xref: /openbmc/linux/drivers/net/ethernet/intel/igb/igb_main.c (revision 49cb93cb090a1515deaef7dcafe905ab25b2809e)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Intel(R) Gigabit Ethernet Linux driver
3  * Copyright(c) 2007-2014 Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  *
14  * You should have received a copy of the GNU General Public License along with
15  * this program; if not, see <http://www.gnu.org/licenses/>.
16  *
17  * The full GNU General Public License is included in this distribution in
18  * the file called "COPYING".
19  *
20  * Contact Information:
21  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
22  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
23  */
24 
25 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/init.h>
30 #include <linux/bitops.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/ipv6.h>
35 #include <linux/slab.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 #include <net/pkt_sched.h>
39 #include <linux/net_tstamp.h>
40 #include <linux/mii.h>
41 #include <linux/ethtool.h>
42 #include <linux/if.h>
43 #include <linux/if_vlan.h>
44 #include <linux/pci.h>
45 #include <linux/pci-aspm.h>
46 #include <linux/delay.h>
47 #include <linux/interrupt.h>
48 #include <linux/ip.h>
49 #include <linux/tcp.h>
50 #include <linux/sctp.h>
51 #include <linux/if_ether.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
54 #include <linux/pm_runtime.h>
55 #include <linux/etherdevice.h>
56 #ifdef CONFIG_IGB_DCA
57 #include <linux/dca.h>
58 #endif
59 #include <linux/i2c.h>
60 #include "igb.h"
61 
62 #define MAJ 5
63 #define MIN 4
64 #define BUILD 0
65 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
66 __stringify(BUILD) "-k"
67 
68 enum queue_mode {
69 	QUEUE_MODE_STRICT_PRIORITY,
70 	QUEUE_MODE_STREAM_RESERVATION,
71 };
72 
73 enum tx_queue_prio {
74 	TX_QUEUE_PRIO_HIGH,
75 	TX_QUEUE_PRIO_LOW,
76 };
77 
78 char igb_driver_name[] = "igb";
79 char igb_driver_version[] = DRV_VERSION;
80 static const char igb_driver_string[] =
81 				"Intel(R) Gigabit Ethernet Network Driver";
82 static const char igb_copyright[] =
83 				"Copyright (c) 2007-2014 Intel Corporation.";
84 
85 static const struct e1000_info *igb_info_tbl[] = {
86 	[board_82575] = &e1000_82575_info,
87 };
88 
89 static const struct pci_device_id igb_pci_tbl[] = {
90 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
91 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
92 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
93 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
94 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
95 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
96 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
97 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
98 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
99 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
100 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
101 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
102 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
103 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
104 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
105 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
106 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
107 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
108 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
109 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
110 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
111 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
112 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
113 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
114 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
115 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
116 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
117 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
118 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
119 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
120 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
121 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
122 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
123 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
124 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
125 	/* required last entry */
126 	{0, }
127 };
128 
129 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
130 
131 static int igb_setup_all_tx_resources(struct igb_adapter *);
132 static int igb_setup_all_rx_resources(struct igb_adapter *);
133 static void igb_free_all_tx_resources(struct igb_adapter *);
134 static void igb_free_all_rx_resources(struct igb_adapter *);
135 static void igb_setup_mrqc(struct igb_adapter *);
136 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
137 static void igb_remove(struct pci_dev *pdev);
138 static int igb_sw_init(struct igb_adapter *);
139 int igb_open(struct net_device *);
140 int igb_close(struct net_device *);
141 static void igb_configure(struct igb_adapter *);
142 static void igb_configure_tx(struct igb_adapter *);
143 static void igb_configure_rx(struct igb_adapter *);
144 static void igb_clean_all_tx_rings(struct igb_adapter *);
145 static void igb_clean_all_rx_rings(struct igb_adapter *);
146 static void igb_clean_tx_ring(struct igb_ring *);
147 static void igb_clean_rx_ring(struct igb_ring *);
148 static void igb_set_rx_mode(struct net_device *);
149 static void igb_update_phy_info(struct timer_list *);
150 static void igb_watchdog(struct timer_list *);
151 static void igb_watchdog_task(struct work_struct *);
152 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
153 static void igb_get_stats64(struct net_device *dev,
154 			    struct rtnl_link_stats64 *stats);
155 static int igb_change_mtu(struct net_device *, int);
156 static int igb_set_mac(struct net_device *, void *);
157 static void igb_set_uta(struct igb_adapter *adapter, bool set);
158 static irqreturn_t igb_intr(int irq, void *);
159 static irqreturn_t igb_intr_msi(int irq, void *);
160 static irqreturn_t igb_msix_other(int irq, void *);
161 static irqreturn_t igb_msix_ring(int irq, void *);
162 #ifdef CONFIG_IGB_DCA
163 static void igb_update_dca(struct igb_q_vector *);
164 static void igb_setup_dca(struct igb_adapter *);
165 #endif /* CONFIG_IGB_DCA */
166 static int igb_poll(struct napi_struct *, int);
167 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
168 static int igb_clean_rx_irq(struct igb_q_vector *, int);
169 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
170 static void igb_tx_timeout(struct net_device *);
171 static void igb_reset_task(struct work_struct *);
172 static void igb_vlan_mode(struct net_device *netdev,
173 			  netdev_features_t features);
174 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
175 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
176 static void igb_restore_vlan(struct igb_adapter *);
177 static void igb_rar_set_index(struct igb_adapter *, u32);
178 static void igb_ping_all_vfs(struct igb_adapter *);
179 static void igb_msg_task(struct igb_adapter *);
180 static void igb_vmm_control(struct igb_adapter *);
181 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
182 static void igb_flush_mac_table(struct igb_adapter *);
183 static int igb_available_rars(struct igb_adapter *, u8);
184 static void igb_set_default_mac_filter(struct igb_adapter *);
185 static int igb_uc_sync(struct net_device *, const unsigned char *);
186 static int igb_uc_unsync(struct net_device *, const unsigned char *);
187 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
188 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
189 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
190 			       int vf, u16 vlan, u8 qos, __be16 vlan_proto);
191 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
192 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
193 				   bool setting);
194 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
195 				bool setting);
196 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
197 				 struct ifla_vf_info *ivi);
198 static void igb_check_vf_rate_limit(struct igb_adapter *);
199 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
200 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
201 
202 #ifdef CONFIG_PCI_IOV
203 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
204 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
205 static int igb_disable_sriov(struct pci_dev *dev);
206 static int igb_pci_disable_sriov(struct pci_dev *dev);
207 #endif
208 
209 static int igb_suspend(struct device *);
210 static int igb_resume(struct device *);
211 static int igb_runtime_suspend(struct device *dev);
212 static int igb_runtime_resume(struct device *dev);
213 static int igb_runtime_idle(struct device *dev);
214 static const struct dev_pm_ops igb_pm_ops = {
215 	SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
216 	SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
217 			igb_runtime_idle)
218 };
219 static void igb_shutdown(struct pci_dev *);
220 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
221 #ifdef CONFIG_IGB_DCA
222 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
223 static struct notifier_block dca_notifier = {
224 	.notifier_call	= igb_notify_dca,
225 	.next		= NULL,
226 	.priority	= 0
227 };
228 #endif
229 #ifdef CONFIG_NET_POLL_CONTROLLER
230 /* for netdump / net console */
231 static void igb_netpoll(struct net_device *);
232 #endif
233 #ifdef CONFIG_PCI_IOV
234 static unsigned int max_vfs;
235 module_param(max_vfs, uint, 0);
236 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
237 #endif /* CONFIG_PCI_IOV */
238 
239 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
240 		     pci_channel_state_t);
241 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
242 static void igb_io_resume(struct pci_dev *);
243 
244 static const struct pci_error_handlers igb_err_handler = {
245 	.error_detected = igb_io_error_detected,
246 	.slot_reset = igb_io_slot_reset,
247 	.resume = igb_io_resume,
248 };
249 
250 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
251 
252 static struct pci_driver igb_driver = {
253 	.name     = igb_driver_name,
254 	.id_table = igb_pci_tbl,
255 	.probe    = igb_probe,
256 	.remove   = igb_remove,
257 #ifdef CONFIG_PM
258 	.driver.pm = &igb_pm_ops,
259 #endif
260 	.shutdown = igb_shutdown,
261 	.sriov_configure = igb_pci_sriov_configure,
262 	.err_handler = &igb_err_handler
263 };
264 
265 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
266 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
267 MODULE_LICENSE("GPL");
268 MODULE_VERSION(DRV_VERSION);
269 
270 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
271 static int debug = -1;
272 module_param(debug, int, 0);
273 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
274 
275 struct igb_reg_info {
276 	u32 ofs;
277 	char *name;
278 };
279 
280 static const struct igb_reg_info igb_reg_info_tbl[] = {
281 
282 	/* General Registers */
283 	{E1000_CTRL, "CTRL"},
284 	{E1000_STATUS, "STATUS"},
285 	{E1000_CTRL_EXT, "CTRL_EXT"},
286 
287 	/* Interrupt Registers */
288 	{E1000_ICR, "ICR"},
289 
290 	/* RX Registers */
291 	{E1000_RCTL, "RCTL"},
292 	{E1000_RDLEN(0), "RDLEN"},
293 	{E1000_RDH(0), "RDH"},
294 	{E1000_RDT(0), "RDT"},
295 	{E1000_RXDCTL(0), "RXDCTL"},
296 	{E1000_RDBAL(0), "RDBAL"},
297 	{E1000_RDBAH(0), "RDBAH"},
298 
299 	/* TX Registers */
300 	{E1000_TCTL, "TCTL"},
301 	{E1000_TDBAL(0), "TDBAL"},
302 	{E1000_TDBAH(0), "TDBAH"},
303 	{E1000_TDLEN(0), "TDLEN"},
304 	{E1000_TDH(0), "TDH"},
305 	{E1000_TDT(0), "TDT"},
306 	{E1000_TXDCTL(0), "TXDCTL"},
307 	{E1000_TDFH, "TDFH"},
308 	{E1000_TDFT, "TDFT"},
309 	{E1000_TDFHS, "TDFHS"},
310 	{E1000_TDFPC, "TDFPC"},
311 
312 	/* List Terminator */
313 	{}
314 };
315 
316 /* igb_regdump - register printout routine */
317 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
318 {
319 	int n = 0;
320 	char rname[16];
321 	u32 regs[8];
322 
323 	switch (reginfo->ofs) {
324 	case E1000_RDLEN(0):
325 		for (n = 0; n < 4; n++)
326 			regs[n] = rd32(E1000_RDLEN(n));
327 		break;
328 	case E1000_RDH(0):
329 		for (n = 0; n < 4; n++)
330 			regs[n] = rd32(E1000_RDH(n));
331 		break;
332 	case E1000_RDT(0):
333 		for (n = 0; n < 4; n++)
334 			regs[n] = rd32(E1000_RDT(n));
335 		break;
336 	case E1000_RXDCTL(0):
337 		for (n = 0; n < 4; n++)
338 			regs[n] = rd32(E1000_RXDCTL(n));
339 		break;
340 	case E1000_RDBAL(0):
341 		for (n = 0; n < 4; n++)
342 			regs[n] = rd32(E1000_RDBAL(n));
343 		break;
344 	case E1000_RDBAH(0):
345 		for (n = 0; n < 4; n++)
346 			regs[n] = rd32(E1000_RDBAH(n));
347 		break;
348 	case E1000_TDBAL(0):
349 		for (n = 0; n < 4; n++)
350 			regs[n] = rd32(E1000_RDBAL(n));
351 		break;
352 	case E1000_TDBAH(0):
353 		for (n = 0; n < 4; n++)
354 			regs[n] = rd32(E1000_TDBAH(n));
355 		break;
356 	case E1000_TDLEN(0):
357 		for (n = 0; n < 4; n++)
358 			regs[n] = rd32(E1000_TDLEN(n));
359 		break;
360 	case E1000_TDH(0):
361 		for (n = 0; n < 4; n++)
362 			regs[n] = rd32(E1000_TDH(n));
363 		break;
364 	case E1000_TDT(0):
365 		for (n = 0; n < 4; n++)
366 			regs[n] = rd32(E1000_TDT(n));
367 		break;
368 	case E1000_TXDCTL(0):
369 		for (n = 0; n < 4; n++)
370 			regs[n] = rd32(E1000_TXDCTL(n));
371 		break;
372 	default:
373 		pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
374 		return;
375 	}
376 
377 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
378 	pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
379 		regs[2], regs[3]);
380 }
381 
382 /* igb_dump - Print registers, Tx-rings and Rx-rings */
383 static void igb_dump(struct igb_adapter *adapter)
384 {
385 	struct net_device *netdev = adapter->netdev;
386 	struct e1000_hw *hw = &adapter->hw;
387 	struct igb_reg_info *reginfo;
388 	struct igb_ring *tx_ring;
389 	union e1000_adv_tx_desc *tx_desc;
390 	struct my_u0 { u64 a; u64 b; } *u0;
391 	struct igb_ring *rx_ring;
392 	union e1000_adv_rx_desc *rx_desc;
393 	u32 staterr;
394 	u16 i, n;
395 
396 	if (!netif_msg_hw(adapter))
397 		return;
398 
399 	/* Print netdevice Info */
400 	if (netdev) {
401 		dev_info(&adapter->pdev->dev, "Net device Info\n");
402 		pr_info("Device Name     state            trans_start\n");
403 		pr_info("%-15s %016lX %016lX\n", netdev->name,
404 			netdev->state, dev_trans_start(netdev));
405 	}
406 
407 	/* Print Registers */
408 	dev_info(&adapter->pdev->dev, "Register Dump\n");
409 	pr_info(" Register Name   Value\n");
410 	for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
411 	     reginfo->name; reginfo++) {
412 		igb_regdump(hw, reginfo);
413 	}
414 
415 	/* Print TX Ring Summary */
416 	if (!netdev || !netif_running(netdev))
417 		goto exit;
418 
419 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
420 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
421 	for (n = 0; n < adapter->num_tx_queues; n++) {
422 		struct igb_tx_buffer *buffer_info;
423 		tx_ring = adapter->tx_ring[n];
424 		buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
425 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
426 			n, tx_ring->next_to_use, tx_ring->next_to_clean,
427 			(u64)dma_unmap_addr(buffer_info, dma),
428 			dma_unmap_len(buffer_info, len),
429 			buffer_info->next_to_watch,
430 			(u64)buffer_info->time_stamp);
431 	}
432 
433 	/* Print TX Rings */
434 	if (!netif_msg_tx_done(adapter))
435 		goto rx_ring_summary;
436 
437 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
438 
439 	/* Transmit Descriptor Formats
440 	 *
441 	 * Advanced Transmit Descriptor
442 	 *   +--------------------------------------------------------------+
443 	 * 0 |         Buffer Address [63:0]                                |
444 	 *   +--------------------------------------------------------------+
445 	 * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
446 	 *   +--------------------------------------------------------------+
447 	 *   63      46 45    40 39 38 36 35 32 31   24             15       0
448 	 */
449 
450 	for (n = 0; n < adapter->num_tx_queues; n++) {
451 		tx_ring = adapter->tx_ring[n];
452 		pr_info("------------------------------------\n");
453 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
454 		pr_info("------------------------------------\n");
455 		pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
456 
457 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
458 			const char *next_desc;
459 			struct igb_tx_buffer *buffer_info;
460 			tx_desc = IGB_TX_DESC(tx_ring, i);
461 			buffer_info = &tx_ring->tx_buffer_info[i];
462 			u0 = (struct my_u0 *)tx_desc;
463 			if (i == tx_ring->next_to_use &&
464 			    i == tx_ring->next_to_clean)
465 				next_desc = " NTC/U";
466 			else if (i == tx_ring->next_to_use)
467 				next_desc = " NTU";
468 			else if (i == tx_ring->next_to_clean)
469 				next_desc = " NTC";
470 			else
471 				next_desc = "";
472 
473 			pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
474 				i, le64_to_cpu(u0->a),
475 				le64_to_cpu(u0->b),
476 				(u64)dma_unmap_addr(buffer_info, dma),
477 				dma_unmap_len(buffer_info, len),
478 				buffer_info->next_to_watch,
479 				(u64)buffer_info->time_stamp,
480 				buffer_info->skb, next_desc);
481 
482 			if (netif_msg_pktdata(adapter) && buffer_info->skb)
483 				print_hex_dump(KERN_INFO, "",
484 					DUMP_PREFIX_ADDRESS,
485 					16, 1, buffer_info->skb->data,
486 					dma_unmap_len(buffer_info, len),
487 					true);
488 		}
489 	}
490 
491 	/* Print RX Rings Summary */
492 rx_ring_summary:
493 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
494 	pr_info("Queue [NTU] [NTC]\n");
495 	for (n = 0; n < adapter->num_rx_queues; n++) {
496 		rx_ring = adapter->rx_ring[n];
497 		pr_info(" %5d %5X %5X\n",
498 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
499 	}
500 
501 	/* Print RX Rings */
502 	if (!netif_msg_rx_status(adapter))
503 		goto exit;
504 
505 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
506 
507 	/* Advanced Receive Descriptor (Read) Format
508 	 *    63                                           1        0
509 	 *    +-----------------------------------------------------+
510 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
511 	 *    +----------------------------------------------+------+
512 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
513 	 *    +-----------------------------------------------------+
514 	 *
515 	 *
516 	 * Advanced Receive Descriptor (Write-Back) Format
517 	 *
518 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
519 	 *   +------------------------------------------------------+
520 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
521 	 *   | Checksum   Ident  |   |           |    | Type | Type |
522 	 *   +------------------------------------------------------+
523 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
524 	 *   +------------------------------------------------------+
525 	 *   63       48 47    32 31            20 19               0
526 	 */
527 
528 	for (n = 0; n < adapter->num_rx_queues; n++) {
529 		rx_ring = adapter->rx_ring[n];
530 		pr_info("------------------------------------\n");
531 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
532 		pr_info("------------------------------------\n");
533 		pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
534 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
535 
536 		for (i = 0; i < rx_ring->count; i++) {
537 			const char *next_desc;
538 			struct igb_rx_buffer *buffer_info;
539 			buffer_info = &rx_ring->rx_buffer_info[i];
540 			rx_desc = IGB_RX_DESC(rx_ring, i);
541 			u0 = (struct my_u0 *)rx_desc;
542 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
543 
544 			if (i == rx_ring->next_to_use)
545 				next_desc = " NTU";
546 			else if (i == rx_ring->next_to_clean)
547 				next_desc = " NTC";
548 			else
549 				next_desc = "";
550 
551 			if (staterr & E1000_RXD_STAT_DD) {
552 				/* Descriptor Done */
553 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
554 					"RWB", i,
555 					le64_to_cpu(u0->a),
556 					le64_to_cpu(u0->b),
557 					next_desc);
558 			} else {
559 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
560 					"R  ", i,
561 					le64_to_cpu(u0->a),
562 					le64_to_cpu(u0->b),
563 					(u64)buffer_info->dma,
564 					next_desc);
565 
566 				if (netif_msg_pktdata(adapter) &&
567 				    buffer_info->dma && buffer_info->page) {
568 					print_hex_dump(KERN_INFO, "",
569 					  DUMP_PREFIX_ADDRESS,
570 					  16, 1,
571 					  page_address(buffer_info->page) +
572 						      buffer_info->page_offset,
573 					  igb_rx_bufsz(rx_ring), true);
574 				}
575 			}
576 		}
577 	}
578 
579 exit:
580 	return;
581 }
582 
583 /**
584  *  igb_get_i2c_data - Reads the I2C SDA data bit
585  *  @hw: pointer to hardware structure
586  *  @i2cctl: Current value of I2CCTL register
587  *
588  *  Returns the I2C data bit value
589  **/
590 static int igb_get_i2c_data(void *data)
591 {
592 	struct igb_adapter *adapter = (struct igb_adapter *)data;
593 	struct e1000_hw *hw = &adapter->hw;
594 	s32 i2cctl = rd32(E1000_I2CPARAMS);
595 
596 	return !!(i2cctl & E1000_I2C_DATA_IN);
597 }
598 
599 /**
600  *  igb_set_i2c_data - Sets the I2C data bit
601  *  @data: pointer to hardware structure
602  *  @state: I2C data value (0 or 1) to set
603  *
604  *  Sets the I2C data bit
605  **/
606 static void igb_set_i2c_data(void *data, int state)
607 {
608 	struct igb_adapter *adapter = (struct igb_adapter *)data;
609 	struct e1000_hw *hw = &adapter->hw;
610 	s32 i2cctl = rd32(E1000_I2CPARAMS);
611 
612 	if (state)
613 		i2cctl |= E1000_I2C_DATA_OUT;
614 	else
615 		i2cctl &= ~E1000_I2C_DATA_OUT;
616 
617 	i2cctl &= ~E1000_I2C_DATA_OE_N;
618 	i2cctl |= E1000_I2C_CLK_OE_N;
619 	wr32(E1000_I2CPARAMS, i2cctl);
620 	wrfl();
621 
622 }
623 
624 /**
625  *  igb_set_i2c_clk - Sets the I2C SCL clock
626  *  @data: pointer to hardware structure
627  *  @state: state to set clock
628  *
629  *  Sets the I2C clock line to state
630  **/
631 static void igb_set_i2c_clk(void *data, int state)
632 {
633 	struct igb_adapter *adapter = (struct igb_adapter *)data;
634 	struct e1000_hw *hw = &adapter->hw;
635 	s32 i2cctl = rd32(E1000_I2CPARAMS);
636 
637 	if (state) {
638 		i2cctl |= E1000_I2C_CLK_OUT;
639 		i2cctl &= ~E1000_I2C_CLK_OE_N;
640 	} else {
641 		i2cctl &= ~E1000_I2C_CLK_OUT;
642 		i2cctl &= ~E1000_I2C_CLK_OE_N;
643 	}
644 	wr32(E1000_I2CPARAMS, i2cctl);
645 	wrfl();
646 }
647 
648 /**
649  *  igb_get_i2c_clk - Gets the I2C SCL clock state
650  *  @data: pointer to hardware structure
651  *
652  *  Gets the I2C clock state
653  **/
654 static int igb_get_i2c_clk(void *data)
655 {
656 	struct igb_adapter *adapter = (struct igb_adapter *)data;
657 	struct e1000_hw *hw = &adapter->hw;
658 	s32 i2cctl = rd32(E1000_I2CPARAMS);
659 
660 	return !!(i2cctl & E1000_I2C_CLK_IN);
661 }
662 
663 static const struct i2c_algo_bit_data igb_i2c_algo = {
664 	.setsda		= igb_set_i2c_data,
665 	.setscl		= igb_set_i2c_clk,
666 	.getsda		= igb_get_i2c_data,
667 	.getscl		= igb_get_i2c_clk,
668 	.udelay		= 5,
669 	.timeout	= 20,
670 };
671 
672 /**
673  *  igb_get_hw_dev - return device
674  *  @hw: pointer to hardware structure
675  *
676  *  used by hardware layer to print debugging information
677  **/
678 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
679 {
680 	struct igb_adapter *adapter = hw->back;
681 	return adapter->netdev;
682 }
683 
684 /**
685  *  igb_init_module - Driver Registration Routine
686  *
687  *  igb_init_module is the first routine called when the driver is
688  *  loaded. All it does is register with the PCI subsystem.
689  **/
690 static int __init igb_init_module(void)
691 {
692 	int ret;
693 
694 	pr_info("%s - version %s\n",
695 	       igb_driver_string, igb_driver_version);
696 	pr_info("%s\n", igb_copyright);
697 
698 #ifdef CONFIG_IGB_DCA
699 	dca_register_notify(&dca_notifier);
700 #endif
701 	ret = pci_register_driver(&igb_driver);
702 	return ret;
703 }
704 
705 module_init(igb_init_module);
706 
707 /**
708  *  igb_exit_module - Driver Exit Cleanup Routine
709  *
710  *  igb_exit_module is called just before the driver is removed
711  *  from memory.
712  **/
713 static void __exit igb_exit_module(void)
714 {
715 #ifdef CONFIG_IGB_DCA
716 	dca_unregister_notify(&dca_notifier);
717 #endif
718 	pci_unregister_driver(&igb_driver);
719 }
720 
721 module_exit(igb_exit_module);
722 
723 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
724 /**
725  *  igb_cache_ring_register - Descriptor ring to register mapping
726  *  @adapter: board private structure to initialize
727  *
728  *  Once we know the feature-set enabled for the device, we'll cache
729  *  the register offset the descriptor ring is assigned to.
730  **/
731 static void igb_cache_ring_register(struct igb_adapter *adapter)
732 {
733 	int i = 0, j = 0;
734 	u32 rbase_offset = adapter->vfs_allocated_count;
735 
736 	switch (adapter->hw.mac.type) {
737 	case e1000_82576:
738 		/* The queues are allocated for virtualization such that VF 0
739 		 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
740 		 * In order to avoid collision we start at the first free queue
741 		 * and continue consuming queues in the same sequence
742 		 */
743 		if (adapter->vfs_allocated_count) {
744 			for (; i < adapter->rss_queues; i++)
745 				adapter->rx_ring[i]->reg_idx = rbase_offset +
746 							       Q_IDX_82576(i);
747 		}
748 		/* Fall through */
749 	case e1000_82575:
750 	case e1000_82580:
751 	case e1000_i350:
752 	case e1000_i354:
753 	case e1000_i210:
754 	case e1000_i211:
755 		/* Fall through */
756 	default:
757 		for (; i < adapter->num_rx_queues; i++)
758 			adapter->rx_ring[i]->reg_idx = rbase_offset + i;
759 		for (; j < adapter->num_tx_queues; j++)
760 			adapter->tx_ring[j]->reg_idx = rbase_offset + j;
761 		break;
762 	}
763 }
764 
765 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
766 {
767 	struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
768 	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
769 	u32 value = 0;
770 
771 	if (E1000_REMOVED(hw_addr))
772 		return ~value;
773 
774 	value = readl(&hw_addr[reg]);
775 
776 	/* reads should not return all F's */
777 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
778 		struct net_device *netdev = igb->netdev;
779 		hw->hw_addr = NULL;
780 		netdev_err(netdev, "PCIe link lost\n");
781 	}
782 
783 	return value;
784 }
785 
786 /**
787  *  igb_write_ivar - configure ivar for given MSI-X vector
788  *  @hw: pointer to the HW structure
789  *  @msix_vector: vector number we are allocating to a given ring
790  *  @index: row index of IVAR register to write within IVAR table
791  *  @offset: column offset of in IVAR, should be multiple of 8
792  *
793  *  This function is intended to handle the writing of the IVAR register
794  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
795  *  each containing an cause allocation for an Rx and Tx ring, and a
796  *  variable number of rows depending on the number of queues supported.
797  **/
798 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
799 			   int index, int offset)
800 {
801 	u32 ivar = array_rd32(E1000_IVAR0, index);
802 
803 	/* clear any bits that are currently set */
804 	ivar &= ~((u32)0xFF << offset);
805 
806 	/* write vector and valid bit */
807 	ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
808 
809 	array_wr32(E1000_IVAR0, index, ivar);
810 }
811 
812 #define IGB_N0_QUEUE -1
813 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
814 {
815 	struct igb_adapter *adapter = q_vector->adapter;
816 	struct e1000_hw *hw = &adapter->hw;
817 	int rx_queue = IGB_N0_QUEUE;
818 	int tx_queue = IGB_N0_QUEUE;
819 	u32 msixbm = 0;
820 
821 	if (q_vector->rx.ring)
822 		rx_queue = q_vector->rx.ring->reg_idx;
823 	if (q_vector->tx.ring)
824 		tx_queue = q_vector->tx.ring->reg_idx;
825 
826 	switch (hw->mac.type) {
827 	case e1000_82575:
828 		/* The 82575 assigns vectors using a bitmask, which matches the
829 		 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
830 		 * or more queues to a vector, we write the appropriate bits
831 		 * into the MSIXBM register for that vector.
832 		 */
833 		if (rx_queue > IGB_N0_QUEUE)
834 			msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
835 		if (tx_queue > IGB_N0_QUEUE)
836 			msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
837 		if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
838 			msixbm |= E1000_EIMS_OTHER;
839 		array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
840 		q_vector->eims_value = msixbm;
841 		break;
842 	case e1000_82576:
843 		/* 82576 uses a table that essentially consists of 2 columns
844 		 * with 8 rows.  The ordering is column-major so we use the
845 		 * lower 3 bits as the row index, and the 4th bit as the
846 		 * column offset.
847 		 */
848 		if (rx_queue > IGB_N0_QUEUE)
849 			igb_write_ivar(hw, msix_vector,
850 				       rx_queue & 0x7,
851 				       (rx_queue & 0x8) << 1);
852 		if (tx_queue > IGB_N0_QUEUE)
853 			igb_write_ivar(hw, msix_vector,
854 				       tx_queue & 0x7,
855 				       ((tx_queue & 0x8) << 1) + 8);
856 		q_vector->eims_value = BIT(msix_vector);
857 		break;
858 	case e1000_82580:
859 	case e1000_i350:
860 	case e1000_i354:
861 	case e1000_i210:
862 	case e1000_i211:
863 		/* On 82580 and newer adapters the scheme is similar to 82576
864 		 * however instead of ordering column-major we have things
865 		 * ordered row-major.  So we traverse the table by using
866 		 * bit 0 as the column offset, and the remaining bits as the
867 		 * row index.
868 		 */
869 		if (rx_queue > IGB_N0_QUEUE)
870 			igb_write_ivar(hw, msix_vector,
871 				       rx_queue >> 1,
872 				       (rx_queue & 0x1) << 4);
873 		if (tx_queue > IGB_N0_QUEUE)
874 			igb_write_ivar(hw, msix_vector,
875 				       tx_queue >> 1,
876 				       ((tx_queue & 0x1) << 4) + 8);
877 		q_vector->eims_value = BIT(msix_vector);
878 		break;
879 	default:
880 		BUG();
881 		break;
882 	}
883 
884 	/* add q_vector eims value to global eims_enable_mask */
885 	adapter->eims_enable_mask |= q_vector->eims_value;
886 
887 	/* configure q_vector to set itr on first interrupt */
888 	q_vector->set_itr = 1;
889 }
890 
891 /**
892  *  igb_configure_msix - Configure MSI-X hardware
893  *  @adapter: board private structure to initialize
894  *
895  *  igb_configure_msix sets up the hardware to properly
896  *  generate MSI-X interrupts.
897  **/
898 static void igb_configure_msix(struct igb_adapter *adapter)
899 {
900 	u32 tmp;
901 	int i, vector = 0;
902 	struct e1000_hw *hw = &adapter->hw;
903 
904 	adapter->eims_enable_mask = 0;
905 
906 	/* set vector for other causes, i.e. link changes */
907 	switch (hw->mac.type) {
908 	case e1000_82575:
909 		tmp = rd32(E1000_CTRL_EXT);
910 		/* enable MSI-X PBA support*/
911 		tmp |= E1000_CTRL_EXT_PBA_CLR;
912 
913 		/* Auto-Mask interrupts upon ICR read. */
914 		tmp |= E1000_CTRL_EXT_EIAME;
915 		tmp |= E1000_CTRL_EXT_IRCA;
916 
917 		wr32(E1000_CTRL_EXT, tmp);
918 
919 		/* enable msix_other interrupt */
920 		array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
921 		adapter->eims_other = E1000_EIMS_OTHER;
922 
923 		break;
924 
925 	case e1000_82576:
926 	case e1000_82580:
927 	case e1000_i350:
928 	case e1000_i354:
929 	case e1000_i210:
930 	case e1000_i211:
931 		/* Turn on MSI-X capability first, or our settings
932 		 * won't stick.  And it will take days to debug.
933 		 */
934 		wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
935 		     E1000_GPIE_PBA | E1000_GPIE_EIAME |
936 		     E1000_GPIE_NSICR);
937 
938 		/* enable msix_other interrupt */
939 		adapter->eims_other = BIT(vector);
940 		tmp = (vector++ | E1000_IVAR_VALID) << 8;
941 
942 		wr32(E1000_IVAR_MISC, tmp);
943 		break;
944 	default:
945 		/* do nothing, since nothing else supports MSI-X */
946 		break;
947 	} /* switch (hw->mac.type) */
948 
949 	adapter->eims_enable_mask |= adapter->eims_other;
950 
951 	for (i = 0; i < adapter->num_q_vectors; i++)
952 		igb_assign_vector(adapter->q_vector[i], vector++);
953 
954 	wrfl();
955 }
956 
957 /**
958  *  igb_request_msix - Initialize MSI-X interrupts
959  *  @adapter: board private structure to initialize
960  *
961  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
962  *  kernel.
963  **/
964 static int igb_request_msix(struct igb_adapter *adapter)
965 {
966 	struct net_device *netdev = adapter->netdev;
967 	int i, err = 0, vector = 0, free_vector = 0;
968 
969 	err = request_irq(adapter->msix_entries[vector].vector,
970 			  igb_msix_other, 0, netdev->name, adapter);
971 	if (err)
972 		goto err_out;
973 
974 	for (i = 0; i < adapter->num_q_vectors; i++) {
975 		struct igb_q_vector *q_vector = adapter->q_vector[i];
976 
977 		vector++;
978 
979 		q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
980 
981 		if (q_vector->rx.ring && q_vector->tx.ring)
982 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
983 				q_vector->rx.ring->queue_index);
984 		else if (q_vector->tx.ring)
985 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
986 				q_vector->tx.ring->queue_index);
987 		else if (q_vector->rx.ring)
988 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
989 				q_vector->rx.ring->queue_index);
990 		else
991 			sprintf(q_vector->name, "%s-unused", netdev->name);
992 
993 		err = request_irq(adapter->msix_entries[vector].vector,
994 				  igb_msix_ring, 0, q_vector->name,
995 				  q_vector);
996 		if (err)
997 			goto err_free;
998 	}
999 
1000 	igb_configure_msix(adapter);
1001 	return 0;
1002 
1003 err_free:
1004 	/* free already assigned IRQs */
1005 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
1006 
1007 	vector--;
1008 	for (i = 0; i < vector; i++) {
1009 		free_irq(adapter->msix_entries[free_vector++].vector,
1010 			 adapter->q_vector[i]);
1011 	}
1012 err_out:
1013 	return err;
1014 }
1015 
1016 /**
1017  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
1018  *  @adapter: board private structure to initialize
1019  *  @v_idx: Index of vector to be freed
1020  *
1021  *  This function frees the memory allocated to the q_vector.
1022  **/
1023 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1024 {
1025 	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1026 
1027 	adapter->q_vector[v_idx] = NULL;
1028 
1029 	/* igb_get_stats64() might access the rings on this vector,
1030 	 * we must wait a grace period before freeing it.
1031 	 */
1032 	if (q_vector)
1033 		kfree_rcu(q_vector, rcu);
1034 }
1035 
1036 /**
1037  *  igb_reset_q_vector - Reset config for interrupt vector
1038  *  @adapter: board private structure to initialize
1039  *  @v_idx: Index of vector to be reset
1040  *
1041  *  If NAPI is enabled it will delete any references to the
1042  *  NAPI struct. This is preparation for igb_free_q_vector.
1043  **/
1044 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1045 {
1046 	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1047 
1048 	/* Coming from igb_set_interrupt_capability, the vectors are not yet
1049 	 * allocated. So, q_vector is NULL so we should stop here.
1050 	 */
1051 	if (!q_vector)
1052 		return;
1053 
1054 	if (q_vector->tx.ring)
1055 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1056 
1057 	if (q_vector->rx.ring)
1058 		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1059 
1060 	netif_napi_del(&q_vector->napi);
1061 
1062 }
1063 
1064 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1065 {
1066 	int v_idx = adapter->num_q_vectors;
1067 
1068 	if (adapter->flags & IGB_FLAG_HAS_MSIX)
1069 		pci_disable_msix(adapter->pdev);
1070 	else if (adapter->flags & IGB_FLAG_HAS_MSI)
1071 		pci_disable_msi(adapter->pdev);
1072 
1073 	while (v_idx--)
1074 		igb_reset_q_vector(adapter, v_idx);
1075 }
1076 
1077 /**
1078  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1079  *  @adapter: board private structure to initialize
1080  *
1081  *  This function frees the memory allocated to the q_vectors.  In addition if
1082  *  NAPI is enabled it will delete any references to the NAPI struct prior
1083  *  to freeing the q_vector.
1084  **/
1085 static void igb_free_q_vectors(struct igb_adapter *adapter)
1086 {
1087 	int v_idx = adapter->num_q_vectors;
1088 
1089 	adapter->num_tx_queues = 0;
1090 	adapter->num_rx_queues = 0;
1091 	adapter->num_q_vectors = 0;
1092 
1093 	while (v_idx--) {
1094 		igb_reset_q_vector(adapter, v_idx);
1095 		igb_free_q_vector(adapter, v_idx);
1096 	}
1097 }
1098 
1099 /**
1100  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1101  *  @adapter: board private structure to initialize
1102  *
1103  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1104  *  MSI-X interrupts allocated.
1105  */
1106 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1107 {
1108 	igb_free_q_vectors(adapter);
1109 	igb_reset_interrupt_capability(adapter);
1110 }
1111 
1112 /**
1113  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1114  *  @adapter: board private structure to initialize
1115  *  @msix: boolean value of MSIX capability
1116  *
1117  *  Attempt to configure interrupts using the best available
1118  *  capabilities of the hardware and kernel.
