1 /*******************************************************************************
2 
3   Intel(R) Gigabit Ethernet Linux driver
4   Copyright(c) 2007-2013 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21 
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 
26 *******************************************************************************/
27 
28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/bitops.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/netdevice.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if.h>
45 #include <linux/if_vlan.h>
46 #include <linux/pci.h>
47 #include <linux/pci-aspm.h>
48 #include <linux/delay.h>
49 #include <linux/interrupt.h>
50 #include <linux/ip.h>
51 #include <linux/tcp.h>
52 #include <linux/sctp.h>
53 #include <linux/if_ether.h>
54 #include <linux/aer.h>
55 #include <linux/prefetch.h>
56 #include <linux/pm_runtime.h>
57 #ifdef CONFIG_IGB_DCA
58 #include <linux/dca.h>
59 #endif
60 #include <linux/i2c.h>
61 #include "igb.h"
62 
63 #define MAJ 5
64 #define MIN 0
65 #define BUILD 3
66 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
67 __stringify(BUILD) "-k"
68 char igb_driver_name[] = "igb";
69 char igb_driver_version[] = DRV_VERSION;
70 static const char igb_driver_string[] =
71 				"Intel(R) Gigabit Ethernet Network Driver";
72 static const char igb_copyright[] =
73 				"Copyright (c) 2007-2013 Intel Corporation.";
74 
75 static const struct e1000_info *igb_info_tbl[] = {
76 	[board_82575] = &e1000_82575_info,
77 };
78 
79 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
80 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
81 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
82 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
83 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
84 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
85 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
86 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
87 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
88 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
89 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
90 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
91 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
92 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
93 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
94 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
95 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
96 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
97 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
98 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
99 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
100 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
101 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
102 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
103 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
104 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
105 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
106 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
107 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
108 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
109 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
110 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
111 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
112 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
113 	/* required last entry */
114 	{0, }
115 };
116 
117 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
118 
119 void igb_reset(struct igb_adapter *);
120 static int igb_setup_all_tx_resources(struct igb_adapter *);
121 static int igb_setup_all_rx_resources(struct igb_adapter *);
122 static void igb_free_all_tx_resources(struct igb_adapter *);
123 static void igb_free_all_rx_resources(struct igb_adapter *);
124 static void igb_setup_mrqc(struct igb_adapter *);
125 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
126 static void igb_remove(struct pci_dev *pdev);
127 static int igb_sw_init(struct igb_adapter *);
128 static int igb_open(struct net_device *);
129 static int igb_close(struct net_device *);
130 static void igb_configure(struct igb_adapter *);
131 static void igb_configure_tx(struct igb_adapter *);
132 static void igb_configure_rx(struct igb_adapter *);
133 static void igb_clean_all_tx_rings(struct igb_adapter *);
134 static void igb_clean_all_rx_rings(struct igb_adapter *);
135 static void igb_clean_tx_ring(struct igb_ring *);
136 static void igb_clean_rx_ring(struct igb_ring *);
137 static void igb_set_rx_mode(struct net_device *);
138 static void igb_update_phy_info(unsigned long);
139 static void igb_watchdog(unsigned long);
140 static void igb_watchdog_task(struct work_struct *);
141 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
142 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
143 						 struct rtnl_link_stats64 *stats);
144 static int igb_change_mtu(struct net_device *, int);
145 static int igb_set_mac(struct net_device *, void *);
