xref: /linux/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c (revision 05ee19c18c2bb3dea69e29219017367c4a77e65a)
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 <linux/numa.h>
31 #include <scsi/fc/fc_fcoe.h>
32 #include <net/udp_tunnel.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_act/tc_gact.h>
35 #include <net/tc_act/tc_mirred.h>
36 #include <net/vxlan.h>
37 #include <net/mpls.h>
38 #include <net/xdp_sock.h>
39 #include <net/xfrm.h>
40 
41 #include "ixgbe.h"
42 #include "ixgbe_common.h"
43 #include "ixgbe_dcb_82599.h"
44 #include "ixgbe_phy.h"
45 #include "ixgbe_sriov.h"
46 #include "ixgbe_model.h"
47 #include "ixgbe_txrx_common.h"
48 
49 char ixgbe_driver_name[] = "ixgbe";
50 static const char ixgbe_driver_string[] =
51 			      "Intel(R) 10 Gigabit PCI Express Network Driver";
52 #ifdef IXGBE_FCOE
53 char ixgbe_default_device_descr[] =
54 			      "Intel(R) 10 Gigabit Network Connection";
55 #else
56 static char ixgbe_default_device_descr[] =
57 			      "Intel(R) 10 Gigabit Network Connection";
58 #endif
59 #define DRV_VERSION "5.1.0-k"
60 const char ixgbe_driver_version[] = DRV_VERSION;
61 static const char ixgbe_copyright[] =
62 				"Copyright (c) 1999-2016 Intel Corporation.";
63 
64 static const char ixgbe_overheat_msg[] = "Network adapter has been stopped because it has over heated. Restart the computer. If the problem persists, power off the system and replace the adapter";
65 
66 static const struct ixgbe_info *ixgbe_info_tbl[] = {
67 	[board_82598]		= &ixgbe_82598_info,
68 	[board_82599]		= &ixgbe_82599_info,
69 	[board_X540]		= &ixgbe_X540_info,
70 	[board_X550]		= &ixgbe_X550_info,
71 	[board_X550EM_x]	= &ixgbe_X550EM_x_info,
72 	[board_x550em_x_fw]	= &ixgbe_x550em_x_fw_info,
73 	[board_x550em_a]	= &ixgbe_x550em_a_info,
74 	[board_x550em_a_fw]	= &ixgbe_x550em_a_fw_info,
75 };
76 
77 /* ixgbe_pci_tbl - PCI Device ID Table
78  *
79  * Wildcard entries (PCI_ANY_ID) should come last
80  * Last entry must be all 0s
81  *
82  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
83  *   Class, Class Mask, private data (not used) }
84  */
85 static const struct pci_device_id ixgbe_pci_tbl[] = {
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
94 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
95 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
96 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
97 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
98 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
99 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
100 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
101 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
102 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
103 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
104 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
105 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
106 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
107 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
108 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
109 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
110 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
111 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
112 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
113 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
114 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
115 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
116 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
117 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T1), board_X550},
118 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
119 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_XFI), board_X550EM_x},
120 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
121 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
122 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
123 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_1G_T), board_x550em_x_fw},
124 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR), board_x550em_a },
125 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_KR_L), board_x550em_a },
126 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP_N), board_x550em_a },
127 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII), board_x550em_a },
128 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SGMII_L), board_x550em_a },
129 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_10G_T), board_x550em_a},
130 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_SFP), board_x550em_a },
131 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T), board_x550em_a_fw },
132 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_1G_T_L), board_x550em_a_fw },
133 	/* required last entry */
134 	{0, }
135 };
136 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
137 
138 #ifdef CONFIG_IXGBE_DCA
139 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
140 			    void *p);
141 static struct notifier_block dca_notifier = {
142 	.notifier_call = ixgbe_notify_dca,
143 	.next          = NULL,
144 	.priority      = 0
145 };
146 #endif
147 
148 #ifdef CONFIG_PCI_IOV
149 static unsigned int max_vfs;
150 module_param(max_vfs, uint, 0);
151 MODULE_PARM_DESC(max_vfs,
152 		 "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
153 #endif /* CONFIG_PCI_IOV */
154 
155 static unsigned int allow_unsupported_sfp;
156 module_param(allow_unsupported_sfp, uint, 0);
157 MODULE_PARM_DESC(allow_unsupported_sfp,
158 		 "Allow unsupported and untested SFP+ modules on 82599-based adapters");
159 
160 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
161 static int debug = -1;
162 module_param(debug, int, 0);
163 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
164 
165 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
166 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
167 MODULE_LICENSE("GPL v2");
168 MODULE_VERSION(DRV_VERSION);
169 
170 static struct workqueue_struct *ixgbe_wq;
171 
172 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
173 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
174 
175 static const struct net_device_ops ixgbe_netdev_ops;
176 
177 static bool netif_is_ixgbe(struct net_device *dev)
178 {
179 	return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
180 }
181 
182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
183 					  u32 reg, u16 *value)
184 {
185 	struct pci_dev *parent_dev;
186 	struct pci_bus *parent_bus;
187 
188 	parent_bus = adapter->pdev->bus->parent;
189 	if (!parent_bus)
190 		return -1;
191 
192 	parent_dev = parent_bus->self;
193 	if (!parent_dev)
194 		return -1;
195 
196 	if (!pci_is_pcie(parent_dev))
197 		return -1;
198 
199 	pcie_capability_read_word(parent_dev, reg, value);
200 	if (*value == IXGBE_FAILED_READ_CFG_WORD &&
201 	    ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
202 		return -1;
203 	return 0;
204 }
205 
206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
207 {
208 	struct ixgbe_hw *hw = &adapter->hw;
209 	u16 link_status = 0;
210 	int err;
211 
212 	hw->bus.type = ixgbe_bus_type_pci_express;
213 
214 	/* Get the negotiated link width and speed from PCI config space of the
215 	 * parent, as this device is behind a switch
216 	 */
217 	err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
218 
219 	/* assume caller will handle error case */
220 	if (err)
221 		return err;
222 
223 	hw->bus.width = ixgbe_convert_bus_width(link_status);
224 	hw->bus.speed = ixgbe_convert_bus_speed(link_status);
225 
226 	return 0;
227 }
228 
229 /**
230  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
231  * @hw: hw specific details
232  *
233  * This function is used by probe to determine whether a device's PCI-Express
234  * bandwidth details should be gathered from the parent bus instead of from the
235  * device. Used to ensure that various locations all have the correct device ID
236  * checks.
237  */
238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
239 {
240 	switch (hw->device_id) {
241 	case IXGBE_DEV_ID_82599_SFP_SF_QP:
242 	case IXGBE_DEV_ID_82599_QSFP_SF_QP:
243 		return true;
244 	default:
245 		return false;
246 	}
247 }
248 
249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
250 				     int expected_gts)
251 {
252 	struct ixgbe_hw *hw = &adapter->hw;
253 	struct pci_dev *pdev;
254 
255 	/* Some devices are not connected over PCIe and thus do not negotiate
256 	 * speed. These devices do not have valid bus info, and thus any report
257 	 * we generate may not be correct.
258 	 */
259 	if (hw->bus.type == ixgbe_bus_type_internal)
260 		return;
261 
262 	/* determine whether to use the parent device */
263 	if (ixgbe_pcie_from_parent(&adapter->hw))
264 		pdev = adapter->pdev->bus->parent->self;
265 	else
266 		pdev = adapter->pdev;
267 
268 	pcie_print_link_status(pdev);
269 }
270 
271 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
272 {
273 	if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
274 	    !test_bit(__IXGBE_REMOVING, &adapter->state) &&
275 	    !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
276 		queue_work(ixgbe_wq, &adapter->service_task);
277 }
278 
279 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
280 {
281 	struct ixgbe_adapter *adapter = hw->back;
282 
283 	if (!hw->hw_addr)
284 		return;
285 	hw->hw_addr = NULL;
286 	e_dev_err("Adapter removed\n");
287 	if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
288 		ixgbe_service_event_schedule(adapter);
289 }
290 
291 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
292 {
293 	u8 __iomem *reg_addr;
294 	u32 value;
295 	int i;
296 
297 	reg_addr = READ_ONCE(hw->hw_addr);
298 	if (ixgbe_removed(reg_addr))
299 		return IXGBE_FAILED_READ_REG;
300 
301 	/* Register read of 0xFFFFFFF can indicate the adapter has been removed,
302 	 * so perform several status register reads to determine if the adapter
303 	 * has been removed.
304 	 */
305 	for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
306 		value = readl(reg_addr + IXGBE_STATUS);
307 		if (value != IXGBE_FAILED_READ_REG)
308 			break;
309 		mdelay(3);
310 	}
311 
312 	if (value == IXGBE_FAILED_READ_REG)
313 		ixgbe_remove_adapter(hw);
314 	else
315 		value = readl(reg_addr + reg);
316 	return value;
317 }
318 
319 /**
320  * ixgbe_read_reg - Read from device register
321  * @hw: hw specific details
322  * @reg: offset of register to read
323  *
324  * Returns : value read or IXGBE_FAILED_READ_REG if removed
325  *
326  * This function is used to read device registers. It checks for device
327  * removal by confirming any read that returns all ones by checking the
328  * status register value for all ones. This function avoids reading from
329  * the hardware if a removal was previously detected in which case it
330  * returns IXGBE_FAILED_READ_REG (all ones).
331  */
332 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
333 {
334 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
335 	u32 value;
336 
337 	if (ixgbe_removed(reg_addr))
338 		return IXGBE_FAILED_READ_REG;
339 	if (unlikely(hw->phy.nw_mng_if_sel &
340 		     IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
341 		struct ixgbe_adapter *adapter;
342 		int i;
343 
344 		for (i = 0; i < 200; ++i) {
345 			value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
346 			if (likely(!value))
347 				goto writes_completed;
348 			if (value == IXGBE_FAILED_READ_REG) {
349 				ixgbe_remove_adapter(hw);
350 				return IXGBE_FAILED_READ_REG;
351 			}
352 			udelay(5);
353 		}
354 
355 		adapter = hw->back;
356 		e_warn(hw, "register writes incomplete %08x\n", value);
357 	}
358 
359 writes_completed:
360 	value = readl(reg_addr + reg);
361 	if (unlikely(value == IXGBE_FAILED_READ_REG))
362 		value = ixgbe_check_remove(hw, reg);
363 	return value;
364 }
365 
366 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
367 {
368 	u16 value;
369 
370 	pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
371 	if (value == IXGBE_FAILED_READ_CFG_WORD) {
372 		ixgbe_remove_adapter(hw);
373 		return true;
374 	}
375 	return false;
376 }
377 
378 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
379 {
380 	struct ixgbe_adapter *adapter = hw->back;
381 	u16 value;
382 
383 	if (ixgbe_removed(hw->hw_addr))
384 		return IXGBE_FAILED_READ_CFG_WORD;
385 	pci_read_config_word(adapter->pdev, reg, &value);
386 	if (value == IXGBE_FAILED_READ_CFG_WORD &&
387 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
388 		return IXGBE_FAILED_READ_CFG_WORD;
389 	return value;
390 }
391 
392 #ifdef CONFIG_PCI_IOV
393 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
394 {
395 	struct ixgbe_adapter *adapter = hw->back;
396 	u32 value;
397 
398 	if (ixgbe_removed(hw->hw_addr))
399 		return IXGBE_FAILED_READ_CFG_DWORD;
400 	pci_read_config_dword(adapter->pdev, reg, &value);
401 	if (value == IXGBE_FAILED_READ_CFG_DWORD &&
402 	    ixgbe_check_cfg_remove(hw, adapter->pdev))
403 		return IXGBE_FAILED_READ_CFG_DWORD;
404 	return value;
405 }
406 #endif /* CONFIG_PCI_IOV */
407 
408 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
409 {
410 	struct ixgbe_adapter *adapter = hw->back;
411 
412 	if (ixgbe_removed(hw->hw_addr))
413 		return;
414 	pci_write_config_word(adapter->pdev, reg, value);
415 }
416 
417 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
418 {
419 	BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
420 
421 	/* flush memory to make sure state is correct before next watchdog */
422 	smp_mb__before_atomic();
423 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
424 }
425 
426 struct ixgbe_reg_info {
427 	u32 ofs;
428 	char *name;
429 };
430 
431 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
432 
433 	/* General Registers */
434 	{IXGBE_CTRL, "CTRL"},
435 	{IXGBE_STATUS, "STATUS"},
436 	{IXGBE_CTRL_EXT, "CTRL_EXT"},
437 
438 	/* Interrupt Registers */
439 	{IXGBE_EICR, "EICR"},
440 
441 	/* RX Registers */
442 	{IXGBE_SRRCTL(0), "SRRCTL"},
443 	{IXGBE_DCA_RXCTRL(0), "DRXCTL"},
444 	{IXGBE_RDLEN(0), "RDLEN"},
445 	{IXGBE_RDH(0), "RDH"},
446 	{IXGBE_RDT(0), "RDT"},
447 	{IXGBE_RXDCTL(0), "RXDCTL"},
448 	{IXGBE_RDBAL(0), "RDBAL"},
449 	{IXGBE_RDBAH(0), "RDBAH"},
450 
451 	/* TX Registers */
452 	{IXGBE_TDBAL(0), "TDBAL"},
453 	{IXGBE_TDBAH(0), "TDBAH"},
454 	{IXGBE_TDLEN(0), "TDLEN"},
455 	{IXGBE_TDH(0), "TDH"},
456 	{IXGBE_TDT(0), "TDT"},
457 	{IXGBE_TXDCTL(0), "TXDCTL"},
458 
459 	/* List Terminator */
460 	{ .name = NULL }
461 };
462 
463 
464 /*
465  * ixgbe_regdump - register printout routine
466  */
467 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
468 {
469 	int i;
470 	char rname[16];
471 	u32 regs[64];
472 
473 	switch (reginfo->ofs) {
474 	case IXGBE_SRRCTL(0):
475 		for (i = 0; i < 64; i++)
476 			regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
477 		break;
478 	case IXGBE_DCA_RXCTRL(0):
479 		for (i = 0; i < 64; i++)
480 			regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
481 		break;
482 	case IXGBE_RDLEN(0):
483 		for (i = 0; i < 64; i++)
484 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
485 		break;
486 	case IXGBE_RDH(0):
487 		for (i = 0; i < 64; i++)
488 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
489 		break;
490 	case IXGBE_RDT(0):
491 		for (i = 0; i < 64; i++)
492 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
493 		break;
494 	case IXGBE_RXDCTL(0):
495 		for (i = 0; i < 64; i++)
496 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
497 		break;
498 	case IXGBE_RDBAL(0):
499 		for (i = 0; i < 64; i++)
500 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
501 		break;
502 	case IXGBE_RDBAH(0):
503 		for (i = 0; i < 64; i++)
504 			regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
505 		break;
506 	case IXGBE_TDBAL(0):
507 		for (i = 0; i < 64; i++)
508 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
509 		break;
510 	case IXGBE_TDBAH(0):
511 		for (i = 0; i < 64; i++)
512 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
513 		break;
514 	case IXGBE_TDLEN(0):
515 		for (i = 0; i < 64; i++)
516 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
517 		break;
518 	case IXGBE_TDH(0):
519 		for (i = 0; i < 64; i++)
520 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
521 		break;
522 	case IXGBE_TDT(0):
523 		for (i = 0; i < 64; i++)
524 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
525 		break;
526 	case IXGBE_TXDCTL(0):
527 		for (i = 0; i < 64; i++)
528 			regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
529 		break;
530 	default:
531 		pr_info("%-15s %08x\n",
532 			reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
533 		return;
534 	}
535 
536 	i = 0;
537 	while (i < 64) {
538 		int j;
539 		char buf[9 * 8 + 1];
540 		char *p = buf;
541 
542 		snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
543 		for (j = 0; j < 8; j++)
544 			p += sprintf(p, " %08x", regs[i++]);
545 		pr_err("%-15s%s\n", rname, buf);
546 	}
547 
548 }
549 
550 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
551 {
552 	struct ixgbe_tx_buffer *tx_buffer;
553 
554 	tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
555 	pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
556 		n, ring->next_to_use, ring->next_to_clean,
557 		(u64)dma_unmap_addr(tx_buffer, dma),
558 		dma_unmap_len(tx_buffer, len),
559 		tx_buffer->next_to_watch,
560 		(u64)tx_buffer->time_stamp);
561 }
562 
563 /*
564  * ixgbe_dump - Print registers, tx-rings and rx-rings
565  */
566 static void ixgbe_dump(struct ixgbe_adapter *adapter)
567 {
568 	struct net_device *netdev = adapter->netdev;
569 	struct ixgbe_hw *hw = &adapter->hw;
570 	struct ixgbe_reg_info *reginfo;
571 	int n = 0;
572 	struct ixgbe_ring *ring;
573 	struct ixgbe_tx_buffer *tx_buffer;
574 	union ixgbe_adv_tx_desc *tx_desc;
575 	struct my_u0 { u64 a; u64 b; } *u0;
576 	struct ixgbe_ring *rx_ring;
577 	union ixgbe_adv_rx_desc *rx_desc;
578 	struct ixgbe_rx_buffer *rx_buffer_info;
579 	int i = 0;
580 
581 	if (!netif_msg_hw(adapter))
582 		return;
583 
584 	/* Print netdevice Info */
585 	if (netdev) {
586 		dev_info(&adapter->pdev->dev, "Net device Info\n");
587 		pr_info("Device Name     state            "
588 			"trans_start\n");
589 		pr_info("%-15s %016lX %016lX\n",
590 			netdev->name,
591 			netdev->state,
592 			dev_trans_start(netdev));
593 	}
594 
595 	/* Print Registers */
596 	dev_info(&adapter->pdev->dev, "Register Dump\n");
597 	pr_info(" Register Name   Value\n");
598 	for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
599 	     reginfo->name; reginfo++) {
600 		ixgbe_regdump(hw, reginfo);
601 	}
602 
603 	/* Print TX Ring Summary */
604 	if (!netdev || !netif_running(netdev))
605 		return;
606 
607 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
608 	pr_info(" %s     %s              %s        %s\n",
609 		"Queue [NTU] [NTC] [bi(ntc)->dma  ]",
610 		"leng", "ntw", "timestamp");
611 	for (n = 0; n < adapter->num_tx_queues; n++) {
612 		ring = adapter->tx_ring[n];
613 		ixgbe_print_buffer(ring, n);
614 	}
615 
616 	for (n = 0; n < adapter->num_xdp_queues; n++) {
617 		ring = adapter->xdp_ring[n];
618 		ixgbe_print_buffer(ring, n);
619 	}
620 
621 	/* Print TX Rings */
622 	if (!netif_msg_tx_done(adapter))
623 		goto rx_ring_summary;
624 
625 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
626 
627 	/* Transmit Descriptor Formats
628 	 *
629 	 * 82598 Advanced Transmit Descriptor
630 	 *   +--------------------------------------------------------------+
631 	 * 0 |         Buffer Address [63:0]                                |
632 	 *   +--------------------------------------------------------------+
633 	 * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
634 	 *   +--------------------------------------------------------------+
635 	 *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
636 	 *
637 	 * 82598 Advanced Transmit Descriptor (Write-Back Format)
638 	 *   +--------------------------------------------------------------+
639 	 * 0 |                          RSV [63:0]                          |
640 	 *   +--------------------------------------------------------------+
641 	 * 8 |            RSV           |  STA  |          NXTSEQ           |
642 	 *   +--------------------------------------------------------------+
643 	 *   63                       36 35   32 31                         0
644 	 *
645 	 * 82599+ Advanced Transmit Descriptor
646 	 *   +--------------------------------------------------------------+
647 	 * 0 |         Buffer Address [63:0]                                |
648 	 *   +--------------------------------------------------------------+
649 	 * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
650 	 *   +--------------------------------------------------------------+
651 	 *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
652 	 *
653 	 * 82599+ Advanced Transmit Descriptor (Write-Back Format)
654 	 *   +--------------------------------------------------------------+
655 	 * 0 |                          RSV [63:0]                          |
656 	 *   +--------------------------------------------------------------+
657 	 * 8 |            RSV           |  STA  |           RSV             |
658 	 *   +--------------------------------------------------------------+
659 	 *   63                       36 35   32 31                         0
660 	 */
661 
662 	for (n = 0; n < adapter->num_tx_queues; n++) {
663 		ring = adapter->tx_ring[n];
664 		pr_info("------------------------------------\n");
665 		pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
666 		pr_info("------------------------------------\n");
667 		pr_info("%s%s    %s              %s        %s          %s\n",
668 			"T [desc]     [address 63:0  ] ",
669 			"[PlPOIdStDDt Ln] [bi->dma       ] ",
670 			"leng", "ntw", "timestamp", "bi->skb");
671 
672 		for (i = 0; ring->desc && (i < ring->count); i++) {
673 			tx_desc = IXGBE_TX_DESC(ring, i);
674 			tx_buffer = &ring->tx_buffer_info[i];
675 			u0 = (struct my_u0 *)tx_desc;
676 			if (dma_unmap_len(tx_buffer, len) > 0) {
677 				const char *ring_desc;
678 
679 				if (i == ring->next_to_use &&
680 				    i == ring->next_to_clean)
681 					ring_desc = " NTC/U";
682 				else if (i == ring->next_to_use)
683 					ring_desc = " NTU";
684 				else if (i == ring->next_to_clean)
685 					ring_desc = " NTC";
686 				else
687 					ring_desc = "";
688 				pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
689 					i,
690 					le64_to_cpu((__force __le64)u0->a),
691 					le64_to_cpu((__force __le64)u0->b),
692 					(u64)dma_unmap_addr(tx_buffer, dma),
693 					dma_unmap_len(tx_buffer, len),
694 					tx_buffer->next_to_watch,
695 					(u64)tx_buffer->time_stamp,
696 					tx_buffer->skb,
697 					ring_desc);
698 
699 				if (netif_msg_pktdata(adapter) &&
700 				    tx_buffer->skb)
701 					print_hex_dump(KERN_INFO, "",
702 						DUMP_PREFIX_ADDRESS, 16, 1,
703 						tx_buffer->skb->data,
704 						dma_unmap_len(tx_buffer, len),
705 						true);
706 			}
707 		}
708 	}
709 
710 	/* Print RX Rings Summary */
711 rx_ring_summary:
712 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
713 	pr_info("Queue [NTU] [NTC]\n");
714 	for (n = 0; n < adapter->num_rx_queues; n++) {
715 		rx_ring = adapter->rx_ring[n];
716 		pr_info("%5d %5X %5X\n",
717 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
718 	}
719 
720 	/* Print RX Rings */
721 	if (!netif_msg_rx_status(adapter))
722 		return;
723 
724 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
725 
726 	/* Receive Descriptor Formats
727 	 *
728 	 * 82598 Advanced Receive Descriptor (Read) Format
729 	 *    63                                           1        0
730 	 *    +-----------------------------------------------------+
731 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
732 	 *    +----------------------------------------------+------+
733 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
734 	 *    +-----------------------------------------------------+
735 	 *
736 	 *
737 	 * 82598 Advanced Receive Descriptor (Write-Back) Format
738 	 *
739 	 *   63       48 47    32 31  30      21 20 16 15   4 3     0
740 	 *   +------------------------------------------------------+
741 	 * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
742 	 *   | Packet   | IP     |   |          |     | Type | Type |
743 	 *   | Checksum | Ident  |   |          |     |      |      |
744 	 *   +------------------------------------------------------+
745 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
746 	 *   +------------------------------------------------------+
747 	 *   63       48 47    32 31            20 19               0
748 	 *
749 	 * 82599+ Advanced Receive Descriptor (Read) Format
750 	 *    63                                           1        0
751 	 *    +-----------------------------------------------------+
752 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
753 	 *    +----------------------------------------------+------+
754 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
755 	 *    +-----------------------------------------------------+
756 	 *
757 	 *
758 	 * 82599+ Advanced Receive Descriptor (Write-Back) Format
759 	 *
760 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
761 	 *   +------------------------------------------------------+
762 	 * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
763 	 *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
764 	 *   |/ Flow Dir Flt ID  |   |          |     |      |      |
765 	 *   +------------------------------------------------------+
766 	 * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
767 	 *   +------------------------------------------------------+
768 	 *   63       48 47    32 31          20 19                 0
769 	 */
770 
771 	for (n = 0; n < adapter->num_rx_queues; n++) {
772 		rx_ring = adapter->rx_ring[n];
773 		pr_info("------------------------------------\n");
774 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
775 		pr_info("------------------------------------\n");
776 		pr_info("%s%s%s\n",
777 			"R  [desc]      [ PktBuf     A0] ",
778 			"[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
779 			"<-- Adv Rx Read format");
780 		pr_info("%s%s%s\n",
781 			"RWB[desc]      [PcsmIpSHl PtRs] ",
782 			"[vl er S cks ln] ---------------- [bi->skb       ] ",
783 			"<-- Adv Rx Write-Back format");
784 
785 		for (i = 0; i < rx_ring->count; i++) {
786 			const char *ring_desc;
787 
788 			if (i == rx_ring->next_to_use)
789 				ring_desc = " NTU";
790 			else if (i == rx_ring->next_to_clean)
791 				ring_desc = " NTC";
792 			else
793 				ring_desc = "";
794 
795 			rx_buffer_info = &rx_ring->rx_buffer_info[i];
796 			rx_desc = IXGBE_RX_DESC(rx_ring, i);
797 			u0 = (struct my_u0 *)rx_desc;
798 			if (rx_desc->wb.upper.length) {
799 				/* Descriptor Done */
800 				pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
801 					i,
802 					le64_to_cpu((__force __le64)u0->a),
803 					le64_to_cpu((__force __le64)u0->b),
804 					rx_buffer_info->skb,
805 					ring_desc);
806 			} else {
807 				pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
808 					i,
809 					le64_to_cpu((__force __le64)u0->a),
810 					le64_to_cpu((__force __le64)u0->b),
811 					(u64)rx_buffer_info->dma,
812 					rx_buffer_info->skb,
813 					ring_desc);
814 
815 				if (netif_msg_pktdata(adapter) &&
816 				    rx_buffer_info->dma) {
817 					print_hex_dump(KERN_INFO, "",
818 					   DUMP_PREFIX_ADDRESS, 16, 1,
819 					   page_address(rx_buffer_info->page) +
820 						    rx_buffer_info->page_offset,
821 					   ixgbe_rx_bufsz(rx_ring), true);
822 				}
823 			}
824 		}
825 	}
826 }
827 
828 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
829 {
830 	u32 ctrl_ext;
831 
832 	/* Let firmware take over control of h/w */
833 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
834 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
835 			ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
836 }
837 
838 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
839 {
840 	u32 ctrl_ext;
841 
842 	/* Let firmware know the driver has taken over */
843 	ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
844 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
845 			ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
846 }
847 
848 /**
849  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
850  * @adapter: pointer to adapter struct
851  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
852  * @queue: queue to map the corresponding interrupt to
853  * @msix_vector: the vector to map to the corresponding queue
854  *
855  */
856 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
857 			   u8 queue, u8 msix_vector)
858 {
859 	u32 ivar, index;
860 	struct ixgbe_hw *hw = &adapter->hw;
861 	switch (hw->mac.type) {
862 	case ixgbe_mac_82598EB:
863 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
864 		if (direction == -1)
865 			direction = 0;
866 		index = (((direction * 64) + queue) >> 2) & 0x1F;
867 		ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
868 		ivar &= ~(0xFF << (8 * (queue & 0x3)));
869 		ivar |= (msix_vector << (8 * (queue & 0x3)));
870 		IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
871 		break;
872 	case ixgbe_mac_82599EB:
873 	case ixgbe_mac_X540:
874 	case ixgbe_mac_X550:
875 	case ixgbe_mac_X550EM_x:
876 	case ixgbe_mac_x550em_a:
877 		if (direction == -1) {
878 			/* other causes */
879 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
880 			index = ((queue & 1) * 8);
881 			ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
882 			ivar &= ~(0xFF << index);
883 			ivar |= (msix_vector << index);
884 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
885 			break;
886 		} else {
887 			/* tx or rx causes */
888 			msix_vector |= IXGBE_IVAR_ALLOC_VAL;
889 			index = ((16 * (queue & 1)) + (8 * direction));
890 			ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
891 			ivar &= ~(0xFF << index);
892 			ivar |= (msix_vector << index);
893 			IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
894 			break;
895 		}
896 	default:
897 		break;
898 	}
899 }
900 
901 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
902 			    u64 qmask)
903 {
904 	u32 mask;
905 
906 	switch (adapter->hw.mac.type) {
907 	case ixgbe_mac_82598EB:
908 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
909 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
910 		break;
911 	case ixgbe_mac_82599EB:
912 	case ixgbe_mac_X540:
913 	case ixgbe_mac_X550:
914 	case ixgbe_mac_X550EM_x:
915 	case ixgbe_mac_x550em_a:
916 		mask = (qmask & 0xFFFFFFFF);
917 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
918 		mask = (qmask >> 32);
919 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
920 		break;
921 	default:
922 		break;
923 	}
924 }
925 
926 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
927 {
928 	struct ixgbe_hw *hw = &adapter->hw;
929 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
930 	int i;
931 	u32 data;
932 
933 	if ((hw->fc.current_mode != ixgbe_fc_full) &&
934 	    (hw->fc.current_mode != ixgbe_fc_rx_pause))
935 		return;
936 
937 	switch (hw->mac.type) {
938 	case ixgbe_mac_82598EB:
939 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
940 		break;
941 	default:
942 		data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
943 	}
944 	hwstats->lxoffrxc += data;
945 
946 	/* refill credits (no tx hang) if we received xoff */
947 	if (!data)
948 		return;
949 
950 	for (i = 0; i < adapter->num_tx_queues; i++)
951 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
952 			  &adapter->tx_ring[i]->state);
953 
954 	for (i = 0; i < adapter->num_xdp_queues; i++)
955 		clear_bit(__IXGBE_HANG_CHECK_ARMED,
956 			  &adapter->xdp_ring[i]->state);
957 }
958 
959 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
960 {
961 	struct ixgbe_hw *hw = &adapter->hw;
962 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
963 	u32 xoff[8] = {0};
964 	u8 tc;
965 	int i;
966 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
967 
968 	if (adapter->ixgbe_ieee_pfc)
969 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
970 
971 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
972 		ixgbe_update_xoff_rx_lfc(adapter);
973 		return;
974 	}
975 
976 	/* update stats for each tc, only valid with PFC enabled */
977 	for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
978 		u32 pxoffrxc;
979 
980 		switch (hw->mac.type) {
981 		case ixgbe_mac_82598EB:
982 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
983 			break;
984 		default:
985 			pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
986 		}
987 		hwstats->pxoffrxc[i] += pxoffrxc;
988 		/* Get the TC for given UP */
989 		tc = netdev_get_prio_tc_map(adapter->netdev, i);
990 		xoff[tc] += pxoffrxc;
991 	}
992 
993 	/* disarm tx queues that have received xoff frames */
994 	for (i = 0; i < adapter->num_tx_queues; i++) {
995 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
996 
997 		tc = tx_ring->dcb_tc;
998 		if (xoff[tc])
999 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1000 	}
1001 
1002 	for (i = 0; i < adapter->num_xdp_queues; i++) {
1003 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1004 
1005 		tc = xdp_ring->dcb_tc;
1006 		if (xoff[tc])
1007 			clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1008 	}
1009 }
1010 
1011 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1012 {
1013 	return ring->stats.packets;
1014 }
1015 
1016 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1017 {
1018 	unsigned int head, tail;
1019 
1020 	head = ring->next_to_clean;
1021 	tail = ring->next_to_use;
1022 
1023 	return ((head <= tail) ? tail : tail + ring->count) - head;
1024 }
1025 
1026 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1027 {
1028 	u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1029 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1030 	u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1031 
1032 	clear_check_for_tx_hang(tx_ring);
1033 
1034 	/*
1035 	 * Check for a hung queue, but be thorough. This verifies
1036 	 * that a transmit has been completed since the previous
1037 	 * check AND there is at least one packet pending. The
1038 	 * ARMED bit is set to indicate a potential hang. The
1039 	 * bit is cleared if a pause frame is received to remove
1040 	 * false hang detection due to PFC or 802.3x frames. By
1041 	 * requiring this to fail twice we avoid races with
1042 	 * pfc clearing the ARMED bit and conditions where we
1043 	 * run the check_tx_hang logic with a transmit completion
1044 	 * pending but without time to complete it yet.
1045 	 */
1046 	if (tx_done_old == tx_done && tx_pending)
1047 		/* make sure it is true for two checks in a row */
1048 		return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1049 					&tx_ring->state);
1050 	/* update completed stats and continue */
1051 	tx_ring->tx_stats.tx_done_old = tx_done;
1052 	/* reset the countdown */
1053 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1054 
1055 	return false;
1056 }
1057 
1058 /**
1059  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1060  * @adapter: driver private struct
1061  **/
1062 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1063 {
1064 
1065 	/* Do the reset outside of interrupt context */
1066 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1067 		set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1068 		e_warn(drv, "initiating reset due to tx timeout\n");
1069 		ixgbe_service_event_schedule(adapter);
1070 	}
1071 }
1072 
1073 /**
1074  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1075  * @netdev: network interface device structure
1076  * @queue_index: Tx queue to set
1077  * @maxrate: desired maximum transmit bitrate
1078  **/
1079 static int ixgbe_tx_maxrate(struct net_device *netdev,
1080 			    int queue_index, u32 maxrate)
1081 {
1082 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
1083 	struct ixgbe_hw *hw = &adapter->hw;
1084 	u32 bcnrc_val = ixgbe_link_mbps(adapter);
1085 
1086 	if (!maxrate)
1087 		return 0;
1088 
1089 	/* Calculate the rate factor values to set */
1090 	bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1091 	bcnrc_val /= maxrate;
1092 
1093 	/* clear everything but the rate factor */
1094 	bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1095 	IXGBE_RTTBCNRC_RF_DEC_MASK;
1096 
1097 	/* enable the rate scheduler */
1098 	bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1099 
1100 	IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1101 	IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1102 
1103 	return 0;
1104 }
1105 
1106 /**
1107  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1108  * @q_vector: structure containing interrupt and ring information
1109  * @tx_ring: tx ring to clean
1110  * @napi_budget: Used to determine if we are in netpoll
1111  **/
1112 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1113 			       struct ixgbe_ring *tx_ring, int napi_budget)
1114 {
1115 	struct ixgbe_adapter *adapter = q_vector->adapter;
1116 	struct ixgbe_tx_buffer *tx_buffer;
1117 	union ixgbe_adv_tx_desc *tx_desc;
1118 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1119 	unsigned int budget = q_vector->tx.work_limit;
1120 	unsigned int i = tx_ring->next_to_clean;
1121 
1122 	if (test_bit(__IXGBE_DOWN, &adapter->state))
1123 		return true;
1124 
1125 	tx_buffer = &tx_ring->tx_buffer_info[i];
1126 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
1127 	i -= tx_ring->count;
1128 
1129 	do {
1130 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1131 
1132 		/* if next_to_watch is not set then there is no work pending */
1133 		if (!eop_desc)
1134 			break;
1135 
1136 		/* prevent any other reads prior to eop_desc */
1137 		smp_rmb();
1138 
1139 		/* if DD is not set pending work has not been completed */
1140 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1141 			break;
1142 
1143 		/* clear next_to_watch to prevent false hangs */
1144 		tx_buffer->next_to_watch = NULL;
1145 
1146 		/* update the statistics for this packet */
1147 		total_bytes += tx_buffer->bytecount;
1148 		total_packets += tx_buffer->gso_segs;
1149 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1150 			total_ipsec++;
1151 
1152 		/* free the skb */
1153 		if (ring_is_xdp(tx_ring))
1154 			xdp_return_frame(tx_buffer->xdpf);
1155 		else
1156 			napi_consume_skb(tx_buffer->skb, napi_budget);
1157 
1158 		/* unmap skb header data */
1159 		dma_unmap_single(tx_ring->dev,
1160 				 dma_unmap_addr(tx_buffer, dma),
1161 				 dma_unmap_len(tx_buffer, len),
1162 				 DMA_TO_DEVICE);
1163 
1164 		/* clear tx_buffer data */
1165 		dma_unmap_len_set(tx_buffer, len, 0);
1166 
1167 		/* unmap remaining buffers */
1168 		while (tx_desc != eop_desc) {
1169 			tx_buffer++;
1170 			tx_desc++;
1171 			i++;
1172 			if (unlikely(!i)) {
1173 				i -= tx_ring->count;
1174 				tx_buffer = tx_ring->tx_buffer_info;
1175 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1176 			}
1177 
1178 			/* unmap any remaining paged data */
1179 			if (dma_unmap_len(tx_buffer, len)) {
1180 				dma_unmap_page(tx_ring->dev,
1181 					       dma_unmap_addr(tx_buffer, dma),
1182 					       dma_unmap_len(tx_buffer, len),
1183 					       DMA_TO_DEVICE);
1184 				dma_unmap_len_set(tx_buffer, len, 0);
1185 			}
1186 		}
1187 
1188 		/* move us one more past the eop_desc for start of next pkt */
1189 		tx_buffer++;
1190 		tx_desc++;
1191 		i++;
1192 		if (unlikely(!i)) {
1193 			i -= tx_ring->count;
1194 			tx_buffer = tx_ring->tx_buffer_info;
1195 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1196 		}
1197 
1198 		/* issue prefetch for next Tx descriptor */
1199 		prefetch(tx_desc);
1200 
1201 		/* update budget accounting */
1202 		budget--;
1203 	} while (likely(budget));
1204 
1205 	i += tx_ring->count;
1206 	tx_ring->next_to_clean = i;
1207 	u64_stats_update_begin(&tx_ring->syncp);
1208 	tx_ring->stats.bytes += total_bytes;
1209 	tx_ring->stats.packets += total_packets;
1210 	u64_stats_update_end(&tx_ring->syncp);
1211 	q_vector->tx.total_bytes += total_bytes;
1212 	q_vector->tx.total_packets += total_packets;
1213 	adapter->tx_ipsec += total_ipsec;
1214 
1215 	if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1216 		/* schedule immediate reset if we believe we hung */
1217 		struct ixgbe_hw *hw = &adapter->hw;
1218 		e_err(drv, "Detected Tx Unit Hang %s\n"
1219 			"  Tx Queue             <%d>\n"
1220 			"  TDH, TDT             <%x>, <%x>\n"
1221 			"  next_to_use          <%x>\n"
1222 			"  next_to_clean        <%x>\n"
1223 			"tx_buffer_info[next_to_clean]\n"
1224 			"  time_stamp           <%lx>\n"
1225 			"  jiffies              <%lx>\n",
1226 			ring_is_xdp(tx_ring) ? "(XDP)" : "",
1227 			tx_ring->queue_index,
1228 			IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1229 			IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1230 			tx_ring->next_to_use, i,
1231 			tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1232 
1233 		if (!ring_is_xdp(tx_ring))
1234 			netif_stop_subqueue(tx_ring->netdev,
1235 					    tx_ring->queue_index);
1236 
1237 		e_info(probe,
1238 		       "tx hang %d detected on queue %d, resetting adapter\n",
1239 			adapter->tx_timeout_count + 1, tx_ring->queue_index);
1240 
1241 		/* schedule immediate reset if we believe we hung */
1242 		ixgbe_tx_timeout_reset(adapter);
1243 
1244 		/* the adapter is about to reset, no point in enabling stuff */
1245 		return true;
1246 	}
1247 
1248 	if (ring_is_xdp(tx_ring))
1249 		return !!budget;
1250 
1251 	netdev_tx_completed_queue(txring_txq(tx_ring),
1252 				  total_packets, total_bytes);
1253 
1254 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1255 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1256 		     (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1257 		/* Make sure that anybody stopping the queue after this
1258 		 * sees the new next_to_clean.
1259 		 */
1260 		smp_mb();
1261 		if (__netif_subqueue_stopped(tx_ring->netdev,
1262 					     tx_ring->queue_index)
1263 		    && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1264 			netif_wake_subqueue(tx_ring->netdev,
1265 					    tx_ring->queue_index);
1266 			++tx_ring->tx_stats.restart_queue;
1267 		}
1268 	}
1269 
1270 	return !!budget;
1271 }
1272 
1273 #ifdef CONFIG_IXGBE_DCA
1274 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1275 				struct ixgbe_ring *tx_ring,
1276 				int cpu)
1277 {
1278 	struct ixgbe_hw *hw = &adapter->hw;
1279 	u32 txctrl = 0;
1280 	u16 reg_offset;
1281 
1282 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1283 		txctrl = dca3_get_tag(tx_ring->dev, cpu);
1284 
1285 	switch (hw->mac.type) {
1286 	case ixgbe_mac_82598EB:
1287 		reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1288 		break;
1289 	case ixgbe_mac_82599EB:
1290 	case ixgbe_mac_X540:
1291 		reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1292 		txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1293 		break;
1294 	default:
1295 		/* for unknown hardware do not write register */
1296 		return;
1297 	}
1298 
1299 	/*
1300 	 * We can enable relaxed ordering for reads, but not writes when
1301 	 * DCA is enabled.  This is due to a known issue in some chipsets
1302 	 * which will cause the DCA tag to be cleared.
1303 	 */
1304 	txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1305 		  IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1306 		  IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1307 
1308 	IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1309 }
1310 
1311 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1312 				struct ixgbe_ring *rx_ring,
1313 				int cpu)
1314 {
1315 	struct ixgbe_hw *hw = &adapter->hw;
1316 	u32 rxctrl = 0;
1317 	u8 reg_idx = rx_ring->reg_idx;
1318 
1319 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1320 		rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1321 
1322 	switch (hw->mac.type) {
1323 	case ixgbe_mac_82599EB:
1324 	case ixgbe_mac_X540:
1325 		rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1326 		break;
1327 	default:
1328 		break;
1329 	}
1330 
1331 	/*
1332 	 * We can enable relaxed ordering for reads, but not writes when
1333 	 * DCA is enabled.  This is due to a known issue in some chipsets
1334 	 * which will cause the DCA tag to be cleared.
1335 	 */
1336 	rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1337 		  IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1338 		  IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1339 
1340 	IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1341 }
1342 
1343 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1344 {
1345 	struct ixgbe_adapter *adapter = q_vector->adapter;
1346 	struct ixgbe_ring *ring;
1347 	int cpu = get_cpu();
1348 
1349 	if (q_vector->cpu == cpu)
1350 		goto out_no_update;
1351 
1352 	ixgbe_for_each_ring(ring, q_vector->tx)
1353 		ixgbe_update_tx_dca(adapter, ring, cpu);
1354 
1355 	ixgbe_for_each_ring(ring, q_vector->rx)
1356 		ixgbe_update_rx_dca(adapter, ring, cpu);
1357 
1358 	q_vector->cpu = cpu;
1359 out_no_update:
1360 	put_cpu();
1361 }
1362 
1363 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1364 {
1365 	int i;
1366 
1367 	/* always use CB2 mode, difference is masked in the CB driver */
1368 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1369 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1370 				IXGBE_DCA_CTRL_DCA_MODE_CB2);
1371 	else
1372 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1373 				IXGBE_DCA_CTRL_DCA_DISABLE);
1374 
1375 	for (i = 0; i < adapter->num_q_vectors; i++) {
1376 		adapter->q_vector[i]->cpu = -1;
1377 		ixgbe_update_dca(adapter->q_vector[i]);
1378 	}
1379 }
1380 
1381 static int __ixgbe_notify_dca(struct device *dev, void *data)
1382 {
1383 	struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1384 	unsigned long event = *(unsigned long *)data;
1385 
1386 	if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1387 		return 0;
1388 
1389 	switch (event) {
1390 	case DCA_PROVIDER_ADD:
1391 		/* if we're already enabled, don't do it again */
1392 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1393 			break;
1394 		if (dca_add_requester(dev) == 0) {
1395 			adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1396 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1397 					IXGBE_DCA_CTRL_DCA_MODE_CB2);
1398 			break;
1399 		}
1400 		/* fall through - DCA is disabled. */
1401 	case DCA_PROVIDER_REMOVE:
1402 		if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1403 			dca_remove_requester(dev);
1404 			adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1405 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1406 					IXGBE_DCA_CTRL_DCA_DISABLE);
1407 		}
1408 		break;
1409 	}
1410 
1411 	return 0;
1412 }
1413 
1414 #endif /* CONFIG_IXGBE_DCA */
1415 
1416 #define IXGBE_RSS_L4_TYPES_MASK \
1417 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1418 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1419 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1420 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1421 
1422 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1423 				 union ixgbe_adv_rx_desc *rx_desc,
1424 				 struct sk_buff *skb)
1425 {
1426 	u16 rss_type;
1427 
1428 	if (!(ring->netdev->features & NETIF_F_RXHASH))
1429 		return;
1430 
1431 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1432 		   IXGBE_RXDADV_RSSTYPE_MASK;
1433 
1434 	if (!rss_type)
1435 		return;
1436 
1437 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1438 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1439 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1440 }
1441 
1442 #ifdef IXGBE_FCOE
1443 /**
1444  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1445  * @ring: structure containing ring specific data
1446  * @rx_desc: advanced rx descriptor
1447  *
1448  * Returns : true if it is FCoE pkt
1449  */
1450 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1451 				    union ixgbe_adv_rx_desc *rx_desc)
1452 {
1453 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1454 
1455 	return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1456 	       ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1457 		(cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1458 			     IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1459 }
1460 
1461 #endif /* IXGBE_FCOE */
1462 /**
1463  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1464  * @ring: structure containing ring specific data
1465  * @rx_desc: current Rx descriptor being processed
1466  * @skb: skb currently being received and modified
1467  **/
1468 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1469 				     union ixgbe_adv_rx_desc *rx_desc,
1470 				     struct sk_buff *skb)
1471 {
1472 	__le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1473 	bool encap_pkt = false;
1474 
1475 	skb_checksum_none_assert(skb);
1476 
1477 	/* Rx csum disabled */
1478 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1479 		return;
1480 
1481 	/* check for VXLAN and Geneve packets */
1482 	if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1483 		encap_pkt = true;
1484 		skb->encapsulation = 1;
1485 	}
1486 
1487 	/* if IP and error */
1488 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1489 	    ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1490 		ring->rx_stats.csum_err++;
1491 		return;
1492 	}
1493 
1494 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1495 		return;
1496 
1497 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1498 		/*
1499 		 * 82599 errata, UDP frames with a 0 checksum can be marked as
1500 		 * checksum errors.
