYPE_L2HDR |
4167 		      IXGBE_PSRTYPE_IPV6HDR;
4168 
4169 	if (hw->mac.type == ixgbe_mac_82598EB)
4170 		return;
4171 
4172 	if (rss_i > 3)
4173 		psrtype |= 2u << 29;
4174 	else if (rss_i > 1)
4175 		psrtype |= 1u << 29;
4176 
4177 	while (pool--)
4178 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4179 }
4180 
4181 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4182 {
4183 	struct ixgbe_hw *hw = &adapter->hw;
4184 	u16 pool = adapter->num_rx_pools;
4185 	u32 reg_offset, vf_shift, vmolr;
4186 	u32 gcr_ext, vmdctl;
4187 	int i;
4188 
4189 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4190 		return;
4191 
4192 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4193 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4194 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4195 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4196 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4197 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4198 
4199 	/* accept untagged packets until a vlan tag is
4200 	 * specifically set for the VMDQ queue/pool
4201 	 */
4202 	vmolr = IXGBE_VMOLR_AUPE;
4203 	while (pool--)
4204 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4205 
4206 	vf_shift = VMDQ_P(0) % 32;
4207 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4208 
4209 	/* Enable only the PF's pool for Tx/Rx */
4210 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4211 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4212 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4213 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4214 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4215 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4216 
4217 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4218 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4219 
4220 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4221 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4222 
4223 	/*
4224 	 * Set up VF register offsets for selected VT Mode,
4225 	 * i.e. 32 or 64 VFs for SR-IOV
4226 	 */
4227 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4228 	case IXGBE_82599_VMDQ_8Q_MASK:
4229 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4230 		break;
4231 	case IXGBE_82599_VMDQ_4Q_MASK:
4232 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4233 		break;
4234 	default:
4235 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4236 		break;
4237 	}
4238 
4239 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4240 
4241 	for (i = 0; i < adapter->num_vfs; i++) {
4242 		/* configure spoof checking */
4243 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4244 					  adapter->vfinfo[i].spoofchk_enabled);
4245 
4246 		/* Enable/Disable RSS query feature  */
4247 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4248 					  adapter->vfinfo[i].rss_query_enabled);
4249 	}
4250 }
4251 
4252 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4253 {
4254 	struct ixgbe_hw *hw = &adapter->hw;
4255 	struct net_device *netdev = adapter->netdev;
4256 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4257 	struct ixgbe_ring *rx_ring;
4258 	int i;
4259 	u32 mhadd, hlreg0;
4260 
4261 #ifdef IXGBE_FCOE
4262 	/* adjust max frame to be able to do baby jumbo for FCoE */
4263 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4264 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4265 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4266 
4267 #endif /* IXGBE_FCOE */
4268 
4269 	/* adjust max frame to be at least the size of a standard frame */
4270 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4271 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4272 
4273 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4274 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4275 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4276 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4277 
4278 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4279 	}
4280 
4281 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4282 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4283 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4284 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4285 
4286 	/*
4287 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4288 	 * the Base and Length of the Rx Descriptor Ring
4289 	 */
4290 	for (i = 0; i < adapter->num_rx_queues; i++) {
4291 		rx_ring = adapter->rx_ring[i];
4292 
4293 		clear_ring_rsc_enabled(rx_ring);
4294 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4295 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4296 
4297 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4298 			set_ring_rsc_enabled(rx_ring);
4299 
4300 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4301 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4302 
4303 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4304 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4305 			continue;
4306 
4307 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4308 
4309 #if (PAGE_SIZE < 8192)
4310 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4311 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4312 
4313 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4314 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4315 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4316 #endif
4317 	}
4318 }
4319 
4320 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4321 {
4322 	struct ixgbe_hw *hw = &adapter->hw;
4323 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4324 
4325 	switch (hw->mac.type) {
4326 	case ixgbe_mac_82598EB:
4327 		/*
4328 		 * For VMDq support of different descriptor types or
4329 		 * buffer sizes through the use of multiple SRRCTL
4330 		 * registers, RDRXCTL.MVMEN must be set to 1
4331 		 *
4332 		 * also, the manual doesn't mention it clearly but DCA hints
4333 		 * will only use queue 0's tags unless this bit is set.  Side
4334 		 * effects of setting this bit are only that SRRCTL must be
4335 		 * fully programmed [0..15]
4336 		 */
4337 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4338 		break;
4339 	case ixgbe_mac_X550:
4340 	case ixgbe_mac_X550EM_x:
4341 	case ixgbe_mac_x550em_a:
4342 		if (adapter->num_vfs)
4343 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4344 		/* fall through */
4345 	case ixgbe_mac_82599EB:
4346 	case ixgbe_mac_X540:
4347 		/* Disable RSC for ACK packets */
4348 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4349 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4350 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4351 		/* hardware requires some bits to be set by default */
4352 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4353 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4354 		break;
4355 	default:
4356 		/* We should do nothing since we don't know this hardware */
4357 		return;
4358 	}
4359 
4360 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4361 }
4362 
4363 /**
4364  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4365  * @adapter: board private structure
4366  *
4367  * Configure the Rx unit of the MAC after a reset.
4368  **/
4369 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4370 {
4371 	struct ixgbe_hw *hw = &adapter->hw;
4372 	int i;
4373 	u32 rxctrl, rfctl;
4374 
4375 	/* disable receives while setting up the descriptors */
4376 	hw->mac.ops.disable_rx(hw);
4377 
4378 	ixgbe_setup_psrtype(adapter);
4379 	ixgbe_setup_rdrxctl(adapter);
4380 
4381 	/* RSC Setup */
4382 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4383 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4384 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4385 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4386 
4387 	/* disable NFS filtering */
4388 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4389 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4390 
4391 	/* Program registers for the distribution of queues */
4392 	ixgbe_setup_mrqc(adapter);
4393 
4394 	/* set_rx_buffer_len must be called before ring initialization */
4395 	ixgbe_set_rx_buffer_len(adapter);
4396 
4397 	/*
4398 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4399 	 * the Base and Length of the Rx Descriptor Ring
4400 	 */
4401 	for (i = 0; i < adapter->num_rx_queues; i++)
4402 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4403 
4404 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4405 	/* disable drop enable for 82598 parts */
4406 	if (hw->mac.type == ixgbe_mac_82598EB)
4407 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4408 
4409 	/* enable all receives */
4410 	rxctrl |= IXGBE_RXCTRL_RXEN;
4411 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4412 }
4413 
4414 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4415 				 __be16 proto, u16 vid)
4416 {
4417 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4418 	struct ixgbe_hw *hw = &adapter->hw;
4419 
4420 	/* add VID to filter table */
4421 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4422 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4423 
4424 	set_bit(vid, adapter->active_vlans);
4425 
4426 	return 0;
4427 }
4428 
4429 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4430 {
4431 	u32 vlvf;
4432 	int idx;
4433 
4434 	/* short cut the special case */
4435 	if (vlan == 0)
4436 		return 0;
4437 
4438 	/* Search for the vlan id in the VLVF entries */
4439 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4440 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4441 		if ((vlvf & VLAN_VID_MASK) == vlan)
4442 			break;
4443 	}
4444 
4445 	return idx;
4446 }
4447 
4448 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4449 {
4450 	struct ixgbe_hw *hw = &adapter->hw;
4451 	u32 bits, word;
4452 	int idx;
4453 
4454 	idx = ixgbe_find_vlvf_entry(hw, vid);
4455 	if (!idx)
4456 		return;
4457 
4458 	/* See if any other pools are set for this VLAN filter
4459 	 * entry other than the PF.
4460 	 */
4461 	word = idx * 2 + (VMDQ_P(0) / 32);
4462 	bits = ~BIT(VMDQ_P(0) % 32);
4463 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4464 
4465 	/* Disable the filter so this falls into the default pool. */
4466 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4467 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4468 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4469 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4470 	}
4471 }
4472 
4473 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4474 				  __be16 proto, u16 vid)
4475 {
4476 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4477 	struct ixgbe_hw *hw = &adapter->hw;
4478 
4479 	/* remove VID from filter table */
4480 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4481 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4482 
4483 	clear_bit(vid, adapter->active_vlans);
4484 
4485 	return 0;
4486 }
4487 
4488 /**
4489  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4490  * @adapter: driver data
4491  */
4492 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4493 {
4494 	struct ixgbe_hw *hw = &adapter->hw;
4495 	u32 vlnctrl;
4496 	int i, j;
4497 
4498 	switch (hw->mac.type) {
4499 	case ixgbe_mac_82598EB:
4500 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4501 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4502 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4503 		break;
4504 	case ixgbe_mac_82599EB:
4505 	case ixgbe_mac_X540:
4506 	case ixgbe_mac_X550:
4507 	case ixgbe_mac_X550EM_x:
4508 	case ixgbe_mac_x550em_a:
4509 		for (i = 0; i < adapter->num_rx_queues; i++) {
4510 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4511 
4512 			if (!netif_is_ixgbe(ring->netdev))
4513 				continue;
4514 
4515 			j = ring->reg_idx;
4516 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4517 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4518 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4519 		}
4520 		break;
4521 	default:
4522 		break;
4523 	}
4524 }
4525 
4526 /**
4527  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4528  * @adapter: driver data
4529  */
4530 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4531 {
4532 	struct ixgbe_hw *hw = &adapter->hw;
4533 	u32 vlnctrl;
4534 	int i, j;
4535 
4536 	switch (hw->mac.type) {
4537 	case ixgbe_mac_82598EB:
4538 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4539 		vlnctrl |= IXGBE_VLNCTRL_VME;
4540 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4541 		break;
4542 	case ixgbe_mac_82599EB:
4543 	case ixgbe_mac_X540:
4544 	case ixgbe_mac_X550:
4545 	case ixgbe_mac_X550EM_x:
4546 	case ixgbe_mac_x550em_a:
4547 		for (i = 0; i < adapter->num_rx_queues; i++) {
4548 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4549 
4550 			if (!netif_is_ixgbe(ring->netdev))
4551 				continue;
4552 
4553 			j = ring->reg_idx;
4554 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4555 			vlnctrl |= IXGBE_RXDCTL_VME;
4556 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4557 		}
4558 		break;
4559 	default:
4560 		break;
4561 	}
4562 }
4563 
4564 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4565 {
4566 	struct ixgbe_hw *hw = &adapter->hw;
4567 	u32 vlnctrl, i;
4568 
4569 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4570 
4571 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4572 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4573 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4574 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4575 	} else {
4576 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4577 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4578 		return;
4579 	}
4580 
4581 	/* Nothing to do for 82598 */
4582 	if (hw->mac.type == ixgbe_mac_82598EB)
4583 		return;
4584 
4585 	/* We are already in VLAN promisc, nothing to do */
4586 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4587 		return;
4588 
4589 	/* Set flag so we don't redo unnecessary work */
4590 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4591 
4592 	/* Add PF to all active pools */
4593 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4594 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4595 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4596 
4597 		vlvfb |= BIT(VMDQ_P(0) % 32);
4598 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4599 	}
4600 
4601 	/* Set all bits in the VLAN filter table array */
4602 	for (i = hw->mac.vft_size; i--;)
4603 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4604 }
4605 
4606 #define VFTA_BLOCK_SIZE 8
4607 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4608 {
4609 	struct ixgbe_hw *hw = &adapter->hw;
4610 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4611 	u32 vid_start = vfta_offset * 32;
4612 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4613 	u32 i, vid, word, bits;
4614 
4615 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4616 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4617 
4618 		/* pull VLAN ID from VLVF */
4619 		vid = vlvf & VLAN_VID_MASK;
4620 
4621 		/* only concern outselves with a certain range */
4622 		if (vid < vid_start || vid >= vid_end)
4623 			continue;
4624 
4625 		if (vlvf) {
4626 			/* record VLAN ID in VFTA */
4627 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4628 
4629 			/* if PF is part of this then continue */
4630 			if (test_bit(vid, adapter->active_vlans))
4631 				continue;
4632 		}
4633 
4634 		/* remove PF from the pool */
4635 		word = i * 2 + VMDQ_P(0) / 32;
4636 		bits = ~BIT(VMDQ_P(0) % 32);
4637 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4638 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4639 	}
4640 
4641 	/* extract values from active_vlans and write back to VFTA */
4642 	for (i = VFTA_BLOCK_SIZE; i--;) {
4643 		vid = (vfta_offset + i) * 32;
4644 		word = vid / BITS_PER_LONG;
4645 		bits = vid % BITS_PER_LONG;
4646 
4647 		vfta[i] |= adapter->active_vlans[word] >> bits;
4648 
4649 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4650 	}
4651 }
4652 
4653 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4654 {
4655 	struct ixgbe_hw *hw = &adapter->hw;
4656 	u32 vlnctrl, i;
4657 
4658 	/* Set VLAN filtering to enabled */
4659 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4660 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4661 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4662 
4663 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4664 	    hw->mac.type == ixgbe_mac_82598EB)
4665 		return;
4666 
4667 	/* We are not in VLAN promisc, nothing to do */
4668 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4669 		return;
4670 
4671 	/* Set flag so we don't redo unnecessary work */
4672 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4673 
4674 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4675 		ixgbe_scrub_vfta(adapter, i);
4676 }
4677 
4678 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4679 {
4680 	u16 vid = 1;
4681 
4682 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4683 
4684 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4685 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4686 }
4687 
4688 /**
4689  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4690  * @netdev: network interface device structure
4691  *
4692  * Writes multicast address list to the MTA hash table.
4693  * Returns: -ENOMEM on failure
4694  *                0 on no addresses written
4695  *                X on writing X addresses to MTA
4696  **/
4697 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4698 {
4699 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4700 	struct ixgbe_hw *hw = &adapter->hw;
4701 
4702 	if (!netif_running(netdev))
4703 		return 0;
4704 
4705 	if (hw->mac.ops.update_mc_addr_list)
4706 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4707 	else
4708 		return -ENOMEM;
4709 
4710 #ifdef CONFIG_PCI_IOV
4711 	ixgbe_restore_vf_multicasts(adapter);
4712 #endif
4713 
4714 	return netdev_mc_count(netdev);
4715 }
4716 
4717 #ifdef CONFIG_PCI_IOV
4718 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4719 {
4720 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4721 	struct ixgbe_hw *hw = &adapter->hw;
4722 	int i;
4723 
4724 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4725 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4726 
4727 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4728 			hw->mac.ops.set_rar(hw, i,
4729 					    mac_table->addr,
4730 					    mac_table->pool,
4731 					    IXGBE_RAH_AV);
4732 		else
4733 			hw->mac.ops.clear_rar(hw, i);
4734 	}
4735 }
4736 
4737 #endif
4738 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4739 {
4740 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4741 	struct ixgbe_hw *hw = &adapter->hw;
4742 	int i;
4743 
4744 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4745 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4746 			continue;
4747 
4748 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4749 
4750 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4751 			hw->mac.ops.set_rar(hw, i,
4752 					    mac_table->addr,
4753 					    mac_table->pool,
4754 					    IXGBE_RAH_AV);
4755 		else
4756 			hw->mac.ops.clear_rar(hw, i);
4757 	}
4758 }
4759 
4760 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4761 {
4762 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4763 	struct ixgbe_hw *hw = &adapter->hw;
4764 	int i;
4765 
4766 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4767 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4768 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4769 	}
4770 
4771 	ixgbe_sync_mac_table(adapter);
4772 }
4773 
4774 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4775 {
4776 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4777 	struct ixgbe_hw *hw = &adapter->hw;
4778 	int i, count = 0;
4779 
4780 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4781 		/* do not count default RAR as available */
4782 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4783 			continue;
4784 
4785 		/* only count unused and addresses that belong to us */
4786 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4787 			if (mac_table->pool != pool)
4788 				continue;
4789 		}
4790 
4791 		count++;
4792 	}
4793 
4794 	return count;
4795 }
4796 
4797 /* this function destroys the first RAR entry */
4798 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4799 {
4800 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4801 	struct ixgbe_hw *hw = &adapter->hw;
4802 
4803 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4804 	mac_table->pool = VMDQ_P(0);
4805 
4806 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4807 
4808 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4809 			    IXGBE_RAH_AV);
4810 }
4811 
4812 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4813 			 const u8 *addr, u16 pool)
4814 {
4815 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4816 	struct ixgbe_hw *hw = &adapter->hw;
4817 	int i;
4818 
4819 	if (is_zero_ether_addr(addr))
4820 		return -EINVAL;
4821 
4822 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4823 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4824 			continue;
4825 
4826 		ether_addr_copy(mac_table->addr, addr);
4827 		mac_table->pool = pool;
4828 
4829 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4830 				    IXGBE_MAC_STATE_IN_USE;
4831 
4832 		ixgbe_sync_mac_table(adapter);
4833 
4834 		return i;
4835 	}
4836 
4837 	return -ENOMEM;
4838 }
4839 
4840 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4841 			 const u8 *addr, u16 pool)
4842 {
4843 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4844 	struct ixgbe_hw *hw = &adapter->hw;
4845 	int i;
4846 
4847 	if (is_zero_ether_addr(addr))
4848 		return -EINVAL;
4849 
4850 	/* search table for addr, if found clear IN_USE flag and sync */
4851 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4852 		/* we can only delete an entry if it is in use */
4853 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4854 			continue;
4855 		/* we only care about entries that belong to the given pool */
4856 		if (mac_table->pool != pool)
4857 			continue;
4858 		/* we only care about a specific MAC address */
4859 		if (!ether_addr_equal(addr, mac_table->addr))
4860 			continue;
4861 
4862 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4863 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4864 
4865 		ixgbe_sync_mac_table(adapter);
4866 
4867 		return 0;
4868 	}
4869 
4870 	return -ENOMEM;
4871 }
4872 
4873 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4874 {
4875 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4876 	int ret;
4877 
4878 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4879 
4880 	return min_t(int, ret, 0);
4881 }
4882 
4883 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4884 {
4885 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4886 
4887 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4888 
4889 	return 0;
4890 }
4891 
4892 /**
4893  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4894  * @netdev: network interface device structure
4895  *
4896  * The set_rx_method entry point is called whenever the unicast/multicast
4897  * address list or the network interface flags are updated.  This routine is
4898  * responsible for configuring the hardware for proper unicast, multicast and
4899  * promiscuous mode.
4900  **/
4901 void ixgbe_set_rx_mode(struct net_device *netdev)
4902 {
4903 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4904 	struct ixgbe_hw *hw = &adapter->hw;
4905 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4906 	netdev_features_t features = netdev->features;
4907 	int count;
4908 
4909 	/* Check for Promiscuous and All Multicast modes */
4910 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4911 
4912 	/* set all bits that we expect to always be set */
4913 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4914 	fctrl |= IXGBE_FCTRL_BAM;
4915 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4916 	fctrl |= IXGBE_FCTRL_PMCF;
4917 
4918 	/* clear the bits we are changing the status of */
4919 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4920 	if (netdev->flags & IFF_PROMISC) {
4921 		hw->addr_ctrl.user_set_promisc = true;
4922 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4923 		vmolr |= IXGBE_VMOLR_MPE;
4924 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4925 	} else {
4926 		if (netdev->flags & IFF_ALLMULTI) {
4927 			fctrl |= IXGBE_FCTRL_MPE;
4928 			vmolr |= IXGBE_VMOLR_MPE;
4929 		}
4930 		hw->addr_ctrl.user_set_promisc = false;
4931 	}
4932 
4933 	/*
4934 	 * Write addresses to available RAR registers, if there is not
4935 	 * sufficient space to store all the addresses then enable
4936 	 * unicast promiscuous mode
4937 	 */
4938 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4939 		fctrl |= IXGBE_FCTRL_UPE;
4940 		vmolr |= IXGBE_VMOLR_ROPE;
4941 	}
4942 
4943 	/* Write addresses to the MTA, if the attempt fails
4944 	 * then we should just turn on promiscuous mode so
4945 	 * that we can at least receive multicast traffic
4946 	 */
4947 	count = ixgbe_write_mc_addr_list(netdev);
4948 	if (count < 0) {
4949 		fctrl |= IXGBE_FCTRL_MPE;
4950 		vmolr |= IXGBE_VMOLR_MPE;
4951 	} else if (count) {
4952 		vmolr |= IXGBE_VMOLR_ROMPE;
4953 	}
4954 
4955 	if (hw->mac.type != ixgbe_mac_82598EB) {
4956 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4957 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4958 			   IXGBE_VMOLR_ROPE);
4959 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4960 	}
4961 
4962 	/* This is useful for sniffing bad packets. */
4963 	if (features & NETIF_F_RXALL) {
4964 		/* UPE and MPE will be handled by normal PROMISC logic
4965 		 * in e1000e_set_rx_mode */
4966 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4967 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4968 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4969 
4970 		fctrl &= ~(IXGBE_FCTRL_DPF);
4971 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
4972 	}
4973 
4974 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4975 
4976 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
4977 		ixgbe_vlan_strip_enable(adapter);
4978 	else
4979 		ixgbe_vlan_strip_disable(adapter);
4980 
4981 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4982 		ixgbe_vlan_promisc_disable(adapter);
4983 	else
4984 		ixgbe_vlan_promisc_enable(adapter);
4985 }
4986 
4987 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4988 {
4989 	int q_idx;
4990 
4991 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4992 		napi_enable(&adapter->q_vector[q_idx]->napi);
4993 }
4994 
4995 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4996 {
4997 	int q_idx;
4998 
4999 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5000 		napi_disable(&adapter->q_vector[q_idx]->napi);
5001 }
5002 
5003 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5004 {
5005 	struct ixgbe_hw *hw = &adapter->hw;
5006 	u32 vxlanctrl;
5007 
5008 	if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5009 				IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5010 		return;
5011 
5012 	vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5013 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5014 
5015 	if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5016 		adapter->vxlan_port = 0;
5017 
5018 	if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5019 		adapter->geneve_port = 0;
5020 }
5021 
5022 #ifdef CONFIG_IXGBE_DCB
5023 /**
5024  * ixgbe_configure_dcb - Configure DCB hardware
5025  * @adapter: ixgbe adapter struct
5026  *
5027  * This is called by the driver on open to configure the DCB hardware.
5028  * This is also called by the gennetlink interface when reconfiguring
5029  * the DCB state.
5030  */
5031 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5032 {
5033 	struct ixgbe_hw *hw = &adapter->hw;
5034 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5035 
5036 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5037 		if (hw->mac.type == ixgbe_mac_82598EB)
5038 			netif_set_gso_max_size(adapter->netdev, 65536);
5039 		return;
5040 	}
5041 
5042 	if (hw->mac.type == ixgbe_mac_82598EB)
5043 		netif_set_gso_max_size(adapter->netdev, 32768);
5044 
5045 #ifdef IXGBE_FCOE
5046 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5047 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5048 #endif
5049 
5050 	/* reconfigure the hardware */
5051 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5052 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5053 						DCB_TX_CONFIG);
5054 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5055 						DCB_RX_CONFIG);
5056 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5057 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5058 		ixgbe_dcb_hw_ets(&adapter->hw,
5059 				 adapter->ixgbe_ieee_ets,
5060 				 max_frame);
5061 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5062 					adapter->ixgbe_ieee_pfc->pfc_en,
5063 					adapter->ixgbe_ieee_ets->prio_tc);
5064 	}
5065 
5066 	/* Enable RSS Hash per TC */
5067 	if (hw->mac.type != ixgbe_mac_82598EB) {
5068 		u32 msb = 0;
5069 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5070 
5071 		while (rss_i) {
5072 			msb++;
5073 			rss_i >>= 1;
5074 		}
5075 
5076 		/* write msb to all 8 TCs in one write */
5077 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5078 	}
5079 }
5080 #endif
5081 
5082 /* Additional bittime to account for IXGBE framing */
5083 #define IXGBE_ETH_FRAMING 20
5084 
5085 /**
5086  * ixgbe_hpbthresh - calculate high water mark for flow control
5087  *
5088  * @adapter: board private structure to calculate for
5089  * @pb: packet buffer to calculate
5090  */
5091 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5092 {
5093 	struct ixgbe_hw *hw = &adapter->hw;
5094 	struct net_device *dev = adapter->netdev;
5095 	int link, tc, kb, marker;
5096 	u32 dv_id, rx_pba;
5097 
5098 	/* Calculate max LAN frame size */
5099 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5100 
5101 #ifdef IXGBE_FCOE
5102 	/* FCoE traffic class uses FCOE jumbo frames */
5103 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5104 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5105 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5106 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5107 #endif
5108 
5109 	/* Calculate delay value for device */
5110 	switch (hw->mac.type) {
5111 	case ixgbe_mac_X540:
5112 	case ixgbe_mac_X550:
5113 	case ixgbe_mac_X550EM_x:
5114 	case ixgbe_mac_x550em_a:
5115 		dv_id = IXGBE_DV_X540(link, tc);
5116 		break;
5117 	default:
5118 		dv_id = IXGBE_DV(link, tc);
5119 		break;
5120 	}
5121 
5122 	/* Loopback switch introduces additional latency */
5123 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5124 		dv_id += IXGBE_B2BT(tc);
5125 
5126 	/* Delay value is calculated in bit times convert to KB */
5127 	kb = IXGBE_BT2KB(dv_id);
5128 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5129 
5130 	marker = rx_pba - kb;
5131 
5132 	/* It is possible that the packet buffer is not large enough
5133 	 * to provide required headroom. In this case throw an error
5134 	 * to user and a do the best we can.
5135 	 */
5136 	if (marker < 0) {
5137 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5138 			    "headroom to support flow control."