1119  **/
1120 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1121 {
1122 	int err;
1123 	int numvecs, i;
1124 
1125 	if (!msix)
1126 		goto msi_only;
1127 	adapter->flags |= IGB_FLAG_HAS_MSIX;
1128 
1129 	/* Number of supported queues. */
1130 	adapter->num_rx_queues = adapter->rss_queues;
1131 	if (adapter->vfs_allocated_count)
1132 		adapter->num_tx_queues = 1;
1133 	else
1134 		adapter->num_tx_queues = adapter->rss_queues;
1135 
1136 	/* start with one vector for every Rx queue */
1137 	numvecs = adapter->num_rx_queues;
1138 
1139 	/* if Tx handler is separate add 1 for every Tx queue */
1140 	if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1141 		numvecs += adapter->num_tx_queues;
1142 
1143 	/* store the number of vectors reserved for queues */
1144 	adapter->num_q_vectors = numvecs;
1145 
1146 	/* add 1 vector for link status interrupts */
1147 	numvecs++;
1148 	for (i = 0; i < numvecs; i++)
1149 		adapter->msix_entries[i].entry = i;
1150 
1151 	err = pci_enable_msix_range(adapter->pdev,
1152 				    adapter->msix_entries,
1153 				    numvecs,
1154 				    numvecs);
1155 	if (err > 0)
1156 		return;
1157 
1158 	igb_reset_interrupt_capability(adapter);
1159 
1160 	/* If we can't do MSI-X, try MSI */
1161 msi_only:
1162 	adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1163 #ifdef CONFIG_PCI_IOV
1164 	/* disable SR-IOV for non MSI-X configurations */
1165 	if (adapter->vf_data) {
1166 		struct e1000_hw *hw = &adapter->hw;
1167 		/* disable iov and allow time for transactions to clear */
1168 		pci_disable_sriov(adapter->pdev);
1169 		msleep(500);
1170 
1171 		kfree(adapter->vf_mac_list);
1172 		adapter->vf_mac_list = NULL;
1173 		kfree(adapter->vf_data);
1174 		adapter->vf_data = NULL;
1175 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1176 		wrfl();
1177 		msleep(100);
1178 		dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1179 	}
1180 #endif
1181 	adapter->vfs_allocated_count = 0;
1182 	adapter->rss_queues = 1;
1183 	adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1184 	adapter->num_rx_queues = 1;
1185 	adapter->num_tx_queues = 1;
1186 	adapter->num_q_vectors = 1;
1187 	if (!pci_enable_msi(adapter->pdev))
1188 		adapter->flags |= IGB_FLAG_HAS_MSI;
1189 }
1190 
1191 static void igb_add_ring(struct igb_ring *ring,
1192 			 struct igb_ring_container *head)
1193 {
1194 	head->ring = ring;
1195 	head->count++;
1196 }
1197 
1198 /**
1199  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1200  *  @adapter: board private structure to initialize
1201  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1202  *  @v_idx: index of vector in adapter struct
1203  *  @txr_count: total number of Tx rings to allocate
1204  *  @txr_idx: index of first Tx ring to allocate
1205  *  @rxr_count: total number of Rx rings to allocate
1206  *  @rxr_idx: index of first Rx ring to allocate
1207  *
1208  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1209  **/
1210 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1211 			      int v_count, int v_idx,
1212 			      int txr_count, int txr_idx,
1213 			      int rxr_count, int rxr_idx)
1214 {
1215 	struct igb_q_vector *q_vector;
1216 	struct igb_ring *ring;
1217 	int ring_count, size;
1218 
1219 	/* igb only supports 1 Tx and/or 1 Rx queue per vector */
1220 	if (txr_count > 1 || rxr_count > 1)
1221 		return -ENOMEM;
1222 
1223 	ring_count = txr_count + rxr_count;
1224 	size = sizeof(struct igb_q_vector) +
1225 	       (sizeof(struct igb_ring) * ring_count);
1226 
1227 	/* allocate q_vector and rings */
1228 	q_vector = adapter->q_vector[v_idx];
1229 	if (!q_vector) {
1230 		q_vector = kzalloc(size, GFP_KERNEL);
1231 	} else if (size > ksize(q_vector)) {
1232 		kfree_rcu(q_vector, rcu);
1233 		q_vector = kzalloc(size, GFP_KERNEL);
1234 	} else {
1235 		memset(q_vector, 0, size);
1236 	}
1237 	if (!q_vector)
1238 		return -ENOMEM;
1239 
1240 	/* initialize NAPI */
1241 	netif_napi_add(adapter->netdev, &q_vector->napi,
1242 		       igb_poll, 64);
1243 
1244 	/* tie q_vector and adapter together */
1245 	adapter->q_vector[v_idx] = q_vector;
1246 	q_vector->adapter = adapter;
1247 
1248 	/* initialize work limits */
1249 	q_vector->tx.work_limit = adapter->tx_work_limit;
1250 
1251 	/* initialize ITR configuration */
1252 	q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1253 	q_vector->itr_val = IGB_START_ITR;
1254 
1255 	/* initialize pointer to rings */
1256 	ring = q_vector->ring;
1257 
1258 	/* intialize ITR */
1259 	if (rxr_count) {
1260 		/* rx or rx/tx vector */
1261 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1262 			q_vector->itr_val = adapter->rx_itr_setting;
1263 	} else {
1264 		/* tx only vector */
1265 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1266 			q_vector->itr_val = adapter->tx_itr_setting;
1267 	}
1268 
1269 	if (txr_count) {
1270 		/* assign generic ring traits */
1271 		ring->dev = &adapter->pdev->dev;
1272 		ring->netdev = adapter->netdev;
1273 
1274 		/* configure backlink on ring */
1275 		ring->q_vector = q_vector;
1276 
1277 		/* update q_vector Tx values */
1278 		igb_add_ring(ring, &q_vector->tx);
1279 
1280 		/* For 82575, context index must be unique per ring. */
1281 		if (adapter->hw.mac.type == e1000_82575)
1282 			set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1283 
1284 		/* apply Tx specific ring traits */
1285 		ring->count = adapter->tx_ring_count;
1286 		ring->queue_index = txr_idx;
1287 
1288 		ring->cbs_enable = false;
1289 		ring->idleslope = 0;
1290 		ring->sendslope = 0;
1291 		ring->hicredit = 0;
1292 		ring->locredit = 0;
1293 
1294 		u64_stats_init(&ring->tx_syncp);
1295 		u64_stats_init(&ring->tx_syncp2);
1296 
1297 		/* assign ring to adapter */
1298 		adapter->tx_ring[txr_idx] = ring;
1299 
1300 		/* push pointer to next ring */
1301 		ring++;
1302 	}
1303 
1304 	if (rxr_count) {
1305 		/* assign generic ring traits */
1306 		ring->dev = &adapter->pdev->dev;
1307 		ring->netdev = adapter->netdev;
1308 
1309 		/* configure backlink on ring */
1310 		ring->q_vector = q_vector;
1311 
1312 		/* update q_vector Rx values */
1313 		igb_add_ring(ring, &q_vector->rx);
1314 
1315 		/* set flag indicating ring supports SCTP checksum offload */
1316 		if (adapter->hw.mac.type >= e1000_82576)
1317 			set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1318 
1319 		/* On i350, i354, i210, and i211, loopback VLAN packets
1320 		 * have the tag byte-swapped.
1321 		 */
1322 		if (adapter->hw.mac.type >= e1000_i350)
1323 			set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1324 
1325 		/* apply Rx specific ring traits */
1326 		ring->count = adapter->rx_ring_count;
1327 		ring->queue_index = rxr_idx;
1328 
1329 		u64_stats_init(&ring->rx_syncp);
1330 
1331 		/* assign ring to adapter */
1332 		adapter->rx_ring[rxr_idx] = ring;
1333 	}
1334 
1335 	return 0;
1336 }
1337 
1338 
1339 /**
1340  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1341  *  @adapter: board private structure to initialize
1342  *
1343  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1344  *  return -ENOMEM.
1345  **/
1346 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1347 {
1348 	int q_vectors = adapter->num_q_vectors;
1349 	int rxr_remaining = adapter->num_rx_queues;
1350 	int txr_remaining = adapter->num_tx_queues;
1351 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1352 	int err;
1353 
1354 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
1355 		for (; rxr_remaining; v_idx++) {
1356 			err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1357 						 0, 0, 1, rxr_idx);
1358 
1359 			if (err)
1360 				goto err_out;
1361 
1362 			/* update counts and index */
1363 			rxr_remaining--;
1364 			rxr_idx++;
1365 		}
1366 	}
1367 
1368 	for (; v_idx < q_vectors; v_idx++) {
1369 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1370 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1371 
1372 		err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1373 					 tqpv, txr_idx, rqpv, rxr_idx);
1374 
1375 		if (err)
1376 			goto err_out;
1377 
1378 		/* update counts and index */
1379 		rxr_remaining -= rqpv;
1380 		txr_remaining -= tqpv;
1381 		rxr_idx++;
1382 		txr_idx++;
1383 	}
1384 
1385 	return 0;
1386 
1387 err_out:
1388 	adapter->num_tx_queues = 0;
1389 	adapter->num_rx_queues = 0;
1390 	adapter->num_q_vectors = 0;
1391 
1392 	while (v_idx--)
1393 		igb_free_q_vector(adapter, v_idx);
1394 
1395 	return -ENOMEM;
1396 }
1397 
1398 /**
1399  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1400  *  @adapter: board private structure to initialize
1401  *  @msix: boolean value of MSIX capability
1402  *
1403  *  This function initializes the interrupts and allocates all of the queues.
1404  **/
1405 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1406 {
1407 	struct pci_dev *pdev = adapter->pdev;
1408 	int err;
1409 
1410 	igb_set_interrupt_capability(adapter, msix);
1411 
1412 	err = igb_alloc_q_vectors(adapter);
1413 	if (err) {
1414 		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1415 		goto err_alloc_q_vectors;
1416 	}
1417 
1418 	igb_cache_ring_register(adapter);
1419 
1420 	return 0;
1421 
1422 err_alloc_q_vectors:
1423 	igb_reset_interrupt_capability(adapter);
1424 	return err;
1425 }
1426 
1427 /**
1428  *  igb_request_irq - initialize interrupts
1429  *  @adapter: board private structure to initialize
1430  *
1431  *  Attempts to configure interrupts using the best available
1432  *  capabilities of the hardware and kernel.
1433  **/
1434 static int igb_request_irq(struct igb_adapter *adapter)
1435 {
1436 	struct net_device *netdev = adapter->netdev;
1437 	struct pci_dev *pdev = adapter->pdev;
1438 	int err = 0;
1439 
1440 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1441 		err = igb_request_msix(adapter);
1442 		if (!err)
1443 			goto request_done;
1444 		/* fall back to MSI */
1445 		igb_free_all_tx_resources(adapter);
1446 		igb_free_all_rx_resources(adapter);
1447 
1448 		igb_clear_interrupt_scheme(adapter);
1449 		err = igb_init_interrupt_scheme(adapter, false);
1450 		if (err)
1451 			goto request_done;
1452 
1453 		igb_setup_all_tx_resources(adapter);
1454 		igb_setup_all_rx_resources(adapter);
1455 		igb_configure(adapter);
1456 	}
1457 
1458 	igb_assign_vector(adapter->q_vector[0], 0);
1459 
1460 	if (adapter->flags & IGB_FLAG_HAS_MSI) {
1461 		err = request_irq(pdev->irq, igb_intr_msi, 0,
1462 				  netdev->name, adapter);
1463 		if (!err)
1464 			goto request_done;
1465 
1466 		/* fall back to legacy interrupts */
1467 		igb_reset_interrupt_capability(adapter);
1468 		adapter->flags &= ~IGB_FLAG_HAS_MSI;
1469 	}
1470 
1471 	err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1472 			  netdev->name, adapter);
1473 
1474 	if (err)
1475 		dev_err(&pdev->dev, "Error %d getting interrupt\n",
1476 			err);
1477 
1478 request_done:
1479 	return err;
1480 }
1481 
1482 static void igb_free_irq(struct igb_adapter *adapter)
1483 {
1484 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1485 		int vector = 0, i;
1486 
1487 		free_irq(adapter->msix_entries[vector++].vector, adapter);
1488 
1489 		for (i = 0; i < adapter->num_q_vectors; i++)
1490 			free_irq(adapter->msix_entries[vector++].vector,
1491 				 adapter->q_vector[i]);
1492 	} else {
1493 		free_irq(adapter->pdev->irq, adapter);
1494 	}
1495 }
1496 
1497 /**
1498  *  igb_irq_disable - Mask off interrupt generation on the NIC
1499  *  @adapter: board private structure
1500  **/
1501 static void igb_irq_disable(struct igb_adapter *adapter)
1502 {
1503 	struct e1000_hw *hw = &adapter->hw;
1504 
1505 	/* we need to be careful when disabling interrupts.  The VFs are also
1506 	 * mapped into these registers and so clearing the bits can cause
1507 	 * issues on the VF drivers so we only need to clear what we set
1508 	 */
1509 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1510 		u32 regval = rd32(E1000_EIAM);
1511 
1512 		wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1513 		wr32(E1000_EIMC, adapter->eims_enable_mask);
1514 		regval = rd32(E1000_EIAC);
1515 		wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1516 	}
1517 
1518 	wr32(E1000_IAM, 0);
1519 	wr32(E1000_IMC, ~0);
1520 	wrfl();
1521 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1522 		int i;
1523 
1524 		for (i = 0; i < adapter->num_q_vectors; i++)
1525 			synchronize_irq(adapter->msix_entries[i].vector);
1526 	} else {
1527 		synchronize_irq(adapter->pdev->irq);
1528 	}
1529 }
1530 
1531 /**
1532  *  igb_irq_enable - Enable default interrupt generation settings
1533  *  @adapter: board private structure
1534  **/
1535 static void igb_irq_enable(struct igb_adapter *adapter)
1536 {
1537 	struct e1000_hw *hw = &adapter->hw;
1538 
1539 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1540 		u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1541 		u32 regval = rd32(E1000_EIAC);
1542 
1543 		wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1544 		regval = rd32(E1000_EIAM);
1545 		wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1546 		wr32(E1000_EIMS, adapter->eims_enable_mask);
1547 		if (adapter->vfs_allocated_count) {
1548 			wr32(E1000_MBVFIMR, 0xFF);
1549 			ims |= E1000_IMS_VMMB;
1550 		}
1551 		wr32(E1000_IMS, ims);
1552 	} else {
1553 		wr32(E1000_IMS, IMS_ENABLE_MASK |
1554 				E1000_IMS_DRSTA);
1555 		wr32(E1000_IAM, IMS_ENABLE_MASK |
1556 				E1000_IMS_DRSTA);
1557 	}
1558 }
1559 
1560 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1561 {
1562 	struct e1000_hw *hw = &adapter->hw;
1563 	u16 pf_id = adapter->vfs_allocated_count;
1564 	u16 vid = adapter->hw.mng_cookie.vlan_id;
1565 	u16 old_vid = adapter->mng_vlan_id;
1566 
1567 	if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1568 		/* add VID to filter table */
1569 		igb_vfta_set(hw, vid, pf_id, true, true);
1570 		adapter->mng_vlan_id = vid;
1571 	} else {
1572 		adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1573 	}
1574 
1575 	if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1576 	    (vid != old_vid) &&
1577 	    !test_bit(old_vid, adapter->active_vlans)) {
1578 		/* remove VID from filter table */
1579 		igb_vfta_set(hw, vid, pf_id, false, true);
1580 	}
1581 }
1582 
1583 /**
1584  *  igb_release_hw_control - release control of the h/w to f/w
1585  *  @adapter: address of board private structure
1586  *
1587  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1588  *  For ASF and Pass Through versions of f/w this means that the
1589  *  driver is no longer loaded.
1590  **/
1591 static void igb_release_hw_control(struct igb_adapter *adapter)
1592 {
1593 	struct e1000_hw *hw = &adapter->hw;
1594 	u32 ctrl_ext;
1595 
1596 	/* Let firmware take over control of h/w */
1597 	ctrl_ext = rd32(E1000_CTRL_EXT);
1598 	wr32(E1000_CTRL_EXT,
1599 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1600 }
1601 
1602 /**
1603  *  igb_get_hw_control - get control of the h/w from f/w
1604  *  @adapter: address of board private structure
1605  *
1606  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1607  *  For ASF and Pass Through versions of f/w this means that
1608  *  the driver is loaded.
1609  **/
1610 static void igb_get_hw_control(struct igb_adapter *adapter)
1611 {
1612 	struct e1000_hw *hw = &adapter->hw;
1613 	u32 ctrl_ext;
1614 
1615 	/* Let firmware know the driver has taken over */
1616 	ctrl_ext = rd32(E1000_CTRL_EXT);
1617 	wr32(E1000_CTRL_EXT,
1618 			ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1619 }
1620 
1621 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1622 {
1623 	struct net_device *netdev = adapter->netdev;
1624 	struct e1000_hw *hw = &adapter->hw;
1625 
1626 	WARN_ON(hw->mac.type != e1000_i210);
1627 
1628 	if (enable)
1629 		adapter->flags |= IGB_FLAG_FQTSS;
1630 	else
1631 		adapter->flags &= ~IGB_FLAG_FQTSS;
1632 
1633 	if (netif_running(netdev))
1634 		schedule_work(&adapter->reset_task);
1635 }
1636 
1637 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1638 {
1639 	return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1640 }
1641 
1642 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1643 				   enum tx_queue_prio prio)
1644 {
1645 	u32 val;
1646 
1647 	WARN_ON(hw->mac.type != e1000_i210);
1648 	WARN_ON(queue < 0 || queue > 4);
1649 
1650 	val = rd32(E1000_I210_TXDCTL(queue));
1651 
1652 	if (prio == TX_QUEUE_PRIO_HIGH)
1653 		val |= E1000_TXDCTL_PRIORITY;
1654 	else
1655 		val &= ~E1000_TXDCTL_PRIORITY;
1656 
1657 	wr32(E1000_I210_TXDCTL(queue), val);
1658 }
1659 
1660 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1661 {
1662 	u32 val;
1663 
1664 	WARN_ON(hw->mac.type != e1000_i210);
1665 	WARN_ON(queue < 0 || queue > 1);
1666 
1667 	val = rd32(E1000_I210_TQAVCC(queue));
1668 
1669 	if (mode == QUEUE_MODE_STREAM_RESERVATION)
1670 		val |= E1000_TQAVCC_QUEUEMODE;
1671 	else
1672 		val &= ~E1000_TQAVCC_QUEUEMODE;
1673 
1674 	wr32(E1000_I210_TQAVCC(queue), val);
1675 }
1676 
1677 /**
1678  *  igb_configure_cbs - Configure Credit-Based Shaper (CBS)
1679  *  @adapter: pointer to adapter struct
1680  *  @queue: queue number
1681  *  @enable: true = enable CBS, false = disable CBS
1682  *  @idleslope: idleSlope in kbps
1683  *  @sendslope: sendSlope in kbps
1684  *  @hicredit: hiCredit in bytes
1685  *  @locredit: loCredit in bytes
1686  *
1687  *  Configure CBS for a given hardware queue. When disabling, idleslope,
1688  *  sendslope, hicredit, locredit arguments are ignored. Returns 0 if
1689  *  success. Negative otherwise.
1690  **/
1691 static void igb_configure_cbs(struct igb_adapter *adapter, int queue,
1692 			      bool enable, int idleslope, int sendslope,
1693 			      int hicredit, int locredit)
1694 {
1695 	struct net_device *netdev = adapter->netdev;
1696 	struct e1000_hw *hw = &adapter->hw;
1697 	u32 tqavcc;
1698 	u16 value;
1699 
1700 	WARN_ON(hw->mac.type != e1000_i210);
1701 	WARN_ON(queue < 0 || queue > 1);
1702 
1703 	if (enable || queue == 0) {
1704 		/* i210 does not allow the queue 0 to be in the Strict
1705 		 * Priority mode while the Qav mode is enabled, so,
1706 		 * instead of disabling strict priority mode, we give
1707 		 * queue 0 the maximum of credits possible.
1708 		 *
1709 		 * See section 8.12.19 of the i210 datasheet, "Note:
1710 		 * Queue0 QueueMode must be set to 1b when
1711 		 * TransmitMode is set to Qav."
1712 		 */
1713 		if (queue == 0 && !enable) {
1714 			/* max "linkspeed" idleslope in kbps */
1715 			idleslope = 1000000;
1716 			hicredit = ETH_FRAME_LEN;
1717 		}
1718 
1719 		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1720 		set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1721 
1722 		/* According to i210 datasheet section 7.2.7.7, we should set
1723 		 * the 'idleSlope' field from TQAVCC register following the
1724 		 * equation:
1725 		 *
1726 		 * For 100 Mbps link speed:
1727 		 *
1728 		 *     value = BW * 0x7735 * 0.2                          (E1)
1729 		 *
1730 		 * For 1000Mbps link speed:
1731 		 *
1732 		 *     value = BW * 0x7735 * 2                            (E2)
1733 		 *
1734 		 * E1 and E2 can be merged into one equation as shown below.
1735 		 * Note that 'link-speed' is in Mbps.
1736 		 *
1737 		 *     value = BW * 0x7735 * 2 * link-speed
1738 		 *                           --------------               (E3)
1739 		 *                                1000
1740 		 *
1741 		 * 'BW' is the percentage bandwidth out of full link speed
1742 		 * which can be found with the following equation. Note that
1743 		 * idleSlope here is the parameter from this function which
1744 		 * is in kbps.
1745 		 *
1746 		 *     BW =     idleSlope
1747 		 *          -----------------                             (E4)
1748 		 *          link-speed * 1000
1749 		 *
1750 		 * That said, we can come up with a generic equation to
1751 		 * calculate the value we should set it TQAVCC register by
1752 		 * replacing 'BW' in E3 by E4. The resulting equation is:
1753 		 *
1754 		 * value =     idleSlope     * 0x7735 * 2 * link-speed
1755 		 *         -----------------            --------------    (E5)
1756 		 *         link-speed * 1000                 1000
1757 		 *
1758 		 * 'link-speed' is present in both sides of the fraction so
1759 		 * it is canceled out. The final equation is the following:
1760 		 *
1761 		 *     value = idleSlope * 61034
1762 		 *             -----------------                          (E6)
1763 		 *                  1000000
1764 		 *
1765 		 * NOTE: For i210, given the above, we can see that idleslope
1766 		 *       is represented in 16.38431 kbps units by the value at
1767 		 *       the TQAVCC register (1Gbps / 61034), which reduces
1768 		 *       the granularity for idleslope increments.
1769 		 *       For instance, if you want to configure a 2576kbps
1770 		 *       idleslope, the value to be written on the register
1771 		 *       would have to be 157.23. If rounded down, you end
1772 		 *       up with less bandwidth available than originally
1773 		 *       required (~2572 kbps). If rounded up, you end up
1774 		 *       with a higher bandwidth (~2589 kbps). Below the
1775 		 *       approach we take is to always round up the
1776 		 *       calculated value, so the resulting bandwidth might
1777 		 *       be slightly higher for some configurations.
1778 		 */
1779 		value = DIV_ROUND_UP_ULL(idleslope * 61034ULL, 1000000);
1780 
1781 		tqavcc = rd32(E1000_I210_TQAVCC(queue));
1782 		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1783 		tqavcc |= value;
1784 		wr32(E1000_I210_TQAVCC(queue), tqavcc);
1785 
1786 		wr32(E1000_I210_TQAVHC(queue), 0x80000000 + hicredit * 0x7735);
1787 	} else {
1788 		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1789 		set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1790 
1791 		/* Set idleSlope to zero. */
1792 		tqavcc = rd32(E1000_I210_TQAVCC(queue));
1793 		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1794 		wr32(E1000_I210_TQAVCC(queue), tqavcc);
1795 
1796 		/* Set hiCredit to zero. */
1797 		wr32(E1000_I210_TQAVHC(queue), 0);
1798 	}
1799 
1800 	/* XXX: In i210 controller the sendSlope and loCredit parameters from
1801 	 * CBS are not configurable by software so we don't do any 'controller
1802 	 * configuration' in respect to these parameters.
1803 	 */
1804 
1805 	netdev_dbg(netdev, "CBS %s: queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1806 		   (enable) ? "enabled" : "disabled", queue,
1807 		   idleslope, sendslope, hicredit, locredit);
1808 }
1809 
1810 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1811 			       bool enable, int idleslope, int sendslope,
1812 			       int hicredit, int locredit)
1813 {
1814 	struct igb_ring *ring;
1815 
1816 	if (queue < 0 || queue > adapter->num_tx_queues)
1817 		return -EINVAL;
1818 
1819 	ring = adapter->tx_ring[queue];
1820 
1821 	ring->cbs_enable = enable;
1822 	ring->idleslope = idleslope;
1823 	ring->sendslope = sendslope;
1824 	ring->hicredit = hicredit;
1825 	ring->locredit = locredit;
1826 
1827 	return 0;
1828 }
1829 
1830 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1831 {
1832 	struct igb_ring *ring;
1833 	int i;
1834 
1835 	for (i = 0; i < adapter->num_tx_queues; i++) {
1836 		ring = adapter->tx_ring[i];
1837 
1838 		if (ring->cbs_enable)
1839 			return true;
1840 	}
1841 
1842 	return false;
1843 }
1844 
1845 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1846 {
1847 	struct net_device *netdev = adapter->netdev;
1848 	struct e1000_hw *hw = &adapter->hw;
1849 	u32 val;
1850 
1851 	/* Only i210 controller supports changing the transmission mode. */
1852 	if (hw->mac.type != e1000_i210)
1853 		return;
1854 
1855 	if (is_fqtss_enabled(adapter)) {
1856 		int i, max_queue;
1857 
1858 		/* Configure TQAVCTRL register: set transmit mode to 'Qav',
1859 		 * set data fetch arbitration to 'round robin' and set data
1860 		 * transfer arbitration to 'credit shaper algorithm.
1861 		 */
1862 		val = rd32(E1000_I210_TQAVCTRL);
1863 		val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_DATATRANARB;
1864 		val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1865 		wr32(E1000_I210_TQAVCTRL, val);
1866 
1867 		/* Configure Tx and Rx packet buffers sizes as described in
1868 		 * i210 datasheet section 7.2.7.7.
1869 		 */
1870 		val = rd32(E1000_TXPBS);
1871 		val &= ~I210_TXPBSIZE_MASK;
1872 		val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB |
1873 			I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB;
1874 		wr32(E1000_TXPBS, val);
1875 
1876 		val = rd32(E1000_RXPBS);
1877 		val &= ~I210_RXPBSIZE_MASK;
1878 		val |= I210_RXPBSIZE_PB_32KB;
1879 		wr32(E1000_RXPBS, val);
1880 
1881 		/* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1882 		 * register should not exceed the buffer size programmed in
1883 		 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1884 		 * so according to the datasheet we should set MAX_TPKT_SIZE to
1885 		 * 4kB / 64.
1886 		 *
1887 		 * However, when we do so, no frame from queue 2 and 3 are
1888 		 * transmitted.  It seems the MAX_TPKT_SIZE should not be great
1889 		 * or _equal_ to the buffer size programmed in TXPBS. For this
1890 		 * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1891 		 */
1892 		val = (4096 - 1) / 64;
1893 		wr32(E1000_I210_DTXMXPKTSZ, val);
1894 
1895 		/* Since FQTSS mode is enabled, apply any CBS configuration
1896 		 * previously set. If no previous CBS configuration has been
1897 		 * done, then the initial configuration is applied, which means
1898 		 * CBS is disabled.
1899 		 */
1900 		max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1901 			    adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1902 
1903 		for (i = 0; i < max_queue; i++) {
1904 			struct igb_ring *ring = adapter->tx_ring[i];
1905 
1906 			igb_configure_cbs(adapter, i, ring->cbs_enable,
1907 					  ring->idleslope, ring->sendslope,
1908 					  ring->hicredit, ring->locredit);
1909 		}
1910 	} else {
1911 		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1912 		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1913 		wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1914 
1915 		val = rd32(E1000_I210_TQAVCTRL);
1916 		/* According to Section 8.12.21, the other flags we've set when
1917 		 * enabling FQTSS are not relevant when disabling FQTSS so we
1918 		 * don't set they here.
1919 		 */
1920 		val &= ~E1000_TQAVCTRL_XMIT_MODE;
1921 		wr32(E1000_I210_TQAVCTRL, val);
1922 	}
1923 
1924 	netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1925 		   "enabled" : "disabled");
1926 }
1927 
1928 /**
1929  *  igb_configure - configure the hardware for RX and TX
1930  *  @adapter: private board structure
1931  **/
1932 static void igb_configure(struct igb_adapter *adapter)
1933 {
1934 	struct net_device *netdev = adapter->netdev;
1935 	int i;
1936 
1937 	igb_get_hw_control(adapter);
1938 	igb_set_rx_mode(netdev);
1939 	igb_setup_tx_mode(adapter);
1940 
1941 	igb_restore_vlan(adapter);
1942 
1943 	igb_setup_tctl(adapter);
1944 	igb_setup_mrqc(adapter);
1945 	igb_setup_rctl(adapter);
1946 
1947 	igb_nfc_filter_restore(adapter);
1948 	igb_configure_tx(adapter);
1949 	igb_configure_rx(adapter);
1950 
1951 	igb_rx_fifo_flush_82575(&adapter->hw);
1952 
1953 	/* call igb_desc_unused which always leaves
1954 	 * at least 1 descriptor unused to make sure
1955 	 * next_to_use != next_to_clean
1956 	 */
1957 	for (i = 0; i < adapter->num_rx_queues; i++) {
1958 		struct igb_ring *ring = adapter->rx_ring[i];
1959 		igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1960 	}
1961 }
1962 
1963 /**
1964  *  igb_power_up_link - Power up the phy/serdes link
1965  *  @adapter: address of board private structure
1966  **/
1967 void igb_power_up_link(struct igb_adapter *adapter)
1968 {
1969 	igb_reset_phy(&adapter->hw);
1970 
1971 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1972 		igb_power_up_phy_copper(&adapter->hw);
1973 	else
1974 		igb_power_up_serdes_link_82575(&adapter->hw);
1975 
1976 	igb_setup_link(&adapter->hw);
1977 }
1978 
1979 /**
1980  *  igb_power_down_link - Power down the phy/serdes link
1981  *  @adapter: address of board private structure
1982  */
1983 static void igb_power_down_link(struct igb_adapter *adapter)
1984 {
1985 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1986 		igb_power_down_phy_copper_82575(&adapter->hw);
1987 	else
1988 		igb_shutdown_serdes_link_82575(&adapter->hw);
1989 }
1990 
1991 /**
1992  * Detect and switch function for Media Auto Sense
1993  * @adapter: address of the board private structure
1994  **/
1995 static void igb_check_swap_media(struct igb_adapter *adapter)
1996 {
1997 	struct e1000_hw *hw = &adapter->hw;
1998 	u32 ctrl_ext, connsw;
1999 	bool swap_now = false;
2000 
2001 	ctrl_ext = rd32(E1000_CTRL_EXT);
2002 	connsw = rd32(E1000_CONNSW);
2003 
2004 	/* need to live swap if current media is copper and we have fiber/serdes
2005 	 * to go to.
2006 	 */
2007 
2008 	if ((hw->phy.media_type == e1000_media_type_copper) &&
2009 	    (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2010 		swap_now = true;
2011 	} else if (!(connsw & E1000_CONNSW_SERDESD)) {
2012 		/* copper signal takes time to appear */
2013 		if (adapter->copper_tries < 4) {
2014 			adapter->copper_tries++;
2015 			connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2016 			wr32(E1000_CONNSW, connsw);
2017 			return;
2018 		} else {
2019 			adapter->copper_tries = 0;
2020 			if ((connsw & E1000_CONNSW_PHYSD) &&
2021 			    (!(connsw & E1000_CONNSW_PHY_PDN))) {
2022 				swap_now = true;
2023 				connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2024 				wr32(E1000_CONNSW, connsw);
2025 			}
2026 		}
2027 	}
2028 
2029 	if (!swap_now)
2030 		return;
2031 
2032 	switch (hw->phy.media_type) {
2033 	case e1000_media_type_copper:
2034 		netdev_info(adapter->netdev,
2035 			"MAS: changing media to fiber/serdes\n");
2036 		ctrl_ext |=
2037 			E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2038 		adapter->flags |= IGB_FLAG_MEDIA_RESET;
2039 		adapter->copper_tries = 0;
2040 		break;
2041 	case e1000_media_type_internal_serdes:
2042 	case e1000_media_type_fiber:
2043 		netdev_info(adapter->netdev,
2044 			"MAS: changing media to copper\n");
2045 		ctrl_ext &=
2046 			~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2047 		adapter->flags |= IGB_FLAG_MEDIA_RESET;
2048 		break;
2049 	default:
2050 		/* shouldn't get here during regular operation */
2051 		netdev_err(adapter->netdev,
2052 			"AMS: Invalid media type found, returning\n");
2053 		break;
2054 	}
2055 	wr32(E1000_CTRL_EXT, ctrl_ext);
2056 }
2057 
2058 /**
2059  *  igb_up - Open the interface and prepare it to handle traffic
2060  *  @adapter: board private structure
2061  **/
2062 int igb_up(struct igb_adapter *adapter)
2063 {
2064 	struct e1000_hw *hw = &adapter->hw;
2065 	int i;
2066 
2067 	/* hardware has been reset, we need to reload some things */
2068 	igb_configure(adapter);
2069 
2070 	clear_bit(__IGB_DOWN, &adapter->state);
2071 
2072 	for (i = 0; i < adapter->num_q_vectors; i++)
2073 		napi_enable(&(adapter->q_vector[i]->napi));
2074 
2075 	if (adapter->flags & IGB_FLAG_HAS_MSIX)
2076 		igb_configure_msix(adapter);
2077 	else
2078 		igb_assign_vector(adapter->q_vector[0], 0);
2079 
2080 	/* Clear any pending interrupts. */
2081 	rd32(E1000_ICR);
2082 	igb_irq_enable(adapter);
2083 
2084 	/* notify VFs that reset has been completed */
2085 	if (adapter->vfs_allocated_count) {
2086 		u32 reg_data = rd32(E1000_CTRL_EXT);
2087 
2088 		reg_data |= E1000_CTRL_EXT_PFRSTD;
2089 		wr32(E1000_CTRL_EXT, reg_data);
2090 	}
2091 
2092 	netif_tx_start_all_queues(adapter->netdev);
2093 
2094 	/* start the watchdog. */
2095 	hw->mac.get_link_status = 1;
2096 	schedule_work(&adapter->watchdog_task);
2097 
2098 	if ((adapter->flags & IGB_FLAG_EEE) &&
2099 	    (!hw->dev_spec._82575.eee_disable))
2100 		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2101 
2102 	return 0;
2103 }
2104 
2105 void igb_down(struct igb_adapter *adapter)
2106 {
2107 	struct net_device *netdev = adapter->netdev;
2108 	struct e1000_hw *hw = &adapter->hw;
2109 	u32 tctl, rctl;
2110 	int i;
2111 
2112 	/* signal that we're down so the interrupt handler does not
2113 	 * reschedule our watchdog timer
2114 	 */
2115 	set_bit(__IGB_DOWN, &adapter->state);
2116 
2117 	/* disable receives in the hardware */
2118 	rctl = rd32(E1000_RCTL);
2119 	wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2120 	/* flush and sleep below */
2121 
2122 	igb_nfc_filter_exit(adapter);
2123 
2124 	netif_carrier_off(netdev);
2125 	netif_tx_stop_all_queues(netdev);
2126 
2127 	/* disable transmits in the hardware */
2128 	tctl = rd32(E1000_TCTL);
2129 	tctl &= ~E1000_TCTL_EN;
2130 	wr32(E1000_TCTL, tctl);
2131 	/* flush both disables and wait for them to finish */
2132 	wrfl();
2133 	usleep_range(10000, 11000);
2134 
2135 	igb_irq_disable(adapter);
2136 
2137 	adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2138 
2139 	for (i = 0; i < adapter->num_q_vectors; i++) {
2140 		if (adapter->q_vector[i]) {
2141 			napi_synchronize(&adapter->q_vector[i]->napi);
2142 			napi_disable(&adapter->q_vector[i]->napi);
2143 		}
2144 	}
2145 
2146 	del_timer_sync(&adapter->watchdog_timer);
2147 	del_timer_sync(&adapter->phy_info_timer);
2148 
2149 	/* record the stats before reset*/
2150 	spin_lock(&adapter->stats64_lock);
2151 	igb_update_stats(adapter);
2152 	spin_unlock(&adapter->stats64_lock);
2153 
2154 	adapter->link_speed = 0;
2155 	adapter->link_duplex = 0;
2156 
2157 	if (!pci_channel_offline(adapter->pdev))
2158 		igb_reset(adapter);
2159 
2160 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
2161 	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2162 
2163 	igb_clean_all_tx_rings(adapter);
2164 	igb_clean_all_rx_rings(adapter);
2165 #ifdef CONFIG_IGB_DCA
2166 
2167 	/* since we reset the hardware DCA settings were cleared */
2168 	igb_setup_dca(adapter);
2169 #endif
2170 }
2171 
2172 void igb_reinit_locked(struct igb_adapter *adapter)
2173 {
2174 	WARN_ON(in_interrupt());
2175 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2176 		usleep_range(1000, 2000);
2177 	igb_down(adapter);
2178 	igb_up(adapter);
2179 	clear_bit(__IGB_RESETTING, &adapter->state);
2180 }
2181 
2182 /** igb_enable_mas - Media Autosense re-enable after swap
2183  *
2184  * @adapter: adapter struct
2185  **/
2186 static void igb_enable_mas(struct igb_adapter *adapter)
2187 {
2188 	struct e1000_hw *hw = &adapter->hw;
2189 	u32 connsw = rd32(E1000_CONNSW);
2190 
2191 	/* configure for SerDes media detect */
2192 	if ((hw->phy.media_type == e1000_media_type_copper) &&
2193 	    (!(connsw & E1000_CONNSW_SERDESD))) {
2194 		connsw |= E1000_CONNSW_ENRGSRC;
2195 		connsw |= E1000_CONNSW_AUTOSENSE_EN;
2196 		wr32(E1000_CONNSW, connsw);
2197 		wrfl();
2198 	}
2199 }
2200 
2201 void igb_reset(struct igb_adapter *adapter)
2202 {
2203 	struct pci_dev *pdev = adapter->pdev;
2204 	struct e1000_hw *hw = &adapter->hw;
2205 	struct e1000_mac_info *mac = &hw->mac;
2206 	struct e1000_fc_info *fc = &hw->fc;
2207 	u32 pba, hwm;
2208 
2209 	/* Repartition Pba for greater than 9k mtu
2210 	 * To take effect CTRL.RST is required.
2211 	 */
2212 	switch (mac->type) {
2213 	case e1000_i350:
2214 	case e1000_i354:
2215 	case e1000_82580:
2216 		pba = rd32(E1000_RXPBS);
2217 		pba = igb_rxpbs_adjust_82580(pba);
2218 		break;
2219 	case e1000_82576:
2220 		pba = rd32(E1000_RXPBS);
2221 		pba &= E1000_RXPBS_SIZE_MASK_82576;
2222 		break;
2223 	case e1000_82575:
2224 	case e1000_i210:
2225 	case e1000_i211:
2226 	default:
2227 		pba = E1000_PBA_34K;
2228 		break;
2229 	}
2230 
2231 	if (mac->type == e1000_82575) {
2232 		u32 min_rx_space, min_tx_space, needed_tx_space;
2233 
2234 		/* write Rx PBA so that hardware can report correct Tx PBA */
2235 		wr32(E1000_PBA, pba);
2236 
2237 		/* To maintain wire speed transmits, the Tx FIFO should be
2238 		 * large enough to accommodate two full transmit packets,
2239 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
2240 		 * the Rx FIFO should be large enough to accommodate at least
2241 		 * one full receive packet and is similarly rounded up and
2242 		 * expressed in KB.
2243 		 */
2244 		min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2245 
2246 		/* The Tx FIFO also stores 16 bytes of information about the Tx
2247 		 * but don't include Ethernet FCS because hardware appends it.
2248 		 * We only need to round down to the nearest 512 byte block
2249 		 * count since the value we care about is 2 frames, not 1.
2250 		 */
2251 		min_tx_space = adapter->max_frame_size;
2252 		min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2253 		min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2254 
2255 		/* upper 16 bits has Tx packet buffer allocation size in KB */
2256 		needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2257 
2258 		/* If current Tx allocation is less than the min Tx FIFO size,
2259 		 * and the min Tx FIFO size is less than the current Rx FIFO
2260 		 * allocation, take space away from current Rx allocation.
2261 		 */
2262 		if (needed_tx_space < pba) {
2263 			pba -= needed_tx_space;
2264 
2265 			/* if short on Rx space, Rx wins and must trump Tx
2266 			 * adjustment
2267 			 */
2268 			if (pba < min_rx_space)
2269 				pba = min_rx_space;
2270 		}
2271 
2272 		/* adjust PBA for jumbo frames */
2273 		wr32(E1000_PBA, pba);
2274 	}
2275 
2276 	/* flow control settings
2277 	 * The high water mark must be low enough to fit one full frame
2278 	 * after transmitting the pause frame.  As such we must have enough
2279 	 * space to allow for us to complete our current transmit and then
2280 	 * receive the frame that is in progress from the link partner.