146 static void igb_set_uta(struct igb_adapter *adapter);
147 static irqreturn_t igb_intr(int irq, void *);
148 static irqreturn_t igb_intr_msi(int irq, void *);
149 static irqreturn_t igb_msix_other(int irq, void *);
150 static irqreturn_t igb_msix_ring(int irq, void *);
151 #ifdef CONFIG_IGB_DCA
152 static void igb_update_dca(struct igb_q_vector *);
153 static void igb_setup_dca(struct igb_adapter *);
154 #endif /* CONFIG_IGB_DCA */
155 static int igb_poll(struct napi_struct *, int);
156 static bool igb_clean_tx_irq(struct igb_q_vector *);
157 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
158 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
159 static void igb_tx_timeout(struct net_device *);
160 static void igb_reset_task(struct work_struct *);
161 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
162 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
163 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
164 static void igb_restore_vlan(struct igb_adapter *);
165 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
166 static void igb_ping_all_vfs(struct igb_adapter *);
167 static void igb_msg_task(struct igb_adapter *);
168 static void igb_vmm_control(struct igb_adapter *);
169 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
170 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
171 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
172 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
173 			       int vf, u16 vlan, u8 qos);
174 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
175 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
176 				   bool setting);
177 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
178 				 struct ifla_vf_info *ivi);
179 static void igb_check_vf_rate_limit(struct igb_adapter *);
180 
181 #ifdef CONFIG_PCI_IOV
182 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
183 #endif
184 
185 #ifdef CONFIG_PM
186 #ifdef CONFIG_PM_SLEEP
187 static int igb_suspend(struct device *);
188 #endif
189 static int igb_resume(struct device *);
190 #ifdef CONFIG_PM_RUNTIME
191 static int igb_runtime_suspend(struct device *dev);
192 static int igb_runtime_resume(struct device *dev);
193 static int igb_runtime_idle(struct device *dev);
194 #endif
195 static const struct dev_pm_ops igb_pm_ops = {
196 	SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
197 	SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
198 			igb_runtime_idle)
199 };
200 #endif
201 static void igb_shutdown(struct pci_dev *);
202 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
203 #ifdef CONFIG_IGB_DCA
204 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
205 static struct notifier_block dca_notifier = {
206 	.notifier_call	= igb_notify_dca,
207 	.next		= NULL,
208 	.priority	= 0
209 };
210 #endif
211 #ifdef CONFIG_NET_POLL_CONTROLLER
212 /* for netdump / net console */
213 static void igb_netpoll(struct net_device *);
214 #endif
215 #ifdef CONFIG_PCI_IOV
216 static unsigned int max_vfs = 0;
217 module_param(max_vfs, uint, 0);
218 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
219                  "per physical function");
220 #endif /* CONFIG_PCI_IOV */
221 
222 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
223 		     pci_channel_state_t);
224 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
225 static void igb_io_resume(struct pci_dev *);
226 
227 static const struct pci_error_handlers igb_err_handler = {
228 	.error_detected = igb_io_error_detected,
229 	.slot_reset = igb_io_slot_reset,
230 	.resume = igb_io_resume,
231 };
232 
233 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
234 
235 static struct pci_driver igb_driver = {
236 	.name     = igb_driver_name,
237 	.id_table = igb_pci_tbl,
238 	.probe    = igb_probe,
239 	.remove   = igb_remove,
240 #ifdef CONFIG_PM
241 	.driver.pm = &igb_pm_ops,
242 #endif
243 	.shutdown = igb_shutdown,
244 	.sriov_configure = igb_pci_sriov_configure,
245 	.err_handler = &igb_err_handler
246 };
247 
248 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
249 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
250 MODULE_LICENSE("GPL");
251 MODULE_VERSION(DRV_VERSION);
252 
253 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
254 static int debug = -1;
255 module_param(debug, int, 0);
256 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
257 
258 struct igb_reg_info {
259 	u32 ofs;
260 	char *name;
261 };
262 
263 static const struct igb_reg_info igb_reg_info_tbl[] = {
264 
265 	/* General Registers */
266 	{E1000_CTRL, "CTRL"},
267 	{E1000_STATUS, "STATUS"},
268 	{E1000_CTRL_EXT, "CTRL_EXT"},
269 
270 	/* Interrupt Registers */
271 	{E1000_ICR, "ICR"},
272 
273 	/* RX Registers */
274 	{E1000_RCTL, "RCTL"},
275 	{E1000_RDLEN(0), "RDLEN"},
276 	{E1000_RDH(0), "RDH"},
277 	{E1000_RDT(0), "RDT"},
278 	{E1000_RXDCTL(0), "RXDCTL"},
279 	{E1000_RDBAL(0), "RDBAL"},
280 	{E1000_RDBAH(0), "RDBAH"},
281 
282 	/* TX Registers */
283 	{E1000_TCTL, "TCTL"},
284 	{E1000_TDBAL(0), "TDBAL"},
285 	{E1000_TDBAH(0), "TDBAH"},
286 	{E1000_TDLEN(0), "TDLEN"},