5139 			    "Decrease MTU or number of traffic classes\n", pb);
5140 		marker = tc + 1;
5141 	}
5142 
5143 	return marker;
5144 }
5145 
5146 /**
5147  * ixgbe_lpbthresh - calculate low water mark for for flow control
5148  *
5149  * @adapter: board private structure to calculate for
5150  * @pb: packet buffer to calculate
5151  */
5152 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5153 {
5154 	struct ixgbe_hw *hw = &adapter->hw;
5155 	struct net_device *dev = adapter->netdev;
5156 	int tc;
5157 	u32 dv_id;
5158 
5159 	/* Calculate max LAN frame size */
5160 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5161 
5162 #ifdef IXGBE_FCOE
5163 	/* FCoE traffic class uses FCOE jumbo frames */
5164 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5165 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5166 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5167 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5168 #endif
5169 
5170 	/* Calculate delay value for device */
5171 	switch (hw->mac.type) {
5172 	case ixgbe_mac_X540:
5173 	case ixgbe_mac_X550:
5174 	case ixgbe_mac_X550EM_x:
5175 	case ixgbe_mac_x550em_a:
5176 		dv_id = IXGBE_LOW_DV_X540(tc);
5177 		break;
5178 	default:
5179 		dv_id = IXGBE_LOW_DV(tc);
5180 		break;
5181 	}
5182 
5183 	/* Delay value is calculated in bit times convert to KB */
5184 	return IXGBE_BT2KB(dv_id);
5185 }
5186 
5187 /*
5188  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5189  */
5190 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5191 {
5192 	struct ixgbe_hw *hw = &adapter->hw;
5193 	int num_tc = adapter->hw_tcs;
5194 	int i;
5195 
5196 	if (!num_tc)
5197 		num_tc = 1;
5198 
5199 	for (i = 0; i < num_tc; i++) {
5200 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5201 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5202 
5203 		/* Low water marks must not be larger than high water marks */
5204 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5205 			hw->fc.low_water[i] = 0;
5206 	}
5207 
5208 	for (; i < MAX_TRAFFIC_CLASS; i++)
5209 		hw->fc.high_water[i] = 0;
5210 }
5211 
5212 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5213 {
5214 	struct ixgbe_hw *hw = &adapter->hw;
5215 	int hdrm;
5216 	u8 tc = adapter->hw_tcs;
5217 
5218 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5219 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5220 		hdrm = 32 << adapter->fdir_pballoc;
5221 	else
5222 		hdrm = 0;
5223 
5224 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5225 	ixgbe_pbthresh_setup(adapter);
5226 }
5227 
5228 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5229 {
5230 	struct ixgbe_hw *hw = &adapter->hw;
5231 	struct hlist_node *node2;
5232 	struct ixgbe_fdir_filter *filter;
5233 	u64 action;
5234 
5235 	spin_lock(&adapter->fdir_perfect_lock);
5236 
5237 	if (!hlist_empty(&adapter->fdir_filter_list))
5238 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5239 
5240 	hlist_for_each_entry_safe(filter, node2,
5241 				  &adapter->fdir_filter_list, fdir_node) {
5242 		action = filter->action;
5243 		if (action != IXGBE_FDIR_DROP_QUEUE && action != 0)
5244 			action =
5245 			(action >> ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF) - 1;
5246 
5247 		ixgbe_fdir_write_perfect_filter_82599(hw,
5248 				&filter->filter,
5249 				filter->sw_idx,
5250 				(action == IXGBE_FDIR_DROP_QUEUE) ?
5251 				IXGBE_FDIR_DROP_QUEUE :
5252 				adapter->rx_ring[action]->reg_idx);
5253 	}
5254 
5255 	spin_unlock(&adapter->fdir_perfect_lock);
5256 }
5257 
5258 /**
5259  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5260  * @rx_ring: ring to free buffers from
5261  **/
5262 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5263 {
5264 	u16 i = rx_ring->next_to_clean;
5265 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5266 
5267 	if (rx_ring->xsk_umem) {
5268 		ixgbe_xsk_clean_rx_ring(rx_ring);
5269 		goto skip_free;
5270 	}
5271 
5272 	/* Free all the Rx ring sk_buffs */
5273 	while (i != rx_ring->next_to_alloc) {
5274 		if (rx_buffer->skb) {
5275 			struct sk_buff *skb = rx_buffer->skb;
5276 			if (IXGBE_CB(skb)->page_released)
5277 				dma_unmap_page_attrs(rx_ring->dev,
5278 						     IXGBE_CB(skb)->dma,
5279 						     ixgbe_rx_pg_size(rx_ring),
5280 						     DMA_FROM_DEVICE,
5281 						     IXGBE_RX_DMA_ATTR);
5282 			dev_kfree_skb(skb);
5283 		}
5284 
5285 		/* Invalidate cache lines that may have been written to by
5286 		 * device so that we avoid corrupting memory.
5287 		 */
5288 		dma_sync_single_range_for_cpu(rx_ring->dev,
5289 					      rx_buffer->dma,
5290 					      rx_buffer->page_offset,
5291 					      ixgbe_rx_bufsz(rx_ring),
5292 					      DMA_FROM_DEVICE);
5293 
5294 		/* free resources associated with mapping */
5295 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5296 				     ixgbe_rx_pg_size(rx_ring),
5297 				     DMA_FROM_DEVICE,
5298 				     IXGBE_RX_DMA_ATTR);
5299 		__page_frag_cache_drain(rx_buffer->page,
5300 					rx_buffer->pagecnt_bias);
5301 
5302 		i++;
5303 		rx_buffer++;
5304 		if (i == rx_ring->count) {
5305 			i = 0;
5306 			rx_buffer = rx_ring->rx_buffer_info;
5307 		}
5308 	}
5309 
5310 skip_free:
5311 	rx_ring->next_to_alloc = 0;
5312 	rx_ring->next_to_clean = 0;
5313 	rx_ring->next_to_use = 0;
5314 }
5315 
5316 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5317 			     struct ixgbe_fwd_adapter *accel)
5318 {
5319 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5320 	int num_tc = netdev_get_num_tc(adapter->netdev);
5321 	struct net_device *vdev = accel->netdev;
5322 	int i, baseq, err;
5323 
5324 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5325 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5326 		   accel->pool, adapter->num_rx_pools,
5327 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5328 
5329 	accel->rx_base_queue = baseq;
5330 	accel->tx_base_queue = baseq;
5331 
5332 	/* record configuration for macvlan interface in vdev */
5333 	for (i = 0; i < num_tc; i++)
5334 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5335 					     i, rss_i, baseq + (rss_i * i));
5336 
5337 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5338 		adapter->rx_ring[baseq + i]->netdev = vdev;
5339 
5340 	/* Guarantee all rings are updated before we update the
5341 	 * MAC address filter.
5342 	 */
5343 	wmb();
5344 
5345 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5346 	 * need to only treat it as an error value if it is negative.
5347 	 */
5348 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5349 				   VMDQ_P(accel->pool));
5350 	if (err >= 0)
5351 		return 0;
5352 
5353 	/* if we cannot add the MAC rule then disable the offload */
5354 	macvlan_release_l2fw_offload(vdev);
5355 
5356 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5357 		adapter->rx_ring[baseq + i]->netdev = NULL;
5358 
5359 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5360 
5361 	/* unbind the queues and drop the subordinate channel config */
5362 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5363 	netdev_set_sb_channel(vdev, 0);
5364 
5365 	clear_bit(accel->pool, adapter->fwd_bitmask);
5366 	kfree(accel);
5367 
5368 	return err;
5369 }
5370 
5371 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5372 {
5373 	struct ixgbe_adapter *adapter = data;
5374 	struct ixgbe_fwd_adapter *accel;
5375 
5376 	if (!netif_is_macvlan(vdev))
5377 		return 0;
5378 
5379 	accel = macvlan_accel_priv(vdev);
5380 	if (!accel)
5381 		return 0;
5382 
5383 	ixgbe_fwd_ring_up(adapter, accel);
5384 
5385 	return 0;
5386 }
5387 
5388 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5389 {
5390 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5391 				      ixgbe_macvlan_up, adapter);
5392 }
5393 
5394 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5395 {
5396 	struct ixgbe_hw *hw = &adapter->hw;
5397 
5398 	ixgbe_configure_pb(adapter);
5399 #ifdef CONFIG_IXGBE_DCB
5400 	ixgbe_configure_dcb(adapter);
5401 #endif
5402 	/*
5403 	 * We must restore virtualization before VLANs or else
5404 	 * the VLVF registers will not be populated
5405 	 */
5406 	ixgbe_configure_virtualization(adapter);
5407 
5408 	ixgbe_set_rx_mode(adapter->netdev);
5409 	ixgbe_restore_vlan(adapter);
5410 	ixgbe_ipsec_restore(adapter);
5411 
5412 	switch (hw->mac.type) {
5413 	case ixgbe_mac_82599EB:
5414 	case ixgbe_mac_X540:
5415 		hw->mac.ops.disable_rx_buff(hw);
5416 		break;
5417 	default:
5418 		break;
5419 	}
5420 
5421 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5422 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5423 						adapter->fdir_pballoc);
5424 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5425 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5426 					      adapter->fdir_pballoc);
5427 		ixgbe_fdir_filter_restore(adapter);
5428 	}
5429 
5430 	switch (hw->mac.type) {
5431 	case ixgbe_mac_82599EB:
5432 	case ixgbe_mac_X540:
5433 		hw->mac.ops.enable_rx_buff(hw);
5434 		break;
5435 	default:
5436 		break;
5437 	}
5438 
5439 #ifdef CONFIG_IXGBE_DCA
5440 	/* configure DCA */
5441 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5442 		ixgbe_setup_dca(adapter);
5443 #endif /* CONFIG_IXGBE_DCA */
5444 
5445 #ifdef IXGBE_FCOE
5446 	/* configure FCoE L2 filters, redirection table, and Rx control */
5447 	ixgbe_configure_fcoe(adapter);
5448 
5449 #endif /* IXGBE_FCOE */
5450 	ixgbe_configure_tx(adapter);
5451 	ixgbe_configure_rx(adapter);
5452 	ixgbe_configure_dfwd(adapter);
5453 }
5454 
5455 /**
5456  * ixgbe_sfp_link_config - set up SFP+ link
5457  * @adapter: pointer to private adapter struct
5458  **/
5459 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5460 {
5461 	/*
5462 	 * We are assuming the worst case scenario here, and that
5463 	 * is that an SFP was inserted/removed after the reset
5464 	 * but before SFP detection was enabled.  As such the best
5465 	 * solution is to just start searching as soon as we start
5466 	 */
5467 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5468 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5469 
5470 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5471 	adapter->sfp_poll_time = 0;
5472 }
5473 
5474 /**
5475  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5476  * @hw: pointer to private hardware struct
5477  *
5478  * Returns 0 on success, negative on failure
5479  **/
5480 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5481 {
5482 	u32 speed;
5483 	bool autoneg, link_up = false;
5484 	int ret = IXGBE_ERR_LINK_SETUP;
5485 
5486 	if (hw->mac.ops.check_link)
5487 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5488 
5489 	if (ret)
5490 		return ret;
5491 
5492 	speed = hw->phy.autoneg_advertised;
5493 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
5494 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5495 							&autoneg);
5496 	if (ret)
5497 		return ret;
5498 
5499 	if (hw->mac.ops.setup_link)
5500 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5501 
5502 	return ret;
5503 }
5504 
5505 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5506 {
5507 	struct ixgbe_hw *hw = &adapter->hw;
5508 	u32 gpie = 0;
5509 
5510 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5511 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5512 		       IXGBE_GPIE_OCD;
5513 		gpie |= IXGBE_GPIE_EIAME;
5514 		/*
5515 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5516 		 * this saves a register write for every interrupt
5517 		 */
5518 		switch (hw->mac.type) {
5519 		case ixgbe_mac_82598EB:
5520 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5521 			break;
5522 		case ixgbe_mac_82599EB:
5523 		case ixgbe_mac_X540:
5524 		case ixgbe_mac_X550:
5525 		case ixgbe_mac_X550EM_x:
5526 		case ixgbe_mac_x550em_a:
5527 		default:
5528 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5529 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5530 			break;
5531 		}
5532 	} else {
5533 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5534 		 * specifically only auto mask tx and rx interrupts */
5535 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5536 	}
5537 
5538 	/* XXX: to interrupt immediately for EICS writes, enable this */
5539 	/* gpie |= IXGBE_GPIE_EIMEN; */
5540 
5541 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5542 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5543 
5544 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5545 		case IXGBE_82599_VMDQ_8Q_MASK:
5546 			gpie |= IXGBE_GPIE_VTMODE_16;
5547 			break;
5548 		case IXGBE_82599_VMDQ_4Q_MASK:
5549 			gpie |= IXGBE_GPIE_VTMODE_32;
5550 			break;
5551 		default:
5552 			gpie |= IXGBE_GPIE_VTMODE_64;
5553 			break;
5554 		}
5555 	}
5556 
5557 	/* Enable Thermal over heat sensor interrupt */
5558 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5559 		switch (adapter->hw.mac.type) {
5560 		case ixgbe_mac_82599EB:
5561 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5562 			break;
5563 		default:
5564 			break;
5565 		}
5566 	}
5567 
5568 	/* Enable fan failure interrupt */
5569 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5570 		gpie |= IXGBE_SDP1_GPIEN(hw);
5571 
5572 	switch (hw->mac.type) {
5573 	case ixgbe_mac_82599EB:
5574 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5575 		break;
5576 	case ixgbe_mac_X550EM_x:
5577 	case ixgbe_mac_x550em_a:
5578 		gpie |= IXGBE_SDP0_GPIEN_X540;
5579 		break;
5580 	default:
5581 		break;
5582 	}
5583 
5584 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5585 }
5586 
5587 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5588 {
5589 	struct ixgbe_hw *hw = &adapter->hw;
5590 	int err;
5591 	u32 ctrl_ext;
5592 
5593 	ixgbe_get_hw_control(adapter);
5594 	ixgbe_setup_gpie(adapter);
5595 
5596 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5597 		ixgbe_configure_msix(adapter);
5598 	else
5599 		ixgbe_configure_msi_and_legacy(adapter);
5600 
5601 	/* enable the optics for 82599 SFP+ fiber */
5602 	if (hw->mac.ops.enable_tx_laser)
5603 		hw->mac.ops.enable_tx_laser(hw);
5604 
5605 	if (hw->phy.ops.set_phy_power)
5606 		hw->phy.ops.set_phy_power(hw, true);
5607 
5608 	smp_mb__before_atomic();
5609 	clear_bit(__IXGBE_DOWN, &adapter->state);
5610 	ixgbe_napi_enable_all(adapter);
5611 
5612 	if (ixgbe_is_sfp(hw)) {
5613 		ixgbe_sfp_link_config(adapter);
5614 	} else {
5615 		err = ixgbe_non_sfp_link_config(hw);
5616 		if (err)
5617 			e_err(probe, "link_config FAILED %d\n", err);
5618 	}
5619 
5620 	/* clear any pending interrupts, may auto mask */
5621 	IXGBE_READ_REG(hw, IXGBE_EICR);
5622 	ixgbe_irq_enable(adapter, true, true);
5623 
5624 	/*
5625 	 * If this adapter has a fan, check to see if we had a failure
5626 	 * before we enabled the interrupt.
5627 	 */
5628 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5629 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5630 		if (esdp & IXGBE_ESDP_SDP1)
5631 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5632 	}
5633 
5634 	/* bring the link up in the watchdog, this could race with our first
5635 	 * link up interrupt but shouldn't be a problem */
5636 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5637 	adapter->link_check_timeout = jiffies;
5638 	mod_timer(&adapter->service_timer, jiffies);
5639 
5640 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5641 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5642 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5643 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5644 }
5645 
5646 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5647 {
5648 	WARN_ON(in_interrupt());
5649 	/* put off any impending NetWatchDogTimeout */
5650 	netif_trans_update(adapter->netdev);
5651 
5652 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5653 		usleep_range(1000, 2000);
5654 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5655 		ixgbe_watchdog_link_is_down(adapter);
5656 	ixgbe_down(adapter);
5657 	/*
5658 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5659 	 * back up to give the VFs time to respond to the reset.  The
5660 	 * two second wait is based upon the watchdog timer cycle in
5661 	 * the VF driver.
5662 	 */
5663 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5664 		msleep(2000);
5665 	ixgbe_up(adapter);
5666 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5667 }
5668 
5669 void ixgbe_up(struct ixgbe_adapter *adapter)
5670 {
5671 	/* hardware has been reset, we need to reload some things */
5672 	ixgbe_configure(adapter);
5673 
5674 	ixgbe_up_complete(adapter);
5675 }
5676 
5677 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5678 {
5679 	u16 devctl2;
5680 
5681 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5682 
5683 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5684 	case IXGBE_PCIDEVCTRL2_17_34s:
5685 	case IXGBE_PCIDEVCTRL2_4_8s:
5686 		/* For now we cap the upper limit on delay to 2 seconds
5687 		 * as we end up going up to 34 seconds of delay in worst
5688 		 * case timeout value.
5689 		 */
5690 	case IXGBE_PCIDEVCTRL2_1_2s:
5691 		return 2000000ul;	/* 2.0 s */
5692 	case IXGBE_PCIDEVCTRL2_260_520ms:
5693 		return 520000ul;	/* 520 ms */
5694 	case IXGBE_PCIDEVCTRL2_65_130ms:
5695 		return 130000ul;	/* 130 ms */
5696 	case IXGBE_PCIDEVCTRL2_16_32ms:
5697 		return 32000ul;		/* 32 ms */
5698 	case IXGBE_PCIDEVCTRL2_1_2ms:
5699 		return 2000ul;		/* 2 ms */
5700 	case IXGBE_PCIDEVCTRL2_50_100us:
5701 		return 100ul;		/* 100 us */
5702 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5703 		return 32000ul;		/* 32 ms */
5704 	default:
5705 		break;
5706 	}
5707 
5708 	/* We shouldn't need to hit this path, but just in case default as
5709 	 * though completion timeout is not supported and support 32ms.
5710 	 */
5711 	return 32000ul;
5712 }
5713 
5714 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5715 {
5716 	unsigned long wait_delay, delay_interval;
5717 	struct ixgbe_hw *hw = &adapter->hw;
5718 	int i, wait_loop;
5719 	u32 rxdctl;
5720 
5721 	/* disable receives */
5722 	hw->mac.ops.disable_rx(hw);
5723 
5724 	if (ixgbe_removed(hw->hw_addr))
5725 		return;
5726 
5727 	/* disable all enabled Rx queues */
5728 	for (i = 0; i < adapter->num_rx_queues; i++) {
5729 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5730 		u8 reg_idx = ring->reg_idx;
5731 
5732 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5733 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5734 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5735 
5736 		/* write value back with RXDCTL.ENABLE bit cleared */
5737 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5738 	}
5739 
5740 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5741 	if (hw->mac.type == ixgbe_mac_82598EB &&
5742 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5743 		return;
5744 
5745 	/* Determine our minimum delay interval. We will increase this value
5746 	 * with each subsequent test. This way if the device returns quickly
5747 	 * we should spend as little time as possible waiting, however as
5748 	 * the time increases we will wait for larger periods of time.
5749 	 *
5750 	 * The trick here is that we increase the interval using the
5751 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5752 	 * of that wait is that it totals up to 100x whatever interval we
5753 	 * choose. Since our minimum wait is 100us we can just divide the
5754 	 * total timeout by 100 to get our minimum delay interval.
5755 	 */
5756 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5757 
5758 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5759 	wait_delay = delay_interval;
5760 
5761 	while (wait_loop--) {
5762 		usleep_range(wait_delay, wait_delay + 10);
5763 		wait_delay += delay_interval * 2;
5764 		rxdctl = 0;
5765 
5766 		/* OR together the reading of all the active RXDCTL registers,
5767 		 * and then test the result. We need the disable to complete
5768 		 * before we start freeing the memory and invalidating the
5769 		 * DMA mappings.
5770 		 */
5771 		for (i = 0; i < adapter->num_rx_queues; i++) {
5772 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5773 			u8 reg_idx = ring->reg_idx;
5774 
5775 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5776 		}
5777 
5778 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5779 			return;
5780 	}
5781 
5782 	e_err(drv,
5783 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5784 }
5785 
5786 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5787 {
5788 	unsigned long wait_delay, delay_interval;
5789 	struct ixgbe_hw *hw = &adapter->hw;
5790 	int i, wait_loop;
5791 	u32 txdctl;
5792 
5793 	if (ixgbe_removed(hw->hw_addr))
5794 		return;
5795 
5796 	/* disable all enabled Tx queues */
5797 	for (i = 0; i < adapter->num_tx_queues; i++) {
5798 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5799 		u8 reg_idx = ring->reg_idx;
5800 
5801 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5802 	}
5803 
5804 	/* disable all enabled XDP Tx queues */
5805 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5806 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5807 		u8 reg_idx = ring->reg_idx;
5808 
5809 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5810 	}
5811 
5812 	/* If the link is not up there shouldn't be much in the way of
5813 	 * pending transactions. Those that are left will be flushed out
5814 	 * when the reset logic goes through the flush sequence to clean out
5815 	 * the pending Tx transactions.
5816 	 */
5817 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5818 		goto dma_engine_disable;
5819 
5820 	/* Determine our minimum delay interval. We will increase this value
5821 	 * with each subsequent test. This way if the device returns quickly
5822 	 * we should spend as little time as possible waiting, however as
5823 	 * the time increases we will wait for larger periods of time.
5824 	 *
5825 	 * The trick here is that we increase the interval using the
5826 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5827 	 * of that wait is that it totals up to 100x whatever interval we
5828 	 * choose. Since our minimum wait is 100us we can just divide the
5829 	 * total timeout by 100 to get our minimum delay interval.
5830 	 */
5831 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5832 
5833 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5834 	wait_delay = delay_interval;
5835 
5836 	while (wait_loop--) {
5837 		usleep_range(wait_delay, wait_delay + 10);
5838 		wait_delay += delay_interval * 2;
5839 		txdctl = 0;
5840 
5841 		/* OR together the reading of all the active TXDCTL registers,
5842 		 * and then test the result. We need the disable to complete
5843 		 * before we start freeing the memory and invalidating the
5844 		 * DMA mappings.
5845 		 */
5846 		for (i = 0; i < adapter->num_tx_queues; i++) {
5847 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5848 			u8 reg_idx = ring->reg_idx;
5849 
5850 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5851 		}
5852 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5853 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5854 			u8 reg_idx = ring->reg_idx;
5855 
5856 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5857 		}
5858 
5859 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5860 			goto dma_engine_disable;
5861 	}
5862 
5863 	e_err(drv,
5864 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5865 
5866 dma_engine_disable:
5867 	/* Disable the Tx DMA engine on 82599 and later MAC */
5868 	switch (hw->mac.type) {
5869 	case ixgbe_mac_82599EB:
5870 	case ixgbe_mac_X540:
5871 	case ixgbe_mac_X550:
5872 	case ixgbe_mac_X550EM_x:
5873 	case ixgbe_mac_x550em_a:
5874 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5875 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5876 				 ~IXGBE_DMATXCTL_TE));
5877 		/* fall through */
5878 	default:
5879 		break;
5880 	}
5881 }
5882 
5883 void ixgbe_reset(struct ixgbe_adapter *adapter)
5884 {
5885 	struct ixgbe_hw *hw = &adapter->hw;
5886 	struct net_device *netdev = adapter->netdev;
5887 	int err;
5888 
5889 	if (ixgbe_removed(hw->hw_addr))
5890 		return;
5891 	/* lock SFP init bit to prevent race conditions with the watchdog */
5892 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5893 		usleep_range(1000, 2000);
5894 
5895 	/* clear all SFP and link config related flags while holding SFP_INIT */
5896 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5897 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5898 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5899 
5900 	err = hw->mac.ops.init_hw(hw);
5901 	switch (err) {
5902 	case 0:
5903 	case IXGBE_ERR_SFP_NOT_PRESENT:
5904 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5905 		break;
5906 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5907 		e_dev_err("master disable timed out\n");
5908 		break;
5909 	case IXGBE_ERR_EEPROM_VERSION:
5910 		/* We are running on a pre-production device, log a warning */
5911 		e_dev_warn("This device is a pre-production adapter/LOM. "
5912 			   "Please be aware there may be issues associated with "
5913 			   "your hardware.  If you are experiencing problems "
5914 			   "please contact your Intel or hardware "
5915 			   "representative who provided you with this "
5916 			   "hardware.\n");
5917 		break;
5918 	default:
5919 		e_dev_err("Hardware Error: %d\n", err);
5920 	}
5921 
5922 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5923 
5924 	/* flush entries out of MAC table */
5925 	ixgbe_flush_sw_mac_table(adapter);
5926 	__dev_uc_unsync(netdev, NULL);
5927 
5928 	/* do not flush user set addresses */
5929 	ixgbe_mac_set_default_filter(adapter);
5930 
5931 	/* update SAN MAC vmdq pool selection */
5932 	if (hw->mac.san_mac_rar_index)
5933 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5934 
5935 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5936 		ixgbe_ptp_reset(adapter);
5937 
5938 	if (hw->phy.ops.set_phy_power) {
5939 		if (!netif_running(adapter->netdev) && !adapter->wol)
5940 			hw->phy.ops.set_phy_power(hw, false);
5941 		else
5942 			hw->phy.ops.set_phy_power(hw, true);
5943 	}
5944 }
5945 
5946 /**
5947  * ixgbe_clean_tx_ring - Free Tx Buffers
5948  * @tx_ring: ring to be cleaned
5949  **/
5950 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5951 {
5952 	u16 i = tx_ring->next_to_clean;
5953 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5954 
5955 	if (tx_ring->xsk_umem) {
5956 		ixgbe_xsk_clean_tx_ring(tx_ring);
5957 		goto out;
5958 	}
5959 
5960 	while (i != tx_ring->next_to_use) {
5961 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5962 
5963 		/* Free all the Tx ring sk_buffs */
5964 		if (ring_is_xdp(tx_ring))
5965 			xdp_return_frame(tx_buffer->xdpf);
5966 		else
5967 			dev_kfree_skb_any(tx_buffer->skb);
5968 
5969 		/* unmap skb header data */
5970 		dma_unmap_single(tx_ring->dev,
5971 				 dma_unmap_addr(tx_buffer, dma),
5972 				 dma_unmap_len(tx_buffer, len),
5973 				 DMA_TO_DEVICE);
5974 
5975 		/* check for eop_desc to determine the end of the packet */
5976 		eop_desc = tx_buffer->next_to_watch;
5977 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
5978 
5979 		/* unmap remaining buffers */
5980 		while (tx_desc != eop_desc) {
5981 			tx_buffer++;
5982 			tx_desc++;
5983 			i++;
5984 			if (unlikely(i == tx_ring->count)) {
5985 				i = 0;
5986 				tx_buffer = tx_ring->tx_buffer_info;
5987 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5988 			}
5989 
5990 			/* unmap any remaining paged data */
5991 			if (dma_unmap_len(tx_buffer, len))
5992 				dma_unmap_page(tx_ring->dev,
5993 					       dma_unmap_addr(tx_buffer, dma),
5994 					       dma_unmap_len(tx_buffer, len),
5995 					       DMA_TO_DEVICE);
5996 		}
5997 
5998 		/* move us one more past the eop_desc for start of next pkt */
5999 		tx_buffer++;
6000 		i++;
6001 		if (unlikely(i == tx_ring->count)) {
6002 			i = 0;
6003 			tx_buffer = tx_ring->tx_buffer_info;
6004 		}
6005 	}
6006 
6007 	/* reset BQL for queue */
6008 	if (!ring_is_xdp(tx_ring))
6009 		netdev_tx_reset_queue(txring_txq(tx_ring));
6010 
6011 out:
6012 	/* reset next_to_use and next_to_clean */
6013 	tx_ring->next_to_use = 0;
6014 	tx_ring->next_to_clean = 0;
6015 }
6016 
6017 /**
6018  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6019  * @adapter: board private structure
6020  **/
6021 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6022 {
6023 	int i;
6024 
6025 	for (i = 0; i < adapter->num_rx_queues; i++)
6026 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6027 }
6028 
6029 /**
6030  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6031  * @adapter: board private structure
6032  **/
6033 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6034 {
6035 	int i;
6036 
6037 	for (i = 0; i < adapter->num_tx_queues; i++)
6038 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6039 	for (i = 0; i < adapter->num_xdp_queues; i++)
6040 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6041 }
6042 
6043 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6044 {
6045 	struct hlist_node *node2;
6046 	struct ixgbe_fdir_filter *filter;
6047 
6048 	spin_lock(&adapter->fdir_perfect_lock);
6049 
6050 	hlist_for_each_entry_safe(filter, node2,
6051 				  &adapter->fdir_filter_list, fdir_node) {
6052 		hlist_del(&filter->fdir_node);
6053 		kfree(filter);
6054 	}
6055 	adapter->fdir_filter_count = 0;
6056 
6057 	spin_unlock(&adapter->fdir_perfect_lock);
6058 }
6059 
6060 void ixgbe_down(struct ixgbe_adapter *adapter)
6061 {
6062 	struct net_device *netdev = adapter->netdev;
6063 	struct ixgbe_hw *hw = &adapter->hw;
6064 	int i;
6065 
6066 	/* signal that we are down to the interrupt handler */
6067 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6068 		return; /* do nothing if already down */
6069 
6070 	/* Shut off incoming Tx traffic */
6071 	netif_tx_stop_all_queues(netdev);
6072 
6073 	/* call carrier off first to avoid false dev_watchdog timeouts */
6074 	netif_carrier_off(netdev);
6075 	netif_tx_disable(netdev);
6076 
6077 	/* Disable Rx */
6078 	ixgbe_disable_rx(adapter);
6079 
6080 	/* synchronize_sched() needed for pending XDP buffers to drain */
6081 	if (adapter->xdp_ring[0])
6082 		synchronize_sched();
6083 
6084 	ixgbe_irq_disable(adapter);
6085 
6086 	ixgbe_napi_disable_all(adapter);
6087 
6088 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6089 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6090 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6091 
6092 	del_timer_sync(&adapter->service_timer);
6093 
6094 	if (adapter->num_vfs) {
6095 		/* Clear EITR Select mapping */
6096 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6097 
6098 		/* Mark all the VFs as inactive */
6099 		for (i = 0 ; i < adapter->num_vfs; i++)
6100 			adapter->vfinfo[i].clear_to_send = false;
6101 
6102 		/* ping all the active vfs to let them know we are going down */
6103 		ixgbe_ping_all_vfs(adapter);
6104 
6105 		/* Disable all VFTE/VFRE TX/RX */
6106 		ixgbe_disable_tx_rx(adapter);
6107 	}
6108 
6109 	/* disable transmits in the hardware now that interrupts are off */
6110 	ixgbe_disable_tx(adapter);
6111 
6112 	if (!pci_channel_offline(adapter->pdev))
6113 		ixgbe_reset(adapter);
6114 
6115 	/* power down the optics for 82599 SFP+ fiber */
6116 	if (hw->mac.ops.disable_tx_laser)
6117 		hw->mac.ops.disable_tx_laser(hw);
6118 
6119 	ixgbe_clean_all_tx_rings(adapter);
6120 	ixgbe_clean_all_rx_rings(adapter);
6121 }
6122 
6123 /**
6124  * ixgbe_eee_capable - helper function to determine EEE support on X550
6125  * @adapter: board private structure
6126  */
6127 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6128 {
6129 	struct ixgbe_hw *hw = &adapter->hw;
6130 
6131 	switch (hw->device_id) {
6132 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6133 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6134 		if (!hw->phy.eee_speeds_supported)
6135 			break;
6136 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6137 		if (!hw->phy.eee_speeds_advertised)
6138 			break;
6139 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6140 		break;
6141 	default:
6142 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6143 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6144 		break;
6145 	}
6146 }
6147 
6148 /**
6149  * ixgbe_tx_timeout - Respond to a Tx Hang
6150  * @netdev: network interface device structure
6151  **/
6152 static void ixgbe_tx_timeout(struct net_device *netdev)
6153 {
6154 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6155 
6156 	/* Do the reset outside of interrupt context */
6157 	ixgbe_tx_timeout_reset(adapter);
6158 }
6159 
6160 #ifdef CONFIG_IXGBE_DCB
6161 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6162 {
6163 	struct ixgbe_hw *hw = &adapter->hw;
6164 	struct tc_configuration *tc;
6165 	int j;
6166 
6167 	switch (hw->mac.type) {
6168 	case ixgbe_mac_82598EB:
6169 	case ixgbe_mac_82599EB:
6170 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6171 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6172 		break;
6173 	case ixgbe_mac_X540:
6174 	case ixgbe_mac_X550:
6175 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6176 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6177 		break;
6178 	case ixgbe_mac_X550EM_x:
6179 	case ixgbe_mac_x550em_a:
6180 	default:
6181 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6182 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6183 		break;
6184 	}
6185 
6186 	/* Configure DCB traffic classes */
6187 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6188 		tc = &adapter->dcb_cfg.tc_config[j];
6189 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6190 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6191 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6192 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6193 		tc->dcb_pfc = pfc_disabled;
6194 	}
6195 
6196 	/* Initialize default user to priority mapping, UPx->TC0 */
6197 	tc = &adapter->dcb_cfg.tc_config[0];
6198 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6199 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6200 
6201 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6202 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6203 	adapter->dcb_cfg.pfc_mode_enable = false;
6204 	adapter->dcb_set_bitmap = 0x00;
6205 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6206 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6207 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6208 	       sizeof(adapter->temp_dcb_cfg));
6209 }
6210 #endif
6211 
6212 /**
6213  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6214  * @adapter: board private structure to initialize
6215  * @ii: pointer to ixgbe_info for device
6216  *
6217  * ixgbe_sw_init initializes the Adapter private data structure.