2281 	 * Set it to:
2282 	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2283 	 */
2284 	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2285 
2286 	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
2287 	fc->low_water = fc->high_water - 16;
2288 	fc->pause_time = 0xFFFF;
2289 	fc->send_xon = 1;
2290 	fc->current_mode = fc->requested_mode;
2291 
2292 	/* disable receive for all VFs and wait one second */
2293 	if (adapter->vfs_allocated_count) {
2294 		int i;
2295 
2296 		for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2297 			adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2298 
2299 		/* ping all the active vfs to let them know we are going down */
2300 		igb_ping_all_vfs(adapter);
2301 
2302 		/* disable transmits and receives */
2303 		wr32(E1000_VFRE, 0);
2304 		wr32(E1000_VFTE, 0);
2305 	}
2306 
2307 	/* Allow time for pending master requests to run */
2308 	hw->mac.ops.reset_hw(hw);
2309 	wr32(E1000_WUC, 0);
2310 
2311 	if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2312 		/* need to resetup here after media swap */
2313 		adapter->ei.get_invariants(hw);
2314 		adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2315 	}
2316 	if ((mac->type == e1000_82575) &&
2317 	    (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2318 		igb_enable_mas(adapter);
2319 	}
2320 	if (hw->mac.ops.init_hw(hw))
2321 		dev_err(&pdev->dev, "Hardware Error\n");
2322 
2323 	/* RAR registers were cleared during init_hw, clear mac table */
2324 	igb_flush_mac_table(adapter);
2325 	__dev_uc_unsync(adapter->netdev, NULL);
2326 
2327 	/* Recover default RAR entry */
2328 	igb_set_default_mac_filter(adapter);
2329 
2330 	/* Flow control settings reset on hardware reset, so guarantee flow
2331 	 * control is off when forcing speed.
2332 	 */
2333 	if (!hw->mac.autoneg)
2334 		igb_force_mac_fc(hw);
2335 
2336 	igb_init_dmac(adapter, pba);
2337 #ifdef CONFIG_IGB_HWMON
2338 	/* Re-initialize the thermal sensor on i350 devices. */
2339 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
2340 		if (mac->type == e1000_i350 && hw->bus.func == 0) {
2341 			/* If present, re-initialize the external thermal sensor
2342 			 * interface.
2343 			 */
2344 			if (adapter->ets)
2345 				mac->ops.init_thermal_sensor_thresh(hw);
2346 		}
2347 	}
2348 #endif
2349 	/* Re-establish EEE setting */
2350 	if (hw->phy.media_type == e1000_media_type_copper) {
2351 		switch (mac->type) {
2352 		case e1000_i350:
2353 		case e1000_i210:
2354 		case e1000_i211:
2355 			igb_set_eee_i350(hw, true, true);
2356 			break;
2357 		case e1000_i354:
2358 			igb_set_eee_i354(hw, true, true);
2359 			break;
2360 		default:
2361 			break;
2362 		}
2363 	}
2364 	if (!netif_running(adapter->netdev))
2365 		igb_power_down_link(adapter);
2366 
2367 	igb_update_mng_vlan(adapter);
2368 
2369 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2370 	wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2371 
2372 	/* Re-enable PTP, where applicable. */
2373 	if (adapter->ptp_flags & IGB_PTP_ENABLED)
2374 		igb_ptp_reset(adapter);
2375 
2376 	igb_get_phy_info(hw);
2377 }
2378 
2379 static netdev_features_t igb_fix_features(struct net_device *netdev,
2380 	netdev_features_t features)
2381 {
2382 	/* Since there is no support for separate Rx/Tx vlan accel
2383 	 * enable/disable make sure Tx flag is always in same state as Rx.
2384 	 */
2385 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2386 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2387 	else
2388 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2389 
2390 	return features;
2391 }
2392 
2393 static int igb_set_features(struct net_device *netdev,
2394 	netdev_features_t features)
2395 {
2396 	netdev_features_t changed = netdev->features ^ features;
2397 	struct igb_adapter *adapter = netdev_priv(netdev);
2398 
2399 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2400 		igb_vlan_mode(netdev, features);
2401 
2402 	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2403 		return 0;
2404 
2405 	if (!(features & NETIF_F_NTUPLE)) {
2406 		struct hlist_node *node2;
2407 		struct igb_nfc_filter *rule;
2408 
2409 		spin_lock(&adapter->nfc_lock);
2410 		hlist_for_each_entry_safe(rule, node2,
2411 					  &adapter->nfc_filter_list, nfc_node) {
2412 			igb_erase_filter(adapter, rule);
2413 			hlist_del(&rule->nfc_node);
2414 			kfree(rule);
2415 		}
2416 		spin_unlock(&adapter->nfc_lock);
2417 		adapter->nfc_filter_count = 0;
2418 	}
2419 
2420 	netdev->features = features;
2421 
2422 	if (netif_running(netdev))
2423 		igb_reinit_locked(adapter);
2424 	else
2425 		igb_reset(adapter);
2426 
2427 	return 0;
2428 }
2429 
2430 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2431 			   struct net_device *dev,
2432 			   const unsigned char *addr, u16 vid,
2433 			   u16 flags)
2434 {
2435 	/* guarantee we can provide a unique filter for the unicast address */
2436 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2437 		struct igb_adapter *adapter = netdev_priv(dev);
2438 		int vfn = adapter->vfs_allocated_count;
2439 
2440 		if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2441 			return -ENOMEM;
2442 	}
2443 
2444 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2445 }
2446 
2447 #define IGB_MAX_MAC_HDR_LEN	127
2448 #define IGB_MAX_NETWORK_HDR_LEN	511
2449 
2450 static netdev_features_t
2451 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2452 		   netdev_features_t features)
2453 {
2454 	unsigned int network_hdr_len, mac_hdr_len;
2455 
2456 	/* Make certain the headers can be described by a context descriptor */
2457 	mac_hdr_len = skb_network_header(skb) - skb->data;
2458 	if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2459 		return features & ~(NETIF_F_HW_CSUM |
2460 				    NETIF_F_SCTP_CRC |
2461 				    NETIF_F_HW_VLAN_CTAG_TX |
2462 				    NETIF_F_TSO |
2463 				    NETIF_F_TSO6);
2464 
2465 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2466 	if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2467 		return features & ~(NETIF_F_HW_CSUM |
2468 				    NETIF_F_SCTP_CRC |
2469 				    NETIF_F_TSO |
2470 				    NETIF_F_TSO6);
2471 
2472 	/* We can only support IPV4 TSO in tunnels if we can mangle the
2473 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2474 	 */
2475 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2476 		features &= ~NETIF_F_TSO;
2477 
2478 	return features;
2479 }
2480 
2481 static int igb_offload_cbs(struct igb_adapter *adapter,
2482 			   struct tc_cbs_qopt_offload *qopt)
2483 {
2484 	struct e1000_hw *hw = &adapter->hw;
2485 	int err;
2486 
2487 	/* CBS offloading is only supported by i210 controller. */
2488 	if (hw->mac.type != e1000_i210)
2489 		return -EOPNOTSUPP;
2490 
2491 	/* CBS offloading is only supported by queue 0 and queue 1. */
2492 	if (qopt->queue < 0 || qopt->queue > 1)
2493 		return -EINVAL;
2494 
2495 	err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2496 				  qopt->idleslope, qopt->sendslope,
2497 				  qopt->hicredit, qopt->locredit);
2498 	if (err)
2499 		return err;
2500 
2501 	if (is_fqtss_enabled(adapter)) {
2502 		igb_configure_cbs(adapter, qopt->queue, qopt->enable,
2503 				  qopt->idleslope, qopt->sendslope,
2504 				  qopt->hicredit, qopt->locredit);
2505 
2506 		if (!is_any_cbs_enabled(adapter))
2507 			enable_fqtss(adapter, false);
2508 
2509 	} else {
2510 		enable_fqtss(adapter, true);
2511 	}
2512 
2513 	return 0;
2514 }
2515 
2516 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2517 			void *type_data)
2518 {
2519 	struct igb_adapter *adapter = netdev_priv(dev);
2520 
2521 	switch (type) {
2522 	case TC_SETUP_QDISC_CBS:
2523 		return igb_offload_cbs(adapter, type_data);
2524 
2525 	default:
2526 		return -EOPNOTSUPP;
2527 	}
2528 }
2529 
2530 static const struct net_device_ops igb_netdev_ops = {
2531 	.ndo_open		= igb_open,
2532 	.ndo_stop		= igb_close,
2533 	.ndo_start_xmit		= igb_xmit_frame,
2534 	.ndo_get_stats64	= igb_get_stats64,
2535 	.ndo_set_rx_mode	= igb_set_rx_mode,
2536 	.ndo_set_mac_address	= igb_set_mac,
2537 	.ndo_change_mtu		= igb_change_mtu,
2538 	.ndo_do_ioctl		= igb_ioctl,
2539 	.ndo_tx_timeout		= igb_tx_timeout,
2540 	.ndo_validate_addr	= eth_validate_addr,
2541 	.ndo_vlan_rx_add_vid	= igb_vlan_rx_add_vid,
2542 	.ndo_vlan_rx_kill_vid	= igb_vlan_rx_kill_vid,
2543 	.ndo_set_vf_mac		= igb_ndo_set_vf_mac,
2544 	.ndo_set_vf_vlan	= igb_ndo_set_vf_vlan,
2545 	.ndo_set_vf_rate	= igb_ndo_set_vf_bw,
2546 	.ndo_set_vf_spoofchk	= igb_ndo_set_vf_spoofchk,
2547 	.ndo_set_vf_trust	= igb_ndo_set_vf_trust,
2548 	.ndo_get_vf_config	= igb_ndo_get_vf_config,
2549 #ifdef CONFIG_NET_POLL_CONTROLLER
2550 	.ndo_poll_controller	= igb_netpoll,
2551 #endif
2552 	.ndo_fix_features	= igb_fix_features,
2553 	.ndo_set_features	= igb_set_features,
2554 	.ndo_fdb_add		= igb_ndo_fdb_add,
2555 	.ndo_features_check	= igb_features_check,
2556 	.ndo_setup_tc		= igb_setup_tc,
2557 };
2558 
2559 /**
2560  * igb_set_fw_version - Configure version string for ethtool
2561  * @adapter: adapter struct
2562  **/
2563 void igb_set_fw_version(struct igb_adapter *adapter)
2564 {
2565 	struct e1000_hw *hw = &adapter->hw;
2566 	struct e1000_fw_version fw;
2567 
2568 	igb_get_fw_version(hw, &fw);
2569 
2570 	switch (hw->mac.type) {
2571 	case e1000_i210:
2572 	case e1000_i211:
2573 		if (!(igb_get_flash_presence_i210(hw))) {
2574 			snprintf(adapter->fw_version,
2575 				 sizeof(adapter->fw_version),
2576 				 "%2d.%2d-%d",
2577 				 fw.invm_major, fw.invm_minor,
2578 				 fw.invm_img_type);
2579 			break;
2580 		}
2581 		/* fall through */
2582 	default:
2583 		/* if option is rom valid, display its version too */
2584 		if (fw.or_valid) {
2585 			snprintf(adapter->fw_version,
2586 				 sizeof(adapter->fw_version),
2587 				 "%d.%d, 0x%08x, %d.%d.%d",
2588 				 fw.eep_major, fw.eep_minor, fw.etrack_id,
2589 				 fw.or_major, fw.or_build, fw.or_patch);
2590 		/* no option rom */
2591 		} else if (fw.etrack_id != 0X0000) {
2592 			snprintf(adapter->fw_version,
2593 			    sizeof(adapter->fw_version),
2594 			    "%d.%d, 0x%08x",
2595 			    fw.eep_major, fw.eep_minor, fw.etrack_id);
2596 		} else {
2597 		snprintf(adapter->fw_version,
2598 		    sizeof(adapter->fw_version),
2599 		    "%d.%d.%d",
2600 		    fw.eep_major, fw.eep_minor, fw.eep_build);
2601 		}
2602 		break;
2603 	}
2604 }
2605 
2606 /**
2607  * igb_init_mas - init Media Autosense feature if enabled in the NVM
2608  *
2609  * @adapter: adapter struct
2610  **/
2611 static void igb_init_mas(struct igb_adapter *adapter)
2612 {
2613 	struct e1000_hw *hw = &adapter->hw;
2614 	u16 eeprom_data;
2615 
2616 	hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2617 	switch (hw->bus.func) {
2618 	case E1000_FUNC_0:
2619 		if (eeprom_data & IGB_MAS_ENABLE_0) {
2620 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2621 			netdev_info(adapter->netdev,
2622 				"MAS: Enabling Media Autosense for port %d\n",
2623 				hw->bus.func);
2624 		}
2625 		break;
2626 	case E1000_FUNC_1:
2627 		if (eeprom_data & IGB_MAS_ENABLE_1) {
2628 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2629 			netdev_info(adapter->netdev,
2630 				"MAS: Enabling Media Autosense for port %d\n",
2631 				hw->bus.func);
2632 		}
2633 		break;
2634 	case E1000_FUNC_2:
2635 		if (eeprom_data & IGB_MAS_ENABLE_2) {
2636 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2637 			netdev_info(adapter->netdev,
2638 				"MAS: Enabling Media Autosense for port %d\n",
2639 				hw->bus.func);
2640 		}
2641 		break;
2642 	case E1000_FUNC_3:
2643 		if (eeprom_data & IGB_MAS_ENABLE_3) {
2644 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2645 			netdev_info(adapter->netdev,
2646 				"MAS: Enabling Media Autosense for port %d\n",
2647 				hw->bus.func);
2648 		}
2649 		break;
2650 	default:
2651 		/* Shouldn't get here */
2652 		netdev_err(adapter->netdev,
2653 			"MAS: Invalid port configuration, returning\n");
2654 		break;
2655 	}
2656 }
2657 
2658 /**
2659  *  igb_init_i2c - Init I2C interface
2660  *  @adapter: pointer to adapter structure
2661  **/
2662 static s32 igb_init_i2c(struct igb_adapter *adapter)
2663 {
2664 	s32 status = 0;
2665 
2666 	/* I2C interface supported on i350 devices */
2667 	if (adapter->hw.mac.type != e1000_i350)
2668 		return 0;
2669 
2670 	/* Initialize the i2c bus which is controlled by the registers.
2671 	 * This bus will use the i2c_algo_bit structue that implements
2672 	 * the protocol through toggling of the 4 bits in the register.
2673 	 */
2674 	adapter->i2c_adap.owner = THIS_MODULE;
2675 	adapter->i2c_algo = igb_i2c_algo;
2676 	adapter->i2c_algo.data = adapter;
2677 	adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2678 	adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2679 	strlcpy(adapter->i2c_adap.name, "igb BB",
2680 		sizeof(adapter->i2c_adap.name));
2681 	status = i2c_bit_add_bus(&adapter->i2c_adap);
2682 	return status;
2683 }
2684 
2685 /**
2686  *  igb_probe - Device Initialization Routine
2687  *  @pdev: PCI device information struct
2688  *  @ent: entry in igb_pci_tbl
2689  *
2690  *  Returns 0 on success, negative on failure
2691  *
2692  *  igb_probe initializes an adapter identified by a pci_dev structure.
2693  *  The OS initialization, configuring of the adapter private structure,
2694  *  and a hardware reset occur.
2695  **/
2696 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2697 {
2698 	struct net_device *netdev;
2699 	struct igb_adapter *adapter;
2700 	struct e1000_hw *hw;
2701 	u16 eeprom_data = 0;
2702 	s32 ret_val;
2703 	static int global_quad_port_a; /* global quad port a indication */
2704 	const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2705 	int err, pci_using_dac;
2706 	u8 part_str[E1000_PBANUM_LENGTH];
2707 
2708 	/* Catch broken hardware that put the wrong VF device ID in
2709 	 * the PCIe SR-IOV capability.
2710 	 */
2711 	if (pdev->is_virtfn) {
2712 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2713 			pci_name(pdev), pdev->vendor, pdev->device);
2714 		return -EINVAL;
2715 	}
2716 
2717 	err = pci_enable_device_mem(pdev);
2718 	if (err)
2719 		return err;
2720 
2721 	pci_using_dac = 0;
2722 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2723 	if (!err) {
2724 		pci_using_dac = 1;
2725 	} else {
2726 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2727 		if (err) {
2728 			dev_err(&pdev->dev,
2729 				"No usable DMA configuration, aborting\n");
2730 			goto err_dma;
2731 		}
2732 	}
2733 
2734 	err = pci_request_mem_regions(pdev, igb_driver_name);
2735 	if (err)
2736 		goto err_pci_reg;
2737 
2738 	pci_enable_pcie_error_reporting(pdev);
2739 
2740 	pci_set_master(pdev);
2741 	pci_save_state(pdev);
2742 
2743 	err = -ENOMEM;
2744 	netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2745 				   IGB_MAX_TX_QUEUES);
2746 	if (!netdev)
2747 		goto err_alloc_etherdev;
2748 
2749 	SET_NETDEV_DEV(netdev, &pdev->dev);
2750 
2751 	pci_set_drvdata(pdev, netdev);
2752 	adapter = netdev_priv(netdev);
2753 	adapter->netdev = netdev;
2754 	adapter->pdev = pdev;
2755 	hw = &adapter->hw;
2756 	hw->back = adapter;
2757 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2758 
2759 	err = -EIO;
2760 	adapter->io_addr = pci_iomap(pdev, 0, 0);
2761 	if (!adapter->io_addr)
2762 		goto err_ioremap;
2763 	/* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
2764 	hw->hw_addr = adapter->io_addr;
2765 
2766 	netdev->netdev_ops = &igb_netdev_ops;
2767 	igb_set_ethtool_ops(netdev);
2768 	netdev->watchdog_timeo = 5 * HZ;
2769 
2770 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2771 
2772 	netdev->mem_start = pci_resource_start(pdev, 0);
2773 	netdev->mem_end = pci_resource_end(pdev, 0);
2774 
2775 	/* PCI config space info */
2776 	hw->vendor_id = pdev->vendor;
2777 	hw->device_id = pdev->device;
2778 	hw->revision_id = pdev->revision;
2779 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2780 	hw->subsystem_device_id = pdev->subsystem_device;
2781 
2782 	/* Copy the default MAC, PHY and NVM function pointers */
2783 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2784 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2785 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2786 	/* Initialize skew-specific constants */
2787 	err = ei->get_invariants(hw);
2788 	if (err)
2789 		goto err_sw_init;
2790 
2791 	/* setup the private structure */
2792 	err = igb_sw_init(adapter);
2793 	if (err)
2794 		goto err_sw_init;
2795 
2796 	igb_get_bus_info_pcie(hw);
2797 
2798 	hw->phy.autoneg_wait_to_complete = false;
2799 
2800 	/* Copper options */
2801 	if (hw->phy.media_type == e1000_media_type_copper) {
2802 		hw->phy.mdix = AUTO_ALL_MODES;
2803 		hw->phy.disable_polarity_correction = false;
2804 		hw->phy.ms_type = e1000_ms_hw_default;
2805 	}
2806 
2807 	if (igb_check_reset_block(hw))
2808 		dev_info(&pdev->dev,
2809 			"PHY reset is blocked due to SOL/IDER session.\n");
2810 
2811 	/* features is initialized to 0 in allocation, it might have bits
2812 	 * set by igb_sw_init so we should use an or instead of an
2813 	 * assignment.
2814 	 */
2815 	netdev->features |= NETIF_F_SG |
2816 			    NETIF_F_TSO |
2817 			    NETIF_F_TSO6 |
2818 			    NETIF_F_RXHASH |
2819 			    NETIF_F_RXCSUM |
2820 			    NETIF_F_HW_CSUM;
2821 
2822 	if (hw->mac.type >= e1000_82576)
2823 		netdev->features |= NETIF_F_SCTP_CRC;
2824 
2825 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2826 				  NETIF_F_GSO_GRE_CSUM | \
2827 				  NETIF_F_GSO_IPXIP4 | \
2828 				  NETIF_F_GSO_IPXIP6 | \
2829 				  NETIF_F_GSO_UDP_TUNNEL | \
2830 				  NETIF_F_GSO_UDP_TUNNEL_CSUM)
2831 
2832 	netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
2833 	netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
2834 
2835 	/* copy netdev features into list of user selectable features */
2836 	netdev->hw_features |= netdev->features |
2837 			       NETIF_F_HW_VLAN_CTAG_RX |
2838 			       NETIF_F_HW_VLAN_CTAG_TX |
2839 			       NETIF_F_RXALL;
2840 
2841 	if (hw->mac.type >= e1000_i350)
2842 		netdev->hw_features |= NETIF_F_NTUPLE;
2843 
2844 	if (pci_using_dac)
2845 		netdev->features |= NETIF_F_HIGHDMA;
2846 
2847 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2848 	netdev->mpls_features |= NETIF_F_HW_CSUM;
2849 	netdev->hw_enc_features |= netdev->vlan_features;
2850 
2851 	/* set this bit last since it cannot be part of vlan_features */
2852 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2853 			    NETIF_F_HW_VLAN_CTAG_RX |
2854 			    NETIF_F_HW_VLAN_CTAG_TX;
2855 
2856 	netdev->priv_flags |= IFF_SUPP_NOFCS;
2857 
2858 	netdev->priv_flags |= IFF_UNICAST_FLT;
2859 
2860 	/* MTU range: 68 - 9216 */
2861 	netdev->min_mtu = ETH_MIN_MTU;
2862 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2863 
2864 	adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2865 
2866 	/* before reading the NVM, reset the controller to put the device in a
2867 	 * known good starting state
2868 	 */
2869 	hw->mac.ops.reset_hw(hw);
2870 
2871 	/* make sure the NVM is good , i211/i210 parts can have special NVM
2872 	 * that doesn't contain a checksum
2873 	 */
2874 	switch (hw->mac.type) {
2875 	case e1000_i210:
2876 	case e1000_i211:
2877 		if (igb_get_flash_presence_i210(hw)) {
2878 			if (hw->nvm.ops.validate(hw) < 0) {
2879 				dev_err(&pdev->dev,
2880 					"The NVM Checksum Is Not Valid\n");
2881 				err = -EIO;
2882 				goto err_eeprom;
2883 			}
2884 		}
2885 		break;
2886 	default:
2887 		if (hw->nvm.ops.validate(hw) < 0) {
2888 			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2889 			err = -EIO;
2890 			goto err_eeprom;
2891 		}
2892 		break;
2893 	}
2894 
2895 	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
2896 		/* copy the MAC address out of the NVM */
2897 		if (hw->mac.ops.read_mac_addr(hw))
2898 			dev_err(&pdev->dev, "NVM Read Error\n");
2899 	}
2900 
2901 	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2902 
2903 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2904 		dev_err(&pdev->dev, "Invalid MAC Address\n");
2905 		err = -EIO;
2906 		goto err_eeprom;
2907 	}
2908 
2909 	igb_set_default_mac_filter(adapter);
2910 
2911 	/* get firmware version for ethtool -i */
2912 	igb_set_fw_version(adapter);
2913 
2914 	/* configure RXPBSIZE and TXPBSIZE */
2915 	if (hw->mac.type == e1000_i210) {
2916 		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
2917 		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
2918 	}
2919 
2920 	timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
2921 	timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
2922 
2923 	INIT_WORK(&adapter->reset_task, igb_reset_task);
2924 	INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2925 
2926 	/* Initialize link properties that are user-changeable */
2927 	adapter->fc_autoneg = true;
2928 	hw->mac.autoneg = true;
2929 	hw->phy.autoneg_advertised = 0x2f;
2930 
2931 	hw->fc.requested_mode = e1000_fc_default;
2932 	hw->fc.current_mode = e1000_fc_default;
2933 
2934 	igb_validate_mdi_setting(hw);
2935 
2936 	/* By default, support wake on port A */
2937 	if (hw->bus.func == 0)
2938 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2939 
2940 	/* Check the NVM for wake support on non-port A ports */
2941 	if (hw->mac.type >= e1000_82580)
2942 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2943 				 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2944 				 &eeprom_data);
2945 	else if (hw->bus.func == 1)
2946 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2947 
2948 	if (eeprom_data & IGB_EEPROM_APME)
2949 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2950 
2951 	/* now that we have the eeprom settings, apply the special cases where
2952 	 * the eeprom may be wrong or the board simply won't support wake on
2953 	 * lan on a particular port
2954 	 */
2955 	switch (pdev->device) {
2956 	case E1000_DEV_ID_82575GB_QUAD_COPPER:
2957 		adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2958 		break;
2959 	case E1000_DEV_ID_82575EB_FIBER_SERDES:
2960 	case E1000_DEV_ID_82576_FIBER:
2961 	case E1000_DEV_ID_82576_SERDES:
2962 		/* Wake events only supported on port A for dual fiber
2963 		 * regardless of eeprom setting
2964 		 */
2965 		if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2966 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2967 		break;
2968 	case E1000_DEV_ID_82576_QUAD_COPPER:
2969 	case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2970 		/* if quad port adapter, disable WoL on all but port A */
2971 		if (global_quad_port_a != 0)
2972 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2973 		else
2974 			adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2975 		/* Reset for multiple quad port adapters */
2976 		if (++global_quad_port_a == 4)
2977 			global_quad_port_a = 0;
2978 		break;
2979 	default:
2980 		/* If the device can't wake, don't set software support */
2981 		if (!device_can_wakeup(&adapter->pdev->dev))
2982 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2983 	}
2984 
2985 	/* initialize the wol settings based on the eeprom settings */
2986 	if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2987 		adapter->wol |= E1000_WUFC_MAG;
2988 
2989 	/* Some vendors want WoL disabled by default, but still supported */
2990 	if ((hw->mac.type == e1000_i350) &&
2991 	    (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2992 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2993 		adapter->wol = 0;
2994 	}
2995 
2996 	/* Some vendors want the ability to Use the EEPROM setting as
2997 	 * enable/disable only, and not for capability
2998 	 */
2999 	if (((hw->mac.type == e1000_i350) ||
3000 	     (hw->mac.type == e1000_i354)) &&
3001 	    (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3002 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3003 		adapter->wol = 0;
3004 	}
3005 	if (hw->mac.type == e1000_i350) {
3006 		if (((pdev->subsystem_device == 0x5001) ||
3007 		     (pdev->subsystem_device == 0x5002)) &&
3008 				(hw->bus.func == 0)) {
3009 			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3010 			adapter->wol = 0;
3011 		}
3012 		if (pdev->subsystem_device == 0x1F52)
3013 			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3014 	}
3015 
3016 	device_set_wakeup_enable(&adapter->pdev->dev,
3017 				 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3018 
3019 	/* reset the hardware with the new settings */
3020 	igb_reset(adapter);
3021 
3022 	/* Init the I2C interface */
3023 	err = igb_init_i2c(adapter);
3024 	if (err) {
3025 		dev_err(&pdev->dev, "failed to init i2c interface\n");
3026 		goto err_eeprom;
3027 	}
3028 
3029 	/* let the f/w know that the h/w is now under the control of the
3030 	 * driver.
3031 	 */
3032 	igb_get_hw_control(adapter);
3033 
3034 	strcpy(netdev->name, "eth%d");
3035 	err = register_netdev(netdev);
3036 	if (err)
3037 		goto err_register;
3038 
3039 	/* carrier off reporting is important to ethtool even BEFORE open */
3040 	netif_carrier_off(netdev);
3041 
3042 #ifdef CONFIG_IGB_DCA
3043 	if (dca_add_requester(&pdev->dev) == 0) {
3044 		adapter->flags |= IGB_FLAG_DCA_ENABLED;
3045 		dev_info(&pdev->dev, "DCA enabled\n");
3046 		igb_setup_dca(adapter);
3047 	}
3048 
3049 #endif
3050 #ifdef CONFIG_IGB_HWMON
3051 	/* Initialize the thermal sensor on i350 devices. */
3052 	if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3053 		u16 ets_word;
3054 
3055 		/* Read the NVM to determine if this i350 device supports an
3056 		 * external thermal sensor.
3057 		 */
3058 		hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3059 		if (ets_word != 0x0000 && ets_word != 0xFFFF)
3060 			adapter->ets = true;
3061 		else
3062 			adapter->ets = false;
3063 		if (igb_sysfs_init(adapter))
3064 			dev_err(&pdev->dev,
3065 				"failed to allocate sysfs resources\n");
3066 	} else {
3067 		adapter->ets = false;
3068 	}
3069 #endif
3070 	/* Check if Media Autosense is enabled */
3071 	adapter->ei = *ei;
3072 	if (hw->dev_spec._82575.mas_capable)
3073 		igb_init_mas(adapter);
3074 
3075 	/* do hw tstamp init after resetting */
3076 	igb_ptp_init(adapter);
3077 
3078 	dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3079 	/* print bus type/speed/width info, not applicable to i354 */
3080 	if (hw->mac.type != e1000_i354) {
3081 		dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3082 			 netdev->name,
3083 			 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3084 			  (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3085 			   "unknown"),
3086 			 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3087 			  "Width x4" :
3088 			  (hw->bus.width == e1000_bus_width_pcie_x2) ?
3089 			  "Width x2" :
3090 			  (hw->bus.width == e1000_bus_width_pcie_x1) ?
3091 			  "Width x1" : "unknown"), netdev->dev_addr);
3092 	}
3093 
3094 	if ((hw->mac.type >= e1000_i210 ||
3095 	     igb_get_flash_presence_i210(hw))) {
3096 		ret_val = igb_read_part_string(hw, part_str,
3097 					       E1000_PBANUM_LENGTH);
3098 	} else {
3099 		ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3100 	}
3101 
3102 	if (ret_val)
3103 		strcpy(part_str, "Unknown");
3104 	dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3105 	dev_info(&pdev->dev,
3106 		"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3107 		(adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3108 		(adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3109 		adapter->num_rx_queues, adapter->num_tx_queues);
3110 	if (hw->phy.media_type == e1000_media_type_copper) {
3111 		switch (hw->mac.type) {
3112 		case e1000_i350:
3113 		case e1000_i210:
3114 		case e1000_i211:
3115 			/* Enable EEE for internal copper PHY devices */
3116 			err = igb_set_eee_i350(hw, true, true);
3117 			if ((!err) &&
3118 			    (!hw->dev_spec._82575.eee_disable)) {
3119 				adapter->eee_advert =
3120 					MDIO_EEE_100TX | MDIO_EEE_1000T;
3121 				adapter->flags |= IGB_FLAG_EEE;
3122 			}
3123 			break;
3124 		case e1000_i354:
3125 			if ((rd32(E1000_CTRL_EXT) &
3126 			    E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3127 				err = igb_set_eee_i354(hw, true, true);
3128 				if ((!err) &&
3129 					(!hw->dev_spec._82575.eee_disable)) {
3130 					adapter->eee_advert =
3131 					   MDIO_EEE_100TX | MDIO_EEE_1000T;
3132 					adapter->flags |= IGB_FLAG_EEE;
3133 				}
3134 			}
3135 			break;
3136 		default:
3137 			break;
3138 		}
3139 	}
3140 	pm_runtime_put_noidle(&pdev->dev);
3141 	return 0;
3142 
3143 err_register:
3144 	igb_release_hw_control(adapter);
3145 	memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3146 err_eeprom:
3147 	if (!igb_check_reset_block(hw))
3148 		igb_reset_phy(hw);
3149 
3150 	if (hw->flash_address)
3151 		iounmap(hw->flash_address);
3152 err_sw_init:
3153 	kfree(adapter->mac_table);
3154 	kfree(adapter->shadow_vfta);
3155 	igb_clear_interrupt_scheme(adapter);
3156 #ifdef CONFIG_PCI_IOV
3157 	igb_disable_sriov(pdev);
3158 #endif
3159 	pci_iounmap(pdev, adapter->io_addr);
3160 err_ioremap:
3161 	free_netdev(netdev);
3162 err_alloc_etherdev:
3163 	pci_release_mem_regions(pdev);
3164 err_pci_reg:
3165 err_dma:
3166 	pci_disable_device(pdev);
3167 	return err;
3168 }
3169 
3170 #ifdef CONFIG_PCI_IOV
3171 static int igb_disable_sriov(struct pci_dev *pdev)
3172 {
3173 	struct net_device *netdev = pci_get_drvdata(pdev);
3174 	struct igb_adapter *adapter = netdev_priv(netdev);
3175 	struct e1000_hw *hw = &adapter->hw;
3176 
3177 	/* reclaim resources allocated to VFs */
3178 	if (adapter->vf_data) {
3179 		/* disable iov and allow time for transactions to clear */
3180 		if (pci_vfs_assigned(pdev)) {
3181 			dev_warn(&pdev->dev,
3182 				 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3183 			return -EPERM;
3184 		} else {
3185 			pci_disable_sriov(pdev);
3186 			msleep(500);
3187 		}
3188 
3189 		kfree(adapter->vf_mac_list);
3190 		adapter->vf_mac_list = NULL;
3191 		kfree(adapter->vf_data);
3192 		adapter->vf_data = NULL;
3193 		adapter->vfs_allocated_count = 0;
3194 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3195 		wrfl();
3196 		msleep(100);
3197 		dev_info(&pdev->dev, "IOV Disabled\n");
3198 
3199 		/* Re-enable DMA Coalescing flag since IOV is turned off */
3200 		adapter->flags |= IGB_FLAG_DMAC;
3201 	}
3202 
3203 	return 0;
3204 }
3205 
3206 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3207 {
3208 	struct net_device *netdev = pci_get_drvdata(pdev);
3209 	struct igb_adapter *adapter = netdev_priv(netdev);
3210 	int old_vfs = pci_num_vf(pdev);
3211 	struct vf_mac_filter *mac_list;
3212 	int err = 0;
3213 	int num_vf_mac_filters, i;
3214 
3215 	if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3216 		err = -EPERM;
3217 		goto out;
3218 	}
3219 	if (!num_vfs)
3220 		goto out;
3221 
3222 	if (old_vfs) {
3223 		dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3224 			 old_vfs, max_vfs);
3225 		adapter->vfs_allocated_count = old_vfs;
3226 	} else
3227 		adapter->vfs_allocated_count = num_vfs;
3228 
3229 	adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3230 				sizeof(struct vf_data_storage), GFP_KERNEL);
3231 
3232 	/* if allocation failed then we do not support SR-IOV */
3233 	if (!adapter->vf_data) {
3234 		adapter->vfs_allocated_count = 0;
3235 		err = -ENOMEM;
3236 		goto out;
3237 	}
3238 
3239 	/* Due to the limited number of RAR entries calculate potential
3240 	 * number of MAC filters available for the VFs. Reserve entries
3241 	 * for PF default MAC, PF MAC filters and at least one RAR entry
3242 	 * for each VF for VF MAC.
3243 	 */
3244 	num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3245 			     (1 + IGB_PF_MAC_FILTERS_RESERVED +
3246 			      adapter->vfs_allocated_count);
3247 
3248 	adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3249 				       sizeof(struct vf_mac_filter),
3250 				       GFP_KERNEL);
3251 
3252 	mac_list = adapter->vf_mac_list;
3253 	INIT_LIST_HEAD(&adapter->vf_macs.l);
3254 
3255 	if (adapter->vf_mac_list) {
3256 		/* Initialize list of VF MAC filters */
3257 		for (i = 0; i < num_vf_mac_filters; i++) {
3258 			mac_list->vf = -1;
3259 			mac_list->free = true;
3260 			list_add(&mac_list->l, &adapter->vf_macs.l);
3261 			mac_list++;
3262 		}
3263 	} else {
3264 		/* If we could not allocate memory for the VF MAC filters
3265 		 * we can continue without this feature but warn user.
3266 		 */
3267 		dev_err(&pdev->dev,
3268 			"Unable to allocate memory for VF MAC filter list\n");
3269 	}
3270 
3271 	/* only call pci_enable_sriov() if no VFs are allocated already */
3272 	if (!old_vfs) {
3273 		err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3274 		if (err)
3275 			goto err_out;
3276 	}
3277 	dev_info(&pdev->dev, "%d VFs allocated\n",
3278 		 adapter->vfs_allocated_count);
3279 	for (i = 0; i < adapter->vfs_allocated_count; i++)
3280 		igb_vf_configure(adapter, i);
3281 
3282 	/* DMA Coalescing is not supported in IOV mode. */
3283 	adapter->flags &= ~IGB_FLAG_DMAC;
3284 	goto out;
3285 
3286 err_out:
3287 	kfree(adapter->vf_mac_list);
3288 	adapter->vf_mac_list = NULL;
3289 	kfree(adapter->vf_data);
3290 	adapter->vf_data = NULL;
3291 	adapter->vfs_allocated_count = 0;
3292 out:
3293 	return err;
3294 }
3295 
3296 #endif
3297 /**
3298  *  igb_remove_i2c - Cleanup  I2C interface
3299  *  @adapter: pointer to adapter structure
3300  **/
3301 static void igb_remove_i2c(struct igb_adapter *adapter)
3302 {
3303 	/* free the adapter bus structure */
3304 	i2c_del_adapter(&adapter->i2c_adap);
3305 }
3306 
3307 /**
3308  *  igb_remove - Device Removal Routine
3309  *  @pdev: PCI device information struct
3310  *
3311  *  igb_remove is called by the PCI subsystem to alert the driver
3312  *  that it should release a PCI device.  The could be caused by a
3313  *  Hot-Plug event, or because the driver is going to be removed from
3314  *  memory.
3315  **/
3316 static void igb_remove(struct pci_dev *pdev)
3317 {
3318 	struct net_device *netdev = pci_get_drvdata(pdev);
3319 	struct igb_adapter *adapter = netdev_priv(netdev);
3320 	struct e1000_hw *hw = &adapter->hw;
3321 
3322 	pm_runtime_get_noresume(&pdev->dev);
3323 #ifdef CONFIG_IGB_HWMON
3324 	igb_sysfs_exit(adapter);
3325 #endif
3326 	igb_remove_i2c(adapter);
3327 	igb_ptp_stop(adapter);
3328 	/* The watchdog timer may be rescheduled, so explicitly
3329 	 * disable watchdog from being rescheduled.
3330 	 */
3331 	set_bit(__IGB_DOWN, &adapter->state);
3332 	del_timer_sync(&adapter->watchdog_timer);
3333 	del_timer_sync(&adapter->phy_info_timer);
3334 
3335 	cancel_work_sync(&adapter->reset_task);
3336 	cancel_work_sync(&adapter->watchdog_task);
3337 
3338 #ifdef CONFIG_IGB_DCA
3339 	if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3340 		dev_info(&pdev->dev, "DCA disabled\n");
3341 		dca_remove_requester(&pdev->dev);
3342 		adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3343 		wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3344 	}
3345 #endif
3346 
3347 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
3348 	 * would have already happened in close and is redundant.
3349 	 */
3350 	igb_release_hw_control(adapter);
3351 
3352 #ifdef CONFIG_PCI_IOV
3353 	igb_disable_sriov(pdev);
3354 #endif
3355 
3356 	unregister_netdev(netdev);
3357 
3358 	igb_clear_interrupt_scheme(adapter);
3359 
3360 	pci_iounmap(pdev, adapter->io_addr);
3361 	if (hw->flash_address)
3362 		iounmap(hw->flash_address);
3363 	pci_release_mem_regions(pdev);
3364 
3365 	kfree(adapter->mac_table);
3366 	kfree(adapter->shadow_vfta);
3367 	free_netdev(netdev);
3368 
3369 	pci_disable_pcie_error_reporting(pdev);
3370 
3371 	pci_disable_device(pdev);
3372 }
3373 
3374 /**
3375  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3376  *  @adapter: board private structure to initialize
3377  *
3378  *  This function initializes the vf specific data storage and then attempts to
3379  *  allocate the VFs.  The reason for ordering it this way is because it is much
3380  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3381  *  the memory for the VFs.