287 	{E1000_TDH(0), "TDH"},
288 	{E1000_TDT(0), "TDT"},
289 	{E1000_TXDCTL(0), "TXDCTL"},
290 	{E1000_TDFH, "TDFH"},
291 	{E1000_TDFT, "TDFT"},
292 	{E1000_TDFHS, "TDFHS"},
293 	{E1000_TDFPC, "TDFPC"},
294 
295 	/* List Terminator */
296 	{}
297 };
298 
299 /* igb_regdump - register printout routine */
300 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
301 {
302 	int n = 0;
303 	char rname[16];
304 	u32 regs[8];
305 
306 	switch (reginfo->ofs) {
307 	case E1000_RDLEN(0):
308 		for (n = 0; n < 4; n++)
309 			regs[n] = rd32(E1000_RDLEN(n));
310 		break;
311 	case E1000_RDH(0):
312 		for (n = 0; n < 4; n++)
313 			regs[n] = rd32(E1000_RDH(n));
314 		break;
315 	case E1000_RDT(0):
316 		for (n = 0; n < 4; n++)
317 			regs[n] = rd32(E1000_RDT(n));
318 		break;
319 	case E1000_RXDCTL(0):
320 		for (n = 0; n < 4; n++)
321 			regs[n] = rd32(E1000_RXDCTL(n));
322 		break;
323 	case E1000_RDBAL(0):
324 		for (n = 0; n < 4; n++)
325 			regs[n] = rd32(E1000_RDBAL(n));
326 		break;
327 	case E1000_RDBAH(0):
328 		for (n = 0; n < 4; n++)
329 			regs[n] = rd32(E1000_RDBAH(n));
330 		break;
331 	case E1000_TDBAL(0):
332 		for (n = 0; n < 4; n++)
333 			regs[n] = rd32(E1000_RDBAL(n));
334 		break;
335 	case E1000_TDBAH(0):
336 		for (n = 0; n < 4; n++)
337 			regs[n] = rd32(E1000_TDBAH(n));
338 		break;
339 	case E1000_TDLEN(0):
340 		for (n = 0; n < 4; n++)
341 			regs[n] = rd32(E1000_TDLEN(n));
342 		break;
343 	case E1000_TDH(0):
344 		for (n = 0; n < 4; n++)
345 			regs[n] = rd32(E1000_TDH(n));
346 		break;
347 	case E1000_TDT(0):
348 		for (n = 0; n < 4; n++)
349 			regs[n] = rd32(E1000_TDT(n));
350 		break;
351 	case E1000_TXDCTL(0):
352 		for (n = 0; n < 4; n++)
353 			regs[n] = rd32(E1000_TXDCTL(n));
354 		break;
355 	default:
356 		pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
357 		return;
358 	}
359 
360 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
361 	pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
362 		regs[2], regs[3]);
363 }
364 
365 /* igb_dump - Print registers, Tx-rings and Rx-rings */
366 static void igb_dump(struct igb_adapter *adapter)
367 {
368 	struct net_device *netdev = adapter->netdev;
369 	struct e1000_hw *hw = &adapter->hw;
370 	struct igb_reg_info *reginfo;
371 	struct igb_ring *tx_ring;
372 	union e1000_adv_tx_desc *tx_desc;
373 	struct my_u0 { u64 a; u64 b; } *u0;
374 	struct igb_ring *rx_ring;
375 	union e1000_adv_rx_desc *rx_desc;
376 	u32 staterr;
377 	u16 i, n;
378 
379 	if (!netif_msg_hw(adapter))
380 		return;
381 
382 	/* Print netdevice Info */
383 	if (netdev) {
384 		dev_info(&adapter->pdev->dev, "Net device Info\n");
385 		pr_info("Device Name     state            trans_start      "
386 			"last_rx\n");
387 		pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
388 			netdev->state, netdev->trans_start, netdev->last_rx);
389 	}
390 
391 	/* Print Registers */
392 	dev_info(&adapter->pdev->dev, "Register Dump\n");
393 	pr_info(" Register Name   Value\n");
394 	for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
395 	     reginfo->name; reginfo++) {
396 		igb_regdump(hw, reginfo);
397 	}
398 
399 	/* Print TX Ring Summary */
400 	if (!netdev || !netif_running(netdev))
401 		goto exit;
402 
403 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
404 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
405 	for (n = 0; n < adapter->num_tx_queues; n++) {
406 		struct igb_tx_buffer *buffer_info;
407 		tx_ring = adapter->tx_ring[n];
408 		buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
409 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
410 			n, tx_ring->next_to_use, tx_ring->next_to_clean,
411 			(u64)dma_unmap_addr(buffer_info, dma),
412 			dma_unmap_len(buffer_info, len),
413 			buffer_info->next_to_watch,
414 			(u64)buffer_info->time_stamp);
415 	}
416 
417 	/* Print TX Rings */
418 	if (!netif_msg_tx_done(adapter))
419 		goto rx_ring_summary;
420 
421 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
422 
423 	/* Transmit Descriptor Formats
424 	 *
425 	 * Advanced Transmit Descriptor
426 	 *   +--------------------------------------------------------------+
427 	 * 0 |         Buffer Address [63:0]                                |
428 	 *   +--------------------------------------------------------------+
429 	 * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
430 	 *   +--------------------------------------------------------------+
431 	 *   63      46 45    40 39 38 36 35 32 31   24             15       0
432 	 */
433 
434 	for (n = 0; n < adapter->num_tx_queues; n++) {
435 		tx_ring = adapter->tx_ring[n];
436 		pr_info("------------------------------------\n");
437 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
438 		pr_info("------------------------------------\n");
439 		pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] "
440 			"[bi->dma       ] leng  ntw timestamp        "
441 			"bi->skb\n");
442 
443 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
444 			const char *next_desc;
445 			struct igb_tx_buffer *buffer_info;