1501 		 */
1502 		if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1503 		    test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1504 			return;
1505 
1506 		ring->rx_stats.csum_err++;
1507 		return;
1508 	}
1509 
1510 	/* It must be a TCP or UDP packet with a valid checksum */
1511 	skb->ip_summed = CHECKSUM_UNNECESSARY;
1512 	if (encap_pkt) {
1513 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1514 			return;
1515 
1516 		if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1517 			skb->ip_summed = CHECKSUM_NONE;
1518 			return;
1519 		}
1520 		/* If we checked the outer header let the stack know */
1521 		skb->csum_level = 1;
1522 	}
1523 }
1524 
1525 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1526 {
1527 	return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1528 }
1529 
1530 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1531 				    struct ixgbe_rx_buffer *bi)
1532 {
1533 	struct page *page = bi->page;
1534 	dma_addr_t dma;
1535 
1536 	/* since we are recycling buffers we should seldom need to alloc */
1537 	if (likely(page))
1538 		return true;
1539 
1540 	/* alloc new page for storage */
1541 	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1542 	if (unlikely(!page)) {
1543 		rx_ring->rx_stats.alloc_rx_page_failed++;
1544 		return false;
1545 	}
1546 
1547 	/* map page for use */
1548 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1549 				 ixgbe_rx_pg_size(rx_ring),
1550 				 DMA_FROM_DEVICE,
1551 				 IXGBE_RX_DMA_ATTR);
1552 
1553 	/*
1554 	 * if mapping failed free memory back to system since
1555 	 * there isn't much point in holding memory we can't use
1556 	 */
1557 	if (dma_mapping_error(rx_ring->dev, dma)) {
1558 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
1559 
1560 		rx_ring->rx_stats.alloc_rx_page_failed++;
1561 		return false;
1562 	}
1563 
1564 	bi->dma = dma;
1565 	bi->page = page;
1566 	bi->page_offset = ixgbe_rx_offset(rx_ring);
1567 	page_ref_add(page, USHRT_MAX - 1);
1568 	bi->pagecnt_bias = USHRT_MAX;
1569 	rx_ring->rx_stats.alloc_rx_page++;
1570 
1571 	return true;
1572 }
1573 
1574 /**
1575  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1576  * @rx_ring: ring to place buffers on
1577  * @cleaned_count: number of buffers to replace
1578  **/
1579 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1580 {
1581 	union ixgbe_adv_rx_desc *rx_desc;
1582 	struct ixgbe_rx_buffer *bi;
1583 	u16 i = rx_ring->next_to_use;
1584 	u16 bufsz;
1585 
1586 	/* nothing to do */
1587 	if (!cleaned_count)
1588 		return;
1589 
1590 	rx_desc = IXGBE_RX_DESC(rx_ring, i);
1591 	bi = &rx_ring->rx_buffer_info[i];
1592 	i -= rx_ring->count;
1593 
1594 	bufsz = ixgbe_rx_bufsz(rx_ring);
1595 
1596 	do {
1597 		if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1598 			break;
1599 
1600 		/* sync the buffer for use by the device */
1601 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1602 						 bi->page_offset, bufsz,
1603 						 DMA_FROM_DEVICE);
1604 
1605 		/*
1606 		 * Refresh the desc even if buffer_addrs didn't change
1607 		 * because each write-back erases this info.
1608 		 */
1609 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1610 
1611 		rx_desc++;
1612 		bi++;
1613 		i++;
1614 		if (unlikely(!i)) {
1615 			rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1616 			bi = rx_ring->rx_buffer_info;
1617 			i -= rx_ring->count;
1618 		}
1619 
1620 		/* clear the length for the next_to_use descriptor */
1621 		rx_desc->wb.upper.length = 0;
1622 
1623 		cleaned_count--;
1624 	} while (cleaned_count);
1625 
1626 	i += rx_ring->count;
1627 
1628 	if (rx_ring->next_to_use != i) {
1629 		rx_ring->next_to_use = i;
1630 
1631 		/* update next to alloc since we have filled the ring */
1632 		rx_ring->next_to_alloc = i;
1633 
1634 		/* Force memory writes to complete before letting h/w
1635 		 * know there are new descriptors to fetch.  (Only
1636 		 * applicable for weak-ordered memory model archs,
1637 		 * such as IA-64).
1638 		 */
1639 		wmb();
1640 		writel(i, rx_ring->tail);
1641 	}
1642 }
1643 
1644 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1645 				   struct sk_buff *skb)
1646 {
1647 	u16 hdr_len = skb_headlen(skb);
1648 
1649 	/* set gso_size to avoid messing up TCP MSS */
1650 	skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1651 						 IXGBE_CB(skb)->append_cnt);
1652 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1653 }
1654 
1655 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1656 				   struct sk_buff *skb)
1657 {
1658 	/* if append_cnt is 0 then frame is not RSC */
1659 	if (!IXGBE_CB(skb)->append_cnt)
1660 		return;
1661 
1662 	rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1663 	rx_ring->rx_stats.rsc_flush++;
1664 
1665 	ixgbe_set_rsc_gso_size(rx_ring, skb);
1666 
1667 	/* gso_size is computed using append_cnt so always clear it last */
1668 	IXGBE_CB(skb)->append_cnt = 0;
1669 }
1670 
1671 /**
1672  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1673  * @rx_ring: rx descriptor ring packet is being transacted on
1674  * @rx_desc: pointer to the EOP Rx descriptor
1675  * @skb: pointer to current skb being populated
1676  *
1677  * This function checks the ring, descriptor, and packet information in
1678  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1679  * other fields within the skb.
1680  **/
1681 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1682 			      union ixgbe_adv_rx_desc *rx_desc,
1683 			      struct sk_buff *skb)
1684 {
1685 	struct net_device *dev = rx_ring->netdev;
1686 	u32 flags = rx_ring->q_vector->adapter->flags;
1687 
1688 	ixgbe_update_rsc_stats(rx_ring, skb);
1689 
1690 	ixgbe_rx_hash(rx_ring, rx_desc, skb);
1691 
1692 	ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1693 
1694 	if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1695 		ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1696 
1697 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1698 	    ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1699 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1700 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1701 	}
1702 
1703 	if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1704 		ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1705 
1706 	/* record Rx queue, or update MACVLAN statistics */
1707 	if (netif_is_ixgbe(dev))
1708 		skb_record_rx_queue(skb, rx_ring->queue_index);
1709 	else
1710 		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1711 				 false);
1712 
1713 	skb->protocol = eth_type_trans(skb, dev);
1714 }
1715 
1716 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1717 		  struct sk_buff *skb)
1718 {
1719 	napi_gro_receive(&q_vector->napi, skb);
1720 }
1721 
1722 /**
1723  * ixgbe_is_non_eop - process handling of non-EOP buffers
1724  * @rx_ring: Rx ring being processed
1725  * @rx_desc: Rx descriptor for current buffer
1726  * @skb: Current socket buffer containing buffer in progress
1727  *
1728  * This function updates next to clean.  If the buffer is an EOP buffer
1729  * this function exits returning false, otherwise it will place the
1730  * sk_buff in the next buffer to be chained and return true indicating
1731  * that this is in fact a non-EOP buffer.
1732  **/
1733 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1734 			     union ixgbe_adv_rx_desc *rx_desc,
1735 			     struct sk_buff *skb)
1736 {
1737 	u32 ntc = rx_ring->next_to_clean + 1;
1738 
1739 	/* fetch, update, and store next to clean */
1740 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1741 	rx_ring->next_to_clean = ntc;
1742 
1743 	prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1744 
1745 	/* update RSC append count if present */
1746 	if (ring_is_rsc_enabled(rx_ring)) {
1747 		__le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1748 				     cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1749 
1750 		if (unlikely(rsc_enabled)) {
1751 			u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1752 
1753 			rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1754 			IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1755 
1756 			/* update ntc based on RSC value */
1757 			ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1758 			ntc &= IXGBE_RXDADV_NEXTP_MASK;
1759 			ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1760 		}
1761 	}
1762 
1763 	/* if we are the last buffer then there is nothing else to do */
1764 	if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1765 		return false;
1766 
1767 	/* place skb in next buffer to be received */
1768 	rx_ring->rx_buffer_info[ntc].skb = skb;
1769 	rx_ring->rx_stats.non_eop_descs++;
1770 
1771 	return true;
1772 }
1773 
1774 /**
1775  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1776  * @rx_ring: rx descriptor ring packet is being transacted on
1777  * @skb: pointer to current skb being adjusted
1778  *
1779  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1780  * main difference between this version and the original function is that
1781  * this function can make several assumptions about the state of things
1782  * that allow for significant optimizations versus the standard function.
1783  * As a result we can do things like drop a frag and maintain an accurate
1784  * truesize for the skb.
1785  */
1786 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1787 			    struct sk_buff *skb)
1788 {
1789 	skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1790 	unsigned char *va;
1791 	unsigned int pull_len;
1792 
1793 	/*
1794 	 * it is valid to use page_address instead of kmap since we are
1795 	 * working with pages allocated out of the lomem pool per
1796 	 * alloc_page(GFP_ATOMIC)
1797 	 */
1798 	va = skb_frag_address(frag);
1799 
1800 	/*
1801 	 * we need the header to contain the greater of either ETH_HLEN or
1802 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
1803 	 */
1804 	pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1805 
1806 	/* align pull length to size of long to optimize memcpy performance */
1807 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1808 
1809 	/* update all of the pointers */
1810 	skb_frag_size_sub(frag, pull_len);
1811 	skb_frag_off_add(frag, pull_len);
1812 	skb->data_len -= pull_len;
1813 	skb->tail += pull_len;
1814 }
1815 
1816 /**
1817  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1818  * @rx_ring: rx descriptor ring packet is being transacted on
1819  * @skb: pointer to current skb being updated
1820  *
1821  * This function provides a basic DMA sync up for the first fragment of an
1822  * skb.  The reason for doing this is that the first fragment cannot be
1823  * unmapped until we have reached the end of packet descriptor for a buffer
1824  * chain.
1825  */
1826 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1827 				struct sk_buff *skb)
1828 {
1829 	if (ring_uses_build_skb(rx_ring)) {
1830 		unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1831 
1832 		dma_sync_single_range_for_cpu(rx_ring->dev,
1833 					      IXGBE_CB(skb)->dma,
1834 					      offset,
1835 					      skb_headlen(skb),
1836 					      DMA_FROM_DEVICE);
1837 	} else {
1838 		skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
1839 
1840 		dma_sync_single_range_for_cpu(rx_ring->dev,
1841 					      IXGBE_CB(skb)->dma,
1842 					      skb_frag_off(frag),
1843 					      skb_frag_size(frag),
1844 					      DMA_FROM_DEVICE);
1845 	}
1846 
1847 	/* If the page was released, just unmap it. */
1848 	if (unlikely(IXGBE_CB(skb)->page_released)) {
1849 		dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1850 				     ixgbe_rx_pg_size(rx_ring),
1851 				     DMA_FROM_DEVICE,
1852 				     IXGBE_RX_DMA_ATTR);
1853 	}
1854 }
1855 
1856 /**
1857  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1858  * @rx_ring: rx descriptor ring packet is being transacted on
1859  * @rx_desc: pointer to the EOP Rx descriptor
1860  * @skb: pointer to current skb being fixed
1861  *
1862  * Check if the skb is valid in the XDP case it will be an error pointer.
1863  * Return true in this case to abort processing and advance to next
1864  * descriptor.
1865  *
1866  * Check for corrupted packet headers caused by senders on the local L2
1867  * embedded NIC switch not setting up their Tx Descriptors right.  These
1868  * should be very rare.
1869  *
1870  * Also address the case where we are pulling data in on pages only
1871  * and as such no data is present in the skb header.
1872  *
1873  * In addition if skb is not at least 60 bytes we need to pad it so that
1874  * it is large enough to qualify as a valid Ethernet frame.
1875  *
1876  * Returns true if an error was encountered and skb was freed.
1877  **/
1878 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1879 			   union ixgbe_adv_rx_desc *rx_desc,
1880 			   struct sk_buff *skb)
1881 {
1882 	struct net_device *netdev = rx_ring->netdev;
1883 
1884 	/* XDP packets use error pointer so abort at this point */
1885 	if (IS_ERR(skb))
1886 		return true;
1887 
1888 	/* Verify netdev is present, and that packet does not have any
1889 	 * errors that would be unacceptable to the netdev.
1890 	 */
1891 	if (!netdev ||
1892 	    (unlikely(ixgbe_test_staterr(rx_desc,
1893 					 IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1894 	     !(netdev->features & NETIF_F_RXALL)))) {
1895 		dev_kfree_skb_any(skb);
1896 		return true;
1897 	}
1898 
1899 	/* place header in linear portion of buffer */
1900 	if (!skb_headlen(skb))
1901 		ixgbe_pull_tail(rx_ring, skb);
1902 
1903 #ifdef IXGBE_FCOE
1904 	/* do not attempt to pad FCoE Frames as this will disrupt DDP */
1905 	if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1906 		return false;
1907 
1908 #endif
1909 	/* if eth_skb_pad returns an error the skb was freed */
1910 	if (eth_skb_pad(skb))
1911 		return true;
1912 
1913 	return false;
1914 }
1915 
1916 /**
1917  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1918  * @rx_ring: rx descriptor ring to store buffers on
1919  * @old_buff: donor buffer to have page reused
1920  *
1921  * Synchronizes page for reuse by the adapter
1922  **/
1923 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1924 				struct ixgbe_rx_buffer *old_buff)
1925 {
1926 	struct ixgbe_rx_buffer *new_buff;
1927 	u16 nta = rx_ring->next_to_alloc;
1928 
1929 	new_buff = &rx_ring->rx_buffer_info[nta];
1930 
1931 	/* update, and store next to alloc */
1932 	nta++;
1933 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1934 
1935 	/* Transfer page from old buffer to new buffer.
1936 	 * Move each member individually to avoid possible store
1937 	 * forwarding stalls and unnecessary copy of skb.
1938 	 */
1939 	new_buff->dma		= old_buff->dma;
1940 	new_buff->page		= old_buff->page;
1941 	new_buff->page_offset	= old_buff->page_offset;
1942 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1943 }
1944 
1945 static inline bool ixgbe_page_is_reserved(struct page *page)
1946 {
1947 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1948 }
1949 
1950 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1951 {
1952 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1953 	struct page *page = rx_buffer->page;
1954 
1955 	/* avoid re-using remote pages */
1956 	if (unlikely(ixgbe_page_is_reserved(page)))
1957 		return false;
1958 
1959 #if (PAGE_SIZE < 8192)
1960 	/* if we are only owner of page we can reuse it */
1961 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1962 		return false;
1963 #else
1964 	/* The last offset is a bit aggressive in that we assume the
1965 	 * worst case of FCoE being enabled and using a 3K buffer.
1966 	 * However this should have minimal impact as the 1K extra is
1967 	 * still less than one buffer in size.
1968 	 */
1969 #define IXGBE_LAST_OFFSET \
1970 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1971 	if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1972 		return false;
1973 #endif
1974 
1975 	/* If we have drained the page fragment pool we need to update
1976 	 * the pagecnt_bias and page count so that we fully restock the
1977 	 * number of references the driver holds.
1978 	 */
1979 	if (unlikely(pagecnt_bias == 1)) {
1980 		page_ref_add(page, USHRT_MAX - 1);
1981 		rx_buffer->pagecnt_bias = USHRT_MAX;
1982 	}
1983 
1984 	return true;
1985 }
1986 
1987 /**
1988  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1989  * @rx_ring: rx descriptor ring to transact packets on
1990  * @rx_buffer: buffer containing page to add
1991  * @skb: sk_buff to place the data into
1992  * @size: size of data in rx_buffer
1993  *
1994  * This function will add the data contained in rx_buffer->page to the skb.
1995  * This is done either through a direct copy if the data in the buffer is
1996  * less than the skb header size, otherwise it will just attach the page as
1997  * a frag to the skb.
1998  *
1999  * The function will then update the page offset if necessary and return
2000  * true if the buffer can be reused by the adapter.
2001  **/
2002 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2003 			      struct ixgbe_rx_buffer *rx_buffer,
2004 			      struct sk_buff *skb,
2005 			      unsigned int size)
2006 {
2007 #if (PAGE_SIZE < 8192)
2008 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2009 #else
2010 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2011 				SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2012 				SKB_DATA_ALIGN(size);
2013 #endif
2014 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2015 			rx_buffer->page_offset, size, truesize);
2016 #if (PAGE_SIZE < 8192)
2017 	rx_buffer->page_offset ^= truesize;
2018 #else
2019 	rx_buffer->page_offset += truesize;
2020 #endif
2021 }
2022 
2023 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2024 						   union ixgbe_adv_rx_desc *rx_desc,
2025 						   struct sk_buff **skb,
2026 						   const unsigned int size)
2027 {
2028 	struct ixgbe_rx_buffer *rx_buffer;
2029 
2030 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2031 	prefetchw(rx_buffer->page);
2032 	*skb = rx_buffer->skb;
2033 
2034 	/* Delay unmapping of the first packet. It carries the header
2035 	 * information, HW may still access the header after the writeback.
2036 	 * Only unmap it when EOP is reached
2037 	 */
2038 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2039 		if (!*skb)
2040 			goto skip_sync;
2041 	} else {
2042 		if (*skb)
2043 			ixgbe_dma_sync_frag(rx_ring, *skb);
2044 	}
2045 
2046 	/* we are reusing so sync this buffer for CPU use */
2047 	dma_sync_single_range_for_cpu(rx_ring->dev,
2048 				      rx_buffer->dma,
2049 				      rx_buffer->page_offset,
2050 				      size,
2051 				      DMA_FROM_DEVICE);
2052 skip_sync:
2053 	rx_buffer->pagecnt_bias--;
2054 
2055 	return rx_buffer;
2056 }
2057 
2058 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2059 				struct ixgbe_rx_buffer *rx_buffer,
2060 				struct sk_buff *skb)
2061 {
2062 	if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2063 		/* hand second half of page back to the ring */
2064 		ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2065 	} else {
2066 		if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2067 			/* the page has been released from the ring */
2068 			IXGBE_CB(skb)->page_released = true;
2069 		} else {
2070 			/* we are not reusing the buffer so unmap it */
2071 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2072 					     ixgbe_rx_pg_size(rx_ring),
2073 					     DMA_FROM_DEVICE,
2074 					     IXGBE_RX_DMA_ATTR);
2075 		}
2076 		__page_frag_cache_drain(rx_buffer->page,
2077 					rx_buffer->pagecnt_bias);
2078 	}
2079 
2080 	/* clear contents of rx_buffer */
2081 	rx_buffer->page = NULL;
2082 	rx_buffer->skb = NULL;
2083 }
2084 
2085 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2086 					   struct ixgbe_rx_buffer *rx_buffer,
2087 					   struct xdp_buff *xdp,
2088 					   union ixgbe_adv_rx_desc *rx_desc)
2089 {
2090 	unsigned int size = xdp->data_end - xdp->data;
2091 #if (PAGE_SIZE < 8192)
2092 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2093 #else
2094 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2095 					       xdp->data_hard_start);
2096 #endif
2097 	struct sk_buff *skb;
2098 
2099 	/* prefetch first cache line of first page */
2100 	prefetch(xdp->data);
2101 #if L1_CACHE_BYTES < 128
2102 	prefetch(xdp->data + L1_CACHE_BYTES);
2103 #endif
2104 	/* Note, we get here by enabling legacy-rx via:
2105 	 *
2106 	 *    ethtool --set-priv-flags <dev> legacy-rx on
2107 	 *
2108 	 * In this mode, we currently get 0 extra XDP headroom as
2109 	 * opposed to having legacy-rx off, where we process XDP
2110 	 * packets going to stack via ixgbe_build_skb(). The latter
2111 	 * provides us currently with 192 bytes of headroom.
2112 	 *
2113 	 * For ixgbe_construct_skb() mode it means that the
2114 	 * xdp->data_meta will always point to xdp->data, since
2115 	 * the helper cannot expand the head. Should this ever
2116 	 * change in future for legacy-rx mode on, then lets also
2117 	 * add xdp->data_meta handling here.
2118 	 */
2119 
2120 	/* allocate a skb to store the frags */
2121 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2122 	if (unlikely(!skb))
2123 		return NULL;
2124 
2125 	if (size > IXGBE_RX_HDR_SIZE) {
2126 		if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2127 			IXGBE_CB(skb)->dma = rx_buffer->dma;
2128 
2129 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2130 				xdp->data - page_address(rx_buffer->page),
2131 				size, truesize);
2132 #if (PAGE_SIZE < 8192)
2133 		rx_buffer->page_offset ^= truesize;
2134 #else
2135 		rx_buffer->page_offset += truesize;
2136 #endif
2137 	} else {
2138 		memcpy(__skb_put(skb, size),
2139 		       xdp->data, ALIGN(size, sizeof(long)));
2140 		rx_buffer->pagecnt_bias++;
2141 	}
2142 
2143 	return skb;
2144 }
2145 
2146 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2147 				       struct ixgbe_rx_buffer *rx_buffer,
2148 				       struct xdp_buff *xdp,
2149 				       union ixgbe_adv_rx_desc *rx_desc)
2150 {
2151 	unsigned int metasize = xdp->data - xdp->data_meta;
2152 #if (PAGE_SIZE < 8192)
2153 	unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2154 #else
2155 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2156 				SKB_DATA_ALIGN(xdp->data_end -
2157 					       xdp->data_hard_start);
2158 #endif
2159 	struct sk_buff *skb;
2160 
2161 	/* Prefetch first cache line of first page. If xdp->data_meta
2162 	 * is unused, this points extactly as xdp->data, otherwise we
2163 	 * likely have a consumer accessing first few bytes of meta
2164 	 * data, and then actual data.
2165 	 */
2166 	prefetch(xdp->data_meta);
2167 #if L1_CACHE_BYTES < 128
2168 	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2169 #endif
2170 
2171 	/* build an skb to around the page buffer */
2172 	skb = build_skb(xdp->data_hard_start, truesize);
2173 	if (unlikely(!skb))
2174 		return NULL;
2175 
2176 	/* update pointers within the skb to store the data */
2177 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2178 	__skb_put(skb, xdp->data_end - xdp->data);
2179 	if (metasize)
2180 		skb_metadata_set(skb, metasize);
2181 
2182 	/* record DMA address if this is the start of a chain of buffers */
2183 	if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2184 		IXGBE_CB(skb)->dma = rx_buffer->dma;
2185 
2186 	/* update buffer offset */
2187 #if (PAGE_SIZE < 8192)
2188 	rx_buffer->page_offset ^= truesize;
2189 #else
2190 	rx_buffer->page_offset += truesize;
2191 #endif
2192 
2193 	return skb;
2194 }
2195 
2196 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2197 				     struct ixgbe_ring *rx_ring,
2198 				     struct xdp_buff *xdp)
2199 {
2200 	int err, result = IXGBE_XDP_PASS;
2201 	struct bpf_prog *xdp_prog;
2202 	struct xdp_frame *xdpf;
2203 	u32 act;
2204 
2205 	rcu_read_lock();
2206 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2207 
2208 	if (!xdp_prog)
2209 		goto xdp_out;
2210 
2211 	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2212 
2213 	act = bpf_prog_run_xdp(xdp_prog, xdp);
2214 	switch (act) {
2215 	case XDP_PASS:
2216 		break;
2217 	case XDP_TX:
2218 		xdpf = convert_to_xdp_frame(xdp);
2219 		if (unlikely(!xdpf)) {
2220 			result = IXGBE_XDP_CONSUMED;
2221 			break;
2222 		}
2223 		result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2224 		break;
2225 	case XDP_REDIRECT:
2226 		err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2227 		if (!err)
2228 			result = IXGBE_XDP_REDIR;
2229 		else
2230 			result = IXGBE_XDP_CONSUMED;
2231 		break;
2232 	default:
2233 		bpf_warn_invalid_xdp_action(act);
2234 		/* fallthrough */
2235 	case XDP_ABORTED:
2236 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2237 		/* fallthrough -- handle aborts by dropping packet */
2238 	case XDP_DROP:
2239 		result = IXGBE_XDP_CONSUMED;
2240 		break;
2241 	}
2242 xdp_out:
2243 	rcu_read_unlock();
2244 	return ERR_PTR(-result);
2245 }
2246 
2247 static unsigned int ixgbe_rx_frame_truesize(struct ixgbe_ring *rx_ring,
2248 					    unsigned int size)
2249 {
2250 	unsigned int truesize;
2251 
2252 #if (PAGE_SIZE < 8192)
2253 	truesize = ixgbe_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
2254 #else
2255 	truesize = ring_uses_build_skb(rx_ring) ?
2256 		SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) +
2257 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
2258 		SKB_DATA_ALIGN(size);
2259 #endif
2260 	return truesize;
2261 }
2262 
2263 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2264 				 struct ixgbe_rx_buffer *rx_buffer,
2265 				 unsigned int size)
2266 {
2267 	unsigned int truesize = ixgbe_rx_frame_truesize(rx_ring, size);
2268 #if (PAGE_SIZE < 8192)
2269 	rx_buffer->page_offset ^= truesize;
2270 #else
2271 	rx_buffer->page_offset += truesize;
2272 #endif
2273 }
2274 
2275 /**
2276  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2277  * @q_vector: structure containing interrupt and ring information
2278  * @rx_ring: rx descriptor ring to transact packets on
2279  * @budget: Total limit on number of packets to process
2280  *
2281  * This function provides a "bounce buffer" approach to Rx interrupt
2282  * processing.  The advantage to this is that on systems that have
2283  * expensive overhead for IOMMU access this provides a means of avoiding
2284  * it by maintaining the mapping of the page to the syste.
2285  *
2286  * Returns amount of work completed
2287  **/
2288 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2289 			       struct ixgbe_ring *rx_ring,
2290 			       const int budget)
2291 {
2292 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2293 	struct ixgbe_adapter *adapter = q_vector->adapter;
2294 #ifdef IXGBE_FCOE
2295 	int ddp_bytes;
2296 	unsigned int mss = 0;
2297 #endif /* IXGBE_FCOE */
2298 	u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2299 	unsigned int xdp_xmit = 0;
2300 	struct xdp_buff xdp;
2301 
2302 	xdp.rxq = &rx_ring->xdp_rxq;
2303 
2304 	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
2305 #if (PAGE_SIZE < 8192)
2306 	xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, 0);
2307 #endif
2308 
2309 	while (likely(total_rx_packets < budget)) {
2310 		union ixgbe_adv_rx_desc *rx_desc;
2311 		struct ixgbe_rx_buffer *rx_buffer;
2312 		struct sk_buff *skb;
2313 		unsigned int size;
2314 
2315 		/* return some buffers to hardware, one at a time is too slow */
2316 		if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2317 			ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2318 			cleaned_count = 0;
2319 		}
2320 
2321 		rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2322 		size = le16_to_cpu(rx_desc->wb.upper.length);
2323 		if (!size)
2324 			break;
2325 
2326 		/* This memory barrier is needed to keep us from reading
2327 		 * any other fields out of the rx_desc until we know the
2328 		 * descriptor has been written back
2329 		 */
2330 		dma_rmb();
2331 
2332 		rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2333 
2334 		/* retrieve a buffer from the ring */
2335 		if (!skb) {
2336 			xdp.data = page_address(rx_buffer->page) +
2337 				   rx_buffer->page_offset;
2338 			xdp.data_meta = xdp.data;
2339 			xdp.data_hard_start = xdp.data -
2340 					      ixgbe_rx_offset(rx_ring);
2341 			xdp.data_end = xdp.data + size;
2342 #if (PAGE_SIZE > 4096)
2343 			/* At larger PAGE_SIZE, frame_sz depend on len size */
2344 			xdp.frame_sz = ixgbe_rx_frame_truesize(rx_ring, size);
2345 #endif
2346 			skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2347 		}
2348 
2349 		if (IS_ERR(skb)) {
2350 			unsigned int xdp_res = -PTR_ERR(skb);
2351 
2352 			if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2353 				xdp_xmit |= xdp_res;
2354 				ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2355 			} else {
2356 				rx_buffer->pagecnt_bias++;
2357 			}
2358 			total_rx_packets++;
2359 			total_rx_bytes += size;
2360 		} else if (skb) {
2361 			ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2362 		} else if (ring_uses_build_skb(rx_ring)) {
2363 			skb = ixgbe_build_skb(rx_ring, rx_buffer,
2364 					      &xdp, rx_desc);
2365 		} else {
2366 			skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2367 						  &xdp, rx_desc);
2368 		}
2369 
2370 		/* exit if we failed to retrieve a buffer */
2371 		if (!skb) {
2372 			rx_ring->rx_stats.alloc_rx_buff_failed++;
2373 			rx_buffer->pagecnt_bias++;
2374 			break;
2375 		}
2376 
2377 		ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2378 		cleaned_count++;
2379 
2380 		/* place incomplete frames back on ring for completion */
2381 		if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2382 			continue;
2383 
2384 		/* verify the packet layout is correct */
2385 		if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2386 			continue;
2387 
2388 		/* probably a little skewed due to removing CRC */
2389 		total_rx_bytes += skb->len;
2390 
2391 		/* populate checksum, timestamp, VLAN, and protocol */
2392 		ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2393 
2394 #ifdef IXGBE_FCOE
2395 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
2396 		if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2397 			ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2398 			/* include DDPed FCoE data */
2399 			if (ddp_bytes > 0) {
2400 				if (!mss) {
2401 					mss = rx_ring->netdev->mtu -
2402 						sizeof(struct fcoe_hdr) -
2403 						sizeof(struct fc_frame_header) -
2404 						sizeof(struct fcoe_crc_eof);
2405 					if (mss > 512)
2406 						mss &= ~511;
2407 				}
2408 				total_rx_bytes += ddp_bytes;
2409 				total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2410 								 mss);
2411 			}
2412 			if (!ddp_bytes) {
2413 				dev_kfree_skb_any(skb);
2414 				continue;
2415 			}
2416 		}
2417 
2418 #endif /* IXGBE_FCOE */
2419 		ixgbe_rx_skb(q_vector, skb);
2420 
2421 		/* update budget accounting */
2422 		total_rx_packets++;
2423 	}
2424 
2425 	if (xdp_xmit & IXGBE_XDP_REDIR)
2426 		xdp_do_flush_map();
2427 
2428 	if (xdp_xmit & IXGBE_XDP_TX) {
2429 		struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2430 
2431 		/* Force memory writes to complete before letting h/w
2432 		 * know there are new descriptors to fetch.
2433 		 */
2434 		wmb();
2435 		writel(ring->next_to_use, ring->tail);
2436 	}
2437 
2438 	u64_stats_update_begin(&rx_ring->syncp);
2439 	rx_ring->stats.packets += total_rx_packets;
2440 	rx_ring->stats.bytes += total_rx_bytes;
2441 	u64_stats_update_end(&rx_ring->syncp);
2442 	q_vector->rx.total_packets += total_rx_packets;
2443 	q_vector->rx.total_bytes += total_rx_bytes;
2444 
2445 	return total_rx_packets;
2446 }
2447 
2448 /**
2449  * ixgbe_configure_msix - Configure MSI-X hardware
2450  * @adapter: board private structure
2451  *
2452  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2453  * interrupts.
2454  **/
2455 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2456 {
2457 	struct ixgbe_q_vector *q_vector;
2458 	int v_idx;
2459 	u32 mask;
2460 
2461 	/* Populate MSIX to EITR Select */
2462 	if (adapter->num_vfs > 32) {
2463 		u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2464 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2465 	}
2466 
2467 	/*
2468 	 * Populate the IVAR table and set the ITR values to the
2469 	 * corresponding register.
2470 	 */
2471 	for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2472 		struct ixgbe_ring *ring;
2473 		q_vector = adapter->q_vector[v_idx];
2474 
2475 		ixgbe_for_each_ring(ring, q_vector->rx)
2476 			ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2477 
2478 		ixgbe_for_each_ring(ring, q_vector->tx)
2479 			ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2480 
2481 		ixgbe_write_eitr(q_vector);
2482 	}
2483 
2484 	switch (adapter->hw.mac.type) {
2485 	case ixgbe_mac_82598EB:
2486 		ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2487 			       v_idx);
2488 		break;
2489 	case ixgbe_mac_82599EB:
2490 	case ixgbe_mac_X540:
2491 	case ixgbe_mac_X550:
2492 	case ixgbe_mac_X550EM_x:
2493 	case ixgbe_mac_x550em_a:
2494 		ixgbe_set_ivar(adapter, -1, 1, v_idx);
2495 		break;
2496 	default:
2497 		break;
2498 	}
2499 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2500 
2501 	/* set up to autoclear timer, and the vectors */
2502 	mask = IXGBE_EIMS_ENABLE_MASK;
2503 	mask &= ~(IXGBE_EIMS_OTHER |
2504 		  IXGBE_EIMS_MAILBOX |
2505 		  IXGBE_EIMS_LSC);
2506 
2507 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2508 }
2509 
2510 /**
2511  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2512  * @q_vector: structure containing interrupt and ring information
2513  * @ring_container: structure containing ring performance data
2514  *
2515  *      Stores a new ITR value based on packets and byte
2516  *      counts during the last interrupt.  The advantage of per interrupt
2517  *      computation is faster updates and more accurate ITR for the current
2518  *      traffic pattern.  Constants in this function were computed
2519  *      based on theoretical maximum wire speed and thresholds were set based
2520  *      on testing data as well as attempting to minimize response time
2521  *      while increasing bulk throughput.
2522  **/
2523 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2524 			     struct ixgbe_ring_container *ring_container)
2525 {
2526 	unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2527 			   IXGBE_ITR_ADAPTIVE_LATENCY;
2528 	unsigned int avg_wire_size, packets, bytes;
2529 	unsigned long next_update = jiffies;
2530 
2531 	/* If we don't have any rings just leave ourselves set for maximum
2532 	 * possible latency so we take ourselves out of the equation.
2533 	 */
2534 	if (!ring_container->ring)
2535 		return;
2536 
2537 	/* If we didn't update within up to 1 - 2 jiffies we can assume
2538 	 * that either packets are coming in so slow there hasn't been
2539 	 * any work, or that there is so much work that NAPI is dealing
2540 	 * with interrupt moderation and we don't need to do anything.
2541 	 */
2542 	if (time_after(next_update, ring_container->next_update))
2543 		goto clear_counts;
2544 
2545 	packets = ring_container->total_packets;
2546 
2547 	/* We have no packets to actually measure against. This means
2548 	 * either one of the other queues on this vector is active or
2549 	 * we are a Tx queue doing TSO with too high of an interrupt rate.
2550 	 *
2551 	 * When this occurs just tick up our delay by the minimum value
2552 	 * and hope that this extra delay will prevent us from being called
2553 	 * without any work on our queue.
2554 	 */
2555 	if (!packets) {
2556 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2557 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2558 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2559 		itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2560 		goto clear_counts;
2561 	}
2562 
2563 	bytes = ring_container->total_bytes;
2564 
2565 	/* If packets are less than 4 or bytes are less than 9000 assume
2566 	 * insufficient data to use bulk rate limiting approach. We are
2567 	 * likely latency driven.
2568 	 */
2569 	if (packets < 4 && bytes < 9000) {
2570 		itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2571 		goto adjust_by_size;
2572 	}
2573 
2574 	/* Between 4 and 48 we can assume that our current interrupt delay
2575 	 * is only slightly too low. As such we should increase it by a small
2576 	 * fixed amount.
2577 	 */
2578 	if (packets < 48) {
2579 		itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2580 		if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2581 			itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2582 		goto clear_counts;
2583 	}
2584 
2585 	/* Between 48 and 96 is our "goldilocks" zone where we are working
2586 	 * out "just right". Just report that our current ITR is good for us.
2587 	 */
2588 	if (packets < 96) {
2589 		itr = q_vector->itr >> 2;
2590 		goto clear_counts;
2591 	}
2592 
2593 	/* If packet count is 96 or greater we are likely looking at a slight
2594 	 * overrun of the delay we want. Try halving our delay to see if that
2595 	 * will cut the number of packets in half per interrupt.
2596 	 */
2597 	if (packets < 256) {
2598 		itr = q_vector->itr >> 3;
2599 		if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2600 			itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2601 		goto clear_counts;
2602 	}
2603 
2604 	/* The paths below assume we are dealing with a bulk ITR since number
2605 	 * of packets is 256 or greater. We are just going to have to compute
2606 	 * a value and try to bring the count under control, though for smaller
2607 	 * packet sizes there isn't much we can do as NAPI polling will likely
2608 	 * be kicking in sooner rather than later.
2609 	 */
2610 	itr = IXGBE_ITR_ADAPTIVE_BULK;
2611 
2612 adjust_by_size:
2613 	/* If packet counts are 256 or greater we can assume we have a gross
2614 	 * overestimation of what the rate should be. Instead of trying to fine
2615 	 * tune it just use the formula below to try and dial in an exact value
2616 	 * give the current packet size of the frame.
2617 	 */
2618 	avg_wire_size = bytes / packets;
2619 
2620 	/* The following is a crude approximation of:
2621 	 *  wmem_default / (size + overhead) = desired_pkts_per_int
2622 	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2623 	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2624 	 *
2625 	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
2626 	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2627 	 * formula down to
2628 	 *
2629 	 *  (170 * (size + 24)) / (size + 640) = ITR
2630 	 *
2631 	 * We first do some math on the packet size and then finally bitshift
2632 	 * by 8 after rounding up. We also have to account for PCIe link speed
2633 	 * difference as ITR scales based on this.
2634 	 */
2635 	if (avg_wire_size <= 60) {
2636 		/* Start at 50k ints/sec */
2637 		avg_wire_size = 5120;
2638 	} else if (avg_wire_size <= 316) {
2639 		/* 50K ints/sec to 16K ints/sec */
2640 		avg_wire_size *= 40;
2641 		avg_wire_size += 2720;
2642 	} else if (avg_wire_size <= 1084) {
2643 		/* 16K ints/sec to 9.2K ints/sec */
2644 		avg_wire_size *= 15;
2645 		avg_wire_size += 11452;
2646 	} else if (avg_wire_size < 1968) {
2647 		/* 9.2K ints/sec to 8K ints/sec */
2648 		avg_wire_size *= 5;
2649 		avg_wire_size += 22420;
2650 	} else {
2651 		/* plateau at a limit of 8K ints/sec */
2652 		avg_wire_size = 32256;
2653 	}
2654 
2655 	/* If we are in low latency mode half our delay which doubles the rate
2656 	 * to somewhere between 100K to 16K ints/sec
2657 	 */
2658 	if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2659 		avg_wire_size >>= 1;
2660 
2661 	/* Resultant value is 256 times larger than it needs to be. This
2662 	 * gives us room to adjust the value as needed to either increase
2663 	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2664 	 *
2665 	 * Use addition as we have already recorded the new latency flag
2666 	 * for the ITR value.
2667 	 */
2668 	switch (q_vector->adapter->link_speed) {
2669 	case IXGBE_LINK_SPEED_10GB_FULL:
2670 	case IXGBE_LINK_SPEED_100_FULL:
2671 	default:
2672 		itr += DIV_ROUND_UP(avg_wire_size,
2673 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2674 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2675 		break;
2676 	case IXGBE_LINK_SPEED_2_5GB_FULL:
2677 	case IXGBE_LINK_SPEED_1GB_FULL:
2678 	case IXGBE_LINK_SPEED_10_FULL:
2679 		if (avg_wire_size > 8064)
2680 			avg_wire_size = 8064;
2681 		itr += DIV_ROUND_UP(avg_wire_size,
2682 				    IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2683 		       IXGBE_ITR_ADAPTIVE_MIN_INC;
2684 		break;
2685 	}
2686 
2687 clear_counts:
2688 	/* write back value */
2689 	ring_container->itr = itr;
2690 
2691 	/* next update should occur within next jiffy */
2692 	ring_container->next_update = next_update + 1;
2693 
2694 	ring_container->total_bytes = 0;
2695 	ring_container->total_packets = 0;
2696 }
2697 
2698 /**
2699  * ixgbe_write_eitr - write EITR register in hardware specific way
2700  * @q_vector: structure containing interrupt and ring information
2701  *
2702  * This function is made to be called by ethtool and by the driver
2703  * when it needs to update EITR registers at runtime.  Hardware
2704  * specific quirks/differences are taken care of here.
2705  */
2706 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2707 {
2708 	struct ixgbe_adapter *adapter = q_vector->adapter;
2709 	struct ixgbe_hw *hw = &adapter->hw;
2710 	int v_idx = q_vector->v_idx;
2711 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2712 
2713 	switch (adapter->hw.mac.type) {
2714 	case ixgbe_mac_82598EB:
2715 		/* must write high and low 16 bits to reset counter */
2716 		itr_reg |= (itr_reg << 16);
2717 		break;
2718 	case ixgbe_mac_82599EB:
2719 	case ixgbe_mac_X540:
2720 	case ixgbe_mac_X550:
2721 	case ixgbe_mac_X550EM_x:
2722 	case ixgbe_mac_x550em_a:
2723 		/*
2724 		 * set the WDIS bit to not clear the timer bits and cause an
2725 		 * immediate assertion of the interrupt
2726 		 */
2727 		itr_reg |= IXGBE_EITR_CNT_WDIS;
2728 		break;
2729 	default:
2730 		break;
2731 	}
2732 	IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2733 }
2734 
2735 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2736 {
2737 	u32 new_itr;
2738 
2739 	ixgbe_update_itr(q_vector, &q_vector->tx);
2740 	ixgbe_update_itr(q_vector, &q_vector->rx);
2741 
2742 	/* use the smallest value of new ITR delay calculations */
2743 	new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2744 
2745 	/* Clear latency flag if set, shift into correct position */
2746 	new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2747 	new_itr <<= 2;
2748 
2749 	if (new_itr != q_vector->itr) {
2750 		/* save the algorithm value here */
2751 		q_vector->itr = new_itr;
2752 
2753 		ixgbe_write_eitr(q_vector);
2754 	}
2755 }
2756 
2757 /**
2758  * ixgbe_check_overtemp_subtask - check for over temperature
2759  * @adapter: pointer to adapter
2760  **/
2761 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2762 {
2763 	struct ixgbe_hw *hw = &adapter->hw;
2764 	u32 eicr = adapter->interrupt_event;
2765 	s32 rc;
2766 
2767 	if (test_bit(__IXGBE_DOWN, &adapter->state))
2768 		return;
2769 
2770 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2771 		return;
2772 
2773 	adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2774 
2775 	switch (hw->device_id) {
2776 	case IXGBE_DEV_ID_82599_T3_LOM:
2777 		/*
2778 		 * Since the warning interrupt is for both ports
2779 		 * we don't have to check if:
2780 		 *  - This interrupt wasn't for our port.