3382  **/
3383 static void igb_probe_vfs(struct igb_adapter *adapter)
3384 {
3385 #ifdef CONFIG_PCI_IOV
3386 	struct pci_dev *pdev = adapter->pdev;
3387 	struct e1000_hw *hw = &adapter->hw;
3388 
3389 	/* Virtualization features not supported on i210 family. */
3390 	if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3391 		return;
3392 
3393 	/* Of the below we really only want the effect of getting
3394 	 * IGB_FLAG_HAS_MSIX set (if available), without which
3395 	 * igb_enable_sriov() has no effect.
3396 	 */
3397 	igb_set_interrupt_capability(adapter, true);
3398 	igb_reset_interrupt_capability(adapter);
3399 
3400 	pci_sriov_set_totalvfs(pdev, 7);
3401 	igb_enable_sriov(pdev, max_vfs);
3402 
3403 #endif /* CONFIG_PCI_IOV */
3404 }
3405 
3406 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3407 {
3408 	struct e1000_hw *hw = &adapter->hw;
3409 	unsigned int max_rss_queues;
3410 
3411 	/* Determine the maximum number of RSS queues supported. */
3412 	switch (hw->mac.type) {
3413 	case e1000_i211:
3414 		max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3415 		break;
3416 	case e1000_82575:
3417 	case e1000_i210:
3418 		max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3419 		break;
3420 	case e1000_i350:
3421 		/* I350 cannot do RSS and SR-IOV at the same time */
3422 		if (!!adapter->vfs_allocated_count) {
3423 			max_rss_queues = 1;
3424 			break;
3425 		}
3426 		/* fall through */
3427 	case e1000_82576:
3428 		if (!!adapter->vfs_allocated_count) {
3429 			max_rss_queues = 2;
3430 			break;
3431 		}
3432 		/* fall through */
3433 	case e1000_82580:
3434 	case e1000_i354:
3435 	default:
3436 		max_rss_queues = IGB_MAX_RX_QUEUES;
3437 		break;
3438 	}
3439 
3440 	return max_rss_queues;
3441 }
3442 
3443 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3444 {
3445 	u32 max_rss_queues;
3446 
3447 	max_rss_queues = igb_get_max_rss_queues(adapter);
3448 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3449 
3450 	igb_set_flag_queue_pairs(adapter, max_rss_queues);
3451 }
3452 
3453 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3454 			      const u32 max_rss_queues)
3455 {
3456 	struct e1000_hw *hw = &adapter->hw;
3457 
3458 	/* Determine if we need to pair queues. */
3459 	switch (hw->mac.type) {
3460 	case e1000_82575:
3461 	case e1000_i211:
3462 		/* Device supports enough interrupts without queue pairing. */
3463 		break;
3464 	case e1000_82576:
3465 	case e1000_82580:
3466 	case e1000_i350:
3467 	case e1000_i354:
3468 	case e1000_i210:
3469 	default:
3470 		/* If rss_queues > half of max_rss_queues, pair the queues in
3471 		 * order to conserve interrupts due to limited supply.
3472 		 */
3473 		if (adapter->rss_queues > (max_rss_queues / 2))
3474 			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3475 		else
3476 			adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3477 		break;
3478 	}
3479 }
3480 
3481 /**
3482  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3483  *  @adapter: board private structure to initialize
3484  *
3485  *  igb_sw_init initializes the Adapter private data structure.
3486  *  Fields are initialized based on PCI device information and
3487  *  OS network device settings (MTU size).
3488  **/
3489 static int igb_sw_init(struct igb_adapter *adapter)
3490 {
3491 	struct e1000_hw *hw = &adapter->hw;
3492 	struct net_device *netdev = adapter->netdev;
3493 	struct pci_dev *pdev = adapter->pdev;
3494 
3495 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3496 
3497 	/* set default ring sizes */
3498 	adapter->tx_ring_count = IGB_DEFAULT_TXD;
3499 	adapter->rx_ring_count = IGB_DEFAULT_RXD;
3500 
3501 	/* set default ITR values */
3502 	adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3503 	adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3504 
3505 	/* set default work limits */
3506 	adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3507 
3508 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3509 				  VLAN_HLEN;
3510 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3511 
3512 	spin_lock_init(&adapter->nfc_lock);
3513 	spin_lock_init(&adapter->stats64_lock);
3514 #ifdef CONFIG_PCI_IOV
3515 	switch (hw->mac.type) {
3516 	case e1000_82576:
3517 	case e1000_i350:
3518 		if (max_vfs > 7) {
3519 			dev_warn(&pdev->dev,
3520 				 "Maximum of 7 VFs per PF, using max\n");
3521 			max_vfs = adapter->vfs_allocated_count = 7;
3522 		} else
3523 			adapter->vfs_allocated_count = max_vfs;
3524 		if (adapter->vfs_allocated_count)
3525 			dev_warn(&pdev->dev,
3526 				 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3527 		break;
3528 	default:
3529 		break;
3530 	}
3531 #endif /* CONFIG_PCI_IOV */
3532 
3533 	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
3534 	adapter->flags |= IGB_FLAG_HAS_MSIX;
3535 
3536 	adapter->mac_table = kzalloc(sizeof(struct igb_mac_addr) *
3537 				     hw->mac.rar_entry_count, GFP_ATOMIC);
3538 	if (!adapter->mac_table)
3539 		return -ENOMEM;
3540 
3541 	igb_probe_vfs(adapter);
3542 
3543 	igb_init_queue_configuration(adapter);
3544 
3545 	/* Setup and initialize a copy of the hw vlan table array */
3546 	adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3547 				       GFP_ATOMIC);
3548 	if (!adapter->shadow_vfta)
3549 		return -ENOMEM;
3550 
3551 	/* This call may decrease the number of queues */
3552 	if (igb_init_interrupt_scheme(adapter, true)) {
3553 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3554 		return -ENOMEM;
3555 	}
3556 
3557 	/* Explicitly disable IRQ since the NIC can be in any state. */
3558 	igb_irq_disable(adapter);
3559 
3560 	if (hw->mac.type >= e1000_i350)
3561 		adapter->flags &= ~IGB_FLAG_DMAC;
3562 
3563 	set_bit(__IGB_DOWN, &adapter->state);
3564 	return 0;
3565 }
3566 
3567 /**
3568  *  igb_open - Called when a network interface is made active
3569  *  @netdev: network interface device structure
3570  *
3571  *  Returns 0 on success, negative value on failure
3572  *
3573  *  The open entry point is called when a network interface is made
3574  *  active by the system (IFF_UP).  At this point all resources needed
3575  *  for transmit and receive operations are allocated, the interrupt
3576  *  handler is registered with the OS, the watchdog timer is started,
3577  *  and the stack is notified that the interface is ready.
3578  **/
3579 static int __igb_open(struct net_device *netdev, bool resuming)
3580 {
3581 	struct igb_adapter *adapter = netdev_priv(netdev);
3582 	struct e1000_hw *hw = &adapter->hw;
3583 	struct pci_dev *pdev = adapter->pdev;
3584 	int err;
3585 	int i;
3586 
3587 	/* disallow open during test */
3588 	if (test_bit(__IGB_TESTING, &adapter->state)) {
3589 		WARN_ON(resuming);
3590 		return -EBUSY;
3591 	}
3592 
3593 	if (!resuming)
3594 		pm_runtime_get_sync(&pdev->dev);
3595 
3596 	netif_carrier_off(netdev);
3597 
3598 	/* allocate transmit descriptors */
3599 	err = igb_setup_all_tx_resources(adapter);
3600 	if (err)
3601 		goto err_setup_tx;
3602 
3603 	/* allocate receive descriptors */
3604 	err = igb_setup_all_rx_resources(adapter);
3605 	if (err)
3606 		goto err_setup_rx;
3607 
3608 	igb_power_up_link(adapter);
3609 
3610 	/* before we allocate an interrupt, we must be ready to handle it.
3611 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3612 	 * as soon as we call pci_request_irq, so we have to setup our
3613 	 * clean_rx handler before we do so.
3614 	 */
3615 	igb_configure(adapter);
3616 
3617 	err = igb_request_irq(adapter);
3618 	if (err)
3619 		goto err_req_irq;
3620 
3621 	/* Notify the stack of the actual queue counts. */
3622 	err = netif_set_real_num_tx_queues(adapter->netdev,
3623 					   adapter->num_tx_queues);
3624 	if (err)
3625 		goto err_set_queues;
3626 
3627 	err = netif_set_real_num_rx_queues(adapter->netdev,
3628 					   adapter->num_rx_queues);
3629 	if (err)
3630 		goto err_set_queues;
3631 
3632 	/* From here on the code is the same as igb_up() */
3633 	clear_bit(__IGB_DOWN, &adapter->state);
3634 
3635 	for (i = 0; i < adapter->num_q_vectors; i++)
3636 		napi_enable(&(adapter->q_vector[i]->napi));
3637 
3638 	/* Clear any pending interrupts. */
3639 	rd32(E1000_ICR);
3640 
3641 	igb_irq_enable(adapter);
3642 
3643 	/* notify VFs that reset has been completed */
3644 	if (adapter->vfs_allocated_count) {
3645 		u32 reg_data = rd32(E1000_CTRL_EXT);
3646 
3647 		reg_data |= E1000_CTRL_EXT_PFRSTD;
3648 		wr32(E1000_CTRL_EXT, reg_data);
3649 	}
3650 
3651 	netif_tx_start_all_queues(netdev);
3652 
3653 	if (!resuming)
3654 		pm_runtime_put(&pdev->dev);
3655 
3656 	/* start the watchdog. */
3657 	hw->mac.get_link_status = 1;
3658 	schedule_work(&adapter->watchdog_task);
3659 
3660 	return 0;
3661 
3662 err_set_queues:
3663 	igb_free_irq(adapter);
3664 err_req_irq:
3665 	igb_release_hw_control(adapter);
3666 	igb_power_down_link(adapter);
3667 	igb_free_all_rx_resources(adapter);
3668 err_setup_rx:
3669 	igb_free_all_tx_resources(adapter);
3670 err_setup_tx:
3671 	igb_reset(adapter);
3672 	if (!resuming)
3673 		pm_runtime_put(&pdev->dev);
3674 
3675 	return err;
3676 }
3677 
3678 int igb_open(struct net_device *netdev)
3679 {
3680 	return __igb_open(netdev, false);
3681 }
3682 
3683 /**
3684  *  igb_close - Disables a network interface
3685  *  @netdev: network interface device structure
3686  *
3687  *  Returns 0, this is not allowed to fail
3688  *
3689  *  The close entry point is called when an interface is de-activated
3690  *  by the OS.  The hardware is still under the driver's control, but
3691  *  needs to be disabled.  A global MAC reset is issued to stop the
3692  *  hardware, and all transmit and receive resources are freed.
3693  **/
3694 static int __igb_close(struct net_device *netdev, bool suspending)
3695 {
3696 	struct igb_adapter *adapter = netdev_priv(netdev);
3697 	struct pci_dev *pdev = adapter->pdev;
3698 
3699 	WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3700 
3701 	if (!suspending)
3702 		pm_runtime_get_sync(&pdev->dev);
3703 
3704 	igb_down(adapter);
3705 	igb_free_irq(adapter);
3706 
3707 	igb_free_all_tx_resources(adapter);
3708 	igb_free_all_rx_resources(adapter);
3709 
3710 	if (!suspending)
3711 		pm_runtime_put_sync(&pdev->dev);
3712 	return 0;
3713 }
3714 
3715 int igb_close(struct net_device *netdev)
3716 {
3717 	if (netif_device_present(netdev) || netdev->dismantle)
3718 		return __igb_close(netdev, false);
3719 	return 0;
3720 }
3721 
3722 /**
3723  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
3724  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
3725  *
3726  *  Return 0 on success, negative on failure
3727  **/
3728 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3729 {
3730 	struct device *dev = tx_ring->dev;
3731 	int size;
3732 
3733 	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3734 
3735 	tx_ring->tx_buffer_info = vmalloc(size);
3736 	if (!tx_ring->tx_buffer_info)
3737 		goto err;
3738 
3739 	/* round up to nearest 4K */
3740 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3741 	tx_ring->size = ALIGN(tx_ring->size, 4096);
3742 
3743 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3744 					   &tx_ring->dma, GFP_KERNEL);
3745 	if (!tx_ring->desc)
3746 		goto err;
3747 
3748 	tx_ring->next_to_use = 0;
3749 	tx_ring->next_to_clean = 0;
3750 
3751 	return 0;
3752 
3753 err:
3754 	vfree(tx_ring->tx_buffer_info);
3755 	tx_ring->tx_buffer_info = NULL;
3756 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3757 	return -ENOMEM;
3758 }
3759 
3760 /**
3761  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
3762  *				 (Descriptors) for all queues
3763  *  @adapter: board private structure
3764  *
3765  *  Return 0 on success, negative on failure
3766  **/
3767 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3768 {
3769 	struct pci_dev *pdev = adapter->pdev;
3770 	int i, err = 0;
3771 
3772 	for (i = 0; i < adapter->num_tx_queues; i++) {
3773 		err = igb_setup_tx_resources(adapter->tx_ring[i]);
3774 		if (err) {
3775 			dev_err(&pdev->dev,
3776 				"Allocation for Tx Queue %u failed\n", i);
3777 			for (i--; i >= 0; i--)
3778 				igb_free_tx_resources(adapter->tx_ring[i]);
3779 			break;
3780 		}
3781 	}
3782 
3783 	return err;
3784 }
3785 
3786 /**
3787  *  igb_setup_tctl - configure the transmit control registers
3788  *  @adapter: Board private structure
3789  **/
3790 void igb_setup_tctl(struct igb_adapter *adapter)
3791 {
3792 	struct e1000_hw *hw = &adapter->hw;
3793 	u32 tctl;
3794 
3795 	/* disable queue 0 which is enabled by default on 82575 and 82576 */
3796 	wr32(E1000_TXDCTL(0), 0);
3797 
3798 	/* Program the Transmit Control Register */
3799 	tctl = rd32(E1000_TCTL);
3800 	tctl &= ~E1000_TCTL_CT;
3801 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3802 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3803 
3804 	igb_config_collision_dist(hw);
3805 
3806 	/* Enable transmits */
3807 	tctl |= E1000_TCTL_EN;
3808 
3809 	wr32(E1000_TCTL, tctl);
3810 }
3811 
3812 /**
3813  *  igb_configure_tx_ring - Configure transmit ring after Reset
3814  *  @adapter: board private structure
3815  *  @ring: tx ring to configure
3816  *
3817  *  Configure a transmit ring after a reset.
3818  **/
3819 void igb_configure_tx_ring(struct igb_adapter *adapter,
3820 			   struct igb_ring *ring)
3821 {
3822 	struct e1000_hw *hw = &adapter->hw;
3823 	u32 txdctl = 0;
3824 	u64 tdba = ring->dma;
3825 	int reg_idx = ring->reg_idx;
3826 
3827 	/* disable the queue */
3828 	wr32(E1000_TXDCTL(reg_idx), 0);
3829 	wrfl();
3830 	mdelay(10);
3831 
3832 	wr32(E1000_TDLEN(reg_idx),
3833 	     ring->count * sizeof(union e1000_adv_tx_desc));
3834 	wr32(E1000_TDBAL(reg_idx),
3835 	     tdba & 0x00000000ffffffffULL);
3836 	wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3837 
3838 	ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
3839 	wr32(E1000_TDH(reg_idx), 0);
3840 	writel(0, ring->tail);
3841 
3842 	txdctl |= IGB_TX_PTHRESH;
3843 	txdctl |= IGB_TX_HTHRESH << 8;
3844 	txdctl |= IGB_TX_WTHRESH << 16;
3845 
3846 	/* reinitialize tx_buffer_info */
3847 	memset(ring->tx_buffer_info, 0,
3848 	       sizeof(struct igb_tx_buffer) * ring->count);
3849 
3850 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3851 	wr32(E1000_TXDCTL(reg_idx), txdctl);
3852 }
3853 
3854 /**
3855  *  igb_configure_tx - Configure transmit Unit after Reset
3856  *  @adapter: board private structure
3857  *
3858  *  Configure the Tx unit of the MAC after a reset.
3859  **/
3860 static void igb_configure_tx(struct igb_adapter *adapter)
3861 {
3862 	int i;
3863 
3864 	for (i = 0; i < adapter->num_tx_queues; i++)
3865 		igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3866 }
3867 
3868 /**
3869  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3870  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3871  *
3872  *  Returns 0 on success, negative on failure
3873  **/
3874 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3875 {
3876 	struct device *dev = rx_ring->dev;
3877 	int size;
3878 
3879 	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3880 
3881 	rx_ring->rx_buffer_info = vmalloc(size);
3882 	if (!rx_ring->rx_buffer_info)
3883 		goto err;
3884 
3885 	/* Round up to nearest 4K */
3886 	rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3887 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3888 
3889 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3890 					   &rx_ring->dma, GFP_KERNEL);
3891 	if (!rx_ring->desc)
3892 		goto err;
3893 
3894 	rx_ring->next_to_alloc = 0;
3895 	rx_ring->next_to_clean = 0;
3896 	rx_ring->next_to_use = 0;
3897 
3898 	return 0;
3899 
3900 err:
3901 	vfree(rx_ring->rx_buffer_info);
3902 	rx_ring->rx_buffer_info = NULL;
3903 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3904 	return -ENOMEM;
3905 }
3906 
3907 /**
3908  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3909  *				 (Descriptors) for all queues
3910  *  @adapter: board private structure
3911  *
3912  *  Return 0 on success, negative on failure
3913  **/
3914 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3915 {
3916 	struct pci_dev *pdev = adapter->pdev;
3917 	int i, err = 0;
3918 
3919 	for (i = 0; i < adapter->num_rx_queues; i++) {
3920 		err = igb_setup_rx_resources(adapter->rx_ring[i]);
3921 		if (err) {
3922 			dev_err(&pdev->dev,
3923 				"Allocation for Rx Queue %u failed\n", i);
3924 			for (i--; i >= 0; i--)
3925 				igb_free_rx_resources(adapter->rx_ring[i]);
3926 			break;
3927 		}
3928 	}
3929 
3930 	return err;
3931 }
3932 
3933 /**
3934  *  igb_setup_mrqc - configure the multiple receive queue control registers
3935  *  @adapter: Board private structure
3936  **/
3937 static void igb_setup_mrqc(struct igb_adapter *adapter)
3938 {
3939 	struct e1000_hw *hw = &adapter->hw;
3940 	u32 mrqc, rxcsum;
3941 	u32 j, num_rx_queues;
3942 	u32 rss_key[10];
3943 
3944 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
3945 	for (j = 0; j < 10; j++)
3946 		wr32(E1000_RSSRK(j), rss_key[j]);
3947 
3948 	num_rx_queues = adapter->rss_queues;
3949 
3950 	switch (hw->mac.type) {
3951 	case e1000_82576:
3952 		/* 82576 supports 2 RSS queues for SR-IOV */
3953 		if (adapter->vfs_allocated_count)
3954 			num_rx_queues = 2;
3955 		break;
3956 	default:
3957 		break;
3958 	}
3959 
3960 	if (adapter->rss_indir_tbl_init != num_rx_queues) {
3961 		for (j = 0; j < IGB_RETA_SIZE; j++)
3962 			adapter->rss_indir_tbl[j] =
3963 			(j * num_rx_queues) / IGB_RETA_SIZE;
3964 		adapter->rss_indir_tbl_init = num_rx_queues;
3965 	}
3966 	igb_write_rss_indir_tbl(adapter);
3967 
3968 	/* Disable raw packet checksumming so that RSS hash is placed in
3969 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3970 	 * offloads as they are enabled by default
3971 	 */
3972 	rxcsum = rd32(E1000_RXCSUM);
3973 	rxcsum |= E1000_RXCSUM_PCSD;
3974 
3975 	if (adapter->hw.mac.type >= e1000_82576)
3976 		/* Enable Receive Checksum Offload for SCTP */
3977 		rxcsum |= E1000_RXCSUM_CRCOFL;
3978 
3979 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
3980 	wr32(E1000_RXCSUM, rxcsum);
3981 
3982 	/* Generate RSS hash based on packet types, TCP/UDP
3983 	 * port numbers and/or IPv4/v6 src and dst addresses
3984 	 */
3985 	mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3986 	       E1000_MRQC_RSS_FIELD_IPV4_TCP |
3987 	       E1000_MRQC_RSS_FIELD_IPV6 |
3988 	       E1000_MRQC_RSS_FIELD_IPV6_TCP |
3989 	       E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3990 
3991 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3992 		mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3993 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3994 		mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3995 
3996 	/* If VMDq is enabled then we set the appropriate mode for that, else
3997 	 * we default to RSS so that an RSS hash is calculated per packet even
3998 	 * if we are only using one queue
3999 	 */
4000 	if (adapter->vfs_allocated_count) {
4001 		if (hw->mac.type > e1000_82575) {
4002 			/* Set the default pool for the PF's first queue */
4003 			u32 vtctl = rd32(E1000_VT_CTL);
4004 
4005 			vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4006 				   E1000_VT_CTL_DISABLE_DEF_POOL);
4007 			vtctl |= adapter->vfs_allocated_count <<
4008 				E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4009 			wr32(E1000_VT_CTL, vtctl);
4010 		}
4011 		if (adapter->rss_queues > 1)
4012 			mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4013 		else
4014 			mrqc |= E1000_MRQC_ENABLE_VMDQ;
4015 	} else {
4016 		if (hw->mac.type != e1000_i211)
4017 			mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4018 	}
4019 	igb_vmm_control(adapter);
4020 
4021 	wr32(E1000_MRQC, mrqc);
4022 }
4023 
4024 /**
4025  *  igb_setup_rctl - configure the receive control registers
4026  *  @adapter: Board private structure
4027  **/
4028 void igb_setup_rctl(struct igb_adapter *adapter)
4029 {
4030 	struct e1000_hw *hw = &adapter->hw;
4031 	u32 rctl;
4032 
4033 	rctl = rd32(E1000_RCTL);
4034 
4035 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4036 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4037 
4038 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4039 		(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4040 
4041 	/* enable stripping of CRC. It's unlikely this will break BMC
4042 	 * redirection as it did with e1000. Newer features require
4043 	 * that the HW strips the CRC.
4044 	 */
4045 	rctl |= E1000_RCTL_SECRC;
4046 
4047 	/* disable store bad packets and clear size bits. */
4048 	rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4049 
4050 	/* enable LPE to allow for reception of jumbo frames */
4051 	rctl |= E1000_RCTL_LPE;
4052 
4053 	/* disable queue 0 to prevent tail write w/o re-config */
4054 	wr32(E1000_RXDCTL(0), 0);
4055 
4056 	/* Attention!!!  For SR-IOV PF driver operations you must enable
4057 	 * queue drop for all VF and PF queues to prevent head of line blocking
4058 	 * if an un-trusted VF does not provide descriptors to hardware.
4059 	 */
4060 	if (adapter->vfs_allocated_count) {
4061 		/* set all queue drop enable bits */
4062 		wr32(E1000_QDE, ALL_QUEUES);
4063 	}
4064 
4065 	/* This is useful for sniffing bad packets. */
4066 	if (adapter->netdev->features & NETIF_F_RXALL) {
4067 		/* UPE and MPE will be handled by normal PROMISC logic
4068 		 * in e1000e_set_rx_mode
4069 		 */
4070 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4071 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4072 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4073 
4074 		rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4075 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4076 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4077 		 * and that breaks VLANs.
4078 		 */
4079 	}
4080 
4081 	wr32(E1000_RCTL, rctl);
4082 }
4083 
4084 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4085 				   int vfn)
4086 {
4087 	struct e1000_hw *hw = &adapter->hw;
4088 	u32 vmolr;
4089 
4090 	if (size > MAX_JUMBO_FRAME_SIZE)
4091 		size = MAX_JUMBO_FRAME_SIZE;
4092 
4093 	vmolr = rd32(E1000_VMOLR(vfn));
4094 	vmolr &= ~E1000_VMOLR_RLPML_MASK;
4095 	vmolr |= size | E1000_VMOLR_LPE;
4096 	wr32(E1000_VMOLR(vfn), vmolr);
4097 
4098 	return 0;
4099 }
4100 
4101 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4102 					 int vfn, bool enable)
4103 {
4104 	struct e1000_hw *hw = &adapter->hw;
4105 	u32 val, reg;
4106 
4107 	if (hw->mac.type < e1000_82576)
4108 		return;
4109 
4110 	if (hw->mac.type == e1000_i350)
4111 		reg = E1000_DVMOLR(vfn);
4112 	else
4113 		reg = E1000_VMOLR(vfn);
4114 
4115 	val = rd32(reg);
4116 	if (enable)
4117 		val |= E1000_VMOLR_STRVLAN;
4118 	else
4119 		val &= ~(E1000_VMOLR_STRVLAN);
4120 	wr32(reg, val);
4121 }
4122 
4123 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4124 				 int vfn, bool aupe)
4125 {
4126 	struct e1000_hw *hw = &adapter->hw;
4127 	u32 vmolr;
4128 
4129 	/* This register exists only on 82576 and newer so if we are older then
4130 	 * we should exit and do nothing
4131 	 */
4132 	if (hw->mac.type < e1000_82576)
4133 		return;
4134 
4135 	vmolr = rd32(E1000_VMOLR(vfn));
4136 	if (aupe)
4137 		vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4138 	else
4139 		vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4140 
4141 	/* clear all bits that might not be set */
4142 	vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4143 
4144 	if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4145 		vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4146 	/* for VMDq only allow the VFs and pool 0 to accept broadcast and
4147 	 * multicast packets
4148 	 */
4149 	if (vfn <= adapter->vfs_allocated_count)
4150 		vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4151 
4152 	wr32(E1000_VMOLR(vfn), vmolr);
4153 }
4154 
4155 /**
4156  *  igb_configure_rx_ring - Configure a receive ring after Reset
4157  *  @adapter: board private structure
4158  *  @ring: receive ring to be configured
4159  *
4160  *  Configure the Rx unit of the MAC after a reset.
4161  **/
4162 void igb_configure_rx_ring(struct igb_adapter *adapter,
4163 			   struct igb_ring *ring)
4164 {
4165 	struct e1000_hw *hw = &adapter->hw;
4166 	union e1000_adv_rx_desc *rx_desc;
4167 	u64 rdba = ring->dma;
4168 	int reg_idx = ring->reg_idx;
4169 	u32 srrctl = 0, rxdctl = 0;
4170 
4171 	/* disable the queue */
4172 	wr32(E1000_RXDCTL(reg_idx), 0);
4173 
4174 	/* Set DMA base address registers */
4175 	wr32(E1000_RDBAL(reg_idx),
4176 	     rdba & 0x00000000ffffffffULL);
4177 	wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4178 	wr32(E1000_RDLEN(reg_idx),
4179 	     ring->count * sizeof(union e1000_adv_rx_desc));
4180 
4181 	/* initialize head and tail */
4182 	ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4183 	wr32(E1000_RDH(reg_idx), 0);
4184 	writel(0, ring->tail);
4185 
4186 	/* set descriptor configuration */
4187 	srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4188 	if (ring_uses_large_buffer(ring))
4189 		srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4190 	else
4191 		srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4192 	srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4193 	if (hw->mac.type >= e1000_82580)
4194 		srrctl |= E1000_SRRCTL_TIMESTAMP;
4195 	/* Only set Drop Enable if we are supporting multiple queues */
4196 	if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
4197 		srrctl |= E1000_SRRCTL_DROP_EN;
4198 
4199 	wr32(E1000_SRRCTL(reg_idx), srrctl);
4200 
4201 	/* set filtering for VMDQ pools */
4202 	igb_set_vmolr(adapter, reg_idx & 0x7, true);
4203 
4204 	rxdctl |= IGB_RX_PTHRESH;
4205 	rxdctl |= IGB_RX_HTHRESH << 8;
4206 	rxdctl |= IGB_RX_WTHRESH << 16;
4207 
4208 	/* initialize rx_buffer_info */
4209 	memset(ring->rx_buffer_info, 0,
4210 	       sizeof(struct igb_rx_buffer) * ring->count);
4211 
4212 	/* initialize Rx descriptor 0 */
4213 	rx_desc = IGB_RX_DESC(ring, 0);
4214 	rx_desc->wb.upper.length = 0;
4215 
4216 	/* enable receive descriptor fetching */
4217 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4218 	wr32(E1000_RXDCTL(reg_idx), rxdctl);
4219 }
4220 
4221 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4222 				  struct igb_ring *rx_ring)
4223 {
4224 	/* set build_skb and buffer size flags */
4225 	clear_ring_build_skb_enabled(rx_ring);
4226 	clear_ring_uses_large_buffer(rx_ring);
4227 
4228 	if (adapter->flags & IGB_FLAG_RX_LEGACY)
4229 		return;
4230 
4231 	set_ring_build_skb_enabled(rx_ring);
4232 
4233 #if (PAGE_SIZE < 8192)
4234 	if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4235 		return;
4236 
4237 	set_ring_uses_large_buffer(rx_ring);
4238 #endif
4239 }
4240 
4241 /**
4242  *  igb_configure_rx - Configure receive Unit after Reset
4243  *  @adapter: board private structure
4244  *
4245  *  Configure the Rx unit of the MAC after a reset.
4246  **/
4247 static void igb_configure_rx(struct igb_adapter *adapter)
4248 {
4249 	int i;
4250 
4251 	/* set the correct pool for the PF default MAC address in entry 0 */
4252 	igb_set_default_mac_filter(adapter);
4253 
4254 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
4255 	 * the Base and Length of the Rx Descriptor Ring
4256 	 */
4257 	for (i = 0; i < adapter->num_rx_queues; i++) {
4258 		struct igb_ring *rx_ring = adapter->rx_ring[i];
4259 
4260 		igb_set_rx_buffer_len(adapter, rx_ring);
4261 		igb_configure_rx_ring(adapter, rx_ring);
4262 	}
4263 }
4264 
4265 /**
4266  *  igb_free_tx_resources - Free Tx Resources per Queue
4267  *  @tx_ring: Tx descriptor ring for a specific queue
4268  *
4269  *  Free all transmit software resources
4270  **/
4271 void igb_free_tx_resources(struct igb_ring *tx_ring)
4272 {
4273 	igb_clean_tx_ring(tx_ring);
4274 
4275 	vfree(tx_ring->tx_buffer_info);
4276 	tx_ring->tx_buffer_info = NULL;
4277 
4278 	/* if not set, then don't free */
4279 	if (!tx_ring->desc)
4280 		return;
4281 
4282 	dma_free_coherent(tx_ring->dev, tx_ring->size,
4283 			  tx_ring->desc, tx_ring->dma);
4284 
4285 	tx_ring->desc = NULL;
4286 }
4287 
4288 /**
4289  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
4290  *  @adapter: board private structure
4291  *
4292  *  Free all transmit software resources
4293  **/
4294 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4295 {
4296 	int i;
4297 
4298 	for (i = 0; i < adapter->num_tx_queues; i++)
4299 		if (adapter->tx_ring[i])
4300 			igb_free_tx_resources(adapter->tx_ring[i]);
4301 }
4302 
4303 /**
4304  *  igb_clean_tx_ring - Free Tx Buffers
4305  *  @tx_ring: ring to be cleaned
4306  **/
4307 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4308 {
4309 	u16 i = tx_ring->next_to_clean;
4310 	struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4311 
4312 	while (i != tx_ring->next_to_use) {
4313 		union e1000_adv_tx_desc *eop_desc, *tx_desc;
4314 
4315 		/* Free all the Tx ring sk_buffs */
4316 		dev_kfree_skb_any(tx_buffer->skb);
4317 
4318 		/* unmap skb header data */
4319 		dma_unmap_single(tx_ring->dev,
4320 				 dma_unmap_addr(tx_buffer, dma),
4321 				 dma_unmap_len(tx_buffer, len),
4322 				 DMA_TO_DEVICE);
4323 
4324 		/* check for eop_desc to determine the end of the packet */
4325 		eop_desc = tx_buffer->next_to_watch;
4326 		tx_desc = IGB_TX_DESC(tx_ring, i);
4327 
4328 		/* unmap remaining buffers */
4329 		while (tx_desc != eop_desc) {
4330 			tx_buffer++;
4331 			tx_desc++;
4332 			i++;
4333 			if (unlikely(i == tx_ring->count)) {
4334 				i = 0;
4335 				tx_buffer = tx_ring->tx_buffer_info;
4336 				tx_desc = IGB_TX_DESC(tx_ring, 0);
4337 			}
4338 
4339 			/* unmap any remaining paged data */
4340 			if (dma_unmap_len(tx_buffer, len))
4341 				dma_unmap_page(tx_ring->dev,
4342 					       dma_unmap_addr(tx_buffer, dma),
4343 					       dma_unmap_len(tx_buffer, len),
4344 					       DMA_TO_DEVICE);
4345 		}
4346 
4347 		/* move us one more past the eop_desc for start of next pkt */
4348 		tx_buffer++;
4349 		i++;
4350 		if (unlikely(i == tx_ring->count)) {
4351 			i = 0;
4352 			tx_buffer = tx_ring->tx_buffer_info;
4353 		}
4354 	}
4355 
4356 	/* reset BQL for queue */
4357 	netdev_tx_reset_queue(txring_txq(tx_ring));
4358 
4359 	/* reset next_to_use and next_to_clean */
4360 	tx_ring->next_to_use = 0;
4361 	tx_ring->next_to_clean = 0;
4362 }
4363 
4364 /**
4365  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
4366  *  @adapter: board private structure
4367  **/
4368 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4369 {
4370 	int i;
4371 
4372 	for (i = 0; i < adapter->num_tx_queues; i++)
4373 		if (adapter->tx_ring[i])
4374 			igb_clean_tx_ring(adapter->tx_ring[i]);
4375 }
4376 
4377 /**
4378  *  igb_free_rx_resources - Free Rx Resources
4379  *  @rx_ring: ring to clean the resources from
4380  *
4381  *  Free all receive software resources
4382  **/
4383 void igb_free_rx_resources(struct igb_ring *rx_ring)
4384 {
4385 	igb_clean_rx_ring(rx_ring);
4386 
4387 	vfree(rx_ring->rx_buffer_info);
4388 	rx_ring->rx_buffer_info = NULL;
4389 
4390 	/* if not set, then don't free */
4391 	if (!rx_ring->desc)
4392 		return;
4393 
4394 	dma_free_coherent(rx_ring->dev, rx_ring->size,
4395 			  rx_ring->desc, rx_ring->dma);
4396 
4397 	rx_ring->desc = NULL;
4398 }
4399 
4400 /**
4401  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4402  *  @adapter: board private structure
4403  *
4404  *  Free all receive software resources
4405  **/
4406 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4407 {
4408 	int i;
4409 
4410 	for (i = 0; i < adapter->num_rx_queues; i++)
4411 		if (adapter->rx_ring[i])
4412 			igb_free_rx_resources(adapter->rx_ring[i]);
4413 }
4414 
4415 /**
4416  *  igb_clean_rx_ring - Free Rx Buffers per Queue
4417  *  @rx_ring: ring to free buffers from
4418  **/
4419 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4420 {
4421 	u16 i = rx_ring->next_to_clean;
4422 
4423 	if (rx_ring->skb)
4424 		dev_kfree_skb(rx_ring->skb);
4425 	rx_ring->skb = NULL;
4426 
4427 	/* Free all the Rx ring sk_buffs */
4428 	while (i != rx_ring->next_to_alloc) {
4429 		struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4430 
4431 		/* Invalidate cache lines that may have been written to by
4432 		 * device so that we avoid corrupting memory.
4433 		 */
4434 		dma_sync_single_range_for_cpu(rx_ring->dev,
4435 					      buffer_info->dma,
4436 					      buffer_info->page_offset,
4437 					      igb_rx_bufsz(rx_ring),
4438 					      DMA_FROM_DEVICE);
4439 
4440 		/* free resources associated with mapping */
4441 		dma_unmap_page_attrs(rx_ring->dev,
4442 				     buffer_info->dma,
4443 				     igb_rx_pg_size(rx_ring),
4444 				     DMA_FROM_DEVICE,
4445 				     IGB_RX_DMA_ATTR);
4446 		__page_frag_cache_drain(buffer_info->page,
4447 					buffer_info->pagecnt_bias);
4448 
4449 		i++;
4450 		if (i == rx_ring->count)
4451 			i = 0;
4452 	}
4453 
4454 	rx_ring->next_to_alloc = 0;
4455 	rx_ring->next_to_clean = 0;
4456 	rx_ring->next_to_use = 0;
4457 }
4458 
4459 /**
4460  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4461  *  @adapter: board private structure
4462  **/
4463 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4464 {
4465 	int i;
4466 
4467 	for (i = 0; i < adapter->num_rx_queues; i++)
4468 		if (adapter->rx_ring[i])
4469 			igb_clean_rx_ring(adapter->rx_ring[i]);
4470 }
4471 
4472 /**
4473  *  igb_set_mac - Change the Ethernet Address of the NIC
4474  *  @netdev: network interface device structure
4475  *  @p: pointer to an address structure
4476  *
4477  *  Returns 0 on success, negative on failure
4478  **/
4479 static int igb_set_mac(struct net_device *netdev, void *p)
4480 {
4481 	struct igb_adapter *adapter = netdev_priv(netdev);
4482 	struct e1000_hw *hw = &adapter->hw;
4483 	struct sockaddr *addr = p;
4484 
4485 	if (!is_valid_ether_addr(addr->sa_data))
4486 		return -EADDRNOTAVAIL;
4487 
4488 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4489 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4490 
4491 	/* set the correct pool for the new PF MAC address in entry 0 */
4492 	igb_set_default_mac_filter(adapter);
4493 
4494 	return 0;
4495 }
4496 
4497 /**
4498  *  igb_write_mc_addr_list - write multicast addresses to MTA
4499  *  @netdev: network interface device structure
4500  *
4501  *  Writes multicast address list to the MTA hash table.