446 			tx_desc = IGB_TX_DESC(tx_ring, i);
447 			buffer_info = &tx_ring->tx_buffer_info[i];
448 			u0 = (struct my_u0 *)tx_desc;
449 			if (i == tx_ring->next_to_use &&
450 			    i == tx_ring->next_to_clean)
451 				next_desc = " NTC/U";
452 			else if (i == tx_ring->next_to_use)
453 				next_desc = " NTU";
454 			else if (i == tx_ring->next_to_clean)
455 				next_desc = " NTC";
456 			else
457 				next_desc = "";
458 
459 			pr_info("T [0x%03X]    %016llX %016llX %016llX"
460 				" %04X  %p %016llX %p%s\n", i,
461 				le64_to_cpu(u0->a),
462 				le64_to_cpu(u0->b),
463 				(u64)dma_unmap_addr(buffer_info, dma),
464 				dma_unmap_len(buffer_info, len),
465 				buffer_info->next_to_watch,
466 				(u64)buffer_info->time_stamp,
467 				buffer_info->skb, next_desc);
468 
469 			if (netif_msg_pktdata(adapter) && buffer_info->skb)
470 				print_hex_dump(KERN_INFO, "",
471 					DUMP_PREFIX_ADDRESS,
472 					16, 1, buffer_info->skb->data,
473 					dma_unmap_len(buffer_info, len),
474 					true);
475 		}
476 	}
477 
478 	/* Print RX Rings Summary */
479 rx_ring_summary:
480 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
481 	pr_info("Queue [NTU] [NTC]\n");
482 	for (n = 0; n < adapter->num_rx_queues; n++) {
483 		rx_ring = adapter->rx_ring[n];
484 		pr_info(" %5d %5X %5X\n",
485 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
486 	}
487 
488 	/* Print RX Rings */
489 	if (!netif_msg_rx_status(adapter))
490 		goto exit;
491 
492 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
493 
494 	/* Advanced Receive Descriptor (Read) Format
495 	 *    63                                           1        0
496 	 *    +-----------------------------------------------------+
497 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
498 	 *    +----------------------------------------------+------+
499 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
500 	 *    +-----------------------------------------------------+
501 	 *
502 	 *
503 	 * Advanced Receive Descriptor (Write-Back) Format
504 	 *
505 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
506 	 *   +------------------------------------------------------+
507 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
508 	 *   | Checksum   Ident  |   |           |    | Type | Type |
509 	 *   +------------------------------------------------------+
510 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
511 	 *   +------------------------------------------------------+
512 	 *   63       48 47    32 31            20 19               0
513 	 */
514 
515 	for (n = 0; n < adapter->num_rx_queues; n++) {
516 		rx_ring = adapter->rx_ring[n];
517 		pr_info("------------------------------------\n");
518 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
519 		pr_info("------------------------------------\n");
520 		pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] "
521 			"[bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
522 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] -----"
523 			"----------- [bi->skb] <-- Adv Rx Write-Back format\n");
524 
525 		for (i = 0; i < rx_ring->count; i++) {
526 			const char *next_desc;
527 			struct igb_rx_buffer *buffer_info;
528 			buffer_info = &rx_ring->rx_buffer_info[i];
529 			rx_desc = IGB_RX_DESC(rx_ring, i);
530 			u0 = (struct my_u0 *)rx_desc;
531 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
532 
533 			if (i == rx_ring->next_to_use)
534 				next_desc = " NTU";
535 			else if (i == rx_ring->next_to_clean)
536 				next_desc = " NTC";
537 			else
538 				next_desc = "";
539 
540 			if (staterr & E1000_RXD_STAT_DD) {
541 				/* Descriptor Done */
542 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
543 					"RWB", i,
544 					le64_to_cpu(u0->a),
545 					le64_to_cpu(u0->b),
546 					next_desc);
547 			} else {
548 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
549 					"R  ", i,
550 					le64_to_cpu(u0->a),
551 					le64_to_cpu(u0->b),
552 					(u64)buffer_info->dma,
553 					next_desc);
554 
555 				if (netif_msg_pktdata(adapter) &&
556 				    buffer_info->dma && buffer_info->page) {
557 					print_hex_dump(KERN_INFO, "",
558 					  DUMP_PREFIX_ADDRESS,
559 					  16, 1,
560 					  page_address(buffer_info->page) +
561 						      buffer_info->page_offset,
562 					  IGB_RX_BUFSZ, true);
563 				}
564 			}
565 		}
566 	}
567 
568 exit:
569 	return;
570 }
571 
572 /**
573  *  igb_get_i2c_data - Reads the I2C SDA data bit
574  *  @hw: pointer to hardware structure
575  *  @i2cctl: Current value of I2CCTL register
576  *
577  *  Returns the I2C data bit value
578  **/
579 static int igb_get_i2c_data(void *data)
580 {
581 	struct igb_adapter *adapter = (struct igb_adapter *)data;
582 	struct e1000_hw *hw = &adapter->hw;
583 	s32 i2cctl = rd32(E1000_I2CPARAMS);
584 
585 	return ((i2cctl & E1000_I2C_DATA_IN) != 0);
586 }
587 
588 /**
589  *  igb_set_i2c_data - Sets the I2C data bit
590  *  @data: pointer to hardware structure
591  *  @state: I2C data value (0 or 1) to set
592  *
593  *  Sets the I2C data bit
594  **/