2781 		 *  - We may have missed the interrupt so always have to
2782 		 *    check if we  got a LSC
2783 		 */
2784 		if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2785 		    !(eicr & IXGBE_EICR_LSC))
2786 			return;
2787 
2788 		if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2789 			u32 speed;
2790 			bool link_up = false;
2791 
2792 			hw->mac.ops.check_link(hw, &speed, &link_up, false);
2793 
2794 			if (link_up)
2795 				return;
2796 		}
2797 
2798 		/* Check if this is not due to overtemp */
2799 		if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2800 			return;
2801 
2802 		break;
2803 	case IXGBE_DEV_ID_X550EM_A_1G_T:
2804 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2805 		rc = hw->phy.ops.check_overtemp(hw);
2806 		if (rc != IXGBE_ERR_OVERTEMP)
2807 			return;
2808 		break;
2809 	default:
2810 		if (adapter->hw.mac.type >= ixgbe_mac_X540)
2811 			return;
2812 		if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2813 			return;
2814 		break;
2815 	}
2816 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2817 
2818 	adapter->interrupt_event = 0;
2819 }
2820 
2821 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2822 {
2823 	struct ixgbe_hw *hw = &adapter->hw;
2824 
2825 	if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2826 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2827 		e_crit(probe, "Fan has stopped, replace the adapter\n");
2828 		/* write to clear the interrupt */
2829 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2830 	}
2831 }
2832 
2833 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2834 {
2835 	struct ixgbe_hw *hw = &adapter->hw;
2836 
2837 	if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2838 		return;
2839 
2840 	switch (adapter->hw.mac.type) {
2841 	case ixgbe_mac_82599EB:
2842 		/*
2843 		 * Need to check link state so complete overtemp check
2844 		 * on service task
2845 		 */
2846 		if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2847 		     (eicr & IXGBE_EICR_LSC)) &&
2848 		    (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2849 			adapter->interrupt_event = eicr;
2850 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2851 			ixgbe_service_event_schedule(adapter);
2852 			return;
2853 		}
2854 		return;
2855 	case ixgbe_mac_x550em_a:
2856 		if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2857 			adapter->interrupt_event = eicr;
2858 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2859 			ixgbe_service_event_schedule(adapter);
2860 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2861 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2862 			IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2863 					IXGBE_EICR_GPI_SDP0_X550EM_a);
2864 		}
2865 		return;
2866 	case ixgbe_mac_X550:
2867 	case ixgbe_mac_X540:
2868 		if (!(eicr & IXGBE_EICR_TS))
2869 			return;
2870 		break;
2871 	default:
2872 		return;
2873 	}
2874 
2875 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
2876 }
2877 
2878 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2879 {
2880 	switch (hw->mac.type) {
2881 	case ixgbe_mac_82598EB:
2882 		if (hw->phy.type == ixgbe_phy_nl)
2883 			return true;
2884 		return false;
2885 	case ixgbe_mac_82599EB:
2886 	case ixgbe_mac_X550EM_x:
2887 	case ixgbe_mac_x550em_a:
2888 		switch (hw->mac.ops.get_media_type(hw)) {
2889 		case ixgbe_media_type_fiber:
2890 		case ixgbe_media_type_fiber_qsfp:
2891 			return true;
2892 		default:
2893 			return false;
2894 		}
2895 	default:
2896 		return false;
2897 	}
2898 }
2899 
2900 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2901 {
2902 	struct ixgbe_hw *hw = &adapter->hw;
2903 	u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2904 
2905 	if (!ixgbe_is_sfp(hw))
2906 		return;
2907 
2908 	/* Later MAC's use different SDP */
2909 	if (hw->mac.type >= ixgbe_mac_X540)
2910 		eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2911 
2912 	if (eicr & eicr_mask) {
2913 		/* Clear the interrupt */
2914 		IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2915 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2916 			adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2917 			adapter->sfp_poll_time = 0;
2918 			ixgbe_service_event_schedule(adapter);
2919 		}
2920 	}
2921 
2922 	if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2923 	    (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2924 		/* Clear the interrupt */
2925 		IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2926 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2927 			adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2928 			ixgbe_service_event_schedule(adapter);
2929 		}
2930 	}
2931 }
2932 
2933 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2934 {
2935 	struct ixgbe_hw *hw = &adapter->hw;
2936 
2937 	adapter->lsc_int++;
2938 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2939 	adapter->link_check_timeout = jiffies;
2940 	if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2941 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2942 		IXGBE_WRITE_FLUSH(hw);
2943 		ixgbe_service_event_schedule(adapter);
2944 	}
2945 }
2946 
2947 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2948 					   u64 qmask)
2949 {
2950 	u32 mask;
2951 	struct ixgbe_hw *hw = &adapter->hw;
2952 
2953 	switch (hw->mac.type) {
2954 	case ixgbe_mac_82598EB:
2955 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2956 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2957 		break;
2958 	case ixgbe_mac_82599EB:
2959 	case ixgbe_mac_X540:
2960 	case ixgbe_mac_X550:
2961 	case ixgbe_mac_X550EM_x:
2962 	case ixgbe_mac_x550em_a:
2963 		mask = (qmask & 0xFFFFFFFF);
2964 		if (mask)
2965 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2966 		mask = (qmask >> 32);
2967 		if (mask)
2968 			IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2969 		break;
2970 	default:
2971 		break;
2972 	}
2973 	/* skip the flush */
2974 }
2975 
2976 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2977 					    u64 qmask)
2978 {
2979 	u32 mask;
2980 	struct ixgbe_hw *hw = &adapter->hw;
2981 
2982 	switch (hw->mac.type) {
2983 	case ixgbe_mac_82598EB:
2984 		mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2985 		IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2986 		break;
2987 	case ixgbe_mac_82599EB:
2988 	case ixgbe_mac_X540:
2989 	case ixgbe_mac_X550:
2990 	case ixgbe_mac_X550EM_x:
2991 	case ixgbe_mac_x550em_a:
2992 		mask = (qmask & 0xFFFFFFFF);
2993 		if (mask)
2994 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2995 		mask = (qmask >> 32);
2996 		if (mask)
2997 			IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2998 		break;
2999 	default:
3000 		break;
3001 	}
3002 	/* skip the flush */
3003 }
3004 
3005 /**
3006  * ixgbe_irq_enable - Enable default interrupt generation settings
3007  * @adapter: board private structure
3008  * @queues: enable irqs for queues
3009  * @flush: flush register write
3010  **/
3011 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
3012 				    bool flush)
3013 {
3014 	struct ixgbe_hw *hw = &adapter->hw;
3015 	u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
3016 
3017 	/* don't reenable LSC while waiting for link */
3018 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
3019 		mask &= ~IXGBE_EIMS_LSC;
3020 
3021 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3022 		switch (adapter->hw.mac.type) {
3023 		case ixgbe_mac_82599EB:
3024 			mask |= IXGBE_EIMS_GPI_SDP0(hw);
3025 			break;
3026 		case ixgbe_mac_X540:
3027 		case ixgbe_mac_X550:
3028 		case ixgbe_mac_X550EM_x:
3029 		case ixgbe_mac_x550em_a:
3030 			mask |= IXGBE_EIMS_TS;
3031 			break;
3032 		default:
3033 			break;
3034 		}
3035 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3036 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3037 	switch (adapter->hw.mac.type) {
3038 	case ixgbe_mac_82599EB:
3039 		mask |= IXGBE_EIMS_GPI_SDP1(hw);
3040 		mask |= IXGBE_EIMS_GPI_SDP2(hw);
3041 		/* fall through */
3042 	case ixgbe_mac_X540:
3043 	case ixgbe_mac_X550:
3044 	case ixgbe_mac_X550EM_x:
3045 	case ixgbe_mac_x550em_a:
3046 		if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3047 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3048 		    adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3049 			mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3050 		if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3051 			mask |= IXGBE_EICR_GPI_SDP0_X540;
3052 		mask |= IXGBE_EIMS_ECC;
3053 		mask |= IXGBE_EIMS_MAILBOX;
3054 		break;
3055 	default:
3056 		break;
3057 	}
3058 
3059 	if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3060 	    !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3061 		mask |= IXGBE_EIMS_FLOW_DIR;
3062 
3063 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3064 	if (queues)
3065 		ixgbe_irq_enable_queues(adapter, ~0);
3066 	if (flush)
3067 		IXGBE_WRITE_FLUSH(&adapter->hw);
3068 }
3069 
3070 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3071 {
3072 	struct ixgbe_adapter *adapter = data;
3073 	struct ixgbe_hw *hw = &adapter->hw;
3074 	u32 eicr;
3075 
3076 	/*
3077 	 * Workaround for Silicon errata.  Use clear-by-write instead
3078 	 * of clear-by-read.  Reading with EICS will return the
3079 	 * interrupt causes without clearing, which later be done
3080 	 * with the write to EICR.
3081 	 */
3082 	eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3083 
3084 	/* The lower 16bits of the EICR register are for the queue interrupts
3085 	 * which should be masked here in order to not accidentally clear them if
3086 	 * the bits are high when ixgbe_msix_other is called. There is a race
3087 	 * condition otherwise which results in possible performance loss
3088 	 * especially if the ixgbe_msix_other interrupt is triggering
3089 	 * consistently (as it would when PPS is turned on for the X540 device)
3090 	 */
3091 	eicr &= 0xFFFF0000;
3092 
3093 	IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3094 
3095 	if (eicr & IXGBE_EICR_LSC)
3096 		ixgbe_check_lsc(adapter);
3097 
3098 	if (eicr & IXGBE_EICR_MAILBOX)
3099 		ixgbe_msg_task(adapter);
3100 
3101 	switch (hw->mac.type) {
3102 	case ixgbe_mac_82599EB:
3103 	case ixgbe_mac_X540:
3104 	case ixgbe_mac_X550:
3105 	case ixgbe_mac_X550EM_x:
3106 	case ixgbe_mac_x550em_a:
3107 		if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3108 		    (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3109 			adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3110 			ixgbe_service_event_schedule(adapter);
3111 			IXGBE_WRITE_REG(hw, IXGBE_EICR,
3112 					IXGBE_EICR_GPI_SDP0_X540);
3113 		}
3114 		if (eicr & IXGBE_EICR_ECC) {
3115 			e_info(link, "Received ECC Err, initiating reset\n");
3116 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3117 			ixgbe_service_event_schedule(adapter);
3118 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3119 		}
3120 		/* Handle Flow Director Full threshold interrupt */
3121 		if (eicr & IXGBE_EICR_FLOW_DIR) {
3122 			int reinit_count = 0;
3123 			int i;
3124 			for (i = 0; i < adapter->num_tx_queues; i++) {
3125 				struct ixgbe_ring *ring = adapter->tx_ring[i];
3126 				if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3127 						       &ring->state))
3128 					reinit_count++;
3129 			}
3130 			if (reinit_count) {
3131 				/* no more flow director interrupts until after init */
3132 				IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3133 				adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3134 				ixgbe_service_event_schedule(adapter);
3135 			}
3136 		}
3137 		ixgbe_check_sfp_event(adapter, eicr);
3138 		ixgbe_check_overtemp_event(adapter, eicr);
3139 		break;
3140 	default:
3141 		break;
3142 	}
3143 
3144 	ixgbe_check_fan_failure(adapter, eicr);
3145 
3146 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3147 		ixgbe_ptp_check_pps_event(adapter);
3148 
3149 	/* re-enable the original interrupt state, no lsc, no queues */
3150 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3151 		ixgbe_irq_enable(adapter, false, false);
3152 
3153 	return IRQ_HANDLED;
3154 }
3155 
3156 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3157 {
3158 	struct ixgbe_q_vector *q_vector = data;
3159 
3160 	/* EIAM disabled interrupts (on this vector) for us */
3161 
3162 	if (q_vector->rx.ring || q_vector->tx.ring)
3163 		napi_schedule_irqoff(&q_vector->napi);
3164 
3165 	return IRQ_HANDLED;
3166 }
3167 
3168 /**
3169  * ixgbe_poll - NAPI Rx polling callback
3170  * @napi: structure for representing this polling device
3171  * @budget: how many packets driver is allowed to clean
3172  *
3173  * This function is used for legacy and MSI, NAPI mode
3174  **/
3175 int ixgbe_poll(struct napi_struct *napi, int budget)
3176 {
3177 	struct ixgbe_q_vector *q_vector =
3178 				container_of(napi, struct ixgbe_q_vector, napi);
3179 	struct ixgbe_adapter *adapter = q_vector->adapter;
3180 	struct ixgbe_ring *ring;
3181 	int per_ring_budget, work_done = 0;
3182 	bool clean_complete = true;
3183 
3184 #ifdef CONFIG_IXGBE_DCA
3185 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3186 		ixgbe_update_dca(q_vector);
3187 #endif
3188 
3189 	ixgbe_for_each_ring(ring, q_vector->tx) {
3190 		bool wd = ring->xsk_umem ?
3191 			  ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3192 			  ixgbe_clean_tx_irq(q_vector, ring, budget);
3193 
3194 		if (!wd)
3195 			clean_complete = false;
3196 	}
3197 
3198 	/* Exit if we are called by netpoll */
3199 	if (budget <= 0)
3200 		return budget;
3201 
3202 	/* attempt to distribute budget to each queue fairly, but don't allow
3203 	 * the budget to go below 1 because we'll exit polling */
3204 	if (q_vector->rx.count > 1)
3205 		per_ring_budget = max(budget/q_vector->rx.count, 1);
3206 	else
3207 		per_ring_budget = budget;
3208 
3209 	ixgbe_for_each_ring(ring, q_vector->rx) {
3210 		int cleaned = ring->xsk_umem ?
3211 			      ixgbe_clean_rx_irq_zc(q_vector, ring,
3212 						    per_ring_budget) :
3213 			      ixgbe_clean_rx_irq(q_vector, ring,
3214 						 per_ring_budget);
3215 
3216 		work_done += cleaned;
3217 		if (cleaned >= per_ring_budget)
3218 			clean_complete = false;
3219 	}
3220 
3221 	/* If all work not completed, return budget and keep polling */
3222 	if (!clean_complete)
3223 		return budget;
3224 
3225 	/* all work done, exit the polling mode */
3226 	if (likely(napi_complete_done(napi, work_done))) {
3227 		if (adapter->rx_itr_setting & 1)
3228 			ixgbe_set_itr(q_vector);
3229 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3230 			ixgbe_irq_enable_queues(adapter,
3231 						BIT_ULL(q_vector->v_idx));
3232 	}
3233 
3234 	return min(work_done, budget - 1);
3235 }
3236 
3237 /**
3238  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3239  * @adapter: board private structure
3240  *
3241  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3242  * interrupts from the kernel.
3243  **/
3244 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3245 {
3246 	struct net_device *netdev = adapter->netdev;
3247 	unsigned int ri = 0, ti = 0;
3248 	int vector, err;
3249 
3250 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3251 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3252 		struct msix_entry *entry = &adapter->msix_entries[vector];
3253 
3254 		if (q_vector->tx.ring && q_vector->rx.ring) {
3255 			snprintf(q_vector->name, sizeof(q_vector->name),
3256 				 "%s-TxRx-%u", netdev->name, ri++);
3257 			ti++;
3258 		} else if (q_vector->rx.ring) {
3259 			snprintf(q_vector->name, sizeof(q_vector->name),
3260 				 "%s-rx-%u", netdev->name, ri++);
3261 		} else if (q_vector->tx.ring) {
3262 			snprintf(q_vector->name, sizeof(q_vector->name),
3263 				 "%s-tx-%u", netdev->name, ti++);
3264 		} else {
3265 			/* skip this unused q_vector */
3266 			continue;
3267 		}
3268 		err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3269 				  q_vector->name, q_vector);
3270 		if (err) {
3271 			e_err(probe, "request_irq failed for MSIX interrupt "
3272 			      "Error: %d\n", err);
3273 			goto free_queue_irqs;
3274 		}
3275 		/* If Flow Director is enabled, set interrupt affinity */
3276 		if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3277 			/* assign the mask for this irq */
3278 			irq_set_affinity_hint(entry->vector,
3279 					      &q_vector->affinity_mask);
3280 		}
3281 	}
3282 
3283 	err = request_irq(adapter->msix_entries[vector].vector,
3284 			  ixgbe_msix_other, 0, netdev->name, adapter);
3285 	if (err) {
3286 		e_err(probe, "request_irq for msix_other failed: %d\n", err);
3287 		goto free_queue_irqs;
3288 	}
3289 
3290 	return 0;
3291 
3292 free_queue_irqs:
3293 	while (vector) {
3294 		vector--;
3295 		irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3296 				      NULL);
3297 		free_irq(adapter->msix_entries[vector].vector,
3298 			 adapter->q_vector[vector]);
3299 	}
3300 	adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3301 	pci_disable_msix(adapter->pdev);
3302 	kfree(adapter->msix_entries);
3303 	adapter->msix_entries = NULL;
3304 	return err;
3305 }
3306 
3307 /**
3308  * ixgbe_intr - legacy mode Interrupt Handler
3309  * @irq: interrupt number
3310  * @data: pointer to a network interface device structure
3311  **/
3312 static irqreturn_t ixgbe_intr(int irq, void *data)
3313 {
3314 	struct ixgbe_adapter *adapter = data;
3315 	struct ixgbe_hw *hw = &adapter->hw;
3316 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3317 	u32 eicr;
3318 
3319 	/*
3320 	 * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3321 	 * before the read of EICR.
3322 	 */
3323 	IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3324 
3325 	/* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3326 	 * therefore no explicit interrupt disable is necessary */
3327 	eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3328 	if (!eicr) {
3329 		/*
3330 		 * shared interrupt alert!
3331 		 * make sure interrupts are enabled because the read will
3332 		 * have disabled interrupts due to EIAM
3333 		 * finish the workaround of silicon errata on 82598.  Unmask
3334 		 * the interrupt that we masked before the EICR read.
3335 		 */
3336 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
3337 			ixgbe_irq_enable(adapter, true, true);
3338 		return IRQ_NONE;	/* Not our interrupt */
3339 	}
3340 
3341 	if (eicr & IXGBE_EICR_LSC)
3342 		ixgbe_check_lsc(adapter);
3343 
3344 	switch (hw->mac.type) {
3345 	case ixgbe_mac_82599EB:
3346 		ixgbe_check_sfp_event(adapter, eicr);
3347 		/* Fall through */
3348 	case ixgbe_mac_X540:
3349 	case ixgbe_mac_X550:
3350 	case ixgbe_mac_X550EM_x:
3351 	case ixgbe_mac_x550em_a:
3352 		if (eicr & IXGBE_EICR_ECC) {
3353 			e_info(link, "Received ECC Err, initiating reset\n");
3354 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3355 			ixgbe_service_event_schedule(adapter);
3356 			IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3357 		}
3358 		ixgbe_check_overtemp_event(adapter, eicr);
3359 		break;
3360 	default:
3361 		break;
3362 	}
3363 
3364 	ixgbe_check_fan_failure(adapter, eicr);
3365 	if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3366 		ixgbe_ptp_check_pps_event(adapter);
3367 
3368 	/* would disable interrupts here but EIAM disabled it */
3369 	napi_schedule_irqoff(&q_vector->napi);
3370 
3371 	/*
3372 	 * re-enable link(maybe) and non-queue interrupts, no flush.
3373 	 * ixgbe_poll will re-enable the queue interrupts
3374 	 */
3375 	if (!test_bit(__IXGBE_DOWN, &adapter->state))
3376 		ixgbe_irq_enable(adapter, false, false);
3377 
3378 	return IRQ_HANDLED;
3379 }
3380 
3381 /**
3382  * ixgbe_request_irq - initialize interrupts
3383  * @adapter: board private structure
3384  *
3385  * Attempts to configure interrupts using the best available
3386  * capabilities of the hardware and kernel.
3387  **/
3388 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3389 {
3390 	struct net_device *netdev = adapter->netdev;
3391 	int err;
3392 
3393 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3394 		err = ixgbe_request_msix_irqs(adapter);
3395 	else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3396 		err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3397 				  netdev->name, adapter);
3398 	else
3399 		err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3400 				  netdev->name, adapter);
3401 
3402 	if (err)
3403 		e_err(probe, "request_irq failed, Error %d\n", err);
3404 
3405 	return err;
3406 }
3407 
3408 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3409 {
3410 	int vector;
3411 
3412 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3413 		free_irq(adapter->pdev->irq, adapter);
3414 		return;
3415 	}
3416 
3417 	if (!adapter->msix_entries)
3418 		return;
3419 
3420 	for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3421 		struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3422 		struct msix_entry *entry = &adapter->msix_entries[vector];
3423 
3424 		/* free only the irqs that were actually requested */
3425 		if (!q_vector->rx.ring && !q_vector->tx.ring)
3426 			continue;
3427 
3428 		/* clear the affinity_mask in the IRQ descriptor */
3429 		irq_set_affinity_hint(entry->vector, NULL);
3430 
3431 		free_irq(entry->vector, q_vector);
3432 	}
3433 
3434 	free_irq(adapter->msix_entries[vector].vector, adapter);
3435 }
3436 
3437 /**
3438  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3439  * @adapter: board private structure
3440  **/
3441 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3442 {
3443 	switch (adapter->hw.mac.type) {
3444 	case ixgbe_mac_82598EB:
3445 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3446 		break;
3447 	case ixgbe_mac_82599EB:
3448 	case ixgbe_mac_X540:
3449 	case ixgbe_mac_X550:
3450 	case ixgbe_mac_X550EM_x:
3451 	case ixgbe_mac_x550em_a:
3452 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3453 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3454 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3455 		break;
3456 	default:
3457 		break;
3458 	}
3459 	IXGBE_WRITE_FLUSH(&adapter->hw);
3460 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3461 		int vector;
3462 
3463 		for (vector = 0; vector < adapter->num_q_vectors; vector++)
3464 			synchronize_irq(adapter->msix_entries[vector].vector);
3465 
3466 		synchronize_irq(adapter->msix_entries[vector++].vector);
3467 	} else {
3468 		synchronize_irq(adapter->pdev->irq);
3469 	}
3470 }
3471 
3472 /**
3473  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3474  * @adapter: board private structure
3475  *
3476  **/
3477 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3478 {
3479 	struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3480 
3481 	ixgbe_write_eitr(q_vector);
3482 
3483 	ixgbe_set_ivar(adapter, 0, 0, 0);
3484 	ixgbe_set_ivar(adapter, 1, 0, 0);
3485 
3486 	e_info(hw, "Legacy interrupt IVAR setup done\n");
3487 }
3488 
3489 /**
3490  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3491  * @adapter: board private structure
3492  * @ring: structure containing ring specific data
3493  *
3494  * Configure the Tx descriptor ring after a reset.
3495  **/
3496 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3497 			     struct ixgbe_ring *ring)
3498 {
3499 	struct ixgbe_hw *hw = &adapter->hw;
3500 	u64 tdba = ring->dma;
3501 	int wait_loop = 10;
3502 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
3503 	u8 reg_idx = ring->reg_idx;
3504 
3505 	ring->xsk_umem = NULL;
3506 	if (ring_is_xdp(ring))
3507 		ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3508 
3509 	/* disable queue to avoid issues while updating state */
3510 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3511 	IXGBE_WRITE_FLUSH(hw);
3512 
3513 	IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3514 			(tdba & DMA_BIT_MASK(32)));
3515 	IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3516 	IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3517 			ring->count * sizeof(union ixgbe_adv_tx_desc));
3518 	IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3519 	IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3520 	ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3521 
3522 	/*
3523 	 * set WTHRESH to encourage burst writeback, it should not be set
3524 	 * higher than 1 when:
3525 	 * - ITR is 0 as it could cause false TX hangs
3526 	 * - ITR is set to > 100k int/sec and BQL is enabled
3527 	 *
3528 	 * In order to avoid issues WTHRESH + PTHRESH should always be equal
3529 	 * to or less than the number of on chip descriptors, which is
3530 	 * currently 40.
3531 	 */
3532 	if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3533 		txdctl |= 1u << 16;	/* WTHRESH = 1 */
3534 	else
3535 		txdctl |= 8u << 16;	/* WTHRESH = 8 */
3536 
3537 	/*
3538 	 * Setting PTHRESH to 32 both improves performance
3539 	 * and avoids a TX hang with DFP enabled
3540 	 */
3541 	txdctl |= (1u << 8) |	/* HTHRESH = 1 */
3542 		   32;		/* PTHRESH = 32 */
3543 
3544 	/* reinitialize flowdirector state */
3545 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3546 		ring->atr_sample_rate = adapter->atr_sample_rate;
3547 		ring->atr_count = 0;
3548 		set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3549 	} else {
3550 		ring->atr_sample_rate = 0;
3551 	}
3552 
3553 	/* initialize XPS */
3554 	if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3555 		struct ixgbe_q_vector *q_vector = ring->q_vector;
3556 
3557 		if (q_vector)
3558 			netif_set_xps_queue(ring->netdev,
3559 					    &q_vector->affinity_mask,
3560 					    ring->queue_index);
3561 	}
3562 
3563 	clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3564 
3565 	/* reinitialize tx_buffer_info */
3566 	memset(ring->tx_buffer_info, 0,
3567 	       sizeof(struct ixgbe_tx_buffer) * ring->count);
3568 
3569 	/* enable queue */
3570 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3571 
3572 	/* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3573 	if (hw->mac.type == ixgbe_mac_82598EB &&
3574 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3575 		return;
3576 
3577 	/* poll to verify queue is enabled */
3578 	do {
3579 		usleep_range(1000, 2000);
3580 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3581 	} while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3582 	if (!wait_loop)
3583 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3584 }
3585 
3586 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3587 {
3588 	struct ixgbe_hw *hw = &adapter->hw;
3589 	u32 rttdcs, mtqc;
3590 	u8 tcs = adapter->hw_tcs;
3591 
3592 	if (hw->mac.type == ixgbe_mac_82598EB)
3593 		return;
3594 
3595 	/* disable the arbiter while setting MTQC */
3596 	rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3597 	rttdcs |= IXGBE_RTTDCS_ARBDIS;
3598 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3599 
3600 	/* set transmit pool layout */
3601 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3602 		mtqc = IXGBE_MTQC_VT_ENA;
3603 		if (tcs > 4)
3604 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3605 		else if (tcs > 1)
3606 			mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3607 		else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3608 			 IXGBE_82599_VMDQ_4Q_MASK)
3609 			mtqc |= IXGBE_MTQC_32VF;
3610 		else
3611 			mtqc |= IXGBE_MTQC_64VF;
3612 	} else {
3613 		if (tcs > 4) {
3614 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3615 		} else if (tcs > 1) {
3616 			mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3617 		} else {
3618 			u8 max_txq = adapter->num_tx_queues +
3619 				adapter->num_xdp_queues;
3620 			if (max_txq > 63)
3621 				mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3622 			else
3623 				mtqc = IXGBE_MTQC_64Q_1PB;
3624 		}
3625 	}
3626 
3627 	IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3628 
3629 	/* Enable Security TX Buffer IFG for multiple pb */
3630 	if (tcs) {
3631 		u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3632 		sectx |= IXGBE_SECTX_DCB;
3633 		IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3634 	}
3635 
3636 	/* re-enable the arbiter */
3637 	rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3638 	IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3639 }
3640 
3641 /**
3642  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3643  * @adapter: board private structure
3644  *
3645  * Configure the Tx unit of the MAC after a reset.
3646  **/
3647 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3648 {
3649 	struct ixgbe_hw *hw = &adapter->hw;
3650 	u32 dmatxctl;
3651 	u32 i;
3652 
3653 	ixgbe_setup_mtqc(adapter);
3654 
3655 	if (hw->mac.type != ixgbe_mac_82598EB) {
3656 		/* DMATXCTL.EN must be before Tx queues are enabled */
3657 		dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3658 		dmatxctl |= IXGBE_DMATXCTL_TE;
3659 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3660 	}
3661 
3662 	/* Setup the HW Tx Head and Tail descriptor pointers */
3663 	for (i = 0; i < adapter->num_tx_queues; i++)
3664 		ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3665 	for (i = 0; i < adapter->num_xdp_queues; i++)
3666 		ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3667 }
3668 
3669 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3670 				 struct ixgbe_ring *ring)
3671 {
3672 	struct ixgbe_hw *hw = &adapter->hw;
3673 	u8 reg_idx = ring->reg_idx;
3674 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3675 
3676 	srrctl |= IXGBE_SRRCTL_DROP_EN;
3677 
3678 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3679 }
3680 
3681 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3682 				  struct ixgbe_ring *ring)
3683 {
3684 	struct ixgbe_hw *hw = &adapter->hw;
3685 	u8 reg_idx = ring->reg_idx;
3686 	u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3687 
3688 	srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3689 
3690 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3691 }
3692 
3693 #ifdef CONFIG_IXGBE_DCB
3694 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3695 #else
3696 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3697 #endif
3698 {
3699 	int i;
3700 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3701 
3702 	if (adapter->ixgbe_ieee_pfc)
3703 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3704 
3705 	/*
3706 	 * We should set the drop enable bit if:
3707 	 *  SR-IOV is enabled
3708 	 *   or
3709 	 *  Number of Rx queues > 1 and flow control is disabled
3710 	 *
3711 	 *  This allows us to avoid head of line blocking for security
3712 	 *  and performance reasons.
3713 	 */
3714 	if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3715 	    !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3716 		for (i = 0; i < adapter->num_rx_queues; i++)
3717 			ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3718 	} else {
3719 		for (i = 0; i < adapter->num_rx_queues; i++)
3720 			ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3721 	}
3722 }
3723 
3724 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3725 
3726 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3727 				   struct ixgbe_ring *rx_ring)
3728 {
3729 	struct ixgbe_hw *hw = &adapter->hw;
3730 	u32 srrctl;
3731 	u8 reg_idx = rx_ring->reg_idx;
3732 
3733 	if (hw->mac.type == ixgbe_mac_82598EB) {
3734 		u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3735 
3736 		/*
3737 		 * if VMDq is not active we must program one srrctl register
3738 		 * per RSS queue since we have enabled RDRXCTL.MVMEN
3739 		 */
3740 		reg_idx &= mask;
3741 	}
3742 
3743 	/* configure header buffer length, needed for RSC */
3744 	srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3745 
3746 	/* configure the packet buffer length */
3747 	if (rx_ring->xsk_umem) {
3748 		u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3749 				  XDP_PACKET_HEADROOM;
3750 
3751 		/* If the MAC support setting RXDCTL.RLPML, the
3752 		 * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3753 		 * RXDCTL.RLPML is set to the actual UMEM buffer
3754 		 * size. If not, then we are stuck with a 1k buffer
3755 		 * size resolution. In this case frames larger than
3756 		 * the UMEM buffer size viewed in a 1k resolution will
3757 		 * be dropped.
3758 		 */
3759 		if (hw->mac.type != ixgbe_mac_82599EB)
3760 			srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3761 		else
3762 			srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3763 	} else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3764 		srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3765 	} else {
3766 		srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3767 	}
3768 
3769 	/* configure descriptor type */
3770 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3771 
3772 	IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3773 }
3774 
3775 /**
3776  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3777  * @adapter: device handle
3778  *
3779  *  - 82598/82599/X540:     128
3780  *  - X550(non-SRIOV mode): 512
3781  *  - X550(SRIOV mode):     64
3782  */
3783 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3784 {
3785 	if (adapter->hw.mac.type < ixgbe_mac_X550)
3786 		return 128;
3787 	else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3788 		return 64;
3789 	else
3790 		return 512;
3791 }
3792 
3793 /**
3794  * ixgbe_store_key - Write the RSS key to HW
3795  * @adapter: device handle
3796  *
3797  * Write the RSS key stored in adapter.rss_key to HW.
3798  */
3799 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3800 {
3801 	struct ixgbe_hw *hw = &adapter->hw;
3802 	int i;
3803 
3804 	for (i = 0; i < 10; i++)
3805 		IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3806 }
3807 
3808 /**
3809  * ixgbe_init_rss_key - Initialize adapter RSS key
3810  * @adapter: device handle
3811  *
3812  * Allocates and initializes the RSS key if it is not allocated.
3813  **/
3814 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3815 {
3816 	u32 *rss_key;
3817 
3818 	if (!adapter->rss_key) {
3819 		rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3820 		if (unlikely(!rss_key))
3821 			return -ENOMEM;
3822 
3823 		netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3824 		adapter->rss_key = rss_key;
3825 	}
3826 
3827 	return 0;
3828 }
3829 
3830 /**
3831  * ixgbe_store_reta - Write the RETA table to HW
3832  * @adapter: device handle
3833  *
3834  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3835  */
3836 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3837 {
3838 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3839 	struct ixgbe_hw *hw = &adapter->hw;
3840 	u32 reta = 0;
3841 	u32 indices_multi;
3842 	u8 *indir_tbl = adapter->rss_indir_tbl;
3843 
3844 	/* Fill out the redirection table as follows:
3845 	 *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3846 	 *    indices.
3847 	 *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3848 	 *  - X550:       8 bit wide entries containing 6 bit RSS index
3849 	 */
3850 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3851 		indices_multi = 0x11;
3852 	else
3853 		indices_multi = 0x1;
3854 
3855 	/* Write redirection table to HW */
3856 	for (i = 0; i < reta_entries; i++) {
3857 		reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3858 		if ((i & 3) == 3) {
3859 			if (i < 128)
3860 				IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3861 			else
3862 				IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3863 						reta);
3864 			reta = 0;
3865 		}
3866 	}
3867 }
3868 
3869 /**
3870  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3871  * @adapter: device handle
3872  *
3873  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3874  */
3875 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3876 {
3877 	u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3878 	struct ixgbe_hw *hw = &adapter->hw;
3879 	u32 vfreta = 0;
3880 
3881 	/* Write redirection table to HW */
3882 	for (i = 0; i < reta_entries; i++) {
3883 		u16 pool = adapter->num_rx_pools;
3884 
3885 		vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3886 		if ((i & 3) != 3)
3887 			continue;
3888 
3889 		while (pool--)
3890 			IXGBE_WRITE_REG(hw,
3891 					IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3892 					vfreta);
3893 		vfreta = 0;
3894 	}
3895 }
3896 
3897 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3898 {
3899 	u32 i, j;
3900 	u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3901 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3902 
3903 	/* Program table for at least 4 queues w/ SR-IOV so that VFs can
3904 	 * make full use of any rings they may have.  We will use the
3905 	 * PSRTYPE register to control how many rings we use within the PF.
3906 	 */
3907 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3908 		rss_i = 4;
3909 
3910 	/* Fill out hash function seeds */
3911 	ixgbe_store_key(adapter);
3912 
3913 	/* Fill out redirection table */
3914 	memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3915 
3916 	for (i = 0, j = 0; i < reta_entries; i++, j++) {
3917 		if (j == rss_i)
3918 			j = 0;
3919 
3920 		adapter->rss_indir_tbl[i] = j;
3921 	}
3922 
3923 	ixgbe_store_reta(adapter);
3924 }
3925 
3926 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3927 {
3928 	struct ixgbe_hw *hw = &adapter->hw;
3929 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3930 	int i, j;
3931 
3932 	/* Fill out hash function seeds */
3933 	for (i = 0; i < 10; i++) {
3934 		u16 pool = adapter->num_rx_pools;
3935 
3936 		while (pool--)
3937 			IXGBE_WRITE_REG(hw,
3938 					IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3939 					*(adapter->rss_key + i));
3940 	}
3941 
3942 	/* Fill out the redirection table */
3943 	for (i = 0, j = 0; i < 64; i++, j++) {
3944 		if (j == rss_i)
3945 			j = 0;
3946 
3947 		adapter->rss_indir_tbl[i] = j;
3948 	}
3949 
3950 	ixgbe_store_vfreta(adapter);
3951 }
3952 
3953 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3954 {
3955 	struct ixgbe_hw *hw = &adapter->hw;
3956 	u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3957 	u32 rxcsum;
3958 
3959 	/* Disable indicating checksum in descriptor, enables RSS hash */
3960 	rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3961 	rxcsum |= IXGBE_RXCSUM_PCSD;
3962 	IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3963 
3964 	if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3965 		if (adapter->ring_feature[RING_F_RSS].mask)
3966 			mrqc = IXGBE_MRQC_RSSEN;
3967 	} else {
3968 		u8 tcs = adapter->hw_tcs;
3969 
3970 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3971 			if (tcs > 4)
3972 				mrqc = IXGBE_MRQC_VMDQRT8TCEN;	/* 8 TCs */
3973 			else if (tcs > 1)
3974 				mrqc = IXGBE_MRQC_VMDQRT4TCEN;	/* 4 TCs */
3975 			else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3976 				 IXGBE_82599_VMDQ_4Q_MASK)
3977 				mrqc = IXGBE_MRQC_VMDQRSS32EN;
3978 			else
3979 				mrqc = IXGBE_MRQC_VMDQRSS64EN;
3980 
3981 			/* Enable L3/L4 for Tx Switched packets only for X550,
3982 			 * older devices do not support this feature
3983 			 */
3984 			if (hw->mac.type >= ixgbe_mac_X550)
3985 				mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3986 		} else {
3987 			if (tcs > 4)
3988 				mrqc = IXGBE_MRQC_RTRSS8TCEN;
3989 			else if (tcs > 1)
3990 				mrqc = IXGBE_MRQC_RTRSS4TCEN;
3991 			else
3992 				mrqc = IXGBE_MRQC_RSSEN;
3993 		}
3994 	}
3995 
3996 	/* Perform hash on these packet types */
3997 	rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3998 		     IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3999 		     IXGBE_MRQC_RSS_FIELD_IPV6 |
4000 		     IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
4001 
4002 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
4003 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
4004 	if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
4005 		rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
4006 
4007 	if ((hw->mac.type >= ixgbe_mac_X550) &&
4008 	    (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
4009 		u16 pool = adapter->num_rx_pools;
4010 
4011 		/* Enable VF RSS mode */
4012 		mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
4013 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4014 
4015 		/* Setup RSS through the VF registers */
4016 		ixgbe_setup_vfreta(adapter);
4017 		vfmrqc = IXGBE_MRQC_RSSEN;
4018 		vfmrqc |= rss_field;
4019 
4020 		while (pool--)
4021 			IXGBE_WRITE_REG(hw,
4022 					IXGBE_PFVFMRQC(VMDQ_P(pool)),
4023 					vfmrqc);
4024 	} else {
4025 		ixgbe_setup_reta(adapter);
4026 		mrqc |= rss_field;
4027 		IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4028 	}
4029 }
4030 
4031 /**
4032  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4033  * @adapter: address of board private structure
4034  * @ring: structure containing ring specific data
4035  **/
4036 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4037 				   struct ixgbe_ring *ring)
4038 {
4039 	struct ixgbe_hw *hw = &adapter->hw;
4040 	u32 rscctrl;
4041 	u8 reg_idx = ring->reg_idx;
4042 
4043 	if (!ring_is_rsc_enabled(ring))
4044 		return;
4045 
4046 	rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4047 	rscctrl |= IXGBE_RSCCTL_RSCEN;
4048 	/*
4049 	 * we must limit the number of descriptors so that the
4050 	 * total size of max desc * buf_len is not greater
4051 	 * than 65536
4052 	 */
4053 	rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4054 	IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4055 }
4056 
4057 #define IXGBE_MAX_RX_DESC_POLL 10
4058 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4059 				       struct ixgbe_ring *ring)
4060 {
4061 	struct ixgbe_hw *hw = &adapter->hw;
4062 	int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4063 	u32 rxdctl;
4064 	u8 reg_idx = ring->reg_idx;
4065 
4066 	if (ixgbe_removed(hw->hw_addr))
4067 		return;
4068 	/* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4069 	if (hw->mac.type == ixgbe_mac_82598EB &&
4070 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4071 		return;
4072 
4073 	do {
4074 		usleep_range(1000, 2000);
4075 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4076 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4077 
4078 	if (!wait_loop) {
4079 		e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4080 		      "the polling period\n", reg_idx);
4081 	}
4082 }
4083 
4084 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4085 			     struct ixgbe_ring *ring)
4086 {
4087 	struct ixgbe_hw *hw = &adapter->hw;
4088 	union ixgbe_adv_rx_desc *rx_desc;
4089 	u64 rdba = ring->dma;
4090 	u32 rxdctl;
4091 	u8 reg_idx = ring->reg_idx;
4092 
4093 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4094 	ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4095 	if (ring->xsk_umem) {
4096 		ring->zca.free = ixgbe_zca_free;
4097 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4098 						   MEM_TYPE_ZERO_COPY,
4099 						   &ring->zca));
4100 
4101 	} else {
4102 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4103 						   MEM_TYPE_PAGE_SHARED, NULL));
4104 	}
4105 
4106 	/* disable queue to avoid use of these values while updating state */
4107 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4108 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4109 
4110 	/* write value back with RXDCTL.ENABLE bit cleared */
4111 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4112 	IXGBE_WRITE_FLUSH(hw);
4113 
4114 	IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4115 	IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4116 	IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4117 			ring->count * sizeof(union ixgbe_adv_rx_desc));
4118 	/* Force flushing of IXGBE_RDLEN to prevent MDD */
4119 	IXGBE_WRITE_FLUSH(hw);
4120 
4121 	IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4122 	IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4123 	ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4124 
4125 	ixgbe_configure_srrctl(adapter, ring);
4126 	ixgbe_configure_rscctl(adapter, ring);
4127 
4128 	if (hw->mac.type == ixgbe_mac_82598EB) {
4129 		/*
4130 		 * enable cache line friendly hardware writes:
4131 		 * PTHRESH=32 descriptors (half the internal cache),
4132 		 * this also removes ugly rx_no_buffer_count increment
4133 		 * HTHRESH=4 descriptors (to minimize latency on fetch)
4134 		 * WTHRESH=8 burst writeback up to two cache lines
4135 		 */
4136 		rxdctl &= ~0x3FFFFF;
4137 		rxdctl |=  0x080420;
4138 #if (PAGE_SIZE < 8192)
4139 	/* RXDCTL.RLPML does not work on 82599 */
4140 	} else if (hw->mac.type != ixgbe_mac_82599EB) {
4141 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4142 			    IXGBE_RXDCTL_RLPML_EN);
4143 
4144 		/* Limit the maximum frame size so we don't overrun the skb.
4145 		 * This can happen in SRIOV mode when the MTU of the VF is
4146 		 * higher than the MTU of the PF.
4147 		 */
4148 		if (ring_uses_build_skb(ring) &&
4149 		    !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4150 			rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4151 				  IXGBE_RXDCTL_RLPML_EN;
4152 #endif
4153 	}
4154 
4155 	if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4156 		u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4157 				  XDP_PACKET_HEADROOM;
4158 
4159 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4160 			    IXGBE_RXDCTL_RLPML_EN);
4161 		rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4162 
4163 		ring->rx_buf_len = xsk_buf_len;
4164 	}
4165 
4166 	/* initialize rx_buffer_info */
4167 	memset(ring->rx_buffer_info, 0,
4168 	       sizeof(struct ixgbe_rx_buffer) * ring->count);
4169 
4170 	/* initialize Rx descriptor 0 */
4171 	rx_desc = IXGBE_RX_DESC(ring, 0);
4172 	rx_desc->wb.upper.length = 0;
4173 
4174 	/* enable receive descriptor ring */
4175 	rxdctl |= IXGBE_RXDCTL_ENABLE;
4176 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4177 
4178 	ixgbe_rx_desc_queue_enable(adapter, ring);
4179 	if (ring->xsk_umem)
4180 		ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4181 	else
4182 		ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4183 }
4184 
4185 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4186 {
4187 	struct ixgbe_hw *hw = &adapter->hw;
4188 	int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4189 	u16 pool = adapter->num_rx_pools;
4190 
4191 	/* PSRTYPE must be initialized in non 82598 adapters */
4192 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4193 		      IXGBE_PSRTYPE_UDPHDR |
4194 		      IXGBE_PSRTYPE_IPV4HDR |
4195 		      IXGBE_PSRTYPE_L2HDR |
4196 		      IXGBE_PSRTYPE_IPV6HDR;
4197 
4198 	if (hw->mac.type == ixgbe_mac_82598EB)
4199 		return;
4200 
4201 	if (rss_i > 3)
4202 		psrtype |= 2u << 29;
4203 	else if (rss_i > 1)
4204 		psrtype |= 1u << 29;
4205 
4206 	while (pool--)
4207 		IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4208 }
4209 
4210 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4211 {
4212 	struct ixgbe_hw *hw = &adapter->hw;
4213 	u16 pool = adapter->num_rx_pools;
4214 	u32 reg_offset, vf_shift, vmolr;
4215 	u32 gcr_ext, vmdctl;
4216 	int i;
4217 
4218 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4219 		return;
4220 
4221 	vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4222 	vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4223 	vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4224 	vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4225 	vmdctl |= IXGBE_VT_CTL_REPLEN;
4226 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4227 
4228 	/* accept untagged packets until a vlan tag is
4229 	 * specifically set for the VMDQ queue/pool
4230 	 */
4231 	vmolr = IXGBE_VMOLR_AUPE;
4232 	while (pool--)
4233 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4234 
4235 	vf_shift = VMDQ_P(0) % 32;
4236 	reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4237 
4238 	/* Enable only the PF's pool for Tx/Rx */
4239 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4240 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4241 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4242 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4243 	if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4244 		IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4245 
4246 	/* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4247 	hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4248 
4249 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
4250 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4251 
4252 	/*
4253 	 * Set up VF register offsets for selected VT Mode,
4254 	 * i.e. 32 or 64 VFs for SR-IOV
4255 	 */
4256 	switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4257 	case IXGBE_82599_VMDQ_8Q_MASK:
4258 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4259 		break;
4260 	case IXGBE_82599_VMDQ_4Q_MASK:
4261 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4262 		break;
4263 	default:
4264 		gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4265 		break;
4266 	}
4267 
4268 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4269 
4270 	for (i = 0; i < adapter->num_vfs; i++) {
4271 		/* configure spoof checking */
4272 		ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4273 					  adapter->vfinfo[i].spoofchk_enabled);
4274 
4275 		/* Enable/Disable RSS query feature  */
4276 		ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4277 					  adapter->vfinfo[i].rss_query_enabled);
4278 	}
4279 }
4280 
4281 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4282 {
4283 	struct ixgbe_hw *hw = &adapter->hw;
4284 	struct net_device *netdev = adapter->netdev;
4285 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4286 	struct ixgbe_ring *rx_ring;
4287 	int i;
4288 	u32 mhadd, hlreg0;
4289 
4290 #ifdef IXGBE_FCOE
4291 	/* adjust max frame to be able to do baby jumbo for FCoE */
4292 	if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4293 	    (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4294 		max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4295 
4296 #endif /* IXGBE_FCOE */
4297 
4298 	/* adjust max frame to be at least the size of a standard frame */
4299 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4300 		max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4301 
4302 	mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4303 	if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4304 		mhadd &= ~IXGBE_MHADD_MFS_MASK;
4305 		mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4306 
4307 		IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4308 	}
4309 
4310 	hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4311 	/* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4312 	hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4313 	IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4314 
4315 	/*
4316 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4317 	 * the Base and Length of the Rx Descriptor Ring
4318 	 */
4319 	for (i = 0; i < adapter->num_rx_queues; i++) {
4320 		rx_ring = adapter->rx_ring[i];
4321 
4322 		clear_ring_rsc_enabled(rx_ring);
4323 		clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4324 		clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4325 
4326 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4327 			set_ring_rsc_enabled(rx_ring);
4328 
4329 		if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4330 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4331 
4332 		if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4333 			continue;
4334 
4335 		set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4336 
4337 #if (PAGE_SIZE < 8192)
4338 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4339 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4340 
4341 		if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4342 		    (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4343 			set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4344 #endif
4345 	}
4346 }
4347 
4348 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4349 {
4350 	struct ixgbe_hw *hw = &adapter->hw;
4351 	u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4352 
4353 	switch (hw->mac.type) {
4354 	case ixgbe_mac_82598EB:
4355 		/*
4356 		 * For VMDq support of different descriptor types or
4357 		 * buffer sizes through the use of multiple SRRCTL
4358 		 * registers, RDRXCTL.MVMEN must be set to 1
4359 		 *
4360 		 * also, the manual doesn't mention it clearly but DCA hints
4361 		 * will only use queue 0's tags unless this bit is set.  Side
4362 		 * effects of setting this bit are only that SRRCTL must be
4363 		 * fully programmed [0..15]
4364 		 */
4365 		rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4366 		break;
4367 	case ixgbe_mac_X550:
4368 	case ixgbe_mac_X550EM_x:
4369 	case ixgbe_mac_x550em_a:
4370 		if (adapter->num_vfs)
4371 			rdrxctl |= IXGBE_RDRXCTL_PSP;
4372 		/* fall through */
4373 	case ixgbe_mac_82599EB:
4374 	case ixgbe_mac_X540:
4375 		/* Disable RSC for ACK packets */
4376 		IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4377 		   (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4378 		rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4379 		/* hardware requires some bits to be set by default */
4380 		rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4381 		rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4382 		break;
4383 	default:
4384 		/* We should do nothing since we don't know this hardware */
4385 		return;
4386 	}
4387 
4388 	IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4389 }
4390 
4391 /**
4392  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4393  * @adapter: board private structure
4394  *
4395  * Configure the Rx unit of the MAC after a reset.