6218  * Fields are initialized based on PCI device information and
6219  * OS network device settings (MTU size).
6220  **/
6221 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6222 			 const struct ixgbe_info *ii)
6223 {
6224 	struct ixgbe_hw *hw = &adapter->hw;
6225 	struct pci_dev *pdev = adapter->pdev;
6226 	unsigned int rss, fdir;
6227 	u32 fwsm;
6228 	int i;
6229 
6230 	/* PCI config space info */
6231 
6232 	hw->vendor_id = pdev->vendor;
6233 	hw->device_id = pdev->device;
6234 	hw->revision_id = pdev->revision;
6235 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6236 	hw->subsystem_device_id = pdev->subsystem_device;
6237 
6238 	/* get_invariants needs the device IDs */
6239 	ii->get_invariants(hw);
6240 
6241 	/* Set common capability flags and settings */
6242 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6243 	adapter->ring_feature[RING_F_RSS].limit = rss;
6244 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6245 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6246 	adapter->atr_sample_rate = 20;
6247 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6248 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6249 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6250 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6251 #ifdef CONFIG_IXGBE_DCA
6252 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6253 #endif
6254 #ifdef CONFIG_IXGBE_DCB
6255 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6256 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6257 #endif
6258 #ifdef IXGBE_FCOE
6259 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6260 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6261 #ifdef CONFIG_IXGBE_DCB
6262 	/* Default traffic class to use for FCoE */
6263 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6264 #endif /* CONFIG_IXGBE_DCB */
6265 #endif /* IXGBE_FCOE */
6266 
6267 	/* initialize static ixgbe jump table entries */
6268 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6269 					  GFP_KERNEL);
6270 	if (!adapter->jump_tables[0])
6271 		return -ENOMEM;
6272 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6273 
6274 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6275 		adapter->jump_tables[i] = NULL;
6276 
6277 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6278 				     sizeof(struct ixgbe_mac_addr),
6279 				     GFP_KERNEL);
6280 	if (!adapter->mac_table)
6281 		return -ENOMEM;
6282 
6283 	if (ixgbe_init_rss_key(adapter))
6284 		return -ENOMEM;
6285 
6286 	/* Set MAC specific capability flags and exceptions */
6287 	switch (hw->mac.type) {
6288 	case ixgbe_mac_82598EB:
6289 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6290 
6291 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6292 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6293 
6294 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6295 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6296 		adapter->atr_sample_rate = 0;
6297 		adapter->fdir_pballoc = 0;
6298 #ifdef IXGBE_FCOE
6299 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6300 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6301 #ifdef CONFIG_IXGBE_DCB
6302 		adapter->fcoe.up = 0;
6303 #endif /* IXGBE_DCB */
6304 #endif /* IXGBE_FCOE */
6305 		break;
6306 	case ixgbe_mac_82599EB:
6307 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6308 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6309 		break;
6310 	case ixgbe_mac_X540:
6311 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6312 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6313 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6314 		break;
6315 	case ixgbe_mac_x550em_a:
6316 		adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6317 		switch (hw->device_id) {
6318 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6319 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6320 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6321 			break;
6322 		default:
6323 			break;
6324 		}
6325 	/* fall through */
6326 	case ixgbe_mac_X550EM_x:
6327 #ifdef CONFIG_IXGBE_DCB
6328 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6329 #endif
6330 #ifdef IXGBE_FCOE
6331 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6332 #ifdef CONFIG_IXGBE_DCB
6333 		adapter->fcoe.up = 0;
6334 #endif /* IXGBE_DCB */
6335 #endif /* IXGBE_FCOE */
6336 	/* Fall Through */
6337 	case ixgbe_mac_X550:
6338 		if (hw->mac.type == ixgbe_mac_X550)
6339 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6340 #ifdef CONFIG_IXGBE_DCA
6341 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6342 #endif
6343 		adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6344 		break;
6345 	default:
6346 		break;
6347 	}
6348 
6349 #ifdef IXGBE_FCOE
6350 	/* FCoE support exists, always init the FCoE lock */
6351 	spin_lock_init(&adapter->fcoe.lock);
6352 
6353 #endif
6354 	/* n-tuple support exists, always init our spinlock */
6355 	spin_lock_init(&adapter->fdir_perfect_lock);
6356 
6357 #ifdef CONFIG_IXGBE_DCB
6358 	ixgbe_init_dcb(adapter);
6359 #endif
6360 	ixgbe_init_ipsec_offload(adapter);
6361 
6362 	/* default flow control settings */
6363 	hw->fc.requested_mode = ixgbe_fc_full;
6364 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6365 	ixgbe_pbthresh_setup(adapter);
6366 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6367 	hw->fc.send_xon = true;
6368 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6369 
6370 #ifdef CONFIG_PCI_IOV
6371 	if (max_vfs > 0)
6372 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6373 
6374 	/* assign number of SR-IOV VFs */
6375 	if (hw->mac.type != ixgbe_mac_82598EB) {
6376 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6377 			max_vfs = 0;
6378 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6379 		}
6380 	}
6381 #endif /* CONFIG_PCI_IOV */
6382 
6383 	/* enable itr by default in dynamic mode */
6384 	adapter->rx_itr_setting = 1;
6385 	adapter->tx_itr_setting = 1;
6386 
6387 	/* set default ring sizes */
6388 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6389 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6390 
6391 	/* set default work limits */
6392 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6393 
6394 	/* initialize eeprom parameters */
6395 	if (ixgbe_init_eeprom_params_generic(hw)) {
6396 		e_dev_err("EEPROM initialization failed\n");
6397 		return -EIO;
6398 	}
6399 
6400 	/* PF holds first pool slot */
6401 	set_bit(0, adapter->fwd_bitmask);
6402 	set_bit(__IXGBE_DOWN, &adapter->state);
6403 
6404 	return 0;
6405 }
6406 
6407 /**
6408  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6409  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6410  *
6411  * Return 0 on success, negative on failure
6412  **/
6413 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6414 {
6415 	struct device *dev = tx_ring->dev;
6416 	int orig_node = dev_to_node(dev);
6417 	int ring_node = -1;
6418 	int size;
6419 
6420 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6421 
6422 	if (tx_ring->q_vector)
6423 		ring_node = tx_ring->q_vector->numa_node;
6424 
6425 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6426 	if (!tx_ring->tx_buffer_info)
6427 		tx_ring->tx_buffer_info = vmalloc(size);
6428 	if (!tx_ring->tx_buffer_info)
6429 		goto err;
6430 
6431 	/* round up to nearest 4K */
6432 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6433 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6434 
6435 	set_dev_node(dev, ring_node);
6436 	tx_ring->desc = dma_alloc_coherent(dev,
6437 					   tx_ring->size,
6438 					   &tx_ring->dma,
6439 					   GFP_KERNEL);
6440 	set_dev_node(dev, orig_node);
6441 	if (!tx_ring->desc)
6442 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6443 						   &tx_ring->dma, GFP_KERNEL);
6444 	if (!tx_ring->desc)
6445 		goto err;
6446 
6447 	tx_ring->next_to_use = 0;
6448 	tx_ring->next_to_clean = 0;
6449 	return 0;
6450 
6451 err:
6452 	vfree(tx_ring->tx_buffer_info);
6453 	tx_ring->tx_buffer_info = NULL;
6454 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6455 	return -ENOMEM;
6456 }
6457 
6458 /**
6459  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6460  * @adapter: board private structure
6461  *
6462  * If this function returns with an error, then it's possible one or
6463  * more of the rings is populated (while the rest are not).  It is the
6464  * callers duty to clean those orphaned rings.
6465  *
6466  * Return 0 on success, negative on failure
6467  **/
6468 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6469 {
6470 	int i, j = 0, err = 0;
6471 
6472 	for (i = 0; i < adapter->num_tx_queues; i++) {
6473 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6474 		if (!err)
6475 			continue;
6476 
6477 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6478 		goto err_setup_tx;
6479 	}
6480 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6481 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6482 		if (!err)
6483 			continue;
6484 
6485 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6486 		goto err_setup_tx;
6487 	}
6488 
6489 	return 0;
6490 err_setup_tx:
6491 	/* rewind the index freeing the rings as we go */
6492 	while (j--)
6493 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6494 	while (i--)
6495 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6496 	return err;
6497 }
6498 
6499 /**
6500  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6501  * @adapter: pointer to ixgbe_adapter
6502  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6503  *
6504  * Returns 0 on success, negative on failure
6505  **/
6506 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6507 			     struct ixgbe_ring *rx_ring)
6508 {
6509 	struct device *dev = rx_ring->dev;
6510 	int orig_node = dev_to_node(dev);
6511 	int ring_node = -1;
6512 	int size;
6513 
6514 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6515 
6516 	if (rx_ring->q_vector)
6517 		ring_node = rx_ring->q_vector->numa_node;
6518 
6519 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6520 	if (!rx_ring->rx_buffer_info)
6521 		rx_ring->rx_buffer_info = vmalloc(size);
6522 	if (!rx_ring->rx_buffer_info)
6523 		goto err;
6524 
6525 	/* Round up to nearest 4K */
6526 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6527 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6528 
6529 	set_dev_node(dev, ring_node);
6530 	rx_ring->desc = dma_alloc_coherent(dev,
6531 					   rx_ring->size,
6532 					   &rx_ring->dma,
6533 					   GFP_KERNEL);
6534 	set_dev_node(dev, orig_node);
6535 	if (!rx_ring->desc)
6536 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6537 						   &rx_ring->dma, GFP_KERNEL);
6538 	if (!rx_ring->desc)
6539 		goto err;
6540 
6541 	rx_ring->next_to_clean = 0;
6542 	rx_ring->next_to_use = 0;
6543 
6544 	/* XDP RX-queue info */
6545 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6546 			     rx_ring->queue_index) < 0)
6547 		goto err;
6548 
6549 	rx_ring->xdp_prog = adapter->xdp_prog;
6550 
6551 	return 0;
6552 err:
6553 	vfree(rx_ring->rx_buffer_info);
6554 	rx_ring->rx_buffer_info = NULL;
6555 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6556 	return -ENOMEM;
6557 }
6558 
6559 /**
6560  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6561  * @adapter: board private structure
6562  *
6563  * If this function returns with an error, then it's possible one or
6564  * more of the rings is populated (while the rest are not).  It is the
6565  * callers duty to clean those orphaned rings.
6566  *
6567  * Return 0 on success, negative on failure
6568  **/
6569 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6570 {
6571 	int i, err = 0;
6572 
6573 	for (i = 0; i < adapter->num_rx_queues; i++) {
6574 		err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6575 		if (!err)
6576 			continue;
6577 
6578 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6579 		goto err_setup_rx;
6580 	}
6581 
6582 #ifdef IXGBE_FCOE
6583 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
6584 	if (!err)
6585 #endif
6586 		return 0;
6587 err_setup_rx:
6588 	/* rewind the index freeing the rings as we go */
6589 	while (i--)
6590 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
6591 	return err;
6592 }
6593 
6594 /**
6595  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6596  * @tx_ring: Tx descriptor ring for a specific queue
6597  *
6598  * Free all transmit software resources
6599  **/
6600 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6601 {
6602 	ixgbe_clean_tx_ring(tx_ring);
6603 
6604 	vfree(tx_ring->tx_buffer_info);
6605 	tx_ring->tx_buffer_info = NULL;
6606 
6607 	/* if not set, then don't free */
6608 	if (!tx_ring->desc)
6609 		return;
6610 
6611 	dma_free_coherent(tx_ring->dev, tx_ring->size,
6612 			  tx_ring->desc, tx_ring->dma);
6613 
6614 	tx_ring->desc = NULL;
6615 }
6616 
6617 /**
6618  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6619  * @adapter: board private structure
6620  *
6621  * Free all transmit software resources
6622  **/
6623 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6624 {
6625 	int i;
6626 
6627 	for (i = 0; i < adapter->num_tx_queues; i++)
6628 		if (adapter->tx_ring[i]->desc)
6629 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
6630 	for (i = 0; i < adapter->num_xdp_queues; i++)
6631 		if (adapter->xdp_ring[i]->desc)
6632 			ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6633 }
6634 
6635 /**
6636  * ixgbe_free_rx_resources - Free Rx Resources
6637  * @rx_ring: ring to clean the resources from
6638  *
6639  * Free all receive software resources
6640  **/
6641 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6642 {
6643 	ixgbe_clean_rx_ring(rx_ring);
6644 
6645 	rx_ring->xdp_prog = NULL;
6646 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6647 	vfree(rx_ring->rx_buffer_info);
6648 	rx_ring->rx_buffer_info = NULL;
6649 
6650 	/* if not set, then don't free */
6651 	if (!rx_ring->desc)
6652 		return;
6653 
6654 	dma_free_coherent(rx_ring->dev, rx_ring->size,
6655 			  rx_ring->desc, rx_ring->dma);
6656 
6657 	rx_ring->desc = NULL;
6658 }
6659 
6660 /**
6661  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6662  * @adapter: board private structure
6663  *
6664  * Free all receive software resources
6665  **/
6666 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6667 {
6668 	int i;
6669 
6670 #ifdef IXGBE_FCOE
6671 	ixgbe_free_fcoe_ddp_resources(adapter);
6672 
6673 #endif
6674 	for (i = 0; i < adapter->num_rx_queues; i++)
6675 		if (adapter->rx_ring[i]->desc)
6676 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
6677 }
6678 
6679 /**
6680  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6681  * @netdev: network interface device structure
6682  * @new_mtu: new value for maximum frame size
6683  *
6684  * Returns 0 on success, negative on failure
6685  **/
6686 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6687 {
6688 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6689 
6690 	if (adapter->xdp_prog) {
6691 		int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6692 				     VLAN_HLEN;
6693 		int i;
6694 
6695 		for (i = 0; i < adapter->num_rx_queues; i++) {
6696 			struct ixgbe_ring *ring = adapter->rx_ring[i];
6697 
6698 			if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6699 				e_warn(probe, "Requested MTU size is not supported with XDP\n");
6700 				return -EINVAL;
6701 			}
6702 		}
6703 	}
6704 
6705 	/*
6706 	 * For 82599EB we cannot allow legacy VFs to enable their receive
6707 	 * paths when MTU greater than 1500 is configured.  So display a
6708 	 * warning that legacy VFs will be disabled.
6709 	 */
6710 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6711 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6712 	    (new_mtu > ETH_DATA_LEN))
6713 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6714 
6715 	e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6716 
6717 	/* must set new MTU before calling down or up */
6718 	netdev->mtu = new_mtu;
6719 
6720 	if (netif_running(netdev))
6721 		ixgbe_reinit_locked(adapter);
6722 
6723 	return 0;
6724 }
6725 
6726 /**
6727  * ixgbe_open - Called when a network interface is made active
6728  * @netdev: network interface device structure
6729  *
6730  * Returns 0 on success, negative value on failure
6731  *
6732  * The open entry point is called when a network interface is made
6733  * active by the system (IFF_UP).  At this point all resources needed
6734  * for transmit and receive operations are allocated, the interrupt
6735  * handler is registered with the OS, the watchdog timer is started,
6736  * and the stack is notified that the interface is ready.
6737  **/
6738 int ixgbe_open(struct net_device *netdev)
6739 {
6740 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6741 	struct ixgbe_hw *hw = &adapter->hw;
6742 	int err, queues;
6743 
6744 	/* disallow open during test */
6745 	if (test_bit(__IXGBE_TESTING, &adapter->state))
6746 		return -EBUSY;
6747 
6748 	netif_carrier_off(netdev);
6749 
6750 	/* allocate transmit descriptors */
6751 	err = ixgbe_setup_all_tx_resources(adapter);
6752 	if (err)
6753 		goto err_setup_tx;
6754 
6755 	/* allocate receive descriptors */
6756 	err = ixgbe_setup_all_rx_resources(adapter);
6757 	if (err)
6758 		goto err_setup_rx;
6759 
6760 	ixgbe_configure(adapter);
6761 
6762 	err = ixgbe_request_irq(adapter);
6763 	if (err)
6764 		goto err_req_irq;
6765 
6766 	/* Notify the stack of the actual queue counts. */
6767 	queues = adapter->num_tx_queues;
6768 	err = netif_set_real_num_tx_queues(netdev, queues);
6769 	if (err)
6770 		goto err_set_queues;
6771 
6772 	queues = adapter->num_rx_queues;
6773 	err = netif_set_real_num_rx_queues(netdev, queues);
6774 	if (err)
6775 		goto err_set_queues;
6776 
6777 	ixgbe_ptp_init(adapter);
6778 
6779 	ixgbe_up_complete(adapter);
6780 
6781 	ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6782 	udp_tunnel_get_rx_info(netdev);
6783 
6784 	return 0;
6785 
6786 err_set_queues:
6787 	ixgbe_free_irq(adapter);
6788 err_req_irq:
6789 	ixgbe_free_all_rx_resources(adapter);
6790 	if (hw->phy.ops.set_phy_power && !adapter->wol)
6791 		hw->phy.ops.set_phy_power(&adapter->hw, false);
6792 err_setup_rx:
6793 	ixgbe_free_all_tx_resources(adapter);
6794 err_setup_tx:
6795 	ixgbe_reset(adapter);
6796 
6797 	return err;
6798 }
6799 
6800 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6801 {
6802 	ixgbe_ptp_suspend(adapter);
6803 
6804 	if (adapter->hw.phy.ops.enter_lplu) {
6805 		adapter->hw.phy.reset_disable = true;
6806 		ixgbe_down(adapter);
6807 		adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6808 		adapter->hw.phy.reset_disable = false;
6809 	} else {
6810 		ixgbe_down(adapter);
6811 	}
6812 
6813 	ixgbe_free_irq(adapter);
6814 
6815 	ixgbe_free_all_tx_resources(adapter);
6816 	ixgbe_free_all_rx_resources(adapter);
6817 }
6818 
6819 /**
6820  * ixgbe_close - Disables a network interface
6821  * @netdev: network interface device structure
6822  *
6823  * Returns 0, this is not allowed to fail
6824  *
6825  * The close entry point is called when an interface is de-activated
6826  * by the OS.  The hardware is still under the drivers control, but
6827  * needs to be disabled.  A global MAC reset is issued to stop the
6828  * hardware, and all transmit and receive resources are freed.
6829  **/
6830 int ixgbe_close(struct net_device *netdev)
6831 {
6832 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6833 
6834 	ixgbe_ptp_stop(adapter);
6835 
6836 	if (netif_device_present(netdev))
6837 		ixgbe_close_suspend(adapter);
6838 
6839 	ixgbe_fdir_filter_exit(adapter);
6840 
6841 	ixgbe_release_hw_control(adapter);
6842 
6843 	return 0;
6844 }
6845 
6846 #ifdef CONFIG_PM
6847 static int ixgbe_resume(struct pci_dev *pdev)
6848 {
6849 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6850 	struct net_device *netdev = adapter->netdev;
6851 	u32 err;
6852 
6853 	adapter->hw.hw_addr = adapter->io_addr;
6854 	pci_set_power_state(pdev, PCI_D0);
6855 	pci_restore_state(pdev);
6856 	/*
6857 	 * pci_restore_state clears dev->state_saved so call
6858 	 * pci_save_state to restore it.
6859 	 */
6860 	pci_save_state(pdev);
6861 
6862 	err = pci_enable_device_mem(pdev);
6863 	if (err) {
6864 		e_dev_err("Cannot enable PCI device from suspend\n");
6865 		return err;
6866 	}
6867 	smp_mb__before_atomic();
6868 	clear_bit(__IXGBE_DISABLED, &adapter->state);
6869 	pci_set_master(pdev);
6870 
6871 	pci_wake_from_d3(pdev, false);
6872 
6873 	ixgbe_reset(adapter);
6874 
6875 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6876 
6877 	rtnl_lock();
6878 	err = ixgbe_init_interrupt_scheme(adapter);
6879 	if (!err && netif_running(netdev))
6880 		err = ixgbe_open(netdev);
6881 
6882 
6883 	if (!err)
6884 		netif_device_attach(netdev);
6885 	rtnl_unlock();
6886 
6887 	return err;
6888 }
6889 #endif /* CONFIG_PM */
6890 
6891 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6892 {
6893 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6894 	struct net_device *netdev = adapter->netdev;
6895 	struct ixgbe_hw *hw = &adapter->hw;
6896 	u32 ctrl;
6897 	u32 wufc = adapter->wol;
6898 #ifdef CONFIG_PM
6899 	int retval = 0;
6900 #endif
6901 
6902 	rtnl_lock();
6903 	netif_device_detach(netdev);
6904 
6905 	if (netif_running(netdev))
6906 		ixgbe_close_suspend(adapter);
6907 
6908 	ixgbe_clear_interrupt_scheme(adapter);
6909 	rtnl_unlock();
6910 
6911 #ifdef CONFIG_PM
6912 	retval = pci_save_state(pdev);
6913 	if (retval)
6914 		return retval;
6915 
6916 #endif
6917 	if (hw->mac.ops.stop_link_on_d3)
6918 		hw->mac.ops.stop_link_on_d3(hw);
6919 
6920 	if (wufc) {
6921 		u32 fctrl;
6922 
6923 		ixgbe_set_rx_mode(netdev);
6924 
6925 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
6926 		if (hw->mac.ops.enable_tx_laser)
6927 			hw->mac.ops.enable_tx_laser(hw);
6928 
6929 		/* enable the reception of multicast packets */
6930 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6931 		fctrl |= IXGBE_FCTRL_MPE;
6932 		IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6933 
6934 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6935 		ctrl |= IXGBE_CTRL_GIO_DIS;
6936 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6937 
6938 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6939 	} else {
6940 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6941 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6942 	}
6943 
6944 	switch (hw->mac.type) {
6945 	case ixgbe_mac_82598EB:
6946 		pci_wake_from_d3(pdev, false);
6947 		break;
6948 	case ixgbe_mac_82599EB:
6949 	case ixgbe_mac_X540:
6950 	case ixgbe_mac_X550:
6951 	case ixgbe_mac_X550EM_x:
6952 	case ixgbe_mac_x550em_a:
6953 		pci_wake_from_d3(pdev, !!wufc);
6954 		break;
6955 	default:
6956 		break;
6957 	}
6958 
6959 	*enable_wake = !!wufc;
6960 	if (hw->phy.ops.set_phy_power && !*enable_wake)
6961 		hw->phy.ops.set_phy_power(hw, false);
6962 
6963 	ixgbe_release_hw_control(adapter);
6964 
6965 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6966 		pci_disable_device(pdev);
6967 
6968 	return 0;
6969 }
6970 
6971 #ifdef CONFIG_PM
6972 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6973 {
6974 	int retval;
6975 	bool wake;
6976 
6977 	retval = __ixgbe_shutdown(pdev, &wake);
6978 	if (retval)
6979 		return retval;
6980 
6981 	if (wake) {
6982 		pci_prepare_to_sleep(pdev);
6983 	} else {
6984 		pci_wake_from_d3(pdev, false);
6985 		pci_set_power_state(pdev, PCI_D3hot);
6986 	}
6987 
6988 	return 0;
6989 }
6990 #endif /* CONFIG_PM */
6991 
6992 static void ixgbe_shutdown(struct pci_dev *pdev)
6993 {
6994 	bool wake;
6995 
6996 	__ixgbe_shutdown(pdev, &wake);
6997 
6998 	if (system_state == SYSTEM_POWER_OFF) {
6999 		pci_wake_from_d3(pdev, wake);
7000 		pci_set_power_state(pdev, PCI_D3hot);
7001 	}
7002 }
7003 
7004 /**
7005  * ixgbe_update_stats - Update the board statistics counters.