595 static void igb_set_i2c_data(void *data, int state)
596 {
597 	struct igb_adapter *adapter = (struct igb_adapter *)data;
598 	struct e1000_hw *hw = &adapter->hw;
599 	s32 i2cctl = rd32(E1000_I2CPARAMS);
600 
601 	if (state)
602 		i2cctl |= E1000_I2C_DATA_OUT;
603 	else
604 		i2cctl &= ~E1000_I2C_DATA_OUT;
605 
606 	i2cctl &= ~E1000_I2C_DATA_OE_N;
607 	i2cctl |= E1000_I2C_CLK_OE_N;
608 	wr32(E1000_I2CPARAMS, i2cctl);
609 	wrfl();
610 
611 }
612 
613 /**
614  *  igb_set_i2c_clk - Sets the I2C SCL clock
615  *  @data: pointer to hardware structure
616  *  @state: state to set clock
617  *
618  *  Sets the I2C clock line to state
619  **/
620 static void igb_set_i2c_clk(void *data, int state)
621 {
622 	struct igb_adapter *adapter = (struct igb_adapter *)data;
623 	struct e1000_hw *hw = &adapter->hw;
624 	s32 i2cctl = rd32(E1000_I2CPARAMS);
625 
626 	if (state) {
627 		i2cctl |= E1000_I2C_CLK_OUT;
628 		i2cctl &= ~E1000_I2C_CLK_OE_N;
629 	} else {
630 		i2cctl &= ~E1000_I2C_CLK_OUT;
631 		i2cctl &= ~E1000_I2C_CLK_OE_N;
632 	}
633 	wr32(E1000_I2CPARAMS, i2cctl);
634 	wrfl();
635 }
636 
637 /**
638  *  igb_get_i2c_clk - Gets the I2C SCL clock state
639  *  @data: pointer to hardware structure
640  *
641  *  Gets the I2C clock state
642  **/
643 static int igb_get_i2c_clk(void *data)
644 {
645 	struct igb_adapter *adapter = (struct igb_adapter *)data;
646 	struct e1000_hw *hw = &adapter->hw;
647 	s32 i2cctl = rd32(E1000_I2CPARAMS);
648 
649 	return ((i2cctl & E1000_I2C_CLK_IN) != 0);
650 }
651 
652 static const struct i2c_algo_bit_data igb_i2c_algo = {
653 	.setsda		= igb_set_i2c_data,
654 	.setscl		= igb_set_i2c_clk,
655 	.getsda		= igb_get_i2c_data,
656 	.getscl		= igb_get_i2c_clk,
657 	.udelay		= 5,
658 	.timeout	= 20,
659 };
660 
661 /**
662  *  igb_get_hw_dev - return device
663  *  @hw: pointer to hardware structure
664  *
665  *  used by hardware layer to print debugging information
666  **/
667 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
668 {
669 	struct igb_adapter *adapter = hw->back;
670 	return adapter->netdev;
671 }
672 
673 /**
674  *  igb_init_module - Driver Registration Routine
675  *
676  *  igb_init_module is the first routine called when the driver is
677  *  loaded. All it does is register with the PCI subsystem.
678  **/
679 static int __init igb_init_module(void)
680 {
681 	int ret;
682 	pr_info("%s - version %s\n",
683 	       igb_driver_string, igb_driver_version);
684 
685 	pr_info("%s\n", igb_copyright);
686 
687 #ifdef CONFIG_IGB_DCA
688 	dca_register_notify(&dca_notifier);
689 #endif
690 	ret = pci_register_driver(&igb_driver);
691 	return ret;
692 }
693 
694 module_init(igb_init_module);
695 
696 /**
697  *  igb_exit_module - Driver Exit Cleanup Routine
698  *
699  *  igb_exit_module is called just before the driver is removed
700  *  from memory.
701  **/
702 static void __exit igb_exit_module(void)
703 {
704 #ifdef CONFIG_IGB_DCA
705 	dca_unregister_notify(&dca_notifier);
706 #endif
707 	pci_unregister_driver(&igb_driver);
708 }
709 
710 module_exit(igb_exit_module);
711 
712 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
713 /**
714  *  igb_cache_ring_register - Descriptor ring to register mapping
715  *  @adapter: board private structure to initialize
716  *
717  *  Once we know the feature-set enabled for the device, we'll cache
718  *  the register offset the descriptor ring is assigned to.
719  **/
720 static void igb_cache_ring_register(struct igb_adapter *adapter)
721 {
722 	int i = 0, j = 0;
723 	u32 rbase_offset = adapter->vfs_allocated_count;
724 
725 	switch (adapter->hw.mac.type) {
726 	case e1000_82576:
727 		/* The queues are allocated for virtualization such that VF 0
728 		 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
729 		 * In order to avoid collision we start at the first free queue
730 		 * and continue consuming queues in the same sequence
731 		 */
732 		if (adapter->vfs_allocated_count) {
733 			for (; i < adapter->rss_queues; i++)
734 				adapter->rx_ring[i]->reg_idx = rbase_offset +
735 							       Q_IDX_82576(i);
736 		}
737 	case e1000_82575:
738 	case e1000_82580:
739 	case e1000_i350:
740 	case e1000_i354:
741 	case e1000_i210:
742 	case e1000_i211:
743 	default:
744 		for (; i < adapter->num_rx_queues; i++)
745 			adapter->rx_ring[i]->reg_idx = rbase_offset + i;
746 		for (; j < adapter->num_tx_queues; j++)
747 			adapter->tx_ring[j]->reg_idx = rbase_offset + j;
748 		break;
749 	}
750 }
751 
752 /**
753  *  igb_write_ivar - configure ivar for given MSI-X vector
754  *  @hw: pointer to the HW structure
755  *  @msix_vector: vector number we are allocating to a given ring
756  *  @index: row index of IVAR register to write within IVAR table
757  *  @offset: column offset of in IVAR, should be multiple of 8
758  *
759  *  This function is intended to handle the writing of the IVAR register
760  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
761  *  each containing an cause allocation for an Rx and Tx ring, and a
762  *  variable number of rows depending on the number of queues supported.