4396  **/
4397 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4398 {
4399 	struct ixgbe_hw *hw = &adapter->hw;
4400 	int i;
4401 	u32 rxctrl, rfctl;
4402 
4403 	/* disable receives while setting up the descriptors */
4404 	hw->mac.ops.disable_rx(hw);
4405 
4406 	ixgbe_setup_psrtype(adapter);
4407 	ixgbe_setup_rdrxctl(adapter);
4408 
4409 	/* RSC Setup */
4410 	rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4411 	rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4412 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4413 		rfctl |= IXGBE_RFCTL_RSC_DIS;
4414 
4415 	/* disable NFS filtering */
4416 	rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4417 	IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4418 
4419 	/* Program registers for the distribution of queues */
4420 	ixgbe_setup_mrqc(adapter);
4421 
4422 	/* set_rx_buffer_len must be called before ring initialization */
4423 	ixgbe_set_rx_buffer_len(adapter);
4424 
4425 	/*
4426 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
4427 	 * the Base and Length of the Rx Descriptor Ring
4428 	 */
4429 	for (i = 0; i < adapter->num_rx_queues; i++)
4430 		ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4431 
4432 	rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4433 	/* disable drop enable for 82598 parts */
4434 	if (hw->mac.type == ixgbe_mac_82598EB)
4435 		rxctrl |= IXGBE_RXCTRL_DMBYPS;
4436 
4437 	/* enable all receives */
4438 	rxctrl |= IXGBE_RXCTRL_RXEN;
4439 	hw->mac.ops.enable_rx_dma(hw, rxctrl);
4440 }
4441 
4442 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4443 				 __be16 proto, u16 vid)
4444 {
4445 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4446 	struct ixgbe_hw *hw = &adapter->hw;
4447 
4448 	/* add VID to filter table */
4449 	if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4450 		hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4451 
4452 	set_bit(vid, adapter->active_vlans);
4453 
4454 	return 0;
4455 }
4456 
4457 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4458 {
4459 	u32 vlvf;
4460 	int idx;
4461 
4462 	/* short cut the special case */
4463 	if (vlan == 0)
4464 		return 0;
4465 
4466 	/* Search for the vlan id in the VLVF entries */
4467 	for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4468 		vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4469 		if ((vlvf & VLAN_VID_MASK) == vlan)
4470 			break;
4471 	}
4472 
4473 	return idx;
4474 }
4475 
4476 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4477 {
4478 	struct ixgbe_hw *hw = &adapter->hw;
4479 	u32 bits, word;
4480 	int idx;
4481 
4482 	idx = ixgbe_find_vlvf_entry(hw, vid);
4483 	if (!idx)
4484 		return;
4485 
4486 	/* See if any other pools are set for this VLAN filter
4487 	 * entry other than the PF.
4488 	 */
4489 	word = idx * 2 + (VMDQ_P(0) / 32);
4490 	bits = ~BIT(VMDQ_P(0) % 32);
4491 	bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4492 
4493 	/* Disable the filter so this falls into the default pool. */
4494 	if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4495 		if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4496 			IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4497 		IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4498 	}
4499 }
4500 
4501 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4502 				  __be16 proto, u16 vid)
4503 {
4504 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4505 	struct ixgbe_hw *hw = &adapter->hw;
4506 
4507 	/* remove VID from filter table */
4508 	if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4509 		hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4510 
4511 	clear_bit(vid, adapter->active_vlans);
4512 
4513 	return 0;
4514 }
4515 
4516 /**
4517  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4518  * @adapter: driver data
4519  */
4520 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4521 {
4522 	struct ixgbe_hw *hw = &adapter->hw;
4523 	u32 vlnctrl;
4524 	int i, j;
4525 
4526 	switch (hw->mac.type) {
4527 	case ixgbe_mac_82598EB:
4528 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4529 		vlnctrl &= ~IXGBE_VLNCTRL_VME;
4530 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4531 		break;
4532 	case ixgbe_mac_82599EB:
4533 	case ixgbe_mac_X540:
4534 	case ixgbe_mac_X550:
4535 	case ixgbe_mac_X550EM_x:
4536 	case ixgbe_mac_x550em_a:
4537 		for (i = 0; i < adapter->num_rx_queues; i++) {
4538 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4539 
4540 			if (!netif_is_ixgbe(ring->netdev))
4541 				continue;
4542 
4543 			j = ring->reg_idx;
4544 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4545 			vlnctrl &= ~IXGBE_RXDCTL_VME;
4546 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4547 		}
4548 		break;
4549 	default:
4550 		break;
4551 	}
4552 }
4553 
4554 /**
4555  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4556  * @adapter: driver data
4557  */
4558 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4559 {
4560 	struct ixgbe_hw *hw = &adapter->hw;
4561 	u32 vlnctrl;
4562 	int i, j;
4563 
4564 	switch (hw->mac.type) {
4565 	case ixgbe_mac_82598EB:
4566 		vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4567 		vlnctrl |= IXGBE_VLNCTRL_VME;
4568 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4569 		break;
4570 	case ixgbe_mac_82599EB:
4571 	case ixgbe_mac_X540:
4572 	case ixgbe_mac_X550:
4573 	case ixgbe_mac_X550EM_x:
4574 	case ixgbe_mac_x550em_a:
4575 		for (i = 0; i < adapter->num_rx_queues; i++) {
4576 			struct ixgbe_ring *ring = adapter->rx_ring[i];
4577 
4578 			if (!netif_is_ixgbe(ring->netdev))
4579 				continue;
4580 
4581 			j = ring->reg_idx;
4582 			vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4583 			vlnctrl |= IXGBE_RXDCTL_VME;
4584 			IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4585 		}
4586 		break;
4587 	default:
4588 		break;
4589 	}
4590 }
4591 
4592 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4593 {
4594 	struct ixgbe_hw *hw = &adapter->hw;
4595 	u32 vlnctrl, i;
4596 
4597 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4598 
4599 	if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4600 	/* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4601 		vlnctrl |= IXGBE_VLNCTRL_VFE;
4602 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4603 	} else {
4604 		vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4605 		IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4606 		return;
4607 	}
4608 
4609 	/* Nothing to do for 82598 */
4610 	if (hw->mac.type == ixgbe_mac_82598EB)
4611 		return;
4612 
4613 	/* We are already in VLAN promisc, nothing to do */
4614 	if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4615 		return;
4616 
4617 	/* Set flag so we don't redo unnecessary work */
4618 	adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4619 
4620 	/* Add PF to all active pools */
4621 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4622 		u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4623 		u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4624 
4625 		vlvfb |= BIT(VMDQ_P(0) % 32);
4626 		IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4627 	}
4628 
4629 	/* Set all bits in the VLAN filter table array */
4630 	for (i = hw->mac.vft_size; i--;)
4631 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4632 }
4633 
4634 #define VFTA_BLOCK_SIZE 8
4635 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4636 {
4637 	struct ixgbe_hw *hw = &adapter->hw;
4638 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4639 	u32 vid_start = vfta_offset * 32;
4640 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4641 	u32 i, vid, word, bits;
4642 
4643 	for (i = IXGBE_VLVF_ENTRIES; --i;) {
4644 		u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4645 
4646 		/* pull VLAN ID from VLVF */
4647 		vid = vlvf & VLAN_VID_MASK;
4648 
4649 		/* only concern outselves with a certain range */
4650 		if (vid < vid_start || vid >= vid_end)
4651 			continue;
4652 
4653 		if (vlvf) {
4654 			/* record VLAN ID in VFTA */
4655 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4656 
4657 			/* if PF is part of this then continue */
4658 			if (test_bit(vid, adapter->active_vlans))
4659 				continue;
4660 		}
4661 
4662 		/* remove PF from the pool */
4663 		word = i * 2 + VMDQ_P(0) / 32;
4664 		bits = ~BIT(VMDQ_P(0) % 32);
4665 		bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4666 		IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4667 	}
4668 
4669 	/* extract values from active_vlans and write back to VFTA */
4670 	for (i = VFTA_BLOCK_SIZE; i--;) {
4671 		vid = (vfta_offset + i) * 32;
4672 		word = vid / BITS_PER_LONG;
4673 		bits = vid % BITS_PER_LONG;
4674 
4675 		vfta[i] |= adapter->active_vlans[word] >> bits;
4676 
4677 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4678 	}
4679 }
4680 
4681 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4682 {
4683 	struct ixgbe_hw *hw = &adapter->hw;
4684 	u32 vlnctrl, i;
4685 
4686 	/* Set VLAN filtering to enabled */
4687 	vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4688 	vlnctrl |= IXGBE_VLNCTRL_VFE;
4689 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4690 
4691 	if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4692 	    hw->mac.type == ixgbe_mac_82598EB)
4693 		return;
4694 
4695 	/* We are not in VLAN promisc, nothing to do */
4696 	if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4697 		return;
4698 
4699 	/* Set flag so we don't redo unnecessary work */
4700 	adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4701 
4702 	for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4703 		ixgbe_scrub_vfta(adapter, i);
4704 }
4705 
4706 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4707 {
4708 	u16 vid = 1;
4709 
4710 	ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4711 
4712 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4713 		ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4714 }
4715 
4716 /**
4717  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4718  * @netdev: network interface device structure
4719  *
4720  * Writes multicast address list to the MTA hash table.
4721  * Returns: -ENOMEM on failure
4722  *                0 on no addresses written
4723  *                X on writing X addresses to MTA
4724  **/
4725 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4726 {
4727 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4728 	struct ixgbe_hw *hw = &adapter->hw;
4729 
4730 	if (!netif_running(netdev))
4731 		return 0;
4732 
4733 	if (hw->mac.ops.update_mc_addr_list)
4734 		hw->mac.ops.update_mc_addr_list(hw, netdev);
4735 	else
4736 		return -ENOMEM;
4737 
4738 #ifdef CONFIG_PCI_IOV
4739 	ixgbe_restore_vf_multicasts(adapter);
4740 #endif
4741 
4742 	return netdev_mc_count(netdev);
4743 }
4744 
4745 #ifdef CONFIG_PCI_IOV
4746 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4747 {
4748 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4749 	struct ixgbe_hw *hw = &adapter->hw;
4750 	int i;
4751 
4752 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4753 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4754 
4755 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4756 			hw->mac.ops.set_rar(hw, i,
4757 					    mac_table->addr,
4758 					    mac_table->pool,
4759 					    IXGBE_RAH_AV);
4760 		else
4761 			hw->mac.ops.clear_rar(hw, i);
4762 	}
4763 }
4764 
4765 #endif
4766 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4767 {
4768 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4769 	struct ixgbe_hw *hw = &adapter->hw;
4770 	int i;
4771 
4772 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4773 		if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4774 			continue;
4775 
4776 		mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4777 
4778 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4779 			hw->mac.ops.set_rar(hw, i,
4780 					    mac_table->addr,
4781 					    mac_table->pool,
4782 					    IXGBE_RAH_AV);
4783 		else
4784 			hw->mac.ops.clear_rar(hw, i);
4785 	}
4786 }
4787 
4788 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4789 {
4790 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4791 	struct ixgbe_hw *hw = &adapter->hw;
4792 	int i;
4793 
4794 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4795 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4796 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4797 	}
4798 
4799 	ixgbe_sync_mac_table(adapter);
4800 }
4801 
4802 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4803 {
4804 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4805 	struct ixgbe_hw *hw = &adapter->hw;
4806 	int i, count = 0;
4807 
4808 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4809 		/* do not count default RAR as available */
4810 		if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4811 			continue;
4812 
4813 		/* only count unused and addresses that belong to us */
4814 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4815 			if (mac_table->pool != pool)
4816 				continue;
4817 		}
4818 
4819 		count++;
4820 	}
4821 
4822 	return count;
4823 }
4824 
4825 /* this function destroys the first RAR entry */
4826 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4827 {
4828 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4829 	struct ixgbe_hw *hw = &adapter->hw;
4830 
4831 	memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4832 	mac_table->pool = VMDQ_P(0);
4833 
4834 	mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4835 
4836 	hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4837 			    IXGBE_RAH_AV);
4838 }
4839 
4840 int ixgbe_add_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 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4851 		if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4852 			continue;
4853 
4854 		ether_addr_copy(mac_table->addr, addr);
4855 		mac_table->pool = pool;
4856 
4857 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4858 				    IXGBE_MAC_STATE_IN_USE;
4859 
4860 		ixgbe_sync_mac_table(adapter);
4861 
4862 		return i;
4863 	}
4864 
4865 	return -ENOMEM;
4866 }
4867 
4868 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4869 			 const u8 *addr, u16 pool)
4870 {
4871 	struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4872 	struct ixgbe_hw *hw = &adapter->hw;
4873 	int i;
4874 
4875 	if (is_zero_ether_addr(addr))
4876 		return -EINVAL;
4877 
4878 	/* search table for addr, if found clear IN_USE flag and sync */
4879 	for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4880 		/* we can only delete an entry if it is in use */
4881 		if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4882 			continue;
4883 		/* we only care about entries that belong to the given pool */
4884 		if (mac_table->pool != pool)
4885 			continue;
4886 		/* we only care about a specific MAC address */
4887 		if (!ether_addr_equal(addr, mac_table->addr))
4888 			continue;
4889 
4890 		mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4891 		mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4892 
4893 		ixgbe_sync_mac_table(adapter);
4894 
4895 		return 0;
4896 	}
4897 
4898 	return -ENOMEM;
4899 }
4900 
4901 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4902 {
4903 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4904 	int ret;
4905 
4906 	ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4907 
4908 	return min_t(int, ret, 0);
4909 }
4910 
4911 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4912 {
4913 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4914 
4915 	ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4916 
4917 	return 0;
4918 }
4919 
4920 /**
4921  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4922  * @netdev: network interface device structure
4923  *
4924  * The set_rx_method entry point is called whenever the unicast/multicast
4925  * address list or the network interface flags are updated.  This routine is
4926  * responsible for configuring the hardware for proper unicast, multicast and
4927  * promiscuous mode.
4928  **/
4929 void ixgbe_set_rx_mode(struct net_device *netdev)
4930 {
4931 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
4932 	struct ixgbe_hw *hw = &adapter->hw;
4933 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4934 	netdev_features_t features = netdev->features;
4935 	int count;
4936 
4937 	/* Check for Promiscuous and All Multicast modes */
4938 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4939 
4940 	/* set all bits that we expect to always be set */
4941 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4942 	fctrl |= IXGBE_FCTRL_BAM;
4943 	fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4944 	fctrl |= IXGBE_FCTRL_PMCF;
4945 
4946 	/* clear the bits we are changing the status of */
4947 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4948 	if (netdev->flags & IFF_PROMISC) {
4949 		hw->addr_ctrl.user_set_promisc = true;
4950 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4951 		vmolr |= IXGBE_VMOLR_MPE;
4952 		features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4953 	} else {
4954 		if (netdev->flags & IFF_ALLMULTI) {
4955 			fctrl |= IXGBE_FCTRL_MPE;
4956 			vmolr |= IXGBE_VMOLR_MPE;
4957 		}
4958 		hw->addr_ctrl.user_set_promisc = false;
4959 	}
4960 
4961 	/*
4962 	 * Write addresses to available RAR registers, if there is not
4963 	 * sufficient space to store all the addresses then enable
4964 	 * unicast promiscuous mode
4965 	 */
4966 	if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4967 		fctrl |= IXGBE_FCTRL_UPE;
4968 		vmolr |= IXGBE_VMOLR_ROPE;
4969 	}
4970 
4971 	/* Write addresses to the MTA, if the attempt fails
4972 	 * then we should just turn on promiscuous mode so
4973 	 * that we can at least receive multicast traffic
4974 	 */
4975 	count = ixgbe_write_mc_addr_list(netdev);
4976 	if (count < 0) {
4977 		fctrl |= IXGBE_FCTRL_MPE;
4978 		vmolr |= IXGBE_VMOLR_MPE;
4979 	} else if (count) {
4980 		vmolr |= IXGBE_VMOLR_ROMPE;
4981 	}
4982 
4983 	if (hw->mac.type != ixgbe_mac_82598EB) {
4984 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4985 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4986 			   IXGBE_VMOLR_ROPE);
4987 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4988 	}
4989 
4990 	/* This is useful for sniffing bad packets. */
4991 	if (features & NETIF_F_RXALL) {
4992 		/* UPE and MPE will be handled by normal PROMISC logic
4993 		 * in e1000e_set_rx_mode */
4994 		fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4995 			  IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4996 			  IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4997 
4998 		fctrl &= ~(IXGBE_FCTRL_DPF);
4999 		/* NOTE:  VLAN filtering is disabled by setting PROMISC */
5000 	}
5001 
5002 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5003 
5004 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
5005 		ixgbe_vlan_strip_enable(adapter);
5006 	else
5007 		ixgbe_vlan_strip_disable(adapter);
5008 
5009 	if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
5010 		ixgbe_vlan_promisc_disable(adapter);
5011 	else
5012 		ixgbe_vlan_promisc_enable(adapter);
5013 }
5014 
5015 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
5016 {
5017 	int q_idx;
5018 
5019 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5020 		napi_enable(&adapter->q_vector[q_idx]->napi);
5021 }
5022 
5023 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5024 {
5025 	int q_idx;
5026 
5027 	for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5028 		napi_disable(&adapter->q_vector[q_idx]->napi);
5029 }
5030 
5031 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5032 {
5033 	struct ixgbe_hw *hw = &adapter->hw;
5034 	u32 vxlanctrl;
5035 
5036 	if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5037 				IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5038 		return;
5039 
5040 	vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5041 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5042 
5043 	if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5044 		adapter->vxlan_port = 0;
5045 
5046 	if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5047 		adapter->geneve_port = 0;
5048 }
5049 
5050 #ifdef CONFIG_IXGBE_DCB
5051 /**
5052  * ixgbe_configure_dcb - Configure DCB hardware
5053  * @adapter: ixgbe adapter struct
5054  *
5055  * This is called by the driver on open to configure the DCB hardware.
5056  * This is also called by the gennetlink interface when reconfiguring
5057  * the DCB state.
5058  */
5059 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5060 {
5061 	struct ixgbe_hw *hw = &adapter->hw;
5062 	int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5063 
5064 	if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5065 		if (hw->mac.type == ixgbe_mac_82598EB)
5066 			netif_set_gso_max_size(adapter->netdev, 65536);
5067 		return;
5068 	}
5069 
5070 	if (hw->mac.type == ixgbe_mac_82598EB)
5071 		netif_set_gso_max_size(adapter->netdev, 32768);
5072 
5073 #ifdef IXGBE_FCOE
5074 	if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5075 		max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5076 #endif
5077 
5078 	/* reconfigure the hardware */
5079 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5080 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5081 						DCB_TX_CONFIG);
5082 		ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5083 						DCB_RX_CONFIG);
5084 		ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5085 	} else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5086 		ixgbe_dcb_hw_ets(&adapter->hw,
5087 				 adapter->ixgbe_ieee_ets,
5088 				 max_frame);
5089 		ixgbe_dcb_hw_pfc_config(&adapter->hw,
5090 					adapter->ixgbe_ieee_pfc->pfc_en,
5091 					adapter->ixgbe_ieee_ets->prio_tc);
5092 	}
5093 
5094 	/* Enable RSS Hash per TC */
5095 	if (hw->mac.type != ixgbe_mac_82598EB) {
5096 		u32 msb = 0;
5097 		u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5098 
5099 		while (rss_i) {
5100 			msb++;
5101 			rss_i >>= 1;
5102 		}
5103 
5104 		/* write msb to all 8 TCs in one write */
5105 		IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5106 	}
5107 }
5108 #endif
5109 
5110 /* Additional bittime to account for IXGBE framing */
5111 #define IXGBE_ETH_FRAMING 20
5112 
5113 /**
5114  * ixgbe_hpbthresh - calculate high water mark for flow control
5115  *
5116  * @adapter: board private structure to calculate for
5117  * @pb: packet buffer to calculate
5118  */
5119 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5120 {
5121 	struct ixgbe_hw *hw = &adapter->hw;
5122 	struct net_device *dev = adapter->netdev;
5123 	int link, tc, kb, marker;
5124 	u32 dv_id, rx_pba;
5125 
5126 	/* Calculate max LAN frame size */
5127 	tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5128 
5129 #ifdef IXGBE_FCOE
5130 	/* FCoE traffic class uses FCOE jumbo frames */
5131 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5132 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5133 	    (pb == ixgbe_fcoe_get_tc(adapter)))
5134 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5135 #endif
5136 
5137 	/* Calculate delay value for device */
5138 	switch (hw->mac.type) {
5139 	case ixgbe_mac_X540:
5140 	case ixgbe_mac_X550:
5141 	case ixgbe_mac_X550EM_x:
5142 	case ixgbe_mac_x550em_a:
5143 		dv_id = IXGBE_DV_X540(link, tc);
5144 		break;
5145 	default:
5146 		dv_id = IXGBE_DV(link, tc);
5147 		break;
5148 	}
5149 
5150 	/* Loopback switch introduces additional latency */
5151 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5152 		dv_id += IXGBE_B2BT(tc);
5153 
5154 	/* Delay value is calculated in bit times convert to KB */
5155 	kb = IXGBE_BT2KB(dv_id);
5156 	rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5157 
5158 	marker = rx_pba - kb;
5159 
5160 	/* It is possible that the packet buffer is not large enough
5161 	 * to provide required headroom. In this case throw an error
5162 	 * to user and a do the best we can.
5163 	 */
5164 	if (marker < 0) {
5165 		e_warn(drv, "Packet Buffer(%i) can not provide enough"
5166 			    "headroom to support flow control."
5167 			    "Decrease MTU or number of traffic classes\n", pb);
5168 		marker = tc + 1;
5169 	}
5170 
5171 	return marker;
5172 }
5173 
5174 /**
5175  * ixgbe_lpbthresh - calculate low water mark for for flow control
5176  *
5177  * @adapter: board private structure to calculate for
5178  * @pb: packet buffer to calculate
5179  */
5180 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5181 {
5182 	struct ixgbe_hw *hw = &adapter->hw;
5183 	struct net_device *dev = adapter->netdev;
5184 	int tc;
5185 	u32 dv_id;
5186 
5187 	/* Calculate max LAN frame size */
5188 	tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5189 
5190 #ifdef IXGBE_FCOE
5191 	/* FCoE traffic class uses FCOE jumbo frames */
5192 	if ((dev->features & NETIF_F_FCOE_MTU) &&
5193 	    (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5194 	    (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5195 		tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5196 #endif
5197 
5198 	/* Calculate delay value for device */
5199 	switch (hw->mac.type) {
5200 	case ixgbe_mac_X540:
5201 	case ixgbe_mac_X550:
5202 	case ixgbe_mac_X550EM_x:
5203 	case ixgbe_mac_x550em_a:
5204 		dv_id = IXGBE_LOW_DV_X540(tc);
5205 		break;
5206 	default:
5207 		dv_id = IXGBE_LOW_DV(tc);
5208 		break;
5209 	}
5210 
5211 	/* Delay value is calculated in bit times convert to KB */
5212 	return IXGBE_BT2KB(dv_id);
5213 }
5214 
5215 /*
5216  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5217  */
5218 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5219 {
5220 	struct ixgbe_hw *hw = &adapter->hw;
5221 	int num_tc = adapter->hw_tcs;
5222 	int i;
5223 
5224 	if (!num_tc)
5225 		num_tc = 1;
5226 
5227 	for (i = 0; i < num_tc; i++) {
5228 		hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5229 		hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5230 
5231 		/* Low water marks must not be larger than high water marks */
5232 		if (hw->fc.low_water[i] > hw->fc.high_water[i])
5233 			hw->fc.low_water[i] = 0;
5234 	}
5235 
5236 	for (; i < MAX_TRAFFIC_CLASS; i++)
5237 		hw->fc.high_water[i] = 0;
5238 }
5239 
5240 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5241 {
5242 	struct ixgbe_hw *hw = &adapter->hw;
5243 	int hdrm;
5244 	u8 tc = adapter->hw_tcs;
5245 
5246 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5247 	    adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5248 		hdrm = 32 << adapter->fdir_pballoc;
5249 	else
5250 		hdrm = 0;
5251 
5252 	hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5253 	ixgbe_pbthresh_setup(adapter);
5254 }
5255 
5256 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5257 {
5258 	struct ixgbe_hw *hw = &adapter->hw;
5259 	struct hlist_node *node2;
5260 	struct ixgbe_fdir_filter *filter;
5261 	u8 queue;
5262 
5263 	spin_lock(&adapter->fdir_perfect_lock);
5264 
5265 	if (!hlist_empty(&adapter->fdir_filter_list))
5266 		ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5267 
5268 	hlist_for_each_entry_safe(filter, node2,
5269 				  &adapter->fdir_filter_list, fdir_node) {
5270 		if (filter->action == IXGBE_FDIR_DROP_QUEUE) {
5271 			queue = IXGBE_FDIR_DROP_QUEUE;
5272 		} else {
5273 			u32 ring = ethtool_get_flow_spec_ring(filter->action);
5274 			u8 vf = ethtool_get_flow_spec_ring_vf(filter->action);
5275 
5276 			if (!vf && (ring >= adapter->num_rx_queues)) {
5277 				e_err(drv, "FDIR restore failed without VF, ring: %u\n",
5278 				      ring);
5279 				continue;
5280 			} else if (vf &&
5281 				   ((vf > adapter->num_vfs) ||
5282 				     ring >= adapter->num_rx_queues_per_pool)) {
5283 				e_err(drv, "FDIR restore failed with VF, vf: %hhu, ring: %u\n",
5284 				      vf, ring);
5285 				continue;
5286 			}
5287 
5288 			/* Map the ring onto the absolute queue index */
5289 			if (!vf)
5290 				queue = adapter->rx_ring[ring]->reg_idx;
5291 			else
5292 				queue = ((vf - 1) *
5293 					adapter->num_rx_queues_per_pool) + ring;
5294 		}
5295 
5296 		ixgbe_fdir_write_perfect_filter_82599(hw,
5297 				&filter->filter, filter->sw_idx, queue);
5298 	}
5299 
5300 	spin_unlock(&adapter->fdir_perfect_lock);
5301 }
5302 
5303 /**
5304  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5305  * @rx_ring: ring to free buffers from
5306  **/
5307 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5308 {
5309 	u16 i = rx_ring->next_to_clean;
5310 	struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5311 
5312 	if (rx_ring->xsk_umem) {
5313 		ixgbe_xsk_clean_rx_ring(rx_ring);
5314 		goto skip_free;
5315 	}
5316 
5317 	/* Free all the Rx ring sk_buffs */
5318 	while (i != rx_ring->next_to_alloc) {
5319 		if (rx_buffer->skb) {
5320 			struct sk_buff *skb = rx_buffer->skb;
5321 			if (IXGBE_CB(skb)->page_released)
5322 				dma_unmap_page_attrs(rx_ring->dev,
5323 						     IXGBE_CB(skb)->dma,
5324 						     ixgbe_rx_pg_size(rx_ring),
5325 						     DMA_FROM_DEVICE,
5326 						     IXGBE_RX_DMA_ATTR);
5327 			dev_kfree_skb(skb);
5328 		}
5329 
5330 		/* Invalidate cache lines that may have been written to by
5331 		 * device so that we avoid corrupting memory.
5332 		 */
5333 		dma_sync_single_range_for_cpu(rx_ring->dev,
5334 					      rx_buffer->dma,
5335 					      rx_buffer->page_offset,
5336 					      ixgbe_rx_bufsz(rx_ring),
5337 					      DMA_FROM_DEVICE);
5338 
5339 		/* free resources associated with mapping */
5340 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5341 				     ixgbe_rx_pg_size(rx_ring),
5342 				     DMA_FROM_DEVICE,
5343 				     IXGBE_RX_DMA_ATTR);
5344 		__page_frag_cache_drain(rx_buffer->page,
5345 					rx_buffer->pagecnt_bias);
5346 
5347 		i++;
5348 		rx_buffer++;
5349 		if (i == rx_ring->count) {
5350 			i = 0;
5351 			rx_buffer = rx_ring->rx_buffer_info;
5352 		}
5353 	}
5354 
5355 skip_free:
5356 	rx_ring->next_to_alloc = 0;
5357 	rx_ring->next_to_clean = 0;
5358 	rx_ring->next_to_use = 0;
5359 }
5360 
5361 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5362 			     struct ixgbe_fwd_adapter *accel)
5363 {
5364 	u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5365 	int num_tc = netdev_get_num_tc(adapter->netdev);
5366 	struct net_device *vdev = accel->netdev;
5367 	int i, baseq, err;
5368 
5369 	baseq = accel->pool * adapter->num_rx_queues_per_pool;
5370 	netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5371 		   accel->pool, adapter->num_rx_pools,
5372 		   baseq, baseq + adapter->num_rx_queues_per_pool);
5373 
5374 	accel->rx_base_queue = baseq;
5375 	accel->tx_base_queue = baseq;
5376 
5377 	/* record configuration for macvlan interface in vdev */
5378 	for (i = 0; i < num_tc; i++)
5379 		netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5380 					     i, rss_i, baseq + (rss_i * i));
5381 
5382 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5383 		adapter->rx_ring[baseq + i]->netdev = vdev;
5384 
5385 	/* Guarantee all rings are updated before we update the
5386 	 * MAC address filter.
5387 	 */
5388 	wmb();
5389 
5390 	/* ixgbe_add_mac_filter will return an index if it succeeds, so we
5391 	 * need to only treat it as an error value if it is negative.
5392 	 */
5393 	err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5394 				   VMDQ_P(accel->pool));
5395 	if (err >= 0)
5396 		return 0;
5397 
5398 	/* if we cannot add the MAC rule then disable the offload */
5399 	macvlan_release_l2fw_offload(vdev);
5400 
5401 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5402 		adapter->rx_ring[baseq + i]->netdev = NULL;
5403 
5404 	netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5405 
5406 	/* unbind the queues and drop the subordinate channel config */
5407 	netdev_unbind_sb_channel(adapter->netdev, vdev);
5408 	netdev_set_sb_channel(vdev, 0);
5409 
5410 	clear_bit(accel->pool, adapter->fwd_bitmask);
5411 	kfree(accel);
5412 
5413 	return err;
5414 }
5415 
5416 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5417 {
5418 	struct ixgbe_adapter *adapter = data;
5419 	struct ixgbe_fwd_adapter *accel;
5420 
5421 	if (!netif_is_macvlan(vdev))
5422 		return 0;
5423 
5424 	accel = macvlan_accel_priv(vdev);
5425 	if (!accel)
5426 		return 0;
5427 
5428 	ixgbe_fwd_ring_up(adapter, accel);
5429 
5430 	return 0;
5431 }
5432 
5433 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5434 {
5435 	netdev_walk_all_upper_dev_rcu(adapter->netdev,
5436 				      ixgbe_macvlan_up, adapter);
5437 }
5438 
5439 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5440 {
5441 	struct ixgbe_hw *hw = &adapter->hw;
5442 
5443 	ixgbe_configure_pb(adapter);
5444 #ifdef CONFIG_IXGBE_DCB
5445 	ixgbe_configure_dcb(adapter);
5446 #endif
5447 	/*
5448 	 * We must restore virtualization before VLANs or else
5449 	 * the VLVF registers will not be populated
5450 	 */
5451 	ixgbe_configure_virtualization(adapter);
5452 
5453 	ixgbe_set_rx_mode(adapter->netdev);
5454 	ixgbe_restore_vlan(adapter);
5455 	ixgbe_ipsec_restore(adapter);
5456 
5457 	switch (hw->mac.type) {
5458 	case ixgbe_mac_82599EB:
5459 	case ixgbe_mac_X540:
5460 		hw->mac.ops.disable_rx_buff(hw);
5461 		break;
5462 	default:
5463 		break;
5464 	}
5465 
5466 	if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5467 		ixgbe_init_fdir_signature_82599(&adapter->hw,
5468 						adapter->fdir_pballoc);
5469 	} else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5470 		ixgbe_init_fdir_perfect_82599(&adapter->hw,
5471 					      adapter->fdir_pballoc);
5472 		ixgbe_fdir_filter_restore(adapter);
5473 	}
5474 
5475 	switch (hw->mac.type) {
5476 	case ixgbe_mac_82599EB:
5477 	case ixgbe_mac_X540:
5478 		hw->mac.ops.enable_rx_buff(hw);
5479 		break;
5480 	default:
5481 		break;
5482 	}
5483 
5484 #ifdef CONFIG_IXGBE_DCA
5485 	/* configure DCA */
5486 	if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5487 		ixgbe_setup_dca(adapter);
5488 #endif /* CONFIG_IXGBE_DCA */
5489 
5490 #ifdef IXGBE_FCOE
5491 	/* configure FCoE L2 filters, redirection table, and Rx control */
5492 	ixgbe_configure_fcoe(adapter);
5493 
5494 #endif /* IXGBE_FCOE */
5495 	ixgbe_configure_tx(adapter);
5496 	ixgbe_configure_rx(adapter);
5497 	ixgbe_configure_dfwd(adapter);
5498 }
5499 
5500 /**
5501  * ixgbe_sfp_link_config - set up SFP+ link
5502  * @adapter: pointer to private adapter struct
5503  **/
5504 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5505 {
5506 	/*
5507 	 * We are assuming the worst case scenario here, and that
5508 	 * is that an SFP was inserted/removed after the reset
5509 	 * but before SFP detection was enabled.  As such the best
5510 	 * solution is to just start searching as soon as we start
5511 	 */
5512 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5513 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5514 
5515 	adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5516 	adapter->sfp_poll_time = 0;
5517 }
5518 
5519 /**
5520  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5521  * @hw: pointer to private hardware struct
5522  *
5523  * Returns 0 on success, negative on failure
5524  **/
5525 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5526 {
5527 	u32 speed;
5528 	bool autoneg, link_up = false;
5529 	int ret = IXGBE_ERR_LINK_SETUP;
5530 
5531 	if (hw->mac.ops.check_link)
5532 		ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5533 
5534 	if (ret)
5535 		return ret;
5536 
5537 	speed = hw->phy.autoneg_advertised;
5538 	if ((!speed) && (hw->mac.ops.get_link_capabilities))
5539 		ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5540 							&autoneg);
5541 	if (ret)
5542 		return ret;
5543 
5544 	if (hw->mac.ops.setup_link)
5545 		ret = hw->mac.ops.setup_link(hw, speed, link_up);
5546 
5547 	return ret;
5548 }
5549 
5550 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5551 {
5552 	struct ixgbe_hw *hw = &adapter->hw;
5553 	u32 gpie = 0;
5554 
5555 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5556 		gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5557 		       IXGBE_GPIE_OCD;
5558 		gpie |= IXGBE_GPIE_EIAME;
5559 		/*
5560 		 * use EIAM to auto-mask when MSI-X interrupt is asserted
5561 		 * this saves a register write for every interrupt
5562 		 */
5563 		switch (hw->mac.type) {
5564 		case ixgbe_mac_82598EB:
5565 			IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5566 			break;
5567 		case ixgbe_mac_82599EB:
5568 		case ixgbe_mac_X540:
5569 		case ixgbe_mac_X550:
5570 		case ixgbe_mac_X550EM_x:
5571 		case ixgbe_mac_x550em_a:
5572 		default:
5573 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5574 			IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5575 			break;
5576 		}
5577 	} else {
5578 		/* legacy interrupts, use EIAM to auto-mask when reading EICR,
5579 		 * specifically only auto mask tx and rx interrupts */
5580 		IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5581 	}
5582 
5583 	/* XXX: to interrupt immediately for EICS writes, enable this */
5584 	/* gpie |= IXGBE_GPIE_EIMEN; */
5585 
5586 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5587 		gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5588 
5589 		switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5590 		case IXGBE_82599_VMDQ_8Q_MASK:
5591 			gpie |= IXGBE_GPIE_VTMODE_16;
5592 			break;
5593 		case IXGBE_82599_VMDQ_4Q_MASK:
5594 			gpie |= IXGBE_GPIE_VTMODE_32;
5595 			break;
5596 		default:
5597 			gpie |= IXGBE_GPIE_VTMODE_64;
5598 			break;
5599 		}
5600 	}
5601 
5602 	/* Enable Thermal over heat sensor interrupt */
5603 	if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5604 		switch (adapter->hw.mac.type) {
5605 		case ixgbe_mac_82599EB:
5606 			gpie |= IXGBE_SDP0_GPIEN_8259X;
5607 			break;
5608 		default:
5609 			break;
5610 		}
5611 	}
5612 
5613 	/* Enable fan failure interrupt */
5614 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5615 		gpie |= IXGBE_SDP1_GPIEN(hw);
5616 
5617 	switch (hw->mac.type) {
5618 	case ixgbe_mac_82599EB:
5619 		gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5620 		break;
5621 	case ixgbe_mac_X550EM_x:
5622 	case ixgbe_mac_x550em_a:
5623 		gpie |= IXGBE_SDP0_GPIEN_X540;
5624 		break;
5625 	default:
5626 		break;
5627 	}
5628 
5629 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5630 }
5631 
5632 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5633 {
5634 	struct ixgbe_hw *hw = &adapter->hw;
5635 	int err;
5636 	u32 ctrl_ext;
5637 
5638 	ixgbe_get_hw_control(adapter);
5639 	ixgbe_setup_gpie(adapter);
5640 
5641 	if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5642 		ixgbe_configure_msix(adapter);
5643 	else
5644 		ixgbe_configure_msi_and_legacy(adapter);
5645 
5646 	/* enable the optics for 82599 SFP+ fiber */
5647 	if (hw->mac.ops.enable_tx_laser)
5648 		hw->mac.ops.enable_tx_laser(hw);
5649 
5650 	if (hw->phy.ops.set_phy_power)
5651 		hw->phy.ops.set_phy_power(hw, true);
5652 
5653 	smp_mb__before_atomic();
5654 	clear_bit(__IXGBE_DOWN, &adapter->state);
5655 	ixgbe_napi_enable_all(adapter);
5656 
5657 	if (ixgbe_is_sfp(hw)) {
5658 		ixgbe_sfp_link_config(adapter);
5659 	} else {
5660 		err = ixgbe_non_sfp_link_config(hw);
5661 		if (err)
5662 			e_err(probe, "link_config FAILED %d\n", err);
5663 	}
5664 
5665 	/* clear any pending interrupts, may auto mask */
5666 	IXGBE_READ_REG(hw, IXGBE_EICR);
5667 	ixgbe_irq_enable(adapter, true, true);
5668 
5669 	/*
5670 	 * If this adapter has a fan, check to see if we had a failure
5671 	 * before we enabled the interrupt.
5672 	 */
5673 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5674 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5675 		if (esdp & IXGBE_ESDP_SDP1)
5676 			e_crit(drv, "Fan has stopped, replace the adapter\n");
5677 	}
5678 
5679 	/* bring the link up in the watchdog, this could race with our first
5680 	 * link up interrupt but shouldn't be a problem */
5681 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5682 	adapter->link_check_timeout = jiffies;
5683 	mod_timer(&adapter->service_timer, jiffies);
5684 
5685 	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
5686 	ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5687 	ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5688 	IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5689 }
5690 
5691 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5692 {
5693 	WARN_ON(in_interrupt());
5694 	/* put off any impending NetWatchDogTimeout */
5695 	netif_trans_update(adapter->netdev);
5696 
5697 	while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5698 		usleep_range(1000, 2000);
5699 	if (adapter->hw.phy.type == ixgbe_phy_fw)
5700 		ixgbe_watchdog_link_is_down(adapter);
5701 	ixgbe_down(adapter);
5702 	/*
5703 	 * If SR-IOV enabled then wait a bit before bringing the adapter
5704 	 * back up to give the VFs time to respond to the reset.  The
5705 	 * two second wait is based upon the watchdog timer cycle in
5706 	 * the VF driver.
5707 	 */
5708 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5709 		msleep(2000);
5710 	ixgbe_up(adapter);
5711 	clear_bit(__IXGBE_RESETTING, &adapter->state);
5712 }
5713 
5714 void ixgbe_up(struct ixgbe_adapter *adapter)
5715 {
5716 	/* hardware has been reset, we need to reload some things */
5717 	ixgbe_configure(adapter);
5718 
5719 	ixgbe_up_complete(adapter);
5720 }
5721 
5722 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5723 {
5724 	u16 devctl2;
5725 
5726 	pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5727 
5728 	switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5729 	case IXGBE_PCIDEVCTRL2_17_34s:
5730 	case IXGBE_PCIDEVCTRL2_4_8s:
5731 		/* For now we cap the upper limit on delay to 2 seconds
5732 		 * as we end up going up to 34 seconds of delay in worst
5733 		 * case timeout value.
5734 		 */
5735 	case IXGBE_PCIDEVCTRL2_1_2s:
5736 		return 2000000ul;	/* 2.0 s */
5737 	case IXGBE_PCIDEVCTRL2_260_520ms:
5738 		return 520000ul;	/* 520 ms */
5739 	case IXGBE_PCIDEVCTRL2_65_130ms:
5740 		return 130000ul;	/* 130 ms */
5741 	case IXGBE_PCIDEVCTRL2_16_32ms:
5742 		return 32000ul;		/* 32 ms */
5743 	case IXGBE_PCIDEVCTRL2_1_2ms:
5744 		return 2000ul;		/* 2 ms */
5745 	case IXGBE_PCIDEVCTRL2_50_100us:
5746 		return 100ul;		/* 100 us */
5747 	case IXGBE_PCIDEVCTRL2_16_32ms_def:
5748 		return 32000ul;		/* 32 ms */
5749 	default:
5750 		break;
5751 	}
5752 
5753 	/* We shouldn't need to hit this path, but just in case default as
5754 	 * though completion timeout is not supported and support 32ms.
5755 	 */
5756 	return 32000ul;
5757 }
5758 
5759 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5760 {
5761 	unsigned long wait_delay, delay_interval;
5762 	struct ixgbe_hw *hw = &adapter->hw;
5763 	int i, wait_loop;
5764 	u32 rxdctl;
5765 
5766 	/* disable receives */
5767 	hw->mac.ops.disable_rx(hw);
5768 
5769 	if (ixgbe_removed(hw->hw_addr))
5770 		return;
5771 
5772 	/* disable all enabled Rx queues */
5773 	for (i = 0; i < adapter->num_rx_queues; i++) {
5774 		struct ixgbe_ring *ring = adapter->rx_ring[i];
5775 		u8 reg_idx = ring->reg_idx;
5776 
5777 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5778 		rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5779 		rxdctl |= IXGBE_RXDCTL_SWFLSH;
5780 
5781 		/* write value back with RXDCTL.ENABLE bit cleared */
5782 		IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5783 	}
5784 
5785 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5786 	if (hw->mac.type == ixgbe_mac_82598EB &&
5787 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5788 		return;
5789 
5790 	/* Determine our minimum delay interval. We will increase this value
5791 	 * with each subsequent test. This way if the device returns quickly
5792 	 * we should spend as little time as possible waiting, however as
5793 	 * the time increases we will wait for larger periods of time.
5794 	 *
5795 	 * The trick here is that we increase the interval using the
5796 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5797 	 * of that wait is that it totals up to 100x whatever interval we
5798 	 * choose. Since our minimum wait is 100us we can just divide the
5799 	 * total timeout by 100 to get our minimum delay interval.
5800 	 */
5801 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5802 
5803 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5804 	wait_delay = delay_interval;
5805 
5806 	while (wait_loop--) {
5807 		usleep_range(wait_delay, wait_delay + 10);
5808 		wait_delay += delay_interval * 2;
5809 		rxdctl = 0;
5810 
5811 		/* OR together the reading of all the active RXDCTL registers,
5812 		 * and then test the result. We need the disable to complete
5813 		 * before we start freeing the memory and invalidating the
5814 		 * DMA mappings.
5815 		 */
5816 		for (i = 0; i < adapter->num_rx_queues; i++) {
5817 			struct ixgbe_ring *ring = adapter->rx_ring[i];
5818 			u8 reg_idx = ring->reg_idx;
5819 
5820 			rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5821 		}
5822 
5823 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5824 			return;
5825 	}
5826 
5827 	e_err(drv,
5828 	      "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5829 }
5830 
5831 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5832 {
5833 	unsigned long wait_delay, delay_interval;
5834 	struct ixgbe_hw *hw = &adapter->hw;
5835 	int i, wait_loop;
5836 	u32 txdctl;
5837 
5838 	if (ixgbe_removed(hw->hw_addr))
5839 		return;
5840 
5841 	/* disable all enabled Tx queues */
5842 	for (i = 0; i < adapter->num_tx_queues; i++) {
5843 		struct ixgbe_ring *ring = adapter->tx_ring[i];
5844 		u8 reg_idx = ring->reg_idx;
5845 
5846 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5847 	}
5848 
5849 	/* disable all enabled XDP Tx queues */
5850 	for (i = 0; i < adapter->num_xdp_queues; i++) {
5851 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
5852 		u8 reg_idx = ring->reg_idx;
5853 
5854 		IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5855 	}
5856 
5857 	/* If the link is not up there shouldn't be much in the way of
5858 	 * pending transactions. Those that are left will be flushed out
5859 	 * when the reset logic goes through the flush sequence to clean out
5860 	 * the pending Tx transactions.
5861 	 */
5862 	if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5863 		goto dma_engine_disable;
5864 
5865 	/* Determine our minimum delay interval. We will increase this value
5866 	 * with each subsequent test. This way if the device returns quickly
5867 	 * we should spend as little time as possible waiting, however as
5868 	 * the time increases we will wait for larger periods of time.
5869 	 *
5870 	 * The trick here is that we increase the interval using the
5871 	 * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5872 	 * of that wait is that it totals up to 100x whatever interval we
5873 	 * choose. Since our minimum wait is 100us we can just divide the
5874 	 * total timeout by 100 to get our minimum delay interval.