4502  *  Returns: -ENOMEM on failure
4503  *           0 on no addresses written
4504  *           X on writing X addresses to MTA
4505  **/
4506 static int igb_write_mc_addr_list(struct net_device *netdev)
4507 {
4508 	struct igb_adapter *adapter = netdev_priv(netdev);
4509 	struct e1000_hw *hw = &adapter->hw;
4510 	struct netdev_hw_addr *ha;
4511 	u8  *mta_list;
4512 	int i;
4513 
4514 	if (netdev_mc_empty(netdev)) {
4515 		/* nothing to program, so clear mc list */
4516 		igb_update_mc_addr_list(hw, NULL, 0);
4517 		igb_restore_vf_multicasts(adapter);
4518 		return 0;
4519 	}
4520 
4521 	mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
4522 	if (!mta_list)
4523 		return -ENOMEM;
4524 
4525 	/* The shared function expects a packed array of only addresses. */
4526 	i = 0;
4527 	netdev_for_each_mc_addr(ha, netdev)
4528 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
4529 
4530 	igb_update_mc_addr_list(hw, mta_list, i);
4531 	kfree(mta_list);
4532 
4533 	return netdev_mc_count(netdev);
4534 }
4535 
4536 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4537 {
4538 	struct e1000_hw *hw = &adapter->hw;
4539 	u32 i, pf_id;
4540 
4541 	switch (hw->mac.type) {
4542 	case e1000_i210:
4543 	case e1000_i211:
4544 	case e1000_i350:
4545 		/* VLAN filtering needed for VLAN prio filter */
4546 		if (adapter->netdev->features & NETIF_F_NTUPLE)
4547 			break;
4548 		/* fall through */
4549 	case e1000_82576:
4550 	case e1000_82580:
4551 	case e1000_i354:
4552 		/* VLAN filtering needed for pool filtering */
4553 		if (adapter->vfs_allocated_count)
4554 			break;
4555 		/* fall through */
4556 	default:
4557 		return 1;
4558 	}
4559 
4560 	/* We are already in VLAN promisc, nothing to do */
4561 	if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4562 		return 0;
4563 
4564 	if (!adapter->vfs_allocated_count)
4565 		goto set_vfta;
4566 
4567 	/* Add PF to all active pools */
4568 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4569 
4570 	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4571 		u32 vlvf = rd32(E1000_VLVF(i));
4572 
4573 		vlvf |= BIT(pf_id);
4574 		wr32(E1000_VLVF(i), vlvf);
4575 	}
4576 
4577 set_vfta:
4578 	/* Set all bits in the VLAN filter table array */
4579 	for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4580 		hw->mac.ops.write_vfta(hw, i, ~0U);
4581 
4582 	/* Set flag so we don't redo unnecessary work */
4583 	adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4584 
4585 	return 0;
4586 }
4587 
4588 #define VFTA_BLOCK_SIZE 8
4589 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4590 {
4591 	struct e1000_hw *hw = &adapter->hw;
4592 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4593 	u32 vid_start = vfta_offset * 32;
4594 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4595 	u32 i, vid, word, bits, pf_id;
4596 
4597 	/* guarantee that we don't scrub out management VLAN */
4598 	vid = adapter->mng_vlan_id;
4599 	if (vid >= vid_start && vid < vid_end)
4600 		vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4601 
4602 	if (!adapter->vfs_allocated_count)
4603 		goto set_vfta;
4604 
4605 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4606 
4607 	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4608 		u32 vlvf = rd32(E1000_VLVF(i));
4609 
4610 		/* pull VLAN ID from VLVF */
4611 		vid = vlvf & VLAN_VID_MASK;
4612 
4613 		/* only concern ourselves with a certain range */
4614 		if (vid < vid_start || vid >= vid_end)
4615 			continue;
4616 
4617 		if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4618 			/* record VLAN ID in VFTA */
4619 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4620 
4621 			/* if PF is part of this then continue */
4622 			if (test_bit(vid, adapter->active_vlans))
4623 				continue;
4624 		}
4625 
4626 		/* remove PF from the pool */
4627 		bits = ~BIT(pf_id);
4628 		bits &= rd32(E1000_VLVF(i));
4629 		wr32(E1000_VLVF(i), bits);
4630 	}
4631 
4632 set_vfta:
4633 	/* extract values from active_vlans and write back to VFTA */
4634 	for (i = VFTA_BLOCK_SIZE; i--;) {
4635 		vid = (vfta_offset + i) * 32;
4636 		word = vid / BITS_PER_LONG;
4637 		bits = vid % BITS_PER_LONG;
4638 
4639 		vfta[i] |= adapter->active_vlans[word] >> bits;
4640 
4641 		hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4642 	}
4643 }
4644 
4645 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4646 {
4647 	u32 i;
4648 
4649 	/* We are not in VLAN promisc, nothing to do */
4650 	if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4651 		return;
4652 
4653 	/* Set flag so we don't redo unnecessary work */
4654 	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4655 
4656 	for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4657 		igb_scrub_vfta(adapter, i);
4658 }
4659 
4660 /**
4661  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4662  *  @netdev: network interface device structure
4663  *
4664  *  The set_rx_mode entry point is called whenever the unicast or multicast
4665  *  address lists or the network interface flags are updated.  This routine is
4666  *  responsible for configuring the hardware for proper unicast, multicast,
4667  *  promiscuous mode, and all-multi behavior.
4668  **/
4669 static void igb_set_rx_mode(struct net_device *netdev)
4670 {
4671 	struct igb_adapter *adapter = netdev_priv(netdev);
4672 	struct e1000_hw *hw = &adapter->hw;
4673 	unsigned int vfn = adapter->vfs_allocated_count;
4674 	u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
4675 	int count;
4676 
4677 	/* Check for Promiscuous and All Multicast modes */
4678 	if (netdev->flags & IFF_PROMISC) {
4679 		rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
4680 		vmolr |= E1000_VMOLR_MPME;
4681 
4682 		/* enable use of UTA filter to force packets to default pool */
4683 		if (hw->mac.type == e1000_82576)
4684 			vmolr |= E1000_VMOLR_ROPE;
4685 	} else {
4686 		if (netdev->flags & IFF_ALLMULTI) {
4687 			rctl |= E1000_RCTL_MPE;
4688 			vmolr |= E1000_VMOLR_MPME;
4689 		} else {
4690 			/* Write addresses to the MTA, if the attempt fails
4691 			 * then we should just turn on promiscuous mode so
4692 			 * that we can at least receive multicast traffic
4693 			 */
4694 			count = igb_write_mc_addr_list(netdev);
4695 			if (count < 0) {
4696 				rctl |= E1000_RCTL_MPE;
4697 				vmolr |= E1000_VMOLR_MPME;
4698 			} else if (count) {
4699 				vmolr |= E1000_VMOLR_ROMPE;
4700 			}
4701 		}
4702 	}
4703 
4704 	/* Write addresses to available RAR registers, if there is not
4705 	 * sufficient space to store all the addresses then enable
4706 	 * unicast promiscuous mode
4707 	 */
4708 	if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
4709 		rctl |= E1000_RCTL_UPE;
4710 		vmolr |= E1000_VMOLR_ROPE;
4711 	}
4712 
4713 	/* enable VLAN filtering by default */
4714 	rctl |= E1000_RCTL_VFE;
4715 
4716 	/* disable VLAN filtering for modes that require it */
4717 	if ((netdev->flags & IFF_PROMISC) ||
4718 	    (netdev->features & NETIF_F_RXALL)) {
4719 		/* if we fail to set all rules then just clear VFE */
4720 		if (igb_vlan_promisc_enable(adapter))
4721 			rctl &= ~E1000_RCTL_VFE;
4722 	} else {
4723 		igb_vlan_promisc_disable(adapter);
4724 	}
4725 
4726 	/* update state of unicast, multicast, and VLAN filtering modes */
4727 	rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
4728 				     E1000_RCTL_VFE);
4729 	wr32(E1000_RCTL, rctl);
4730 
4731 #if (PAGE_SIZE < 8192)
4732 	if (!adapter->vfs_allocated_count) {
4733 		if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4734 			rlpml = IGB_MAX_FRAME_BUILD_SKB;
4735 	}
4736 #endif
4737 	wr32(E1000_RLPML, rlpml);
4738 
4739 	/* In order to support SR-IOV and eventually VMDq it is necessary to set
4740 	 * the VMOLR to enable the appropriate modes.  Without this workaround
4741 	 * we will have issues with VLAN tag stripping not being done for frames
4742 	 * that are only arriving because we are the default pool
4743 	 */
4744 	if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4745 		return;
4746 
4747 	/* set UTA to appropriate mode */
4748 	igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
4749 
4750 	vmolr |= rd32(E1000_VMOLR(vfn)) &
4751 		 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
4752 
4753 	/* enable Rx jumbo frames, restrict as needed to support build_skb */
4754 	vmolr &= ~E1000_VMOLR_RLPML_MASK;
4755 #if (PAGE_SIZE < 8192)
4756 	if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4757 		vmolr |= IGB_MAX_FRAME_BUILD_SKB;
4758 	else
4759 #endif
4760 		vmolr |= MAX_JUMBO_FRAME_SIZE;
4761 	vmolr |= E1000_VMOLR_LPE;
4762 
4763 	wr32(E1000_VMOLR(vfn), vmolr);
4764 
4765 	igb_restore_vf_multicasts(adapter);
4766 }
4767 
4768 static void igb_check_wvbr(struct igb_adapter *adapter)
4769 {
4770 	struct e1000_hw *hw = &adapter->hw;
4771 	u32 wvbr = 0;
4772 
4773 	switch (hw->mac.type) {
4774 	case e1000_82576:
4775 	case e1000_i350:
4776 		wvbr = rd32(E1000_WVBR);
4777 		if (!wvbr)
4778 			return;
4779 		break;
4780 	default:
4781 		break;
4782 	}
4783 
4784 	adapter->wvbr |= wvbr;
4785 }
4786 
4787 #define IGB_STAGGERED_QUEUE_OFFSET 8
4788 
4789 static void igb_spoof_check(struct igb_adapter *adapter)
4790 {
4791 	int j;
4792 
4793 	if (!adapter->wvbr)
4794 		return;
4795 
4796 	for (j = 0; j < adapter->vfs_allocated_count; j++) {
4797 		if (adapter->wvbr & BIT(j) ||
4798 		    adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
4799 			dev_warn(&adapter->pdev->dev,
4800 				"Spoof event(s) detected on VF %d\n", j);
4801 			adapter->wvbr &=
4802 				~(BIT(j) |
4803 				  BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
4804 		}
4805 	}
4806 }
4807 
4808 /* Need to wait a few seconds after link up to get diagnostic information from
4809  * the phy
4810  */
4811 static void igb_update_phy_info(struct timer_list *t)
4812 {
4813 	struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4814 	igb_get_phy_info(&adapter->hw);
4815 }
4816 
4817 /**
4818  *  igb_has_link - check shared code for link and determine up/down
4819  *  @adapter: pointer to driver private info
4820  **/
4821 bool igb_has_link(struct igb_adapter *adapter)
4822 {
4823 	struct e1000_hw *hw = &adapter->hw;
4824 	bool link_active = false;
4825 
4826 	/* get_link_status is set on LSC (link status) interrupt or
4827 	 * rx sequence error interrupt.  get_link_status will stay
4828 	 * false until the e1000_check_for_link establishes link
4829 	 * for copper adapters ONLY
4830 	 */
4831 	switch (hw->phy.media_type) {
4832 	case e1000_media_type_copper:
4833 		if (!hw->mac.get_link_status)
4834 			return true;
4835 	case e1000_media_type_internal_serdes:
4836 		hw->mac.ops.check_for_link(hw);
4837 		link_active = !hw->mac.get_link_status;
4838 		break;
4839 	default:
4840 	case e1000_media_type_unknown:
4841 		break;
4842 	}
4843 
4844 	if (((hw->mac.type == e1000_i210) ||
4845 	     (hw->mac.type == e1000_i211)) &&
4846 	     (hw->phy.id == I210_I_PHY_ID)) {
4847 		if (!netif_carrier_ok(adapter->netdev)) {
4848 			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4849 		} else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4850 			adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4851 			adapter->link_check_timeout = jiffies;
4852 		}
4853 	}
4854 
4855 	return link_active;
4856 }
4857 
4858 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4859 {
4860 	bool ret = false;
4861 	u32 ctrl_ext, thstat;
4862 
4863 	/* check for thermal sensor event on i350 copper only */
4864 	if (hw->mac.type == e1000_i350) {
4865 		thstat = rd32(E1000_THSTAT);
4866 		ctrl_ext = rd32(E1000_CTRL_EXT);
4867 
4868 		if ((hw->phy.media_type == e1000_media_type_copper) &&
4869 		    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4870 			ret = !!(thstat & event);
4871 	}
4872 
4873 	return ret;
4874 }
4875 
4876 /**
4877  *  igb_check_lvmmc - check for malformed packets received
4878  *  and indicated in LVMMC register
4879  *  @adapter: pointer to adapter
4880  **/
4881 static void igb_check_lvmmc(struct igb_adapter *adapter)
4882 {
4883 	struct e1000_hw *hw = &adapter->hw;
4884 	u32 lvmmc;
4885 
4886 	lvmmc = rd32(E1000_LVMMC);
4887 	if (lvmmc) {
4888 		if (unlikely(net_ratelimit())) {
4889 			netdev_warn(adapter->netdev,
4890 				    "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4891 				    lvmmc);
4892 		}
4893 	}
4894 }
4895 
4896 /**
4897  *  igb_watchdog - Timer Call-back
4898  *  @data: pointer to adapter cast into an unsigned long
4899  **/
4900 static void igb_watchdog(struct timer_list *t)
4901 {
4902 	struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
4903 	/* Do the rest outside of interrupt context */
4904 	schedule_work(&adapter->watchdog_task);
4905 }
4906 
4907 static void igb_watchdog_task(struct work_struct *work)
4908 {
4909 	struct igb_adapter *adapter = container_of(work,
4910 						   struct igb_adapter,
4911 						   watchdog_task);
4912 	struct e1000_hw *hw = &adapter->hw;
4913 	struct e1000_phy_info *phy = &hw->phy;
4914 	struct net_device *netdev = adapter->netdev;
4915 	u32 link;
4916 	int i;
4917 	u32 connsw;
4918 	u16 phy_data, retry_count = 20;
4919 
4920 	link = igb_has_link(adapter);
4921 
4922 	if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4923 		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4924 			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4925 		else
4926 			link = false;
4927 	}
4928 
4929 	/* Force link down if we have fiber to swap to */
4930 	if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4931 		if (hw->phy.media_type == e1000_media_type_copper) {
4932 			connsw = rd32(E1000_CONNSW);
4933 			if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4934 				link = 0;
4935 		}
4936 	}
4937 	if (link) {
4938 		/* Perform a reset if the media type changed. */
4939 		if (hw->dev_spec._82575.media_changed) {
4940 			hw->dev_spec._82575.media_changed = false;
4941 			adapter->flags |= IGB_FLAG_MEDIA_RESET;
4942 			igb_reset(adapter);
4943 		}
4944 		/* Cancel scheduled suspend requests. */
4945 		pm_runtime_resume(netdev->dev.parent);
4946 
4947 		if (!netif_carrier_ok(netdev)) {
4948 			u32 ctrl;
4949 
4950 			hw->mac.ops.get_speed_and_duplex(hw,
4951 							 &adapter->link_speed,
4952 							 &adapter->link_duplex);
4953 
4954 			ctrl = rd32(E1000_CTRL);
4955 			/* Links status message must follow this format */
4956 			netdev_info(netdev,
4957 			       "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4958 			       netdev->name,
4959 			       adapter->link_speed,
4960 			       adapter->link_duplex == FULL_DUPLEX ?
4961 			       "Full" : "Half",
4962 			       (ctrl & E1000_CTRL_TFCE) &&
4963 			       (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4964 			       (ctrl & E1000_CTRL_RFCE) ?  "RX" :
4965 			       (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
4966 
4967 			/* disable EEE if enabled */
4968 			if ((adapter->flags & IGB_FLAG_EEE) &&
4969 				(adapter->link_duplex == HALF_DUPLEX)) {
4970 				dev_info(&adapter->pdev->dev,
4971 				"EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4972 				adapter->hw.dev_spec._82575.eee_disable = true;
4973 				adapter->flags &= ~IGB_FLAG_EEE;
4974 			}
4975 
4976 			/* check if SmartSpeed worked */
4977 			igb_check_downshift(hw);
4978 			if (phy->speed_downgraded)
4979 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4980 
4981 			/* check for thermal sensor event */
4982 			if (igb_thermal_sensor_event(hw,
4983 			    E1000_THSTAT_LINK_THROTTLE))
4984 				netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
4985 
4986 			/* adjust timeout factor according to speed/duplex */
4987 			adapter->tx_timeout_factor = 1;
4988 			switch (adapter->link_speed) {
4989 			case SPEED_10:
4990 				adapter->tx_timeout_factor = 14;
4991 				break;
4992 			case SPEED_100:
4993 				/* maybe add some timeout factor ? */
4994 				break;
4995 			}
4996 
4997 			if (adapter->link_speed != SPEED_1000)
4998 				goto no_wait;
4999 
5000 			/* wait for Remote receiver status OK */
5001 retry_read_status:
5002 			if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5003 					      &phy_data)) {
5004 				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5005 				    retry_count) {
5006 					msleep(100);
5007 					retry_count--;
5008 					goto retry_read_status;
5009 				} else if (!retry_count) {
5010 					dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5011 				}
5012 			} else {
5013 				dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5014 			}
5015 no_wait:
5016 			netif_carrier_on(netdev);
5017 
5018 			igb_ping_all_vfs(adapter);
5019 			igb_check_vf_rate_limit(adapter);
5020 
5021 			/* link state has changed, schedule phy info update */
5022 			if (!test_bit(__IGB_DOWN, &adapter->state))
5023 				mod_timer(&adapter->phy_info_timer,
5024 					  round_jiffies(jiffies + 2 * HZ));
5025 		}
5026 	} else {
5027 		if (netif_carrier_ok(netdev)) {
5028 			adapter->link_speed = 0;
5029 			adapter->link_duplex = 0;
5030 
5031 			/* check for thermal sensor event */
5032 			if (igb_thermal_sensor_event(hw,
5033 			    E1000_THSTAT_PWR_DOWN)) {
5034 				netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5035 			}
5036 
5037 			/* Links status message must follow this format */
5038 			netdev_info(netdev, "igb: %s NIC Link is Down\n",
5039 			       netdev->name);
5040 			netif_carrier_off(netdev);
5041 
5042 			igb_ping_all_vfs(adapter);
5043 
5044 			/* link state has changed, schedule phy info update */
5045 			if (!test_bit(__IGB_DOWN, &adapter->state))
5046 				mod_timer(&adapter->phy_info_timer,
5047 					  round_jiffies(jiffies + 2 * HZ));
5048 
5049 			/* link is down, time to check for alternate media */
5050 			if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5051 				igb_check_swap_media(adapter);
5052 				if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5053 					schedule_work(&adapter->reset_task);
5054 					/* return immediately */
5055 					return;
5056 				}
5057 			}
5058 			pm_schedule_suspend(netdev->dev.parent,
5059 					    MSEC_PER_SEC * 5);
5060 
5061 		/* also check for alternate media here */
5062 		} else if (!netif_carrier_ok(netdev) &&
5063 			   (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5064 			igb_check_swap_media(adapter);
5065 			if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5066 				schedule_work(&adapter->reset_task);
5067 				/* return immediately */
5068 				return;
5069 			}
5070 		}
5071 	}
5072 
5073 	spin_lock(&adapter->stats64_lock);
5074 	igb_update_stats(adapter);
5075 	spin_unlock(&adapter->stats64_lock);
5076 
5077 	for (i = 0; i < adapter->num_tx_queues; i++) {
5078 		struct igb_ring *tx_ring = adapter->tx_ring[i];
5079 		if (!netif_carrier_ok(netdev)) {
5080 			/* We've lost link, so the controller stops DMA,
5081 			 * but we've got queued Tx work that's never going
5082 			 * to get done, so reset controller to flush Tx.
5083 			 * (Do the reset outside of interrupt context).
5084 			 */
5085 			if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5086 				adapter->tx_timeout_count++;
5087 				schedule_work(&adapter->reset_task);
5088 				/* return immediately since reset is imminent */
5089 				return;
5090 			}
5091 		}
5092 
5093 		/* Force detection of hung controller every watchdog period */
5094 		set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5095 	}
5096 
5097 	/* Cause software interrupt to ensure Rx ring is cleaned */
5098 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5099 		u32 eics = 0;
5100 
5101 		for (i = 0; i < adapter->num_q_vectors; i++)
5102 			eics |= adapter->q_vector[i]->eims_value;
5103 		wr32(E1000_EICS, eics);
5104 	} else {
5105 		wr32(E1000_ICS, E1000_ICS_RXDMT0);
5106 	}
5107 
5108 	igb_spoof_check(adapter);
5109 	igb_ptp_rx_hang(adapter);
5110 	igb_ptp_tx_hang(adapter);
5111 
5112 	/* Check LVMMC register on i350/i354 only */
5113 	if ((adapter->hw.mac.type == e1000_i350) ||
5114 	    (adapter->hw.mac.type == e1000_i354))
5115 		igb_check_lvmmc(adapter);
5116 
5117 	/* Reset the timer */
5118 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
5119 		if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5120 			mod_timer(&adapter->watchdog_timer,
5121 				  round_jiffies(jiffies +  HZ));
5122 		else
5123 			mod_timer(&adapter->watchdog_timer,
5124 				  round_jiffies(jiffies + 2 * HZ));
5125 	}
5126 }
5127 
5128 enum latency_range {
5129 	lowest_latency = 0,
5130 	low_latency = 1,
5131 	bulk_latency = 2,
5132 	latency_invalid = 255
5133 };
5134 
5135 /**
5136  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
5137  *  @q_vector: pointer to q_vector
5138  *
5139  *  Stores a new ITR value based on strictly on packet size.  This
5140  *  algorithm is less sophisticated than that used in igb_update_itr,
5141  *  due to the difficulty of synchronizing statistics across multiple
5142  *  receive rings.  The divisors and thresholds used by this function
5143  *  were determined based on theoretical maximum wire speed and testing
5144  *  data, in order to minimize response time while increasing bulk
5145  *  throughput.
5146  *  This functionality is controlled by ethtool's coalescing settings.
5147  *  NOTE:  This function is called only when operating in a multiqueue
5148  *         receive environment.
5149  **/
5150 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5151 {
5152 	int new_val = q_vector->itr_val;
5153 	int avg_wire_size = 0;
5154 	struct igb_adapter *adapter = q_vector->adapter;
5155 	unsigned int packets;
5156 
5157 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
5158 	 * ints/sec - ITR timer value of 120 ticks.
5159 	 */
5160 	if (adapter->link_speed != SPEED_1000) {
5161 		new_val = IGB_4K_ITR;
5162 		goto set_itr_val;
5163 	}
5164 
5165 	packets = q_vector->rx.total_packets;
5166 	if (packets)
5167 		avg_wire_size = q_vector->rx.total_bytes / packets;
5168 
5169 	packets = q_vector->tx.total_packets;
5170 	if (packets)
5171 		avg_wire_size = max_t(u32, avg_wire_size,
5172 				      q_vector->tx.total_bytes / packets);
5173 
5174 	/* if avg_wire_size isn't set no work was done */
5175 	if (!avg_wire_size)
5176 		goto clear_counts;
5177 
5178 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
5179 	avg_wire_size += 24;
5180 
5181 	/* Don't starve jumbo frames */
5182 	avg_wire_size = min(avg_wire_size, 3000);
5183 
5184 	/* Give a little boost to mid-size frames */
5185 	if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5186 		new_val = avg_wire_size / 3;
5187 	else
5188 		new_val = avg_wire_size / 2;
5189 
5190 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
5191 	if (new_val < IGB_20K_ITR &&
5192 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5193 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5194 		new_val = IGB_20K_ITR;
5195 
5196 set_itr_val:
5197 	if (new_val != q_vector->itr_val) {
5198 		q_vector->itr_val = new_val;
5199 		q_vector->set_itr = 1;
5200 	}
5201 clear_counts:
5202 	q_vector->rx.total_bytes = 0;
5203 	q_vector->rx.total_packets = 0;
5204 	q_vector->tx.total_bytes = 0;
5205 	q_vector->tx.total_packets = 0;
5206 }
5207 
5208 /**
5209  *  igb_update_itr - update the dynamic ITR value based on statistics
5210  *  @q_vector: pointer to q_vector
5211  *  @ring_container: ring info to update the itr for
5212  *
5213  *  Stores a new ITR value based on packets and byte
5214  *  counts during the last interrupt.  The advantage of per interrupt
5215  *  computation is faster updates and more accurate ITR for the current
5216  *  traffic pattern.  Constants in this function were computed
5217  *  based on theoretical maximum wire speed and thresholds were set based
5218  *  on testing data as well as attempting to minimize response time
5219  *  while increasing bulk throughput.
5220  *  This functionality is controlled by ethtool's coalescing settings.
5221  *  NOTE:  These calculations are only valid when operating in a single-
5222  *         queue environment.
5223  **/
5224 static void igb_update_itr(struct igb_q_vector *q_vector,
5225 			   struct igb_ring_container *ring_container)
5226 {
5227 	unsigned int packets = ring_container->total_packets;
5228 	unsigned int bytes = ring_container->total_bytes;
5229 	u8 itrval = ring_container->itr;
5230 
5231 	/* no packets, exit with status unchanged */
5232 	if (packets == 0)
5233 		return;
5234 
5235 	switch (itrval) {
5236 	case lowest_latency:
5237 		/* handle TSO and jumbo frames */
5238 		if (bytes/packets > 8000)
5239 			itrval = bulk_latency;
5240 		else if ((packets < 5) && (bytes > 512))
5241 			itrval = low_latency;
5242 		break;
5243 	case low_latency:  /* 50 usec aka 20000 ints/s */
5244 		if (bytes > 10000) {
5245 			/* this if handles the TSO accounting */
5246 			if (bytes/packets > 8000)
5247 				itrval = bulk_latency;
5248 			else if ((packets < 10) || ((bytes/packets) > 1200))
5249 				itrval = bulk_latency;
5250 			else if ((packets > 35))
5251 				itrval = lowest_latency;
5252 		} else if (bytes/packets > 2000) {
5253 			itrval = bulk_latency;
5254 		} else if (packets <= 2 && bytes < 512) {
5255 			itrval = lowest_latency;
5256 		}
5257 		break;
5258 	case bulk_latency: /* 250 usec aka 4000 ints/s */
5259 		if (bytes > 25000) {
5260 			if (packets > 35)
5261 				itrval = low_latency;
5262 		} else if (bytes < 1500) {
5263 			itrval = low_latency;
5264 		}
5265 		break;
5266 	}
5267 
5268 	/* clear work counters since we have the values we need */
5269 	ring_container->total_bytes = 0;
5270 	ring_container->total_packets = 0;
5271 
5272 	/* write updated itr to ring container */
5273 	ring_container->itr = itrval;
5274 }
5275 
5276 static void igb_set_itr(struct igb_q_vector *q_vector)
5277 {
5278 	struct igb_adapter *adapter = q_vector->adapter;
5279 	u32 new_itr = q_vector->itr_val;
5280 	u8 current_itr = 0;
5281 
5282 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5283 	if (adapter->link_speed != SPEED_1000) {
5284 		current_itr = 0;
5285 		new_itr = IGB_4K_ITR;
5286 		goto set_itr_now;
5287 	}
5288 
5289 	igb_update_itr(q_vector, &q_vector->tx);
5290 	igb_update_itr(q_vector, &q_vector->rx);
5291 
5292 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5293 
5294 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
5295 	if (current_itr == lowest_latency &&
5296 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5297 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5298 		current_itr = low_latency;
5299 
5300 	switch (current_itr) {
5301 	/* counts and packets in update_itr are dependent on these numbers */
5302 	case lowest_latency:
5303 		new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5304 		break;
5305 	case low_latency:
5306 		new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5307 		break;
5308 	case bulk_latency:
5309 		new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
5310 		break;
5311 	default:
5312 		break;
5313 	}
5314 
5315 set_itr_now:
5316 	if (new_itr != q_vector->itr_val) {
5317 		/* this attempts to bias the interrupt rate towards Bulk
5318 		 * by adding intermediate steps when interrupt rate is
5319 		 * increasing
5320 		 */
5321 		new_itr = new_itr > q_vector->itr_val ?
5322 			  max((new_itr * q_vector->itr_val) /
5323 			  (new_itr + (q_vector->itr_val >> 2)),
5324 			  new_itr) : new_itr;
5325 		/* Don't write the value here; it resets the adapter's
5326 		 * internal timer, and causes us to delay far longer than
5327 		 * we should between interrupts.  Instead, we write the ITR
5328 		 * value at the beginning of the next interrupt so the timing
5329 		 * ends up being correct.
5330 		 */
5331 		q_vector->itr_val = new_itr;
5332 		q_vector->set_itr = 1;
5333 	}
5334 }
5335 
5336 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
5337 			    u32 type_tucmd, u32 mss_l4len_idx)
5338 {
5339 	struct e1000_adv_tx_context_desc *context_desc;
5340 	u16 i = tx_ring->next_to_use;
5341 
5342 	context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5343 
5344 	i++;
5345 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5346 
5347 	/* set bits to identify this as an advanced context descriptor */
5348 	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5349 
5350 	/* For 82575, context index must be unique per ring. */
5351 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5352 		mss_l4len_idx |= tx_ring->reg_idx << 4;
5353 
5354 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
5355 	context_desc->seqnum_seed	= 0;
5356 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
5357 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
5358 }
5359 
5360 static int igb_tso(struct igb_ring *tx_ring,
5361 		   struct igb_tx_buffer *first,
5362 		   u8 *hdr_len)
5363 {
5364 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5365 	struct sk_buff *skb = first->skb;
5366 	union {
5367 		struct iphdr *v4;
5368 		struct ipv6hdr *v6;
5369 		unsigned char *hdr;
5370 	} ip;
5371 	union {
5372 		struct tcphdr *tcp;
5373 		unsigned char *hdr;
5374 	} l4;
5375 	u32 paylen, l4_offset;
5376 	int err;
5377 
5378 	if (skb->ip_summed != CHECKSUM_PARTIAL)
5379 		return 0;
5380 
5381 	if (!skb_is_gso(skb))
5382 		return 0;
5383 
5384 	err = skb_cow_head(skb, 0);
5385 	if (err < 0)
5386 		return err;
5387 
5388 	ip.hdr = skb_network_header(skb);
5389 	l4.hdr = skb_checksum_start(skb);
5390 
5391 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5392 	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5393 
5394 	/* initialize outer IP header fields */
5395 	if (ip.v4->version == 4) {
5396 		unsigned char *csum_start = skb_checksum_start(skb);
5397 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5398 
5399 		/* IP header will have to cancel out any data that
5400 		 * is not a part of the outer IP header
5401 		 */
5402 		ip.v4->check = csum_fold(csum_partial(trans_start,
5403 						      csum_start - trans_start,
5404 						      0));
5405 		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5406 
5407 		ip.v4->tot_len = 0;
5408 		first->tx_flags |= IGB_TX_FLAGS_TSO |
5409 				   IGB_TX_FLAGS_CSUM |
5410 				   IGB_TX_FLAGS_IPV4;
5411 	} else {
5412 		ip.v6->payload_len = 0;
5413 		first->tx_flags |= IGB_TX_FLAGS_TSO |
5414 				   IGB_TX_FLAGS_CSUM;
5415 	}
5416 
5417 	/* determine offset of inner transport header */
5418 	l4_offset = l4.hdr - skb->data;
5419 
5420 	/* compute length of segmentation header */
5421 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
5422 
5423 	/* remove payload length from inner checksum */
5424 	paylen = skb->len - l4_offset;
5425 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
5426 
5427 	/* update gso size and bytecount with header size */
5428 	first->gso_segs = skb_shinfo(skb)->gso_segs;
5429 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
5430 
5431 	/* MSS L4LEN IDX */
5432 	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5433 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5434 
5435 	/* VLAN MACLEN IPLEN */
5436 	vlan_macip_lens = l4.hdr - ip.hdr;
5437 	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5438 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5439 
5440 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
5441 
5442 	return 1;
5443 }
5444 
5445 static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
5446 {
5447 	unsigned int offset = 0;
5448 
5449 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
5450 
5451 	return offset == skb_checksum_start_offset(skb);
5452 }
5453 
5454 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5455 {
5456 	struct sk_buff *skb = first->skb;
5457 	u32 vlan_macip_lens = 0;
5458 	u32 type_tucmd = 0;
5459 
5460 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
5461 csum_failed:
5462 		if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
5463 			return;
5464 		goto no_csum;
5465 	}
5466 
5467 	switch (skb->csum_offset) {
5468 	case offsetof(struct tcphdr, check):
5469 		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5470 		/* fall through */
5471 	case offsetof(struct udphdr, check):
5472 		break;
5473 	case offsetof(struct sctphdr, checksum):
5474 		/* validate that this is actually an SCTP request */
5475 		if (((first->protocol == htons(ETH_P_IP)) &&
5476 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
5477 		    ((first->protocol == htons(ETH_P_IPV6)) &&
5478 		     igb_ipv6_csum_is_sctp(skb))) {
5479 			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5480 			break;
5481 		}
5482 	default:
5483 		skb_checksum_help(skb);
5484 		goto csum_failed;
5485 	}
5486 
5487 	/* update TX checksum flag */
5488 	first->tx_flags |= IGB_TX_FLAGS_CSUM;
5489 	vlan_macip_lens = skb_checksum_start_offset(skb) -
5490 			  skb_network_offset(skb);
5491 no_csum:
5492 	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5493 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5494 
5495 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
5496 }
5497 
5498 #define IGB_SET_FLAG(_input, _flag, _result) \
5499 	((_flag <= _result) ? \
5500 	 ((u32)(_input & _flag) * (_result / _flag)) : \
5501 	 ((u32)(_input & _flag) / (_flag / _result)))
5502 
5503 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
5504 {
5505 	/* set type for advanced descriptor with frame checksum insertion */
5506 	u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
5507 		       E1000_ADVTXD_DCMD_DEXT |
5508 		       E1000_ADVTXD_DCMD_IFCS;
5509 
5510 	/* set HW vlan bit if vlan is present */
5511 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
5512 				 (E1000_ADVTXD_DCMD_VLE));
5513 
5514 	/* set segmentation bits for TSO */
5515 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
5516 				 (E1000_ADVTXD_DCMD_TSE));
5517 
5518 	/* set timestamp bit if present */
5519 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
5520 				 (E1000_ADVTXD_MAC_TSTAMP));
5521 
5522 	/* insert frame checksum */
5523 	cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
5524 
5525 	return cmd_type;
5526 }
5527 
5528 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
5529 				 union e1000_adv_tx_desc *tx_desc,
5530 				 u32 tx_flags, unsigned int paylen)
5531 {
5532 	u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
5533 
5534 	/* 82575 requires a unique index per ring */
5535 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5536 		olinfo_status |= tx_ring->reg_idx << 4;
5537 
5538 	/* insert L4 checksum */
5539 	olinfo_status |= IGB_SET_FLAG(tx_flags,
5540 				      IGB_TX_FLAGS_CSUM,
5541 				      (E1000_TXD_POPTS_TXSM << 8));
5542 
5543 	/* insert IPv4 checksum */
5544 	olinfo_status |= IGB_SET_FLAG(tx_flags,
5545 				      IGB_TX_FLAGS_IPV4,
5546 				      (E1000_TXD_POPTS_IXSM << 8));
5547 
5548 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5549 }
5550 
5551 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5552 {
5553 	struct net_device *netdev = tx_ring->netdev;
5554 
5555 	netif_stop_subqueue(netdev, tx_ring->queue_index);
5556 
5557 	/* Herbert's original patch had:
5558 	 *  smp_mb__after_netif_stop_queue();
5559 	 * but since that doesn't exist yet, just open code it.
5560 	 */
5561 	smp_mb();
5562 
5563 	/* We need to check again in a case another CPU has just
5564 	 * made room available.
5565 	 */
5566 	if (igb_desc_unused(tx_ring) < size)
5567 		return -EBUSY;
5568 
5569 	/* A reprieve! */
5570 	netif_wake_subqueue(netdev, tx_ring->queue_index);
5571 
5572 	u64_stats_update_begin(&tx_ring->tx_syncp2);
5573 	tx_ring->tx_stats.restart_queue2++;
5574 	u64_stats_update_end(&tx_ring->tx_syncp2);
5575 
5576 	return 0;
5577 }
5578 
5579 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5580 {
5581 	if (igb_desc_unused(tx_ring) >= size)
5582 		return 0;
5583 	return __igb_maybe_stop_tx(tx_ring, size);
5584 }
5585 
5586 static int igb_tx_map(struct igb_ring *tx_ring,
5587 		      struct igb_tx_buffer *first,
5588 		      const u8 hdr_len)
5589 {
5590 	struct sk_buff *skb = first->skb;
5591 	struct igb_tx_buffer *tx_buffer;
5592 	union e1000_adv_tx_desc *tx_desc;
5593 	struct skb_frag_struct *frag;
5594 	dma_addr_t dma;
5595 	unsigned int data_len, size;
5596 	u32 tx_flags = first->tx_flags;
5597 	u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5598 	u16 i = tx_ring->next_to_use;
5599 
5600 	tx_desc = IGB_TX_DESC(tx_ring, i);
5601 
5602 	igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5603 
5604 	size = skb_headlen(skb);
5605 	data_len = skb->data_len;
5606 
5607 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5608 
5609 	tx_buffer = first;
5610 
5611 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5612 		if (dma_mapping_error(tx_ring->dev, dma))
5613 			goto dma_error;
5614 
5615 		/* record length, and DMA address */
5616 		dma_unmap_len_set(tx_buffer, len, size);
5617 		dma_unmap_addr_set(tx_buffer, dma, dma);
5618 
5619 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
5620 
5621 		while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5622 			tx_desc->read.cmd_type_len =
5623 				cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5624 
5625 			i++;
5626 			tx_desc++;
5627 			if (i == tx_ring->count) {
5628 				tx_desc = IGB_TX_DESC(tx_ring, 0);
5629 				i = 0;
5630 			}
5631 			tx_desc->read.olinfo_status = 0;
5632 
5633 			dma += IGB_MAX_DATA_PER_TXD;
5634 			size -= IGB_MAX_DATA_PER_TXD;
5635 
5636 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
5637 		}
5638 
5639 		if (likely(!data_len))
5640 			break;
5641 
5642 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
5643 
5644 		i++;
5645 		tx_desc++;
5646 		if (i == tx_ring->count) {
5647 			tx_desc = IGB_TX_DESC(tx_ring, 0);
5648 			i = 0;
5649 		}
5650 		tx_desc->read.olinfo_status = 0;
5651 
5652 		size = skb_frag_size(frag);
5653 		data_len -= size;
5654 
5655 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
5656 				       size, DMA_TO_DEVICE);
5657 
5658 		tx_buffer = &tx_ring->tx_buffer_info[i];
5659 	}
5660 
5661 	/* write last descriptor with RS and EOP bits */
5662 	cmd_type |= size | IGB_TXD_DCMD;
5663 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
5664 
5665 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5666 
5667 	/* set the timestamp */
5668 	first->time_stamp = jiffies;
5669 
5670 	/* Force memory writes to complete before letting h/w know there
5671 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
5672 	 * memory model archs, such as IA-64).
5673 	 *
5674 	 * We also need this memory barrier to make certain all of the
5675 	 * status bits have been updated before next_to_watch is written.