763  **/
764 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
765 			   int index, int offset)
766 {
767 	u32 ivar = array_rd32(E1000_IVAR0, index);
768 
769 	/* clear any bits that are currently set */
770 	ivar &= ~((u32)0xFF << offset);
771 
772 	/* write vector and valid bit */
773 	ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
774 
775 	array_wr32(E1000_IVAR0, index, ivar);
776 }
777 
778 #define IGB_N0_QUEUE -1
779 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
780 {
781 	struct igb_adapter *adapter = q_vector->adapter;
782 	struct e1000_hw *hw = &adapter->hw;
783 	int rx_queue = IGB_N0_QUEUE;
784 	int tx_queue = IGB_N0_QUEUE;
785 	u32 msixbm = 0;
786 
787 	if (q_vector->rx.ring)
788 		rx_queue = q_vector->rx.ring->reg_idx;
789 	if (q_vector->tx.ring)
790 		tx_queue = q_vector->tx.ring->reg_idx;
791 
792 	switch (hw->mac.type) {
793 	case e1000_82575:
794 		/* The 82575 assigns vectors using a bitmask, which matches the
795 		 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
796 		 * or more queues to a vector, we write the appropriate bits
797 		 * into the MSIXBM register for that vector.
798 		 */
799 		if (rx_queue > IGB_N0_QUEUE)
800 			msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
801 		if (tx_queue > IGB_N0_QUEUE)
802 			msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
803 		if (!adapter->msix_entries && msix_vector == 0)
804 			msixbm |= E1000_EIMS_OTHER;
805 		array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
806 		q_vector->eims_value = msixbm;
807 		break;
808 	case e1000_82576:
809 		/* 82576 uses a table that essentially consists of 2 columns
810 		 * with 8 rows.  The ordering is column-major so we use the
811 		 * lower 3 bits as the row index, and the 4th bit as the
812 		 * column offset.
813 		 */
814 		if (rx_queue > IGB_N0_QUEUE)
815 			igb_write_ivar(hw, msix_vector,
816 				       rx_queue & 0x7,
817 				       (rx_queue & 0x8) << 1);
818 		if (tx_queue > IGB_N0_QUEUE)
819 			igb_write_ivar(hw, msix_vector,
820 				       tx_queue & 0x7,
821 				       ((tx_queue & 0x8) << 1) + 8);
822 		q_vector->eims_value = 1 << msix_vector;
823 		break;
824 	case e1000_82580:
825 	case e1000_i350:
826 	case e1000_i354:
827 	case e1000_i210:
828 	case e1000_i211:
829 		/* On 82580 and newer adapters the scheme is similar to 82576
830 		 * however instead of ordering column-major we have things
831 		 * ordered row-major.  So we traverse the table by using
832 		 * bit 0 as the column offset, and the remaining bits as the
833 		 * row index.
834 		 */
835 		if (rx_queue > IGB_N0_QUEUE)
836 			igb_write_ivar(hw, msix_vector,
837 				       rx_queue >> 1,
838 				       (rx_queue & 0x1) << 4);
839 		if (tx_queue > IGB_N0_QUEUE)
840 			igb_write_ivar(hw, msix_vector,
841 				       tx_queue >> 1,
842 				       ((tx_queue & 0x1) << 4) + 8);
843 		q_vector->eims_value = 1 << msix_vector;
844 		break;
845 	default:
846 		BUG();
847 		break;
848 	}
849 
850 	/* add q_vector eims value to global eims_enable_mask */
851 	adapter->eims_enable_mask |= q_vector->eims_value;
852 
853 	/* configure q_vector to set itr on first interrupt */
854 	q_vector->set_itr = 1;
855 }
856 
857 /**
858  *  igb_configure_msix - Configure MSI-X hardware
859  *  @adapter: board private structure to initialize
860  *
861  *  igb_configure_msix sets up the hardware to properly
862  *  generate MSI-X interrupts.