7006  * @adapter: board private structure
7007  **/
7008 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7009 {
7010 	struct net_device *netdev = adapter->netdev;
7011 	struct ixgbe_hw *hw = &adapter->hw;
7012 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
7013 	u64 total_mpc = 0;
7014 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7015 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7016 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7017 	u64 alloc_rx_page = 0;
7018 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7019 
7020 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7021 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7022 		return;
7023 
7024 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7025 		u64 rsc_count = 0;
7026 		u64 rsc_flush = 0;
7027 		for (i = 0; i < adapter->num_rx_queues; i++) {
7028 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7029 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7030 		}
7031 		adapter->rsc_total_count = rsc_count;
7032 		adapter->rsc_total_flush = rsc_flush;
7033 	}
7034 
7035 	for (i = 0; i < adapter->num_rx_queues; i++) {
7036 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7037 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7038 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7039 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7040 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7041 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7042 		bytes += rx_ring->stats.bytes;
7043 		packets += rx_ring->stats.packets;
7044 	}
7045 	adapter->non_eop_descs = non_eop_descs;
7046 	adapter->alloc_rx_page = alloc_rx_page;
7047 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7048 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7049 	adapter->hw_csum_rx_error = hw_csum_rx_error;
7050 	netdev->stats.rx_bytes = bytes;
7051 	netdev->stats.rx_packets = packets;
7052 
7053 	bytes = 0;
7054 	packets = 0;
7055 	/* gather some stats to the adapter struct that are per queue */
7056 	for (i = 0; i < adapter->num_tx_queues; i++) {
7057 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7058 		restart_queue += tx_ring->tx_stats.restart_queue;
7059 		tx_busy += tx_ring->tx_stats.tx_busy;
7060 		bytes += tx_ring->stats.bytes;
7061 		packets += tx_ring->stats.packets;
7062 	}
7063 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7064 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7065 
7066 		restart_queue += xdp_ring->tx_stats.restart_queue;
7067 		tx_busy += xdp_ring->tx_stats.tx_busy;
7068 		bytes += xdp_ring->stats.bytes;
7069 		packets += xdp_ring->stats.packets;
7070 	}
7071 	adapter->restart_queue = restart_queue;
7072 	adapter->tx_busy = tx_busy;
7073 	netdev->stats.tx_bytes = bytes;
7074 	netdev->stats.tx_packets = packets;
7075 
7076 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7077 
7078 	/* 8 register reads */
7079 	for (i = 0; i < 8; i++) {
7080 		/* for packet buffers not used, the register should read 0 */
7081 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7082 		missed_rx += mpc;
7083 		hwstats->mpc[i] += mpc;
7084 		total_mpc += hwstats->mpc[i];
7085 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7086 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7087 		switch (hw->mac.type) {
7088 		case ixgbe_mac_82598EB:
7089 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7090 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7091 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7092 			hwstats->pxonrxc[i] +=
7093 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7094 			break;
7095 		case ixgbe_mac_82599EB:
7096 		case ixgbe_mac_X540:
7097 		case ixgbe_mac_X550:
7098 		case ixgbe_mac_X550EM_x:
7099 		case ixgbe_mac_x550em_a:
7100 			hwstats->pxonrxc[i] +=
7101 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7102 			break;
7103 		default:
7104 			break;
7105 		}
7106 	}
7107 
7108 	/*16 register reads */
7109 	for (i = 0; i < 16; i++) {
7110 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7111 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7112 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
7113 		    (hw->mac.type == ixgbe_mac_X540) ||
7114 		    (hw->mac.type == ixgbe_mac_X550) ||
7115 		    (hw->mac.type == ixgbe_mac_X550EM_x) ||
7116 		    (hw->mac.type == ixgbe_mac_x550em_a)) {
7117 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7118 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7119 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7120 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7121 		}
7122 	}
7123 
7124 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7125 	/* work around hardware counting issue */
7126 	hwstats->gprc -= missed_rx;
7127 
7128 	ixgbe_update_xoff_received(adapter);
7129 
7130 	/* 82598 hardware only has a 32 bit counter in the high register */
7131 	switch (hw->mac.type) {
7132 	case ixgbe_mac_82598EB:
7133 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7134 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7135 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7136 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7137 		break;
7138 	case ixgbe_mac_X540:
7139 	case ixgbe_mac_X550:
7140 	case ixgbe_mac_X550EM_x:
7141 	case ixgbe_mac_x550em_a:
7142 		/* OS2BMC stats are X540 and later */
7143 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7144 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7145 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7146 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7147 		/* fall through */
7148 	case ixgbe_mac_82599EB:
7149 		for (i = 0; i < 16; i++)
7150 			adapter->hw_rx_no_dma_resources +=
7151 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7152 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7153 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7154 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7155 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7156 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7157 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7158 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7159 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7160 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7161 #ifdef IXGBE_FCOE
7162 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7163 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7164 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7165 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7166 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7167 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7168 		/* Add up per cpu counters for total ddp aloc fail */
7169 		if (adapter->fcoe.ddp_pool) {
7170 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7171 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
7172 			unsigned int cpu;
7173 			u64 noddp = 0, noddp_ext_buff = 0;
7174 			for_each_possible_cpu(cpu) {
7175 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7176 				noddp += ddp_pool->noddp;
7177 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
7178 			}
7179 			hwstats->fcoe_noddp = noddp;
7180 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7181 		}
7182 #endif /* IXGBE_FCOE */
7183 		break;
7184 	default:
7185 		break;
7186 	}
7187 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7188 	hwstats->bprc += bprc;
7189 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7190 	if (hw->mac.type == ixgbe_mac_82598EB)
7191 		hwstats->mprc -= bprc;
7192 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7193 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7194 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7195 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7196 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7197 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7198 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7199 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7200 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7201 	hwstats->lxontxc += lxon;
7202 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7203 	hwstats->lxofftxc += lxoff;
7204 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7205 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7206 	/*
7207 	 * 82598 errata - tx of flow control packets is included in tx counters
7208 	 */
7209 	xon_off_tot = lxon + lxoff;
7210 	hwstats->gptc -= xon_off_tot;
7211 	hwstats->mptc -= xon_off_tot;
7212 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7213 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7214 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7215 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7216 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7217 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7218 	hwstats->ptc64 -= xon_off_tot;
7219 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7220 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7221 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7222 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7223 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7224 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7225 
7226 	/* Fill out the OS statistics structure */
7227 	netdev->stats.multicast = hwstats->mprc;
7228 
7229 	/* Rx Errors */
7230 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7231 	netdev->stats.rx_dropped = 0;
7232 	netdev->stats.rx_length_errors = hwstats->rlec;
7233 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
7234 	netdev->stats.rx_missed_errors = total_mpc;
7235 }
7236 
7237 /**
7238  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7239  * @adapter: pointer to the device adapter structure
7240  **/
7241 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7242 {
7243 	struct ixgbe_hw *hw = &adapter->hw;
7244 	int i;
7245 
7246 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7247 		return;
7248 
7249 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7250 
7251 	/* if interface is down do nothing */
7252 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7253 		return;
7254 
7255 	/* do nothing if we are not using signature filters */
7256 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7257 		return;
7258 
7259 	adapter->fdir_overflow++;
7260 
7261 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7262 		for (i = 0; i < adapter->num_tx_queues; i++)
7263 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7264 				&(adapter->tx_ring[i]->state));
7265 		for (i = 0; i < adapter->num_xdp_queues; i++)
7266 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7267 				&adapter->xdp_ring[i]->state);
7268 		/* re-enable flow director interrupts */
7269 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7270 	} else {
7271 		e_err(probe, "failed to finish FDIR re-initialization, "
7272 		      "ignored adding FDIR ATR filters\n");
7273 	}
7274 }
7275 
7276 /**
7277  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7278  * @adapter: pointer to the device adapter structure
7279  *
7280  * This function serves two purposes.  First it strobes the interrupt lines
7281  * in order to make certain interrupts are occurring.  Secondly it sets the
7282  * bits needed to check for TX hangs.  As a result we should immediately
7283  * determine if a hang has occurred.
7284  */
7285 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7286 {
7287 	struct ixgbe_hw *hw = &adapter->hw;
7288 	u64 eics = 0;
7289 	int i;
7290 
7291 	/* If we're down, removing or resetting, just bail */
7292 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7293 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7294 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7295 		return;
7296 
7297 	/* Force detection of hung controller */
7298 	if (netif_carrier_ok(adapter->netdev)) {
7299 		for (i = 0; i < adapter->num_tx_queues; i++)
7300 			set_check_for_tx_hang(adapter->tx_ring[i]);
7301 		for (i = 0; i < adapter->num_xdp_queues; i++)
7302 			set_check_for_tx_hang(adapter->xdp_ring[i]);
7303 	}
7304 
7305 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7306 		/*
7307 		 * for legacy and MSI interrupts don't set any bits
7308 		 * that are enabled for EIAM, because this operation
7309 		 * would set *both* EIMS and EICS for any bit in EIAM
7310 		 */
7311 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
7312 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7313 	} else {
7314 		/* get one bit for every active tx/rx interrupt vector */
7315 		for (i = 0; i < adapter->num_q_vectors; i++) {
7316 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
7317 			if (qv->rx.ring || qv->tx.ring)
7318 				eics |= BIT_ULL(i);
7319 		}
7320 	}
7321 
7322 	/* Cause software interrupt to ensure rings are cleaned */
7323 	ixgbe_irq_rearm_queues(adapter, eics);
7324 }
7325 
7326 /**
7327  * ixgbe_watchdog_update_link - update the link status
7328  * @adapter: pointer to the device adapter structure
7329  **/
7330 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7331 {
7332 	struct ixgbe_hw *hw = &adapter->hw;
7333 	u32 link_speed = adapter->link_speed;
7334 	bool link_up = adapter->link_up;
7335 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7336 
7337 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7338 		return;
7339 
7340 	if (hw->mac.ops.check_link) {
7341 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7342 	} else {
7343 		/* always assume link is up, if no check link function */
7344 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7345 		link_up = true;
7346 	}
7347 
7348 	if (adapter->ixgbe_ieee_pfc)
7349 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7350 
7351 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7352 		hw->mac.ops.fc_enable(hw);
7353 		ixgbe_set_rx_drop_en(adapter);
7354 	}
7355 
7356 	if (link_up ||
7357 	    time_after(jiffies, (adapter->link_check_timeout +
7358 				 IXGBE_TRY_LINK_TIMEOUT))) {
7359 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7360 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7361 		IXGBE_WRITE_FLUSH(hw);
7362 	}
7363 
7364 	adapter->link_up = link_up;
7365 	adapter->link_speed = link_speed;
7366 }
7367 
7368 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7369 {
7370 #ifdef CONFIG_IXGBE_DCB
7371 	struct net_device *netdev = adapter->netdev;
7372 	struct dcb_app app = {
7373 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7374 			      .protocol = 0,
7375 			     };
7376 	u8 up = 0;
7377 
7378 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7379 		up = dcb_ieee_getapp_mask(netdev, &app);
7380 
7381 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7382 #endif
7383 }
7384 
7385 /**
7386  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7387  *                             print link up message
7388  * @adapter: pointer to the device adapter structure
7389  **/
7390 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7391 {
7392 	struct net_device *netdev = adapter->netdev;
7393 	struct ixgbe_hw *hw = &adapter->hw;
7394 	u32 link_speed = adapter->link_speed;
7395 	const char *speed_str;
7396 	bool flow_rx, flow_tx;
7397 
7398 	/* only continue if link was previously down */
7399 	if (netif_carrier_ok(netdev))
7400 		return;
7401 
7402 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7403 
7404 	switch (hw->mac.type) {
7405 	case ixgbe_mac_82598EB: {
7406 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7407 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7408 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7409 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7410 	}
7411 		break;
7412 	case ixgbe_mac_X540:
7413 	case ixgbe_mac_X550:
7414 	case ixgbe_mac_X550EM_x:
7415 	case ixgbe_mac_x550em_a:
7416 	case ixgbe_mac_82599EB: {
7417 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7418 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7419 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7420 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7421 	}
7422 		break;
7423 	default:
7424 		flow_tx = false;
7425 		flow_rx = false;
7426 		break;
7427 	}
7428 
7429 	adapter->last_rx_ptp_check = jiffies;
7430 
7431 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7432 		ixgbe_ptp_start_cyclecounter(adapter);
7433 
7434 	switch (link_speed) {
7435 	case IXGBE_LINK_SPEED_10GB_FULL:
7436 		speed_str = "10 Gbps";
7437 		break;
7438 	case IXGBE_LINK_SPEED_5GB_FULL:
7439 		speed_str = "5 Gbps";
7440 		break;
7441 	case IXGBE_LINK_SPEED_2_5GB_FULL:
7442 		speed_str = "2.5 Gbps";
7443 		break;
7444 	case IXGBE_LINK_SPEED_1GB_FULL:
7445 		speed_str = "1 Gbps";
7446 		break;
7447 	case IXGBE_LINK_SPEED_100_FULL:
7448 		speed_str = "100 Mbps";
7449 		break;
7450 	case IXGBE_LINK_SPEED_10_FULL:
7451 		speed_str = "10 Mbps";
7452 		break;
7453 	default:
7454 		speed_str = "unknown speed";
7455 		break;
7456 	}
7457 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7458 	       ((flow_rx && flow_tx) ? "RX/TX" :
7459 	       (flow_rx ? "RX" :
7460 	       (flow_tx ? "TX" : "None"))));
7461 
7462 	netif_carrier_on(netdev);
7463 	ixgbe_check_vf_rate_limit(adapter);
7464 
7465 	/* enable transmits */
7466 	netif_tx_wake_all_queues(adapter->netdev);
7467 
7468 	/* update the default user priority for VFs */
7469 	ixgbe_update_default_up(adapter);
7470 
7471 	/* ping all the active vfs to let them know link has changed */
7472 	ixgbe_ping_all_vfs(adapter);
7473 }
7474 
7475 /**
7476  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7477  *                               print link down message
7478  * @adapter: pointer to the adapter structure
7479  **/
7480 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7481 {
7482 	struct net_device *netdev = adapter->netdev;
7483 	struct ixgbe_hw *hw = &adapter->hw;
7484 
7485 	adapter->link_up = false;
7486 	adapter->link_speed = 0;
7487 
7488 	/* only continue if link was up previously */
7489 	if (!netif_carrier_ok(netdev))
7490 		return;
7491 
7492 	/* poll for SFP+ cable when link is down */
7493 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7494 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7495 
7496 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7497 		ixgbe_ptp_start_cyclecounter(adapter);
7498 
7499 	e_info(drv, "NIC Link is Down\n");
7500 	netif_carrier_off(netdev);
7501 
7502 	/* ping all the active vfs to let them know link has changed */
7503 	ixgbe_ping_all_vfs(adapter);
7504 }
7505 
7506 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7507 {
7508 	int i;
7509 
7510 	for (i = 0; i < adapter->num_tx_queues; i++) {
7511 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7512 
7513 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
7514 			return true;
7515 	}
7516 
7517 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7518 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
7519 
7520 		if (ring->next_to_use != ring->next_to_clean)
7521 			return true;
7522 	}
7523 
7524 	return false;
7525 }
7526 
7527 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7528 {
7529 	struct ixgbe_hw *hw = &adapter->hw;
7530 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7531 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7532 
7533 	int i, j;
7534 
7535 	if (!adapter->num_vfs)
7536 		return false;
7537 
7538 	/* resetting the PF is only needed for MAC before X550 */
7539 	if (hw->mac.type >= ixgbe_mac_X550)
7540 		return false;
7541 
7542 	for (i = 0; i < adapter->num_vfs; i++) {
7543 		for (j = 0; j < q_per_pool; j++) {
7544 			u32 h, t;
7545 
7546 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7547 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7548 
7549 			if (h != t)
7550 				return true;
7551 		}
7552 	}
7553 
7554 	return false;
7555 }
7556 
7557 /**
7558  * ixgbe_watchdog_flush_tx - flush queues on link down
7559  * @adapter: pointer to the device adapter structure
7560  **/
7561 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7562 {
7563 	if (!netif_carrier_ok(adapter->netdev)) {
7564 		if (ixgbe_ring_tx_pending(adapter) ||
7565 		    ixgbe_vf_tx_pending(adapter)) {
7566 			/* We've lost link, so the controller stops DMA,
7567 			 * but we've got queued Tx work that's never going
7568 			 * to get done, so reset controller to flush Tx.
7569 			 * (Do the reset outside of interrupt context).
7570 			 */
7571 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7572 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7573 		}
7574 	}
7575 }
7576 
7577 #ifdef CONFIG_PCI_IOV
7578 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7579 {
7580 	struct ixgbe_hw *hw = &adapter->hw;
7581 	struct pci_dev *pdev = adapter->pdev;
7582 	unsigned int vf;
7583 	u32 gpc;
7584 
7585 	if (!(netif_carrier_ok(adapter->netdev)))
7586 		return;
7587 
7588 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7589 	if (gpc) /* If incrementing then no need for the check below */
7590 		return;
7591 	/* Check to see if a bad DMA write target from an errant or
7592 	 * malicious VF has caused a PCIe error.  If so then we can
7593 	 * issue a VFLR to the offending VF(s) and then resume without
7594 	 * requesting a full slot reset.
7595 	 */
7596 
7597 	if (!pdev)
7598 		return;
7599 
7600 	/* check status reg for all VFs owned by this PF */
7601 	for (vf = 0; vf < adapter->num_vfs; ++vf) {
7602 		struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7603 		u16 status_reg;
7604 
7605 		if (!vfdev)
7606 			continue;
7607 		pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7608 		if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7609 		    status_reg & PCI_STATUS_REC_MASTER_ABORT)
7610 			pcie_flr(vfdev);
7611 	}
7612 }
7613 
7614 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7615 {
7616 	u32 ssvpc;
7617 
7618 	/* Do not perform spoof check for 82598 or if not in IOV mode */
7619 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7620 	    adapter->num_vfs == 0)
7621 		return;
7622 
7623 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7624 
7625 	/*
7626 	 * ssvpc register is cleared on read, if zero then no
7627 	 * spoofed packets in the last interval.
7628 	 */
7629 	if (!ssvpc)
7630 		return;
7631 
7632 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7633 }
7634 #else
7635 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7636 {
7637 }
7638 
7639 static void
7640 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7641 {
7642 }
7643 #endif /* CONFIG_PCI_IOV */
7644 
7645 
7646 /**
7647  * ixgbe_watchdog_subtask - check and bring link up
7648  * @adapter: pointer to the device adapter structure
7649  **/
7650 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7651 {
7652 	/* if interface is down, removing or resetting, do nothing */
7653 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7654 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7655 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7656 		return;
7657 
7658 	ixgbe_watchdog_update_link(adapter);
7659 
7660 	if (adapter->link_up)
7661 		ixgbe_watchdog_link_is_up(adapter);
7662 	else
7663 		ixgbe_watchdog_link_is_down(adapter);
7664 
7665 	ixgbe_check_for_bad_vf(adapter);
7666 	ixgbe_spoof_check(adapter);
7667 	ixgbe_update_stats(adapter);
7668 
7669 	ixgbe_watchdog_flush_tx(adapter);
7670 }
7671 
7672 /**
7673  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7674  * @adapter: the ixgbe adapter structure
7675  **/
7676 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7677 {
7678 	struct ixgbe_hw *hw = &adapter->hw;
7679 	s32 err;
7680 
7681 	/* not searching for SFP so there is nothing to do here */
7682 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7683 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7684 		return;
7685 
7686 	if (adapter->sfp_poll_time &&
7687 	    time_after(adapter->sfp_poll_time, jiffies))
7688 		return; /* If not yet time to poll for SFP */
7689 
7690 	/* someone else is in init, wait until next service event */
7691 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7692 		return;
7693 
7694 	adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7695 
7696 	err = hw->phy.ops.identify_sfp(hw);
7697 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7698 		goto sfp_out;
7699 
7700 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7701 		/* If no cable is present, then we need to reset
7702 		 * the next time we find a good cable. */
7703 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7704 	}
7705 
7706 	/* exit on error */
7707 	if (err)
7708 		goto sfp_out;
7709 
7710 	/* exit if reset not needed */
7711 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7712 		goto sfp_out;
7713 
7714 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7715 
7716 	/*
7717 	 * A module may be identified correctly, but the EEPROM may not have
7718 	 * support for that module.  setup_sfp() will fail in that case, so
7719 	 * we should not allow that module to load.
7720 	 */
7721 	if (hw->mac.type == ixgbe_mac_82598EB)
7722 		err = hw->phy.ops.reset(hw);
7723 	else
7724 		err = hw->mac.ops.setup_sfp(hw);
7725 
7726 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7727 		goto sfp_out;
7728 
7729 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7730 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7731 
7732 sfp_out:
7733 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7734 
7735 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7736 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7737 		e_dev_err("failed to initialize because an unsupported "
7738 			  "SFP+ module type was detected.\n");
7739 		e_dev_err("Reload the driver after installing a "
7740 			  "supported module.\n");
7741 		unregister_netdev(adapter->netdev);
7742 	}
7743 }
7744 
7745 /**
7746  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7747  * @adapter: the ixgbe adapter structure
7748  **/
7749 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7750 {
7751 	struct ixgbe_hw *hw = &adapter->hw;
7752 	u32 cap_speed;
7753 	u32 speed;
7754 	bool autoneg = false;
7755 
7756 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7757 		return;
7758 
7759 	/* someone else is in init, wait until next service event */
7760 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7761 		return;
7762 
7763 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7764 
7765 	hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7766 
7767 	/* advertise highest capable link speed */
7768 	if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7769 		speed = IXGBE_LINK_SPEED_10GB_FULL;
7770 	else
7771 		speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7772 				     IXGBE_LINK_SPEED_1GB_FULL);
7773 
7774 	if (hw->mac.ops.setup_link)
7775 		hw->mac.ops.setup_link(hw, speed, true);
7776 
7777 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7778 	adapter->link_check_timeout = jiffies;
7779 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7780 }
7781 
7782 /**
7783  * ixgbe_service_timer - Timer Call-back
7784  * @t: pointer to timer_list structure
7785  **/
7786 static void ixgbe_service_timer(struct timer_list *t)
7787 {
7788 	struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7789 	unsigned long next_event_offset;
7790 
7791 	/* poll faster when waiting for link */
7792 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7793 		next_event_offset = HZ / 10;
7794 	else
7795 		next_event_offset = HZ * 2;
7796 
7797 	/* Reset the timer */
7798 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7799 
7800 	ixgbe_service_event_schedule(adapter);
7801 }
7802 
7803 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7804 {
7805 	struct ixgbe_hw *hw = &adapter->hw;
7806 	u32 status;
7807 
7808 	if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7809 		return;
7810 
7811 	adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7812 
7813 	if (!hw->phy.ops.handle_lasi)
7814 		return;
7815 
7816 	status = hw->phy.ops.handle_lasi(&adapter->hw);
7817 	if (status != IXGBE_ERR_OVERTEMP)
7818 		return;
7819 
7820 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
7821 }
7822 
7823 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7824 {
7825 	if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7826 		return;
7827 
7828 	rtnl_lock();
7829 	/* If we're already down, removing or resetting, just bail */
7830 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7831 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7832 	    test_bit(__IXGBE_RESETTING, &adapter->state)) {
7833 		rtnl_unlock();
7834 		return;
7835 	}
7836 
7837 	ixgbe_dump(adapter);
7838 	netdev_err(adapter->netdev, "Reset adapter\n");
7839 	adapter->tx_timeout_count++;
7840 
7841 	ixgbe_reinit_locked(adapter);
7842 	rtnl_unlock();
7843 }
7844 
7845 /**
7846  * ixgbe_check_fw_error - Check firmware for errors
7847  * @adapter: the adapter private structure
7848  *
7849  * Check firmware errors in register FWSM
7850  */
7851 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7852 {
7853 	struct ixgbe_hw *hw = &adapter->hw;
7854 	u32 fwsm;
7855 
7856 	/* read fwsm.ext_err_ind register and log errors */
7857 	fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7858 
7859 	if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7860 	    !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7861 		e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7862 			   fwsm);
7863 
7864 	if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7865 		e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7866 		return true;
7867 	}
7868 
7869 	return false;
7870 }
7871 
7872 /**
7873  * ixgbe_service_task - manages and runs subtasks
7874  * @work: pointer to work_struct containing our data
7875  **/
7876 static void ixgbe_service_task(struct work_struct *work)
7877 {
7878 	struct ixgbe_adapter *adapter = container_of(work,
7879 						     struct ixgbe_adapter,
7880 						     service_task);
7881 	if (ixgbe_removed(adapter->hw.hw_addr)) {
7882 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7883 			rtnl_lock();
7884 			ixgbe_down(adapter);
7885 			rtnl_unlock();
7886 		}
7887 		ixgbe_service_event_complete(adapter);
7888 		return;
7889 	}
7890 	if (ixgbe_check_fw_error(adapter)) {
7891 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7892 			rtnl_lock();
7893 			unregister_netdev(adapter->netdev);
7894 			rtnl_unlock();
7895 		}
7896 		ixgbe_service_event_complete(adapter);
7897 		return;
7898 	}
7899 	if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7900 		rtnl_lock();
7901 		adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7902 		udp_tunnel_get_rx_info(adapter->netdev);
7903 		rtnl_unlock();
7904 	}
7905 	ixgbe_reset_subtask(adapter);
7906 	ixgbe_phy_interrupt_subtask(adapter);
7907 	ixgbe_sfp_detection_subtask(adapter);
7908 	ixgbe_sfp_link_config_subtask(adapter);
7909 	ixgbe_check_overtemp_subtask(adapter);
7910 	ixgbe_watchdog_subtask(adapter);
7911 	ixgbe_fdir_reinit_subtask(adapter);
7912 	ixgbe_check_hang_subtask(adapter);
7913 
7914 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7915 		ixgbe_ptp_overflow_check(adapter);
7916 		if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7917 			ixgbe_ptp_rx_hang(adapter);
7918 		ixgbe_ptp_tx_hang(adapter);
7919 	}
7920 
7921 	ixgbe_service_event_complete(adapter);
7922 }
7923 
7924 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7925 		     struct ixgbe_tx_buffer *first,
7926 		     u8 *hdr_len,
7927 		     struct ixgbe_ipsec_tx_data *itd)
7928 {
7929 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7930 	struct sk_buff *skb = first->skb;
7931 	union {
7932 		struct iphdr *v4;
7933 		struct ipv6hdr *v6;
7934 		unsigned char *hdr;
7935 	} ip;
7936 	union {
7937 		struct tcphdr *tcp;
7938 		unsigned char *hdr;
7939 	} l4;
7940 	u32 paylen, l4_offset;
7941 	u32 fceof_saidx = 0;
7942 	int err;
7943 
7944 	if (skb->ip_summed != CHECKSUM_PARTIAL)
7945 		return 0;
7946 
7947 	if (!skb_is_gso(skb))
7948 		return 0;
7949 
7950 	err = skb_cow_head(skb, 0);
7951 	if (err < 0)
7952 		return err;
7953 
7954 	if (eth_p_mpls(first->protocol))
7955 		ip.hdr = skb_inner_network_header(skb);
7956 	else
7957 		ip.hdr = skb_network_header(skb);
7958 	l4.hdr = skb_checksum_start(skb);
7959 
7960 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7961 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7962 
7963 	/* initialize outer IP header fields */
7964 	if (ip.v4->version == 4) {
7965 		unsigned char *csum_start = skb_checksum_start(skb);
7966 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7967 		int len = csum_start - trans_start;
7968 
7969 		/* IP header will have to cancel out any data that
7970 		 * is not a part of the outer IP header, so set to
7971 		 * a reverse csum if needed, else init check to 0.