5676 	 */
5677 	wmb();
5678 
5679 	/* set next_to_watch value indicating a packet is present */
5680 	first->next_to_watch = tx_desc;
5681 
5682 	i++;
5683 	if (i == tx_ring->count)
5684 		i = 0;
5685 
5686 	tx_ring->next_to_use = i;
5687 
5688 	/* Make sure there is space in the ring for the next send. */
5689 	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
5690 
5691 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
5692 		writel(i, tx_ring->tail);
5693 
5694 		/* we need this if more than one processor can write to our tail
5695 		 * at a time, it synchronizes IO on IA64/Altix systems
5696 		 */
5697 		mmiowb();
5698 	}
5699 	return 0;
5700 
5701 dma_error:
5702 	dev_err(tx_ring->dev, "TX DMA map failed\n");
5703 	tx_buffer = &tx_ring->tx_buffer_info[i];
5704 
5705 	/* clear dma mappings for failed tx_buffer_info map */
5706 	while (tx_buffer != first) {
5707 		if (dma_unmap_len(tx_buffer, len))
5708 			dma_unmap_page(tx_ring->dev,
5709 				       dma_unmap_addr(tx_buffer, dma),
5710 				       dma_unmap_len(tx_buffer, len),
5711 				       DMA_TO_DEVICE);
5712 		dma_unmap_len_set(tx_buffer, len, 0);
5713 
5714 		if (i-- == 0)
5715 			i += tx_ring->count;
5716 		tx_buffer = &tx_ring->tx_buffer_info[i];
5717 	}
5718 
5719 	if (dma_unmap_len(tx_buffer, len))
5720 		dma_unmap_single(tx_ring->dev,
5721 				 dma_unmap_addr(tx_buffer, dma),
5722 				 dma_unmap_len(tx_buffer, len),
5723 				 DMA_TO_DEVICE);
5724 	dma_unmap_len_set(tx_buffer, len, 0);
5725 
5726 	dev_kfree_skb_any(tx_buffer->skb);
5727 	tx_buffer->skb = NULL;
5728 
5729 	tx_ring->next_to_use = i;
5730 
5731 	return -1;
5732 }
5733 
5734 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
5735 				struct igb_ring *tx_ring)
5736 {
5737 	struct igb_tx_buffer *first;
5738 	int tso;
5739 	u32 tx_flags = 0;
5740 	unsigned short f;
5741 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
5742 	__be16 protocol = vlan_get_protocol(skb);
5743 	u8 hdr_len = 0;
5744 
5745 	/* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
5746 	 *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
5747 	 *       + 2 desc gap to keep tail from touching head,
5748 	 *       + 1 desc for context descriptor,
5749 	 * otherwise try next time
5750 	 */
5751 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5752 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5753 
5754 	if (igb_maybe_stop_tx(tx_ring, count + 3)) {
5755 		/* this is a hard error */
5756 		return NETDEV_TX_BUSY;
5757 	}
5758 
5759 	/* record the location of the first descriptor for this packet */
5760 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
5761 	first->skb = skb;
5762 	first->bytecount = skb->len;
5763 	first->gso_segs = 1;
5764 
5765 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
5766 		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5767 
5768 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
5769 		    !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
5770 					   &adapter->state)) {
5771 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5772 			tx_flags |= IGB_TX_FLAGS_TSTAMP;
5773 
5774 			adapter->ptp_tx_skb = skb_get(skb);
5775 			adapter->ptp_tx_start = jiffies;
5776 			if (adapter->hw.mac.type == e1000_82576)
5777 				schedule_work(&adapter->ptp_tx_work);
5778 		} else {
5779 			adapter->tx_hwtstamp_skipped++;
5780 		}
5781 	}
5782 
5783 	skb_tx_timestamp(skb);
5784 
5785 	if (skb_vlan_tag_present(skb)) {
5786 		tx_flags |= IGB_TX_FLAGS_VLAN;
5787 		tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5788 	}
5789 
5790 	/* record initial flags and protocol */
5791 	first->tx_flags = tx_flags;
5792 	first->protocol = protocol;
5793 
5794 	tso = igb_tso(tx_ring, first, &hdr_len);
5795 	if (tso < 0)
5796 		goto out_drop;
5797 	else if (!tso)
5798 		igb_tx_csum(tx_ring, first);
5799 
5800 	if (igb_tx_map(tx_ring, first, hdr_len))
5801 		goto cleanup_tx_tstamp;
5802 
5803 	return NETDEV_TX_OK;
5804 
5805 out_drop:
5806 	dev_kfree_skb_any(first->skb);
5807 	first->skb = NULL;
5808 cleanup_tx_tstamp:
5809 	if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
5810 		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5811 
5812 		dev_kfree_skb_any(adapter->ptp_tx_skb);
5813 		adapter->ptp_tx_skb = NULL;
5814 		if (adapter->hw.mac.type == e1000_82576)
5815 			cancel_work_sync(&adapter->ptp_tx_work);
5816 		clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
5817 	}
5818 
5819 	return NETDEV_TX_OK;
5820 }
5821 
5822 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5823 						    struct sk_buff *skb)
5824 {
5825 	unsigned int r_idx = skb->queue_mapping;
5826 
5827 	if (r_idx >= adapter->num_tx_queues)
5828 		r_idx = r_idx % adapter->num_tx_queues;
5829 
5830 	return adapter->tx_ring[r_idx];
5831 }
5832 
5833 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5834 				  struct net_device *netdev)
5835 {
5836 	struct igb_adapter *adapter = netdev_priv(netdev);
5837 
5838 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5839 	 * in order to meet this minimum size requirement.
5840 	 */
5841 	if (skb_put_padto(skb, 17))
5842 		return NETDEV_TX_OK;
5843 
5844 	return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5845 }
5846 
5847 /**
5848  *  igb_tx_timeout - Respond to a Tx Hang
5849  *  @netdev: network interface device structure
5850  **/
5851 static void igb_tx_timeout(struct net_device *netdev)
5852 {
5853 	struct igb_adapter *adapter = netdev_priv(netdev);
5854 	struct e1000_hw *hw = &adapter->hw;
5855 
5856 	/* Do the reset outside of interrupt context */
5857 	adapter->tx_timeout_count++;
5858 
5859 	if (hw->mac.type >= e1000_82580)
5860 		hw->dev_spec._82575.global_device_reset = true;
5861 
5862 	schedule_work(&adapter->reset_task);
5863 	wr32(E1000_EICS,
5864 	     (adapter->eims_enable_mask & ~adapter->eims_other));
5865 }
5866 
5867 static void igb_reset_task(struct work_struct *work)
5868 {
5869 	struct igb_adapter *adapter;
5870 	adapter = container_of(work, struct igb_adapter, reset_task);
5871 
5872 	igb_dump(adapter);
5873 	netdev_err(adapter->netdev, "Reset adapter\n");
5874 	igb_reinit_locked(adapter);
5875 }
5876 
5877 /**
5878  *  igb_get_stats64 - Get System Network Statistics
5879  *  @netdev: network interface device structure
5880  *  @stats: rtnl_link_stats64 pointer
5881  **/
5882 static void igb_get_stats64(struct net_device *netdev,
5883 			    struct rtnl_link_stats64 *stats)
5884 {
5885 	struct igb_adapter *adapter = netdev_priv(netdev);
5886 
5887 	spin_lock(&adapter->stats64_lock);
5888 	igb_update_stats(adapter);
5889 	memcpy(stats, &adapter->stats64, sizeof(*stats));
5890 	spin_unlock(&adapter->stats64_lock);
5891 }
5892 
5893 /**
5894  *  igb_change_mtu - Change the Maximum Transfer Unit
5895  *  @netdev: network interface device structure
5896  *  @new_mtu: new value for maximum frame size
5897  *
5898  *  Returns 0 on success, negative on failure
5899  **/
5900 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5901 {
5902 	struct igb_adapter *adapter = netdev_priv(netdev);
5903 	struct pci_dev *pdev = adapter->pdev;
5904 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5905 
5906 	/* adjust max frame to be at least the size of a standard frame */
5907 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5908 		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5909 
5910 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5911 		usleep_range(1000, 2000);
5912 
5913 	/* igb_down has a dependency on max_frame_size */
5914 	adapter->max_frame_size = max_frame;
5915 
5916 	if (netif_running(netdev))
5917 		igb_down(adapter);
5918 
5919 	dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5920 		 netdev->mtu, new_mtu);
5921 	netdev->mtu = new_mtu;
5922 
5923 	if (netif_running(netdev))
5924 		igb_up(adapter);
5925 	else
5926 		igb_reset(adapter);
5927 
5928 	clear_bit(__IGB_RESETTING, &adapter->state);
5929 
5930 	return 0;
5931 }
5932 
5933 /**
5934  *  igb_update_stats - Update the board statistics counters
5935  *  @adapter: board private structure
5936  **/
5937 void igb_update_stats(struct igb_adapter *adapter)
5938 {
5939 	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
5940 	struct e1000_hw *hw = &adapter->hw;
5941 	struct pci_dev *pdev = adapter->pdev;
5942 	u32 reg, mpc;
5943 	int i;
5944 	u64 bytes, packets;
5945 	unsigned int start;
5946 	u64 _bytes, _packets;
5947 
5948 	/* Prevent stats update while adapter is being reset, or if the pci
5949 	 * connection is down.
5950 	 */
5951 	if (adapter->link_speed == 0)
5952 		return;
5953 	if (pci_channel_offline(pdev))
5954 		return;
5955 
5956 	bytes = 0;
5957 	packets = 0;
5958 
5959 	rcu_read_lock();
5960 	for (i = 0; i < adapter->num_rx_queues; i++) {
5961 		struct igb_ring *ring = adapter->rx_ring[i];
5962 		u32 rqdpc = rd32(E1000_RQDPC(i));
5963 		if (hw->mac.type >= e1000_i210)
5964 			wr32(E1000_RQDPC(i), 0);
5965 
5966 		if (rqdpc) {
5967 			ring->rx_stats.drops += rqdpc;
5968 			net_stats->rx_fifo_errors += rqdpc;
5969 		}
5970 
5971 		do {
5972 			start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5973 			_bytes = ring->rx_stats.bytes;
5974 			_packets = ring->rx_stats.packets;
5975 		} while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5976 		bytes += _bytes;
5977 		packets += _packets;
5978 	}
5979 
5980 	net_stats->rx_bytes = bytes;
5981 	net_stats->rx_packets = packets;
5982 
5983 	bytes = 0;
5984 	packets = 0;
5985 	for (i = 0; i < adapter->num_tx_queues; i++) {
5986 		struct igb_ring *ring = adapter->tx_ring[i];
5987 		do {
5988 			start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5989 			_bytes = ring->tx_stats.bytes;
5990 			_packets = ring->tx_stats.packets;
5991 		} while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5992 		bytes += _bytes;
5993 		packets += _packets;
5994 	}
5995 	net_stats->tx_bytes = bytes;
5996 	net_stats->tx_packets = packets;
5997 	rcu_read_unlock();
5998 
5999 	/* read stats registers */
6000 	adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6001 	adapter->stats.gprc += rd32(E1000_GPRC);
6002 	adapter->stats.gorc += rd32(E1000_GORCL);
6003 	rd32(E1000_GORCH); /* clear GORCL */
6004 	adapter->stats.bprc += rd32(E1000_BPRC);
6005 	adapter->stats.mprc += rd32(E1000_MPRC);
6006 	adapter->stats.roc += rd32(E1000_ROC);
6007 
6008 	adapter->stats.prc64 += rd32(E1000_PRC64);
6009 	adapter->stats.prc127 += rd32(E1000_PRC127);
6010 	adapter->stats.prc255 += rd32(E1000_PRC255);
6011 	adapter->stats.prc511 += rd32(E1000_PRC511);
6012 	adapter->stats.prc1023 += rd32(E1000_PRC1023);
6013 	adapter->stats.prc1522 += rd32(E1000_PRC1522);
6014 	adapter->stats.symerrs += rd32(E1000_SYMERRS);
6015 	adapter->stats.sec += rd32(E1000_SEC);
6016 
6017 	mpc = rd32(E1000_MPC);
6018 	adapter->stats.mpc += mpc;
6019 	net_stats->rx_fifo_errors += mpc;
6020 	adapter->stats.scc += rd32(E1000_SCC);
6021 	adapter->stats.ecol += rd32(E1000_ECOL);
6022 	adapter->stats.mcc += rd32(E1000_MCC);
6023 	adapter->stats.latecol += rd32(E1000_LATECOL);
6024 	adapter->stats.dc += rd32(E1000_DC);
6025 	adapter->stats.rlec += rd32(E1000_RLEC);
6026 	adapter->stats.xonrxc += rd32(E1000_XONRXC);
6027 	adapter->stats.xontxc += rd32(E1000_XONTXC);
6028 	adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6029 	adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6030 	adapter->stats.fcruc += rd32(E1000_FCRUC);
6031 	adapter->stats.gptc += rd32(E1000_GPTC);
6032 	adapter->stats.gotc += rd32(E1000_GOTCL);
6033 	rd32(E1000_GOTCH); /* clear GOTCL */
6034 	adapter->stats.rnbc += rd32(E1000_RNBC);
6035 	adapter->stats.ruc += rd32(E1000_RUC);
6036 	adapter->stats.rfc += rd32(E1000_RFC);
6037 	adapter->stats.rjc += rd32(E1000_RJC);
6038 	adapter->stats.tor += rd32(E1000_TORH);
6039 	adapter->stats.tot += rd32(E1000_TOTH);
6040 	adapter->stats.tpr += rd32(E1000_TPR);
6041 
6042 	adapter->stats.ptc64 += rd32(E1000_PTC64);
6043 	adapter->stats.ptc127 += rd32(E1000_PTC127);
6044 	adapter->stats.ptc255 += rd32(E1000_PTC255);
6045 	adapter->stats.ptc511 += rd32(E1000_PTC511);
6046 	adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6047 	adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6048 
6049 	adapter->stats.mptc += rd32(E1000_MPTC);
6050 	adapter->stats.bptc += rd32(E1000_BPTC);
6051 
6052 	adapter->stats.tpt += rd32(E1000_TPT);
6053 	adapter->stats.colc += rd32(E1000_COLC);
6054 
6055 	adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6056 	/* read internal phy specific stats */
6057 	reg = rd32(E1000_CTRL_EXT);
6058 	if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6059 		adapter->stats.rxerrc += rd32(E1000_RXERRC);
6060 
6061 		/* this stat has invalid values on i210/i211 */
6062 		if ((hw->mac.type != e1000_i210) &&
6063 		    (hw->mac.type != e1000_i211))
6064 			adapter->stats.tncrs += rd32(E1000_TNCRS);
6065 	}
6066 
6067 	adapter->stats.tsctc += rd32(E1000_TSCTC);
6068 	adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6069 
6070 	adapter->stats.iac += rd32(E1000_IAC);
6071 	adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6072 	adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6073 	adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6074 	adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6075 	adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6076 	adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6077 	adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6078 	adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6079 
6080 	/* Fill out the OS statistics structure */
6081 	net_stats->multicast = adapter->stats.mprc;
6082 	net_stats->collisions = adapter->stats.colc;
6083 
6084 	/* Rx Errors */
6085 
6086 	/* RLEC on some newer hardware can be incorrect so build
6087 	 * our own version based on RUC and ROC
6088 	 */
6089 	net_stats->rx_errors = adapter->stats.rxerrc +
6090 		adapter->stats.crcerrs + adapter->stats.algnerrc +
6091 		adapter->stats.ruc + adapter->stats.roc +
6092 		adapter->stats.cexterr;
6093 	net_stats->rx_length_errors = adapter->stats.ruc +
6094 				      adapter->stats.roc;
6095 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
6096 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
6097 	net_stats->rx_missed_errors = adapter->stats.mpc;
6098 
6099 	/* Tx Errors */
6100 	net_stats->tx_errors = adapter->stats.ecol +
6101 			       adapter->stats.latecol;
6102 	net_stats->tx_aborted_errors = adapter->stats.ecol;
6103 	net_stats->tx_window_errors = adapter->stats.latecol;
6104 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
6105 
6106 	/* Tx Dropped needs to be maintained elsewhere */
6107 
6108 	/* Management Stats */
6109 	adapter->stats.mgptc += rd32(E1000_MGTPTC);
6110 	adapter->stats.mgprc += rd32(E1000_MGTPRC);
6111 	adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6112 
6113 	/* OS2BMC Stats */
6114 	reg = rd32(E1000_MANC);
6115 	if (reg & E1000_MANC_EN_BMC2OS) {
6116 		adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6117 		adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6118 		adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6119 		adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6120 	}
6121 }
6122 
6123 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6124 {
6125 	struct e1000_hw *hw = &adapter->hw;
6126 	struct ptp_clock_event event;
6127 	struct timespec64 ts;
6128 	u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
6129 
6130 	if (tsicr & TSINTR_SYS_WRAP) {
6131 		event.type = PTP_CLOCK_PPS;
6132 		if (adapter->ptp_caps.pps)
6133 			ptp_clock_event(adapter->ptp_clock, &event);
6134 		ack |= TSINTR_SYS_WRAP;
6135 	}
6136 
6137 	if (tsicr & E1000_TSICR_TXTS) {
6138 		/* retrieve hardware timestamp */
6139 		schedule_work(&adapter->ptp_tx_work);
6140 		ack |= E1000_TSICR_TXTS;
6141 	}
6142 
6143 	if (tsicr & TSINTR_TT0) {
6144 		spin_lock(&adapter->tmreg_lock);
6145 		ts = timespec64_add(adapter->perout[0].start,
6146 				    adapter->perout[0].period);
6147 		/* u32 conversion of tv_sec is safe until y2106 */
6148 		wr32(E1000_TRGTTIML0, ts.tv_nsec);
6149 		wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
6150 		tsauxc = rd32(E1000_TSAUXC);
6151 		tsauxc |= TSAUXC_EN_TT0;
6152 		wr32(E1000_TSAUXC, tsauxc);
6153 		adapter->perout[0].start = ts;
6154 		spin_unlock(&adapter->tmreg_lock);
6155 		ack |= TSINTR_TT0;
6156 	}
6157 
6158 	if (tsicr & TSINTR_TT1) {
6159 		spin_lock(&adapter->tmreg_lock);
6160 		ts = timespec64_add(adapter->perout[1].start,
6161 				    adapter->perout[1].period);
6162 		wr32(E1000_TRGTTIML1, ts.tv_nsec);
6163 		wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
6164 		tsauxc = rd32(E1000_TSAUXC);
6165 		tsauxc |= TSAUXC_EN_TT1;
6166 		wr32(E1000_TSAUXC, tsauxc);
6167 		adapter->perout[1].start = ts;
6168 		spin_unlock(&adapter->tmreg_lock);
6169 		ack |= TSINTR_TT1;
6170 	}
6171 
6172 	if (tsicr & TSINTR_AUTT0) {
6173 		nsec = rd32(E1000_AUXSTMPL0);
6174 		sec  = rd32(E1000_AUXSTMPH0);
6175 		event.type = PTP_CLOCK_EXTTS;
6176 		event.index = 0;
6177 		event.timestamp = sec * 1000000000ULL + nsec;
6178 		ptp_clock_event(adapter->ptp_clock, &event);
6179 		ack |= TSINTR_AUTT0;
6180 	}
6181 
6182 	if (tsicr & TSINTR_AUTT1) {
6183 		nsec = rd32(E1000_AUXSTMPL1);
6184 		sec  = rd32(E1000_AUXSTMPH1);
6185 		event.type = PTP_CLOCK_EXTTS;
6186 		event.index = 1;
6187 		event.timestamp = sec * 1000000000ULL + nsec;
6188 		ptp_clock_event(adapter->ptp_clock, &event);
6189 		ack |= TSINTR_AUTT1;
6190 	}
6191 
6192 	/* acknowledge the interrupts */
6193 	wr32(E1000_TSICR, ack);
6194 }
6195 
6196 static irqreturn_t igb_msix_other(int irq, void *data)
6197 {
6198 	struct igb_adapter *adapter = data;
6199 	struct e1000_hw *hw = &adapter->hw;
6200 	u32 icr = rd32(E1000_ICR);
6201 	/* reading ICR causes bit 31 of EICR to be cleared */
6202 
6203 	if (icr & E1000_ICR_DRSTA)
6204 		schedule_work(&adapter->reset_task);
6205 
6206 	if (icr & E1000_ICR_DOUTSYNC) {
6207 		/* HW is reporting DMA is out of sync */
6208 		adapter->stats.doosync++;
6209 		/* The DMA Out of Sync is also indication of a spoof event
6210 		 * in IOV mode. Check the Wrong VM Behavior register to
6211 		 * see if it is really a spoof event.
6212 		 */
6213 		igb_check_wvbr(adapter);
6214 	}
6215 
6216 	/* Check for a mailbox event */
6217 	if (icr & E1000_ICR_VMMB)
6218 		igb_msg_task(adapter);
6219 
6220 	if (icr & E1000_ICR_LSC) {
6221 		hw->mac.get_link_status = 1;
6222 		/* guard against interrupt when we're going down */
6223 		if (!test_bit(__IGB_DOWN, &adapter->state))
6224 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
6225 	}
6226 
6227 	if (icr & E1000_ICR_TS)
6228 		igb_tsync_interrupt(adapter);
6229 
6230 	wr32(E1000_EIMS, adapter->eims_other);
6231 
6232 	return IRQ_HANDLED;
6233 }
6234 
6235 static void igb_write_itr(struct igb_q_vector *q_vector)
6236 {
6237 	struct igb_adapter *adapter = q_vector->adapter;
6238 	u32 itr_val = q_vector->itr_val & 0x7FFC;
6239 
6240 	if (!q_vector->set_itr)
6241 		return;
6242 
6243 	if (!itr_val)
6244 		itr_val = 0x4;
6245 
6246 	if (adapter->hw.mac.type == e1000_82575)
6247 		itr_val |= itr_val << 16;
6248 	else
6249 		itr_val |= E1000_EITR_CNT_IGNR;
6250 
6251 	writel(itr_val, q_vector->itr_register);
6252 	q_vector->set_itr = 0;
6253 }
6254 
6255 static irqreturn_t igb_msix_ring(int irq, void *data)
6256 {
6257 	struct igb_q_vector *q_vector = data;
6258 
6259 	/* Write the ITR value calculated from the previous interrupt. */
6260 	igb_write_itr(q_vector);
6261 
6262 	napi_schedule(&q_vector->napi);
6263 
6264 	return IRQ_HANDLED;
6265 }
6266 
6267 #ifdef CONFIG_IGB_DCA
6268 static void igb_update_tx_dca(struct igb_adapter *adapter,
6269 			      struct igb_ring *tx_ring,
6270 			      int cpu)
6271 {
6272 	struct e1000_hw *hw = &adapter->hw;
6273 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
6274 
6275 	if (hw->mac.type != e1000_82575)
6276 		txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
6277 
6278 	/* We can enable relaxed ordering for reads, but not writes when
6279 	 * DCA is enabled.  This is due to a known issue in some chipsets
6280 	 * which will cause the DCA tag to be cleared.
6281 	 */
6282 	txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
6283 		  E1000_DCA_TXCTRL_DATA_RRO_EN |
6284 		  E1000_DCA_TXCTRL_DESC_DCA_EN;
6285 
6286 	wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
6287 }
6288 
6289 static void igb_update_rx_dca(struct igb_adapter *adapter,
6290 			      struct igb_ring *rx_ring,
6291 			      int cpu)
6292 {
6293 	struct e1000_hw *hw = &adapter->hw;
6294 	u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
6295 
6296 	if (hw->mac.type != e1000_82575)
6297 		rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
6298 
6299 	/* We can enable relaxed ordering for reads, but not writes when
6300 	 * DCA is enabled.  This is due to a known issue in some chipsets
6301 	 * which will cause the DCA tag to be cleared.
6302 	 */
6303 	rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
6304 		  E1000_DCA_RXCTRL_DESC_DCA_EN;
6305 
6306 	wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
6307 }
6308 
6309 static void igb_update_dca(struct igb_q_vector *q_vector)
6310 {
6311 	struct igb_adapter *adapter = q_vector->adapter;
6312 	int cpu = get_cpu();
6313 
6314 	if (q_vector->cpu == cpu)
6315 		goto out_no_update;
6316 
6317 	if (q_vector->tx.ring)
6318 		igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
6319 
6320 	if (q_vector->rx.ring)
6321 		igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
6322 
6323 	q_vector->cpu = cpu;
6324 out_no_update:
6325 	put_cpu();
6326 }
6327 
6328 static void igb_setup_dca(struct igb_adapter *adapter)
6329 {
6330 	struct e1000_hw *hw = &adapter->hw;
6331 	int i;
6332 
6333 	if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
6334 		return;
6335 
6336 	/* Always use CB2 mode, difference is masked in the CB driver. */
6337 	wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
6338 
6339 	for (i = 0; i < adapter->num_q_vectors; i++) {
6340 		adapter->q_vector[i]->cpu = -1;
6341 		igb_update_dca(adapter->q_vector[i]);
6342 	}
6343 }
6344 
6345 static int __igb_notify_dca(struct device *dev, void *data)
6346 {
6347 	struct net_device *netdev = dev_get_drvdata(dev);
6348 	struct igb_adapter *adapter = netdev_priv(netdev);
6349 	struct pci_dev *pdev = adapter->pdev;
6350 	struct e1000_hw *hw = &adapter->hw;
6351 	unsigned long event = *(unsigned long *)data;
6352 
6353 	switch (event) {
6354 	case DCA_PROVIDER_ADD:
6355 		/* if already enabled, don't do it again */
6356 		if (adapter->flags & IGB_FLAG_DCA_ENABLED)
6357 			break;
6358 		if (dca_add_requester(dev) == 0) {
6359 			adapter->flags |= IGB_FLAG_DCA_ENABLED;
6360 			dev_info(&pdev->dev, "DCA enabled\n");
6361 			igb_setup_dca(adapter);
6362 			break;
6363 		}
6364 		/* Fall Through since DCA is disabled. */
6365 	case DCA_PROVIDER_REMOVE:
6366 		if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
6367 			/* without this a class_device is left
6368 			 * hanging around in the sysfs model
6369 			 */
6370 			dca_remove_requester(dev);
6371 			dev_info(&pdev->dev, "DCA disabled\n");
6372 			adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
6373 			wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
6374 		}
6375 		break;
6376 	}
6377 
6378 	return 0;
6379 }
6380 
6381 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
6382 			  void *p)
6383 {
6384 	int ret_val;
6385 
6386 	ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6387 					 __igb_notify_dca);
6388 
6389 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6390 }
6391 #endif /* CONFIG_IGB_DCA */
6392 
6393 #ifdef CONFIG_PCI_IOV
6394 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6395 {
6396 	unsigned char mac_addr[ETH_ALEN];
6397 
6398 	eth_zero_addr(mac_addr);
6399 	igb_set_vf_mac(adapter, vf, mac_addr);
6400 
6401 	/* By default spoof check is enabled for all VFs */
6402 	adapter->vf_data[vf].spoofchk_enabled = true;
6403 
6404 	/* By default VFs are not trusted */
6405 	adapter->vf_data[vf].trusted = false;
6406 
6407 	return 0;
6408 }
6409 
6410 #endif
6411 static void igb_ping_all_vfs(struct igb_adapter *adapter)
6412 {
6413 	struct e1000_hw *hw = &adapter->hw;
6414 	u32 ping;
6415 	int i;
6416 
6417 	for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
6418 		ping = E1000_PF_CONTROL_MSG;
6419 		if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
6420 			ping |= E1000_VT_MSGTYPE_CTS;
6421 		igb_write_mbx(hw, &ping, 1, i);
6422 	}
6423 }
6424 
6425 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6426 {
6427 	struct e1000_hw *hw = &adapter->hw;
6428 	u32 vmolr = rd32(E1000_VMOLR(vf));
6429 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6430 
6431 	vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
6432 			    IGB_VF_FLAG_MULTI_PROMISC);
6433 	vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6434 
6435 	if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
6436 		vmolr |= E1000_VMOLR_MPME;
6437 		vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
6438 		*msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
6439 	} else {
6440 		/* if we have hashes and we are clearing a multicast promisc
6441 		 * flag we need to write the hashes to the MTA as this step
6442 		 * was previously skipped
6443 		 */
6444 		if (vf_data->num_vf_mc_hashes > 30) {
6445 			vmolr |= E1000_VMOLR_MPME;
6446 		} else if (vf_data->num_vf_mc_hashes) {
6447 			int j;
6448 
6449 			vmolr |= E1000_VMOLR_ROMPE;
6450 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6451 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6452 		}
6453 	}
6454 
6455 	wr32(E1000_VMOLR(vf), vmolr);
6456 
6457 	/* there are flags left unprocessed, likely not supported */
6458 	if (*msgbuf & E1000_VT_MSGINFO_MASK)
6459 		return -EINVAL;
6460 
6461 	return 0;
6462 }
6463 
6464 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
6465 				  u32 *msgbuf, u32 vf)
6466 {
6467 	int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6468 	u16 *hash_list = (u16 *)&msgbuf[1];
6469 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6470 	int i;
6471 
6472 	/* salt away the number of multicast addresses assigned
6473 	 * to this VF for later use to restore when the PF multi cast
6474 	 * list changes
6475 	 */
6476 	vf_data->num_vf_mc_hashes = n;
6477 
6478 	/* only up to 30 hash values supported */
6479 	if (n > 30)
6480 		n = 30;
6481 
6482 	/* store the hashes for later use */
6483 	for (i = 0; i < n; i++)
6484 		vf_data->vf_mc_hashes[i] = hash_list[i];
6485 
6486 	/* Flush and reset the mta with the new values */
6487 	igb_set_rx_mode(adapter->netdev);
6488 
6489 	return 0;
6490 }
6491 
6492 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
6493 {
6494 	struct e1000_hw *hw = &adapter->hw;
6495 	struct vf_data_storage *vf_data;
6496 	int i, j;
6497 
6498 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
6499 		u32 vmolr = rd32(E1000_VMOLR(i));
6500 
6501 		vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6502 
6503 		vf_data = &adapter->vf_data[i];
6504 
6505 		if ((vf_data->num_vf_mc_hashes > 30) ||
6506 		    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
6507 			vmolr |= E1000_VMOLR_MPME;
6508 		} else if (vf_data->num_vf_mc_hashes) {
6509 			vmolr |= E1000_VMOLR_ROMPE;
6510 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6511 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6512 		}
6513 		wr32(E1000_VMOLR(i), vmolr);
6514 	}
6515 }
6516 
6517 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
6518 {
6519 	struct e1000_hw *hw = &adapter->hw;
6520 	u32 pool_mask, vlvf_mask, i;
6521 
6522 	/* create mask for VF and other pools */
6523 	pool_mask = E1000_VLVF_POOLSEL_MASK;
6524 	vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
6525 
6526 	/* drop PF from pool bits */
6527 	pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
6528 			     adapter->vfs_allocated_count);
6529 
6530 	/* Find the vlan filter for this id */
6531 	for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
6532 		u32 vlvf = rd32(E1000_VLVF(i));
6533 		u32 vfta_mask, vid, vfta;
6534 
6535 		/* remove the vf from the pool */
6536 		if (!(vlvf & vlvf_mask))
6537 			continue;
6538 
6539 		/* clear out bit from VLVF */
6540 		vlvf ^= vlvf_mask;
6541 
6542 		/* if other pools are present, just remove ourselves */
6543 		if (vlvf & pool_mask)
6544 			goto update_vlvfb;
6545 
6546 		/* if PF is present, leave VFTA */
6547 		if (vlvf & E1000_VLVF_POOLSEL_MASK)
6548 			goto update_vlvf;
6549 
6550 		vid = vlvf & E1000_VLVF_VLANID_MASK;
6551 		vfta_mask = BIT(vid % 32);
6552 
6553 		/* clear bit from VFTA */
6554 		vfta = adapter->shadow_vfta[vid / 32];
6555 		if (vfta & vfta_mask)
6556 			hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
6557 update_vlvf:
6558 		/* clear pool selection enable */
6559 		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6560 			vlvf &= E1000_VLVF_POOLSEL_MASK;
6561 		else
6562 			vlvf = 0;
6563 update_vlvfb:
6564 		/* clear pool bits */
6565 		wr32(E1000_VLVF(i), vlvf);
6566 	}
6567 }
6568 
6569 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6570 {
6571 	u32 vlvf;
6572 	int idx;
6573 
6574 	/* short cut the special case */
6575 	if (vlan == 0)
6576 		return 0;
6577 
6578 	/* Search for the VLAN id in the VLVF entries */
6579 	for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6580 		vlvf = rd32(E1000_VLVF(idx));
6581 		if ((vlvf & VLAN_VID_MASK) == vlan)
6582 			break;
6583 	}
6584 
6585 	return idx;
6586 }
6587 
6588 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6589 {
6590 	struct e1000_hw *hw = &adapter->hw;
6591 	u32 bits, pf_id;
6592 	int idx;
6593 
6594 	idx = igb_find_vlvf_entry(hw, vid);
6595 	if (!idx)
6596 		return;
6597 
6598 	/* See if any other pools are set for this VLAN filter
6599 	 * entry other than the PF.
6600 	 */
6601 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6602 	bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
6603 	bits &= rd32(E1000_VLVF(idx));
6604 
6605 	/* Disable the filter so this falls into the default pool. */
6606 	if (!bits) {
6607 		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6608 			wr32(E1000_VLVF(idx), BIT(pf_id));
6609 		else
6610 			wr32(E1000_VLVF(idx), 0);
6611 	}
6612 }
6613 
6614 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6615 			   bool add, u32 vf)
6616 {
6617 	int pf_id = adapter->vfs_allocated_count;
6618 	struct e1000_hw *hw = &adapter->hw;
6619 	int err;
6620 
6621 	/* If VLAN overlaps with one the PF is currently monitoring make
6622 	 * sure that we are able to allocate a VLVF entry.  This may be
6623 	 * redundant but it guarantees PF will maintain visibility to
6624 	 * the VLAN.
6625 	 */
6626 	if (add && test_bit(vid, adapter->active_vlans)) {
6627 		err = igb_vfta_set(hw, vid, pf_id, true, false);
6628 		if (err)
6629 			return err;
6630 	}
6631 
6632 	err = igb_vfta_set(hw, vid, vf, add, false);
6633 
6634 	if (add && !err)
6635 		return err;
6636 
6637 	/* If we failed to add the VF VLAN or we are removing the VF VLAN
6638 	 * we may need to drop the PF pool bit in order to allow us to free
6639 	 * up the VLVF resources.
6640 	 */
6641 	if (test_bit(vid, adapter->active_vlans) ||
6642 	    (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6643 		igb_update_pf_vlvf(adapter, vid);
6644 
6645 	return err;
6646 }
6647 
6648 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
6649 {
6650 	struct e1000_hw *hw = &adapter->hw;
6651 
6652 	if (vid)
6653 		wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
6654 	else
6655 		wr32(E1000_VMVIR(vf), 0);
6656 }
6657 
6658 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
6659 				u16 vlan, u8 qos)
6660 {
6661 	int err;
6662 
6663 	err = igb_set_vf_vlan(adapter, vlan, true, vf);
6664 	if (err)
6665 		return err;
6666 
6667 	igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
6668 	igb_set_vmolr(adapter, vf, !vlan);
6669 
6670 	/* revoke access to previous VLAN */
6671 	if (vlan != adapter->vf_data[vf].pf_vlan)
6672 		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6673 				false, vf);
6674 
6675 	adapter->vf_data[vf].pf_vlan = vlan;
6676 	adapter->vf_data[vf].pf_qos = qos;
6677 	igb_set_vf_vlan_strip(adapter, vf, true);
6678 	dev_info(&adapter->pdev->dev,
6679 		 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
6680 	if (test_bit(__IGB_DOWN, &adapter->state)) {
6681 		dev_warn(&adapter->pdev->dev,
6682 			 "The VF VLAN has been set, but the PF device is not up.\n");
6683 		dev_warn(&adapter->pdev->dev,
6684 			 "Bring the PF device up before attempting to use the VF device.\n");
6685 	}
6686 
6687 	return err;
6688 }
6689 
6690 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
6691 {
6692 	/* Restore tagless access via VLAN 0 */
6693 	igb_set_vf_vlan(adapter, 0, true, vf);
6694 
6695 	igb_set_vmvir(adapter, 0, vf);
6696 	igb_set_vmolr(adapter, vf, true);
6697 
6698 	/* Remove any PF assigned VLAN */
6699 	if (adapter->vf_data[vf].pf_vlan)
6700 		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6701 				false, vf);
6702 
6703 	adapter->vf_data[vf].pf_vlan = 0;
6704 	adapter->vf_data[vf].pf_qos = 0;
6705 	igb_set_vf_vlan_strip(adapter, vf, false);
6706 
6707 	return 0;
6708 }
6709 
6710 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
6711 			       u16 vlan, u8 qos, __be16 vlan_proto)
6712 {
6713 	struct igb_adapter *adapter = netdev_priv(netdev);
6714 
6715 	if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
6716 		return -EINVAL;
6717 
6718 	if (vlan_proto != htons(ETH_P_8021Q))
6719 		return -EPROTONOSUPPORT;
6720 
6721 	return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
6722 			       igb_disable_port_vlan(adapter, vf);
6723 }
6724 
6725 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6726 {
6727 	int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6728 	int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
6729 	int ret;
6730 
6731 	if (adapter->vf_data[vf].pf_vlan)
6732 		return -1;
6733 
6734 	/* VLAN 0 is a special case, don't allow it to be removed */
6735 	if (!vid && !add)
6736 		return 0;
6737 
6738 	ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
6739 	if (!ret)
6740 		igb_set_vf_vlan_strip(adapter, vf, !!vid);
6741 	return ret;
6742 }
6743 
6744 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
6745 {
6746 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6747 
6748 	/* clear flags - except flag that indicates PF has set the MAC */
6749 	vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
6750 	vf_data->last_nack = jiffies;
6751 
6752 	/* reset vlans for device */
6753 	igb_clear_vf_vfta(adapter, vf);
6754 	igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
6755 	igb_set_vmvir(adapter, vf_data->pf_vlan |
6756 			       (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
6757 	igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
6758 	igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
6759 
6760 	/* reset multicast table array for vf */
6761 	adapter->vf_data[vf].num_vf_mc_hashes = 0;
6762 
6763 	/* Flush and reset the mta with the new values */
6764 	igb_set_rx_mode(adapter->netdev);
6765 }
6766 
6767 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
6768 {
6769 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6770 
6771 	/* clear mac address as we were hotplug removed/added */
6772 	if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
6773 		eth_zero_addr(vf_mac);
6774 
6775 	/* process remaining reset events */
6776 	igb_vf_reset(adapter, vf);
6777 }
6778 
6779 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
6780 {
6781 	struct e1000_hw *hw = &adapter->hw;
6782 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6783 	u32 reg, msgbuf[3];
6784 	u8 *addr = (u8 *)(&msgbuf[1]);
6785 
6786 	/* process all the same items cleared in a function level reset */
6787 	igb_vf_reset(adapter, vf);
6788 
6789 	/* set vf mac address */
6790 	igb_set_vf_mac(adapter, vf, vf_mac);
6791 
6792 	/* enable transmit and receive for vf */
6793 	reg = rd32(E1000_VFTE);
6794 	wr32(E1000_VFTE, reg | BIT(vf));
6795 	reg = rd32(E1000_VFRE);
6796 	wr32(E1000_VFRE, reg | BIT(vf));
6797 
6798 	adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
6799 
6800 	/* reply to reset with ack and vf mac address */
6801 	if (!is_zero_ether_addr(vf_mac)) {
6802 		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
6803 		memcpy(addr, vf_mac, ETH_ALEN);
6804 	} else {
6805 		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
6806 	}
6807 	igb_write_mbx(hw, msgbuf, 3, vf);
6808 }
6809 
6810 static void igb_flush_mac_table(struct igb_adapter *adapter)
6811 {
6812 	struct e1000_hw *hw = &adapter->hw;
6813 	int i;
6814 
6815 	for (i = 0; i < hw->mac.rar_entry_count; i++) {
6816 		adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6817 		memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6818 		adapter->mac_table[i].queue = 0;
6819 		igb_rar_set_index(adapter, i);
6820 	}
6821 }
6822 
6823 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
6824 {
6825 	struct e1000_hw *hw = &adapter->hw;
6826 	/* do not count rar entries reserved for VFs MAC addresses */
6827 	int rar_entries = hw->mac.rar_entry_count -
6828 			  adapter->vfs_allocated_count;
6829 	int i, count = 0;
6830 
6831 	for (i = 0; i < rar_entries; i++) {
6832 		/* do not count default entries */
6833 		if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
6834 			continue;
6835 
6836 		/* do not count "in use" entries for different queues */
6837 		if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
6838 		    (adapter->mac_table[i].queue != queue))
6839 			continue;
6840 
6841 		count++;
6842 	}
6843 
6844 	return count;
6845 }
6846 
6847 /* Set default MAC address for the PF in the first RAR entry */
6848 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
6849 {
6850 	struct igb_mac_addr *mac_table = &adapter->mac_table[0];
6851 
6852 	ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
6853 	mac_table->queue = adapter->vfs_allocated_count;
6854 	mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
6855 
6856 	igb_rar_set_index(adapter, 0);
6857 }
6858 
6859 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6860 			      const u8 queue)
6861 {
6862 	struct e1000_hw *hw = &adapter->hw;
6863 	int rar_entries = hw->mac.rar_entry_count -
6864 			  adapter->vfs_allocated_count;
6865 	int i;
6866 
6867 	if (is_zero_ether_addr(addr))
6868 		return -EINVAL;
6869 
6870 	/* Search for the first empty entry in the MAC table.