863  **/
864 static void igb_configure_msix(struct igb_adapter *adapter)
865 {
866 	u32 tmp;
867 	int i, vector = 0;
868 	struct e1000_hw *hw = &adapter->hw;
869 
870 	adapter->eims_enable_mask = 0;
871 
872 	/* set vector for other causes, i.e. link changes */
873 	switch (hw->mac.type) {
874 	case e1000_82575:
875 		tmp = rd32(E1000_CTRL_EXT);
876 		/* enable MSI-X PBA support*/
877 		tmp |= E1000_CTRL_EXT_PBA_CLR;
878 
879 		/* Auto-Mask interrupts upon ICR read. */
880 		tmp |= E1000_CTRL_EXT_EIAME;
881 		tmp |= E1000_CTRL_EXT_IRCA;
882 
883 		wr32(E1000_CTRL_EXT, tmp);
884 
885 		/* enable msix_other interrupt */
886 		array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
887 		adapter->eims_other = E1000_EIMS_OTHER;
888 
889 		break;
890 
891 	case e1000_82576:
892 	case e1000_82580:
893 	case e1000_i350:
894 	case e1000_i354:
895 	case e1000_i210:
896 	case e1000_i211:
897 		/* Turn on MSI-X capability first, or our settings
898 		 * won't stick.  And it will take days to debug.
899 		 */
900 		wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
901 		     E1000_GPIE_PBA | E1000_GPIE_EIAME |
902 		     E1000_GPIE_NSICR);
903 
904 		/* enable msix_other interrupt */
905 		adapter->eims_other = 1 << vector;
906 		tmp = (vector++ | E1000_IVAR_VALID) << 8;
907 
908 		wr32(E1000_IVAR_MISC, tmp);
909 		break;
910 	default:
911 		/* do nothing, since nothing else supports MSI-X */
912 		break;
913 	} /* switch (hw->mac.type) */
914 
915 	adapter->eims_enable_mask |= adapter->eims_other;
916 
917 	for (i = 0; i < adapter->num_q_vectors; i++)
918 		igb_assign_vector(adapter->q_vector[i], vector++);
919 
920 	wrfl();
921 }
922 
923 /**
924  *  igb_request_msix - Initialize MSI-X interrupts
925  *  @adapter: board private structure to initialize
926  *
927  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
928  *  kernel.
929  **/
930 static int igb_request_msix(struct igb_adapter *adapter)
931 {
932 	struct net_device *netdev = adapter->netdev;
933 	struct e1000_hw *hw = &adapter->hw;
934 	int i, err = 0, vector = 0, free_vector = 0;
935 
936 	err = request_irq(adapter->msix_entries[vector].vector,
937 			  igb_msix_other, 0, netdev->name, adapter);
938 	if (err)
939 		goto err_out;
940 
941 	for (i = 0; i < adapter->num_q_vectors; i++) {
942 		struct igb_q_vector *q_vector = adapter->q_vector[i];
943 
944 		vector++;
945 
946 		q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
947 
948 		if (q_vector->rx.ring && q_vector->tx.ring)
949 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
950 				q_vector->rx.ring->queue_index);
951 		else if (q_vector->tx.ring)
952 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
953 				q_vector->tx.ring->queue_index);
954 		else if (q_vector->rx.ring)
955 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
956 				q_vector->rx.ring->queue_index);
957 		else
958 			sprintf(q_vector->name, "%s-unused", netdev->name);
959 
960 		err = request_irq(adapter->msix_entries[vector].vector,
961 				  igb_msix_ring, 0, q_vector->name,
962 				  q_vector);
963 		if (err)
964 			goto err_free;
965 	}
966 
967 	igb_configure_msix(adapter);
968 	return 0;
969 
970 err_free:
971 	/* free already assigned IRQs */
972 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
973 
974 	vector--;
975 	for (i = 0; i < vector; i++) {
976 		free_irq(adapter->msix_entries[free_vector++].vector,
977 			 adapter->q_vector[i]);
978 	}
979 err_out:
980 	return err;
981 }
982 
983 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
984 {
985 	if (adapter->msix_entries) {
986 		pci_disable_msix(adapter->pdev);
987 		kfree(adapter->msix_entries);
988 		adapter->msix_entries = NULL;
989 	} else if (adapter->flags & IGB_FLAG_HAS_MSI) {
990 		pci_disable_msi(adapter->pdev);
991 	}
992 }
993 
994 /**
995  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
996  *  @adapter: board private structure to initialize
997  *  @v_idx: Index of vector to be freed
998  *
999  *  This function frees the memory allocated to the q_vector.  In addition if
1000  *  NAPI is enabled it will delete any references to the NAPI struct prior
1001  *  to freeing the q_vector.
1002  **/
1003 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1004 {
1005 	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1006 
1007 	if (q_vector->tx.ring)
1008 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1009 
1010 	if (q_vector->rx.ring)
1011 		adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1012 
1013 	adapter->q_vector[v_idx] = NULL;
1014 	netif_napi_del(&q_vector->napi);
1015 
1016 	/* ixgbe_get_stats64() might access the rings on this vector,
1017 	 * we must wait a grace period before freeing it.
1018 	 */
1019 	kfree_rcu(q_vector, rcu);
1020 }
1021 
1022 /**
1023  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1024  *  @adapter: board private structure to initialize
1025  *
1026  *  This function frees the memory allocated to the q_vectors.  In addition if
1027  *  NAPI is enabled it will delete any references to the NAPI struct prior
1028  *  to freeing the q_vector.