7972 		 */
7973 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7974 					   csum_fold(csum_partial(trans_start,
7975 								  len, 0)) : 0;
7976 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7977 
7978 		ip.v4->tot_len = 0;
7979 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7980 				   IXGBE_TX_FLAGS_CSUM |
7981 				   IXGBE_TX_FLAGS_IPV4;
7982 	} else {
7983 		ip.v6->payload_len = 0;
7984 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7985 				   IXGBE_TX_FLAGS_CSUM;
7986 	}
7987 
7988 	/* determine offset of inner transport header */
7989 	l4_offset = l4.hdr - skb->data;
7990 
7991 	/* compute length of segmentation header */
7992 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
7993 
7994 	/* remove payload length from inner checksum */
7995 	paylen = skb->len - l4_offset;
7996 	csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
7997 
7998 	/* update gso size and bytecount with header size */
7999 	first->gso_segs = skb_shinfo(skb)->gso_segs;
8000 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
8001 
8002 	/* mss_l4len_id: use 0 as index for TSO */
8003 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8004 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8005 
8006 	fceof_saidx |= itd->sa_idx;
8007 	type_tucmd |= itd->flags | itd->trailer_len;
8008 
8009 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8010 	vlan_macip_lens = l4.hdr - ip.hdr;
8011 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8012 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8013 
8014 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8015 			  mss_l4len_idx);
8016 
8017 	return 1;
8018 }
8019 
8020 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8021 {
8022 	unsigned int offset = 0;
8023 
8024 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8025 
8026 	return offset == skb_checksum_start_offset(skb);
8027 }
8028 
8029 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8030 			  struct ixgbe_tx_buffer *first,
8031 			  struct ixgbe_ipsec_tx_data *itd)
8032 {
8033 	struct sk_buff *skb = first->skb;
8034 	u32 vlan_macip_lens = 0;
8035 	u32 fceof_saidx = 0;
8036 	u32 type_tucmd = 0;
8037 
8038 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8039 csum_failed:
8040 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8041 					 IXGBE_TX_FLAGS_CC)))
8042 			return;
8043 		goto no_csum;
8044 	}
8045 
8046 	switch (skb->csum_offset) {
8047 	case offsetof(struct tcphdr, check):
8048 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8049 		/* fall through */
8050 	case offsetof(struct udphdr, check):
8051 		break;
8052 	case offsetof(struct sctphdr, checksum):
8053 		/* validate that this is actually an SCTP request */
8054 		if (((first->protocol == htons(ETH_P_IP)) &&
8055 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8056 		    ((first->protocol == htons(ETH_P_IPV6)) &&
8057 		     ixgbe_ipv6_csum_is_sctp(skb))) {
8058 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8059 			break;
8060 		}
8061 		/* fall through */
8062 	default:
8063 		skb_checksum_help(skb);
8064 		goto csum_failed;
8065 	}
8066 
8067 	/* update TX checksum flag */
8068 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8069 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8070 			  skb_network_offset(skb);
8071 no_csum:
8072 	/* vlan_macip_lens: MACLEN, VLAN tag */
8073 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8074 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8075 
8076 	fceof_saidx |= itd->sa_idx;
8077 	type_tucmd |= itd->flags | itd->trailer_len;
8078 
8079 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8080 }
8081 
8082 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8083 	((_flag <= _result) ? \
8084 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8085 	 ((u32)(_input & _flag) / (_flag / _result)))
8086 
8087 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8088 {
8089 	/* set type for advanced descriptor with frame checksum insertion */
8090 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8091 		       IXGBE_ADVTXD_DCMD_DEXT |
8092 		       IXGBE_ADVTXD_DCMD_IFCS;
8093 
8094 	/* set HW vlan bit if vlan is present */
8095 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8096 				   IXGBE_ADVTXD_DCMD_VLE);
8097 
8098 	/* set segmentation enable bits for TSO/FSO */
8099 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8100 				   IXGBE_ADVTXD_DCMD_TSE);
8101 
8102 	/* set timestamp bit if present */
8103 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8104 				   IXGBE_ADVTXD_MAC_TSTAMP);
8105 
8106 	/* insert frame checksum */
8107 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8108 
8109 	return cmd_type;
8110 }
8111 
8112 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8113 				   u32 tx_flags, unsigned int paylen)
8114 {
8115 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8116 
8117 	/* enable L4 checksum for TSO and TX checksum offload */
8118 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8119 					IXGBE_TX_FLAGS_CSUM,
8120 					IXGBE_ADVTXD_POPTS_TXSM);
8121 
8122 	/* enable IPv4 checksum for TSO */
8123 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8124 					IXGBE_TX_FLAGS_IPV4,
8125 					IXGBE_ADVTXD_POPTS_IXSM);
8126 
8127 	/* enable IPsec */
8128 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8129 					IXGBE_TX_FLAGS_IPSEC,
8130 					IXGBE_ADVTXD_POPTS_IPSEC);
8131 
8132 	/*
8133 	 * Check Context must be set if Tx switch is enabled, which it
8134 	 * always is for case where virtual functions are running
8135 	 */
8136 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8137 					IXGBE_TX_FLAGS_CC,
8138 					IXGBE_ADVTXD_CC);
8139 
8140 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8141 }
8142 
8143 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8144 {
8145 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8146 
8147 	/* Herbert's original patch had:
8148 	 *  smp_mb__after_netif_stop_queue();
8149 	 * but since that doesn't exist yet, just open code it.
8150 	 */
8151 	smp_mb();
8152 
8153 	/* We need to check again in a case another CPU has just
8154 	 * made room available.
8155 	 */
8156 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8157 		return -EBUSY;
8158 
8159 	/* A reprieve! - use start_queue because it doesn't call schedule */
8160 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8161 	++tx_ring->tx_stats.restart_queue;
8162 	return 0;
8163 }
8164 
8165 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8166 {
8167 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8168 		return 0;
8169 
8170 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8171 }
8172 
8173 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8174 			struct ixgbe_tx_buffer *first,
8175 			const u8 hdr_len)
8176 {
8177 	struct sk_buff *skb = first->skb;
8178 	struct ixgbe_tx_buffer *tx_buffer;
8179 	union ixgbe_adv_tx_desc *tx_desc;
8180 	struct skb_frag_struct *frag;
8181 	dma_addr_t dma;
8182 	unsigned int data_len, size;
8183 	u32 tx_flags = first->tx_flags;
8184 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8185 	u16 i = tx_ring->next_to_use;
8186 
8187 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8188 
8189 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8190 
8191 	size = skb_headlen(skb);
8192 	data_len = skb->data_len;
8193 
8194 #ifdef IXGBE_FCOE
8195 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8196 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8197 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8198 			data_len = 0;
8199 		} else {
8200 			data_len -= sizeof(struct fcoe_crc_eof);
8201 		}
8202 	}
8203 
8204 #endif
8205 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8206 
8207 	tx_buffer = first;
8208 
8209 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8210 		if (dma_mapping_error(tx_ring->dev, dma))
8211 			goto dma_error;
8212 
8213 		/* record length, and DMA address */
8214 		dma_unmap_len_set(tx_buffer, len, size);
8215 		dma_unmap_addr_set(tx_buffer, dma, dma);
8216 
8217 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8218 
8219 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8220 			tx_desc->read.cmd_type_len =
8221 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8222 
8223 			i++;
8224 			tx_desc++;
8225 			if (i == tx_ring->count) {
8226 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8227 				i = 0;
8228 			}
8229 			tx_desc->read.olinfo_status = 0;
8230 
8231 			dma += IXGBE_MAX_DATA_PER_TXD;
8232 			size -= IXGBE_MAX_DATA_PER_TXD;
8233 
8234 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8235 		}
8236 
8237 		if (likely(!data_len))
8238 			break;
8239 
8240 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8241 
8242 		i++;
8243 		tx_desc++;
8244 		if (i == tx_ring->count) {
8245 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8246 			i = 0;
8247 		}
8248 		tx_desc->read.olinfo_status = 0;
8249 
8250 #ifdef IXGBE_FCOE
8251 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8252 #else
8253 		size = skb_frag_size(frag);
8254 #endif
8255 		data_len -= size;
8256 
8257 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8258 				       DMA_TO_DEVICE);
8259 
8260 		tx_buffer = &tx_ring->tx_buffer_info[i];
8261 	}
8262 
8263 	/* write last descriptor with RS and EOP bits */
8264 	cmd_type |= size | IXGBE_TXD_CMD;
8265 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8266 
8267 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8268 
8269 	/* set the timestamp */
8270 	first->time_stamp = jiffies;
8271 
8272 	/*
8273 	 * Force memory writes to complete before letting h/w know there
8274 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8275 	 * memory model archs, such as IA-64).
8276 	 *
8277 	 * We also need this memory barrier to make certain all of the
8278 	 * status bits have been updated before next_to_watch is written.
8279 	 */
8280 	wmb();
8281 
8282 	/* set next_to_watch value indicating a packet is present */
8283 	first->next_to_watch = tx_desc;
8284 
8285 	i++;
8286 	if (i == tx_ring->count)
8287 		i = 0;
8288 
8289 	tx_ring->next_to_use = i;
8290 
8291 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8292 
8293 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
8294 		writel(i, tx_ring->tail);
8295 
8296 		/* we need this if more than one processor can write to our tail
8297 		 * at a time, it synchronizes IO on IA64/Altix systems
8298 		 */
8299 		mmiowb();
8300 	}
8301 
8302 	return 0;
8303 dma_error:
8304 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8305 
8306 	/* clear dma mappings for failed tx_buffer_info map */
8307 	for (;;) {
8308 		tx_buffer = &tx_ring->tx_buffer_info[i];
8309 		if (dma_unmap_len(tx_buffer, len))
8310 			dma_unmap_page(tx_ring->dev,
8311 				       dma_unmap_addr(tx_buffer, dma),
8312 				       dma_unmap_len(tx_buffer, len),
8313 				       DMA_TO_DEVICE);
8314 		dma_unmap_len_set(tx_buffer, len, 0);
8315 		if (tx_buffer == first)
8316 			break;
8317 		if (i == 0)
8318 			i += tx_ring->count;
8319 		i--;
8320 	}
8321 
8322 	dev_kfree_skb_any(first->skb);
8323 	first->skb = NULL;
8324 
8325 	tx_ring->next_to_use = i;
8326 
8327 	return -1;
8328 }
8329 
8330 static void ixgbe_atr(struct ixgbe_ring *ring,
8331 		      struct ixgbe_tx_buffer *first)
8332 {
8333 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8334 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8335 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8336 	union {
8337 		unsigned char *network;
8338 		struct iphdr *ipv4;
8339 		struct ipv6hdr *ipv6;
8340 	} hdr;
8341 	struct tcphdr *th;
8342 	unsigned int hlen;
8343 	struct sk_buff *skb;
8344 	__be16 vlan_id;
8345 	int l4_proto;
8346 
8347 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8348 	if (!q_vector)
8349 		return;
8350 
8351 	/* do nothing if sampling is disabled */
8352 	if (!ring->atr_sample_rate)
8353 		return;
8354 
8355 	ring->atr_count++;
8356 
8357 	/* currently only IPv4/IPv6 with TCP is supported */
8358 	if ((first->protocol != htons(ETH_P_IP)) &&
8359 	    (first->protocol != htons(ETH_P_IPV6)))
8360 		return;
8361 
8362 	/* snag network header to get L4 type and address */
8363 	skb = first->skb;
8364 	hdr.network = skb_network_header(skb);
8365 	if (unlikely(hdr.network <= skb->data))
8366 		return;
8367 	if (skb->encapsulation &&
8368 	    first->protocol == htons(ETH_P_IP) &&
8369 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8370 		struct ixgbe_adapter *adapter = q_vector->adapter;
8371 
8372 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8373 			     VXLAN_HEADROOM))
8374 			return;
8375 
8376 		/* verify the port is recognized as VXLAN */
8377 		if (adapter->vxlan_port &&
8378 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8379 			hdr.network = skb_inner_network_header(skb);
8380 
8381 		if (adapter->geneve_port &&
8382 		    udp_hdr(skb)->dest == adapter->geneve_port)
8383 			hdr.network = skb_inner_network_header(skb);
8384 	}
8385 
8386 	/* Make sure we have at least [minimum IPv4 header + TCP]
8387 	 * or [IPv6 header] bytes
8388 	 */
8389 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8390 		return;
8391 
8392 	/* Currently only IPv4/IPv6 with TCP is supported */
8393 	switch (hdr.ipv4->version) {
8394 	case IPVERSION:
8395 		/* access ihl as u8 to avoid unaligned access on ia64 */
8396 		hlen = (hdr.network[0] & 0x0F) << 2;
8397 		l4_proto = hdr.ipv4->protocol;
8398 		break;
8399 	case 6:
8400 		hlen = hdr.network - skb->data;
8401 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8402 		hlen -= hdr.network - skb->data;
8403 		break;
8404 	default:
8405 		return;
8406 	}
8407 
8408 	if (l4_proto != IPPROTO_TCP)
8409 		return;
8410 
8411 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8412 		     hlen + sizeof(struct tcphdr)))
8413 		return;
8414 
8415 	th = (struct tcphdr *)(hdr.network + hlen);
8416 
8417 	/* skip this packet since the socket is closing */
8418 	if (th->fin)
8419 		return;
8420 
8421 	/* sample on all syn packets or once every atr sample count */
8422 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8423 		return;
8424 
8425 	/* reset sample count */
8426 	ring->atr_count = 0;
8427 
8428 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8429 
8430 	/*
8431 	 * src and dst are inverted, think how the receiver sees them
8432 	 *
8433 	 * The input is broken into two sections, a non-compressed section
8434 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8435 	 * is XORed together and stored in the compressed dword.
8436 	 */
8437 	input.formatted.vlan_id = vlan_id;
8438 
8439 	/*
8440 	 * since src port and flex bytes occupy the same word XOR them together
8441 	 * and write the value to source port portion of compressed dword
8442 	 */
8443 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8444 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8445 	else
8446 		common.port.src ^= th->dest ^ first->protocol;
8447 	common.port.dst ^= th->source;
8448 
8449 	switch (hdr.ipv4->version) {
8450 	case IPVERSION:
8451 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8452 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8453 		break;
8454 	case 6:
8455 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8456 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8457 			     hdr.ipv6->saddr.s6_addr32[1] ^
8458 			     hdr.ipv6->saddr.s6_addr32[2] ^
8459 			     hdr.ipv6->saddr.s6_addr32[3] ^
8460 			     hdr.ipv6->daddr.s6_addr32[0] ^
8461 			     hdr.ipv6->daddr.s6_addr32[1] ^
8462 			     hdr.ipv6->daddr.s6_addr32[2] ^
8463 			     hdr.ipv6->daddr.s6_addr32[3];
8464 		break;
8465 	default:
8466 		break;
8467 	}
8468 
8469 	if (hdr.network != skb_network_header(skb))
8470 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8471 
8472 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8473 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8474 					      input, common, ring->queue_index);
8475 }
8476 
8477 #ifdef IXGBE_FCOE
8478 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8479 			      struct net_device *sb_dev,
8480 			      select_queue_fallback_t fallback)
8481 {
8482 	struct ixgbe_adapter *adapter;
8483 	struct ixgbe_ring_feature *f;
8484 	int txq;
8485 
8486 	if (sb_dev) {
8487 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8488 		struct net_device *vdev = sb_dev;
8489 
8490 		txq = vdev->tc_to_txq[tc].offset;
8491 		txq += reciprocal_scale(skb_get_hash(skb),
8492 					vdev->tc_to_txq[tc].count);
8493 
8494 		return txq;
8495 	}
8496 
8497 	/*
8498 	 * only execute the code below if protocol is FCoE
8499 	 * or FIP and we have FCoE enabled on the adapter
8500 	 */
8501 	switch (vlan_get_protocol(skb)) {
8502 	case htons(ETH_P_FCOE):
8503 	case htons(ETH_P_FIP):
8504 		adapter = netdev_priv(dev);
8505 
8506 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8507 			break;
8508 		/* fall through */
8509 	default:
8510 		return fallback(dev, skb, sb_dev);
8511 	}
8512 
8513 	f = &adapter->ring_feature[RING_F_FCOE];
8514 
8515 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8516 					   smp_processor_id();
8517 
8518 	while (txq >= f->indices)
8519 		txq -= f->indices;
8520 
8521 	return txq + f->offset;
8522 }
8523 
8524 #endif
8525 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8526 			struct xdp_frame *xdpf)
8527 {
8528 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8529 	struct ixgbe_tx_buffer *tx_buffer;
8530 	union ixgbe_adv_tx_desc *tx_desc;
8531 	u32 len, cmd_type;
8532 	dma_addr_t dma;
8533 	u16 i;
8534 
8535 	len = xdpf->len;
8536 
8537 	if (unlikely(!ixgbe_desc_unused(ring)))
8538 		return IXGBE_XDP_CONSUMED;
8539 
8540 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8541 	if (dma_mapping_error(ring->dev, dma))
8542 		return IXGBE_XDP_CONSUMED;
8543 
8544 	/* record the location of the first descriptor for this packet */
8545 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8546 	tx_buffer->bytecount = len;
8547 	tx_buffer->gso_segs = 1;
8548 	tx_buffer->protocol = 0;
8549 
8550 	i = ring->next_to_use;
8551 	tx_desc = IXGBE_TX_DESC(ring, i);
8552 
8553 	dma_unmap_len_set(tx_buffer, len, len);
8554 	dma_unmap_addr_set(tx_buffer, dma, dma);
8555 	tx_buffer->xdpf = xdpf;
8556 
8557 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8558 
8559 	/* put descriptor type bits */
8560 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8561 		   IXGBE_ADVTXD_DCMD_DEXT |
8562 		   IXGBE_ADVTXD_DCMD_IFCS;
8563 	cmd_type |= len | IXGBE_TXD_CMD;
8564 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8565 	tx_desc->read.olinfo_status =
8566 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8567 
8568 	/* Avoid any potential race with xdp_xmit and cleanup */
8569 	smp_wmb();
8570 
8571 	/* set next_to_watch value indicating a packet is present */
8572 	i++;
8573 	if (i == ring->count)
8574 		i = 0;
8575 
8576 	tx_buffer->next_to_watch = tx_desc;
8577 	ring->next_to_use = i;
8578 
8579 	return IXGBE_XDP_TX;
8580 }
8581 
8582 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8583 			  struct ixgbe_adapter *adapter,
8584 			  struct ixgbe_ring *tx_ring)
8585 {
8586 	struct ixgbe_tx_buffer *first;
8587 	int tso;
8588 	u32 tx_flags = 0;
8589 	unsigned short f;
8590 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8591 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8592 	__be16 protocol = skb->protocol;
8593 	u8 hdr_len = 0;
8594 
8595 	/*
8596 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8597 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8598 	 *       + 2 desc gap to keep tail from touching head,
8599 	 *       + 1 desc for context descriptor,
8600 	 * otherwise try next time
8601 	 */
8602 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8603 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
8604 
8605 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8606 		tx_ring->tx_stats.tx_busy++;
8607 		return NETDEV_TX_BUSY;
8608 	}
8609 
8610 	/* record the location of the first descriptor for this packet */
8611 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8612 	first->skb = skb;
8613 	first->bytecount = skb->len;
8614 	first->gso_segs = 1;
8615 
8616 	/* if we have a HW VLAN tag being added default to the HW one */
8617 	if (skb_vlan_tag_present(skb)) {
8618 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8619 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8620 	/* else if it is a SW VLAN check the next protocol and store the tag */
8621 	} else if (protocol == htons(ETH_P_8021Q)) {
8622 		struct vlan_hdr *vhdr, _vhdr;
8623 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8624 		if (!vhdr)
8625 			goto out_drop;
8626 
8627 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8628 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8629 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8630 	}
8631 	protocol = vlan_get_protocol(skb);
8632 
8633 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8634 	    adapter->ptp_clock) {
8635 		if (!test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8636 					   &adapter->state)) {
8637 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8638 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8639 
8640 			/* schedule check for Tx timestamp */
8641 			adapter->ptp_tx_skb = skb_get(skb);
8642 			adapter->ptp_tx_start = jiffies;
8643 			schedule_work(&adapter->ptp_tx_work);
8644 		} else {
8645 			adapter->tx_hwtstamp_skipped++;
8646 		}
8647 	}
8648 
8649 	skb_tx_timestamp(skb);
8650 
8651 #ifdef CONFIG_PCI_IOV
8652 	/*
8653 	 * Use the l2switch_enable flag - would be false if the DMA
8654 	 * Tx switch had been disabled.
8655 	 */
8656 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8657 		tx_flags |= IXGBE_TX_FLAGS_CC;
8658 
8659 #endif
8660 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8661 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8662 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8663 	     (skb->priority != TC_PRIO_CONTROL))) {
8664 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8665 		tx_flags |= (skb->priority & 0x7) <<
8666 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8667 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8668 			struct vlan_ethhdr *vhdr;
8669 
8670 			if (skb_cow_head(skb, 0))
8671 				goto out_drop;
8672 			vhdr = (struct vlan_ethhdr *)skb->data;
8673 			vhdr->h_vlan_TCI = htons(tx_flags >>
8674 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8675 		} else {
8676 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8677 		}
8678 	}
8679 
8680 	/* record initial flags and protocol */
8681 	first->tx_flags = tx_flags;
8682 	first->protocol = protocol;
8683 
8684 #ifdef IXGBE_FCOE
8685 	/* setup tx offload for FCoE */
8686 	if ((protocol == htons(ETH_P_FCOE)) &&
8687 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8688 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8689 		if (tso < 0)
8690 			goto out_drop;
8691 
8692 		goto xmit_fcoe;
8693 	}
8694 
8695 #endif /* IXGBE_FCOE */
8696 
8697 #ifdef CONFIG_IXGBE_IPSEC
8698 	if (skb->sp && !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8699 		goto out_drop;
8700 #endif
8701 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8702 	if (tso < 0)
8703 		goto out_drop;
8704 	else if (!tso)
8705 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8706 
8707 	/* add the ATR filter if ATR is on */
8708 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8709 		ixgbe_atr(tx_ring, first);
8710 
8711 #ifdef IXGBE_FCOE
8712 xmit_fcoe:
8713 #endif /* IXGBE_FCOE */
8714 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8715 		goto cleanup_tx_timestamp;
8716 
8717 	return NETDEV_TX_OK;
8718 
8719 out_drop:
8720 	dev_kfree_skb_any(first->skb);
8721 	first->skb = NULL;
8722 cleanup_tx_timestamp:
8723 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8724 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8725 		adapter->ptp_tx_skb = NULL;
8726 		cancel_work_sync(&adapter->ptp_tx_work);
8727 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8728 	}
8729 
8730 	return NETDEV_TX_OK;
8731 }
8732 
8733 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8734 				      struct net_device *netdev,
8735 				      struct ixgbe_ring *ring)
8736 {
8737 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8738 	struct ixgbe_ring *tx_ring;
8739 
8740 	/*
8741 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8742 	 * in order to meet this minimum size requirement.