5875 	 */
5876 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5877 
5878 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
5879 	wait_delay = delay_interval;
5880 
5881 	while (wait_loop--) {
5882 		usleep_range(wait_delay, wait_delay + 10);
5883 		wait_delay += delay_interval * 2;
5884 		txdctl = 0;
5885 
5886 		/* OR together the reading of all the active TXDCTL registers,
5887 		 * and then test the result. We need the disable to complete
5888 		 * before we start freeing the memory and invalidating the
5889 		 * DMA mappings.
5890 		 */
5891 		for (i = 0; i < adapter->num_tx_queues; i++) {
5892 			struct ixgbe_ring *ring = adapter->tx_ring[i];
5893 			u8 reg_idx = ring->reg_idx;
5894 
5895 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5896 		}
5897 		for (i = 0; i < adapter->num_xdp_queues; i++) {
5898 			struct ixgbe_ring *ring = adapter->xdp_ring[i];
5899 			u8 reg_idx = ring->reg_idx;
5900 
5901 			txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5902 		}
5903 
5904 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5905 			goto dma_engine_disable;
5906 	}
5907 
5908 	e_err(drv,
5909 	      "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5910 
5911 dma_engine_disable:
5912 	/* Disable the Tx DMA engine on 82599 and later MAC */
5913 	switch (hw->mac.type) {
5914 	case ixgbe_mac_82599EB:
5915 	case ixgbe_mac_X540:
5916 	case ixgbe_mac_X550:
5917 	case ixgbe_mac_X550EM_x:
5918 	case ixgbe_mac_x550em_a:
5919 		IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5920 				(IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5921 				 ~IXGBE_DMATXCTL_TE));
5922 		/* fall through */
5923 	default:
5924 		break;
5925 	}
5926 }
5927 
5928 void ixgbe_reset(struct ixgbe_adapter *adapter)
5929 {
5930 	struct ixgbe_hw *hw = &adapter->hw;
5931 	struct net_device *netdev = adapter->netdev;
5932 	int err;
5933 
5934 	if (ixgbe_removed(hw->hw_addr))
5935 		return;
5936 	/* lock SFP init bit to prevent race conditions with the watchdog */
5937 	while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5938 		usleep_range(1000, 2000);
5939 
5940 	/* clear all SFP and link config related flags while holding SFP_INIT */
5941 	adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5942 			     IXGBE_FLAG2_SFP_NEEDS_RESET);
5943 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5944 
5945 	err = hw->mac.ops.init_hw(hw);
5946 	switch (err) {
5947 	case 0:
5948 	case IXGBE_ERR_SFP_NOT_PRESENT:
5949 	case IXGBE_ERR_SFP_NOT_SUPPORTED:
5950 		break;
5951 	case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5952 		e_dev_err("master disable timed out\n");
5953 		break;
5954 	case IXGBE_ERR_EEPROM_VERSION:
5955 		/* We are running on a pre-production device, log a warning */
5956 		e_dev_warn("This device is a pre-production adapter/LOM. "
5957 			   "Please be aware there may be issues associated with "
5958 			   "your hardware.  If you are experiencing problems "
5959 			   "please contact your Intel or hardware "
5960 			   "representative who provided you with this "
5961 			   "hardware.\n");
5962 		break;
5963 	default:
5964 		e_dev_err("Hardware Error: %d\n", err);
5965 	}
5966 
5967 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5968 
5969 	/* flush entries out of MAC table */
5970 	ixgbe_flush_sw_mac_table(adapter);
5971 	__dev_uc_unsync(netdev, NULL);
5972 
5973 	/* do not flush user set addresses */
5974 	ixgbe_mac_set_default_filter(adapter);
5975 
5976 	/* update SAN MAC vmdq pool selection */
5977 	if (hw->mac.san_mac_rar_index)
5978 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5979 
5980 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5981 		ixgbe_ptp_reset(adapter);
5982 
5983 	if (hw->phy.ops.set_phy_power) {
5984 		if (!netif_running(adapter->netdev) && !adapter->wol)
5985 			hw->phy.ops.set_phy_power(hw, false);
5986 		else
5987 			hw->phy.ops.set_phy_power(hw, true);
5988 	}
5989 }
5990 
5991 /**
5992  * ixgbe_clean_tx_ring - Free Tx Buffers
5993  * @tx_ring: ring to be cleaned
5994  **/
5995 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5996 {
5997 	u16 i = tx_ring->next_to_clean;
5998 	struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5999 
6000 	if (tx_ring->xsk_umem) {
6001 		ixgbe_xsk_clean_tx_ring(tx_ring);
6002 		goto out;
6003 	}
6004 
6005 	while (i != tx_ring->next_to_use) {
6006 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
6007 
6008 		/* Free all the Tx ring sk_buffs */
6009 		if (ring_is_xdp(tx_ring))
6010 			xdp_return_frame(tx_buffer->xdpf);
6011 		else
6012 			dev_kfree_skb_any(tx_buffer->skb);
6013 
6014 		/* unmap skb header data */
6015 		dma_unmap_single(tx_ring->dev,
6016 				 dma_unmap_addr(tx_buffer, dma),
6017 				 dma_unmap_len(tx_buffer, len),
6018 				 DMA_TO_DEVICE);
6019 
6020 		/* check for eop_desc to determine the end of the packet */
6021 		eop_desc = tx_buffer->next_to_watch;
6022 		tx_desc = IXGBE_TX_DESC(tx_ring, i);
6023 
6024 		/* unmap remaining buffers */
6025 		while (tx_desc != eop_desc) {
6026 			tx_buffer++;
6027 			tx_desc++;
6028 			i++;
6029 			if (unlikely(i == tx_ring->count)) {
6030 				i = 0;
6031 				tx_buffer = tx_ring->tx_buffer_info;
6032 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
6033 			}
6034 
6035 			/* unmap any remaining paged data */
6036 			if (dma_unmap_len(tx_buffer, len))
6037 				dma_unmap_page(tx_ring->dev,
6038 					       dma_unmap_addr(tx_buffer, dma),
6039 					       dma_unmap_len(tx_buffer, len),
6040 					       DMA_TO_DEVICE);
6041 		}
6042 
6043 		/* move us one more past the eop_desc for start of next pkt */
6044 		tx_buffer++;
6045 		i++;
6046 		if (unlikely(i == tx_ring->count)) {
6047 			i = 0;
6048 			tx_buffer = tx_ring->tx_buffer_info;
6049 		}
6050 	}
6051 
6052 	/* reset BQL for queue */
6053 	if (!ring_is_xdp(tx_ring))
6054 		netdev_tx_reset_queue(txring_txq(tx_ring));
6055 
6056 out:
6057 	/* reset next_to_use and next_to_clean */
6058 	tx_ring->next_to_use = 0;
6059 	tx_ring->next_to_clean = 0;
6060 }
6061 
6062 /**
6063  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6064  * @adapter: board private structure
6065  **/
6066 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6067 {
6068 	int i;
6069 
6070 	for (i = 0; i < adapter->num_rx_queues; i++)
6071 		ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6072 }
6073 
6074 /**
6075  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6076  * @adapter: board private structure
6077  **/
6078 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6079 {
6080 	int i;
6081 
6082 	for (i = 0; i < adapter->num_tx_queues; i++)
6083 		ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6084 	for (i = 0; i < adapter->num_xdp_queues; i++)
6085 		ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6086 }
6087 
6088 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6089 {
6090 	struct hlist_node *node2;
6091 	struct ixgbe_fdir_filter *filter;
6092 
6093 	spin_lock(&adapter->fdir_perfect_lock);
6094 
6095 	hlist_for_each_entry_safe(filter, node2,
6096 				  &adapter->fdir_filter_list, fdir_node) {
6097 		hlist_del(&filter->fdir_node);
6098 		kfree(filter);
6099 	}
6100 	adapter->fdir_filter_count = 0;
6101 
6102 	spin_unlock(&adapter->fdir_perfect_lock);
6103 }
6104 
6105 void ixgbe_down(struct ixgbe_adapter *adapter)
6106 {
6107 	struct net_device *netdev = adapter->netdev;
6108 	struct ixgbe_hw *hw = &adapter->hw;
6109 	int i;
6110 
6111 	/* signal that we are down to the interrupt handler */
6112 	if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6113 		return; /* do nothing if already down */
6114 
6115 	/* Shut off incoming Tx traffic */
6116 	netif_tx_stop_all_queues(netdev);
6117 
6118 	/* call carrier off first to avoid false dev_watchdog timeouts */
6119 	netif_carrier_off(netdev);
6120 	netif_tx_disable(netdev);
6121 
6122 	/* Disable Rx */
6123 	ixgbe_disable_rx(adapter);
6124 
6125 	/* synchronize_rcu() needed for pending XDP buffers to drain */
6126 	if (adapter->xdp_ring[0])
6127 		synchronize_rcu();
6128 
6129 	ixgbe_irq_disable(adapter);
6130 
6131 	ixgbe_napi_disable_all(adapter);
6132 
6133 	clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6134 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6135 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6136 
6137 	del_timer_sync(&adapter->service_timer);
6138 
6139 	if (adapter->num_vfs) {
6140 		/* Clear EITR Select mapping */
6141 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6142 
6143 		/* Mark all the VFs as inactive */
6144 		for (i = 0 ; i < adapter->num_vfs; i++)
6145 			adapter->vfinfo[i].clear_to_send = false;
6146 
6147 		/* ping all the active vfs to let them know we are going down */
6148 		ixgbe_ping_all_vfs(adapter);
6149 
6150 		/* Disable all VFTE/VFRE TX/RX */
6151 		ixgbe_disable_tx_rx(adapter);
6152 	}
6153 
6154 	/* disable transmits in the hardware now that interrupts are off */
6155 	ixgbe_disable_tx(adapter);
6156 
6157 	if (!pci_channel_offline(adapter->pdev))
6158 		ixgbe_reset(adapter);
6159 
6160 	/* power down the optics for 82599 SFP+ fiber */
6161 	if (hw->mac.ops.disable_tx_laser)
6162 		hw->mac.ops.disable_tx_laser(hw);
6163 
6164 	ixgbe_clean_all_tx_rings(adapter);
6165 	ixgbe_clean_all_rx_rings(adapter);
6166 }
6167 
6168 /**
6169  * ixgbe_eee_capable - helper function to determine EEE support on X550
6170  * @adapter: board private structure
6171  */
6172 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6173 {
6174 	struct ixgbe_hw *hw = &adapter->hw;
6175 
6176 	switch (hw->device_id) {
6177 	case IXGBE_DEV_ID_X550EM_A_1G_T:
6178 	case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6179 		if (!hw->phy.eee_speeds_supported)
6180 			break;
6181 		adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6182 		if (!hw->phy.eee_speeds_advertised)
6183 			break;
6184 		adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6185 		break;
6186 	default:
6187 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6188 		adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6189 		break;
6190 	}
6191 }
6192 
6193 /**
6194  * ixgbe_tx_timeout - Respond to a Tx Hang
6195  * @netdev: network interface device structure
6196  **/
6197 static void ixgbe_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6198 {
6199 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6200 
6201 	/* Do the reset outside of interrupt context */
6202 	ixgbe_tx_timeout_reset(adapter);
6203 }
6204 
6205 #ifdef CONFIG_IXGBE_DCB
6206 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6207 {
6208 	struct ixgbe_hw *hw = &adapter->hw;
6209 	struct tc_configuration *tc;
6210 	int j;
6211 
6212 	switch (hw->mac.type) {
6213 	case ixgbe_mac_82598EB:
6214 	case ixgbe_mac_82599EB:
6215 		adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6216 		adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6217 		break;
6218 	case ixgbe_mac_X540:
6219 	case ixgbe_mac_X550:
6220 		adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6221 		adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6222 		break;
6223 	case ixgbe_mac_X550EM_x:
6224 	case ixgbe_mac_x550em_a:
6225 	default:
6226 		adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6227 		adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6228 		break;
6229 	}
6230 
6231 	/* Configure DCB traffic classes */
6232 	for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6233 		tc = &adapter->dcb_cfg.tc_config[j];
6234 		tc->path[DCB_TX_CONFIG].bwg_id = 0;
6235 		tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6236 		tc->path[DCB_RX_CONFIG].bwg_id = 0;
6237 		tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6238 		tc->dcb_pfc = pfc_disabled;
6239 	}
6240 
6241 	/* Initialize default user to priority mapping, UPx->TC0 */
6242 	tc = &adapter->dcb_cfg.tc_config[0];
6243 	tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6244 	tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6245 
6246 	adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6247 	adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6248 	adapter->dcb_cfg.pfc_mode_enable = false;
6249 	adapter->dcb_set_bitmap = 0x00;
6250 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6251 		adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6252 	memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6253 	       sizeof(adapter->temp_dcb_cfg));
6254 }
6255 #endif
6256 
6257 /**
6258  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6259  * @adapter: board private structure to initialize
6260  * @ii: pointer to ixgbe_info for device
6261  *
6262  * ixgbe_sw_init initializes the Adapter private data structure.
6263  * Fields are initialized based on PCI device information and
6264  * OS network device settings (MTU size).
6265  **/
6266 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6267 			 const struct ixgbe_info *ii)
6268 {
6269 	struct ixgbe_hw *hw = &adapter->hw;
6270 	struct pci_dev *pdev = adapter->pdev;
6271 	unsigned int rss, fdir;
6272 	u32 fwsm;
6273 	int i;
6274 
6275 	/* PCI config space info */
6276 
6277 	hw->vendor_id = pdev->vendor;
6278 	hw->device_id = pdev->device;
6279 	hw->revision_id = pdev->revision;
6280 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6281 	hw->subsystem_device_id = pdev->subsystem_device;
6282 
6283 	/* get_invariants needs the device IDs */
6284 	ii->get_invariants(hw);
6285 
6286 	/* Set common capability flags and settings */
6287 	rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6288 	adapter->ring_feature[RING_F_RSS].limit = rss;
6289 	adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6290 	adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6291 	adapter->atr_sample_rate = 20;
6292 	fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6293 	adapter->ring_feature[RING_F_FDIR].limit = fdir;
6294 	adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6295 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
6296 #ifdef CONFIG_IXGBE_DCA
6297 	adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6298 #endif
6299 #ifdef CONFIG_IXGBE_DCB
6300 	adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6301 	adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6302 #endif
6303 #ifdef IXGBE_FCOE
6304 	adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6305 	adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6306 #ifdef CONFIG_IXGBE_DCB
6307 	/* Default traffic class to use for FCoE */
6308 	adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6309 #endif /* CONFIG_IXGBE_DCB */
6310 #endif /* IXGBE_FCOE */
6311 
6312 	/* initialize static ixgbe jump table entries */
6313 	adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6314 					  GFP_KERNEL);
6315 	if (!adapter->jump_tables[0])
6316 		return -ENOMEM;
6317 	adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6318 
6319 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6320 		adapter->jump_tables[i] = NULL;
6321 
6322 	adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6323 				     sizeof(struct ixgbe_mac_addr),
6324 				     GFP_KERNEL);
6325 	if (!adapter->mac_table)
6326 		return -ENOMEM;
6327 
6328 	if (ixgbe_init_rss_key(adapter))
6329 		return -ENOMEM;
6330 
6331 	adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6332 	if (!adapter->af_xdp_zc_qps)
6333 		return -ENOMEM;
6334 
6335 	/* Set MAC specific capability flags and exceptions */
6336 	switch (hw->mac.type) {
6337 	case ixgbe_mac_82598EB:
6338 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6339 
6340 		if (hw->device_id == IXGBE_DEV_ID_82598AT)
6341 			adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6342 
6343 		adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6344 		adapter->ring_feature[RING_F_FDIR].limit = 0;
6345 		adapter->atr_sample_rate = 0;
6346 		adapter->fdir_pballoc = 0;
6347 #ifdef IXGBE_FCOE
6348 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6349 		adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6350 #ifdef CONFIG_IXGBE_DCB
6351 		adapter->fcoe.up = 0;
6352 #endif /* IXGBE_DCB */
6353 #endif /* IXGBE_FCOE */
6354 		break;
6355 	case ixgbe_mac_82599EB:
6356 		if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6357 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6358 		break;
6359 	case ixgbe_mac_X540:
6360 		fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6361 		if (fwsm & IXGBE_FWSM_TS_ENABLED)
6362 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6363 		break;
6364 	case ixgbe_mac_x550em_a:
6365 		adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6366 		switch (hw->device_id) {
6367 		case IXGBE_DEV_ID_X550EM_A_1G_T:
6368 		case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6369 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6370 			break;
6371 		default:
6372 			break;
6373 		}
6374 	/* fall through */
6375 	case ixgbe_mac_X550EM_x:
6376 #ifdef CONFIG_IXGBE_DCB
6377 		adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6378 #endif
6379 #ifdef IXGBE_FCOE
6380 		adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6381 #ifdef CONFIG_IXGBE_DCB
6382 		adapter->fcoe.up = 0;
6383 #endif /* IXGBE_DCB */
6384 #endif /* IXGBE_FCOE */
6385 	/* Fall Through */
6386 	case ixgbe_mac_X550:
6387 		if (hw->mac.type == ixgbe_mac_X550)
6388 			adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6389 #ifdef CONFIG_IXGBE_DCA
6390 		adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6391 #endif
6392 		adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6393 		break;
6394 	default:
6395 		break;
6396 	}
6397 
6398 #ifdef IXGBE_FCOE
6399 	/* FCoE support exists, always init the FCoE lock */
6400 	spin_lock_init(&adapter->fcoe.lock);
6401 
6402 #endif
6403 	/* n-tuple support exists, always init our spinlock */
6404 	spin_lock_init(&adapter->fdir_perfect_lock);
6405 
6406 #ifdef CONFIG_IXGBE_DCB
6407 	ixgbe_init_dcb(adapter);
6408 #endif
6409 	ixgbe_init_ipsec_offload(adapter);
6410 
6411 	/* default flow control settings */
6412 	hw->fc.requested_mode = ixgbe_fc_full;
6413 	hw->fc.current_mode = ixgbe_fc_full;	/* init for ethtool output */
6414 	ixgbe_pbthresh_setup(adapter);
6415 	hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6416 	hw->fc.send_xon = true;
6417 	hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6418 
6419 #ifdef CONFIG_PCI_IOV
6420 	if (max_vfs > 0)
6421 		e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6422 
6423 	/* assign number of SR-IOV VFs */
6424 	if (hw->mac.type != ixgbe_mac_82598EB) {
6425 		if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6426 			max_vfs = 0;
6427 			e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6428 		}
6429 	}
6430 #endif /* CONFIG_PCI_IOV */
6431 
6432 	/* enable itr by default in dynamic mode */
6433 	adapter->rx_itr_setting = 1;
6434 	adapter->tx_itr_setting = 1;
6435 
6436 	/* set default ring sizes */
6437 	adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6438 	adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6439 
6440 	/* set default work limits */
6441 	adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6442 
6443 	/* initialize eeprom parameters */
6444 	if (ixgbe_init_eeprom_params_generic(hw)) {
6445 		e_dev_err("EEPROM initialization failed\n");
6446 		return -EIO;
6447 	}
6448 
6449 	/* PF holds first pool slot */
6450 	set_bit(0, adapter->fwd_bitmask);
6451 	set_bit(__IXGBE_DOWN, &adapter->state);
6452 
6453 	return 0;
6454 }
6455 
6456 /**
6457  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6458  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6459  *
6460  * Return 0 on success, negative on failure
6461  **/
6462 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6463 {
6464 	struct device *dev = tx_ring->dev;
6465 	int orig_node = dev_to_node(dev);
6466 	int ring_node = NUMA_NO_NODE;
6467 	int size;
6468 
6469 	size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6470 
6471 	if (tx_ring->q_vector)
6472 		ring_node = tx_ring->q_vector->numa_node;
6473 
6474 	tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6475 	if (!tx_ring->tx_buffer_info)
6476 		tx_ring->tx_buffer_info = vmalloc(size);
6477 	if (!tx_ring->tx_buffer_info)
6478 		goto err;
6479 
6480 	/* round up to nearest 4K */
6481 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6482 	tx_ring->size = ALIGN(tx_ring->size, 4096);
6483 
6484 	set_dev_node(dev, ring_node);
6485 	tx_ring->desc = dma_alloc_coherent(dev,
6486 					   tx_ring->size,
6487 					   &tx_ring->dma,
6488 					   GFP_KERNEL);
6489 	set_dev_node(dev, orig_node);
6490 	if (!tx_ring->desc)
6491 		tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6492 						   &tx_ring->dma, GFP_KERNEL);
6493 	if (!tx_ring->desc)
6494 		goto err;
6495 
6496 	tx_ring->next_to_use = 0;
6497 	tx_ring->next_to_clean = 0;
6498 	return 0;
6499 
6500 err:
6501 	vfree(tx_ring->tx_buffer_info);
6502 	tx_ring->tx_buffer_info = NULL;
6503 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6504 	return -ENOMEM;
6505 }
6506 
6507 /**
6508  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6509  * @adapter: board private structure
6510  *
6511  * If this function returns with an error, then it's possible one or
6512  * more of the rings is populated (while the rest are not).  It is the
6513  * callers duty to clean those orphaned rings.
6514  *
6515  * Return 0 on success, negative on failure
6516  **/
6517 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6518 {
6519 	int i, j = 0, err = 0;
6520 
6521 	for (i = 0; i < adapter->num_tx_queues; i++) {
6522 		err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6523 		if (!err)
6524 			continue;
6525 
6526 		e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6527 		goto err_setup_tx;
6528 	}
6529 	for (j = 0; j < adapter->num_xdp_queues; j++) {
6530 		err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6531 		if (!err)
6532 			continue;
6533 
6534 		e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6535 		goto err_setup_tx;
6536 	}
6537 
6538 	return 0;
6539 err_setup_tx:
6540 	/* rewind the index freeing the rings as we go */
6541 	while (j--)
6542 		ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6543 	while (i--)
6544 		ixgbe_free_tx_resources(adapter->tx_ring[i]);
6545 	return err;
6546 }
6547 
6548 /**
6549  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6550  * @adapter: pointer to ixgbe_adapter
6551  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6552  *
6553  * Returns 0 on success, negative on failure
6554  **/
6555 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6556 			     struct ixgbe_ring *rx_ring)
6557 {
6558 	struct device *dev = rx_ring->dev;
6559 	int orig_node = dev_to_node(dev);
6560 	int ring_node = NUMA_NO_NODE;
6561 	int size;
6562 
6563 	size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6564 
6565 	if (rx_ring->q_vector)
6566 		ring_node = rx_ring->q_vector->numa_node;
6567 
6568 	rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6569 	if (!rx_ring->rx_buffer_info)
6570 		rx_ring->rx_buffer_info = vmalloc(size);
6571 	if (!rx_ring->rx_buffer_info)
6572 		goto err;
6573 
6574 	/* Round up to nearest 4K */
6575 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6576 	rx_ring->size = ALIGN(rx_ring->size, 4096);
6577 
6578 	set_dev_node(dev, ring_node);
6579 	rx_ring->desc = dma_alloc_coherent(dev,
6580 					   rx_ring->size,
6581 					   &rx_ring->dma,
6582 					   GFP_KERNEL);
6583 	set_dev_node(dev, orig_node);
6584 	if (!rx_ring->desc)
6585 		rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6586 						   &rx_ring->dma, GFP_KERNEL);
6587 	if (!rx_ring->desc)
6588 		goto err;
6589 
6590 	rx_ring->next_to_clean = 0;
6591 	rx_ring->next_to_use = 0;
6592 
6593 	/* XDP RX-queue info */
6594 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6595 			     rx_ring->queue_index) < 0)
6596 		goto err;
6597 
6598 	rx_ring->xdp_prog = adapter->xdp_prog;
6599 
6600 	return 0;
6601 err:
6602 	vfree(rx_ring->rx_buffer_info);
6603 	rx_ring->rx_buffer_info = NULL;
6604 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6605 	return -ENOMEM;
6606 }
6607 
6608 /**
6609  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6610  * @adapter: board private structure
6611  *
6612  * If this function returns with an error, then it's possible one or
6613  * more of the rings is populated (while the rest are not).  It is the
6614  * callers duty to clean those orphaned rings.
6615  *
6616  * Return 0 on success, negative on failure
6617  **/
6618 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6619 {
6620 	int i, err = 0;
6621 
6622 	for (i = 0; i < adapter->num_rx_queues; i++) {
6623 		err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6624 		if (!err)
6625 			continue;
6626 
6627 		e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6628 		goto err_setup_rx;
6629 	}
6630 
6631 #ifdef IXGBE_FCOE
6632 	err = ixgbe_setup_fcoe_ddp_resources(adapter);
6633 	if (!err)
6634 #endif
6635 		return 0;
6636 err_setup_rx:
6637 	/* rewind the index freeing the rings as we go */
6638 	while (i--)
6639 		ixgbe_free_rx_resources(adapter->rx_ring[i]);
6640 	return err;
6641 }
6642 
6643 /**
6644  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6645  * @tx_ring: Tx descriptor ring for a specific queue
6646  *
6647  * Free all transmit software resources
6648  **/
6649 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6650 {
6651 	ixgbe_clean_tx_ring(tx_ring);
6652 
6653 	vfree(tx_ring->tx_buffer_info);
6654 	tx_ring->tx_buffer_info = NULL;
6655 
6656 	/* if not set, then don't free */
6657 	if (!tx_ring->desc)
6658 		return;
6659 
6660 	dma_free_coherent(tx_ring->dev, tx_ring->size,
6661 			  tx_ring->desc, tx_ring->dma);
6662 
6663 	tx_ring->desc = NULL;
6664 }
6665 
6666 /**
6667  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6668  * @adapter: board private structure
6669  *
6670  * Free all transmit software resources
6671  **/
6672 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6673 {
6674 	int i;
6675 
6676 	for (i = 0; i < adapter->num_tx_queues; i++)
6677 		if (adapter->tx_ring[i]->desc)
6678 			ixgbe_free_tx_resources(adapter->tx_ring[i]);
6679 	for (i = 0; i < adapter->num_xdp_queues; i++)
6680 		if (adapter->xdp_ring[i]->desc)
6681 			ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6682 }
6683 
6684 /**
6685  * ixgbe_free_rx_resources - Free Rx Resources
6686  * @rx_ring: ring to clean the resources from
6687  *
6688  * Free all receive software resources
6689  **/
6690 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6691 {
6692 	ixgbe_clean_rx_ring(rx_ring);
6693 
6694 	rx_ring->xdp_prog = NULL;
6695 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6696 	vfree(rx_ring->rx_buffer_info);
6697 	rx_ring->rx_buffer_info = NULL;
6698 
6699 	/* if not set, then don't free */
6700 	if (!rx_ring->desc)
6701 		return;
6702 
6703 	dma_free_coherent(rx_ring->dev, rx_ring->size,
6704 			  rx_ring->desc, rx_ring->dma);
6705 
6706 	rx_ring->desc = NULL;
6707 }
6708 
6709 /**
6710  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6711  * @adapter: board private structure
6712  *
6713  * Free all receive software resources
6714  **/
6715 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6716 {
6717 	int i;
6718 
6719 #ifdef IXGBE_FCOE
6720 	ixgbe_free_fcoe_ddp_resources(adapter);
6721 
6722 #endif
6723 	for (i = 0; i < adapter->num_rx_queues; i++)
6724 		if (adapter->rx_ring[i]->desc)
6725 			ixgbe_free_rx_resources(adapter->rx_ring[i]);
6726 }
6727 
6728 /**
6729  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6730  * @netdev: network interface device structure
6731  * @new_mtu: new value for maximum frame size
6732  *
6733  * Returns 0 on success, negative on failure
6734  **/
6735 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6736 {
6737 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6738 
6739 	if (adapter->xdp_prog) {
6740 		int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6741 				     VLAN_HLEN;
6742 		int i;
6743 
6744 		for (i = 0; i < adapter->num_rx_queues; i++) {
6745 			struct ixgbe_ring *ring = adapter->rx_ring[i];
6746 
6747 			if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6748 				e_warn(probe, "Requested MTU size is not supported with XDP\n");
6749 				return -EINVAL;
6750 			}
6751 		}
6752 	}
6753 
6754 	/*
6755 	 * For 82599EB we cannot allow legacy VFs to enable their receive
6756 	 * paths when MTU greater than 1500 is configured.  So display a
6757 	 * warning that legacy VFs will be disabled.
6758 	 */
6759 	if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6760 	    (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6761 	    (new_mtu > ETH_DATA_LEN))
6762 		e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6763 
6764 	netdev_dbg(netdev, "changing MTU from %d to %d\n",
6765 		   netdev->mtu, new_mtu);
6766 
6767 	/* must set new MTU before calling down or up */
6768 	netdev->mtu = new_mtu;
6769 
6770 	if (netif_running(netdev))
6771 		ixgbe_reinit_locked(adapter);
6772 
6773 	return 0;
6774 }
6775 
6776 /**
6777  * ixgbe_open - Called when a network interface is made active
6778  * @netdev: network interface device structure
6779  *
6780  * Returns 0 on success, negative value on failure
6781  *
6782  * The open entry point is called when a network interface is made
6783  * active by the system (IFF_UP).  At this point all resources needed
6784  * for transmit and receive operations are allocated, the interrupt
6785  * handler is registered with the OS, the watchdog timer is started,
6786  * and the stack is notified that the interface is ready.
6787  **/
6788 int ixgbe_open(struct net_device *netdev)
6789 {
6790 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6791 	struct ixgbe_hw *hw = &adapter->hw;
6792 	int err, queues;
6793 
6794 	/* disallow open during test */
6795 	if (test_bit(__IXGBE_TESTING, &adapter->state))
6796 		return -EBUSY;
6797 
6798 	netif_carrier_off(netdev);
6799 
6800 	/* allocate transmit descriptors */
6801 	err = ixgbe_setup_all_tx_resources(adapter);
6802 	if (err)
6803 		goto err_setup_tx;
6804 
6805 	/* allocate receive descriptors */
6806 	err = ixgbe_setup_all_rx_resources(adapter);
6807 	if (err)
6808 		goto err_setup_rx;
6809 
6810 	ixgbe_configure(adapter);
6811 
6812 	err = ixgbe_request_irq(adapter);
6813 	if (err)
6814 		goto err_req_irq;
6815 
6816 	/* Notify the stack of the actual queue counts. */
6817 	queues = adapter->num_tx_queues;
6818 	err = netif_set_real_num_tx_queues(netdev, queues);
6819 	if (err)
6820 		goto err_set_queues;
6821 
6822 	queues = adapter->num_rx_queues;
6823 	err = netif_set_real_num_rx_queues(netdev, queues);
6824 	if (err)
6825 		goto err_set_queues;
6826 
6827 	ixgbe_ptp_init(adapter);
6828 
6829 	ixgbe_up_complete(adapter);
6830 
6831 	ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6832 	udp_tunnel_get_rx_info(netdev);
6833 
6834 	return 0;
6835 
6836 err_set_queues:
6837 	ixgbe_free_irq(adapter);
6838 err_req_irq:
6839 	ixgbe_free_all_rx_resources(adapter);
6840 	if (hw->phy.ops.set_phy_power && !adapter->wol)
6841 		hw->phy.ops.set_phy_power(&adapter->hw, false);
6842 err_setup_rx:
6843 	ixgbe_free_all_tx_resources(adapter);
6844 err_setup_tx:
6845 	ixgbe_reset(adapter);
6846 
6847 	return err;
6848 }
6849 
6850 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6851 {
6852 	ixgbe_ptp_suspend(adapter);
6853 
6854 	if (adapter->hw.phy.ops.enter_lplu) {
6855 		adapter->hw.phy.reset_disable = true;
6856 		ixgbe_down(adapter);
6857 		adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6858 		adapter->hw.phy.reset_disable = false;
6859 	} else {
6860 		ixgbe_down(adapter);
6861 	}
6862 
6863 	ixgbe_free_irq(adapter);
6864 
6865 	ixgbe_free_all_tx_resources(adapter);
6866 	ixgbe_free_all_rx_resources(adapter);
6867 }
6868 
6869 /**
6870  * ixgbe_close - Disables a network interface
6871  * @netdev: network interface device structure
6872  *
6873  * Returns 0, this is not allowed to fail
6874  *
6875  * The close entry point is called when an interface is de-activated
6876  * by the OS.  The hardware is still under the drivers control, but
6877  * needs to be disabled.  A global MAC reset is issued to stop the
6878  * hardware, and all transmit and receive resources are freed.
6879  **/
6880 int ixgbe_close(struct net_device *netdev)
6881 {
6882 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
6883 
6884 	ixgbe_ptp_stop(adapter);
6885 
6886 	if (netif_device_present(netdev))
6887 		ixgbe_close_suspend(adapter);
6888 
6889 	ixgbe_fdir_filter_exit(adapter);
6890 
6891 	ixgbe_release_hw_control(adapter);
6892 
6893 	return 0;
6894 }
6895 
6896 #ifdef CONFIG_PM
6897 static int ixgbe_resume(struct pci_dev *pdev)
6898 {
6899 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6900 	struct net_device *netdev = adapter->netdev;
6901 	u32 err;
6902 
6903 	adapter->hw.hw_addr = adapter->io_addr;
6904 	pci_set_power_state(pdev, PCI_D0);
6905 	pci_restore_state(pdev);
6906 	/*
6907 	 * pci_restore_state clears dev->state_saved so call
6908 	 * pci_save_state to restore it.
6909 	 */
6910 	pci_save_state(pdev);
6911 
6912 	err = pci_enable_device_mem(pdev);
6913 	if (err) {
6914 		e_dev_err("Cannot enable PCI device from suspend\n");
6915 		return err;
6916 	}
6917 	smp_mb__before_atomic();
6918 	clear_bit(__IXGBE_DISABLED, &adapter->state);
6919 	pci_set_master(pdev);
6920 
6921 	pci_wake_from_d3(pdev, false);
6922 
6923 	ixgbe_reset(adapter);
6924 
6925 	IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6926 
6927 	rtnl_lock();
6928 	err = ixgbe_init_interrupt_scheme(adapter);
6929 	if (!err && netif_running(netdev))
6930 		err = ixgbe_open(netdev);
6931 
6932 
6933 	if (!err)
6934 		netif_device_attach(netdev);
6935 	rtnl_unlock();
6936 
6937 	return err;
6938 }
6939 #endif /* CONFIG_PM */
6940 
6941 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6942 {
6943 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6944 	struct net_device *netdev = adapter->netdev;
6945 	struct ixgbe_hw *hw = &adapter->hw;
6946 	u32 ctrl;
6947 	u32 wufc = adapter->wol;
6948 #ifdef CONFIG_PM
6949 	int retval = 0;
6950 #endif
6951 
6952 	rtnl_lock();
6953 	netif_device_detach(netdev);
6954 
6955 	if (netif_running(netdev))
6956 		ixgbe_close_suspend(adapter);
6957 
6958 	ixgbe_clear_interrupt_scheme(adapter);
6959 	rtnl_unlock();
6960 
6961 #ifdef CONFIG_PM
6962 	retval = pci_save_state(pdev);
6963 	if (retval)
6964 		return retval;
6965 
6966 #endif
6967 	if (hw->mac.ops.stop_link_on_d3)
6968 		hw->mac.ops.stop_link_on_d3(hw);
6969 
6970 	if (wufc) {
6971 		u32 fctrl;
6972 
6973 		ixgbe_set_rx_mode(netdev);
6974 
6975 		/* enable the optics for 82599 SFP+ fiber as we can WoL */
6976 		if (hw->mac.ops.enable_tx_laser)
6977 			hw->mac.ops.enable_tx_laser(hw);
6978 
6979 		/* enable the reception of multicast packets */
6980 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6981 		fctrl |= IXGBE_FCTRL_MPE;
6982 		IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6983 
6984 		ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6985 		ctrl |= IXGBE_CTRL_GIO_DIS;
6986 		IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6987 
6988 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6989 	} else {
6990 		IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6991 		IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6992 	}
6993 
6994 	switch (hw->mac.type) {
6995 	case ixgbe_mac_82598EB:
6996 		pci_wake_from_d3(pdev, false);
6997 		break;
6998 	case ixgbe_mac_82599EB:
6999 	case ixgbe_mac_X540:
7000 	case ixgbe_mac_X550:
7001 	case ixgbe_mac_X550EM_x:
7002 	case ixgbe_mac_x550em_a:
7003 		pci_wake_from_d3(pdev, !!wufc);
7004 		break;
7005 	default:
7006 		break;
7007 	}
7008 
7009 	*enable_wake = !!wufc;
7010 	if (hw->phy.ops.set_phy_power && !*enable_wake)
7011 		hw->phy.ops.set_phy_power(hw, false);
7012 
7013 	ixgbe_release_hw_control(adapter);
7014 
7015 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
7016 		pci_disable_device(pdev);
7017 
7018 	return 0;
7019 }
7020 
7021 #ifdef CONFIG_PM
7022 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
7023 {
7024 	int retval;
7025 	bool wake;
7026 
7027 	retval = __ixgbe_shutdown(pdev, &wake);
7028 	if (retval)
7029 		return retval;
7030 
7031 	if (wake) {
7032 		pci_prepare_to_sleep(pdev);
7033 	} else {
7034 		pci_wake_from_d3(pdev, false);
7035 		pci_set_power_state(pdev, PCI_D3hot);
7036 	}
7037 
7038 	return 0;
7039 }
7040 #endif /* CONFIG_PM */
7041 
7042 static void ixgbe_shutdown(struct pci_dev *pdev)
7043 {
7044 	bool wake;
7045 
7046 	__ixgbe_shutdown(pdev, &wake);
7047 
7048 	if (system_state == SYSTEM_POWER_OFF) {
7049 		pci_wake_from_d3(pdev, wake);
7050 		pci_set_power_state(pdev, PCI_D3hot);
7051 	}
7052 }
7053 
7054 /**
7055  * ixgbe_update_stats - Update the board statistics counters.
7056  * @adapter: board private structure
7057  **/
7058 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7059 {
7060 	struct net_device *netdev = adapter->netdev;
7061 	struct ixgbe_hw *hw = &adapter->hw;
7062 	struct ixgbe_hw_stats *hwstats = &adapter->stats;
7063 	u64 total_mpc = 0;
7064 	u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7065 	u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7066 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7067 	u64 alloc_rx_page = 0;
7068 	u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7069 
7070 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7071 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7072 		return;
7073 
7074 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7075 		u64 rsc_count = 0;
7076 		u64 rsc_flush = 0;
7077 		for (i = 0; i < adapter->num_rx_queues; i++) {
7078 			rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7079 			rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7080 		}
7081 		adapter->rsc_total_count = rsc_count;
7082 		adapter->rsc_total_flush = rsc_flush;
7083 	}
7084 
7085 	for (i = 0; i < adapter->num_rx_queues; i++) {
7086 		struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7087 		non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7088 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7089 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7090 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7091 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7092 		bytes += rx_ring->stats.bytes;
7093 		packets += rx_ring->stats.packets;
7094 	}
7095 	adapter->non_eop_descs = non_eop_descs;
7096 	adapter->alloc_rx_page = alloc_rx_page;
7097 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7098 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7099 	adapter->hw_csum_rx_error = hw_csum_rx_error;
7100 	netdev->stats.rx_bytes = bytes;
7101 	netdev->stats.rx_packets = packets;
7102 
7103 	bytes = 0;
7104 	packets = 0;
7105 	/* gather some stats to the adapter struct that are per queue */
7106 	for (i = 0; i < adapter->num_tx_queues; i++) {
7107 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7108 		restart_queue += tx_ring->tx_stats.restart_queue;
7109 		tx_busy += tx_ring->tx_stats.tx_busy;
7110 		bytes += tx_ring->stats.bytes;
7111 		packets += tx_ring->stats.packets;
7112 	}
7113 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7114 		struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7115 
7116 		restart_queue += xdp_ring->tx_stats.restart_queue;
7117 		tx_busy += xdp_ring->tx_stats.tx_busy;
7118 		bytes += xdp_ring->stats.bytes;
7119 		packets += xdp_ring->stats.packets;
7120 	}
7121 	adapter->restart_queue = restart_queue;
7122 	adapter->tx_busy = tx_busy;
7123 	netdev->stats.tx_bytes = bytes;
7124 	netdev->stats.tx_packets = packets;
7125 
7126 	hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7127 
7128 	/* 8 register reads */
7129 	for (i = 0; i < 8; i++) {
7130 		/* for packet buffers not used, the register should read 0 */
7131 		mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7132 		missed_rx += mpc;
7133 		hwstats->mpc[i] += mpc;
7134 		total_mpc += hwstats->mpc[i];
7135 		hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7136 		hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7137 		switch (hw->mac.type) {
7138 		case ixgbe_mac_82598EB:
7139 			hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7140 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7141 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7142 			hwstats->pxonrxc[i] +=
7143 				IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7144 			break;
7145 		case ixgbe_mac_82599EB:
7146 		case ixgbe_mac_X540:
7147 		case ixgbe_mac_X550:
7148 		case ixgbe_mac_X550EM_x:
7149 		case ixgbe_mac_x550em_a:
7150 			hwstats->pxonrxc[i] +=
7151 				IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7152 			break;
7153 		default:
7154 			break;
7155 		}
7156 	}
7157 
7158 	/*16 register reads */
7159 	for (i = 0; i < 16; i++) {
7160 		hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7161 		hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7162 		if ((hw->mac.type == ixgbe_mac_82599EB) ||
7163 		    (hw->mac.type == ixgbe_mac_X540) ||
7164 		    (hw->mac.type == ixgbe_mac_X550) ||
7165 		    (hw->mac.type == ixgbe_mac_X550EM_x) ||
7166 		    (hw->mac.type == ixgbe_mac_x550em_a)) {
7167 			hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7168 			IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7169 			hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7170 			IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7171 		}
7172 	}
7173 
7174 	hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7175 	/* work around hardware counting issue */
7176 	hwstats->gprc -= missed_rx;
7177 
7178 	ixgbe_update_xoff_received(adapter);
7179 
7180 	/* 82598 hardware only has a 32 bit counter in the high register */
7181 	switch (hw->mac.type) {
7182 	case ixgbe_mac_82598EB:
7183 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7184 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7185 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7186 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7187 		break;
7188 	case ixgbe_mac_X540:
7189 	case ixgbe_mac_X550:
7190 	case ixgbe_mac_X550EM_x:
7191 	case ixgbe_mac_x550em_a:
7192 		/* OS2BMC stats are X540 and later */
7193 		hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7194 		hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7195 		hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7196 		hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7197 		/* fall through */
7198 	case ixgbe_mac_82599EB:
7199 		for (i = 0; i < 16; i++)
7200 			adapter->hw_rx_no_dma_resources +=
7201 					     IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7202 		hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7203 		IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7204 		hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7205 		IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7206 		hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7207 		IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7208 		hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7209 		hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7210 		hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7211 #ifdef IXGBE_FCOE
7212 		hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7213 		hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7214 		hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7215 		hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7216 		hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7217 		hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7218 		/* Add up per cpu counters for total ddp aloc fail */
7219 		if (adapter->fcoe.ddp_pool) {
7220 			struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7221 			struct ixgbe_fcoe_ddp_pool *ddp_pool;
7222 			unsigned int cpu;
7223 			u64 noddp = 0, noddp_ext_buff = 0;
7224 			for_each_possible_cpu(cpu) {
7225 				ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7226 				noddp += ddp_pool->noddp;
7227 				noddp_ext_buff += ddp_pool->noddp_ext_buff;
7228 			}
7229 			hwstats->fcoe_noddp = noddp;
7230 			hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7231 		}
7232 #endif /* IXGBE_FCOE */
7233 		break;
7234 	default:
7235 		break;
7236 	}
7237 	bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7238 	hwstats->bprc += bprc;
7239 	hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7240 	if (hw->mac.type == ixgbe_mac_82598EB)
7241 		hwstats->mprc -= bprc;
7242 	hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7243 	hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7244 	hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7245 	hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7246 	hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7247 	hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7248 	hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7249 	hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7250 	lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7251 	hwstats->lxontxc += lxon;
7252 	lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7253 	hwstats->lxofftxc += lxoff;
7254 	hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7255 	hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7256 	/*
7257 	 * 82598 errata - tx of flow control packets is included in tx counters
7258 	 */
7259 	xon_off_tot = lxon + lxoff;
7260 	hwstats->gptc -= xon_off_tot;
7261 	hwstats->mptc -= xon_off_tot;
7262 	hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7263 	hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7264 	hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7265 	hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7266 	hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7267 	hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7268 	hwstats->ptc64 -= xon_off_tot;
7269 	hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7270 	hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7271 	hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7272 	hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7273 	hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7274 	hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7275 
7276 	/* Fill out the OS statistics structure */
7277 	netdev->stats.multicast = hwstats->mprc;
7278 
7279 	/* Rx Errors */
7280 	netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7281 	netdev->stats.rx_dropped = 0;
7282 	netdev->stats.rx_length_errors = hwstats->rlec;
7283 	netdev->stats.rx_crc_errors = hwstats->crcerrs;
7284 	netdev->stats.rx_missed_errors = total_mpc;
7285 }
7286 
7287 /**
7288  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7289  * @adapter: pointer to the device adapter structure
7290  **/
7291 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7292 {
7293 	struct ixgbe_hw *hw = &adapter->hw;
7294 	int i;
7295 
7296 	if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7297 		return;
7298 
7299 	adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7300 
7301 	/* if interface is down do nothing */
7302 	if (test_bit(__IXGBE_DOWN, &adapter->state))
7303 		return;
7304 
7305 	/* do nothing if we are not using signature filters */
7306 	if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7307 		return;
7308 
7309 	adapter->fdir_overflow++;
7310 
7311 	if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7312 		for (i = 0; i < adapter->num_tx_queues; i++)
7313 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7314 				&(adapter->tx_ring[i]->state));
7315 		for (i = 0; i < adapter->num_xdp_queues; i++)
7316 			set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7317 				&adapter->xdp_ring[i]->state);
7318 		/* re-enable flow director interrupts */
7319 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7320 	} else {
7321 		e_err(probe, "failed to finish FDIR re-initialization, "
7322 		      "ignored adding FDIR ATR filters\n");
7323 	}
7324 }
7325 
7326 /**
7327  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7328  * @adapter: pointer to the device adapter structure
7329  *
7330  * This function serves two purposes.  First it strobes the interrupt lines
7331  * in order to make certain interrupts are occurring.  Secondly it sets the
7332  * bits needed to check for TX hangs.  As a result we should immediately
7333  * determine if a hang has occurred.