6871 	 * Do not touch entries at the end of the table reserved for the VF MAC
6872 	 * addresses.
6873 	 */
6874 	for (i = 0; i < rar_entries; i++) {
6875 		if (adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE)
6876 			continue;
6877 
6878 		ether_addr_copy(adapter->mac_table[i].addr, addr);
6879 		adapter->mac_table[i].queue = queue;
6880 		adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE;
6881 
6882 		igb_rar_set_index(adapter, i);
6883 		return i;
6884 	}
6885 
6886 	return -ENOSPC;
6887 }
6888 
6889 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6890 			      const u8 queue)
6891 {
6892 	struct e1000_hw *hw = &adapter->hw;
6893 	int rar_entries = hw->mac.rar_entry_count -
6894 			  adapter->vfs_allocated_count;
6895 	int i;
6896 
6897 	if (is_zero_ether_addr(addr))
6898 		return -EINVAL;
6899 
6900 	/* Search for matching entry in the MAC table based on given address
6901 	 * and queue. Do not touch entries at the end of the table reserved
6902 	 * for the VF MAC addresses.
6903 	 */
6904 	for (i = 0; i < rar_entries; i++) {
6905 		if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
6906 			continue;
6907 		if (adapter->mac_table[i].queue != queue)
6908 			continue;
6909 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
6910 			continue;
6911 
6912 		adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6913 		memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6914 		adapter->mac_table[i].queue = 0;
6915 
6916 		igb_rar_set_index(adapter, i);
6917 		return 0;
6918 	}
6919 
6920 	return -ENOENT;
6921 }
6922 
6923 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
6924 {
6925 	struct igb_adapter *adapter = netdev_priv(netdev);
6926 	int ret;
6927 
6928 	ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6929 
6930 	return min_t(int, ret, 0);
6931 }
6932 
6933 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
6934 {
6935 	struct igb_adapter *adapter = netdev_priv(netdev);
6936 
6937 	igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6938 
6939 	return 0;
6940 }
6941 
6942 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
6943 				 const u32 info, const u8 *addr)
6944 {
6945 	struct pci_dev *pdev = adapter->pdev;
6946 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6947 	struct list_head *pos;
6948 	struct vf_mac_filter *entry = NULL;
6949 	int ret = 0;
6950 
6951 	switch (info) {
6952 	case E1000_VF_MAC_FILTER_CLR:
6953 		/* remove all unicast MAC filters related to the current VF */
6954 		list_for_each(pos, &adapter->vf_macs.l) {
6955 			entry = list_entry(pos, struct vf_mac_filter, l);
6956 			if (entry->vf == vf) {
6957 				entry->vf = -1;
6958 				entry->free = true;
6959 				igb_del_mac_filter(adapter, entry->vf_mac, vf);
6960 			}
6961 		}
6962 		break;
6963 	case E1000_VF_MAC_FILTER_ADD:
6964 		if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
6965 		    !vf_data->trusted) {
6966 			dev_warn(&pdev->dev,
6967 				 "VF %d requested MAC filter but is administratively denied\n",
6968 				 vf);
6969 			return -EINVAL;
6970 		}
6971 		if (!is_valid_ether_addr(addr)) {
6972 			dev_warn(&pdev->dev,
6973 				 "VF %d attempted to set invalid MAC filter\n",
6974 				 vf);
6975 			return -EINVAL;
6976 		}
6977 
6978 		/* try to find empty slot in the list */
6979 		list_for_each(pos, &adapter->vf_macs.l) {
6980 			entry = list_entry(pos, struct vf_mac_filter, l);
6981 			if (entry->free)
6982 				break;
6983 		}
6984 
6985 		if (entry && entry->free) {
6986 			entry->free = false;
6987 			entry->vf = vf;
6988 			ether_addr_copy(entry->vf_mac, addr);
6989 
6990 			ret = igb_add_mac_filter(adapter, addr, vf);
6991 			ret = min_t(int, ret, 0);
6992 		} else {
6993 			ret = -ENOSPC;
6994 		}
6995 
6996 		if (ret == -ENOSPC)
6997 			dev_warn(&pdev->dev,
6998 				 "VF %d has requested MAC filter but there is no space for it\n",
6999 				 vf);
7000 		break;
7001 	default:
7002 		ret = -EINVAL;
7003 		break;
7004 	}
7005 
7006 	return ret;
7007 }
7008 
7009 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7010 {
7011 	struct pci_dev *pdev = adapter->pdev;
7012 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7013 	u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7014 
7015 	/* The VF MAC Address is stored in a packed array of bytes
7016 	 * starting at the second 32 bit word of the msg array
7017 	 */
7018 	unsigned char *addr = (unsigned char *)&msg[1];
7019 	int ret = 0;
7020 
7021 	if (!info) {
7022 		if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7023 		    !vf_data->trusted) {
7024 			dev_warn(&pdev->dev,
7025 				 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7026 				 vf);
7027 			return -EINVAL;
7028 		}
7029 
7030 		if (!is_valid_ether_addr(addr)) {
7031 			dev_warn(&pdev->dev,
7032 				 "VF %d attempted to set invalid MAC\n",
7033 				 vf);
7034 			return -EINVAL;
7035 		}
7036 
7037 		ret = igb_set_vf_mac(adapter, vf, addr);
7038 	} else {
7039 		ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7040 	}
7041 
7042 	return ret;
7043 }
7044 
7045 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7046 {
7047 	struct e1000_hw *hw = &adapter->hw;
7048 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7049 	u32 msg = E1000_VT_MSGTYPE_NACK;
7050 
7051 	/* if device isn't clear to send it shouldn't be reading either */
7052 	if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7053 	    time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7054 		igb_write_mbx(hw, &msg, 1, vf);
7055 		vf_data->last_nack = jiffies;
7056 	}
7057 }
7058 
7059 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7060 {
7061 	struct pci_dev *pdev = adapter->pdev;
7062 	u32 msgbuf[E1000_VFMAILBOX_SIZE];
7063 	struct e1000_hw *hw = &adapter->hw;
7064 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7065 	s32 retval;
7066 
7067 	retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7068 
7069 	if (retval) {
7070 		/* if receive failed revoke VF CTS stats and restart init */
7071 		dev_err(&pdev->dev, "Error receiving message from VF\n");
7072 		vf_data->flags &= ~IGB_VF_FLAG_CTS;
7073 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7074 			goto unlock;
7075 		goto out;
7076 	}
7077 
7078 	/* this is a message we already processed, do nothing */
7079 	if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7080 		goto unlock;
7081 
7082 	/* until the vf completes a reset it should not be
7083 	 * allowed to start any configuration.
7084 	 */
7085 	if (msgbuf[0] == E1000_VF_RESET) {
7086 		/* unlocks mailbox */
7087 		igb_vf_reset_msg(adapter, vf);
7088 		return;
7089 	}
7090 
7091 	if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7092 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7093 			goto unlock;
7094 		retval = -1;
7095 		goto out;
7096 	}
7097 
7098 	switch ((msgbuf[0] & 0xFFFF)) {
7099 	case E1000_VF_SET_MAC_ADDR:
7100 		retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7101 		break;
7102 	case E1000_VF_SET_PROMISC:
7103 		retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7104 		break;
7105 	case E1000_VF_SET_MULTICAST:
7106 		retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7107 		break;
7108 	case E1000_VF_SET_LPE:
7109 		retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7110 		break;
7111 	case E1000_VF_SET_VLAN:
7112 		retval = -1;
7113 		if (vf_data->pf_vlan)
7114 			dev_warn(&pdev->dev,
7115 				 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7116 				 vf);
7117 		else
7118 			retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7119 		break;
7120 	default:
7121 		dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7122 		retval = -1;
7123 		break;
7124 	}
7125 
7126 	msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7127 out:
7128 	/* notify the VF of the results of what it sent us */
7129 	if (retval)
7130 		msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7131 	else
7132 		msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7133 
7134 	/* unlocks mailbox */
7135 	igb_write_mbx(hw, msgbuf, 1, vf);
7136 	return;
7137 
7138 unlock:
7139 	igb_unlock_mbx(hw, vf);
7140 }
7141 
7142 static void igb_msg_task(struct igb_adapter *adapter)
7143 {
7144 	struct e1000_hw *hw = &adapter->hw;
7145 	u32 vf;
7146 
7147 	for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7148 		/* process any reset requests */
7149 		if (!igb_check_for_rst(hw, vf))
7150 			igb_vf_reset_event(adapter, vf);
7151 
7152 		/* process any messages pending */
7153 		if (!igb_check_for_msg(hw, vf))
7154 			igb_rcv_msg_from_vf(adapter, vf);
7155 
7156 		/* process any acks */
7157 		if (!igb_check_for_ack(hw, vf))
7158 			igb_rcv_ack_from_vf(adapter, vf);
7159 	}
7160 }
7161 
7162 /**
7163  *  igb_set_uta - Set unicast filter table address
7164  *  @adapter: board private structure
7165  *  @set: boolean indicating if we are setting or clearing bits
7166  *
7167  *  The unicast table address is a register array of 32-bit registers.
7168  *  The table is meant to be used in a way similar to how the MTA is used
7169  *  however due to certain limitations in the hardware it is necessary to
7170  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7171  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
7172  **/
7173 static void igb_set_uta(struct igb_adapter *adapter, bool set)
7174 {
7175 	struct e1000_hw *hw = &adapter->hw;
7176 	u32 uta = set ? ~0 : 0;
7177 	int i;
7178 
7179 	/* we only need to do this if VMDq is enabled */
7180 	if (!adapter->vfs_allocated_count)
7181 		return;
7182 
7183 	for (i = hw->mac.uta_reg_count; i--;)
7184 		array_wr32(E1000_UTA, i, uta);
7185 }
7186 
7187 /**
7188  *  igb_intr_msi - Interrupt Handler
7189  *  @irq: interrupt number
7190  *  @data: pointer to a network interface device structure
7191  **/
7192 static irqreturn_t igb_intr_msi(int irq, void *data)
7193 {
7194 	struct igb_adapter *adapter = data;
7195 	struct igb_q_vector *q_vector = adapter->q_vector[0];
7196 	struct e1000_hw *hw = &adapter->hw;
7197 	/* read ICR disables interrupts using IAM */
7198 	u32 icr = rd32(E1000_ICR);
7199 
7200 	igb_write_itr(q_vector);
7201 
7202 	if (icr & E1000_ICR_DRSTA)
7203 		schedule_work(&adapter->reset_task);
7204 
7205 	if (icr & E1000_ICR_DOUTSYNC) {
7206 		/* HW is reporting DMA is out of sync */
7207 		adapter->stats.doosync++;
7208 	}
7209 
7210 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7211 		hw->mac.get_link_status = 1;
7212 		if (!test_bit(__IGB_DOWN, &adapter->state))
7213 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
7214 	}
7215 
7216 	if (icr & E1000_ICR_TS)
7217 		igb_tsync_interrupt(adapter);
7218 
7219 	napi_schedule(&q_vector->napi);
7220 
7221 	return IRQ_HANDLED;
7222 }
7223 
7224 /**
7225  *  igb_intr - Legacy Interrupt Handler
7226  *  @irq: interrupt number
7227  *  @data: pointer to a network interface device structure
7228  **/
7229 static irqreturn_t igb_intr(int irq, void *data)
7230 {
7231 	struct igb_adapter *adapter = data;
7232 	struct igb_q_vector *q_vector = adapter->q_vector[0];
7233 	struct e1000_hw *hw = &adapter->hw;
7234 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
7235 	 * need for the IMC write
7236 	 */
7237 	u32 icr = rd32(E1000_ICR);
7238 
7239 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
7240 	 * not set, then the adapter didn't send an interrupt
7241 	 */
7242 	if (!(icr & E1000_ICR_INT_ASSERTED))
7243 		return IRQ_NONE;
7244 
7245 	igb_write_itr(q_vector);
7246 
7247 	if (icr & E1000_ICR_DRSTA)
7248 		schedule_work(&adapter->reset_task);
7249 
7250 	if (icr & E1000_ICR_DOUTSYNC) {
7251 		/* HW is reporting DMA is out of sync */
7252 		adapter->stats.doosync++;
7253 	}
7254 
7255 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7256 		hw->mac.get_link_status = 1;
7257 		/* guard against interrupt when we're going down */
7258 		if (!test_bit(__IGB_DOWN, &adapter->state))
7259 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
7260 	}
7261 
7262 	if (icr & E1000_ICR_TS)
7263 		igb_tsync_interrupt(adapter);
7264 
7265 	napi_schedule(&q_vector->napi);
7266 
7267 	return IRQ_HANDLED;
7268 }
7269 
7270 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
7271 {
7272 	struct igb_adapter *adapter = q_vector->adapter;
7273 	struct e1000_hw *hw = &adapter->hw;
7274 
7275 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
7276 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
7277 		if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
7278 			igb_set_itr(q_vector);
7279 		else
7280 			igb_update_ring_itr(q_vector);
7281 	}
7282 
7283 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
7284 		if (adapter->flags & IGB_FLAG_HAS_MSIX)
7285 			wr32(E1000_EIMS, q_vector->eims_value);
7286 		else
7287 			igb_irq_enable(adapter);
7288 	}
7289 }
7290 
7291 /**
7292  *  igb_poll - NAPI Rx polling callback
7293  *  @napi: napi polling structure
7294  *  @budget: count of how many packets we should handle
7295  **/
7296 static int igb_poll(struct napi_struct *napi, int budget)
7297 {
7298 	struct igb_q_vector *q_vector = container_of(napi,
7299 						     struct igb_q_vector,
7300 						     napi);
7301 	bool clean_complete = true;
7302 	int work_done = 0;
7303 
7304 #ifdef CONFIG_IGB_DCA
7305 	if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
7306 		igb_update_dca(q_vector);
7307 #endif
7308 	if (q_vector->tx.ring)
7309 		clean_complete = igb_clean_tx_irq(q_vector, budget);
7310 
7311 	if (q_vector->rx.ring) {
7312 		int cleaned = igb_clean_rx_irq(q_vector, budget);
7313 
7314 		work_done += cleaned;
7315 		if (cleaned >= budget)
7316 			clean_complete = false;
7317 	}
7318 
7319 	/* If all work not completed, return budget and keep polling */
7320 	if (!clean_complete)
7321 		return budget;
7322 
7323 	/* If not enough Rx work done, exit the polling mode */
7324 	napi_complete_done(napi, work_done);
7325 	igb_ring_irq_enable(q_vector);
7326 
7327 	return 0;
7328 }
7329 
7330 /**
7331  *  igb_clean_tx_irq - Reclaim resources after transmit completes
7332  *  @q_vector: pointer to q_vector containing needed info
7333  *  @napi_budget: Used to determine if we are in netpoll
7334  *
7335  *  returns true if ring is completely cleaned
7336  **/
7337 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
7338 {
7339 	struct igb_adapter *adapter = q_vector->adapter;
7340 	struct igb_ring *tx_ring = q_vector->tx.ring;
7341 	struct igb_tx_buffer *tx_buffer;
7342 	union e1000_adv_tx_desc *tx_desc;
7343 	unsigned int total_bytes = 0, total_packets = 0;
7344 	unsigned int budget = q_vector->tx.work_limit;
7345 	unsigned int i = tx_ring->next_to_clean;
7346 
7347 	if (test_bit(__IGB_DOWN, &adapter->state))
7348 		return true;
7349 
7350 	tx_buffer = &tx_ring->tx_buffer_info[i];
7351 	tx_desc = IGB_TX_DESC(tx_ring, i);
7352 	i -= tx_ring->count;
7353 
7354 	do {
7355 		union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
7356 
7357 		/* if next_to_watch is not set then there is no work pending */
7358 		if (!eop_desc)
7359 			break;
7360 
7361 		/* prevent any other reads prior to eop_desc */
7362 		smp_rmb();
7363 
7364 		/* if DD is not set pending work has not been completed */
7365 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
7366 			break;
7367 
7368 		/* clear next_to_watch to prevent false hangs */
7369 		tx_buffer->next_to_watch = NULL;
7370 
7371 		/* update the statistics for this packet */
7372 		total_bytes += tx_buffer->bytecount;
7373 		total_packets += tx_buffer->gso_segs;
7374 
7375 		/* free the skb */
7376 		napi_consume_skb(tx_buffer->skb, napi_budget);
7377 
7378 		/* unmap skb header data */
7379 		dma_unmap_single(tx_ring->dev,
7380 				 dma_unmap_addr(tx_buffer, dma),
7381 				 dma_unmap_len(tx_buffer, len),
7382 				 DMA_TO_DEVICE);
7383 
7384 		/* clear tx_buffer data */
7385 		dma_unmap_len_set(tx_buffer, len, 0);
7386 
7387 		/* clear last DMA location and unmap remaining buffers */
7388 		while (tx_desc != eop_desc) {
7389 			tx_buffer++;
7390 			tx_desc++;
7391 			i++;
7392 			if (unlikely(!i)) {
7393 				i -= tx_ring->count;
7394 				tx_buffer = tx_ring->tx_buffer_info;
7395 				tx_desc = IGB_TX_DESC(tx_ring, 0);
7396 			}
7397 
7398 			/* unmap any remaining paged data */
7399 			if (dma_unmap_len(tx_buffer, len)) {
7400 				dma_unmap_page(tx_ring->dev,
7401 					       dma_unmap_addr(tx_buffer, dma),
7402 					       dma_unmap_len(tx_buffer, len),
7403 					       DMA_TO_DEVICE);
7404 				dma_unmap_len_set(tx_buffer, len, 0);
7405 			}
7406 		}
7407 
7408 		/* move us one more past the eop_desc for start of next pkt */
7409 		tx_buffer++;
7410 		tx_desc++;
7411 		i++;
7412 		if (unlikely(!i)) {
7413 			i -= tx_ring->count;
7414 			tx_buffer = tx_ring->tx_buffer_info;
7415 			tx_desc = IGB_TX_DESC(tx_ring, 0);
7416 		}
7417 
7418 		/* issue prefetch for next Tx descriptor */
7419 		prefetch(tx_desc);
7420 
7421 		/* update budget accounting */
7422 		budget--;
7423 	} while (likely(budget));
7424 
7425 	netdev_tx_completed_queue(txring_txq(tx_ring),
7426 				  total_packets, total_bytes);
7427 	i += tx_ring->count;
7428 	tx_ring->next_to_clean = i;
7429 	u64_stats_update_begin(&tx_ring->tx_syncp);
7430 	tx_ring->tx_stats.bytes += total_bytes;
7431 	tx_ring->tx_stats.packets += total_packets;
7432 	u64_stats_update_end(&tx_ring->tx_syncp);
7433 	q_vector->tx.total_bytes += total_bytes;
7434 	q_vector->tx.total_packets += total_packets;
7435 
7436 	if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
7437 		struct e1000_hw *hw = &adapter->hw;
7438 
7439 		/* Detect a transmit hang in hardware, this serializes the
7440 		 * check with the clearing of time_stamp and movement of i
7441 		 */
7442 		clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
7443 		if (tx_buffer->next_to_watch &&
7444 		    time_after(jiffies, tx_buffer->time_stamp +
7445 			       (adapter->tx_timeout_factor * HZ)) &&
7446 		    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
7447 
7448 			/* detected Tx unit hang */
7449 			dev_err(tx_ring->dev,
7450 				"Detected Tx Unit Hang\n"
7451 				"  Tx Queue             <%d>\n"
7452 				"  TDH                  <%x>\n"
7453 				"  TDT                  <%x>\n"
7454 				"  next_to_use          <%x>\n"
7455 				"  next_to_clean        <%x>\n"
7456 				"buffer_info[next_to_clean]\n"
7457 				"  time_stamp           <%lx>\n"
7458 				"  next_to_watch        <%p>\n"
7459 				"  jiffies              <%lx>\n"
7460 				"  desc.status          <%x>\n",
7461 				tx_ring->queue_index,
7462 				rd32(E1000_TDH(tx_ring->reg_idx)),
7463 				readl(tx_ring->tail),
7464 				tx_ring->next_to_use,
7465 				tx_ring->next_to_clean,
7466 				tx_buffer->time_stamp,
7467 				tx_buffer->next_to_watch,
7468 				jiffies,
7469 				tx_buffer->next_to_watch->wb.status);
7470 			netif_stop_subqueue(tx_ring->netdev,
7471 					    tx_ring->queue_index);
7472 
7473 			/* we are about to reset, no point in enabling stuff */
7474 			return true;
7475 		}
7476 	}
7477 
7478 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
7479 	if (unlikely(total_packets &&
7480 	    netif_carrier_ok(tx_ring->netdev) &&
7481 	    igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
7482 		/* Make sure that anybody stopping the queue after this
7483 		 * sees the new next_to_clean.
7484 		 */
7485 		smp_mb();
7486 		if (__netif_subqueue_stopped(tx_ring->netdev,
7487 					     tx_ring->queue_index) &&
7488 		    !(test_bit(__IGB_DOWN, &adapter->state))) {
7489 			netif_wake_subqueue(tx_ring->netdev,
7490 					    tx_ring->queue_index);
7491 
7492 			u64_stats_update_begin(&tx_ring->tx_syncp);
7493 			tx_ring->tx_stats.restart_queue++;
7494 			u64_stats_update_end(&tx_ring->tx_syncp);
7495 		}
7496 	}
7497 
7498 	return !!budget;
7499 }
7500 
7501 /**
7502  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
7503  *  @rx_ring: rx descriptor ring to store buffers on
7504  *  @old_buff: donor buffer to have page reused
7505  *
7506  *  Synchronizes page for reuse by the adapter
7507  **/
7508 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
7509 			      struct igb_rx_buffer *old_buff)
7510 {
7511 	struct igb_rx_buffer *new_buff;
7512 	u16 nta = rx_ring->next_to_alloc;
7513 
7514 	new_buff = &rx_ring->rx_buffer_info[nta];
7515 
7516 	/* update, and store next to alloc */
7517 	nta++;
7518 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
7519 
7520 	/* Transfer page from old buffer to new buffer.
7521 	 * Move each member individually to avoid possible store
7522 	 * forwarding stalls.
7523 	 */
7524 	new_buff->dma		= old_buff->dma;
7525 	new_buff->page		= old_buff->page;
7526 	new_buff->page_offset	= old_buff->page_offset;
7527 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
7528 }
7529 
7530 static inline bool igb_page_is_reserved(struct page *page)
7531 {
7532 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
7533 }
7534 
7535 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer)
7536 {
7537 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
7538 	struct page *page = rx_buffer->page;
7539 
7540 	/* avoid re-using remote pages */
7541 	if (unlikely(igb_page_is_reserved(page)))
7542 		return false;
7543 
7544 #if (PAGE_SIZE < 8192)
7545 	/* if we are only owner of page we can reuse it */
7546 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
7547 		return false;
7548 #else
7549 #define IGB_LAST_OFFSET \
7550 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
7551 
7552 	if (rx_buffer->page_offset > IGB_LAST_OFFSET)
7553 		return false;
7554 #endif
7555 
7556 	/* If we have drained the page fragment pool we need to update
7557 	 * the pagecnt_bias and page count so that we fully restock the
7558 	 * number of references the driver holds.
7559 	 */
7560 	if (unlikely(!pagecnt_bias)) {
7561 		page_ref_add(page, USHRT_MAX);
7562 		rx_buffer->pagecnt_bias = USHRT_MAX;
7563 	}
7564 
7565 	return true;
7566 }
7567 
7568 /**
7569  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
7570  *  @rx_ring: rx descriptor ring to transact packets on
7571  *  @rx_buffer: buffer containing page to add
7572  *  @skb: sk_buff to place the data into
7573  *  @size: size of buffer to be added
7574  *
7575  *  This function will add the data contained in rx_buffer->page to the skb.
7576  **/
7577 static void igb_add_rx_frag(struct igb_ring *rx_ring,
7578 			    struct igb_rx_buffer *rx_buffer,
7579 			    struct sk_buff *skb,
7580 			    unsigned int size)
7581 {
7582 #if (PAGE_SIZE < 8192)
7583 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7584 #else
7585 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
7586 				SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
7587 				SKB_DATA_ALIGN(size);
7588 #endif
7589 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
7590 			rx_buffer->page_offset, size, truesize);
7591 #if (PAGE_SIZE < 8192)
7592 	rx_buffer->page_offset ^= truesize;
7593 #else
7594 	rx_buffer->page_offset += truesize;
7595 #endif
7596 }
7597 
7598 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
7599 					 struct igb_rx_buffer *rx_buffer,
7600 					 union e1000_adv_rx_desc *rx_desc,
7601 					 unsigned int size)
7602 {
7603 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7604 #if (PAGE_SIZE < 8192)
7605 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7606 #else
7607 	unsigned int truesize = SKB_DATA_ALIGN(size);
7608 #endif
7609 	unsigned int headlen;
7610 	struct sk_buff *skb;
7611 
7612 	/* prefetch first cache line of first page */
7613 	prefetch(va);
7614 #if L1_CACHE_BYTES < 128
7615 	prefetch(va + L1_CACHE_BYTES);
7616 #endif
7617 
7618 	/* allocate a skb to store the frags */
7619 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
7620 	if (unlikely(!skb))
7621 		return NULL;
7622 
7623 	if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
7624 		igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
7625 		va += IGB_TS_HDR_LEN;
7626 		size -= IGB_TS_HDR_LEN;
7627 	}
7628 
7629 	/* Determine available headroom for copy */
7630 	headlen = size;
7631 	if (headlen > IGB_RX_HDR_LEN)
7632 		headlen = eth_get_headlen(va, IGB_RX_HDR_LEN);
7633 
7634 	/* align pull length to size of long to optimize memcpy performance */
7635 	memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
7636 
7637 	/* update all of the pointers */
7638 	size -= headlen;
7639 	if (size) {
7640 		skb_add_rx_frag(skb, 0, rx_buffer->page,
7641 				(va + headlen) - page_address(rx_buffer->page),
7642 				size, truesize);
7643 #if (PAGE_SIZE < 8192)
7644 		rx_buffer->page_offset ^= truesize;
7645 #else
7646 		rx_buffer->page_offset += truesize;
7647 #endif
7648 	} else {
7649 		rx_buffer->pagecnt_bias++;
7650 	}
7651 
7652 	return skb;
7653 }
7654 
7655 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
7656 				     struct igb_rx_buffer *rx_buffer,
7657 				     union e1000_adv_rx_desc *rx_desc,
7658 				     unsigned int size)
7659 {
7660 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7661 #if (PAGE_SIZE < 8192)
7662 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7663 #else
7664 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
7665 				SKB_DATA_ALIGN(IGB_SKB_PAD + size);
7666 #endif
7667 	struct sk_buff *skb;
7668 
7669 	/* prefetch first cache line of first page */
7670 	prefetch(va);
7671 #if L1_CACHE_BYTES < 128
7672 	prefetch(va + L1_CACHE_BYTES);
7673 #endif
7674 
7675 	/* build an skb around the page buffer */
7676 	skb = build_skb(va - IGB_SKB_PAD, truesize);
7677 	if (unlikely(!skb))
7678 		return NULL;
7679 
7680 	/* update pointers within the skb to store the data */
7681 	skb_reserve(skb, IGB_SKB_PAD);
7682 	__skb_put(skb, size);
7683 
7684 	/* pull timestamp out of packet data */
7685 	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
7686 		igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
7687 		__skb_pull(skb, IGB_TS_HDR_LEN);
7688 	}
7689 
7690 	/* update buffer offset */
7691 #if (PAGE_SIZE < 8192)
7692 	rx_buffer->page_offset ^= truesize;
7693 #else
7694 	rx_buffer->page_offset += truesize;
7695 #endif
7696 
7697 	return skb;
7698 }
7699 
7700 static inline void igb_rx_checksum(struct igb_ring *ring,
7701 				   union e1000_adv_rx_desc *rx_desc,
7702 				   struct sk_buff *skb)
7703 {
7704 	skb_checksum_none_assert(skb);
7705 
7706 	/* Ignore Checksum bit is set */
7707 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
7708 		return;
7709 
7710 	/* Rx checksum disabled via ethtool */
7711 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
7712 		return;
7713 
7714 	/* TCP/UDP checksum error bit is set */
7715 	if (igb_test_staterr(rx_desc,
7716 			     E1000_RXDEXT_STATERR_TCPE |
7717 			     E1000_RXDEXT_STATERR_IPE)) {
7718 		/* work around errata with sctp packets where the TCPE aka
7719 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
7720 		 * packets, (aka let the stack check the crc32c)
7721 		 */
7722 		if (!((skb->len == 60) &&
7723 		      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
7724 			u64_stats_update_begin(&ring->rx_syncp);
7725 			ring->rx_stats.csum_err++;
7726 			u64_stats_update_end(&ring->rx_syncp);
7727 		}
7728 		/* let the stack verify checksum errors */
7729 		return;
7730 	}
7731 	/* It must be a TCP or UDP packet with a valid checksum */
7732 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
7733 				      E1000_RXD_STAT_UDPCS))
7734 		skb->ip_summed = CHECKSUM_UNNECESSARY;
7735 
7736 	dev_dbg(ring->dev, "cksum success: bits %08X\n",
7737 		le32_to_cpu(rx_desc->wb.upper.status_error));
7738 }
7739 
7740 static inline void igb_rx_hash(struct igb_ring *ring,
7741 			       union e1000_adv_rx_desc *rx_desc,
7742 			       struct sk_buff *skb)
7743 {
7744 	if (ring->netdev->features & NETIF_F_RXHASH)
7745 		skb_set_hash(skb,
7746 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
7747 			     PKT_HASH_TYPE_L3);
7748 }
7749 
7750 /**
7751  *  igb_is_non_eop - process handling of non-EOP buffers
7752  *  @rx_ring: Rx ring being processed
7753  *  @rx_desc: Rx descriptor for current buffer
7754  *  @skb: current socket buffer containing buffer in progress
7755  *
7756  *  This function updates next to clean.  If the buffer is an EOP buffer
7757  *  this function exits returning false, otherwise it will place the
7758  *  sk_buff in the next buffer to be chained and return true indicating
7759  *  that this is in fact a non-EOP buffer.
7760  **/
7761 static bool igb_is_non_eop(struct igb_ring *rx_ring,
7762 			   union e1000_adv_rx_desc *rx_desc)
7763 {
7764 	u32 ntc = rx_ring->next_to_clean + 1;
7765 
7766 	/* fetch, update, and store next to clean */
7767 	ntc = (ntc < rx_ring->count) ? ntc : 0;
7768 	rx_ring->next_to_clean = ntc;
7769 
7770 	prefetch(IGB_RX_DESC(rx_ring, ntc));
7771 
7772 	if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
7773 		return false;
7774 
7775 	return true;
7776 }
7777 
7778 /**
7779  *  igb_cleanup_headers - Correct corrupted or empty headers
7780  *  @rx_ring: rx descriptor ring packet is being transacted on
7781  *  @rx_desc: pointer to the EOP Rx descriptor
7782  *  @skb: pointer to current skb being fixed
7783  *
7784  *  Address the case where we are pulling data in on pages only
7785  *  and as such no data is present in the skb header.
7786  *
7787  *  In addition if skb is not at least 60 bytes we need to pad it so that
7788  *  it is large enough to qualify as a valid Ethernet frame.
7789  *
7790  *  Returns true if an error was encountered and skb was freed.
7791  **/
7792 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
7793 				union e1000_adv_rx_desc *rx_desc,
7794 				struct sk_buff *skb)
7795 {
7796 	if (unlikely((igb_test_staterr(rx_desc,
7797 				       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
7798 		struct net_device *netdev = rx_ring->netdev;
7799 		if (!(netdev->features & NETIF_F_RXALL)) {
7800 			dev_kfree_skb_any(skb);
7801 			return true;
7802 		}
7803 	}
7804 
7805 	/* if eth_skb_pad returns an error the skb was freed */
7806 	if (eth_skb_pad(skb))
7807 		return true;
7808 
7809 	return false;
7810 }
7811 
7812 /**
7813  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
7814  *  @rx_ring: rx descriptor ring packet is being transacted on
7815  *  @rx_desc: pointer to the EOP Rx descriptor
7816  *  @skb: pointer to current skb being populated
7817  *
7818  *  This function checks the ring, descriptor, and packet information in
7819  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
7820  *  other fields within the skb.
7821  **/
7822 static void igb_process_skb_fields(struct igb_ring *rx_ring,
7823 				   union e1000_adv_rx_desc *rx_desc,
7824 				   struct sk_buff *skb)
7825 {
7826 	struct net_device *dev = rx_ring->netdev;
7827 
7828 	igb_rx_hash(rx_ring, rx_desc, skb);
7829 
7830 	igb_rx_checksum(rx_ring, rx_desc, skb);
7831 
7832 	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
7833 	    !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
7834 		igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
7835 
7836 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
7837 	    igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
7838 		u16 vid;
7839 
7840 		if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
7841 		    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
7842 			vid = be16_to_cpu(rx_desc->wb.upper.vlan);
7843 		else
7844 			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
7845 
7846 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
7847 	}
7848 
7849 	skb_record_rx_queue(skb, rx_ring->queue_index);
7850 
7851 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
7852 }
7853 
7854 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
7855 					       const unsigned int size)
7856 {
7857 	struct igb_rx_buffer *rx_buffer;
7858 
7859 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
7860 	prefetchw(rx_buffer->page);
7861 
7862 	/* we are reusing so sync this buffer for CPU use */
7863 	dma_sync_single_range_for_cpu(rx_ring->dev,
7864 				      rx_buffer->dma,
7865 				      rx_buffer->page_offset,
7866 				      size,
7867 				      DMA_FROM_DEVICE);
7868 
7869 	rx_buffer->pagecnt_bias--;
7870 
7871 	return rx_buffer;
7872 }
7873 
7874 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
7875 			      struct igb_rx_buffer *rx_buffer)
7876 {
7877 	if (igb_can_reuse_rx_page(rx_buffer)) {
7878 		/* hand second half of page back to the ring */
7879 		igb_reuse_rx_page(rx_ring, rx_buffer);
7880 	} else {
7881 		/* We are not reusing the buffer so unmap it and free
7882 		 * any references we are holding to it
7883 		 */
7884 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
7885 				     igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
7886 				     IGB_RX_DMA_ATTR);
7887 		__page_frag_cache_drain(rx_buffer->page,
7888 					rx_buffer->pagecnt_bias);
7889 	}
7890 
7891 	/* clear contents of rx_buffer */
7892 	rx_buffer->page = NULL;
7893 }
7894 
7895 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
7896 {
7897 	struct igb_ring *rx_ring = q_vector->rx.ring;
7898 	struct sk_buff *skb = rx_ring->skb;
7899 	unsigned int total_bytes = 0, total_packets = 0;
7900 	u16 cleaned_count = igb_desc_unused(rx_ring);
7901 
7902 	while (likely(total_packets < budget)) {
7903 		union e1000_adv_rx_desc *rx_desc;
7904 		struct igb_rx_buffer *rx_buffer;
7905 		unsigned int size;
7906 
7907 		/* return some buffers to hardware, one at a time is too slow */
7908 		if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
7909 			igb_alloc_rx_buffers(rx_ring, cleaned_count);
7910 			cleaned_count = 0;
7911 		}
7912 
7913 		rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
7914 		size = le16_to_cpu(rx_desc->wb.upper.length);
7915 		if (!size)
7916 			break;
7917 
7918 		/* This memory barrier is needed to keep us from reading
7919 		 * any other fields out of the rx_desc until we know the
7920 		 * descriptor has been written back
7921 		 */
7922 		dma_rmb();
7923 
7924 		rx_buffer = igb_get_rx_buffer(rx_ring, size);
7925 
7926 		/* retrieve a buffer from the ring */
7927 		if (skb)
7928 			igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
7929 		else if (ring_uses_build_skb(rx_ring))
7930 			skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size);
7931 		else
7932 			skb = igb_construct_skb(rx_ring, rx_buffer,
7933 						rx_desc, size);
7934 
7935 		/* exit if we failed to retrieve a buffer */
7936 		if (!skb) {
7937 			rx_ring->rx_stats.alloc_failed++;
7938 			rx_buffer->pagecnt_bias++;
7939 			break;
7940 		}
7941 
7942 		igb_put_rx_buffer(rx_ring, rx_buffer);
7943 		cleaned_count++;
7944 
7945 		/* fetch next buffer in frame if non-eop */
7946 		if (igb_is_non_eop(rx_ring, rx_desc))
7947 			continue;
7948 
7949 		/* verify the packet layout is correct */
7950 		if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
7951 			skb = NULL;
7952 			continue;
7953 		}
7954 
7955 		/* probably a little skewed due to removing CRC */
7956 		total_bytes += skb->len;
7957 
7958 		/* populate checksum, timestamp, VLAN, and protocol */
7959 		igb_process_skb_fields(rx_ring, rx_desc, skb);
7960 
7961 		napi_gro_receive(&q_vector->napi, skb);
7962 
7963 		/* reset skb pointer */
7964 		skb = NULL;
7965 
7966 		/* update budget accounting */
7967 		total_packets++;
7968 	}
7969 
7970 	/* place incomplete frames back on ring for completion */
7971 	rx_ring->skb = skb;
7972 
7973 	u64_stats_update_begin(&rx_ring->rx_syncp);
7974 	rx_ring->rx_stats.packets += total_packets;
7975 	rx_ring->rx_stats.bytes += total_bytes;
7976 	u64_stats_update_end(&rx_ring->rx_syncp);
7977 	q_vector->rx.total_packets += total_packets;
7978 	q_vector->rx.total_bytes += total_bytes;
7979 
7980 	if (cleaned_count)
7981 		igb_alloc_rx_buffers(rx_ring, cleaned_count);
7982 
7983 	return total_packets;
7984 }
7985 
7986 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
7987 {
7988 	return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
7989 }
7990 
7991 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
7992 				  struct igb_rx_buffer *bi)
7993 {
7994 	struct page *page = bi->page;
7995 	dma_addr_t dma;
7996 
7997 	/* since we are recycling buffers we should seldom need to alloc */
7998 	if (likely(page))
7999 		return true;
8000 
8001 	/* alloc new page for storage */
8002 	page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
8003 	if (unlikely(!page)) {
8004 		rx_ring->rx_stats.alloc_failed++;
8005 		return false;
8006 	}
8007 
8008 	/* map page for use */
8009 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
8010 				 igb_rx_pg_size(rx_ring),
8011 				 DMA_FROM_DEVICE,
8012 				 IGB_RX_DMA_ATTR);
8013 
8014 	/* if mapping failed free memory back to system since
8015 	 * there isn't much point in holding memory we can't use
8016 	 */
8017 	if (dma_mapping_error(rx_ring->dev, dma)) {
8018 		__free_pages(page, igb_rx_pg_order(rx_ring));
8019 
8020 		rx_ring->rx_stats.alloc_failed++;
8021 		return false;
8022 	}
8023 
8024 	bi->dma = dma;
8025 	bi->page = page;
8026 	bi->page_offset = igb_rx_offset(rx_ring);
8027 	bi->pagecnt_bias = 1;
8028 
8029 	return true;
8030 }
8031 
8032 /**
8033  *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
8034  *  @adapter: address of board private structure
8035  **/
8036 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
8037 {
8038 	union e1000_adv_rx_desc *rx_desc;
8039 	struct igb_rx_buffer *bi;
8040 	u16 i = rx_ring->next_to_use;
8041 	u16 bufsz;
8042 
8043 	/* nothing to do */
8044 	if (!cleaned_count)
8045 		return;
8046 
8047 	rx_desc = IGB_RX_DESC(rx_ring, i);
8048 	bi = &rx_ring->rx_buffer_info[i];
8049 	i -= rx_ring->count;
8050 
8051 	bufsz = igb_rx_bufsz(rx_ring);
8052 
8053 	do {
8054 		if (!igb_alloc_mapped_page(rx_ring, bi))
8055 			break;
8056 
8057 		/* sync the buffer for use by the device */
8058 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
8059 						 bi->page_offset, bufsz,
8060 						 DMA_FROM_DEVICE);
8061 
8062 		/* Refresh the desc even if buffer_addrs didn't change
8063 		 * because each write-back erases this info.