8743 	 */
8744 	if (skb_put_padto(skb, 17))
8745 		return NETDEV_TX_OK;
8746 
8747 	tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
8748 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8749 		return NETDEV_TX_BUSY;
8750 
8751 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8752 }
8753 
8754 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8755 				    struct net_device *netdev)
8756 {
8757 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8758 }
8759 
8760 /**
8761  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8762  * @netdev: network interface device structure
8763  * @p: pointer to an address structure
8764  *
8765  * Returns 0 on success, negative on failure
8766  **/
8767 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8768 {
8769 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8770 	struct ixgbe_hw *hw = &adapter->hw;
8771 	struct sockaddr *addr = p;
8772 
8773 	if (!is_valid_ether_addr(addr->sa_data))
8774 		return -EADDRNOTAVAIL;
8775 
8776 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8777 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8778 
8779 	ixgbe_mac_set_default_filter(adapter);
8780 
8781 	return 0;
8782 }
8783 
8784 static int
8785 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8786 {
8787 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8788 	struct ixgbe_hw *hw = &adapter->hw;
8789 	u16 value;
8790 	int rc;
8791 
8792 	if (prtad != hw->phy.mdio.prtad)
8793 		return -EINVAL;
8794 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8795 	if (!rc)
8796 		rc = value;
8797 	return rc;
8798 }
8799 
8800 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8801 			    u16 addr, u16 value)
8802 {
8803 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8804 	struct ixgbe_hw *hw = &adapter->hw;
8805 
8806 	if (prtad != hw->phy.mdio.prtad)
8807 		return -EINVAL;
8808 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8809 }
8810 
8811 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8812 {
8813 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8814 
8815 	switch (cmd) {
8816 	case SIOCSHWTSTAMP:
8817 		return ixgbe_ptp_set_ts_config(adapter, req);
8818 	case SIOCGHWTSTAMP:
8819 		return ixgbe_ptp_get_ts_config(adapter, req);
8820 	case SIOCGMIIPHY:
8821 		if (!adapter->hw.phy.ops.read_reg)
8822 			return -EOPNOTSUPP;
8823 		/* fall through */
8824 	default:
8825 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8826 	}
8827 }
8828 
8829 /**
8830  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8831  * netdev->dev_addrs
8832  * @dev: network interface device structure
8833  *
8834  * Returns non-zero on failure
8835  **/
8836 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8837 {
8838 	int err = 0;
8839 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8840 	struct ixgbe_hw *hw = &adapter->hw;
8841 
8842 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8843 		rtnl_lock();
8844 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8845 		rtnl_unlock();
8846 
8847 		/* update SAN MAC vmdq pool selection */
8848 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8849 	}
8850 	return err;
8851 }
8852 
8853 /**
8854  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8855  * netdev->dev_addrs
8856  * @dev: network interface device structure
8857  *
8858  * Returns non-zero on failure
8859  **/
8860 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8861 {
8862 	int err = 0;
8863 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8864 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8865 
8866 	if (is_valid_ether_addr(mac->san_addr)) {
8867 		rtnl_lock();
8868 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8869 		rtnl_unlock();
8870 	}
8871 	return err;
8872 }
8873 
8874 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8875 				   struct ixgbe_ring *ring)
8876 {
8877 	u64 bytes, packets;
8878 	unsigned int start;
8879 
8880 	if (ring) {
8881 		do {
8882 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8883 			packets = ring->stats.packets;
8884 			bytes   = ring->stats.bytes;
8885 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8886 		stats->tx_packets += packets;
8887 		stats->tx_bytes   += bytes;
8888 	}
8889 }
8890 
8891 static void ixgbe_get_stats64(struct net_device *netdev,
8892 			      struct rtnl_link_stats64 *stats)
8893 {
8894 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8895 	int i;
8896 
8897 	rcu_read_lock();
8898 	for (i = 0; i < adapter->num_rx_queues; i++) {
8899 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8900 		u64 bytes, packets;
8901 		unsigned int start;
8902 
8903 		if (ring) {
8904 			do {
8905 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8906 				packets = ring->stats.packets;
8907 				bytes   = ring->stats.bytes;
8908 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8909 			stats->rx_packets += packets;
8910 			stats->rx_bytes   += bytes;
8911 		}
8912 	}
8913 
8914 	for (i = 0; i < adapter->num_tx_queues; i++) {
8915 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8916 
8917 		ixgbe_get_ring_stats64(stats, ring);
8918 	}
8919 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8920 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8921 
8922 		ixgbe_get_ring_stats64(stats, ring);
8923 	}
8924 	rcu_read_unlock();
8925 
8926 	/* following stats updated by ixgbe_watchdog_task() */
8927 	stats->multicast	= netdev->stats.multicast;
8928 	stats->rx_errors	= netdev->stats.rx_errors;
8929 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
8930 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
8931 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
8932 }
8933 
8934 #ifdef CONFIG_IXGBE_DCB
8935 /**
8936  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8937  * @adapter: pointer to ixgbe_adapter
8938  * @tc: number of traffic classes currently enabled
8939  *
8940  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8941  * 802.1Q priority maps to a packet buffer that exists.
8942  */
8943 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8944 {
8945 	struct ixgbe_hw *hw = &adapter->hw;
8946 	u32 reg, rsave;
8947 	int i;
8948 
8949 	/* 82598 have a static priority to TC mapping that can not
8950 	 * be changed so no validation is needed.
8951 	 */
8952 	if (hw->mac.type == ixgbe_mac_82598EB)
8953 		return;
8954 
8955 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8956 	rsave = reg;
8957 
8958 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8959 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8960 
8961 		/* If up2tc is out of bounds default to zero */
8962 		if (up2tc > tc)
8963 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8964 	}
8965 
8966 	if (reg != rsave)
8967 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8968 
8969 	return;
8970 }
8971 
8972 /**
8973  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8974  * @adapter: Pointer to adapter struct
8975  *
8976  * Populate the netdev user priority to tc map
8977  */
8978 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
8979 {
8980 	struct net_device *dev = adapter->netdev;
8981 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
8982 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
8983 	u8 prio;
8984 
8985 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
8986 		u8 tc = 0;
8987 
8988 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
8989 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
8990 		else if (ets)
8991 			tc = ets->prio_tc[prio];
8992 
8993 		netdev_set_prio_tc_map(dev, prio, tc);
8994 	}
8995 }
8996 
8997 #endif /* CONFIG_IXGBE_DCB */
8998 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
8999 {
9000 	struct ixgbe_adapter *adapter = data;
9001 	struct ixgbe_fwd_adapter *accel;
9002 	int pool;
9003 
9004 	/* we only care about macvlans... */
9005 	if (!netif_is_macvlan(vdev))
9006 		return 0;
9007 
9008 	/* that have hardware offload enabled... */
9009 	accel = macvlan_accel_priv(vdev);
9010 	if (!accel)
9011 		return 0;
9012 
9013 	/* If we can relocate to a different bit do so */
9014 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9015 	if (pool < adapter->num_rx_pools) {
9016 		set_bit(pool, adapter->fwd_bitmask);
9017 		accel->pool = pool;
9018 		return 0;
9019 	}
9020 
9021 	/* if we cannot find a free pool then disable the offload */
9022 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9023 	macvlan_release_l2fw_offload(vdev);
9024 
9025 	/* unbind the queues and drop the subordinate channel config */
9026 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9027 	netdev_set_sb_channel(vdev, 0);
9028 
9029 	kfree(accel);
9030 
9031 	return 0;
9032 }
9033 
9034 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9035 {
9036 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9037 
9038 	/* flush any stale bits out of the fwd bitmask */
9039 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9040 
9041 	/* walk through upper devices reassigning pools */
9042 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9043 				      adapter);
9044 }
9045 
9046 /**
9047  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9048  *
9049  * @dev: net device to configure
9050  * @tc: number of traffic classes to enable
9051  */
9052 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9053 {
9054 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9055 	struct ixgbe_hw *hw = &adapter->hw;
9056 
9057 	/* Hardware supports up to 8 traffic classes */
9058 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9059 		return -EINVAL;
9060 
9061 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9062 		return -EINVAL;
9063 
9064 	/* Hardware has to reinitialize queues and interrupts to
9065 	 * match packet buffer alignment. Unfortunately, the
9066 	 * hardware is not flexible enough to do this dynamically.
9067 	 */
9068 	if (netif_running(dev))
9069 		ixgbe_close(dev);
9070 	else
9071 		ixgbe_reset(adapter);
9072 
9073 	ixgbe_clear_interrupt_scheme(adapter);
9074 
9075 #ifdef CONFIG_IXGBE_DCB
9076 	if (tc) {
9077 		if (adapter->xdp_prog) {
9078 			e_warn(probe, "DCB is not supported with XDP\n");
9079 
9080 			ixgbe_init_interrupt_scheme(adapter);
9081 			if (netif_running(dev))
9082 				ixgbe_open(dev);
9083 			return -EINVAL;
9084 		}
9085 
9086 		netdev_set_num_tc(dev, tc);
9087 		ixgbe_set_prio_tc_map(adapter);
9088 
9089 		adapter->hw_tcs = tc;
9090 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9091 
9092 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9093 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9094 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9095 		}
9096 	} else {
9097 		netdev_reset_tc(dev);
9098 
9099 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9100 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9101 
9102 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9103 		adapter->hw_tcs = tc;
9104 
9105 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9106 		adapter->dcb_cfg.pfc_mode_enable = false;
9107 	}
9108 
9109 	ixgbe_validate_rtr(adapter, tc);
9110 
9111 #endif /* CONFIG_IXGBE_DCB */
9112 	ixgbe_init_interrupt_scheme(adapter);
9113 
9114 	ixgbe_defrag_macvlan_pools(dev);
9115 
9116 	if (netif_running(dev))
9117 		return ixgbe_open(dev);
9118 
9119 	return 0;
9120 }
9121 
9122 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9123 			       struct tc_cls_u32_offload *cls)
9124 {
9125 	u32 hdl = cls->knode.handle;
9126 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9127 	u32 loc = cls->knode.handle & 0xfffff;
9128 	int err = 0, i, j;
9129 	struct ixgbe_jump_table *jump = NULL;
9130 
9131 	if (loc > IXGBE_MAX_HW_ENTRIES)
9132 		return -EINVAL;
9133 
9134 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9135 		return -EINVAL;
9136 
9137 	/* Clear this filter in the link data it is associated with */
9138 	if (uhtid != 0x800) {
9139 		jump = adapter->jump_tables[uhtid];
9140 		if (!jump)
9141 			return -EINVAL;
9142 		if (!test_bit(loc - 1, jump->child_loc_map))
9143 			return -EINVAL;
9144 		clear_bit(loc - 1, jump->child_loc_map);
9145 	}
9146 
9147 	/* Check if the filter being deleted is a link */
9148 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9149 		jump = adapter->jump_tables[i];
9150 		if (jump && jump->link_hdl == hdl) {
9151 			/* Delete filters in the hardware in the child hash
9152 			 * table associated with this link
9153 			 */
9154 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9155 				if (!test_bit(j, jump->child_loc_map))
9156 					continue;
9157 				spin_lock(&adapter->fdir_perfect_lock);
9158 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9159 								      NULL,
9160 								      j + 1);
9161 				spin_unlock(&adapter->fdir_perfect_lock);
9162 				clear_bit(j, jump->child_loc_map);
9163 			}
9164 			/* Remove resources for this link */
9165 			kfree(jump->input);
9166 			kfree(jump->mask);
9167 			kfree(jump);
9168 			adapter->jump_tables[i] = NULL;
9169 			return err;
9170 		}
9171 	}
9172 
9173 	spin_lock(&adapter->fdir_perfect_lock);
9174 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9175 	spin_unlock(&adapter->fdir_perfect_lock);
9176 	return err;
9177 }
9178 
9179 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9180 					    struct tc_cls_u32_offload *cls)
9181 {
9182 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9183 
9184 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9185 		return -EINVAL;
9186 
9187 	/* This ixgbe devices do not support hash tables at the moment
9188 	 * so abort when given hash tables.
9189 	 */
9190 	if (cls->hnode.divisor > 0)
9191 		return -EINVAL;
9192 
9193 	set_bit(uhtid - 1, &adapter->tables);
9194 	return 0;
9195 }
9196 
9197 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9198 					    struct tc_cls_u32_offload *cls)
9199 {
9200 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9201 
9202 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9203 		return -EINVAL;
9204 
9205 	clear_bit(uhtid - 1, &adapter->tables);
9206 	return 0;
9207 }
9208 
9209 #ifdef CONFIG_NET_CLS_ACT
9210 struct upper_walk_data {
9211 	struct ixgbe_adapter *adapter;
9212 	u64 action;
9213 	int ifindex;
9214 	u8 queue;
9215 };
9216 
9217 static int get_macvlan_queue(struct net_device *upper, void *_data)
9218 {
9219 	if (netif_is_macvlan(upper)) {
9220 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9221 		struct upper_walk_data *data = _data;
9222 		struct ixgbe_adapter *adapter = data->adapter;
9223 		int ifindex = data->ifindex;
9224 
9225 		if (vadapter && upper->ifindex == ifindex) {
9226 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9227 			data->action = data->queue;
9228 			return 1;
9229 		}
9230 	}
9231 
9232 	return 0;
9233 }
9234 
9235 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9236 				  u8 *queue, u64 *action)
9237 {
9238 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9239 	unsigned int num_vfs = adapter->num_vfs, vf;
9240 	struct upper_walk_data data;
9241 	struct net_device *upper;
9242 
9243 	/* redirect to a SRIOV VF */
9244 	for (vf = 0; vf < num_vfs; ++vf) {
9245 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9246 		if (upper->ifindex == ifindex) {
9247 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9248 			*action = vf + 1;
9249 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9250 			return 0;
9251 		}
9252 	}
9253 
9254 	/* redirect to a offloaded macvlan netdev */
9255 	data.adapter = adapter;
9256 	data.ifindex = ifindex;
9257 	data.action = 0;
9258 	data.queue = 0;
9259 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9260 					  get_macvlan_queue, &data)) {
9261 		*action = data.action;
9262 		*queue = data.queue;
9263 
9264 		return 0;
9265 	}
9266 
9267 	return -EINVAL;
9268 }
9269 
9270 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9271 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9272 {
9273 	const struct tc_action *a;
9274 	int i;
9275 
9276 	if (!tcf_exts_has_actions(exts))
9277 		return -EINVAL;
9278 
9279 	tcf_exts_for_each_action(i, a, exts) {
9280 		/* Drop action */
9281 		if (is_tcf_gact_shot(a)) {
9282 			*action = IXGBE_FDIR_DROP_QUEUE;
9283 			*queue = IXGBE_FDIR_DROP_QUEUE;
9284 			return 0;
9285 		}
9286 
9287 		/* Redirect to a VF or a offloaded macvlan */
9288 		if (is_tcf_mirred_egress_redirect(a)) {
9289 			struct net_device *dev = tcf_mirred_dev(a);
9290 
9291 			if (!dev)
9292 				return -EINVAL;
9293 			return handle_redirect_action(adapter, dev->ifindex,
9294 						      queue, action);
9295 		}
9296 
9297 		return -EINVAL;
9298 	}
9299 
9300 	return -EINVAL;
9301 }
9302 #else
9303 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9304 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9305 {
9306 	return -EINVAL;
9307 }
9308 #endif /* CONFIG_NET_CLS_ACT */
9309 
9310 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9311 				    union ixgbe_atr_input *mask,
9312 				    struct tc_cls_u32_offload *cls,
9313 				    struct ixgbe_mat_field *field_ptr,
9314 				    struct ixgbe_nexthdr *nexthdr)
9315 {
9316 	int i, j, off;
9317 	__be32 val, m;
9318 	bool found_entry = false, found_jump_field = false;
9319 
9320 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9321 		off = cls->knode.sel->keys[i].off;
9322 		val = cls->knode.sel->keys[i].val;
9323 		m = cls->knode.sel->keys[i].mask;
9324 
9325 		for (j = 0; field_ptr[j].val; j++) {
9326 			if (field_ptr[j].off == off) {
9327 				field_ptr[j].val(input, mask, (__force u32)val,
9328 						 (__force u32)m);
9329 				input->filter.formatted.flow_type |=
9330 					field_ptr[j].type;
9331 				found_entry = true;
9332 				break;
9333 			}
9334 		}
9335 		if (nexthdr) {
9336 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9337 			    nexthdr->val ==
9338 			    (__force u32)cls->knode.sel->keys[i].val &&
9339 			    nexthdr->mask ==
9340 			    (__force u32)cls->knode.sel->keys[i].mask)
9341 				found_jump_field = true;
9342 			else
9343 				continue;
9344 		}
9345 	}
9346 
9347 	if (nexthdr && !found_jump_field)
9348 		return -EINVAL;
9349 
9350 	if (!found_entry)
9351 		return 0;
9352 
9353 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9354 				    IXGBE_ATR_L4TYPE_MASK;
9355 
9356 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9357 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9358 
9359 	return 0;
9360 }
9361 
9362 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9363 				  struct tc_cls_u32_offload *cls)
9364 {
9365 	__be16 protocol = cls->common.protocol;
9366 	u32 loc = cls->knode.handle & 0xfffff;
9367 	struct ixgbe_hw *hw = &adapter->hw;
9368 	struct ixgbe_mat_field *field_ptr;
9369 	struct ixgbe_fdir_filter *input = NULL;
9370 	union ixgbe_atr_input *mask = NULL;
9371 	struct ixgbe_jump_table *jump = NULL;
9372 	int i, err = -EINVAL;
9373 	u8 queue;
9374 	u32 uhtid, link_uhtid;
9375 
9376 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9377 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9378 
9379 	/* At the moment cls_u32 jumps to network layer and skips past
9380 	 * L2 headers. The canonical method to match L2 frames is to use
9381 	 * negative values. However this is error prone at best but really
9382 	 * just broken because there is no way to "know" what sort of hdr
9383 	 * is in front of the network layer. Fix cls_u32 to support L2
9384 	 * headers when needed.
9385 	 */
9386 	if (protocol != htons(ETH_P_IP))
9387 		return err;
9388 
9389 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9390 		e_err(drv, "Location out of range\n");
9391 		return err;
9392 	}
9393 
9394 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9395 	 * links and also the fields used to advance the parser across each
9396 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9397 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9398 	 * To add support for new nodes update ixgbe_model.h parse structures
9399 	 * this function _should_ be generic try not to hardcode values here.
9400 	 */
9401 	if (uhtid == 0x800) {
9402 		field_ptr = (adapter->jump_tables[0])->mat;
9403 	} else {
9404 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9405 			return err;
9406 		if (!adapter->jump_tables[uhtid])
9407 			return err;
9408 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9409 	}
9410 
9411 	if (!field_ptr)
9412 		return err;
9413 
9414 	/* At this point we know the field_ptr is valid and need to either
9415 	 * build cls_u32 link or attach filter. Because adding a link to
9416 	 * a handle that does not exist is invalid and the same for adding
9417 	 * rules to handles that don't exist.
9418 	 */
9419 
9420 	if (link_uhtid) {
9421 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9422 
9423 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9424 			return err;
9425 
9426 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9427 			return err;
9428 
9429 		/* Multiple filters as links to the same hash table are not
9430 		 * supported. To add a new filter with the same next header
9431 		 * but different match/jump conditions, create a new hash table
9432 		 * and link to it.
9433 		 */
9434 		if (adapter->jump_tables[link_uhtid] &&
9435 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9436 			e_err(drv, "Link filter exists for link: %x\n",
9437 			      link_uhtid);
9438 			return err;
9439 		}
9440 
9441 		for (i = 0; nexthdr[i].jump; i++) {
9442 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9443 			    nexthdr[i].s != cls->knode.sel->offshift ||
9444 			    nexthdr[i].m !=
9445 			    (__force u32)cls->knode.sel->offmask)
9446 				return err;
9447 
9448 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9449 			if (!jump)
9450 				return -ENOMEM;
9451 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9452 			if (!input) {
9453 				err = -ENOMEM;
9454 				goto free_jump;
9455 			}
9456 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9457 			if (!mask) {
9458 				err = -ENOMEM;
9459 				goto free_input;
9460 			}
9461 			jump->input = input;
9462 			jump->mask = mask;
9463 			jump->link_hdl = cls->knode.handle;
9464 
9465 			err = ixgbe_clsu32_build_input(input, mask, cls,
9466 						       field_ptr, &nexthdr[i]);
9467 			if (!err) {
9468 				jump->mat = nexthdr[i].jump;
9469 				adapter->jump_tables[link_uhtid] = jump;
9470 				break;
9471 			}
9472 		}
9473 		return 0;
9474 	}
9475 
9476 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9477 	if (!input)
9478 		return -ENOMEM;
9479 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9480 	if (!mask) {
9481 		err = -ENOMEM;
9482 		goto free_input;
9483 	}
9484 
9485 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9486 		if ((adapter->jump_tables[uhtid])->input)
9487 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9488 			       sizeof(*input));
9489 		if ((adapter->jump_tables[uhtid])->mask)
9490 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9491 			       sizeof(*mask));
9492 
9493 		/* Lookup in all child hash tables if this location is already
9494 		 * filled with a filter
9495 		 */
9496 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9497 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9498 
9499 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9500 				e_err(drv, "Filter exists in location: %x\n",
9501 				      loc);
9502 				err = -EINVAL;
9503 				goto err_out;
9504 			}
9505 		}
9506 	}
9507 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9508 	if (err)
9509 		goto err_out;
9510 
9511 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9512 			       &queue);
9513 	if (err < 0)
9514 		goto err_out;
9515 
9516 	input->sw_idx = loc;
9517 
9518 	spin_lock(&adapter->fdir_perfect_lock);
9519 
9520 	if (hlist_empty(&adapter->fdir_filter_list)) {
9521 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9522 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9523 		if (err)
9524 			goto err_out_w_lock;
9525 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9526 		err = -EINVAL;
9527 		goto err_out_w_lock;
9528 	}
9529 
9530 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9531 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9532 						    input->sw_idx, queue);
9533 	if (!err)
9534 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9535 	spin_unlock(&adapter->fdir_perfect_lock);
9536 
9537 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9538 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9539 
9540 	kfree(mask);
9541 	return err;
9542 err_out_w_lock:
9543 	spin_unlock(&adapter->fdir_perfect_lock);
9544 err_out:
9545 	kfree(mask);
9546 free_input:
9547 	kfree(input);
9548 free_jump:
9549 	kfree(jump);
9550 	return err;
9551 }
9552 
9553 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9554 				  struct tc_cls_u32_offload *cls_u32)
9555 {
9556 	switch (cls_u32->command) {
9557 	case TC_CLSU32_NEW_KNODE:
9558 	case TC_CLSU32_REPLACE_KNODE:
9559 		return ixgbe_configure_clsu32(adapter, cls_u32);
9560 	case TC_CLSU32_DELETE_KNODE:
9561 		return ixgbe_delete_clsu32(adapter, cls_u32);
9562 	case TC_CLSU32_NEW_HNODE:
9563 	case TC_CLSU32_REPLACE_HNODE:
9564 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9565 	case TC_CLSU32_DELETE_HNODE:
9566 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9567 	default:
9568 		return -EOPNOTSUPP;
9569 	}
9570 }
9571 
9572 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9573 				   void *cb_priv)
9574 {
9575 	struct ixgbe_adapter *adapter = cb_priv;
9576 
9577 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9578 		return -EOPNOTSUPP;
9579 
9580 	switch (type) {
9581 	case TC_SETUP_CLSU32:
9582 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9583 	default:
9584 		return -EOPNOTSUPP;
9585 	}
9586 }
9587 
9588 static int ixgbe_setup_tc_block(struct net_device *dev,
9589 				struct tc_block_offload *f)
9590 {
9591 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9592 
9593 	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
9594 		return -EOPNOTSUPP;
9595 
9596 	switch (f->command) {
9597 	case TC_BLOCK_BIND:
9598 		return tcf_block_cb_register(f->block, ixgbe_setup_tc_block_cb,
9599 					     adapter, adapter, f->extack);
9600 	case TC_BLOCK_UNBIND:
9601 		tcf_block_cb_unregister(f->block, ixgbe_setup_tc_block_cb,
9602 					adapter);
9603 		return 0;
9604 	default:
9605 		return -EOPNOTSUPP;
9606 	}
9607 }
9608 
9609 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9610 				 struct tc_mqprio_qopt *mqprio)
9611 {
9612 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9613 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9614 }
9615 
9616 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9617 			    void *type_data)
9618 {
9619 	switch (type) {
9620 	case TC_SETUP_BLOCK:
9621 		return ixgbe_setup_tc_block(dev, type_data);
9622 	case TC_SETUP_QDISC_MQPRIO:
9623 		return ixgbe_setup_tc_mqprio(dev, type_data);
9624 	default:
9625 		return -EOPNOTSUPP;
9626 	}
9627 }
9628 
9629 #ifdef CONFIG_PCI_IOV
9630 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9631 {
9632 	struct net_device *netdev = adapter->netdev;
9633 
9634 	rtnl_lock();
9635 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9636 	rtnl_unlock();
9637 }
9638 
9639 #endif
9640 void ixgbe_do_reset(struct net_device *netdev)
9641 {
9642 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9643 
9644 	if (netif_running(netdev))
9645 		ixgbe_reinit_locked(adapter);
9646 	else
9647 		ixgbe_reset(adapter);
9648 }
9649 
9650 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9651 					    netdev_features_t features)
9652 {
9653 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9654 
9655 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9656 	if (!(features & NETIF_F_RXCSUM))
9657 		features &= ~NETIF_F_LRO;
9658 
9659 	/* Turn off LRO if not RSC capable */
9660 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9661 		features &= ~NETIF_F_LRO;
9662 
9663 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9664 		e_dev_err("LRO is not supported with XDP\n");
9665 		features &= ~NETIF_F_LRO;
9666 	}
9667 
9668 	return features;
9669 }
9670 
9671 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9672 {
9673 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9674 			num_online_cpus());
9675 
9676 	/* go back to full RSS if we're not running SR-IOV */
9677 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9678 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9679 				    IXGBE_FLAG_SRIOV_ENABLED);
9680 
9681 	adapter->ring_feature[RING_F_RSS].limit = rss;
9682 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9683 
9684 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9685 }
9686 
9687 static int ixgbe_set_features(struct net_device *netdev,
9688 			      netdev_features_t features)
9689 {
9690 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9691 	netdev_features_t changed = netdev->features ^ features;
9692 	bool need_reset = false;
9693 
9694 	/* Make sure RSC matches LRO, reset if change */
9695 	if (!(features & NETIF_F_LRO)) {
9696 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9697 			need_reset = true;
9698 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9699 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9700 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9701 		if (adapter->rx_itr_setting == 1 ||
9702 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9703 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9704 			need_reset = true;
9705 		} else if ((changed ^ features) & NETIF_F_LRO) {
9706 			e_info(probe, "rx-usecs set too low, "
9707 			       "disabling RSC\n");
9708 		}
9709 	}
9710 
9711 	/*
9712 	 * Check if Flow Director n-tuple support or hw_tc support was
9713 	 * enabled or disabled.  If the state changed, we need to reset.