7334  */
7335 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7336 {
7337 	struct ixgbe_hw *hw = &adapter->hw;
7338 	u64 eics = 0;
7339 	int i;
7340 
7341 	/* If we're down, removing or resetting, just bail */
7342 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7343 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7344 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7345 		return;
7346 
7347 	/* Force detection of hung controller */
7348 	if (netif_carrier_ok(adapter->netdev)) {
7349 		for (i = 0; i < adapter->num_tx_queues; i++)
7350 			set_check_for_tx_hang(adapter->tx_ring[i]);
7351 		for (i = 0; i < adapter->num_xdp_queues; i++)
7352 			set_check_for_tx_hang(adapter->xdp_ring[i]);
7353 	}
7354 
7355 	if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7356 		/*
7357 		 * for legacy and MSI interrupts don't set any bits
7358 		 * that are enabled for EIAM, because this operation
7359 		 * would set *both* EIMS and EICS for any bit in EIAM
7360 		 */
7361 		IXGBE_WRITE_REG(hw, IXGBE_EICS,
7362 			(IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7363 	} else {
7364 		/* get one bit for every active tx/rx interrupt vector */
7365 		for (i = 0; i < adapter->num_q_vectors; i++) {
7366 			struct ixgbe_q_vector *qv = adapter->q_vector[i];
7367 			if (qv->rx.ring || qv->tx.ring)
7368 				eics |= BIT_ULL(i);
7369 		}
7370 	}
7371 
7372 	/* Cause software interrupt to ensure rings are cleaned */
7373 	ixgbe_irq_rearm_queues(adapter, eics);
7374 }
7375 
7376 /**
7377  * ixgbe_watchdog_update_link - update the link status
7378  * @adapter: pointer to the device adapter structure
7379  **/
7380 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7381 {
7382 	struct ixgbe_hw *hw = &adapter->hw;
7383 	u32 link_speed = adapter->link_speed;
7384 	bool link_up = adapter->link_up;
7385 	bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7386 
7387 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7388 		return;
7389 
7390 	if (hw->mac.ops.check_link) {
7391 		hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7392 	} else {
7393 		/* always assume link is up, if no check link function */
7394 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7395 		link_up = true;
7396 	}
7397 
7398 	if (adapter->ixgbe_ieee_pfc)
7399 		pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7400 
7401 	if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7402 		hw->mac.ops.fc_enable(hw);
7403 		ixgbe_set_rx_drop_en(adapter);
7404 	}
7405 
7406 	if (link_up ||
7407 	    time_after(jiffies, (adapter->link_check_timeout +
7408 				 IXGBE_TRY_LINK_TIMEOUT))) {
7409 		adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7410 		IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7411 		IXGBE_WRITE_FLUSH(hw);
7412 	}
7413 
7414 	adapter->link_up = link_up;
7415 	adapter->link_speed = link_speed;
7416 }
7417 
7418 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7419 {
7420 #ifdef CONFIG_IXGBE_DCB
7421 	struct net_device *netdev = adapter->netdev;
7422 	struct dcb_app app = {
7423 			      .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7424 			      .protocol = 0,
7425 			     };
7426 	u8 up = 0;
7427 
7428 	if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7429 		up = dcb_ieee_getapp_mask(netdev, &app);
7430 
7431 	adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7432 #endif
7433 }
7434 
7435 /**
7436  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7437  *                             print link up message
7438  * @adapter: pointer to the device adapter structure
7439  **/
7440 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7441 {
7442 	struct net_device *netdev = adapter->netdev;
7443 	struct ixgbe_hw *hw = &adapter->hw;
7444 	u32 link_speed = adapter->link_speed;
7445 	const char *speed_str;
7446 	bool flow_rx, flow_tx;
7447 
7448 	/* only continue if link was previously down */
7449 	if (netif_carrier_ok(netdev))
7450 		return;
7451 
7452 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7453 
7454 	switch (hw->mac.type) {
7455 	case ixgbe_mac_82598EB: {
7456 		u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7457 		u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7458 		flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7459 		flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7460 	}
7461 		break;
7462 	case ixgbe_mac_X540:
7463 	case ixgbe_mac_X550:
7464 	case ixgbe_mac_X550EM_x:
7465 	case ixgbe_mac_x550em_a:
7466 	case ixgbe_mac_82599EB: {
7467 		u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7468 		u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7469 		flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7470 		flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7471 	}
7472 		break;
7473 	default:
7474 		flow_tx = false;
7475 		flow_rx = false;
7476 		break;
7477 	}
7478 
7479 	adapter->last_rx_ptp_check = jiffies;
7480 
7481 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7482 		ixgbe_ptp_start_cyclecounter(adapter);
7483 
7484 	switch (link_speed) {
7485 	case IXGBE_LINK_SPEED_10GB_FULL:
7486 		speed_str = "10 Gbps";
7487 		break;
7488 	case IXGBE_LINK_SPEED_5GB_FULL:
7489 		speed_str = "5 Gbps";
7490 		break;
7491 	case IXGBE_LINK_SPEED_2_5GB_FULL:
7492 		speed_str = "2.5 Gbps";
7493 		break;
7494 	case IXGBE_LINK_SPEED_1GB_FULL:
7495 		speed_str = "1 Gbps";
7496 		break;
7497 	case IXGBE_LINK_SPEED_100_FULL:
7498 		speed_str = "100 Mbps";
7499 		break;
7500 	case IXGBE_LINK_SPEED_10_FULL:
7501 		speed_str = "10 Mbps";
7502 		break;
7503 	default:
7504 		speed_str = "unknown speed";
7505 		break;
7506 	}
7507 	e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7508 	       ((flow_rx && flow_tx) ? "RX/TX" :
7509 	       (flow_rx ? "RX" :
7510 	       (flow_tx ? "TX" : "None"))));
7511 
7512 	netif_carrier_on(netdev);
7513 	ixgbe_check_vf_rate_limit(adapter);
7514 
7515 	/* enable transmits */
7516 	netif_tx_wake_all_queues(adapter->netdev);
7517 
7518 	/* update the default user priority for VFs */
7519 	ixgbe_update_default_up(adapter);
7520 
7521 	/* ping all the active vfs to let them know link has changed */
7522 	ixgbe_ping_all_vfs(adapter);
7523 }
7524 
7525 /**
7526  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7527  *                               print link down message
7528  * @adapter: pointer to the adapter structure
7529  **/
7530 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7531 {
7532 	struct net_device *netdev = adapter->netdev;
7533 	struct ixgbe_hw *hw = &adapter->hw;
7534 
7535 	adapter->link_up = false;
7536 	adapter->link_speed = 0;
7537 
7538 	/* only continue if link was up previously */
7539 	if (!netif_carrier_ok(netdev))
7540 		return;
7541 
7542 	/* poll for SFP+ cable when link is down */
7543 	if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7544 		adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7545 
7546 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7547 		ixgbe_ptp_start_cyclecounter(adapter);
7548 
7549 	e_info(drv, "NIC Link is Down\n");
7550 	netif_carrier_off(netdev);
7551 
7552 	/* ping all the active vfs to let them know link has changed */
7553 	ixgbe_ping_all_vfs(adapter);
7554 }
7555 
7556 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7557 {
7558 	int i;
7559 
7560 	for (i = 0; i < adapter->num_tx_queues; i++) {
7561 		struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7562 
7563 		if (tx_ring->next_to_use != tx_ring->next_to_clean)
7564 			return true;
7565 	}
7566 
7567 	for (i = 0; i < adapter->num_xdp_queues; i++) {
7568 		struct ixgbe_ring *ring = adapter->xdp_ring[i];
7569 
7570 		if (ring->next_to_use != ring->next_to_clean)
7571 			return true;
7572 	}
7573 
7574 	return false;
7575 }
7576 
7577 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7578 {
7579 	struct ixgbe_hw *hw = &adapter->hw;
7580 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7581 	u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7582 
7583 	int i, j;
7584 
7585 	if (!adapter->num_vfs)
7586 		return false;
7587 
7588 	/* resetting the PF is only needed for MAC before X550 */
7589 	if (hw->mac.type >= ixgbe_mac_X550)
7590 		return false;
7591 
7592 	for (i = 0; i < adapter->num_vfs; i++) {
7593 		for (j = 0; j < q_per_pool; j++) {
7594 			u32 h, t;
7595 
7596 			h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7597 			t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7598 
7599 			if (h != t)
7600 				return true;
7601 		}
7602 	}
7603 
7604 	return false;
7605 }
7606 
7607 /**
7608  * ixgbe_watchdog_flush_tx - flush queues on link down
7609  * @adapter: pointer to the device adapter structure
7610  **/
7611 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7612 {
7613 	if (!netif_carrier_ok(adapter->netdev)) {
7614 		if (ixgbe_ring_tx_pending(adapter) ||
7615 		    ixgbe_vf_tx_pending(adapter)) {
7616 			/* We've lost link, so the controller stops DMA,
7617 			 * but we've got queued Tx work that's never going
7618 			 * to get done, so reset controller to flush Tx.
7619 			 * (Do the reset outside of interrupt context).
7620 			 */
7621 			e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7622 			set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7623 		}
7624 	}
7625 }
7626 
7627 #ifdef CONFIG_PCI_IOV
7628 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7629 {
7630 	struct ixgbe_hw *hw = &adapter->hw;
7631 	struct pci_dev *pdev = adapter->pdev;
7632 	unsigned int vf;
7633 	u32 gpc;
7634 
7635 	if (!(netif_carrier_ok(adapter->netdev)))
7636 		return;
7637 
7638 	gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7639 	if (gpc) /* If incrementing then no need for the check below */
7640 		return;
7641 	/* Check to see if a bad DMA write target from an errant or
7642 	 * malicious VF has caused a PCIe error.  If so then we can
7643 	 * issue a VFLR to the offending VF(s) and then resume without
7644 	 * requesting a full slot reset.
7645 	 */
7646 
7647 	if (!pdev)
7648 		return;
7649 
7650 	/* check status reg for all VFs owned by this PF */
7651 	for (vf = 0; vf < adapter->num_vfs; ++vf) {
7652 		struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7653 		u16 status_reg;
7654 
7655 		if (!vfdev)
7656 			continue;
7657 		pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7658 		if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7659 		    status_reg & PCI_STATUS_REC_MASTER_ABORT)
7660 			pcie_flr(vfdev);
7661 	}
7662 }
7663 
7664 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7665 {
7666 	u32 ssvpc;
7667 
7668 	/* Do not perform spoof check for 82598 or if not in IOV mode */
7669 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7670 	    adapter->num_vfs == 0)
7671 		return;
7672 
7673 	ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7674 
7675 	/*
7676 	 * ssvpc register is cleared on read, if zero then no
7677 	 * spoofed packets in the last interval.
7678 	 */
7679 	if (!ssvpc)
7680 		return;
7681 
7682 	e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7683 }
7684 #else
7685 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7686 {
7687 }
7688 
7689 static void
7690 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7691 {
7692 }
7693 #endif /* CONFIG_PCI_IOV */
7694 
7695 
7696 /**
7697  * ixgbe_watchdog_subtask - check and bring link up
7698  * @adapter: pointer to the device adapter structure
7699  **/
7700 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7701 {
7702 	/* if interface is down, removing or resetting, do nothing */
7703 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7704 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7705 	    test_bit(__IXGBE_RESETTING, &adapter->state))
7706 		return;
7707 
7708 	ixgbe_watchdog_update_link(adapter);
7709 
7710 	if (adapter->link_up)
7711 		ixgbe_watchdog_link_is_up(adapter);
7712 	else
7713 		ixgbe_watchdog_link_is_down(adapter);
7714 
7715 	ixgbe_check_for_bad_vf(adapter);
7716 	ixgbe_spoof_check(adapter);
7717 	ixgbe_update_stats(adapter);
7718 
7719 	ixgbe_watchdog_flush_tx(adapter);
7720 }
7721 
7722 /**
7723  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7724  * @adapter: the ixgbe adapter structure
7725  **/
7726 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7727 {
7728 	struct ixgbe_hw *hw = &adapter->hw;
7729 	s32 err;
7730 
7731 	/* not searching for SFP so there is nothing to do here */
7732 	if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7733 	    !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7734 		return;
7735 
7736 	if (adapter->sfp_poll_time &&
7737 	    time_after(adapter->sfp_poll_time, jiffies))
7738 		return; /* If not yet time to poll for SFP */
7739 
7740 	/* someone else is in init, wait until next service event */
7741 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7742 		return;
7743 
7744 	adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7745 
7746 	err = hw->phy.ops.identify_sfp(hw);
7747 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7748 		goto sfp_out;
7749 
7750 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7751 		/* If no cable is present, then we need to reset
7752 		 * the next time we find a good cable. */
7753 		adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7754 	}
7755 
7756 	/* exit on error */
7757 	if (err)
7758 		goto sfp_out;
7759 
7760 	/* exit if reset not needed */
7761 	if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7762 		goto sfp_out;
7763 
7764 	adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7765 
7766 	/*
7767 	 * A module may be identified correctly, but the EEPROM may not have
7768 	 * support for that module.  setup_sfp() will fail in that case, so
7769 	 * we should not allow that module to load.
7770 	 */
7771 	if (hw->mac.type == ixgbe_mac_82598EB)
7772 		err = hw->phy.ops.reset(hw);
7773 	else
7774 		err = hw->mac.ops.setup_sfp(hw);
7775 
7776 	if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7777 		goto sfp_out;
7778 
7779 	adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7780 	e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7781 
7782 sfp_out:
7783 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7784 
7785 	if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7786 	    (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7787 		e_dev_err("failed to initialize because an unsupported "
7788 			  "SFP+ module type was detected.\n");
7789 		e_dev_err("Reload the driver after installing a "
7790 			  "supported module.\n");
7791 		unregister_netdev(adapter->netdev);
7792 	}
7793 }
7794 
7795 /**
7796  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7797  * @adapter: the ixgbe adapter structure
7798  **/
7799 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7800 {
7801 	struct ixgbe_hw *hw = &adapter->hw;
7802 	u32 cap_speed;
7803 	u32 speed;
7804 	bool autoneg = false;
7805 
7806 	if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7807 		return;
7808 
7809 	/* someone else is in init, wait until next service event */
7810 	if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7811 		return;
7812 
7813 	adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7814 
7815 	hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7816 
7817 	/* advertise highest capable link speed */
7818 	if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7819 		speed = IXGBE_LINK_SPEED_10GB_FULL;
7820 	else
7821 		speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7822 				     IXGBE_LINK_SPEED_1GB_FULL);
7823 
7824 	if (hw->mac.ops.setup_link)
7825 		hw->mac.ops.setup_link(hw, speed, true);
7826 
7827 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7828 	adapter->link_check_timeout = jiffies;
7829 	clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7830 }
7831 
7832 /**
7833  * ixgbe_service_timer - Timer Call-back
7834  * @t: pointer to timer_list structure
7835  **/
7836 static void ixgbe_service_timer(struct timer_list *t)
7837 {
7838 	struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7839 	unsigned long next_event_offset;
7840 
7841 	/* poll faster when waiting for link */
7842 	if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7843 		next_event_offset = HZ / 10;
7844 	else
7845 		next_event_offset = HZ * 2;
7846 
7847 	/* Reset the timer */
7848 	mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7849 
7850 	ixgbe_service_event_schedule(adapter);
7851 }
7852 
7853 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7854 {
7855 	struct ixgbe_hw *hw = &adapter->hw;
7856 	u32 status;
7857 
7858 	if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7859 		return;
7860 
7861 	adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7862 
7863 	if (!hw->phy.ops.handle_lasi)
7864 		return;
7865 
7866 	status = hw->phy.ops.handle_lasi(&adapter->hw);
7867 	if (status != IXGBE_ERR_OVERTEMP)
7868 		return;
7869 
7870 	e_crit(drv, "%s\n", ixgbe_overheat_msg);
7871 }
7872 
7873 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7874 {
7875 	if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7876 		return;
7877 
7878 	rtnl_lock();
7879 	/* If we're already down, removing or resetting, just bail */
7880 	if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7881 	    test_bit(__IXGBE_REMOVING, &adapter->state) ||
7882 	    test_bit(__IXGBE_RESETTING, &adapter->state)) {
7883 		rtnl_unlock();
7884 		return;
7885 	}
7886 
7887 	ixgbe_dump(adapter);
7888 	netdev_err(adapter->netdev, "Reset adapter\n");
7889 	adapter->tx_timeout_count++;
7890 
7891 	ixgbe_reinit_locked(adapter);
7892 	rtnl_unlock();
7893 }
7894 
7895 /**
7896  * ixgbe_check_fw_error - Check firmware for errors
7897  * @adapter: the adapter private structure
7898  *
7899  * Check firmware errors in register FWSM
7900  */
7901 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7902 {
7903 	struct ixgbe_hw *hw = &adapter->hw;
7904 	u32 fwsm;
7905 
7906 	/* read fwsm.ext_err_ind register and log errors */
7907 	fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7908 
7909 	if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7910 	    !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7911 		e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7912 			   fwsm);
7913 
7914 	if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7915 		e_dev_err("Firmware recovery mode detected. Limiting functionality. Refer to the Intel(R) Ethernet Adapters and Devices User Guide for details on firmware recovery mode.\n");
7916 		return true;
7917 	}
7918 
7919 	return false;
7920 }
7921 
7922 /**
7923  * ixgbe_service_task - manages and runs subtasks
7924  * @work: pointer to work_struct containing our data
7925  **/
7926 static void ixgbe_service_task(struct work_struct *work)
7927 {
7928 	struct ixgbe_adapter *adapter = container_of(work,
7929 						     struct ixgbe_adapter,
7930 						     service_task);
7931 	if (ixgbe_removed(adapter->hw.hw_addr)) {
7932 		if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7933 			rtnl_lock();
7934 			ixgbe_down(adapter);
7935 			rtnl_unlock();
7936 		}
7937 		ixgbe_service_event_complete(adapter);
7938 		return;
7939 	}
7940 	if (ixgbe_check_fw_error(adapter)) {
7941 		if (!test_bit(__IXGBE_DOWN, &adapter->state))
7942 			unregister_netdev(adapter->netdev);
7943 		ixgbe_service_event_complete(adapter);
7944 		return;
7945 	}
7946 	if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7947 		rtnl_lock();
7948 		adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7949 		udp_tunnel_get_rx_info(adapter->netdev);
7950 		rtnl_unlock();
7951 	}
7952 	ixgbe_reset_subtask(adapter);
7953 	ixgbe_phy_interrupt_subtask(adapter);
7954 	ixgbe_sfp_detection_subtask(adapter);
7955 	ixgbe_sfp_link_config_subtask(adapter);
7956 	ixgbe_check_overtemp_subtask(adapter);
7957 	ixgbe_watchdog_subtask(adapter);
7958 	ixgbe_fdir_reinit_subtask(adapter);
7959 	ixgbe_check_hang_subtask(adapter);
7960 
7961 	if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7962 		ixgbe_ptp_overflow_check(adapter);
7963 		if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7964 			ixgbe_ptp_rx_hang(adapter);
7965 		ixgbe_ptp_tx_hang(adapter);
7966 	}
7967 
7968 	ixgbe_service_event_complete(adapter);
7969 }
7970 
7971 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7972 		     struct ixgbe_tx_buffer *first,
7973 		     u8 *hdr_len,
7974 		     struct ixgbe_ipsec_tx_data *itd)
7975 {
7976 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7977 	struct sk_buff *skb = first->skb;
7978 	union {
7979 		struct iphdr *v4;
7980 		struct ipv6hdr *v6;
7981 		unsigned char *hdr;
7982 	} ip;
7983 	union {
7984 		struct tcphdr *tcp;
7985 		struct udphdr *udp;
7986 		unsigned char *hdr;
7987 	} l4;
7988 	u32 paylen, l4_offset;
7989 	u32 fceof_saidx = 0;
7990 	int err;
7991 
7992 	if (skb->ip_summed != CHECKSUM_PARTIAL)
7993 		return 0;
7994 
7995 	if (!skb_is_gso(skb))
7996 		return 0;
7997 
7998 	err = skb_cow_head(skb, 0);
7999 	if (err < 0)
8000 		return err;
8001 
8002 	if (eth_p_mpls(first->protocol))
8003 		ip.hdr = skb_inner_network_header(skb);
8004 	else
8005 		ip.hdr = skb_network_header(skb);
8006 	l4.hdr = skb_checksum_start(skb);
8007 
8008 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
8009 	type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
8010 		      IXGBE_ADVTXD_TUCMD_L4T_UDP : IXGBE_ADVTXD_TUCMD_L4T_TCP;
8011 
8012 	/* initialize outer IP header fields */
8013 	if (ip.v4->version == 4) {
8014 		unsigned char *csum_start = skb_checksum_start(skb);
8015 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
8016 		int len = csum_start - trans_start;
8017 
8018 		/* IP header will have to cancel out any data that
8019 		 * is not a part of the outer IP header, so set to
8020 		 * a reverse csum if needed, else init check to 0.
8021 		 */
8022 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
8023 					   csum_fold(csum_partial(trans_start,
8024 								  len, 0)) : 0;
8025 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
8026 
8027 		ip.v4->tot_len = 0;
8028 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8029 				   IXGBE_TX_FLAGS_CSUM |
8030 				   IXGBE_TX_FLAGS_IPV4;
8031 	} else {
8032 		ip.v6->payload_len = 0;
8033 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
8034 				   IXGBE_TX_FLAGS_CSUM;
8035 	}
8036 
8037 	/* determine offset of inner transport header */
8038 	l4_offset = l4.hdr - skb->data;
8039 
8040 	/* remove payload length from inner checksum */
8041 	paylen = skb->len - l4_offset;
8042 
8043 	if (type_tucmd & IXGBE_ADVTXD_TUCMD_L4T_TCP) {
8044 		/* compute length of segmentation header */
8045 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
8046 		csum_replace_by_diff(&l4.tcp->check,
8047 				     (__force __wsum)htonl(paylen));
8048 	} else {
8049 		/* compute length of segmentation header */
8050 		*hdr_len = sizeof(*l4.udp) + l4_offset;
8051 		csum_replace_by_diff(&l4.udp->check,
8052 				     (__force __wsum)htonl(paylen));
8053 	}
8054 
8055 	/* update gso size and bytecount with header size */
8056 	first->gso_segs = skb_shinfo(skb)->gso_segs;
8057 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
8058 
8059 	/* mss_l4len_id: use 0 as index for TSO */
8060 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8061 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8062 
8063 	fceof_saidx |= itd->sa_idx;
8064 	type_tucmd |= itd->flags | itd->trailer_len;
8065 
8066 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8067 	vlan_macip_lens = l4.hdr - ip.hdr;
8068 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8069 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8070 
8071 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8072 			  mss_l4len_idx);
8073 
8074 	return 1;
8075 }
8076 
8077 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8078 {
8079 	unsigned int offset = 0;
8080 
8081 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8082 
8083 	return offset == skb_checksum_start_offset(skb);
8084 }
8085 
8086 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8087 			  struct ixgbe_tx_buffer *first,
8088 			  struct ixgbe_ipsec_tx_data *itd)
8089 {
8090 	struct sk_buff *skb = first->skb;
8091 	u32 vlan_macip_lens = 0;
8092 	u32 fceof_saidx = 0;
8093 	u32 type_tucmd = 0;
8094 
8095 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
8096 csum_failed:
8097 		if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8098 					 IXGBE_TX_FLAGS_CC)))
8099 			return;
8100 		goto no_csum;
8101 	}
8102 
8103 	switch (skb->csum_offset) {
8104 	case offsetof(struct tcphdr, check):
8105 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8106 		/* fall through */
8107 	case offsetof(struct udphdr, check):
8108 		break;
8109 	case offsetof(struct sctphdr, checksum):
8110 		/* validate that this is actually an SCTP request */
8111 		if (((first->protocol == htons(ETH_P_IP)) &&
8112 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8113 		    ((first->protocol == htons(ETH_P_IPV6)) &&
8114 		     ixgbe_ipv6_csum_is_sctp(skb))) {
8115 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8116 			break;
8117 		}
8118 		/* fall through */
8119 	default:
8120 		skb_checksum_help(skb);
8121 		goto csum_failed;
8122 	}
8123 
8124 	/* update TX checksum flag */
8125 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8126 	vlan_macip_lens = skb_checksum_start_offset(skb) -
8127 			  skb_network_offset(skb);
8128 no_csum:
8129 	/* vlan_macip_lens: MACLEN, VLAN tag */
8130 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8131 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8132 
8133 	fceof_saidx |= itd->sa_idx;
8134 	type_tucmd |= itd->flags | itd->trailer_len;
8135 
8136 	ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8137 }
8138 
8139 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8140 	((_flag <= _result) ? \
8141 	 ((u32)(_input & _flag) * (_result / _flag)) : \
8142 	 ((u32)(_input & _flag) / (_flag / _result)))
8143 
8144 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8145 {
8146 	/* set type for advanced descriptor with frame checksum insertion */
8147 	u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8148 		       IXGBE_ADVTXD_DCMD_DEXT |
8149 		       IXGBE_ADVTXD_DCMD_IFCS;
8150 
8151 	/* set HW vlan bit if vlan is present */
8152 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8153 				   IXGBE_ADVTXD_DCMD_VLE);
8154 
8155 	/* set segmentation enable bits for TSO/FSO */
8156 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8157 				   IXGBE_ADVTXD_DCMD_TSE);
8158 
8159 	/* set timestamp bit if present */
8160 	cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8161 				   IXGBE_ADVTXD_MAC_TSTAMP);
8162 
8163 	/* insert frame checksum */
8164 	cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8165 
8166 	return cmd_type;
8167 }
8168 
8169 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8170 				   u32 tx_flags, unsigned int paylen)
8171 {
8172 	u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8173 
8174 	/* enable L4 checksum for TSO and TX checksum offload */
8175 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8176 					IXGBE_TX_FLAGS_CSUM,
8177 					IXGBE_ADVTXD_POPTS_TXSM);
8178 
8179 	/* enable IPv4 checksum for TSO */
8180 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8181 					IXGBE_TX_FLAGS_IPV4,
8182 					IXGBE_ADVTXD_POPTS_IXSM);
8183 
8184 	/* enable IPsec */
8185 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8186 					IXGBE_TX_FLAGS_IPSEC,
8187 					IXGBE_ADVTXD_POPTS_IPSEC);
8188 
8189 	/*
8190 	 * Check Context must be set if Tx switch is enabled, which it
8191 	 * always is for case where virtual functions are running
8192 	 */
8193 	olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8194 					IXGBE_TX_FLAGS_CC,
8195 					IXGBE_ADVTXD_CC);
8196 
8197 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8198 }
8199 
8200 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8201 {
8202 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8203 
8204 	/* Herbert's original patch had:
8205 	 *  smp_mb__after_netif_stop_queue();
8206 	 * but since that doesn't exist yet, just open code it.
8207 	 */
8208 	smp_mb();
8209 
8210 	/* We need to check again in a case another CPU has just
8211 	 * made room available.
8212 	 */
8213 	if (likely(ixgbe_desc_unused(tx_ring) < size))
8214 		return -EBUSY;
8215 
8216 	/* A reprieve! - use start_queue because it doesn't call schedule */
8217 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8218 	++tx_ring->tx_stats.restart_queue;
8219 	return 0;
8220 }
8221 
8222 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8223 {
8224 	if (likely(ixgbe_desc_unused(tx_ring) >= size))
8225 		return 0;
8226 
8227 	return __ixgbe_maybe_stop_tx(tx_ring, size);
8228 }
8229 
8230 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8231 			struct ixgbe_tx_buffer *first,
8232 			const u8 hdr_len)
8233 {
8234 	struct sk_buff *skb = first->skb;
8235 	struct ixgbe_tx_buffer *tx_buffer;
8236 	union ixgbe_adv_tx_desc *tx_desc;
8237 	skb_frag_t *frag;
8238 	dma_addr_t dma;
8239 	unsigned int data_len, size;
8240 	u32 tx_flags = first->tx_flags;
8241 	u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8242 	u16 i = tx_ring->next_to_use;
8243 
8244 	tx_desc = IXGBE_TX_DESC(tx_ring, i);
8245 
8246 	ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8247 
8248 	size = skb_headlen(skb);
8249 	data_len = skb->data_len;
8250 
8251 #ifdef IXGBE_FCOE
8252 	if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8253 		if (data_len < sizeof(struct fcoe_crc_eof)) {
8254 			size -= sizeof(struct fcoe_crc_eof) - data_len;
8255 			data_len = 0;
8256 		} else {
8257 			data_len -= sizeof(struct fcoe_crc_eof);
8258 		}
8259 	}
8260 
8261 #endif
8262 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8263 
8264 	tx_buffer = first;
8265 
8266 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8267 		if (dma_mapping_error(tx_ring->dev, dma))
8268 			goto dma_error;
8269 
8270 		/* record length, and DMA address */
8271 		dma_unmap_len_set(tx_buffer, len, size);
8272 		dma_unmap_addr_set(tx_buffer, dma, dma);
8273 
8274 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
8275 
8276 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8277 			tx_desc->read.cmd_type_len =
8278 				cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8279 
8280 			i++;
8281 			tx_desc++;
8282 			if (i == tx_ring->count) {
8283 				tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8284 				i = 0;
8285 			}
8286 			tx_desc->read.olinfo_status = 0;
8287 
8288 			dma += IXGBE_MAX_DATA_PER_TXD;
8289 			size -= IXGBE_MAX_DATA_PER_TXD;
8290 
8291 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
8292 		}
8293 
8294 		if (likely(!data_len))
8295 			break;
8296 
8297 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8298 
8299 		i++;
8300 		tx_desc++;
8301 		if (i == tx_ring->count) {
8302 			tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8303 			i = 0;
8304 		}
8305 		tx_desc->read.olinfo_status = 0;
8306 
8307 #ifdef IXGBE_FCOE
8308 		size = min_t(unsigned int, data_len, skb_frag_size(frag));
8309 #else
8310 		size = skb_frag_size(frag);
8311 #endif
8312 		data_len -= size;
8313 
8314 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8315 				       DMA_TO_DEVICE);
8316 
8317 		tx_buffer = &tx_ring->tx_buffer_info[i];
8318 	}
8319 
8320 	/* write last descriptor with RS and EOP bits */
8321 	cmd_type |= size | IXGBE_TXD_CMD;
8322 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8323 
8324 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8325 
8326 	/* set the timestamp */
8327 	first->time_stamp = jiffies;
8328 
8329 	skb_tx_timestamp(skb);
8330 
8331 	/*
8332 	 * Force memory writes to complete before letting h/w know there
8333 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
8334 	 * memory model archs, such as IA-64).
8335 	 *
8336 	 * We also need this memory barrier to make certain all of the
8337 	 * status bits have been updated before next_to_watch is written.
8338 	 */
8339 	wmb();
8340 
8341 	/* set next_to_watch value indicating a packet is present */
8342 	first->next_to_watch = tx_desc;
8343 
8344 	i++;
8345 	if (i == tx_ring->count)
8346 		i = 0;
8347 
8348 	tx_ring->next_to_use = i;
8349 
8350 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8351 
8352 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8353 		writel(i, tx_ring->tail);
8354 	}
8355 
8356 	return 0;
8357 dma_error:
8358 	dev_err(tx_ring->dev, "TX DMA map failed\n");
8359 
8360 	/* clear dma mappings for failed tx_buffer_info map */
8361 	for (;;) {
8362 		tx_buffer = &tx_ring->tx_buffer_info[i];
8363 		if (dma_unmap_len(tx_buffer, len))
8364 			dma_unmap_page(tx_ring->dev,
8365 				       dma_unmap_addr(tx_buffer, dma),
8366 				       dma_unmap_len(tx_buffer, len),
8367 				       DMA_TO_DEVICE);
8368 		dma_unmap_len_set(tx_buffer, len, 0);
8369 		if (tx_buffer == first)
8370 			break;
8371 		if (i == 0)
8372 			i += tx_ring->count;
8373 		i--;
8374 	}
8375 
8376 	dev_kfree_skb_any(first->skb);
8377 	first->skb = NULL;
8378 
8379 	tx_ring->next_to_use = i;
8380 
8381 	return -1;
8382 }
8383 
8384 static void ixgbe_atr(struct ixgbe_ring *ring,
8385 		      struct ixgbe_tx_buffer *first)
8386 {
8387 	struct ixgbe_q_vector *q_vector = ring->q_vector;
8388 	union ixgbe_atr_hash_dword input = { .dword = 0 };
8389 	union ixgbe_atr_hash_dword common = { .dword = 0 };
8390 	union {
8391 		unsigned char *network;
8392 		struct iphdr *ipv4;
8393 		struct ipv6hdr *ipv6;
8394 	} hdr;
8395 	struct tcphdr *th;
8396 	unsigned int hlen;
8397 	struct sk_buff *skb;
8398 	__be16 vlan_id;
8399 	int l4_proto;
8400 
8401 	/* if ring doesn't have a interrupt vector, cannot perform ATR */
8402 	if (!q_vector)
8403 		return;
8404 
8405 	/* do nothing if sampling is disabled */
8406 	if (!ring->atr_sample_rate)
8407 		return;
8408 
8409 	ring->atr_count++;
8410 
8411 	/* currently only IPv4/IPv6 with TCP is supported */
8412 	if ((first->protocol != htons(ETH_P_IP)) &&
8413 	    (first->protocol != htons(ETH_P_IPV6)))
8414 		return;
8415 
8416 	/* snag network header to get L4 type and address */
8417 	skb = first->skb;
8418 	hdr.network = skb_network_header(skb);
8419 	if (unlikely(hdr.network <= skb->data))
8420 		return;
8421 	if (skb->encapsulation &&
8422 	    first->protocol == htons(ETH_P_IP) &&
8423 	    hdr.ipv4->protocol == IPPROTO_UDP) {
8424 		struct ixgbe_adapter *adapter = q_vector->adapter;
8425 
8426 		if (unlikely(skb_tail_pointer(skb) < hdr.network +
8427 			     VXLAN_HEADROOM))
8428 			return;
8429 
8430 		/* verify the port is recognized as VXLAN */
8431 		if (adapter->vxlan_port &&
8432 		    udp_hdr(skb)->dest == adapter->vxlan_port)
8433 			hdr.network = skb_inner_network_header(skb);
8434 
8435 		if (adapter->geneve_port &&
8436 		    udp_hdr(skb)->dest == adapter->geneve_port)
8437 			hdr.network = skb_inner_network_header(skb);
8438 	}
8439 
8440 	/* Make sure we have at least [minimum IPv4 header + TCP]
8441 	 * or [IPv6 header] bytes
8442 	 */
8443 	if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8444 		return;
8445 
8446 	/* Currently only IPv4/IPv6 with TCP is supported */
8447 	switch (hdr.ipv4->version) {
8448 	case IPVERSION:
8449 		/* access ihl as u8 to avoid unaligned access on ia64 */
8450 		hlen = (hdr.network[0] & 0x0F) << 2;
8451 		l4_proto = hdr.ipv4->protocol;
8452 		break;
8453 	case 6:
8454 		hlen = hdr.network - skb->data;
8455 		l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8456 		hlen -= hdr.network - skb->data;
8457 		break;
8458 	default:
8459 		return;
8460 	}
8461 
8462 	if (l4_proto != IPPROTO_TCP)
8463 		return;
8464 
8465 	if (unlikely(skb_tail_pointer(skb) < hdr.network +
8466 		     hlen + sizeof(struct tcphdr)))
8467 		return;
8468 
8469 	th = (struct tcphdr *)(hdr.network + hlen);
8470 
8471 	/* skip this packet since the socket is closing */
8472 	if (th->fin)
8473 		return;
8474 
8475 	/* sample on all syn packets or once every atr sample count */
8476 	if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8477 		return;
8478 
8479 	/* reset sample count */
8480 	ring->atr_count = 0;
8481 
8482 	vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8483 
8484 	/*
8485 	 * src and dst are inverted, think how the receiver sees them
8486 	 *
8487 	 * The input is broken into two sections, a non-compressed section
8488 	 * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8489 	 * is XORed together and stored in the compressed dword.
8490 	 */
8491 	input.formatted.vlan_id = vlan_id;
8492 
8493 	/*
8494 	 * since src port and flex bytes occupy the same word XOR them together
8495 	 * and write the value to source port portion of compressed dword
8496 	 */
8497 	if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8498 		common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8499 	else
8500 		common.port.src ^= th->dest ^ first->protocol;
8501 	common.port.dst ^= th->source;
8502 
8503 	switch (hdr.ipv4->version) {
8504 	case IPVERSION:
8505 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8506 		common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8507 		break;
8508 	case 6:
8509 		input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8510 		common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8511 			     hdr.ipv6->saddr.s6_addr32[1] ^
8512 			     hdr.ipv6->saddr.s6_addr32[2] ^
8513 			     hdr.ipv6->saddr.s6_addr32[3] ^
8514 			     hdr.ipv6->daddr.s6_addr32[0] ^
8515 			     hdr.ipv6->daddr.s6_addr32[1] ^
8516 			     hdr.ipv6->daddr.s6_addr32[2] ^
8517 			     hdr.ipv6->daddr.s6_addr32[3];
8518 		break;
8519 	default:
8520 		break;
8521 	}
8522 
8523 	if (hdr.network != skb_network_header(skb))
8524 		input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8525 
8526 	/* This assumes the Rx queue and Tx queue are bound to the same CPU */
8527 	ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8528 					      input, common, ring->queue_index);
8529 }
8530 
8531 #ifdef IXGBE_FCOE
8532 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8533 			      struct net_device *sb_dev)
8534 {
8535 	struct ixgbe_adapter *adapter;
8536 	struct ixgbe_ring_feature *f;
8537 	int txq;
8538 
8539 	if (sb_dev) {
8540 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8541 		struct net_device *vdev = sb_dev;
8542 
8543 		txq = vdev->tc_to_txq[tc].offset;
8544 		txq += reciprocal_scale(skb_get_hash(skb),
8545 					vdev->tc_to_txq[tc].count);
8546 
8547 		return txq;
8548 	}
8549 
8550 	/*
8551 	 * only execute the code below if protocol is FCoE
8552 	 * or FIP and we have FCoE enabled on the adapter
8553 	 */
8554 	switch (vlan_get_protocol(skb)) {
8555 	case htons(ETH_P_FCOE):
8556 	case htons(ETH_P_FIP):
8557 		adapter = netdev_priv(dev);
8558 
8559 		if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8560 			break;
8561 		/* fall through */
8562 	default:
8563 		return netdev_pick_tx(dev, skb, sb_dev);
8564 	}
8565 
8566 	f = &adapter->ring_feature[RING_F_FCOE];
8567 
8568 	txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8569 					   smp_processor_id();
8570 
8571 	while (txq >= f->indices)
8572 		txq -= f->indices;
8573 
8574 	return txq + f->offset;
8575 }
8576 
8577 #endif
8578 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8579 			struct xdp_frame *xdpf)
8580 {
8581 	struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8582 	struct ixgbe_tx_buffer *tx_buffer;
8583 	union ixgbe_adv_tx_desc *tx_desc;
8584 	u32 len, cmd_type;
8585 	dma_addr_t dma;
8586 	u16 i;
8587 
8588 	len = xdpf->len;
8589 
8590 	if (unlikely(!ixgbe_desc_unused(ring)))
8591 		return IXGBE_XDP_CONSUMED;
8592 
8593 	dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8594 	if (dma_mapping_error(ring->dev, dma))
8595 		return IXGBE_XDP_CONSUMED;
8596 
8597 	/* record the location of the first descriptor for this packet */
8598 	tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8599 	tx_buffer->bytecount = len;
8600 	tx_buffer->gso_segs = 1;
8601 	tx_buffer->protocol = 0;
8602 
8603 	i = ring->next_to_use;
8604 	tx_desc = IXGBE_TX_DESC(ring, i);
8605 
8606 	dma_unmap_len_set(tx_buffer, len, len);
8607 	dma_unmap_addr_set(tx_buffer, dma, dma);
8608 	tx_buffer->xdpf = xdpf;
8609 
8610 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
8611 
8612 	/* put descriptor type bits */
8613 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8614 		   IXGBE_ADVTXD_DCMD_DEXT |
8615 		   IXGBE_ADVTXD_DCMD_IFCS;
8616 	cmd_type |= len | IXGBE_TXD_CMD;
8617 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8618 	tx_desc->read.olinfo_status =
8619 		cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8620 
8621 	/* Avoid any potential race with xdp_xmit and cleanup */
8622 	smp_wmb();
8623 
8624 	/* set next_to_watch value indicating a packet is present */
8625 	i++;
8626 	if (i == ring->count)
8627 		i = 0;
8628 
8629 	tx_buffer->next_to_watch = tx_desc;
8630 	ring->next_to_use = i;
8631 
8632 	return IXGBE_XDP_TX;
8633 }
8634 
8635 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8636 			  struct ixgbe_adapter *adapter,
8637 			  struct ixgbe_ring *tx_ring)
8638 {
8639 	struct ixgbe_tx_buffer *first;
8640 	int tso;
8641 	u32 tx_flags = 0;
8642 	unsigned short f;
8643 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
8644 	struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8645 	__be16 protocol = skb->protocol;
8646 	u8 hdr_len = 0;
8647 
8648 	/*
8649 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8650 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8651 	 *       + 2 desc gap to keep tail from touching head,
8652 	 *       + 1 desc for context descriptor,
8653 	 * otherwise try next time
8654 	 */
8655 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8656 		count += TXD_USE_COUNT(skb_frag_size(
8657 						&skb_shinfo(skb)->frags[f]));
8658 
8659 	if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8660 		tx_ring->tx_stats.tx_busy++;
8661 		return NETDEV_TX_BUSY;
8662 	}
8663 
8664 	/* record the location of the first descriptor for this packet */
8665 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8666 	first->skb = skb;
8667 	first->bytecount = skb->len;
8668 	first->gso_segs = 1;
8669 
8670 	/* if we have a HW VLAN tag being added default to the HW one */
8671 	if (skb_vlan_tag_present(skb)) {
8672 		tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8673 		tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8674 	/* else if it is a SW VLAN check the next protocol and store the tag */
8675 	} else if (protocol == htons(ETH_P_8021Q)) {
8676 		struct vlan_hdr *vhdr, _vhdr;
8677 		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8678 		if (!vhdr)
8679 			goto out_drop;
8680 
8681 		tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8682 				  IXGBE_TX_FLAGS_VLAN_SHIFT;
8683 		tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8684 	}
8685 	protocol = vlan_get_protocol(skb);
8686 
8687 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8688 	    adapter->ptp_clock) {
8689 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
8690 		    !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8691 					   &adapter->state)) {
8692 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8693 			tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8694 
8695 			/* schedule check for Tx timestamp */
8696 			adapter->ptp_tx_skb = skb_get(skb);
8697 			adapter->ptp_tx_start = jiffies;
8698 			schedule_work(&adapter->ptp_tx_work);
8699 		} else {
8700 			adapter->tx_hwtstamp_skipped++;
8701 		}
8702 	}
8703 
8704 #ifdef CONFIG_PCI_IOV
8705 	/*
8706 	 * Use the l2switch_enable flag - would be false if the DMA
8707 	 * Tx switch had been disabled.
8708 	 */
8709 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8710 		tx_flags |= IXGBE_TX_FLAGS_CC;
8711 
8712 #endif
8713 	/* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8714 	if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8715 	    ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8716 	     (skb->priority != TC_PRIO_CONTROL))) {
8717 		tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8718 		tx_flags |= (skb->priority & 0x7) <<
8719 					IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8720 		if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8721 			struct vlan_ethhdr *vhdr;
8722 
8723 			if (skb_cow_head(skb, 0))
8724 				goto out_drop;
8725 			vhdr = (struct vlan_ethhdr *)skb->data;
8726 			vhdr->h_vlan_TCI = htons(tx_flags >>
8727 						 IXGBE_TX_FLAGS_VLAN_SHIFT);
8728 		} else {
8729 			tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8730 		}
8731 	}
8732 
8733 	/* record initial flags and protocol */
8734 	first->tx_flags = tx_flags;
8735 	first->protocol = protocol;
8736 
8737 #ifdef IXGBE_FCOE
8738 	/* setup tx offload for FCoE */
8739 	if ((protocol == htons(ETH_P_FCOE)) &&
8740 	    (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8741 		tso = ixgbe_fso(tx_ring, first, &hdr_len);
8742 		if (tso < 0)
8743 			goto out_drop;
8744 
8745 		goto xmit_fcoe;
8746 	}
8747 
8748 #endif /* IXGBE_FCOE */
8749 
8750 #ifdef CONFIG_IXGBE_IPSEC
8751 	if (xfrm_offload(skb) &&
8752 	    !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8753 		goto out_drop;
8754 #endif
8755 	tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8756 	if (tso < 0)
8757 		goto out_drop;
8758 	else if (!tso)
8759 		ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8760 
8761 	/* add the ATR filter if ATR is on */
8762 	if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8763 		ixgbe_atr(tx_ring, first);
8764 
8765 #ifdef IXGBE_FCOE
8766 xmit_fcoe:
8767 #endif /* IXGBE_FCOE */
8768 	if (ixgbe_tx_map(tx_ring, first, hdr_len))
8769 		goto cleanup_tx_timestamp;
8770 
8771 	return NETDEV_TX_OK;
8772 
8773 out_drop:
8774 	dev_kfree_skb_any(first->skb);
8775 	first->skb = NULL;
8776 cleanup_tx_timestamp:
8777 	if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8778 		dev_kfree_skb_any(adapter->ptp_tx_skb);
8779 		adapter->ptp_tx_skb = NULL;
8780 		cancel_work_sync(&adapter->ptp_tx_work);
8781 		clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8782 	}
8783 
8784 	return NETDEV_TX_OK;
8785 }
8786 
8787 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8788 				      struct net_device *netdev,
8789 				      struct ixgbe_ring *ring)
8790 {
8791 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8792 	struct ixgbe_ring *tx_ring;
8793 
8794 	/*
8795 	 * The minimum packet size for olinfo paylen is 17 so pad the skb
8796 	 * in order to meet this minimum size requirement.