8064 		 */
8065 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
8066 
8067 		rx_desc++;
8068 		bi++;
8069 		i++;
8070 		if (unlikely(!i)) {
8071 			rx_desc = IGB_RX_DESC(rx_ring, 0);
8072 			bi = rx_ring->rx_buffer_info;
8073 			i -= rx_ring->count;
8074 		}
8075 
8076 		/* clear the length for the next_to_use descriptor */
8077 		rx_desc->wb.upper.length = 0;
8078 
8079 		cleaned_count--;
8080 	} while (cleaned_count);
8081 
8082 	i += rx_ring->count;
8083 
8084 	if (rx_ring->next_to_use != i) {
8085 		/* record the next descriptor to use */
8086 		rx_ring->next_to_use = i;
8087 
8088 		/* update next to alloc since we have filled the ring */
8089 		rx_ring->next_to_alloc = i;
8090 
8091 		/* Force memory writes to complete before letting h/w
8092 		 * know there are new descriptors to fetch.  (Only
8093 		 * applicable for weak-ordered memory model archs,
8094 		 * such as IA-64).
8095 		 */
8096 		wmb();
8097 		writel(i, rx_ring->tail);
8098 	}
8099 }
8100 
8101 /**
8102  * igb_mii_ioctl -
8103  * @netdev:
8104  * @ifreq:
8105  * @cmd:
8106  **/
8107 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8108 {
8109 	struct igb_adapter *adapter = netdev_priv(netdev);
8110 	struct mii_ioctl_data *data = if_mii(ifr);
8111 
8112 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
8113 		return -EOPNOTSUPP;
8114 
8115 	switch (cmd) {
8116 	case SIOCGMIIPHY:
8117 		data->phy_id = adapter->hw.phy.addr;
8118 		break;
8119 	case SIOCGMIIREG:
8120 		if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
8121 				     &data->val_out))
8122 			return -EIO;
8123 		break;
8124 	case SIOCSMIIREG:
8125 	default:
8126 		return -EOPNOTSUPP;
8127 	}
8128 	return 0;
8129 }
8130 
8131 /**
8132  * igb_ioctl -
8133  * @netdev:
8134  * @ifreq:
8135  * @cmd:
8136  **/
8137 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8138 {
8139 	switch (cmd) {
8140 	case SIOCGMIIPHY:
8141 	case SIOCGMIIREG:
8142 	case SIOCSMIIREG:
8143 		return igb_mii_ioctl(netdev, ifr, cmd);
8144 	case SIOCGHWTSTAMP:
8145 		return igb_ptp_get_ts_config(netdev, ifr);
8146 	case SIOCSHWTSTAMP:
8147 		return igb_ptp_set_ts_config(netdev, ifr);
8148 	default:
8149 		return -EOPNOTSUPP;
8150 	}
8151 }
8152 
8153 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8154 {
8155 	struct igb_adapter *adapter = hw->back;
8156 
8157 	pci_read_config_word(adapter->pdev, reg, value);
8158 }
8159 
8160 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8161 {
8162 	struct igb_adapter *adapter = hw->back;
8163 
8164 	pci_write_config_word(adapter->pdev, reg, *value);
8165 }
8166 
8167 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8168 {
8169 	struct igb_adapter *adapter = hw->back;
8170 
8171 	if (pcie_capability_read_word(adapter->pdev, reg, value))
8172 		return -E1000_ERR_CONFIG;
8173 
8174 	return 0;
8175 }
8176 
8177 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8178 {
8179 	struct igb_adapter *adapter = hw->back;
8180 
8181 	if (pcie_capability_write_word(adapter->pdev, reg, *value))
8182 		return -E1000_ERR_CONFIG;
8183 
8184 	return 0;
8185 }
8186 
8187 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
8188 {
8189 	struct igb_adapter *adapter = netdev_priv(netdev);
8190 	struct e1000_hw *hw = &adapter->hw;
8191 	u32 ctrl, rctl;
8192 	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
8193 
8194 	if (enable) {
8195 		/* enable VLAN tag insert/strip */
8196 		ctrl = rd32(E1000_CTRL);
8197 		ctrl |= E1000_CTRL_VME;
8198 		wr32(E1000_CTRL, ctrl);
8199 
8200 		/* Disable CFI check */
8201 		rctl = rd32(E1000_RCTL);
8202 		rctl &= ~E1000_RCTL_CFIEN;
8203 		wr32(E1000_RCTL, rctl);
8204 	} else {
8205 		/* disable VLAN tag insert/strip */
8206 		ctrl = rd32(E1000_CTRL);
8207 		ctrl &= ~E1000_CTRL_VME;
8208 		wr32(E1000_CTRL, ctrl);
8209 	}
8210 
8211 	igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
8212 }
8213 
8214 static int igb_vlan_rx_add_vid(struct net_device *netdev,
8215 			       __be16 proto, u16 vid)
8216 {
8217 	struct igb_adapter *adapter = netdev_priv(netdev);
8218 	struct e1000_hw *hw = &adapter->hw;
8219 	int pf_id = adapter->vfs_allocated_count;
8220 
8221 	/* add the filter since PF can receive vlans w/o entry in vlvf */
8222 	if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8223 		igb_vfta_set(hw, vid, pf_id, true, !!vid);
8224 
8225 	set_bit(vid, adapter->active_vlans);
8226 
8227 	return 0;
8228 }
8229 
8230 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
8231 				__be16 proto, u16 vid)
8232 {
8233 	struct igb_adapter *adapter = netdev_priv(netdev);
8234 	int pf_id = adapter->vfs_allocated_count;
8235 	struct e1000_hw *hw = &adapter->hw;
8236 
8237 	/* remove VID from filter table */
8238 	if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8239 		igb_vfta_set(hw, vid, pf_id, false, true);
8240 
8241 	clear_bit(vid, adapter->active_vlans);
8242 
8243 	return 0;
8244 }
8245 
8246 static void igb_restore_vlan(struct igb_adapter *adapter)
8247 {
8248 	u16 vid = 1;
8249 
8250 	igb_vlan_mode(adapter->netdev, adapter->netdev->features);
8251 	igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
8252 
8253 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
8254 		igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
8255 }
8256 
8257 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
8258 {
8259 	struct pci_dev *pdev = adapter->pdev;
8260 	struct e1000_mac_info *mac = &adapter->hw.mac;
8261 
8262 	mac->autoneg = 0;
8263 
8264 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
8265 	 * for the switch() below to work
8266 	 */
8267 	if ((spd & 1) || (dplx & ~1))
8268 		goto err_inval;
8269 
8270 	/* Fiber NIC's only allow 1000 gbps Full duplex
8271 	 * and 100Mbps Full duplex for 100baseFx sfp
8272 	 */
8273 	if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
8274 		switch (spd + dplx) {
8275 		case SPEED_10 + DUPLEX_HALF:
8276 		case SPEED_10 + DUPLEX_FULL:
8277 		case SPEED_100 + DUPLEX_HALF:
8278 			goto err_inval;
8279 		default:
8280 			break;
8281 		}
8282 	}
8283 
8284 	switch (spd + dplx) {
8285 	case SPEED_10 + DUPLEX_HALF:
8286 		mac->forced_speed_duplex = ADVERTISE_10_HALF;
8287 		break;
8288 	case SPEED_10 + DUPLEX_FULL:
8289 		mac->forced_speed_duplex = ADVERTISE_10_FULL;
8290 		break;
8291 	case SPEED_100 + DUPLEX_HALF:
8292 		mac->forced_speed_duplex = ADVERTISE_100_HALF;
8293 		break;
8294 	case SPEED_100 + DUPLEX_FULL:
8295 		mac->forced_speed_duplex = ADVERTISE_100_FULL;
8296 		break;
8297 	case SPEED_1000 + DUPLEX_FULL:
8298 		mac->autoneg = 1;
8299 		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
8300 		break;
8301 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
8302 	default:
8303 		goto err_inval;
8304 	}
8305 
8306 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
8307 	adapter->hw.phy.mdix = AUTO_ALL_MODES;
8308 
8309 	return 0;
8310 
8311 err_inval:
8312 	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
8313 	return -EINVAL;
8314 }
8315 
8316 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
8317 			  bool runtime)
8318 {
8319 	struct net_device *netdev = pci_get_drvdata(pdev);
8320 	struct igb_adapter *adapter = netdev_priv(netdev);
8321 	struct e1000_hw *hw = &adapter->hw;
8322 	u32 ctrl, rctl, status;
8323 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
8324 #ifdef CONFIG_PM
8325 	int retval = 0;
8326 #endif
8327 
8328 	rtnl_lock();
8329 	netif_device_detach(netdev);
8330 
8331 	if (netif_running(netdev))
8332 		__igb_close(netdev, true);
8333 
8334 	igb_ptp_suspend(adapter);
8335 
8336 	igb_clear_interrupt_scheme(adapter);
8337 	rtnl_unlock();
8338 
8339 #ifdef CONFIG_PM
8340 	retval = pci_save_state(pdev);
8341 	if (retval)
8342 		return retval;
8343 #endif
8344 
8345 	status = rd32(E1000_STATUS);
8346 	if (status & E1000_STATUS_LU)
8347 		wufc &= ~E1000_WUFC_LNKC;
8348 
8349 	if (wufc) {
8350 		igb_setup_rctl(adapter);
8351 		igb_set_rx_mode(netdev);
8352 
8353 		/* turn on all-multi mode if wake on multicast is enabled */
8354 		if (wufc & E1000_WUFC_MC) {
8355 			rctl = rd32(E1000_RCTL);
8356 			rctl |= E1000_RCTL_MPE;
8357 			wr32(E1000_RCTL, rctl);
8358 		}
8359 
8360 		ctrl = rd32(E1000_CTRL);
8361 		/* advertise wake from D3Cold */
8362 		#define E1000_CTRL_ADVD3WUC 0x00100000
8363 		/* phy power management enable */
8364 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
8365 		ctrl |= E1000_CTRL_ADVD3WUC;
8366 		wr32(E1000_CTRL, ctrl);
8367 
8368 		/* Allow time for pending master requests to run */
8369 		igb_disable_pcie_master(hw);
8370 
8371 		wr32(E1000_WUC, E1000_WUC_PME_EN);
8372 		wr32(E1000_WUFC, wufc);
8373 	} else {
8374 		wr32(E1000_WUC, 0);
8375 		wr32(E1000_WUFC, 0);
8376 	}
8377 
8378 	*enable_wake = wufc || adapter->en_mng_pt;
8379 	if (!*enable_wake)
8380 		igb_power_down_link(adapter);
8381 	else
8382 		igb_power_up_link(adapter);
8383 
8384 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
8385 	 * would have already happened in close and is redundant.
8386 	 */
8387 	igb_release_hw_control(adapter);
8388 
8389 	pci_disable_device(pdev);
8390 
8391 	return 0;
8392 }
8393 
8394 static void igb_deliver_wake_packet(struct net_device *netdev)
8395 {
8396 	struct igb_adapter *adapter = netdev_priv(netdev);
8397 	struct e1000_hw *hw = &adapter->hw;
8398 	struct sk_buff *skb;
8399 	u32 wupl;
8400 
8401 	wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
8402 
8403 	/* WUPM stores only the first 128 bytes of the wake packet.
8404 	 * Read the packet only if we have the whole thing.
8405 	 */
8406 	if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
8407 		return;
8408 
8409 	skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
8410 	if (!skb)
8411 		return;
8412 
8413 	skb_put(skb, wupl);
8414 
8415 	/* Ensure reads are 32-bit aligned */
8416 	wupl = roundup(wupl, 4);
8417 
8418 	memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
8419 
8420 	skb->protocol = eth_type_trans(skb, netdev);
8421 	netif_rx(skb);
8422 }
8423 
8424 static int __maybe_unused igb_suspend(struct device *dev)
8425 {
8426 	int retval;
8427 	bool wake;
8428 	struct pci_dev *pdev = to_pci_dev(dev);
8429 
8430 	retval = __igb_shutdown(pdev, &wake, 0);
8431 	if (retval)
8432 		return retval;
8433 
8434 	if (wake) {
8435 		pci_prepare_to_sleep(pdev);
8436 	} else {
8437 		pci_wake_from_d3(pdev, false);
8438 		pci_set_power_state(pdev, PCI_D3hot);
8439 	}
8440 
8441 	return 0;
8442 }
8443 
8444 static int __maybe_unused igb_resume(struct device *dev)
8445 {
8446 	struct pci_dev *pdev = to_pci_dev(dev);
8447 	struct net_device *netdev = pci_get_drvdata(pdev);
8448 	struct igb_adapter *adapter = netdev_priv(netdev);
8449 	struct e1000_hw *hw = &adapter->hw;
8450 	u32 err, val;
8451 
8452 	pci_set_power_state(pdev, PCI_D0);
8453 	pci_restore_state(pdev);
8454 	pci_save_state(pdev);
8455 
8456 	if (!pci_device_is_present(pdev))
8457 		return -ENODEV;
8458 	err = pci_enable_device_mem(pdev);
8459 	if (err) {
8460 		dev_err(&pdev->dev,
8461 			"igb: Cannot enable PCI device from suspend\n");
8462 		return err;
8463 	}
8464 	pci_set_master(pdev);
8465 
8466 	pci_enable_wake(pdev, PCI_D3hot, 0);
8467 	pci_enable_wake(pdev, PCI_D3cold, 0);
8468 
8469 	if (igb_init_interrupt_scheme(adapter, true)) {
8470 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8471 		return -ENOMEM;
8472 	}
8473 
8474 	igb_reset(adapter);
8475 
8476 	/* let the f/w know that the h/w is now under the control of the
8477 	 * driver.
8478 	 */
8479 	igb_get_hw_control(adapter);
8480 
8481 	val = rd32(E1000_WUS);
8482 	if (val & WAKE_PKT_WUS)
8483 		igb_deliver_wake_packet(netdev);
8484 
8485 	wr32(E1000_WUS, ~0);
8486 
8487 	rtnl_lock();
8488 	if (!err && netif_running(netdev))
8489 		err = __igb_open(netdev, true);
8490 
8491 	if (!err)
8492 		netif_device_attach(netdev);
8493 	rtnl_unlock();
8494 
8495 	return err;
8496 }
8497 
8498 static int __maybe_unused igb_runtime_idle(struct device *dev)
8499 {
8500 	struct pci_dev *pdev = to_pci_dev(dev);
8501 	struct net_device *netdev = pci_get_drvdata(pdev);
8502 	struct igb_adapter *adapter = netdev_priv(netdev);
8503 
8504 	if (!igb_has_link(adapter))
8505 		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
8506 
8507 	return -EBUSY;
8508 }
8509 
8510 static int __maybe_unused igb_runtime_suspend(struct device *dev)
8511 {
8512 	struct pci_dev *pdev = to_pci_dev(dev);
8513 	int retval;
8514 	bool wake;
8515 
8516 	retval = __igb_shutdown(pdev, &wake, 1);
8517 	if (retval)
8518 		return retval;
8519 
8520 	if (wake) {
8521 		pci_prepare_to_sleep(pdev);
8522 	} else {
8523 		pci_wake_from_d3(pdev, false);
8524 		pci_set_power_state(pdev, PCI_D3hot);
8525 	}
8526 
8527 	return 0;
8528 }
8529 
8530 static int __maybe_unused igb_runtime_resume(struct device *dev)
8531 {
8532 	return igb_resume(dev);
8533 }
8534 
8535 static void igb_shutdown(struct pci_dev *pdev)
8536 {
8537 	bool wake;
8538 
8539 	__igb_shutdown(pdev, &wake, 0);
8540 
8541 	if (system_state == SYSTEM_POWER_OFF) {
8542 		pci_wake_from_d3(pdev, wake);
8543 		pci_set_power_state(pdev, PCI_D3hot);
8544 	}
8545 }
8546 
8547 #ifdef CONFIG_PCI_IOV
8548 static int igb_sriov_reinit(struct pci_dev *dev)
8549 {
8550 	struct net_device *netdev = pci_get_drvdata(dev);
8551 	struct igb_adapter *adapter = netdev_priv(netdev);
8552 	struct pci_dev *pdev = adapter->pdev;
8553 
8554 	rtnl_lock();
8555 
8556 	if (netif_running(netdev))
8557 		igb_close(netdev);
8558 	else
8559 		igb_reset(adapter);
8560 
8561 	igb_clear_interrupt_scheme(adapter);
8562 
8563 	igb_init_queue_configuration(adapter);
8564 
8565 	if (igb_init_interrupt_scheme(adapter, true)) {
8566 		rtnl_unlock();
8567 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8568 		return -ENOMEM;
8569 	}
8570 
8571 	if (netif_running(netdev))
8572 		igb_open(netdev);
8573 
8574 	rtnl_unlock();
8575 
8576 	return 0;
8577 }
8578 
8579 static int igb_pci_disable_sriov(struct pci_dev *dev)
8580 {
8581 	int err = igb_disable_sriov(dev);
8582 
8583 	if (!err)
8584 		err = igb_sriov_reinit(dev);
8585 
8586 	return err;
8587 }
8588 
8589 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
8590 {
8591 	int err = igb_enable_sriov(dev, num_vfs);
8592 
8593 	if (err)
8594 		goto out;
8595 
8596 	err = igb_sriov_reinit(dev);
8597 	if (!err)
8598 		return num_vfs;
8599 
8600 out:
8601 	return err;
8602 }
8603 
8604 #endif
8605 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
8606 {
8607 #ifdef CONFIG_PCI_IOV
8608 	if (num_vfs == 0)
8609 		return igb_pci_disable_sriov(dev);
8610 	else
8611 		return igb_pci_enable_sriov(dev, num_vfs);
8612 #endif
8613 	return 0;
8614 }
8615 
8616 #ifdef CONFIG_NET_POLL_CONTROLLER
8617 /* Polling 'interrupt' - used by things like netconsole to send skbs
8618  * without having to re-enable interrupts. It's not called while
8619  * the interrupt routine is executing.
8620  */
8621 static void igb_netpoll(struct net_device *netdev)
8622 {
8623 	struct igb_adapter *adapter = netdev_priv(netdev);
8624 	struct e1000_hw *hw = &adapter->hw;
8625 	struct igb_q_vector *q_vector;
8626 	int i;
8627 
8628 	for (i = 0; i < adapter->num_q_vectors; i++) {
8629 		q_vector = adapter->q_vector[i];
8630 		if (adapter->flags & IGB_FLAG_HAS_MSIX)
8631 			wr32(E1000_EIMC, q_vector->eims_value);
8632 		else
8633 			igb_irq_disable(adapter);
8634 		napi_schedule(&q_vector->napi);
8635 	}
8636 }
8637 #endif /* CONFIG_NET_POLL_CONTROLLER */
8638 
8639 /**
8640  *  igb_io_error_detected - called when PCI error is detected
8641  *  @pdev: Pointer to PCI device
8642  *  @state: The current pci connection state
8643  *
8644  *  This function is called after a PCI bus error affecting
8645  *  this device has been detected.
8646  **/
8647 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
8648 					      pci_channel_state_t state)
8649 {
8650 	struct net_device *netdev = pci_get_drvdata(pdev);
8651 	struct igb_adapter *adapter = netdev_priv(netdev);
8652 
8653 	netif_device_detach(netdev);
8654 
8655 	if (state == pci_channel_io_perm_failure)
8656 		return PCI_ERS_RESULT_DISCONNECT;
8657 
8658 	if (netif_running(netdev))
8659 		igb_down(adapter);
8660 	pci_disable_device(pdev);
8661 
8662 	/* Request a slot slot reset. */
8663 	return PCI_ERS_RESULT_NEED_RESET;
8664 }
8665 
8666 /**
8667  *  igb_io_slot_reset - called after the pci bus has been reset.
8668  *  @pdev: Pointer to PCI device
8669  *
8670  *  Restart the card from scratch, as if from a cold-boot. Implementation
8671  *  resembles the first-half of the igb_resume routine.
8672  **/
8673 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
8674 {
8675 	struct net_device *netdev = pci_get_drvdata(pdev);
8676 	struct igb_adapter *adapter = netdev_priv(netdev);
8677 	struct e1000_hw *hw = &adapter->hw;
8678 	pci_ers_result_t result;
8679 	int err;
8680 
8681 	if (pci_enable_device_mem(pdev)) {
8682 		dev_err(&pdev->dev,
8683 			"Cannot re-enable PCI device after reset.\n");
8684 		result = PCI_ERS_RESULT_DISCONNECT;
8685 	} else {
8686 		pci_set_master(pdev);
8687 		pci_restore_state(pdev);
8688 		pci_save_state(pdev);
8689 
8690 		pci_enable_wake(pdev, PCI_D3hot, 0);
8691 		pci_enable_wake(pdev, PCI_D3cold, 0);
8692 
8693 		/* In case of PCI error, adapter lose its HW address
8694 		 * so we should re-assign it here.
8695 		 */
8696 		hw->hw_addr = adapter->io_addr;
8697 
8698 		igb_reset(adapter);
8699 		wr32(E1000_WUS, ~0);
8700 		result = PCI_ERS_RESULT_RECOVERED;
8701 	}
8702 
8703 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
8704 	if (err) {
8705 		dev_err(&pdev->dev,
8706 			"pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
8707 			err);
8708 		/* non-fatal, continue */
8709 	}
8710 
8711 	return result;
8712 }
8713 
8714 /**
8715  *  igb_io_resume - called when traffic can start flowing again.
8716  *  @pdev: Pointer to PCI device
8717  *
8718  *  This callback is called when the error recovery driver tells us that
8719  *  its OK to resume normal operation. Implementation resembles the
8720  *  second-half of the igb_resume routine.
8721  */
8722 static void igb_io_resume(struct pci_dev *pdev)
8723 {
8724 	struct net_device *netdev = pci_get_drvdata(pdev);
8725 	struct igb_adapter *adapter = netdev_priv(netdev);
8726 
8727 	if (netif_running(netdev)) {
8728 		if (igb_up(adapter)) {
8729 			dev_err(&pdev->dev, "igb_up failed after reset\n");
8730 			return;
8731 		}
8732 	}
8733 
8734 	netif_device_attach(netdev);
8735 
8736 	/* let the f/w know that the h/w is now under the control of the
8737 	 * driver.
8738 	 */
8739 	igb_get_hw_control(adapter);
8740 }
8741 
8742 /**
8743  *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
8744  *  @adapter: Pointer to adapter structure
8745  *  @index: Index of the RAR entry which need to be synced with MAC table
8746  **/
8747 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
8748 {
8749 	struct e1000_hw *hw = &adapter->hw;
8750 	u32 rar_low, rar_high;
8751 	u8 *addr = adapter->mac_table[index].addr;
8752 
8753 	/* HW expects these to be in network order when they are plugged
8754 	 * into the registers which are little endian.  In order to guarantee
8755 	 * that ordering we need to do an leXX_to_cpup here in order to be
8756 	 * ready for the byteswap that occurs with writel
8757 	 */
8758 	rar_low = le32_to_cpup((__le32 *)(addr));
8759 	rar_high = le16_to_cpup((__le16 *)(addr + 4));
8760 
8761 	/* Indicate to hardware the Address is Valid. */
8762 	if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
8763 		if (is_valid_ether_addr(addr))
8764 			rar_high |= E1000_RAH_AV;
8765 
8766 		if (hw->mac.type == e1000_82575)
8767 			rar_high |= E1000_RAH_POOL_1 *
8768 				    adapter->mac_table[index].queue;
8769 		else
8770 			rar_high |= E1000_RAH_POOL_1 <<
8771 				    adapter->mac_table[index].queue;
8772 	}
8773 
8774 	wr32(E1000_RAL(index), rar_low);
8775 	wrfl();
8776 	wr32(E1000_RAH(index), rar_high);
8777 	wrfl();
8778 }
8779 
8780 static int igb_set_vf_mac(struct igb_adapter *adapter,
8781 			  int vf, unsigned char *mac_addr)
8782 {
8783 	struct e1000_hw *hw = &adapter->hw;
8784 	/* VF MAC addresses start at end of receive addresses and moves
8785 	 * towards the first, as a result a collision should not be possible
8786 	 */
8787 	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
8788 	unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
8789 
8790 	ether_addr_copy(vf_mac_addr, mac_addr);
8791 	ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
8792 	adapter->mac_table[rar_entry].queue = vf;
8793 	adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
8794 	igb_rar_set_index(adapter, rar_entry);
8795 
8796 	return 0;
8797 }
8798 
8799 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
8800 {
8801 	struct igb_adapter *adapter = netdev_priv(netdev);
8802 
8803 	if (vf >= adapter->vfs_allocated_count)
8804 		return -EINVAL;
8805 
8806 	/* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
8807 	 * flag and allows to overwrite the MAC via VF netdev.  This
8808 	 * is necessary to allow libvirt a way to restore the original
8809 	 * MAC after unbinding vfio-pci and reloading igbvf after shutting
8810 	 * down a VM.
8811 	 */
8812 	if (is_zero_ether_addr(mac)) {
8813 		adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
8814 		dev_info(&adapter->pdev->dev,
8815 			 "remove administratively set MAC on VF %d\n",
8816 			 vf);
8817 	} else if (is_valid_ether_addr(mac)) {
8818 		adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
8819 		dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
8820 			 mac, vf);
8821 		dev_info(&adapter->pdev->dev,
8822 			 "Reload the VF driver to make this change effective.");
8823 		/* Generate additional warning if PF is down */
8824 		if (test_bit(__IGB_DOWN, &adapter->state)) {
8825 			dev_warn(&adapter->pdev->dev,
8826 				 "The VF MAC address has been set, but the PF device is not up.\n");
8827 			dev_warn(&adapter->pdev->dev,
8828 				 "Bring the PF device up before attempting to use the VF device.\n");
8829 		}
8830 	} else {
8831 		return -EINVAL;
8832 	}
8833 	return igb_set_vf_mac(adapter, vf, mac);
8834 }
8835 
8836 static int igb_link_mbps(int internal_link_speed)
8837 {
8838 	switch (internal_link_speed) {
8839 	case SPEED_100:
8840 		return 100;
8841 	case SPEED_1000:
8842 		return 1000;
8843 	default:
8844 		return 0;
8845 	}
8846 }
8847 
8848 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
8849 				  int link_speed)
8850 {
8851 	int rf_dec, rf_int;
8852 	u32 bcnrc_val;
8853 
8854 	if (tx_rate != 0) {
8855 		/* Calculate the rate factor values to set */
8856 		rf_int = link_speed / tx_rate;
8857 		rf_dec = (link_speed - (rf_int * tx_rate));
8858 		rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
8859 			 tx_rate;
8860 
8861 		bcnrc_val = E1000_RTTBCNRC_RS_ENA;
8862 		bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
8863 			      E1000_RTTBCNRC_RF_INT_MASK);
8864 		bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
8865 	} else {
8866 		bcnrc_val = 0;
8867 	}
8868 
8869 	wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
8870 	/* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
8871 	 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
8872 	 */
8873 	wr32(E1000_RTTBCNRM, 0x14);
8874 	wr32(E1000_RTTBCNRC, bcnrc_val);
8875 }
8876 
8877 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
8878 {
8879 	int actual_link_speed, i;
8880 	bool reset_rate = false;
8881 
8882 	/* VF TX rate limit was not set or not supported */
8883 	if ((adapter->vf_rate_link_speed == 0) ||
8884 	    (adapter->hw.mac.type != e1000_82576))
8885 		return;
8886 
8887 	actual_link_speed = igb_link_mbps(adapter->link_speed);
8888 	if (actual_link_speed != adapter->vf_rate_link_speed) {
8889 		reset_rate = true;
8890 		adapter->vf_rate_link_speed = 0;
8891 		dev_info(&adapter->pdev->dev,
8892 			 "Link speed has been changed. VF Transmit rate is disabled\n");
8893 	}
8894 
8895 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
8896 		if (reset_rate)
8897 			adapter->vf_data[i].tx_rate = 0;
8898 
8899 		igb_set_vf_rate_limit(&adapter->hw, i,
8900 				      adapter->vf_data[i].tx_rate,
8901 				      actual_link_speed);
8902 	}
8903 }
8904 
8905 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
8906 			     int min_tx_rate, int max_tx_rate)
8907 {
8908 	struct igb_adapter *adapter = netdev_priv(netdev);
8909 	struct e1000_hw *hw = &adapter->hw;
8910 	int actual_link_speed;
8911 
8912 	if (hw->mac.type != e1000_82576)
8913 		return -EOPNOTSUPP;
8914 
8915 	if (min_tx_rate)
8916 		return -EINVAL;
8917 
8918 	actual_link_speed = igb_link_mbps(adapter->link_speed);
8919 	if ((vf >= adapter->vfs_allocated_count) ||
8920 	    (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
8921 	    (max_tx_rate < 0) ||
8922 	    (max_tx_rate > actual_link_speed))
8923 		return -EINVAL;
8924 
8925 	adapter->vf_rate_link_speed = actual_link_speed;
8926 	adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
8927 	igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
8928 
8929 	return 0;
8930 }
8931 
8932 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
8933 				   bool setting)
8934 {
8935 	struct igb_adapter *adapter = netdev_priv(netdev);
8936 	struct e1000_hw *hw = &adapter->hw;
8937 	u32 reg_val, reg_offset;
8938 
8939 	if (!adapter->vfs_allocated_count)
8940 		return -EOPNOTSUPP;
8941 
8942 	if (vf >= adapter->vfs_allocated_count)
8943 		return -EINVAL;
8944 
8945 	reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
8946 	reg_val = rd32(reg_offset);
8947 	if (setting)
8948 		reg_val |= (BIT(vf) |
8949 			    BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8950 	else
8951 		reg_val &= ~(BIT(vf) |
8952 			     BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8953 	wr32(reg_offset, reg_val);
8954 
8955 	adapter->vf_data[vf].spoofchk_enabled = setting;
8956 	return 0;
8957 }
8958 
8959 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
8960 {
8961 	struct igb_adapter *adapter = netdev_priv(netdev);
8962 
8963 	if (vf >= adapter->vfs_allocated_count)
8964 		return -EINVAL;
8965 	if (adapter->vf_data[vf].trusted == setting)
8966 		return 0;
8967 
8968 	adapter->vf_data[vf].trusted = setting;
8969 
8970 	dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
8971 		 vf, setting ? "" : "not ");
8972 	return 0;
8973 }
8974 
8975 static int igb_ndo_get_vf_config(struct net_device *netdev,
8976 				 int vf, struct ifla_vf_info *ivi)
8977 {
8978 	struct igb_adapter *adapter = netdev_priv(netdev);
8979 	if (vf >= adapter->vfs_allocated_count)
8980 		return -EINVAL;
8981 	ivi->vf = vf;
8982 	memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
8983 	ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
8984 	ivi->min_tx_rate = 0;
8985 	ivi->vlan = adapter->vf_data[vf].pf_vlan;
8986 	ivi->qos = adapter->vf_data[vf].pf_qos;
8987 	ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
8988 	ivi->trusted = adapter->vf_data[vf].trusted;
8989 	return 0;
8990 }
8991 
8992 static void igb_vmm_control(struct igb_adapter *adapter)
8993 {
8994 	struct e1000_hw *hw = &adapter->hw;
8995 	u32 reg;
8996 
8997 	switch (hw->mac.type) {
8998 	case e1000_82575:
8999 	case e1000_i210:
9000 	case e1000_i211:
9001 	case e1000_i354:
9002 	default:
9003 		/* replication is not supported for 82575 */
9004 		return;
9005 	case e1000_82576:
9006 		/* notify HW that the MAC is adding vlan tags */
9007 		reg = rd32(E1000_DTXCTL);
9008 		reg |= E1000_DTXCTL_VLAN_ADDED;
9009 		wr32(E1000_DTXCTL, reg);
9010 		/* Fall through */
9011 	case e1000_82580:
9012 		/* enable replication vlan tag stripping */
9013 		reg = rd32(E1000_RPLOLR);
9014 		reg |= E1000_RPLOLR_STRVLAN;
9015 		wr32(E1000_RPLOLR, reg);
9016 		/* Fall through */
9017 	case e1000_i350:
9018 		/* none of the above registers are supported by i350 */
9019 		break;
9020 	}
9021 
9022 	if (adapter->vfs_allocated_count) {
9023 		igb_vmdq_set_loopback_pf(hw, true);
9024 		igb_vmdq_set_replication_pf(hw, true);
9025 		igb_vmdq_set_anti_spoofing_pf(hw, true,
9026 					      adapter->vfs_allocated_count);
9027 	} else {
9028 		igb_vmdq_set_loopback_pf(hw, false);
9029 		igb_vmdq_set_replication_pf(hw, false);
9030 	}
9031 }
9032 
9033 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9034 {
9035 	struct e1000_hw *hw = &adapter->hw;
9036 	u32 dmac_thr;
9037 	u16 hwm;
9038 
9039 	if (hw->mac.type > e1000_82580) {
9040 		if (adapter->flags & IGB_FLAG_DMAC) {
9041 			u32 reg;
9042 
9043 			/* force threshold to 0. */
9044 			wr32(E1000_DMCTXTH, 0);
9045 
9046 			/* DMA Coalescing high water mark needs to be greater
9047 			 * than the Rx threshold. Set hwm to PBA - max frame
9048 			 * size in 16B units, capping it at PBA - 6KB.
9049 			 */
9050 			hwm = 64 * (pba - 6);
9051 			reg = rd32(E1000_FCRTC);
9052 			reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9053 			reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9054 				& E1000_FCRTC_RTH_COAL_MASK);
9055 			wr32(E1000_FCRTC, reg);
9056 
9057 			/* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9058 			 * frame size, capping it at PBA - 10KB.
9059 			 */
9060 			dmac_thr = pba - 10;
9061 			reg = rd32(E1000_DMACR);
9062 			reg &= ~E1000_DMACR_DMACTHR_MASK;
9063 			reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9064 				& E1000_DMACR_DMACTHR_MASK);
9065 
9066 			/* transition to L0x or L1 if available..*/
9067 			reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9068 
9069 			/* watchdog timer= +-1000 usec in 32usec intervals */
9070 			reg |= (1000 >> 5);
9071 
9072 			/* Disable BMC-to-OS Watchdog Enable */
9073 			if (hw->mac.type != e1000_i354)
9074 				reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9075 
9076 			wr32(E1000_DMACR, reg);
9077 
9078 			/* no lower threshold to disable
9079 			 * coalescing(smart fifb)-UTRESH=0
9080 			 */
9081 			wr32(E1000_DMCRTRH, 0);
9082 
9083 			reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9084 
9085 			wr32(E1000_DMCTLX, reg);
9086 
9087 			/* free space in tx packet buffer to wake from
9088 			 * DMA coal
9089 			 */
9090 			wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
9091 			     (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
9092 
9093 			/* make low power state decision controlled
9094 			 * by DMA coal
9095 			 */
9096 			reg = rd32(E1000_PCIEMISC);
9097 			reg &= ~E1000_PCIEMISC_LX_DECISION;
9098 			wr32(E1000_PCIEMISC, reg);
9099 		} /* endif adapter->dmac is not disabled */
9100 	} else if (hw->mac.type == e1000_82580) {
9101 		u32 reg = rd32(E1000_PCIEMISC);
9102 
9103 		wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
9104 		wr32(E1000_DMACR, 0);
9105 	}
9106 }
9107 
9108 /**
9109  *  igb_read_i2c_byte - Reads 8 bit word over I2C
9110  *  @hw: pointer to hardware structure
9111  *  @byte_offset: byte offset to read
9112  *  @dev_addr: device address
9113  *  @data: value read
9114  *
9115  *  Performs byte read operation over I2C interface at
9116  *  a specified device address.
9117  **/
9118 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9119 		      u8 dev_addr, u8 *data)
9120 {
9121 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9122 	struct i2c_client *this_client = adapter->i2c_client;
9123 	s32 status;
9124 	u16 swfw_mask = 0;
9125 
9126 	if (!this_client)
9127 		return E1000_ERR_I2C;
9128 
9129 	swfw_mask = E1000_SWFW_PHY0_SM;
9130 
9131 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9132 		return E1000_ERR_SWFW_SYNC;
9133 
9134 	status = i2c_smbus_read_byte_data(this_client, byte_offset);
9135 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9136 
9137 	if (status < 0)
9138 		return E1000_ERR_I2C;
9139 	else {
9140 		*data = status;
9141 		return 0;
9142 	}
9143 }
9144 
9145 /**
9146  *  igb_write_i2c_byte - Writes 8 bit word over I2C
9147  *  @hw: pointer to hardware structure
9148  *  @byte_offset: byte offset to write
9149  *  @dev_addr: device address
9150  *  @data: value to write
9151  *
9152  *  Performs byte write operation over I2C interface at
9153  *  a specified device address.
9154  **/
9155 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9156 		       u8 dev_addr, u8 data)
9157 {
9158 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9159 	struct i2c_client *this_client = adapter->i2c_client;
9160 	s32 status;
9161 	u16 swfw_mask = E1000_SWFW_PHY0_SM;
9162 
9163 	if (!this_client)
9164 		return E1000_ERR_I2C;
9165 
9166 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9167 		return E1000_ERR_SWFW_SYNC;
9168 	status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
9169 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9170 
9171 	if (status)
9172 		return E1000_ERR_I2C;
9173 	else
9174 		return 0;
9175 
9176 }
9177 
9178 int igb_reinit_queues(struct igb_adapter *adapter)
9179 {
9180 	struct net_device *netdev = adapter->netdev;
9181 	struct pci_dev *pdev = adapter->pdev;
9182 	int err = 0;
9183 
9184 	if (netif_running(netdev))
9185 		igb_close(netdev);
9186 
9187 	igb_reset_interrupt_capability(adapter);
9188 
9189 	if (igb_init_interrupt_scheme(adapter, true)) {
9190 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9191 		return -ENOMEM;
9192 	}
9193 
9194 	if (netif_running(netdev))
9195 		err = igb_open(netdev);
9196 
9197 	return err;
9198 }
9199 
9200 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
9201 {
9202 	struct igb_nfc_filter *rule;
9203 
9204 	spin_lock(&adapter->nfc_lock);
9205 
9206 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9207 		igb_erase_filter(adapter, rule);
9208 
9209 	spin_unlock(&adapter->nfc_lock);
9210 }
9211 
9212 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
9213 {
9214 	struct igb_nfc_filter *rule;
9215 
9216 	spin_lock(&adapter->nfc_lock);
9217 
9218 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9219 		igb_add_filter(adapter, rule);
9220 
9221 	spin_unlock(&adapter->nfc_lock);
9222 }
9223 /* igb_main.c */
9224