9714 	 */
9715 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9716 		/* turn off ATR, enable perfect filters and reset */
9717 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9718 			need_reset = true;
9719 
9720 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9721 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9722 	} else {
9723 		/* turn off perfect filters, enable ATR and reset */
9724 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9725 			need_reset = true;
9726 
9727 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9728 
9729 		/* We cannot enable ATR if SR-IOV is enabled */
9730 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9731 		    /* We cannot enable ATR if we have 2 or more tcs */
9732 		    (adapter->hw_tcs > 1) ||
9733 		    /* We cannot enable ATR if RSS is disabled */
9734 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9735 		    /* A sample rate of 0 indicates ATR disabled */
9736 		    (!adapter->atr_sample_rate))
9737 			; /* do nothing not supported */
9738 		else /* otherwise supported and set the flag */
9739 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9740 	}
9741 
9742 	if (changed & NETIF_F_RXALL)
9743 		need_reset = true;
9744 
9745 	netdev->features = features;
9746 
9747 	if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9748 		if (features & NETIF_F_RXCSUM) {
9749 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9750 		} else {
9751 			u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9752 
9753 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9754 		}
9755 	}
9756 
9757 	if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9758 		if (features & NETIF_F_RXCSUM) {
9759 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9760 		} else {
9761 			u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9762 
9763 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9764 		}
9765 	}
9766 
9767 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9768 		ixgbe_reset_l2fw_offload(adapter);
9769 	else if (need_reset)
9770 		ixgbe_do_reset(netdev);
9771 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9772 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9773 		ixgbe_set_rx_mode(netdev);
9774 
9775 	return 0;
9776 }
9777 
9778 /**
9779  * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9780  * @dev: The port's netdev
9781  * @ti: Tunnel endpoint information
9782  **/
9783 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9784 				      struct udp_tunnel_info *ti)
9785 {
9786 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9787 	struct ixgbe_hw *hw = &adapter->hw;
9788 	__be16 port = ti->port;
9789 	u32 port_shift = 0;
9790 	u32 reg;
9791 
9792 	if (ti->sa_family != AF_INET)
9793 		return;
9794 
9795 	switch (ti->type) {
9796 	case UDP_TUNNEL_TYPE_VXLAN:
9797 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9798 			return;
9799 
9800 		if (adapter->vxlan_port == port)
9801 			return;
9802 
9803 		if (adapter->vxlan_port) {
9804 			netdev_info(dev,
9805 				    "VXLAN port %d set, not adding port %d\n",
9806 				    ntohs(adapter->vxlan_port),
9807 				    ntohs(port));
9808 			return;
9809 		}
9810 
9811 		adapter->vxlan_port = port;
9812 		break;
9813 	case UDP_TUNNEL_TYPE_GENEVE:
9814 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9815 			return;
9816 
9817 		if (adapter->geneve_port == port)
9818 			return;
9819 
9820 		if (adapter->geneve_port) {
9821 			netdev_info(dev,
9822 				    "GENEVE port %d set, not adding port %d\n",
9823 				    ntohs(adapter->geneve_port),
9824 				    ntohs(port));
9825 			return;
9826 		}
9827 
9828 		port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9829 		adapter->geneve_port = port;
9830 		break;
9831 	default:
9832 		return;
9833 	}
9834 
9835 	reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9836 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9837 }
9838 
9839 /**
9840  * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9841  * @dev: The port's netdev
9842  * @ti: Tunnel endpoint information
9843  **/
9844 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9845 				      struct udp_tunnel_info *ti)
9846 {
9847 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9848 	u32 port_mask;
9849 
9850 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9851 	    ti->type != UDP_TUNNEL_TYPE_GENEVE)
9852 		return;
9853 
9854 	if (ti->sa_family != AF_INET)
9855 		return;
9856 
9857 	switch (ti->type) {
9858 	case UDP_TUNNEL_TYPE_VXLAN:
9859 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9860 			return;
9861 
9862 		if (adapter->vxlan_port != ti->port) {
9863 			netdev_info(dev, "VXLAN port %d not found\n",
9864 				    ntohs(ti->port));
9865 			return;
9866 		}
9867 
9868 		port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9869 		break;
9870 	case UDP_TUNNEL_TYPE_GENEVE:
9871 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9872 			return;
9873 
9874 		if (adapter->geneve_port != ti->port) {
9875 			netdev_info(dev, "GENEVE port %d not found\n",
9876 				    ntohs(ti->port));
9877 			return;
9878 		}
9879 
9880 		port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9881 		break;
9882 	default:
9883 		return;
9884 	}
9885 
9886 	ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9887 	adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9888 }
9889 
9890 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9891 			     struct net_device *dev,
9892 			     const unsigned char *addr, u16 vid,
9893 			     u16 flags)
9894 {
9895 	/* guarantee we can provide a unique filter for the unicast address */
9896 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9897 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9898 		u16 pool = VMDQ_P(0);
9899 
9900 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9901 			return -ENOMEM;
9902 	}
9903 
9904 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9905 }
9906 
9907 /**
9908  * ixgbe_configure_bridge_mode - set various bridge modes
9909  * @adapter: the private structure
9910  * @mode: requested bridge mode
9911  *
9912  * Configure some settings require for various bridge modes.
9913  **/
9914 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9915 				       __u16 mode)
9916 {
9917 	struct ixgbe_hw *hw = &adapter->hw;
9918 	unsigned int p, num_pools;
9919 	u32 vmdctl;
9920 
9921 	switch (mode) {
9922 	case BRIDGE_MODE_VEPA:
9923 		/* disable Tx loopback, rely on switch hairpin mode */
9924 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9925 
9926 		/* must enable Rx switching replication to allow multicast
9927 		 * packet reception on all VFs, and to enable source address
9928 		 * pruning.
9929 		 */
9930 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9931 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9932 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9933 
9934 		/* enable Rx source address pruning. Note, this requires
9935 		 * replication to be enabled or else it does nothing.
9936 		 */
9937 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9938 		for (p = 0; p < num_pools; p++) {
9939 			if (hw->mac.ops.set_source_address_pruning)
9940 				hw->mac.ops.set_source_address_pruning(hw,
9941 								       true,
9942 								       p);
9943 		}
9944 		break;
9945 	case BRIDGE_MODE_VEB:
9946 		/* enable Tx loopback for internal VF/PF communication */
9947 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9948 				IXGBE_PFDTXGSWC_VT_LBEN);
9949 
9950 		/* disable Rx switching replication unless we have SR-IOV
9951 		 * virtual functions
9952 		 */
9953 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9954 		if (!adapter->num_vfs)
9955 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9956 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9957 
9958 		/* disable Rx source address pruning, since we don't expect to
9959 		 * be receiving external loopback of our transmitted frames.
9960 		 */
9961 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
9962 		for (p = 0; p < num_pools; p++) {
9963 			if (hw->mac.ops.set_source_address_pruning)
9964 				hw->mac.ops.set_source_address_pruning(hw,
9965 								       false,
9966 								       p);
9967 		}
9968 		break;
9969 	default:
9970 		return -EINVAL;
9971 	}
9972 
9973 	adapter->bridge_mode = mode;
9974 
9975 	e_info(drv, "enabling bridge mode: %s\n",
9976 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9977 
9978 	return 0;
9979 }
9980 
9981 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9982 				    struct nlmsghdr *nlh, u16 flags)
9983 {
9984 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9985 	struct nlattr *attr, *br_spec;
9986 	int rem;
9987 
9988 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9989 		return -EOPNOTSUPP;
9990 
9991 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9992 	if (!br_spec)
9993 		return -EINVAL;
9994 
9995 	nla_for_each_nested(attr, br_spec, rem) {
9996 		int status;
9997 		__u16 mode;
9998 
9999 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
10000 			continue;
10001 
10002 		if (nla_len(attr) < sizeof(mode))
10003 			return -EINVAL;
10004 
10005 		mode = nla_get_u16(attr);
10006 		status = ixgbe_configure_bridge_mode(adapter, mode);
10007 		if (status)
10008 			return status;
10009 
10010 		break;
10011 	}
10012 
10013 	return 0;
10014 }
10015 
10016 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10017 				    struct net_device *dev,
10018 				    u32 filter_mask, int nlflags)
10019 {
10020 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10021 
10022 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10023 		return 0;
10024 
10025 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10026 				       adapter->bridge_mode, 0, 0, nlflags,
10027 				       filter_mask, NULL);
10028 }
10029 
10030 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10031 {
10032 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10033 	struct ixgbe_fwd_adapter *accel;
10034 	int tcs = adapter->hw_tcs ? : 1;
10035 	int pool, err;
10036 
10037 	if (adapter->xdp_prog) {
10038 		e_warn(probe, "L2FW offload is not supported with XDP\n");
10039 		return ERR_PTR(-EINVAL);
10040 	}
10041 
10042 	/* The hardware supported by ixgbe only filters on the destination MAC
10043 	 * address. In order to avoid issues we only support offloading modes
10044 	 * where the hardware can actually provide the functionality.
10045 	 */
10046 	if (!macvlan_supports_dest_filter(vdev))
10047 		return ERR_PTR(-EMEDIUMTYPE);
10048 
10049 	/* We need to lock down the macvlan to be a single queue device so that
10050 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10051 	 * the queue mapping to our netdev.
10052 	 */
10053 	if (netif_is_multiqueue(vdev))
10054 		return ERR_PTR(-ERANGE);
10055 
10056 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10057 	if (pool == adapter->num_rx_pools) {
10058 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10059 		u16 reserved_pools;
10060 
10061 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10062 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10063 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10064 			return ERR_PTR(-EBUSY);
10065 
10066 		/* Hardware has a limited number of available pools. Each VF,
10067 		 * and the PF require a pool. Check to ensure we don't
10068 		 * attempt to use more then the available number of pools.
10069 		 */
10070 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10071 			return ERR_PTR(-EBUSY);
10072 
10073 		/* Enable VMDq flag so device will be set in VM mode */
10074 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10075 				  IXGBE_FLAG_SRIOV_ENABLED;
10076 
10077 		/* Try to reserve as many queues per pool as possible,
10078 		 * we start with the configurations that support 4 queues
10079 		 * per pools, followed by 2, and then by just 1 per pool.
10080 		 */
10081 		if (used_pools < 32 && adapter->num_rx_pools < 16)
10082 			reserved_pools = min_t(u16,
10083 					       32 - used_pools,
10084 					       16 - adapter->num_rx_pools);
10085 		else if (adapter->num_rx_pools < 32)
10086 			reserved_pools = min_t(u16,
10087 					       64 - used_pools,
10088 					       32 - adapter->num_rx_pools);
10089 		else
10090 			reserved_pools = 64 - used_pools;
10091 
10092 
10093 		if (!reserved_pools)
10094 			return ERR_PTR(-EBUSY);
10095 
10096 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10097 
10098 		/* Force reinit of ring allocation with VMDQ enabled */
10099 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10100 		if (err)
10101 			return ERR_PTR(err);
10102 
10103 		if (pool >= adapter->num_rx_pools)
10104 			return ERR_PTR(-ENOMEM);
10105 	}
10106 
10107 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10108 	if (!accel)
10109 		return ERR_PTR(-ENOMEM);
10110 
10111 	set_bit(pool, adapter->fwd_bitmask);
10112 	netdev_set_sb_channel(vdev, pool);
10113 	accel->pool = pool;
10114 	accel->netdev = vdev;
10115 
10116 	if (!netif_running(pdev))
10117 		return accel;
10118 
10119 	err = ixgbe_fwd_ring_up(adapter, accel);
10120 	if (err)
10121 		return ERR_PTR(err);
10122 
10123 	return accel;
10124 }
10125 
10126 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10127 {
10128 	struct ixgbe_fwd_adapter *accel = priv;
10129 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10130 	unsigned int rxbase = accel->rx_base_queue;
10131 	unsigned int i;
10132 
10133 	/* delete unicast filter associated with offloaded interface */
10134 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10135 			     VMDQ_P(accel->pool));
10136 
10137 	/* Allow remaining Rx packets to get flushed out of the
10138 	 * Rx FIFO before we drop the netdev for the ring.
10139 	 */
10140 	usleep_range(10000, 20000);
10141 
10142 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10143 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10144 		struct ixgbe_q_vector *qv = ring->q_vector;
10145 
10146 		/* Make sure we aren't processing any packets and clear
10147 		 * netdev to shut down the ring.
10148 		 */
10149 		if (netif_running(adapter->netdev))
10150 			napi_synchronize(&qv->napi);
10151 		ring->netdev = NULL;
10152 	}
10153 
10154 	/* unbind the queues and drop the subordinate channel config */
10155 	netdev_unbind_sb_channel(pdev, accel->netdev);
10156 	netdev_set_sb_channel(accel->netdev, 0);
10157 
10158 	clear_bit(accel->pool, adapter->fwd_bitmask);
10159 	kfree(accel);
10160 }
10161 
10162 #define IXGBE_MAX_MAC_HDR_LEN		127
10163 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10164 
10165 static netdev_features_t
10166 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10167 		     netdev_features_t features)
10168 {
10169 	unsigned int network_hdr_len, mac_hdr_len;
10170 
10171 	/* Make certain the headers can be described by a context descriptor */
10172 	mac_hdr_len = skb_network_header(skb) - skb->data;
10173 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10174 		return features & ~(NETIF_F_HW_CSUM |
10175 				    NETIF_F_SCTP_CRC |
10176 				    NETIF_F_HW_VLAN_CTAG_TX |
10177 				    NETIF_F_TSO |
10178 				    NETIF_F_TSO6);
10179 
10180 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10181 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10182 		return features & ~(NETIF_F_HW_CSUM |
10183 				    NETIF_F_SCTP_CRC |
10184 				    NETIF_F_TSO |
10185 				    NETIF_F_TSO6);
10186 
10187 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10188 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10189 	 * IPsec offoad sets skb->encapsulation but still can handle
10190 	 * the TSO, so it's the exception.
10191 	 */
10192 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10193 #ifdef CONFIG_IXGBE_IPSEC
10194 		if (!skb->sp)
10195 #endif
10196 			features &= ~NETIF_F_TSO;
10197 	}
10198 
10199 	return features;
10200 }
10201 
10202 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10203 {
10204 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10205 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10206 	struct bpf_prog *old_prog;
10207 
10208 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10209 		return -EINVAL;
10210 
10211 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10212 		return -EINVAL;
10213 
10214 	/* verify ixgbe ring attributes are sufficient for XDP */
10215 	for (i = 0; i < adapter->num_rx_queues; i++) {
10216 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10217 
10218 		if (ring_is_rsc_enabled(ring))
10219 			return -EINVAL;
10220 
10221 		if (frame_size > ixgbe_rx_bufsz(ring))
10222 			return -EINVAL;
10223 	}
10224 
10225 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10226 		return -ENOMEM;
10227 
10228 	old_prog = xchg(&adapter->xdp_prog, prog);
10229 
10230 	/* If transitioning XDP modes reconfigure rings */
10231 	if (!!prog != !!old_prog) {
10232 		int err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10233 
10234 		if (err) {
10235 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10236 			return -EINVAL;
10237 		}
10238 	} else {
10239 		for (i = 0; i < adapter->num_rx_queues; i++)
10240 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10241 			    adapter->xdp_prog);
10242 	}
10243 
10244 	if (old_prog)
10245 		bpf_prog_put(old_prog);
10246 
10247 	return 0;
10248 }
10249 
10250 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10251 {
10252 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10253 
10254 	switch (xdp->command) {
10255 	case XDP_SETUP_PROG:
10256 		return ixgbe_xdp_setup(dev, xdp->prog);
10257 	case XDP_QUERY_PROG:
10258 		xdp->prog_id = adapter->xdp_prog ?
10259 			adapter->xdp_prog->aux->id : 0;
10260 		return 0;
10261 	case XDP_QUERY_XSK_UMEM:
10262 		return ixgbe_xsk_umem_query(adapter, &xdp->xsk.umem,
10263 					    xdp->xsk.queue_id);
10264 	case XDP_SETUP_XSK_UMEM:
10265 		return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10266 					    xdp->xsk.queue_id);
10267 
10268 	default:
10269 		return -EINVAL;
10270 	}
10271 }
10272 
10273 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10274 {
10275 	/* Force memory writes to complete before letting h/w know there
10276 	 * are new descriptors to fetch.
10277 	 */
10278 	wmb();
10279 	writel(ring->next_to_use, ring->tail);
10280 }
10281 
10282 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10283 			  struct xdp_frame **frames, u32 flags)
10284 {
10285 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10286 	struct ixgbe_ring *ring;
10287 	int drops = 0;
10288 	int i;
10289 
10290 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10291 		return -ENETDOWN;
10292 
10293 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10294 		return -EINVAL;
10295 
10296 	/* During program transitions its possible adapter->xdp_prog is assigned
10297 	 * but ring has not been configured yet. In this case simply abort xmit.
10298 	 */
10299 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10300 	if (unlikely(!ring))
10301 		return -ENXIO;
10302 
10303 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10304 		return -ENXIO;
10305 
10306 	for (i = 0; i < n; i++) {
10307 		struct xdp_frame *xdpf = frames[i];
10308 		int err;
10309 
10310 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10311 		if (err != IXGBE_XDP_TX) {
10312 			xdp_return_frame_rx_napi(xdpf);
10313 			drops++;
10314 		}
10315 	}
10316 
10317 	if (unlikely(flags & XDP_XMIT_FLUSH))
10318 		ixgbe_xdp_ring_update_tail(ring);
10319 
10320 	return n - drops;
10321 }
10322 
10323 static const struct net_device_ops ixgbe_netdev_ops = {
10324 	.ndo_open		= ixgbe_open,
10325 	.ndo_stop		= ixgbe_close,
10326 	.ndo_start_xmit		= ixgbe_xmit_frame,
10327 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10328 	.ndo_validate_addr	= eth_validate_addr,
10329 	.ndo_set_mac_address	= ixgbe_set_mac,
10330 	.ndo_change_mtu		= ixgbe_change_mtu,
10331 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10332 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10333 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10334 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10335 	.ndo_do_ioctl		= ixgbe_ioctl,
10336 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10337 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10338 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10339 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10340 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10341 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10342 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10343 	.ndo_get_stats64	= ixgbe_get_stats64,
10344 	.ndo_setup_tc		= __ixgbe_setup_tc,
10345 #ifdef IXGBE_FCOE
10346 	.ndo_select_queue	= ixgbe_select_queue,
10347 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10348 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10349 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10350 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10351 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10352 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10353 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10354 #endif /* IXGBE_FCOE */
10355 	.ndo_set_features = ixgbe_set_features,
10356 	.ndo_fix_features = ixgbe_fix_features,
10357 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10358 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10359 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10360 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10361 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10362 	.ndo_udp_tunnel_add	= ixgbe_add_udp_tunnel_port,
10363 	.ndo_udp_tunnel_del	= ixgbe_del_udp_tunnel_port,
10364 	.ndo_features_check	= ixgbe_features_check,
10365 	.ndo_bpf		= ixgbe_xdp,
10366 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10367 	.ndo_xsk_async_xmit	= ixgbe_xsk_async_xmit,
10368 };
10369 
10370 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10371 				 struct ixgbe_ring *tx_ring)
10372 {
10373 	unsigned long wait_delay, delay_interval;
10374 	struct ixgbe_hw *hw = &adapter->hw;
10375 	u8 reg_idx = tx_ring->reg_idx;
10376 	int wait_loop;
10377 	u32 txdctl;
10378 
10379 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10380 
10381 	/* delay mechanism from ixgbe_disable_tx */
10382 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10383 
10384 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10385 	wait_delay = delay_interval;
10386 
10387 	while (wait_loop--) {
10388 		usleep_range(wait_delay, wait_delay + 10);
10389 		wait_delay += delay_interval * 2;
10390 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10391 
10392 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10393 			return;
10394 	}
10395 
10396 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10397 }
10398 
10399 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10400 			      struct ixgbe_ring *tx_ring)
10401 {
10402 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10403 	ixgbe_disable_txr_hw(adapter, tx_ring);
10404 }
10405 
10406 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10407 				 struct ixgbe_ring *rx_ring)
10408 {
10409 	unsigned long wait_delay, delay_interval;
10410 	struct ixgbe_hw *hw = &adapter->hw;
10411 	u8 reg_idx = rx_ring->reg_idx;
10412 	int wait_loop;
10413 	u32 rxdctl;
10414 
10415 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10416 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10417 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10418 
10419 	/* write value back with RXDCTL.ENABLE bit cleared */
10420 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10421 
10422 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10423 	if (hw->mac.type == ixgbe_mac_82598EB &&
10424 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10425 		return;
10426 
10427 	/* delay mechanism from ixgbe_disable_rx */
10428 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10429 
10430 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10431 	wait_delay = delay_interval;
10432 
10433 	while (wait_loop--) {
10434 		usleep_range(wait_delay, wait_delay + 10);
10435 		wait_delay += delay_interval * 2;
10436 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10437 
10438 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10439 			return;
10440 	}
10441 
10442 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10443 }
10444 
10445 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10446 {
10447 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10448 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10449 }
10450 
10451 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10452 {
10453 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10454 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10455 }
10456 
10457 /**
10458  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10459  * @adapter: adapter structure
10460  * @ring: ring index
10461  *
10462  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10463  * assumes that the netdev is running.
10464  **/
10465 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10466 {
10467 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10468 
10469 	rx_ring = adapter->rx_ring[ring];
10470 	tx_ring = adapter->tx_ring[ring];
10471 	xdp_ring = adapter->xdp_ring[ring];
10472 
10473 	ixgbe_disable_txr(adapter, tx_ring);
10474 	if (xdp_ring)
10475 		ixgbe_disable_txr(adapter, xdp_ring);
10476 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10477 
10478 	if (xdp_ring)
10479 		synchronize_sched();
10480 
10481 	/* Rx/Tx/XDP Tx share the same napi context. */
10482 	napi_disable(&rx_ring->q_vector->napi);
10483 
10484 	ixgbe_clean_tx_ring(tx_ring);
10485 	if (xdp_ring)
10486 		ixgbe_clean_tx_ring(xdp_ring);
10487 	ixgbe_clean_rx_ring(rx_ring);
10488 
10489 	ixgbe_reset_txr_stats(tx_ring);
10490 	if (xdp_ring)
10491 		ixgbe_reset_txr_stats(xdp_ring);
10492 	ixgbe_reset_rxr_stats(rx_ring);
10493 }
10494 
10495 /**
10496  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10497  * @adapter: adapter structure
10498  * @ring: ring index
10499  *
10500  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10501  * assumes that the netdev is running.
10502  **/
10503 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10504 {
10505 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10506 
10507 	rx_ring = adapter->rx_ring[ring];
10508 	tx_ring = adapter->tx_ring[ring];
10509 	xdp_ring = adapter->xdp_ring[ring];
10510 
10511 	/* Rx/Tx/XDP Tx share the same napi context. */
10512 	napi_enable(&rx_ring->q_vector->napi);
10513 
10514 	ixgbe_configure_tx_ring(adapter, tx_ring);
10515 	if (xdp_ring)
10516 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10517 	ixgbe_configure_rx_ring(adapter, rx_ring);
10518 
10519 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10520 	clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10521 }
10522 
10523 /**
10524  * ixgbe_enumerate_functions - Get the number of ports this device has
10525  * @adapter: adapter structure
10526  *
10527  * This function enumerates the phsyical functions co-located on a single slot,
10528  * in order to determine how many ports a device has. This is most useful in
10529  * determining the required GT/s of PCIe bandwidth necessary for optimal
10530  * performance.
10531  **/
10532 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10533 {
10534 	struct pci_dev *entry, *pdev = adapter->pdev;
10535 	int physfns = 0;
10536 
10537 	/* Some cards can not use the generic count PCIe functions method,
10538 	 * because they are behind a parent switch, so we hardcode these with
10539 	 * the correct number of functions.
10540 	 */
10541 	if (ixgbe_pcie_from_parent(&adapter->hw))
10542 		physfns = 4;
10543 
10544 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10545 		/* don't count virtual functions */
10546 		if (entry->is_virtfn)
10547 			continue;
10548 
10549 		/* When the devices on the bus don't all match our device ID,
10550 		 * we can't reliably determine the correct number of
10551 		 * functions. This can occur if a function has been direct
10552 		 * attached to a virtual machine using VT-d, for example. In
10553 		 * this case, simply return -1 to indicate this.
10554 		 */
10555 		if ((entry->vendor != pdev->vendor) ||
10556 		    (entry->device != pdev->device))
10557 			return -1;
10558 
10559 		physfns++;
10560 	}
10561 
10562 	return physfns;
10563 }
10564 
10565 /**
10566  * ixgbe_wol_supported - Check whether device supports WoL
10567  * @adapter: the adapter private structure
10568  * @device_id: the device ID
10569  * @subdevice_id: the subsystem device ID
10570  *
10571  * This function is used by probe and ethtool to determine
10572  * which devices have WoL support
10573  *
10574  **/
10575 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10576 			 u16 subdevice_id)
10577 {
10578 	struct ixgbe_hw *hw = &adapter->hw;
10579 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10580 
10581 	/* WOL not supported on 82598 */
10582 	if (hw->mac.type == ixgbe_mac_82598EB)
10583 		return false;
10584 
10585 	/* check eeprom to see if WOL is enabled for X540 and newer */
10586 	if (hw->mac.type >= ixgbe_mac_X540) {
10587 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10588 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10589 		     (hw->bus.func == 0)))
10590 			return true;
10591 	}
10592 
10593 	/* WOL is determined based on device IDs for 82599 MACs */
10594 	switch (device_id) {
10595 	case IXGBE_DEV_ID_82599_SFP:
10596 		/* Only these subdevices could supports WOL */
10597 		switch (subdevice_id) {
10598 		case IXGBE_SUBDEV_ID_82599_560FLR:
10599 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10600 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10601 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10602 			/* only support first port */
10603 			if (hw->bus.func != 0)
10604 				break;
10605 			/* fall through */
10606 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10607 		case IXGBE_SUBDEV_ID_82599_SFP:
10608 		case IXGBE_SUBDEV_ID_82599_RNDC:
10609 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10610 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10611 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10612 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10613 			return true;
10614 		}
10615 		break;
10616 	case IXGBE_DEV_ID_82599EN_SFP:
10617 		/* Only these subdevices support WOL */
10618 		switch (subdevice_id) {
10619 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10620 			return true;
10621 		}
10622 		break;
10623 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10624 		/* All except this subdevice support WOL */
10625 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10626 			return true;
10627 		break;
10628 	case IXGBE_DEV_ID_82599_KX4:
10629 		return  true;
10630 	default:
10631 		break;
10632 	}
10633 
10634 	return false;
10635 }
10636 
10637 /**
10638  * ixgbe_set_fw_version - Set FW version
10639  * @adapter: the adapter private structure
10640  *
10641  * This function is used by probe and ethtool to determine the FW version to
10642  * format to display. The FW version is taken from the EEPROM/NVM.
10643  */
10644 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10645 {
10646 	struct ixgbe_hw *hw = &adapter->hw;
10647 	struct ixgbe_nvm_version nvm_ver;
10648 
10649 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10650 	if (nvm_ver.oem_valid) {
10651 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10652 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10653 			 nvm_ver.oem_release);
10654 		return;
10655 	}
10656 
10657 	ixgbe_get_etk_id(hw, &nvm_ver);
10658 	ixgbe_get_orom_version(hw, &nvm_ver);
10659 
10660 	if (nvm_ver.or_valid) {
10661 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10662 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10663 			 nvm_ver.or_build, nvm_ver.or_patch);
10664 		return;
10665 	}
10666 
10667 	/* Set ETrack ID format */
10668 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10669 		 "0x%08x", nvm_ver.etk_id);
10670 }
10671 
10672 /**
10673  * ixgbe_probe - Device Initialization Routine
10674  * @pdev: PCI device information struct
10675  * @ent: entry in ixgbe_pci_tbl
10676  *
10677  * Returns 0 on success, negative on failure
10678  *
10679  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10680  * The OS initialization, configuring of the adapter private structure,
10681  * and a hardware reset occur.
10682  **/
10683 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10684 {
10685 	struct net_device *netdev;
10686 	struct ixgbe_adapter *adapter = NULL;
10687 	struct ixgbe_hw *hw;
10688 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10689 	int i, err, pci_using_dac, expected_gts;
10690 	unsigned int indices = MAX_TX_QUEUES;
10691 	u8 part_str[IXGBE_PBANUM_LENGTH];
10692 	bool disable_dev = false;
10693 #ifdef IXGBE_FCOE
10694 	u16 device_caps;
10695 #endif
10696 	u32 eec;
10697 
10698 	/* Catch broken hardware that put the wrong VF device ID in
10699 	 * the PCIe SR-IOV capability.