8797 	 */
8798 	if (skb_put_padto(skb, 17))
8799 		return NETDEV_TX_OK;
8800 
8801 	tx_ring = ring ? ring : adapter->tx_ring[skb_get_queue_mapping(skb)];
8802 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8803 		return NETDEV_TX_BUSY;
8804 
8805 	return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8806 }
8807 
8808 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8809 				    struct net_device *netdev)
8810 {
8811 	return __ixgbe_xmit_frame(skb, netdev, NULL);
8812 }
8813 
8814 /**
8815  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8816  * @netdev: network interface device structure
8817  * @p: pointer to an address structure
8818  *
8819  * Returns 0 on success, negative on failure
8820  **/
8821 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8822 {
8823 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8824 	struct ixgbe_hw *hw = &adapter->hw;
8825 	struct sockaddr *addr = p;
8826 
8827 	if (!is_valid_ether_addr(addr->sa_data))
8828 		return -EADDRNOTAVAIL;
8829 
8830 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8831 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8832 
8833 	ixgbe_mac_set_default_filter(adapter);
8834 
8835 	return 0;
8836 }
8837 
8838 static int
8839 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8840 {
8841 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8842 	struct ixgbe_hw *hw = &adapter->hw;
8843 	u16 value;
8844 	int rc;
8845 
8846 	if (adapter->mii_bus) {
8847 		int regnum = addr;
8848 
8849 		if (devad != MDIO_DEVAD_NONE)
8850 			regnum |= (devad << 16) | MII_ADDR_C45;
8851 
8852 		return mdiobus_read(adapter->mii_bus, prtad, regnum);
8853 	}
8854 
8855 	if (prtad != hw->phy.mdio.prtad)
8856 		return -EINVAL;
8857 	rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8858 	if (!rc)
8859 		rc = value;
8860 	return rc;
8861 }
8862 
8863 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8864 			    u16 addr, u16 value)
8865 {
8866 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8867 	struct ixgbe_hw *hw = &adapter->hw;
8868 
8869 	if (adapter->mii_bus) {
8870 		int regnum = addr;
8871 
8872 		if (devad != MDIO_DEVAD_NONE)
8873 			regnum |= (devad << 16) | MII_ADDR_C45;
8874 
8875 		return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8876 	}
8877 
8878 	if (prtad != hw->phy.mdio.prtad)
8879 		return -EINVAL;
8880 	return hw->phy.ops.write_reg(hw, addr, devad, value);
8881 }
8882 
8883 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8884 {
8885 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8886 
8887 	switch (cmd) {
8888 	case SIOCSHWTSTAMP:
8889 		return ixgbe_ptp_set_ts_config(adapter, req);
8890 	case SIOCGHWTSTAMP:
8891 		return ixgbe_ptp_get_ts_config(adapter, req);
8892 	case SIOCGMIIPHY:
8893 		if (!adapter->hw.phy.ops.read_reg)
8894 			return -EOPNOTSUPP;
8895 		/* fall through */
8896 	default:
8897 		return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8898 	}
8899 }
8900 
8901 /**
8902  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8903  * netdev->dev_addrs
8904  * @dev: network interface device structure
8905  *
8906  * Returns non-zero on failure
8907  **/
8908 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8909 {
8910 	int err = 0;
8911 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8912 	struct ixgbe_hw *hw = &adapter->hw;
8913 
8914 	if (is_valid_ether_addr(hw->mac.san_addr)) {
8915 		rtnl_lock();
8916 		err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8917 		rtnl_unlock();
8918 
8919 		/* update SAN MAC vmdq pool selection */
8920 		hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8921 	}
8922 	return err;
8923 }
8924 
8925 /**
8926  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8927  * netdev->dev_addrs
8928  * @dev: network interface device structure
8929  *
8930  * Returns non-zero on failure
8931  **/
8932 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8933 {
8934 	int err = 0;
8935 	struct ixgbe_adapter *adapter = netdev_priv(dev);
8936 	struct ixgbe_mac_info *mac = &adapter->hw.mac;
8937 
8938 	if (is_valid_ether_addr(mac->san_addr)) {
8939 		rtnl_lock();
8940 		err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8941 		rtnl_unlock();
8942 	}
8943 	return err;
8944 }
8945 
8946 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8947 				   struct ixgbe_ring *ring)
8948 {
8949 	u64 bytes, packets;
8950 	unsigned int start;
8951 
8952 	if (ring) {
8953 		do {
8954 			start = u64_stats_fetch_begin_irq(&ring->syncp);
8955 			packets = ring->stats.packets;
8956 			bytes   = ring->stats.bytes;
8957 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8958 		stats->tx_packets += packets;
8959 		stats->tx_bytes   += bytes;
8960 	}
8961 }
8962 
8963 static void ixgbe_get_stats64(struct net_device *netdev,
8964 			      struct rtnl_link_stats64 *stats)
8965 {
8966 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
8967 	int i;
8968 
8969 	rcu_read_lock();
8970 	for (i = 0; i < adapter->num_rx_queues; i++) {
8971 		struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8972 		u64 bytes, packets;
8973 		unsigned int start;
8974 
8975 		if (ring) {
8976 			do {
8977 				start = u64_stats_fetch_begin_irq(&ring->syncp);
8978 				packets = ring->stats.packets;
8979 				bytes   = ring->stats.bytes;
8980 			} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8981 			stats->rx_packets += packets;
8982 			stats->rx_bytes   += bytes;
8983 		}
8984 	}
8985 
8986 	for (i = 0; i < adapter->num_tx_queues; i++) {
8987 		struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8988 
8989 		ixgbe_get_ring_stats64(stats, ring);
8990 	}
8991 	for (i = 0; i < adapter->num_xdp_queues; i++) {
8992 		struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8993 
8994 		ixgbe_get_ring_stats64(stats, ring);
8995 	}
8996 	rcu_read_unlock();
8997 
8998 	/* following stats updated by ixgbe_watchdog_task() */
8999 	stats->multicast	= netdev->stats.multicast;
9000 	stats->rx_errors	= netdev->stats.rx_errors;
9001 	stats->rx_length_errors	= netdev->stats.rx_length_errors;
9002 	stats->rx_crc_errors	= netdev->stats.rx_crc_errors;
9003 	stats->rx_missed_errors	= netdev->stats.rx_missed_errors;
9004 }
9005 
9006 #ifdef CONFIG_IXGBE_DCB
9007 /**
9008  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
9009  * @adapter: pointer to ixgbe_adapter
9010  * @tc: number of traffic classes currently enabled
9011  *
9012  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
9013  * 802.1Q priority maps to a packet buffer that exists.
9014  */
9015 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
9016 {
9017 	struct ixgbe_hw *hw = &adapter->hw;
9018 	u32 reg, rsave;
9019 	int i;
9020 
9021 	/* 82598 have a static priority to TC mapping that can not
9022 	 * be changed so no validation is needed.
9023 	 */
9024 	if (hw->mac.type == ixgbe_mac_82598EB)
9025 		return;
9026 
9027 	reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
9028 	rsave = reg;
9029 
9030 	for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
9031 		u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
9032 
9033 		/* If up2tc is out of bounds default to zero */
9034 		if (up2tc > tc)
9035 			reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
9036 	}
9037 
9038 	if (reg != rsave)
9039 		IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
9040 
9041 	return;
9042 }
9043 
9044 /**
9045  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
9046  * @adapter: Pointer to adapter struct
9047  *
9048  * Populate the netdev user priority to tc map
9049  */
9050 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
9051 {
9052 	struct net_device *dev = adapter->netdev;
9053 	struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9054 	struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9055 	u8 prio;
9056 
9057 	for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9058 		u8 tc = 0;
9059 
9060 		if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9061 			tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9062 		else if (ets)
9063 			tc = ets->prio_tc[prio];
9064 
9065 		netdev_set_prio_tc_map(dev, prio, tc);
9066 	}
9067 }
9068 
9069 #endif /* CONFIG_IXGBE_DCB */
9070 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9071 {
9072 	struct ixgbe_adapter *adapter = data;
9073 	struct ixgbe_fwd_adapter *accel;
9074 	int pool;
9075 
9076 	/* we only care about macvlans... */
9077 	if (!netif_is_macvlan(vdev))
9078 		return 0;
9079 
9080 	/* that have hardware offload enabled... */
9081 	accel = macvlan_accel_priv(vdev);
9082 	if (!accel)
9083 		return 0;
9084 
9085 	/* If we can relocate to a different bit do so */
9086 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9087 	if (pool < adapter->num_rx_pools) {
9088 		set_bit(pool, adapter->fwd_bitmask);
9089 		accel->pool = pool;
9090 		return 0;
9091 	}
9092 
9093 	/* if we cannot find a free pool then disable the offload */
9094 	netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9095 	macvlan_release_l2fw_offload(vdev);
9096 
9097 	/* unbind the queues and drop the subordinate channel config */
9098 	netdev_unbind_sb_channel(adapter->netdev, vdev);
9099 	netdev_set_sb_channel(vdev, 0);
9100 
9101 	kfree(accel);
9102 
9103 	return 0;
9104 }
9105 
9106 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9107 {
9108 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9109 
9110 	/* flush any stale bits out of the fwd bitmask */
9111 	bitmap_clear(adapter->fwd_bitmask, 1, 63);
9112 
9113 	/* walk through upper devices reassigning pools */
9114 	netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9115 				      adapter);
9116 }
9117 
9118 /**
9119  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9120  *
9121  * @dev: net device to configure
9122  * @tc: number of traffic classes to enable
9123  */
9124 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9125 {
9126 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9127 	struct ixgbe_hw *hw = &adapter->hw;
9128 
9129 	/* Hardware supports up to 8 traffic classes */
9130 	if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9131 		return -EINVAL;
9132 
9133 	if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9134 		return -EINVAL;
9135 
9136 	/* Hardware has to reinitialize queues and interrupts to
9137 	 * match packet buffer alignment. Unfortunately, the
9138 	 * hardware is not flexible enough to do this dynamically.
9139 	 */
9140 	if (netif_running(dev))
9141 		ixgbe_close(dev);
9142 	else
9143 		ixgbe_reset(adapter);
9144 
9145 	ixgbe_clear_interrupt_scheme(adapter);
9146 
9147 #ifdef CONFIG_IXGBE_DCB
9148 	if (tc) {
9149 		if (adapter->xdp_prog) {
9150 			e_warn(probe, "DCB is not supported with XDP\n");
9151 
9152 			ixgbe_init_interrupt_scheme(adapter);
9153 			if (netif_running(dev))
9154 				ixgbe_open(dev);
9155 			return -EINVAL;
9156 		}
9157 
9158 		netdev_set_num_tc(dev, tc);
9159 		ixgbe_set_prio_tc_map(adapter);
9160 
9161 		adapter->hw_tcs = tc;
9162 		adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9163 
9164 		if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9165 			adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9166 			adapter->hw.fc.requested_mode = ixgbe_fc_none;
9167 		}
9168 	} else {
9169 		netdev_reset_tc(dev);
9170 
9171 		if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9172 			adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9173 
9174 		adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9175 		adapter->hw_tcs = tc;
9176 
9177 		adapter->temp_dcb_cfg.pfc_mode_enable = false;
9178 		adapter->dcb_cfg.pfc_mode_enable = false;
9179 	}
9180 
9181 	ixgbe_validate_rtr(adapter, tc);
9182 
9183 #endif /* CONFIG_IXGBE_DCB */
9184 	ixgbe_init_interrupt_scheme(adapter);
9185 
9186 	ixgbe_defrag_macvlan_pools(dev);
9187 
9188 	if (netif_running(dev))
9189 		return ixgbe_open(dev);
9190 
9191 	return 0;
9192 }
9193 
9194 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9195 			       struct tc_cls_u32_offload *cls)
9196 {
9197 	u32 hdl = cls->knode.handle;
9198 	u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9199 	u32 loc = cls->knode.handle & 0xfffff;
9200 	int err = 0, i, j;
9201 	struct ixgbe_jump_table *jump = NULL;
9202 
9203 	if (loc > IXGBE_MAX_HW_ENTRIES)
9204 		return -EINVAL;
9205 
9206 	if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9207 		return -EINVAL;
9208 
9209 	/* Clear this filter in the link data it is associated with */
9210 	if (uhtid != 0x800) {
9211 		jump = adapter->jump_tables[uhtid];
9212 		if (!jump)
9213 			return -EINVAL;
9214 		if (!test_bit(loc - 1, jump->child_loc_map))
9215 			return -EINVAL;
9216 		clear_bit(loc - 1, jump->child_loc_map);
9217 	}
9218 
9219 	/* Check if the filter being deleted is a link */
9220 	for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9221 		jump = adapter->jump_tables[i];
9222 		if (jump && jump->link_hdl == hdl) {
9223 			/* Delete filters in the hardware in the child hash
9224 			 * table associated with this link
9225 			 */
9226 			for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9227 				if (!test_bit(j, jump->child_loc_map))
9228 					continue;
9229 				spin_lock(&adapter->fdir_perfect_lock);
9230 				err = ixgbe_update_ethtool_fdir_entry(adapter,
9231 								      NULL,
9232 								      j + 1);
9233 				spin_unlock(&adapter->fdir_perfect_lock);
9234 				clear_bit(j, jump->child_loc_map);
9235 			}
9236 			/* Remove resources for this link */
9237 			kfree(jump->input);
9238 			kfree(jump->mask);
9239 			kfree(jump);
9240 			adapter->jump_tables[i] = NULL;
9241 			return err;
9242 		}
9243 	}
9244 
9245 	spin_lock(&adapter->fdir_perfect_lock);
9246 	err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9247 	spin_unlock(&adapter->fdir_perfect_lock);
9248 	return err;
9249 }
9250 
9251 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9252 					    struct tc_cls_u32_offload *cls)
9253 {
9254 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9255 
9256 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9257 		return -EINVAL;
9258 
9259 	/* This ixgbe devices do not support hash tables at the moment
9260 	 * so abort when given hash tables.
9261 	 */
9262 	if (cls->hnode.divisor > 0)
9263 		return -EINVAL;
9264 
9265 	set_bit(uhtid - 1, &adapter->tables);
9266 	return 0;
9267 }
9268 
9269 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
9270 					    struct tc_cls_u32_offload *cls)
9271 {
9272 	u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9273 
9274 	if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9275 		return -EINVAL;
9276 
9277 	clear_bit(uhtid - 1, &adapter->tables);
9278 	return 0;
9279 }
9280 
9281 #ifdef CONFIG_NET_CLS_ACT
9282 struct upper_walk_data {
9283 	struct ixgbe_adapter *adapter;
9284 	u64 action;
9285 	int ifindex;
9286 	u8 queue;
9287 };
9288 
9289 static int get_macvlan_queue(struct net_device *upper, void *_data)
9290 {
9291 	if (netif_is_macvlan(upper)) {
9292 		struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9293 		struct upper_walk_data *data = _data;
9294 		struct ixgbe_adapter *adapter = data->adapter;
9295 		int ifindex = data->ifindex;
9296 
9297 		if (vadapter && upper->ifindex == ifindex) {
9298 			data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9299 			data->action = data->queue;
9300 			return 1;
9301 		}
9302 	}
9303 
9304 	return 0;
9305 }
9306 
9307 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9308 				  u8 *queue, u64 *action)
9309 {
9310 	struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9311 	unsigned int num_vfs = adapter->num_vfs, vf;
9312 	struct upper_walk_data data;
9313 	struct net_device *upper;
9314 
9315 	/* redirect to a SRIOV VF */
9316 	for (vf = 0; vf < num_vfs; ++vf) {
9317 		upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9318 		if (upper->ifindex == ifindex) {
9319 			*queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9320 			*action = vf + 1;
9321 			*action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9322 			return 0;
9323 		}
9324 	}
9325 
9326 	/* redirect to a offloaded macvlan netdev */
9327 	data.adapter = adapter;
9328 	data.ifindex = ifindex;
9329 	data.action = 0;
9330 	data.queue = 0;
9331 	if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9332 					  get_macvlan_queue, &data)) {
9333 		*action = data.action;
9334 		*queue = data.queue;
9335 
9336 		return 0;
9337 	}
9338 
9339 	return -EINVAL;
9340 }
9341 
9342 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9343 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9344 {
9345 	const struct tc_action *a;
9346 	int i;
9347 
9348 	if (!tcf_exts_has_actions(exts))
9349 		return -EINVAL;
9350 
9351 	tcf_exts_for_each_action(i, a, exts) {
9352 		/* Drop action */
9353 		if (is_tcf_gact_shot(a)) {
9354 			*action = IXGBE_FDIR_DROP_QUEUE;
9355 			*queue = IXGBE_FDIR_DROP_QUEUE;
9356 			return 0;
9357 		}
9358 
9359 		/* Redirect to a VF or a offloaded macvlan */
9360 		if (is_tcf_mirred_egress_redirect(a)) {
9361 			struct net_device *dev = tcf_mirred_dev(a);
9362 
9363 			if (!dev)
9364 				return -EINVAL;
9365 			return handle_redirect_action(adapter, dev->ifindex,
9366 						      queue, action);
9367 		}
9368 
9369 		return -EINVAL;
9370 	}
9371 
9372 	return -EINVAL;
9373 }
9374 #else
9375 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9376 			    struct tcf_exts *exts, u64 *action, u8 *queue)
9377 {
9378 	return -EINVAL;
9379 }
9380 #endif /* CONFIG_NET_CLS_ACT */
9381 
9382 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9383 				    union ixgbe_atr_input *mask,
9384 				    struct tc_cls_u32_offload *cls,
9385 				    struct ixgbe_mat_field *field_ptr,
9386 				    struct ixgbe_nexthdr *nexthdr)
9387 {
9388 	int i, j, off;
9389 	__be32 val, m;
9390 	bool found_entry = false, found_jump_field = false;
9391 
9392 	for (i = 0; i < cls->knode.sel->nkeys; i++) {
9393 		off = cls->knode.sel->keys[i].off;
9394 		val = cls->knode.sel->keys[i].val;
9395 		m = cls->knode.sel->keys[i].mask;
9396 
9397 		for (j = 0; field_ptr[j].val; j++) {
9398 			if (field_ptr[j].off == off) {
9399 				field_ptr[j].val(input, mask, (__force u32)val,
9400 						 (__force u32)m);
9401 				input->filter.formatted.flow_type |=
9402 					field_ptr[j].type;
9403 				found_entry = true;
9404 				break;
9405 			}
9406 		}
9407 		if (nexthdr) {
9408 			if (nexthdr->off == cls->knode.sel->keys[i].off &&
9409 			    nexthdr->val ==
9410 			    (__force u32)cls->knode.sel->keys[i].val &&
9411 			    nexthdr->mask ==
9412 			    (__force u32)cls->knode.sel->keys[i].mask)
9413 				found_jump_field = true;
9414 			else
9415 				continue;
9416 		}
9417 	}
9418 
9419 	if (nexthdr && !found_jump_field)
9420 		return -EINVAL;
9421 
9422 	if (!found_entry)
9423 		return 0;
9424 
9425 	mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9426 				    IXGBE_ATR_L4TYPE_MASK;
9427 
9428 	if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9429 		mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9430 
9431 	return 0;
9432 }
9433 
9434 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9435 				  struct tc_cls_u32_offload *cls)
9436 {
9437 	__be16 protocol = cls->common.protocol;
9438 	u32 loc = cls->knode.handle & 0xfffff;
9439 	struct ixgbe_hw *hw = &adapter->hw;
9440 	struct ixgbe_mat_field *field_ptr;
9441 	struct ixgbe_fdir_filter *input = NULL;
9442 	union ixgbe_atr_input *mask = NULL;
9443 	struct ixgbe_jump_table *jump = NULL;
9444 	int i, err = -EINVAL;
9445 	u8 queue;
9446 	u32 uhtid, link_uhtid;
9447 
9448 	uhtid = TC_U32_USERHTID(cls->knode.handle);
9449 	link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9450 
9451 	/* At the moment cls_u32 jumps to network layer and skips past
9452 	 * L2 headers. The canonical method to match L2 frames is to use
9453 	 * negative values. However this is error prone at best but really
9454 	 * just broken because there is no way to "know" what sort of hdr
9455 	 * is in front of the network layer. Fix cls_u32 to support L2
9456 	 * headers when needed.
9457 	 */
9458 	if (protocol != htons(ETH_P_IP))
9459 		return err;
9460 
9461 	if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9462 		e_err(drv, "Location out of range\n");
9463 		return err;
9464 	}
9465 
9466 	/* cls u32 is a graph starting at root node 0x800. The driver tracks
9467 	 * links and also the fields used to advance the parser across each
9468 	 * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9469 	 * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9470 	 * To add support for new nodes update ixgbe_model.h parse structures
9471 	 * this function _should_ be generic try not to hardcode values here.
9472 	 */
9473 	if (uhtid == 0x800) {
9474 		field_ptr = (adapter->jump_tables[0])->mat;
9475 	} else {
9476 		if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9477 			return err;
9478 		if (!adapter->jump_tables[uhtid])
9479 			return err;
9480 		field_ptr = (adapter->jump_tables[uhtid])->mat;
9481 	}
9482 
9483 	if (!field_ptr)
9484 		return err;
9485 
9486 	/* At this point we know the field_ptr is valid and need to either
9487 	 * build cls_u32 link or attach filter. Because adding a link to
9488 	 * a handle that does not exist is invalid and the same for adding
9489 	 * rules to handles that don't exist.
9490 	 */
9491 
9492 	if (link_uhtid) {
9493 		struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9494 
9495 		if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9496 			return err;
9497 
9498 		if (!test_bit(link_uhtid - 1, &adapter->tables))
9499 			return err;
9500 
9501 		/* Multiple filters as links to the same hash table are not
9502 		 * supported. To add a new filter with the same next header
9503 		 * but different match/jump conditions, create a new hash table
9504 		 * and link to it.
9505 		 */
9506 		if (adapter->jump_tables[link_uhtid] &&
9507 		    (adapter->jump_tables[link_uhtid])->link_hdl) {
9508 			e_err(drv, "Link filter exists for link: %x\n",
9509 			      link_uhtid);
9510 			return err;
9511 		}
9512 
9513 		for (i = 0; nexthdr[i].jump; i++) {
9514 			if (nexthdr[i].o != cls->knode.sel->offoff ||
9515 			    nexthdr[i].s != cls->knode.sel->offshift ||
9516 			    nexthdr[i].m !=
9517 			    (__force u32)cls->knode.sel->offmask)
9518 				return err;
9519 
9520 			jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9521 			if (!jump)
9522 				return -ENOMEM;
9523 			input = kzalloc(sizeof(*input), GFP_KERNEL);
9524 			if (!input) {
9525 				err = -ENOMEM;
9526 				goto free_jump;
9527 			}
9528 			mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9529 			if (!mask) {
9530 				err = -ENOMEM;
9531 				goto free_input;
9532 			}
9533 			jump->input = input;
9534 			jump->mask = mask;
9535 			jump->link_hdl = cls->knode.handle;
9536 
9537 			err = ixgbe_clsu32_build_input(input, mask, cls,
9538 						       field_ptr, &nexthdr[i]);
9539 			if (!err) {
9540 				jump->mat = nexthdr[i].jump;
9541 				adapter->jump_tables[link_uhtid] = jump;
9542 				break;
9543 			} else {
9544 				kfree(mask);
9545 				kfree(input);
9546 				kfree(jump);
9547 			}
9548 		}
9549 		return 0;
9550 	}
9551 
9552 	input = kzalloc(sizeof(*input), GFP_KERNEL);
9553 	if (!input)
9554 		return -ENOMEM;
9555 	mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9556 	if (!mask) {
9557 		err = -ENOMEM;
9558 		goto free_input;
9559 	}
9560 
9561 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9562 		if ((adapter->jump_tables[uhtid])->input)
9563 			memcpy(input, (adapter->jump_tables[uhtid])->input,
9564 			       sizeof(*input));
9565 		if ((adapter->jump_tables[uhtid])->mask)
9566 			memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9567 			       sizeof(*mask));
9568 
9569 		/* Lookup in all child hash tables if this location is already
9570 		 * filled with a filter
9571 		 */
9572 		for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9573 			struct ixgbe_jump_table *link = adapter->jump_tables[i];
9574 
9575 			if (link && (test_bit(loc - 1, link->child_loc_map))) {
9576 				e_err(drv, "Filter exists in location: %x\n",
9577 				      loc);
9578 				err = -EINVAL;
9579 				goto err_out;
9580 			}
9581 		}
9582 	}
9583 	err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9584 	if (err)
9585 		goto err_out;
9586 
9587 	err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9588 			       &queue);
9589 	if (err < 0)
9590 		goto err_out;
9591 
9592 	input->sw_idx = loc;
9593 
9594 	spin_lock(&adapter->fdir_perfect_lock);
9595 
9596 	if (hlist_empty(&adapter->fdir_filter_list)) {
9597 		memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9598 		err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9599 		if (err)
9600 			goto err_out_w_lock;
9601 	} else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9602 		err = -EINVAL;
9603 		goto err_out_w_lock;
9604 	}
9605 
9606 	ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9607 	err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9608 						    input->sw_idx, queue);
9609 	if (!err)
9610 		ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9611 	spin_unlock(&adapter->fdir_perfect_lock);
9612 
9613 	if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9614 		set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9615 
9616 	kfree(mask);
9617 	return err;
9618 err_out_w_lock:
9619 	spin_unlock(&adapter->fdir_perfect_lock);
9620 err_out:
9621 	kfree(mask);
9622 free_input:
9623 	kfree(input);
9624 free_jump:
9625 	kfree(jump);
9626 	return err;
9627 }
9628 
9629 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9630 				  struct tc_cls_u32_offload *cls_u32)
9631 {
9632 	switch (cls_u32->command) {
9633 	case TC_CLSU32_NEW_KNODE:
9634 	case TC_CLSU32_REPLACE_KNODE:
9635 		return ixgbe_configure_clsu32(adapter, cls_u32);
9636 	case TC_CLSU32_DELETE_KNODE:
9637 		return ixgbe_delete_clsu32(adapter, cls_u32);
9638 	case TC_CLSU32_NEW_HNODE:
9639 	case TC_CLSU32_REPLACE_HNODE:
9640 		return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9641 	case TC_CLSU32_DELETE_HNODE:
9642 		return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9643 	default:
9644 		return -EOPNOTSUPP;
9645 	}
9646 }
9647 
9648 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9649 				   void *cb_priv)
9650 {
9651 	struct ixgbe_adapter *adapter = cb_priv;
9652 
9653 	if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9654 		return -EOPNOTSUPP;
9655 
9656 	switch (type) {
9657 	case TC_SETUP_CLSU32:
9658 		return ixgbe_setup_tc_cls_u32(adapter, type_data);
9659 	default:
9660 		return -EOPNOTSUPP;
9661 	}
9662 }
9663 
9664 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9665 				 struct tc_mqprio_qopt *mqprio)
9666 {
9667 	mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9668 	return ixgbe_setup_tc(dev, mqprio->num_tc);
9669 }
9670 
9671 static LIST_HEAD(ixgbe_block_cb_list);
9672 
9673 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9674 			    void *type_data)
9675 {
9676 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9677 
9678 	switch (type) {
9679 	case TC_SETUP_BLOCK:
9680 		return flow_block_cb_setup_simple(type_data,
9681 						  &ixgbe_block_cb_list,
9682 						  ixgbe_setup_tc_block_cb,
9683 						  adapter, adapter, true);
9684 	case TC_SETUP_QDISC_MQPRIO:
9685 		return ixgbe_setup_tc_mqprio(dev, type_data);
9686 	default:
9687 		return -EOPNOTSUPP;
9688 	}
9689 }
9690 
9691 #ifdef CONFIG_PCI_IOV
9692 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9693 {
9694 	struct net_device *netdev = adapter->netdev;
9695 
9696 	rtnl_lock();
9697 	ixgbe_setup_tc(netdev, adapter->hw_tcs);
9698 	rtnl_unlock();
9699 }
9700 
9701 #endif
9702 void ixgbe_do_reset(struct net_device *netdev)
9703 {
9704 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9705 
9706 	if (netif_running(netdev))
9707 		ixgbe_reinit_locked(adapter);
9708 	else
9709 		ixgbe_reset(adapter);
9710 }
9711 
9712 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9713 					    netdev_features_t features)
9714 {
9715 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9716 
9717 	/* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9718 	if (!(features & NETIF_F_RXCSUM))
9719 		features &= ~NETIF_F_LRO;
9720 
9721 	/* Turn off LRO if not RSC capable */
9722 	if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9723 		features &= ~NETIF_F_LRO;
9724 
9725 	if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9726 		e_dev_err("LRO is not supported with XDP\n");
9727 		features &= ~NETIF_F_LRO;
9728 	}
9729 
9730 	return features;
9731 }
9732 
9733 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9734 {
9735 	int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9736 			num_online_cpus());
9737 
9738 	/* go back to full RSS if we're not running SR-IOV */
9739 	if (!adapter->ring_feature[RING_F_VMDQ].offset)
9740 		adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9741 				    IXGBE_FLAG_SRIOV_ENABLED);
9742 
9743 	adapter->ring_feature[RING_F_RSS].limit = rss;
9744 	adapter->ring_feature[RING_F_VMDQ].limit = 1;
9745 
9746 	ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9747 }
9748 
9749 static int ixgbe_set_features(struct net_device *netdev,
9750 			      netdev_features_t features)
9751 {
9752 	struct ixgbe_adapter *adapter = netdev_priv(netdev);
9753 	netdev_features_t changed = netdev->features ^ features;
9754 	bool need_reset = false;
9755 
9756 	/* Make sure RSC matches LRO, reset if change */
9757 	if (!(features & NETIF_F_LRO)) {
9758 		if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9759 			need_reset = true;
9760 		adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9761 	} else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9762 		   !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9763 		if (adapter->rx_itr_setting == 1 ||
9764 		    adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9765 			adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9766 			need_reset = true;
9767 		} else if ((changed ^ features) & NETIF_F_LRO) {
9768 			e_info(probe, "rx-usecs set too low, "
9769 			       "disabling RSC\n");
9770 		}
9771 	}
9772 
9773 	/*
9774 	 * Check if Flow Director n-tuple support or hw_tc support was
9775 	 * enabled or disabled.  If the state changed, we need to reset.
9776 	 */
9777 	if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9778 		/* turn off ATR, enable perfect filters and reset */
9779 		if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9780 			need_reset = true;
9781 
9782 		adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9783 		adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9784 	} else {
9785 		/* turn off perfect filters, enable ATR and reset */
9786 		if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9787 			need_reset = true;
9788 
9789 		adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9790 
9791 		/* We cannot enable ATR if SR-IOV is enabled */
9792 		if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9793 		    /* We cannot enable ATR if we have 2 or more tcs */
9794 		    (adapter->hw_tcs > 1) ||
9795 		    /* We cannot enable ATR if RSS is disabled */
9796 		    (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9797 		    /* A sample rate of 0 indicates ATR disabled */
9798 		    (!adapter->atr_sample_rate))
9799 			; /* do nothing not supported */
9800 		else /* otherwise supported and set the flag */
9801 			adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9802 	}
9803 
9804 	if (changed & NETIF_F_RXALL)
9805 		need_reset = true;
9806 
9807 	netdev->features = features;
9808 
9809 	if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9810 		if (features & NETIF_F_RXCSUM) {
9811 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9812 		} else {
9813 			u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9814 
9815 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9816 		}
9817 	}
9818 
9819 	if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9820 		if (features & NETIF_F_RXCSUM) {
9821 			adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9822 		} else {
9823 			u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9824 
9825 			ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9826 		}
9827 	}
9828 
9829 	if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9830 		ixgbe_reset_l2fw_offload(adapter);
9831 	else if (need_reset)
9832 		ixgbe_do_reset(netdev);
9833 	else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9834 			    NETIF_F_HW_VLAN_CTAG_FILTER))
9835 		ixgbe_set_rx_mode(netdev);
9836 
9837 	return 1;
9838 }
9839 
9840 /**
9841  * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9842  * @dev: The port's netdev
9843  * @ti: Tunnel endpoint information
9844  **/
9845 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9846 				      struct udp_tunnel_info *ti)
9847 {
9848 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9849 	struct ixgbe_hw *hw = &adapter->hw;
9850 	__be16 port = ti->port;
9851 	u32 port_shift = 0;
9852 	u32 reg;
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 == port)
9863 			return;
9864 
9865 		if (adapter->vxlan_port) {
9866 			netdev_info(dev,
9867 				    "VXLAN port %d set, not adding port %d\n",
9868 				    ntohs(adapter->vxlan_port),
9869 				    ntohs(port));
9870 			return;
9871 		}
9872 
9873 		adapter->vxlan_port = port;
9874 		break;
9875 	case UDP_TUNNEL_TYPE_GENEVE:
9876 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9877 			return;
9878 
9879 		if (adapter->geneve_port == port)
9880 			return;
9881 
9882 		if (adapter->geneve_port) {
9883 			netdev_info(dev,
9884 				    "GENEVE port %d set, not adding port %d\n",
9885 				    ntohs(adapter->geneve_port),
9886 				    ntohs(port));
9887 			return;
9888 		}
9889 
9890 		port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9891 		adapter->geneve_port = port;
9892 		break;
9893 	default:
9894 		return;
9895 	}
9896 
9897 	reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9898 	IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9899 }
9900 
9901 /**
9902  * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9903  * @dev: The port's netdev
9904  * @ti: Tunnel endpoint information
9905  **/
9906 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9907 				      struct udp_tunnel_info *ti)
9908 {
9909 	struct ixgbe_adapter *adapter = netdev_priv(dev);
9910 	u32 port_mask;
9911 
9912 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9913 	    ti->type != UDP_TUNNEL_TYPE_GENEVE)
9914 		return;
9915 
9916 	if (ti->sa_family != AF_INET)
9917 		return;
9918 
9919 	switch (ti->type) {
9920 	case UDP_TUNNEL_TYPE_VXLAN:
9921 		if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9922 			return;
9923 
9924 		if (adapter->vxlan_port != ti->port) {
9925 			netdev_info(dev, "VXLAN port %d not found\n",
9926 				    ntohs(ti->port));
9927 			return;
9928 		}
9929 
9930 		port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9931 		break;
9932 	case UDP_TUNNEL_TYPE_GENEVE:
9933 		if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9934 			return;
9935 
9936 		if (adapter->geneve_port != ti->port) {
9937 			netdev_info(dev, "GENEVE port %d not found\n",
9938 				    ntohs(ti->port));
9939 			return;
9940 		}
9941 
9942 		port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9943 		break;
9944 	default:
9945 		return;
9946 	}
9947 
9948 	ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9949 	adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9950 }
9951 
9952 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9953 			     struct net_device *dev,
9954 			     const unsigned char *addr, u16 vid,
9955 			     u16 flags,
9956 			     struct netlink_ext_ack *extack)
9957 {
9958 	/* guarantee we can provide a unique filter for the unicast address */
9959 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9960 		struct ixgbe_adapter *adapter = netdev_priv(dev);
9961 		u16 pool = VMDQ_P(0);
9962 
9963 		if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9964 			return -ENOMEM;
9965 	}
9966 
9967 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9968 }
9969 
9970 /**
9971  * ixgbe_configure_bridge_mode - set various bridge modes
9972  * @adapter: the private structure
9973  * @mode: requested bridge mode
9974  *
9975  * Configure some settings require for various bridge modes.
9976  **/
9977 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9978 				       __u16 mode)
9979 {
9980 	struct ixgbe_hw *hw = &adapter->hw;
9981 	unsigned int p, num_pools;
9982 	u32 vmdctl;
9983 
9984 	switch (mode) {
9985 	case BRIDGE_MODE_VEPA:
9986 		/* disable Tx loopback, rely on switch hairpin mode */
9987 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9988 
9989 		/* must enable Rx switching replication to allow multicast
9990 		 * packet reception on all VFs, and to enable source address
9991 		 * pruning.
9992 		 */
9993 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9994 		vmdctl |= IXGBE_VT_CTL_REPLEN;
9995 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9996 
9997 		/* enable Rx source address pruning. Note, this requires
9998 		 * replication to be enabled or else it does nothing.
9999 		 */
10000 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
10001 		for (p = 0; p < num_pools; p++) {
10002 			if (hw->mac.ops.set_source_address_pruning)
10003 				hw->mac.ops.set_source_address_pruning(hw,
10004 								       true,
10005 								       p);
10006 		}
10007 		break;
10008 	case BRIDGE_MODE_VEB:
10009 		/* enable Tx loopback for internal VF/PF communication */
10010 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
10011 				IXGBE_PFDTXGSWC_VT_LBEN);
10012 
10013 		/* disable Rx switching replication unless we have SR-IOV
10014 		 * virtual functions
10015 		 */
10016 		vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
10017 		if (!adapter->num_vfs)
10018 			vmdctl &= ~IXGBE_VT_CTL_REPLEN;
10019 		IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
10020 
10021 		/* disable Rx source address pruning, since we don't expect to
10022 		 * be receiving external loopback of our transmitted frames.
10023 		 */
10024 		num_pools = adapter->num_vfs + adapter->num_rx_pools;
10025 		for (p = 0; p < num_pools; p++) {
10026 			if (hw->mac.ops.set_source_address_pruning)
10027 				hw->mac.ops.set_source_address_pruning(hw,
10028 								       false,
10029 								       p);
10030 		}
10031 		break;
10032 	default:
10033 		return -EINVAL;
10034 	}
10035 
10036 	adapter->bridge_mode = mode;
10037 
10038 	e_info(drv, "enabling bridge mode: %s\n",
10039 	       mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
10040 
10041 	return 0;
10042 }
10043 
10044 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
10045 				    struct nlmsghdr *nlh, u16 flags,
10046 				    struct netlink_ext_ack *extack)
10047 {
10048 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10049 	struct nlattr *attr, *br_spec;
10050 	int rem;
10051 
10052 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10053 		return -EOPNOTSUPP;
10054 
10055 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
10056 	if (!br_spec)
10057 		return -EINVAL;
10058 
10059 	nla_for_each_nested(attr, br_spec, rem) {
10060 		int status;
10061 		__u16 mode;
10062 
10063 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
10064 			continue;
10065 
10066 		if (nla_len(attr) < sizeof(mode))
10067 			return -EINVAL;
10068 
10069 		mode = nla_get_u16(attr);
10070 		status = ixgbe_configure_bridge_mode(adapter, mode);
10071 		if (status)
10072 			return status;
10073 
10074 		break;
10075 	}
10076 
10077 	return 0;
10078 }
10079 
10080 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10081 				    struct net_device *dev,
10082 				    u32 filter_mask, int nlflags)
10083 {
10084 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10085 
10086 	if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10087 		return 0;
10088 
10089 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10090 				       adapter->bridge_mode, 0, 0, nlflags,
10091 				       filter_mask, NULL);
10092 }
10093 
10094 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10095 {
10096 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10097 	struct ixgbe_fwd_adapter *accel;
10098 	int tcs = adapter->hw_tcs ? : 1;
10099 	int pool, err;
10100 
10101 	if (adapter->xdp_prog) {
10102 		e_warn(probe, "L2FW offload is not supported with XDP\n");
10103 		return ERR_PTR(-EINVAL);
10104 	}
10105 
10106 	/* The hardware supported by ixgbe only filters on the destination MAC
10107 	 * address. In order to avoid issues we only support offloading modes
10108 	 * where the hardware can actually provide the functionality.
10109 	 */
10110 	if (!macvlan_supports_dest_filter(vdev))
10111 		return ERR_PTR(-EMEDIUMTYPE);
10112 
10113 	/* We need to lock down the macvlan to be a single queue device so that
10114 	 * we can reuse the tc_to_txq field in the macvlan netdev to represent
10115 	 * the queue mapping to our netdev.
10116 	 */
10117 	if (netif_is_multiqueue(vdev))
10118 		return ERR_PTR(-ERANGE);
10119 
10120 	pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10121 	if (pool == adapter->num_rx_pools) {
10122 		u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10123 		u16 reserved_pools;
10124 
10125 		if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10126 		     adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10127 		    adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10128 			return ERR_PTR(-EBUSY);
10129 
10130 		/* Hardware has a limited number of available pools. Each VF,
10131 		 * and the PF require a pool. Check to ensure we don't
10132 		 * attempt to use more then the available number of pools.
10133 		 */
10134 		if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10135 			return ERR_PTR(-EBUSY);
10136 
10137 		/* Enable VMDq flag so device will be set in VM mode */
10138 		adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10139 				  IXGBE_FLAG_SRIOV_ENABLED;
10140 
10141 		/* Try to reserve as many queues per pool as possible,
10142 		 * we start with the configurations that support 4 queues
10143 		 * per pools, followed by 2, and then by just 1 per pool.
10144 		 */
10145 		if (used_pools < 32 && adapter->num_rx_pools < 16)
10146 			reserved_pools = min_t(u16,
10147 					       32 - used_pools,
10148 					       16 - adapter->num_rx_pools);
10149 		else if (adapter->num_rx_pools < 32)
10150 			reserved_pools = min_t(u16,
10151 					       64 - used_pools,
10152 					       32 - adapter->num_rx_pools);
10153 		else
10154 			reserved_pools = 64 - used_pools;
10155 
10156 
10157 		if (!reserved_pools)
10158 			return ERR_PTR(-EBUSY);
10159 
10160 		adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10161 
10162 		/* Force reinit of ring allocation with VMDQ enabled */
10163 		err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10164 		if (err)
10165 			return ERR_PTR(err);
10166 
10167 		if (pool >= adapter->num_rx_pools)
10168 			return ERR_PTR(-ENOMEM);
10169 	}
10170 
10171 	accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10172 	if (!accel)
10173 		return ERR_PTR(-ENOMEM);
10174 
10175 	set_bit(pool, adapter->fwd_bitmask);
10176 	netdev_set_sb_channel(vdev, pool);
10177 	accel->pool = pool;
10178 	accel->netdev = vdev;
10179 
10180 	if (!netif_running(pdev))
10181 		return accel;
10182 
10183 	err = ixgbe_fwd_ring_up(adapter, accel);
10184 	if (err)
10185 		return ERR_PTR(err);
10186 
10187 	return accel;
10188 }
10189 
10190 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10191 {
10192 	struct ixgbe_fwd_adapter *accel = priv;
10193 	struct ixgbe_adapter *adapter = netdev_priv(pdev);
10194 	unsigned int rxbase = accel->rx_base_queue;
10195 	unsigned int i;
10196 
10197 	/* delete unicast filter associated with offloaded interface */
10198 	ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10199 			     VMDQ_P(accel->pool));
10200 
10201 	/* Allow remaining Rx packets to get flushed out of the
10202 	 * Rx FIFO before we drop the netdev for the ring.
10203 	 */
10204 	usleep_range(10000, 20000);
10205 
10206 	for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10207 		struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10208 		struct ixgbe_q_vector *qv = ring->q_vector;
10209 
10210 		/* Make sure we aren't processing any packets and clear
10211 		 * netdev to shut down the ring.
10212 		 */
10213 		if (netif_running(adapter->netdev))
10214 			napi_synchronize(&qv->napi);
10215 		ring->netdev = NULL;
10216 	}
10217 
10218 	/* unbind the queues and drop the subordinate channel config */
10219 	netdev_unbind_sb_channel(pdev, accel->netdev);
10220 	netdev_set_sb_channel(accel->netdev, 0);
10221 
10222 	clear_bit(accel->pool, adapter->fwd_bitmask);
10223 	kfree(accel);
10224 }
10225 
10226 #define IXGBE_MAX_MAC_HDR_LEN		127
10227 #define IXGBE_MAX_NETWORK_HDR_LEN	511
10228 
10229 static netdev_features_t
10230 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10231 		     netdev_features_t features)
10232 {
10233 	unsigned int network_hdr_len, mac_hdr_len;
10234 
10235 	/* Make certain the headers can be described by a context descriptor */
10236 	mac_hdr_len = skb_network_header(skb) - skb->data;
10237 	if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10238 		return features & ~(NETIF_F_HW_CSUM |
10239 				    NETIF_F_SCTP_CRC |
10240 				    NETIF_F_GSO_UDP_L4 |
10241 				    NETIF_F_HW_VLAN_CTAG_TX |
10242 				    NETIF_F_TSO |
10243 				    NETIF_F_TSO6);
10244 
10245 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10246 	if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10247 		return features & ~(NETIF_F_HW_CSUM |
10248 				    NETIF_F_SCTP_CRC |
10249 				    NETIF_F_GSO_UDP_L4 |
10250 				    NETIF_F_TSO |
10251 				    NETIF_F_TSO6);
10252 
10253 	/* We can only support IPV4 TSO in tunnels if we can mangle the
10254 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
10255 	 * IPsec offoad sets skb->encapsulation but still can handle
10256 	 * the TSO, so it's the exception.
10257 	 */
10258 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10259 #ifdef CONFIG_IXGBE_IPSEC
10260 		if (!secpath_exists(skb))
10261 #endif
10262 			features &= ~NETIF_F_TSO;
10263 	}
10264 
10265 	return features;
10266 }
10267 
10268 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10269 {
10270 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10271 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10272 	struct bpf_prog *old_prog;
10273 	bool need_reset;
10274 
10275 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10276 		return -EINVAL;
10277 
10278 	if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10279 		return -EINVAL;
10280 
10281 	/* verify ixgbe ring attributes are sufficient for XDP */
10282 	for (i = 0; i < adapter->num_rx_queues; i++) {
10283 		struct ixgbe_ring *ring = adapter->rx_ring[i];
10284 
10285 		if (ring_is_rsc_enabled(ring))
10286 			return -EINVAL;
10287 
10288 		if (frame_size > ixgbe_rx_bufsz(ring))
10289 			return -EINVAL;
10290 	}
10291 
10292 	if (nr_cpu_ids > MAX_XDP_QUEUES)
10293 		return -ENOMEM;
10294 
10295 	old_prog = xchg(&adapter->xdp_prog, prog);
10296 	need_reset = (!!prog != !!old_prog);
10297 
10298 	/* If transitioning XDP modes reconfigure rings */
10299 	if (need_reset) {
10300 		int err;
10301 
10302 		if (!prog)
10303 			/* Wait until ndo_xsk_wakeup completes. */
10304 			synchronize_rcu();
10305 		err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10306 
10307 		if (err) {
10308 			rcu_assign_pointer(adapter->xdp_prog, old_prog);
10309 			return -EINVAL;
10310 		}
10311 	} else {
10312 		for (i = 0; i < adapter->num_rx_queues; i++)
10313 			(void)xchg(&adapter->rx_ring[i]->xdp_prog,
10314 			    adapter->xdp_prog);
10315 	}
10316 
10317 	if (old_prog)
10318 		bpf_prog_put(old_prog);
10319 
10320 	/* Kick start the NAPI context if there is an AF_XDP socket open
10321 	 * on that queue id. This so that receiving will start.
10322 	 */
10323 	if (need_reset && prog)
10324 		for (i = 0; i < adapter->num_rx_queues; i++)
10325 			if (adapter->xdp_ring[i]->xsk_umem)
10326 				(void)ixgbe_xsk_wakeup(adapter->netdev, i,
10327 						       XDP_WAKEUP_RX);
10328 
10329 	return 0;
10330 }
10331 
10332 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10333 {
10334 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10335 
10336 	switch (xdp->command) {
10337 	case XDP_SETUP_PROG:
10338 		return ixgbe_xdp_setup(dev, xdp->prog);
10339 	case XDP_QUERY_PROG:
10340 		xdp->prog_id = adapter->xdp_prog ?
10341 			adapter->xdp_prog->aux->id : 0;
10342 		return 0;
10343 	case XDP_SETUP_XSK_UMEM:
10344 		return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10345 					    xdp->xsk.queue_id);
10346 
10347 	default:
10348 		return -EINVAL;
10349 	}
10350 }
10351 
10352 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10353 {
10354 	/* Force memory writes to complete before letting h/w know there
10355 	 * are new descriptors to fetch.
10356 	 */
10357 	wmb();
10358 	writel(ring->next_to_use, ring->tail);
10359 }
10360 
10361 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10362 			  struct xdp_frame **frames, u32 flags)
10363 {
10364 	struct ixgbe_adapter *adapter = netdev_priv(dev);
10365 	struct ixgbe_ring *ring;
10366 	int drops = 0;
10367 	int i;
10368 
10369 	if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10370 		return -ENETDOWN;
10371 
10372 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10373 		return -EINVAL;
10374 
10375 	/* During program transitions its possible adapter->xdp_prog is assigned
10376 	 * but ring has not been configured yet. In this case simply abort xmit.
10377 	 */
10378 	ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10379 	if (unlikely(!ring))
10380 		return -ENXIO;
10381 
10382 	if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10383 		return -ENXIO;
10384 
10385 	for (i = 0; i < n; i++) {
10386 		struct xdp_frame *xdpf = frames[i];
10387 		int err;
10388 
10389 		err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10390 		if (err != IXGBE_XDP_TX) {
10391 			xdp_return_frame_rx_napi(xdpf);
10392 			drops++;
10393 		}
10394 	}
10395 
10396 	if (unlikely(flags & XDP_XMIT_FLUSH))
10397 		ixgbe_xdp_ring_update_tail(ring);
10398 
10399 	return n - drops;
10400 }
10401 
10402 static const struct net_device_ops ixgbe_netdev_ops = {
10403 	.ndo_open		= ixgbe_open,
10404 	.ndo_stop		= ixgbe_close,
10405 	.ndo_start_xmit		= ixgbe_xmit_frame,
10406 	.ndo_set_rx_mode	= ixgbe_set_rx_mode,
10407 	.ndo_validate_addr	= eth_validate_addr,
10408 	.ndo_set_mac_address	= ixgbe_set_mac,
10409 	.ndo_change_mtu		= ixgbe_change_mtu,
10410 	.ndo_tx_timeout		= ixgbe_tx_timeout,
10411 	.ndo_set_tx_maxrate	= ixgbe_tx_maxrate,
10412 	.ndo_vlan_rx_add_vid	= ixgbe_vlan_rx_add_vid,
10413 	.ndo_vlan_rx_kill_vid	= ixgbe_vlan_rx_kill_vid,
10414 	.ndo_do_ioctl		= ixgbe_ioctl,
10415 	.ndo_set_vf_mac		= ixgbe_ndo_set_vf_mac,
10416 	.ndo_set_vf_vlan	= ixgbe_ndo_set_vf_vlan,
10417 	.ndo_set_vf_rate	= ixgbe_ndo_set_vf_bw,
10418 	.ndo_set_vf_spoofchk	= ixgbe_ndo_set_vf_spoofchk,
10419 	.ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10420 	.ndo_set_vf_trust	= ixgbe_ndo_set_vf_trust,
10421 	.ndo_get_vf_config	= ixgbe_ndo_get_vf_config,
10422 	.ndo_get_stats64	= ixgbe_get_stats64,
10423 	.ndo_setup_tc		= __ixgbe_setup_tc,
10424 #ifdef IXGBE_FCOE
10425 	.ndo_select_queue	= ixgbe_select_queue,
10426 	.ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10427 	.ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10428 	.ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10429 	.ndo_fcoe_enable = ixgbe_fcoe_enable,
10430 	.ndo_fcoe_disable = ixgbe_fcoe_disable,
10431 	.ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10432 	.ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10433 #endif /* IXGBE_FCOE */
10434 	.ndo_set_features = ixgbe_set_features,
10435 	.ndo_fix_features = ixgbe_fix_features,
10436 	.ndo_fdb_add		= ixgbe_ndo_fdb_add,
10437 	.ndo_bridge_setlink	= ixgbe_ndo_bridge_setlink,
10438 	.ndo_bridge_getlink	= ixgbe_ndo_bridge_getlink,
10439 	.ndo_dfwd_add_station	= ixgbe_fwd_add,
10440 	.ndo_dfwd_del_station	= ixgbe_fwd_del,
10441 	.ndo_udp_tunnel_add	= ixgbe_add_udp_tunnel_port,
10442 	.ndo_udp_tunnel_del	= ixgbe_del_udp_tunnel_port,
10443 	.ndo_features_check	= ixgbe_features_check,
10444 	.ndo_bpf		= ixgbe_xdp,
10445 	.ndo_xdp_xmit		= ixgbe_xdp_xmit,
10446 	.ndo_xsk_wakeup         = ixgbe_xsk_wakeup,
10447 };
10448 
10449 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10450 				 struct ixgbe_ring *tx_ring)
10451 {
10452 	unsigned long wait_delay, delay_interval;
10453 	struct ixgbe_hw *hw = &adapter->hw;
10454 	u8 reg_idx = tx_ring->reg_idx;
10455 	int wait_loop;
10456 	u32 txdctl;
10457 
10458 	IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10459 
10460 	/* delay mechanism from ixgbe_disable_tx */
10461 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10462 
10463 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10464 	wait_delay = delay_interval;
10465 
10466 	while (wait_loop--) {
10467 		usleep_range(wait_delay, wait_delay + 10);
10468 		wait_delay += delay_interval * 2;
10469 		txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10470 
10471 		if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10472 			return;
10473 	}
10474 
10475 	e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10476 }
10477 
10478 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10479 			      struct ixgbe_ring *tx_ring)
10480 {
10481 	set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10482 	ixgbe_disable_txr_hw(adapter, tx_ring);
10483 }
10484 
10485 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10486 				 struct ixgbe_ring *rx_ring)
10487 {
10488 	unsigned long wait_delay, delay_interval;
10489 	struct ixgbe_hw *hw = &adapter->hw;
10490 	u8 reg_idx = rx_ring->reg_idx;
10491 	int wait_loop;
10492 	u32 rxdctl;
10493 
10494 	rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10495 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10496 	rxdctl |= IXGBE_RXDCTL_SWFLSH;
10497 
10498 	/* write value back with RXDCTL.ENABLE bit cleared */
10499 	IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10500 
10501 	/* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10502 	if (hw->mac.type == ixgbe_mac_82598EB &&
10503 	    !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10504 		return;
10505 
10506 	/* delay mechanism from ixgbe_disable_rx */
10507 	delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10508 
10509 	wait_loop = IXGBE_MAX_RX_DESC_POLL;
10510 	wait_delay = delay_interval;
10511 
10512 	while (wait_loop--) {
10513 		usleep_range(wait_delay, wait_delay + 10);
10514 		wait_delay += delay_interval * 2;
10515 		rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10516 
10517 		if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10518 			return;
10519 	}
10520 
10521 	e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10522 }
10523 
10524 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10525 {
10526 	memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10527 	memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10528 }
10529 
10530 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10531 {
10532 	memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10533 	memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10534 }
10535 
10536 /**
10537  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10538  * @adapter: adapter structure
10539  * @ring: ring index
10540  *
10541  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10542  * assumes that the netdev is running.
10543  **/
10544 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10545 {
10546 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10547 
10548 	rx_ring = adapter->rx_ring[ring];
10549 	tx_ring = adapter->tx_ring[ring];
10550 	xdp_ring = adapter->xdp_ring[ring];
10551 
10552 	ixgbe_disable_txr(adapter, tx_ring);
10553 	if (xdp_ring)
10554 		ixgbe_disable_txr(adapter, xdp_ring);
10555 	ixgbe_disable_rxr_hw(adapter, rx_ring);
10556 
10557 	if (xdp_ring)
10558 		synchronize_rcu();
10559 
10560 	/* Rx/Tx/XDP Tx share the same napi context. */
10561 	napi_disable(&rx_ring->q_vector->napi);
10562 
10563 	ixgbe_clean_tx_ring(tx_ring);
10564 	if (xdp_ring)
10565 		ixgbe_clean_tx_ring(xdp_ring);
10566 	ixgbe_clean_rx_ring(rx_ring);
10567 
10568 	ixgbe_reset_txr_stats(tx_ring);
10569 	if (xdp_ring)
10570 		ixgbe_reset_txr_stats(xdp_ring);
10571 	ixgbe_reset_rxr_stats(rx_ring);
10572 }
10573 
10574 /**
10575  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10576  * @adapter: adapter structure
10577  * @ring: ring index
10578  *
10579  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10580  * assumes that the netdev is running.
10581  **/
10582 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10583 {
10584 	struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10585 
10586 	rx_ring = adapter->rx_ring[ring];
10587 	tx_ring = adapter->tx_ring[ring];
10588 	xdp_ring = adapter->xdp_ring[ring];
10589 
10590 	/* Rx/Tx/XDP Tx share the same napi context. */
10591 	napi_enable(&rx_ring->q_vector->napi);
10592 
10593 	ixgbe_configure_tx_ring(adapter, tx_ring);
10594 	if (xdp_ring)
10595 		ixgbe_configure_tx_ring(adapter, xdp_ring);
10596 	ixgbe_configure_rx_ring(adapter, rx_ring);
10597 
10598 	clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10599 	if (xdp_ring)
10600 		clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10601 }
10602 
10603 /**
10604  * ixgbe_enumerate_functions - Get the number of ports this device has
10605  * @adapter: adapter structure
10606  *
10607  * This function enumerates the phsyical functions co-located on a single slot,
10608  * in order to determine how many ports a device has. This is most useful in
10609  * determining the required GT/s of PCIe bandwidth necessary for optimal
10610  * performance.
10611  **/
10612 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10613 {
10614 	struct pci_dev *entry, *pdev = adapter->pdev;
10615 	int physfns = 0;
10616 
10617 	/* Some cards can not use the generic count PCIe functions method,
10618 	 * because they are behind a parent switch, so we hardcode these with
10619 	 * the correct number of functions.
10620 	 */
10621 	if (ixgbe_pcie_from_parent(&adapter->hw))
10622 		physfns = 4;
10623 
10624 	list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10625 		/* don't count virtual functions */
10626 		if (entry->is_virtfn)
10627 			continue;
10628 
10629 		/* When the devices on the bus don't all match our device ID,
10630 		 * we can't reliably determine the correct number of
10631 		 * functions. This can occur if a function has been direct
10632 		 * attached to a virtual machine using VT-d, for example. In
10633 		 * this case, simply return -1 to indicate this.
10634 		 */
10635 		if ((entry->vendor != pdev->vendor) ||
10636 		    (entry->device != pdev->device))
10637 			return -1;
10638 
10639 		physfns++;
10640 	}
10641 
10642 	return physfns;
10643 }
10644 
10645 /**
10646  * ixgbe_wol_supported - Check whether device supports WoL
10647  * @adapter: the adapter private structure
10648  * @device_id: the device ID
10649  * @subdevice_id: the subsystem device ID
10650  *
10651  * This function is used by probe and ethtool to determine
10652  * which devices have WoL support
10653  *
10654  **/
10655 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10656 			 u16 subdevice_id)
10657 {
10658 	struct ixgbe_hw *hw = &adapter->hw;
10659 	u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10660 
10661 	/* WOL not supported on 82598 */
10662 	if (hw->mac.type == ixgbe_mac_82598EB)
10663 		return false;
10664 
10665 	/* check eeprom to see if WOL is enabled for X540 and newer */
10666 	if (hw->mac.type >= ixgbe_mac_X540) {
10667 		if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10668 		    ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10669 		     (hw->bus.func == 0)))
10670 			return true;
10671 	}
10672 
10673 	/* WOL is determined based on device IDs for 82599 MACs */
10674 	switch (device_id) {
10675 	case IXGBE_DEV_ID_82599_SFP:
10676 		/* Only these subdevices could supports WOL */
10677 		switch (subdevice_id) {
10678 		case IXGBE_SUBDEV_ID_82599_560FLR:
10679 		case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10680 		case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10681 		case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10682 			/* only support first port */
10683 			if (hw->bus.func != 0)
10684 				break;
10685 			/* fall through */
10686 		case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10687 		case IXGBE_SUBDEV_ID_82599_SFP:
10688 		case IXGBE_SUBDEV_ID_82599_RNDC:
10689 		case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10690 		case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10691 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10692 		case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10693 			return true;
10694 		}
10695 		break;
10696 	case IXGBE_DEV_ID_82599EN_SFP:
10697 		/* Only these subdevices support WOL */
10698 		switch (subdevice_id) {
10699 		case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10700 			return true;
10701 		}
10702 		break;
10703 	case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10704 		/* All except this subdevice support WOL */
10705 		if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10706 			return true;
10707 		break;
10708 	case IXGBE_DEV_ID_82599_KX4:
10709 		return  true;
10710 	default:
10711 		break;
10712 	}
10713 
10714 	return false;
10715 }
10716 
10717 /**
10718  * ixgbe_set_fw_version - Set FW version
10719  * @adapter: the adapter private structure
10720  *
10721  * This function is used by probe and ethtool to determine the FW version to
10722  * format to display. The FW version is taken from the EEPROM/NVM.
10723  */
10724 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10725 {
10726 	struct ixgbe_hw *hw = &adapter->hw;
10727 	struct ixgbe_nvm_version nvm_ver;
10728 
10729 	ixgbe_get_oem_prod_version(hw, &nvm_ver);
10730 	if (nvm_ver.oem_valid) {
10731 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10732 			 "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10733 			 nvm_ver.oem_release);
10734 		return;
10735 	}
10736 
10737 	ixgbe_get_etk_id(hw, &nvm_ver);
10738 	ixgbe_get_orom_version(hw, &nvm_ver);
10739 
10740 	if (nvm_ver.or_valid) {
10741 		snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10742 			 "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10743 			 nvm_ver.or_build, nvm_ver.or_patch);
10744 		return;
10745 	}
10746 
10747 	/* Set ETrack ID format */
10748 	snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10749 		 "0x%08x", nvm_ver.etk_id);
10750 }
10751 
10752 /**
10753  * ixgbe_probe - Device Initialization Routine
10754  * @pdev: PCI device information struct
10755  * @ent: entry in ixgbe_pci_tbl
10756  *
10757  * Returns 0 on success, negative on failure
10758  *
10759  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10760  * The OS initialization, configuring of the adapter private structure,
10761  * and a hardware reset occur.
10762  **/
10763 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10764 {
10765 	struct net_device *netdev;
10766 	struct ixgbe_adapter *adapter = NULL;
10767 	struct ixgbe_hw *hw;
10768 	const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10769 	int i, err, pci_using_dac, expected_gts;
10770 	unsigned int indices = MAX_TX_QUEUES;
10771 	u8 part_str[IXGBE_PBANUM_LENGTH];
10772 	bool disable_dev = false;
10773 #ifdef IXGBE_FCOE
10774 	u16 device_caps;
10775 #endif
10776 	u32 eec;
10777 
10778 	/* Catch broken hardware that put the wrong VF device ID in
10779 	 * the PCIe SR-IOV capability.
10780 	 */
10781 	if (pdev->is_virtfn) {
10782 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10783 		     pci_name(pdev), pdev->vendor, pdev->device);
10784 		return -EINVAL;
10785 	}
10786 
10787 	err = pci_enable_device_mem(pdev);
10788 	if (err)
10789 		return err;
10790 
10791 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10792 		pci_using_dac = 1;
10793 	} else {
10794 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10795 		if (err) {
10796 			dev_err(&pdev->dev,
10797 				"No usable DMA configuration, aborting\n");
10798 			goto err_dma;
10799 		}
10800 		pci_using_dac = 0;
10801 	}
10802 
10803 	err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10804 	if (err) {
10805 		dev_err(&pdev->dev,
10806 			"pci_request_selected_regions failed 0x%x\n", err);
10807 		goto err_pci_reg;
10808 	}
10809 
10810 	pci_enable_pcie_error_reporting(pdev);
10811 
10812 	pci_set_master(pdev);
10813 	pci_save_state(pdev);
10814 
10815 	if (ii->mac == ixgbe_mac_82598EB) {
10816 #ifdef CONFIG_IXGBE_DCB
10817 		/* 8 TC w/ 4 queues per TC */
10818 		indices = 4 * MAX_TRAFFIC_CLASS;
10819 #else
10820 		indices = IXGBE_MAX_RSS_INDICES;
10821 #endif
10822 	}
10823 
10824 	netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10825 	if (!netdev) {
10826 		err = -ENOMEM;
10827 		goto err_alloc_etherdev;
10828 	}
10829 
10830 	SET_NETDEV_DEV(netdev, &pdev->dev);
10831 
10832 	adapter = netdev_priv(netdev);
10833 
10834 	adapter->netdev = netdev;
10835 	adapter->pdev = pdev;
10836 	hw = &adapter->hw;
10837 	hw->back = adapter;
10838 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10839 
10840 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10841 			      pci_resource_len(pdev, 0));
10842 	adapter->io_addr = hw->hw_addr;
10843 	if (!hw->hw_addr) {
10844 		err = -EIO;
10845 		goto err_ioremap;
10846 	}
10847 
10848 	netdev->netdev_ops = &ixgbe_netdev_ops;
10849 	ixgbe_set_ethtool_ops(netdev);
10850 	netdev->watchdog_timeo = 5 * HZ;
10851 	strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10852 
10853 	/* Setup hw api */
10854 	hw->mac.ops   = *ii->mac_ops;
10855 	hw->mac.type  = ii->mac;
10856 	hw->mvals     = ii->mvals;
10857 	if (ii->link_ops)
10858 		hw->link.ops  = *ii->link_ops;
10859 
10860 	/* EEPROM */
10861 	hw->eeprom.ops = *ii->eeprom_ops;
10862 	eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10863 	if (ixgbe_removed(hw->hw_addr)) {
10864 		err = -EIO;
10865 		goto err_ioremap;
10866 	}
10867 	/* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10868 	if (!(eec & BIT(8)))
10869 		hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10870 
10871 	/* PHY */
10872 	hw->phy.ops = *ii->phy_ops;
10873 	hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10874 	/* ixgbe_identify_phy_generic will set prtad and mmds properly */
10875 	hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10876 	hw->phy.mdio.mmds = 0;
10877 	hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10878 	hw->phy.mdio.dev = netdev;
10879 	hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10880 	hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10881 
10882 	/* setup the private structure */
10883 	err = ixgbe_sw_init(adapter, ii);
10884 	if (err)
10885 		goto err_sw_init;
10886 
10887 	/* Make sure the SWFW semaphore is in a valid state */
10888 	if (hw->mac.ops.init_swfw_sync)
10889 		hw->mac.ops.init_swfw_sync(hw);
10890 
10891 	/* Make it possible the adapter to be woken up via WOL */
10892 	switch (adapter->hw.mac.type) {
10893 	case ixgbe_mac_82599EB:
10894 	case ixgbe_mac_X540:
10895 	case ixgbe_mac_X550:
10896 	case ixgbe_mac_X550EM_x:
10897 	case ixgbe_mac_x550em_a:
10898 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10899 		break;
10900 	default:
10901 		break;
10902 	}
10903 
10904 	/*
10905 	 * If there is a fan on this device and it has failed log the
10906 	 * failure.
10907 	 */
10908 	if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10909 		u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10910 		if (esdp & IXGBE_ESDP_SDP1)
10911 			e_crit(probe, "Fan has stopped, replace the adapter\n");
10912 	}
10913 
10914 	if (allow_unsupported_sfp)
10915 		hw->allow_unsupported_sfp = allow_unsupported_sfp;
10916 
10917 	/* reset_hw fills in the perm_addr as well */
10918 	hw->phy.reset_if_overtemp = true;
10919 	err = hw->mac.ops.reset_hw(hw);
10920 	hw->phy.reset_if_overtemp = false;
10921 	ixgbe_set_eee_capable(adapter);
10922 	if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10923 		err = 0;
10924 	} else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10925 		e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10926 		e_dev_err("Reload the driver after installing a supported module.\n");
10927 		goto err_sw_init;
10928 	} else if (err) {
10929 		e_dev_err("HW Init failed: %d\n", err);
10930 		goto err_sw_init;
10931 	}
10932 
10933 #ifdef CONFIG_PCI_IOV
10934 	/* SR-IOV not supported on the 82598 */
10935 	if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10936 		goto skip_sriov;
10937 	/* Mailbox */
10938 	ixgbe_init_mbx_params_pf(hw);
10939 	hw->mbx.ops = ii->mbx_ops;
10940 	pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10941 	ixgbe_enable_sriov(adapter, max_vfs);
10942 skip_sriov:
10943 
10944 #endif
10945 	netdev->features = NETIF_F_SG |
10946 			   NETIF_F_TSO |
10947 			   NETIF_F_TSO6 |
10948 			   NETIF_F_RXHASH |
10949 			   NETIF_F_RXCSUM |
10950 			   NETIF_F_HW_CSUM;
10951 
10952 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10953 				    NETIF_F_GSO_GRE_CSUM | \
10954 				    NETIF_F_GSO_IPXIP4 | \
10955 				    NETIF_F_GSO_IPXIP6 | \
10956 				    NETIF_F_GSO_UDP_TUNNEL | \
10957 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
10958 
10959 	netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10960 	netdev->features |= NETIF_F_GSO_PARTIAL |
10961 			    IXGBE_GSO_PARTIAL_FEATURES;
10962 
10963 	if (hw->mac.type >= ixgbe_mac_82599EB)
10964 		netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
10965 
10966 #ifdef CONFIG_IXGBE_IPSEC
10967 #define IXGBE_ESP_FEATURES	(NETIF_F_HW_ESP | \
10968 				 NETIF_F_HW_ESP_TX_CSUM | \
10969 				 NETIF_F_GSO_ESP)
10970 
10971 	if (adapter->ipsec)
10972 		netdev->features |= IXGBE_ESP_FEATURES;
10973 #endif
10974 	/* copy netdev features into list of user selectable features */
10975 	netdev->hw_features |= netdev->features |
10976 			       NETIF_F_HW_VLAN_CTAG_FILTER |
10977 			       NETIF_F_HW_VLAN_CTAG_RX |
10978 			       NETIF_F_HW_VLAN_CTAG_TX |
10979 			       NETIF_F_RXALL |
10980 			       NETIF_F_HW_L2FW_DOFFLOAD;
10981 
10982 	if (hw->mac.type >= ixgbe_mac_82599EB)
10983 		netdev->hw_features |= NETIF_F_NTUPLE |
10984 				       NETIF_F_HW_TC;
10985 
10986 	if (pci_using_dac)
10987 		netdev->features |= NETIF_F_HIGHDMA;
10988 
10989 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10990 	netdev->hw_enc_features |= netdev->vlan_features;
10991 	netdev->mpls_features |= NETIF_F_SG |
10992 				 NETIF_F_TSO |
10993 				 NETIF_F_TSO6 |
10994 				 NETIF_F_HW_CSUM;
10995 	netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10996 
10997 	/* set this bit last since it cannot be part of vlan_features */
10998 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10999 			    NETIF_F_HW_VLAN_CTAG_RX |
11000 			    NETIF_F_HW_VLAN_CTAG_TX;
11001 
11002 	netdev->priv_flags |= IFF_UNICAST_FLT;
11003 	netdev->priv_flags |= IFF_SUPP_NOFCS;
11004 
11005 	/* MTU range: 68 - 9710 */
11006 	netdev->min_mtu = ETH_MIN_MTU;
11007 	netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
11008 
11009 #ifdef CONFIG_IXGBE_DCB
11010 	if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
11011 		netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
11012 #endif
11013 
11014 #ifdef IXGBE_FCOE
11015 	if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
11016 		unsigned int fcoe_l;
11017 
11018 		if (hw->mac.ops.get_device_caps) {
11019 			hw->mac.ops.get_device_caps(hw, &device_caps);
11020 			if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
11021 				adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
11022 		}
11023 
11024 
11025 		fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
11026 		adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
11027 
11028 		netdev->features |= NETIF_F_FSO |
11029 				    NETIF_F_FCOE_CRC;
11030 
11031 		netdev->vlan_features |= NETIF_F_FSO |
11032 					 NETIF_F_FCOE_CRC |
11033 					 NETIF_F_FCOE_MTU;
11034 	}
11035 #endif /* IXGBE_FCOE */
11036 	if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
11037 		netdev->hw_features |= NETIF_F_LRO;
11038 	if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
11039 		netdev->features |= NETIF_F_LRO;
11040 
11041 	if (ixgbe_check_fw_error(adapter)) {
11042 		err = -EIO;
11043 		goto err_sw_init;
11044 	}
11045 
11046 	/* make sure the EEPROM is good */
11047 	if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
11048 		e_dev_err("The EEPROM Checksum Is Not Valid\n");
11049 		err = -EIO;
11050 		goto err_sw_init;
11051 	}
11052 
11053 	eth_platform_get_mac_address(&adapter->pdev->dev,
11054 				     adapter->hw.mac.perm_addr);
11055 
11056 	memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
11057 
11058 	if (!is_valid_ether_addr(netdev->dev_addr)) {
11059 		e_dev_err("invalid MAC address\n");
11060 		err = -EIO;
11061 		goto err_sw_init;
11062 	}
11063 
11064 	/* Set hw->mac.addr to permanent MAC address */
11065 	ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11066 	ixgbe_mac_set_default_filter(adapter);
11067 
11068 	timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11069 
11070 	if (ixgbe_removed(hw->hw_addr)) {
11071 		err = -EIO;
11072 		goto err_sw_init;
11073 	}
11074 	INIT_WORK(&adapter->service_task, ixgbe_service_task);
11075 	set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11076 	clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11077 
11078 	err = ixgbe_init_interrupt_scheme(adapter);
11079 	if (err)
11080 		goto err_sw_init;
11081 
11082 	for (i = 0; i < adapter->num_rx_queues; i++)
11083 		u64_stats_init(&adapter->rx_ring[i]->syncp);
11084 	for (i = 0; i < adapter->num_tx_queues; i++)
11085 		u64_stats_init(&adapter->tx_ring[i]->syncp);
11086 	for (i = 0; i < adapter->num_xdp_queues; i++)
11087 		u64_stats_init(&adapter->xdp_ring[i]->syncp);
11088 
11089 	/* WOL not supported for all devices */
11090 	adapter->wol = 0;
11091 	hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11092 	hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11093 						pdev->subsystem_device);
11094 	if (hw->wol_enabled)
11095 		adapter->wol = IXGBE_WUFC_MAG;
11096 
11097 	device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11098 
11099 	/* save off EEPROM version number */
11100 	ixgbe_set_fw_version(adapter);
11101 
11102 	/* pick up the PCI bus settings for reporting later */
11103 	if (ixgbe_pcie_from_parent(hw))
11104 		ixgbe_get_parent_bus_info(adapter);
11105 	else
11106 		 hw->mac.ops.get_bus_info(hw);
11107 
11108 	/* calculate the expected PCIe bandwidth required for optimal
11109 	 * performance. Note that some older parts will never have enough
11110 	 * bandwidth due to being older generation PCIe parts. We clamp these
11111 	 * parts to ensure no warning is displayed if it can't be fixed.
11112 	 */
11113 	switch (hw->mac.type) {
11114 	case ixgbe_mac_82598EB:
11115 		expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11116 		break;
11117 	default:
11118 		expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11119 		break;
11120 	}
11121 
11122 	/* don't check link if we failed to enumerate functions */
11123 	if (expected_gts > 0)
11124 		ixgbe_check_minimum_link(adapter, expected_gts);
11125 
11126 	err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11127 	if (err)
11128 		strlcpy(part_str, "Unknown", sizeof(part_str));
11129 	if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11130 		e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11131 			   hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11132 			   part_str);
11133 	else
11134 		e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11135 			   hw->mac.type, hw->phy.type, part_str);
11136 
11137 	e_dev_info("%pM\n", netdev->dev_addr);
11138 
11139 	/* reset the hardware with the new settings */
11140 	err = hw->mac.ops.start_hw(hw);
11141 	if (err == IXGBE_ERR_EEPROM_VERSION) {
11142 		/* We are running on a pre-production device, log a warning */
11143 		e_dev_warn("This device is a pre-production adapter/LOM. "
11144 			   "Please be aware there may be issues associated "
11145 			   "with your hardware.  If you are experiencing "
11146 			   "problems please contact your Intel or hardware "
11147 			   "representative who provided you with this "
11148 			   "hardware.\n");
11149 	}
11150 	strcpy(netdev->name, "eth%d");
11151 	pci_set_drvdata(pdev, adapter);
11152 	err = register_netdev(netdev);
11153 	if (err)
11154 		goto err_register;
11155 
11156 
11157 	/* power down the optics for 82599 SFP+ fiber */
11158 	if (hw->mac.ops.disable_tx_laser)
11159 		hw->mac.ops.disable_tx_laser(hw);
11160 
11161 	/* carrier off reporting is important to ethtool even BEFORE open */
11162 	netif_carrier_off(netdev);
11163 
11164 #ifdef CONFIG_IXGBE_DCA
11165 	if (dca_add_requester(&pdev->dev) == 0) {
11166 		adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11167 		ixgbe_setup_dca(adapter);
11168 	}
11169 #endif
11170 	if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11171 		e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11172 		for (i = 0; i < adapter->num_vfs; i++)
11173 			ixgbe_vf_configuration(pdev, (i | 0x10000000));
11174 	}
11175 
11176 	/* firmware requires driver version to be 0xFFFFFFFF
11177 	 * since os does not support feature
11178 	 */
11179 	if (hw->mac.ops.set_fw_drv_ver)
11180 		hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11181 					   sizeof(ixgbe_driver_version) - 1,
11182 					   ixgbe_driver_version);
11183 
11184 	/* add san mac addr to netdev */
11185 	ixgbe_add_sanmac_netdev(netdev);
11186 
11187 	e_dev_info("%s\n", ixgbe_default_device_descr);
11188 
11189 #ifdef CONFIG_IXGBE_HWMON
11190 	if (ixgbe_sysfs_init(adapter))
11191 		e_err(probe, "failed to allocate sysfs resources\n");
11192 #endif /* CONFIG_IXGBE_HWMON */
11193 
11194 	ixgbe_dbg_adapter_init(adapter);
11195 
11196 	/* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11197 	if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11198 		hw->mac.ops.setup_link(hw,
11199 			IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11200 			true);
11201 
11202 	ixgbe_mii_bus_init(hw);
11203 
11204 	return 0;
11205 
11206 err_register:
11207 	ixgbe_release_hw_control(adapter);
11208 	ixgbe_clear_interrupt_scheme(adapter);
11209 err_sw_init:
11210 	ixgbe_disable_sriov(adapter);
11211 	adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11212 	iounmap(adapter->io_addr);
11213 	kfree(adapter->jump_tables[0]);
11214 	kfree(adapter->mac_table);
11215 	kfree(adapter->rss_key);
11216 	bitmap_free(adapter->af_xdp_zc_qps);
11217 err_ioremap:
11218 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11219 	free_netdev(netdev);
11220 err_alloc_etherdev:
11221 	pci_release_mem_regions(pdev);
11222 err_pci_reg:
11223 err_dma:
11224 	if (!adapter || disable_dev)
11225 		pci_disable_device(pdev);
11226 	return err;
11227 }
11228 
11229 /**
11230  * ixgbe_remove - Device Removal Routine
11231  * @pdev: PCI device information struct
11232  *
11233  * ixgbe_remove is called by the PCI subsystem to alert the driver
11234  * that it should release a PCI device.  The could be caused by a
11235  * Hot-Plug event, or because the driver is going to be removed from
11236  * memory.
11237  **/
11238 static void ixgbe_remove(struct pci_dev *pdev)
11239 {
11240 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11241 	struct net_device *netdev;
11242 	bool disable_dev;
11243 	int i;
11244 
11245 	/* if !adapter then we already cleaned up in probe */
11246 	if (!adapter)
11247 		return;
11248 
11249 	netdev  = adapter->netdev;
11250 	ixgbe_dbg_adapter_exit(adapter);
11251 
11252 	set_bit(__IXGBE_REMOVING, &adapter->state);
11253 	cancel_work_sync(&adapter->service_task);
11254 
11255 	if (adapter->mii_bus)
11256 		mdiobus_unregister(adapter->mii_bus);
11257 
11258 #ifdef CONFIG_IXGBE_DCA
11259 	if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11260 		adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11261 		dca_remove_requester(&pdev->dev);
11262 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11263 				IXGBE_DCA_CTRL_DCA_DISABLE);
11264 	}
11265 
11266 #endif
11267 #ifdef CONFIG_IXGBE_HWMON
11268 	ixgbe_sysfs_exit(adapter);
11269 #endif /* CONFIG_IXGBE_HWMON */
11270 
11271 	/* remove the added san mac */
11272 	ixgbe_del_sanmac_netdev(netdev);
11273 
11274 #ifdef CONFIG_PCI_IOV
11275 	ixgbe_disable_sriov(adapter);
11276 #endif
11277 	if (netdev->reg_state == NETREG_REGISTERED)
11278 		unregister_netdev(netdev);
11279 
11280 	ixgbe_stop_ipsec_offload(adapter);
11281 	ixgbe_clear_interrupt_scheme(adapter);
11282 
11283 	ixgbe_release_hw_control(adapter);
11284 
11285 #ifdef CONFIG_DCB
11286 	kfree(adapter->ixgbe_ieee_pfc);
11287 	kfree(adapter->ixgbe_ieee_ets);
11288 
11289 #endif
11290 	iounmap(adapter->io_addr);
11291 	pci_release_mem_regions(pdev);
11292 
11293 	e_dev_info("complete\n");
11294 
11295 	for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11296 		if (adapter->jump_tables[i]) {
11297 			kfree(adapter->jump_tables[i]->input);
11298 			kfree(adapter->jump_tables[i]->mask);
11299 		}
11300 		kfree(adapter->jump_tables[i]);
11301 	}
11302 
11303 	kfree(adapter->mac_table);
11304 	kfree(adapter->rss_key);
11305 	bitmap_free(adapter->af_xdp_zc_qps);
11306 	disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11307 	free_netdev(netdev);
11308 
11309 	pci_disable_pcie_error_reporting(pdev);
11310 
11311 	if (disable_dev)
11312 		pci_disable_device(pdev);
11313 }
11314 
11315 /**
11316  * ixgbe_io_error_detected - called when PCI error is detected
11317  * @pdev: Pointer to PCI device
11318  * @state: The current pci connection state
11319  *
11320  * This function is called after a PCI bus error affecting
11321  * this device has been detected.
11322  */
11323 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11324 						pci_channel_state_t state)
11325 {
11326 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11327 	struct net_device *netdev = adapter->netdev;
11328 
11329 #ifdef CONFIG_PCI_IOV
11330 	struct ixgbe_hw *hw = &adapter->hw;
11331 	struct pci_dev *bdev, *vfdev;
11332 	u32 dw0, dw1, dw2, dw3;
11333 	int vf, pos;
11334 	u16 req_id, pf_func;
11335 
11336 	if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11337 	    adapter->num_vfs == 0)
11338 		goto skip_bad_vf_detection;
11339 
11340 	bdev = pdev->bus->self;
11341 	while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11342 		bdev = bdev->bus->self;
11343 
11344 	if (!bdev)
11345 		goto skip_bad_vf_detection;
11346 
11347 	pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11348 	if (!pos)
11349 		goto skip_bad_vf_detection;
11350 
11351 	dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11352 	dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11353 	dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11354 	dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11355 	if (ixgbe_removed(hw->hw_addr))
11356 		goto skip_bad_vf_detection;
11357 
11358 	req_id = dw1 >> 16;
11359 	/* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11360 	if (!(req_id & 0x0080))
11361 		goto skip_bad_vf_detection;
11362 
11363 	pf_func = req_id & 0x01;
11364 	if ((pf_func & 1) == (pdev->devfn & 1)) {
11365 		unsigned int device_id;
11366 
11367 		vf = (req_id & 0x7F) >> 1;
11368 		e_dev_err("VF %d has caused a PCIe error\n", vf);
11369 		e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11370 				"%8.8x\tdw3: %8.8x\n",
11371 		dw0, dw1, dw2, dw3);
11372 		switch (adapter->hw.mac.type) {
11373 		case ixgbe_mac_82599EB:
11374 			device_id = IXGBE_82599_VF_DEVICE_ID;
11375 			break;
11376 		case ixgbe_mac_X540:
11377 			device_id = IXGBE_X540_VF_DEVICE_ID;
11378 			break;
11379 		case ixgbe_mac_X550:
11380 			device_id = IXGBE_DEV_ID_X550_VF;
11381 			break;
11382 		case ixgbe_mac_X550EM_x:
11383 			device_id = IXGBE_DEV_ID_X550EM_X_VF;
11384 			break;
11385 		case ixgbe_mac_x550em_a:
11386 			device_id = IXGBE_DEV_ID_X550EM_A_VF;
11387 			break;
11388 		default:
11389 			device_id = 0;
11390 			break;
11391 		}
11392 
11393 		/* Find the pci device of the offending VF */
11394 		vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11395 		while (vfdev) {
11396 			if (vfdev->devfn == (req_id & 0xFF))
11397 				break;
11398 			vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11399 					       device_id, vfdev);
11400 		}
11401 		/*
11402 		 * There's a slim chance the VF could have been hot plugged,
11403 		 * so if it is no longer present we don't need to issue the
11404 		 * VFLR.  Just clean up the AER in that case.
11405 		 */
11406 		if (vfdev) {
11407 			pcie_flr(vfdev);
11408 			/* Free device reference count */
11409 			pci_dev_put(vfdev);
11410 		}
11411 	}
11412 
11413 	/*
11414 	 * Even though the error may have occurred on the other port
11415 	 * we still need to increment the vf error reference count for
11416 	 * both ports because the I/O resume function will be called
11417 	 * for both of them.
11418 	 */
11419 	adapter->vferr_refcount++;
11420 
11421 	return PCI_ERS_RESULT_RECOVERED;
11422 
11423 skip_bad_vf_detection:
11424 #endif /* CONFIG_PCI_IOV */
11425 	if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11426 		return PCI_ERS_RESULT_DISCONNECT;
11427 
11428 	if (!netif_device_present(netdev))
11429 		return PCI_ERS_RESULT_DISCONNECT;
11430 
11431 	rtnl_lock();
11432 	netif_device_detach(netdev);
11433 
11434 	if (netif_running(netdev))
11435 		ixgbe_close_suspend(adapter);
11436 
11437 	if (state == pci_channel_io_perm_failure) {
11438 		rtnl_unlock();
11439 		return PCI_ERS_RESULT_DISCONNECT;
11440 	}
11441 
11442 	if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11443 		pci_disable_device(pdev);
11444 	rtnl_unlock();
11445 
11446 	/* Request a slot reset. */
11447 	return PCI_ERS_RESULT_NEED_RESET;
11448 }
11449 
11450 /**
11451  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11452  * @pdev: Pointer to PCI device
11453  *
11454  * Restart the card from scratch, as if from a cold-boot.
11455  */
11456 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11457 {
11458 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11459 	pci_ers_result_t result;
11460 
11461 	if (pci_enable_device_mem(pdev)) {
11462 		e_err(probe, "Cannot re-enable PCI device after reset.\n");
11463 		result = PCI_ERS_RESULT_DISCONNECT;
11464 	} else {
11465 		smp_mb__before_atomic();
11466 		clear_bit(__IXGBE_DISABLED, &adapter->state);
11467 		adapter->hw.hw_addr = adapter->io_addr;
11468 		pci_set_master(pdev);
11469 		pci_restore_state(pdev);
11470 		pci_save_state(pdev);
11471 
11472 		pci_wake_from_d3(pdev, false);
11473 
11474 		ixgbe_reset(adapter);
11475 		IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11476 		result = PCI_ERS_RESULT_RECOVERED;
11477 	}
11478 
11479 	return result;
11480 }
11481 
11482 /**
11483  * ixgbe_io_resume - called when traffic can start flowing again.
11484  * @pdev: Pointer to PCI device
11485  *
11486  * This callback is called when the error recovery driver tells us that
11487  * its OK to resume normal operation.
11488  */
11489 static void ixgbe_io_resume(struct pci_dev *pdev)
11490 {
11491 	struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11492 	struct net_device *netdev = adapter->netdev;
11493 
11494 #ifdef CONFIG_PCI_IOV
11495 	if (adapter->vferr_refcount) {
11496 		e_info(drv, "Resuming after VF err\n");
11497 		adapter->vferr_refcount--;
11498 		return;
11499 	}
11500 
11501 #endif
11502 	rtnl_lock();
11503 	if (netif_running(netdev))
11504 		ixgbe_open(netdev);
11505 
11506 	netif_device_attach(netdev);
11507 	rtnl_unlock();
11508 }
11509 
11510 static const struct pci_error_handlers ixgbe_err_handler = {
11511 	.error_detected = ixgbe_io_error_detected,
11512 	.slot_reset = ixgbe_io_slot_reset,
11513 	.resume = ixgbe_io_resume,
11514 };
11515 
11516 static struct pci_driver ixgbe_driver = {
11517 	.name     = ixgbe_driver_name,
11518 	.id_table = ixgbe_pci_tbl,
11519 	.probe    = ixgbe_probe,
11520 	.remove   = ixgbe_remove,
11521 #ifdef CONFIG_PM
11522 	.suspend  = ixgbe_suspend,
11523 	.resume   = ixgbe_resume,
11524 #endif
11525 	.shutdown = ixgbe_shutdown,
11526 	.sriov_configure = ixgbe_pci_sriov_configure,
11527 	.err_handler = &ixgbe_err_handler
11528 };
11529 
11530 /**
11531  * ixgbe_init_module - Driver Registration Routine
11532  *
11533  * ixgbe_init_module is the first routine called when the driver is
11534  * loaded. All it does is register with the PCI subsystem.
11535  **/
11536 static int __init ixgbe_init_module(void)
11537 {
11538 	int ret;
11539 	pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11540 	pr_info("%s\n", ixgbe_copyright);
11541 
11542 	ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11543 	if (!ixgbe_wq) {
11544 		pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11545 		return -ENOMEM;
11546 	}
11547 
11548 	ixgbe_dbg_init();
11549 
11550 	ret = pci_register_driver(&ixgbe_driver);
11551 	if (ret) {
11552 		destroy_workqueue(ixgbe_wq);
11553 		ixgbe_dbg_exit();
11554 		return ret;
11555 	}
11556 
11557 #ifdef CONFIG_IXGBE_DCA
11558 	dca_register_notify(&dca_notifier);
11559 #endif
11560 
11561 	return 0;
11562 }
11563 
11564 module_init(ixgbe_init_module);
11565 
11566 /**
11567  * ixgbe_exit_module - Driver Exit Cleanup Routine
11568  *
11569  * ixgbe_exit_module is called just before the driver is removed
11570  * from memory.
11571  **/
11572 static void __exit ixgbe_exit_module(void)
11573 {
11574 #ifdef CONFIG_IXGBE_DCA
11575 	dca_unregister_notify(&dca_notifier);
11576 #endif
11577 	pci_unregister_driver(&ixgbe_driver);
11578 
11579 	ixgbe_dbg_exit();
11580 	if (ixgbe_wq) {
11581 		destroy_workqueue(ixgbe_wq);
11582 		ixgbe_wq = NULL;
11583 	}
11584 }
11585 
11586 #ifdef CONFIG_IXGBE_DCA
11587 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11588 			    void *p)
11589 {
11590 	int ret_val;
11591 
11592 	ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11593 					 __ixgbe_notify_dca);
11594 
11595 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11596 }
11597 
11598 #endif /* CONFIG_IXGBE_DCA */
11599 
11600 module_exit(ixgbe_exit_module);
11601 
11602 /* ixgbe_main.c */
11603