10700 	 */
10701 	if (pdev->is_virtfn) {
10702 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10703 		     pci_name(pdev), pdev->vendor, pdev->device);
10704 		return -EINVAL;
10705 	}
10706 
10707 	err = pci_enable_device_mem(pdev);
10708 	if (err)
10709 		return err;
10710 
10711 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10712 		pci_using_dac = 1;
10713 	} else {
10714 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10715 		if (err) {
10716 			dev_err(&pdev->dev,
10717 				"No usable DMA configuration, aborting\n");
10718 			goto err_dma;
10719 		}
10720 		pci_using_dac = 0;
10721 	}
10722 
10723 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10724 	if (err) {
10725 		dev_err(&pdev->dev,
10726 			"pci_request_selected_regions failed 0x%x\n", err);
10727 		goto err_pci_reg;
10728 	}
10729 
10730 	pci_enable_pcie_error_reporting(pdev);
10731 
10732 	pci_set_master(pdev);
10733 	pci_save_state(pdev);
10734 
10735 	if (ii->mac == ixgbe_mac_82598EB) {
10736 #ifdef CONFIG_IXGBE_DCB
10737 		/* 8 TC w/ 4 queues per TC */
10738 		indices = 4 * MAX_TRAFFIC_CLASS;
10739 #else
10740 		indices = IXGBE_MAX_RSS_INDICES;
10741 #endif
10742 	}
10743 
10744 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10745 	if (!netdev) {
10746 		err = -ENOMEM;
10747 		goto err_alloc_etherdev;
10748 	}
10749 
10750 	SET_NETDEV_DEV(netdev, &pdev->dev);
10751 
10752 	adapter = netdev_priv(netdev);
10753 
10754 	adapter->netdev = netdev;
10755 	adapter->pdev = pdev;
10756 	hw = &adapter->hw;
10757 	hw->back = adapter;
10758 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10759 
10760 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10761 			      pci_resource_len(pdev, 0));
10762 	adapter->io_addr = hw->hw_addr;
10763 	if (!hw->hw_addr) {
10764 		err = -EIO;
10765 		goto err_ioremap;
10766 	}
10767 
10768 	netdev->netdev_ops = &ixgbe_netdev_ops;
10769 	ixgbe_set_ethtool_ops(netdev);
10770 	netdev->watchdog_timeo = 5 * HZ;
10771 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10772 
10773 	/* Setup hw api */
10774 	hw->mac.ops   = *ii->mac_ops;
10775 	hw->mac.type  = ii->mac;
10776 	hw->mvals     = ii->mvals;
10777 	if (ii->link_ops)
10778 		hw->link.ops  = *ii->link_ops;
10779 
10780 	/* EEPROM */
10781 	hw->eeprom.ops = *ii->eeprom_ops;
10782 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10783 	if (ixgbe_removed(hw->hw_addr)) {
10784 		err = -EIO;
10785 		goto err_ioremap;
10786 	}
10787 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10788 	if (!(eec & BIT(8)))
10789 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10790 
10791 	/* PHY */
10792 	hw->phy.ops = *ii->phy_ops;
10793 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10794 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10795 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10796 	hw->phy.mdio.mmds = 0;
10797 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10798 	hw->phy.mdio.dev = netdev;
10799 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10800 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10801 
10802 	/* setup the private structure */
10803 	err = ixgbe_sw_init(adapter, ii);
10804 	if (err)
10805 		goto err_sw_init;
10806 
10807 	/* Make sure the SWFW semaphore is in a valid state */
10808 	if (hw->mac.ops.init_swfw_sync)
10809 		hw->mac.ops.init_swfw_sync(hw);
10810 
10811 	/* Make it possible the adapter to be woken up via WOL */
10812 	switch (adapter->hw.mac.type) {
10813 	case ixgbe_mac_82599EB:
10814 	case ixgbe_mac_X540:
10815 	case ixgbe_mac_X550:
10816 	case ixgbe_mac_X550EM_x:
10817 	case ixgbe_mac_x550em_a:
10818 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10819 		break;
10820 	default:
10821 		break;
10822 	}
10823 
10824 	/*
10825 	 * If there is a fan on this device and it has failed log the
10826 	 * failure.
10827 	 */
10828 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10829 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10830 		if (esdp & IXGBE_ESDP_SDP1)
10831 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10832 	}
10833 
10834 	if (allow_unsupported_sfp)
10835 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10836 
10837 	/* reset_hw fills in the perm_addr as well */
10838 	hw->phy.reset_if_overtemp = true;
10839 	err = hw->mac.ops.reset_hw(hw);
10840 	hw->phy.reset_if_overtemp = false;
10841 	ixgbe_set_eee_capable(adapter);
10842 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10843 		err = 0;
10844 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10845 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10846 		e_dev_err("Reload the driver after installing a supported module.\n");
10847 		goto err_sw_init;
10848 	} else if (err) {
10849 		e_dev_err("HW Init failed: %d\n", err);
10850 		goto err_sw_init;
10851 	}
10852 
10853 #ifdef CONFIG_PCI_IOV
10854 	/* SR-IOV not supported on the 82598 */
10855 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10856 		goto skip_sriov;
10857 	/* Mailbox */
10858 	ixgbe_init_mbx_params_pf(hw);
10859 	hw->mbx.ops = ii->mbx_ops;
10860 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10861 	ixgbe_enable_sriov(adapter, max_vfs);
10862 skip_sriov:
10863 
10864 #endif
10865 	netdev->features = NETIF_F_SG |
10866 			   NETIF_F_TSO |
10867 			   NETIF_F_TSO6 |
10868 			   NETIF_F_RXHASH |
10869 			   NETIF_F_RXCSUM |
10870 			   NETIF_F_HW_CSUM;
10871 
10872 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10873 				    NETIF_F_GSO_GRE_CSUM | \
10874 				    NETIF_F_GSO_IPXIP4 | \
10875 				    NETIF_F_GSO_IPXIP6 | \
10876 				    NETIF_F_GSO_UDP_TUNNEL | \
10877 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10878 
10879 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10880 	netdev->features |= NETIF_F_GSO_PARTIAL |
10881 			    IXGBE_GSO_PARTIAL_FEATURES;
10882 
10883 	if (hw->mac.type >= ixgbe_mac_82599EB)
10884 		netdev->features |= NETIF_F_SCTP_CRC;
10885 
10886 #ifdef CONFIG_IXGBE_IPSEC
10887 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10888 				 NETIF_F_HW_ESP_TX_CSUM | \
10889 				 NETIF_F_GSO_ESP)
10890 
10891 	if (adapter->ipsec)
10892 		netdev->features |= IXGBE_ESP_FEATURES;
10893 #endif
10894 	/* copy netdev features into list of user selectable features */
10895 	netdev->hw_features |= netdev->features |
10896 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10897 			       NETIF_F_HW_VLAN_CTAG_RX |
10898 			       NETIF_F_HW_VLAN_CTAG_TX |
10899 			       NETIF_F_RXALL |
10900 			       NETIF_F_HW_L2FW_DOFFLOAD;
10901 
10902 	if (hw->mac.type >= ixgbe_mac_82599EB)
10903 		netdev->hw_features |= NETIF_F_NTUPLE |
10904 				       NETIF_F_HW_TC;
10905 
10906 	if (pci_using_dac)
10907 		netdev->features |= NETIF_F_HIGHDMA;
10908 
10909 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10910 	netdev->hw_enc_features |= netdev->vlan_features;
10911 	netdev->mpls_features |= NETIF_F_SG |
10912 				 NETIF_F_TSO |
10913 				 NETIF_F_TSO6 |
10914 				 NETIF_F_HW_CSUM;
10915 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10916 
10917 	/* set this bit last since it cannot be part of vlan_features */
10918 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10919 			    NETIF_F_HW_VLAN_CTAG_RX |
10920 			    NETIF_F_HW_VLAN_CTAG_TX;
10921 
10922 	netdev->priv_flags |= IFF_UNICAST_FLT;
10923 	netdev->priv_flags |= IFF_SUPP_NOFCS;
10924 
10925 	/* MTU range: 68 - 9710 */
10926 	netdev->min_mtu = ETH_MIN_MTU;
10927 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10928 
10929 #ifdef CONFIG_IXGBE_DCB
10930 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10931 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10932 #endif
10933 
10934 #ifdef IXGBE_FCOE
10935 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10936 		unsigned int fcoe_l;
10937 
10938 		if (hw->mac.ops.get_device_caps) {
10939 			hw->mac.ops.get_device_caps(hw, &device_caps);
10940 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10941 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10942 		}
10943 
10944 
10945 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10946 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10947 
10948 		netdev->features |= NETIF_F_FSO |
10949 				    NETIF_F_FCOE_CRC;
10950 
10951 		netdev->vlan_features |= NETIF_F_FSO |
10952 					 NETIF_F_FCOE_CRC |
10953 					 NETIF_F_FCOE_MTU;
10954 	}
10955 #endif /* IXGBE_FCOE */
10956 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10957 		netdev->hw_features |= NETIF_F_LRO;
10958 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10959 		netdev->features |= NETIF_F_LRO;
10960 
10961 	if (ixgbe_check_fw_error(adapter)) {
10962 		err = -EIO;
10963 		goto err_sw_init;
10964 	}
10965 
10966 	/* make sure the EEPROM is good */
10967 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10968 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
10969 		err = -EIO;
10970 		goto err_sw_init;
10971 	}
10972 
10973 	eth_platform_get_mac_address(&adapter->pdev->dev,
10974 				     adapter->hw.mac.perm_addr);
10975 
10976 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10977 
10978 	if (!is_valid_ether_addr(netdev->dev_addr)) {
10979 		e_dev_err("invalid MAC address\n");
10980 		err = -EIO;
10981 		goto err_sw_init;
10982 	}
10983 
10984 	/* Set hw->mac.addr to permanent MAC address */
10985 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
10986 	ixgbe_mac_set_default_filter(adapter);
10987 
10988 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
10989 
10990 	if (ixgbe_removed(hw->hw_addr)) {
10991 		err = -EIO;
10992 		goto err_sw_init;
10993 	}
10994 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
10995 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
10996 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
10997 
10998 	err = ixgbe_init_interrupt_scheme(adapter);
10999 	if (err)
11000 		goto err_sw_init;
11001 
11002 	for (i = 0; i < adapter->num_rx_queues; i++)
11003 		u64_stats_init(&adapter->rx_ring[i]->syncp);
11004 	for (i = 0; i < adapter->num_tx_queues; i++)
11005 		u64_stats_init(&adapter->tx_ring[i]->syncp);
11006 	for (i = 0; i < adapter->num_xdp_queues; i++)
11007 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
11008 
11009 	/* WOL not supported for all devices */
11010 	adapter->wol = 0;
11011 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11012 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11013 						pdev->subsystem_device);
11014 	if (hw->wol_enabled)
11015 		adapter->wol = IXGBE_WUFC_MAG;
11016 
11017 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11018 
11019 	/* save off EEPROM version number */
11020 	ixgbe_set_fw_version(adapter);
11021 
11022 	/* pick up the PCI bus settings for reporting later */
11023 	if (ixgbe_pcie_from_parent(hw))
11024 		ixgbe_get_parent_bus_info(adapter);
11025 	else
11026 		 hw->mac.ops.get_bus_info(hw);
11027 
11028 	/* calculate the expected PCIe bandwidth required for optimal
11029 	 * performance. Note that some older parts will never have enough
11030 	 * bandwidth due to being older generation PCIe parts. We clamp these
11031 	 * parts to ensure no warning is displayed if it can't be fixed.
11032 	 */
11033 	switch (hw->mac.type) {
11034 	case ixgbe_mac_82598EB:
11035 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11036 		break;
11037 	default:
11038 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11039 		break;
11040 	}
11041 
11042 	/* don't check link if we failed to enumerate functions */
11043 	if (expected_gts > 0)
11044 		ixgbe_check_minimum_link(adapter, expected_gts);
11045 
11046 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11047 	if (err)
11048 		strlcpy(part_str, "Unknown", sizeof(part_str));
11049 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11050 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11051 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11052 			   part_str);
11053 	else
11054 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11055 			   hw->mac.type, hw->phy.type, part_str);
11056 
11057 	e_dev_info("%pM\n", netdev->dev_addr);
11058 
11059 	/* reset the hardware with the new settings */
11060 	err = hw->mac.ops.start_hw(hw);
11061 	if (err == IXGBE_ERR_EEPROM_VERSION) {
11062 		/* We are running on a pre-production device, log a warning */
11063 		e_dev_warn("This device is a pre-production adapter/LOM. "
11064 			   "Please be aware there may be issues associated "
11065 			   "with your hardware.  If you are experiencing "
11066 			   "problems please contact your Intel or hardware "
11067 			   "representative who provided you with this "
11068 			   "hardware.\n");
11069 	}
11070 	strcpy(netdev->name, "eth%d");
11071 	pci_set_drvdata(pdev, adapter);
11072 	err = register_netdev(netdev);
11073 	if (err)
11074 		goto err_register;
11075 
11076 
11077 	/* power down the optics for 82599 SFP+ fiber */
11078 	if (hw->mac.ops.disable_tx_laser)
11079 		hw->mac.ops.disable_tx_laser(hw);
11080 
11081 	/* carrier off reporting is important to ethtool even BEFORE open */
11082 	netif_carrier_off(netdev);
11083 
11084 #ifdef CONFIG_IXGBE_DCA
11085 	if (dca_add_requester(&pdev->dev) == 0) {
11086 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11087 		ixgbe_setup_dca(adapter);
11088 	}
11089 #endif
11090 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11091 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11092 		for (i = 0; i < adapter->num_vfs; i++)
11093 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11094 	}
11095 
11096 	/* firmware requires driver version to be 0xFFFFFFFF
11097 	 * since os does not support feature
11098 	 */
11099 	if (hw->mac.ops.set_fw_drv_ver)
11100 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11101 					   sizeof(ixgbe_driver_version) - 1,
11102 					   ixgbe_driver_version);
11103 
11104 	/* add san mac addr to netdev */
11105 	ixgbe_add_sanmac_netdev(netdev);
11106 
11107 	e_dev_info("%s\n", ixgbe_default_device_descr);
11108 
11109 #ifdef CONFIG_IXGBE_HWMON
11110 	if (ixgbe_sysfs_init(adapter))
11111 		e_err(probe, "failed to allocate sysfs resources\n");
11112 #endif /* CONFIG_IXGBE_HWMON */
11113 
11114 	ixgbe_dbg_adapter_init(adapter);
11115 
11116 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11117 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11118 		hw->mac.ops.setup_link(hw,
11119 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11120 			true);
11121 
11122 	return 0;
11123 
11124 err_register:
11125 	ixgbe_release_hw_control(adapter);
11126 	ixgbe_clear_interrupt_scheme(adapter);
11127 err_sw_init:
11128 	ixgbe_disable_sriov(adapter);
11129 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11130 	iounmap(adapter->io_addr);
11131 	kfree(adapter->jump_tables[0]);
11132 	kfree(adapter->mac_table);
11133 	kfree(adapter->rss_key);
11134 err_ioremap:
11135 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11136 	free_netdev(netdev);
11137 err_alloc_etherdev:
11138 	pci_release_mem_regions(pdev);
11139 err_pci_reg:
11140 err_dma:
11141 	if (!adapter || disable_dev)
11142 		pci_disable_device(pdev);
11143 	return err;
11144 }
11145 
11146 /**
11147  * ixgbe_remove - Device Removal Routine
11148  * @pdev: PCI device information struct
11149  *
11150  * ixgbe_remove is called by the PCI subsystem to alert the driver
11151  * that it should release a PCI device.  The could be caused by a
11152  * Hot-Plug event, or because the driver is going to be removed from
11153  * memory.
11154  **/
11155 static void ixgbe_remove(struct pci_dev *pdev)
11156 {
11157 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11158 	struct net_device *netdev;
11159 	bool disable_dev;
11160 	int i;
11161 
11162 	/* if !adapter then we already cleaned up in probe */
11163 	if (!adapter)
11164 		return;
11165 
11166 	netdev  = adapter->netdev;
11167 	ixgbe_dbg_adapter_exit(adapter);
11168 
11169 	set_bit(__IXGBE_REMOVING, &adapter->state);
11170 	cancel_work_sync(&adapter->service_task);
11171 
11172 
11173 #ifdef CONFIG_IXGBE_DCA
11174 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11175 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11176 		dca_remove_requester(&pdev->dev);
11177 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11178 				IXGBE_DCA_CTRL_DCA_DISABLE);
11179 	}
11180 
11181 #endif
11182 #ifdef CONFIG_IXGBE_HWMON
11183 	ixgbe_sysfs_exit(adapter);
11184 #endif /* CONFIG_IXGBE_HWMON */
11185 
11186 	/* remove the added san mac */
11187 	ixgbe_del_sanmac_netdev(netdev);
11188 
11189 #ifdef CONFIG_PCI_IOV
11190 	ixgbe_disable_sriov(adapter);
11191 #endif
11192 	if (netdev->reg_state == NETREG_REGISTERED)
11193 		unregister_netdev(netdev);
11194 
11195 	ixgbe_stop_ipsec_offload(adapter);
11196 	ixgbe_clear_interrupt_scheme(adapter);
11197 
11198 	ixgbe_release_hw_control(adapter);
11199 
11200 #ifdef CONFIG_DCB
11201 	kfree(adapter->ixgbe_ieee_pfc);
11202 	kfree(adapter->ixgbe_ieee_ets);
11203 
11204 #endif
11205 	iounmap(adapter->io_addr);
11206 	pci_release_mem_regions(pdev);
11207 
11208 	e_dev_info("complete\n");
11209 
11210 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11211 		if (adapter->jump_tables[i]) {
11212 			kfree(adapter->jump_tables[i]->input);
11213 			kfree(adapter->jump_tables[i]->mask);
11214 		}
11215 		kfree(adapter->jump_tables[i]);
11216 	}
11217 
11218 	kfree(adapter->mac_table);
11219 	kfree(adapter->rss_key);
11220 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11221 	free_netdev(netdev);
11222 
11223 	pci_disable_pcie_error_reporting(pdev);
11224 
11225 	if (disable_dev)
11226 		pci_disable_device(pdev);
11227 }
11228 
11229 /**
11230  * ixgbe_io_error_detected - called when PCI error is detected
11231  * @pdev: Pointer to PCI device
11232  * @state: The current pci connection state
11233  *
11234  * This function is called after a PCI bus error affecting
11235  * this device has been detected.
11236  */
11237 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11238 						pci_channel_state_t state)
11239 {
11240 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11241 	struct net_device *netdev = adapter->netdev;
11242 
11243 #ifdef CONFIG_PCI_IOV
11244 	struct ixgbe_hw *hw = &adapter->hw;
11245 	struct pci_dev *bdev, *vfdev;
11246 	u32 dw0, dw1, dw2, dw3;
11247 	int vf, pos;
11248 	u16 req_id, pf_func;
11249 
11250 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11251 	    adapter->num_vfs == 0)
11252 		goto skip_bad_vf_detection;
11253 
11254 	bdev = pdev->bus->self;
11255 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11256 		bdev = bdev->bus->self;
11257 
11258 	if (!bdev)
11259 		goto skip_bad_vf_detection;
11260 
11261 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11262 	if (!pos)
11263 		goto skip_bad_vf_detection;
11264 
11265 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11266 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11267 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11268 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11269 	if (ixgbe_removed(hw->hw_addr))
11270 		goto skip_bad_vf_detection;
11271 
11272 	req_id = dw1 >> 16;
11273 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11274 	if (!(req_id & 0x0080))
11275 		goto skip_bad_vf_detection;
11276 
11277 	pf_func = req_id & 0x01;
11278 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11279 		unsigned int device_id;
11280 
11281 		vf = (req_id & 0x7F) >> 1;
11282 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11283 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11284 				"%8.8x\tdw3: %8.8x\n",
11285 		dw0, dw1, dw2, dw3);
11286 		switch (adapter->hw.mac.type) {
11287 		case ixgbe_mac_82599EB:
11288 			device_id = IXGBE_82599_VF_DEVICE_ID;
11289 			break;
11290 		case ixgbe_mac_X540:
11291 			device_id = IXGBE_X540_VF_DEVICE_ID;
11292 			break;
11293 		case ixgbe_mac_X550:
11294 			device_id = IXGBE_DEV_ID_X550_VF;
11295 			break;
11296 		case ixgbe_mac_X550EM_x:
11297 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11298 			break;
11299 		case ixgbe_mac_x550em_a:
11300 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11301 			break;
11302 		default:
11303 			device_id = 0;
11304 			break;
11305 		}
11306 
11307 		/* Find the pci device of the offending VF */
11308 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11309 		while (vfdev) {
11310 			if (vfdev->devfn == (req_id & 0xFF))
11311 				break;
11312 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11313 					       device_id, vfdev);
11314 		}
11315 		/*
11316 		 * There's a slim chance the VF could have been hot plugged,
11317 		 * so if it is no longer present we don't need to issue the
11318 		 * VFLR.  Just clean up the AER in that case.
11319 		 */
11320 		if (vfdev) {
11321 			pcie_flr(vfdev);
11322 			/* Free device reference count */
11323 			pci_dev_put(vfdev);
11324 		}
11325 	}
11326 
11327 	/*
11328 	 * Even though the error may have occurred on the other port
11329 	 * we still need to increment the vf error reference count for
11330 	 * both ports because the I/O resume function will be called
11331 	 * for both of them.
11332 	 */
11333 	adapter->vferr_refcount++;
11334 
11335 	return PCI_ERS_RESULT_RECOVERED;
11336 
11337 skip_bad_vf_detection:
11338 #endif /* CONFIG_PCI_IOV */
11339 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11340 		return PCI_ERS_RESULT_DISCONNECT;
11341 
11342 	if (!netif_device_present(netdev))
11343 		return PCI_ERS_RESULT_DISCONNECT;
11344 
11345 	rtnl_lock();
11346 	netif_device_detach(netdev);
11347 
11348 	if (netif_running(netdev))
11349 		ixgbe_close_suspend(adapter);
11350 
11351 	if (state == pci_channel_io_perm_failure) {
11352 		rtnl_unlock();
11353 		return PCI_ERS_RESULT_DISCONNECT;
11354 	}
11355 
11356 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11357 		pci_disable_device(pdev);
11358 	rtnl_unlock();
11359 
11360 	/* Request a slot reset. */
11361 	return PCI_ERS_RESULT_NEED_RESET;
11362 }
11363 
11364 /**
11365  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11366  * @pdev: Pointer to PCI device
11367  *
11368  * Restart the card from scratch, as if from a cold-boot.
11369  */
11370 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11371 {
11372 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11373 	pci_ers_result_t result;
11374 
11375 	if (pci_enable_device_mem(pdev)) {
11376 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11377 		result = PCI_ERS_RESULT_DISCONNECT;
11378 	} else {
11379 		smp_mb__before_atomic();
11380 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11381 		adapter->hw.hw_addr = adapter->io_addr;
11382 		pci_set_master(pdev);
11383 		pci_restore_state(pdev);
11384 		pci_save_state(pdev);
11385 
11386 		pci_wake_from_d3(pdev, false);
11387 
11388 		ixgbe_reset(adapter);
11389 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11390 		result = PCI_ERS_RESULT_RECOVERED;
11391 	}
11392 
11393 	return result;
11394 }
11395 
11396 /**
11397  * ixgbe_io_resume - called when traffic can start flowing again.
11398  * @pdev: Pointer to PCI device
11399  *
11400  * This callback is called when the error recovery driver tells us that
11401  * its OK to resume normal operation.
11402  */
11403 static void ixgbe_io_resume(struct pci_dev *pdev)
11404 {
11405 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11406 	struct net_device *netdev = adapter->netdev;
11407 
11408 #ifdef CONFIG_PCI_IOV
11409 	if (adapter->vferr_refcount) {
11410 		e_info(drv, "Resuming after VF err\n");
11411 		adapter->vferr_refcount--;
11412 		return;
11413 	}
11414 
11415 #endif
11416 	rtnl_lock();
11417 	if (netif_running(netdev))
11418 		ixgbe_open(netdev);
11419 
11420 	netif_device_attach(netdev);
11421 	rtnl_unlock();
11422 }
11423 
11424 static const struct pci_error_handlers ixgbe_err_handler = {
11425 	.error_detected = ixgbe_io_error_detected,
11426 	.slot_reset = ixgbe_io_slot_reset,
11427 	.resume = ixgbe_io_resume,
11428 };
11429 
11430 static struct pci_driver ixgbe_driver = {
11431 	.name     = ixgbe_driver_name,
11432 	.id_table = ixgbe_pci_tbl,
11433 	.probe    = ixgbe_probe,
11434 	.remove   = ixgbe_remove,
11435 #ifdef CONFIG_PM
11436 	.suspend  = ixgbe_suspend,
11437 	.resume   = ixgbe_resume,
11438 #endif
11439 	.shutdown = ixgbe_shutdown,
11440 	.sriov_configure = ixgbe_pci_sriov_configure,
11441 	.err_handler = &ixgbe_err_handler
11442 };
11443 
11444 /**
11445  * ixgbe_init_module - Driver Registration Routine
11446  *
11447  * ixgbe_init_module is the first routine called when the driver is
11448  * loaded. All it does is register with the PCI subsystem.
11449  **/
11450 static int __init ixgbe_init_module(void)
11451 {
11452 	int ret;
11453 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11454 	pr_info("%s\n", ixgbe_copyright);
11455 
11456 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11457 	if (!ixgbe_wq) {
11458 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11459 		return -ENOMEM;
11460 	}
11461 
11462 	ixgbe_dbg_init();
11463 
11464 	ret = pci_register_driver(&ixgbe_driver);
11465 	if (ret) {
11466 		destroy_workqueue(ixgbe_wq);
11467 		ixgbe_dbg_exit();
11468 		return ret;
11469 	}
11470 
11471 #ifdef CONFIG_IXGBE_DCA
11472 	dca_register_notify(&dca_notifier);
11473 #endif
11474 
11475 	return 0;
11476 }
11477 
11478 module_init(ixgbe_init_module);
11479 
11480 /**
11481  * ixgbe_exit_module - Driver Exit Cleanup Routine
11482  *
11483  * ixgbe_exit_module is called just before the driver is removed
11484  * from memory.
11485  **/
11486 static void __exit ixgbe_exit_module(void)
11487 {
11488 #ifdef CONFIG_IXGBE_DCA
11489 	dca_unregister_notify(&dca_notifier);
11490 #endif
11491 	pci_unregister_driver(&ixgbe_driver);
11492 
11493 	ixgbe_dbg_exit();
11494 	if (ixgbe_wq) {
11495 		destroy_workqueue(ixgbe_wq);
11496 		ixgbe_wq = NULL;
11497 	}
11498 }
11499 
11500 #ifdef CONFIG_IXGBE_DCA
11501 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11502 			    void *p)
11503 {
11504 	int ret_val;
11505 
11506 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11507 					 __ixgbe_notify_dca);
11508 
11509 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11510 }
11511 
11512 #endif /* CONFIG_IXGBE_DCA */
11513 
11514 module_exit(ixgbe_exit_module);
11515 
11516 /* ixgbe_main.c */
11517