1029  **/
1030 static void igb_free_q_vectors(struct igb_adapter *adapter)
1031 {
1032 	int v_idx = adapter->num_q_vectors;
1033 
1034 	adapter->num_tx_queues = 0;
1035 	adapter->num_rx_queues = 0;
1036 	adapter->num_q_vectors = 0;
1037 
1038 	while (v_idx--)
1039 		igb_free_q_vector(adapter, v_idx);
1040 }
1041 
1042 /**
1043  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1044  *  @adapter: board private structure to initialize
1045  *
1046  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1047  *  MSI-X interrupts allocated.
1048  */
1049 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1050 {
1051 	igb_free_q_vectors(adapter);
1052 	igb_reset_interrupt_capability(adapter);
1053 }
1054 
1055 /**
1056  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1057  *  @adapter: board private structure to initialize
1058  *  @msix: boolean value of MSIX capability
1059  *
1060  *  Attempt to configure interrupts using the best available
1061  *  capabilities of the hardware and kernel.
1062  **/
1063 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1064 {
1065 	int err;
1066 	int numvecs, i;
1067 
1068 	if (!msix)
1069 		goto msi_only;
1070 
1071 	/* Number of supported queues. */
1072 	adapter->num_rx_queues = adapter->rss_queues;
1073 	if (adapter->vfs_allocated_count)
1074 		adapter->num_tx_queues = 1;
1075 	else
1076 		adapter->num_tx_queues = adapter->rss_queues;
1077 
1078 	/* start with one vector for every Rx queue */
1079 	numvecs = adapter->num_rx_queues;
1080 
1081 	/* if Tx handler is separate add 1 for every Tx queue */
1082 	if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1083 		numvecs += adapter->num_tx_queues;
1084 
1085 	/* store the number of vectors reserved for queues */
1086 	adapter->num_q_vectors = numvecs;
1087 
1088 	/* add 1 vector for link status interrupts */
1089 	numvecs++;
1090 	adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1091 					GFP_KERNEL);
1092 
1093 	if (!adapter->msix_entries)
1094 		goto msi_only;
1095 
1096 	for (i = 0; i < numvecs; i++)
1097 		adapter->msix_entries[i].entry = i;
1098 
1099 	err = pci_enable_msix(adapter->pdev,
1100 			      adapter->msix_entries,
1101 			      numvecs);
1102 	if (err == 0)
1103 		return;
1104 
1105 	igb_reset_interrupt_capability(adapter);
1106 
1107 	/* If we can't do MSI-X, try MSI */
1108 msi_only:
1109 #ifdef CONFIG_PCI_IOV
1110 	/* disable SR-IOV for non MSI-X configurations */
1111 	if (adapter->vf_data) {
1112 		struct e1000_hw *hw = &adapter->hw;
1113 		/* disable iov and allow time for transactions to clear */
1114 		pci_disable_sriov(adapter->pdev);
1115 		msleep(500);
1116 
1117 		kfree(adapter->vf_data);
1118 		adapter->vf_data = NULL;
1119 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1120 		wrfl();
1121 		msleep(100);
1122 		dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1123 	}
1124 #endif
1125 	adapter->vfs_allocated_count = 0;
1126 	adapter->rss_queues = 1;
1127 	adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1128 	adapter->num_rx_queues = 1;
1129 	adapter->num_tx_queues = 1;
1130 	adapter->num_q_vectors = 1;
1131 	if (!pci_enable_msi(adapter->pdev))
1132 		adapter->flags |= IGB_FLAG_HAS_MSI;
1133 }
1134 
1135 static void igb_add_ring(struct igb_ring *ring,
1136 			 struct igb_ring_container *head)
1137 {
1138 	head->ring = ring;
1139 	head->count++;
1140 }
1141 
1142 /**
1143  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1144  *  @adapter: board private structure to initialize
1145  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1146  *  @v_idx: index of vector in adapter struct
1147  *  @txr_count: total number of Tx rings to allocate
1148  *  @txr_idx: index of first Tx ring to allocate
1149  *  @rxr_count: total number of Rx rings to allocate
1150  *  @rxr_idx: index of first Rx ring to allocate
1151  *
1152  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1153  **/
1154 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1155 			      int v_count, int v_idx,
1156 			      int txr_count, int txr_idx,
1157 			      int rxr_count, int rxr_idx)
1158 {
1159 	struct igb_q_vector *q_vector;
1160 	struct igb_ring *ring;
1161 	int ring_count, size;
1162 
1163 	/* igb only supports 1 Tx and/or 1 Rx queue per vector */
1164 	if (txr_count > 1 || rxr_count > 1)
1165 		return -ENOMEM;
1166 
1167 	ring_count = txr_count + rxr_count;
1168 	size = sizeof(struct igb_q_vector) +
1169 	       (sizeof(struct igb_ring) * ring_count);
1170 
1171 	/* allocate q_vector and rings */
1172 	q_vector = kzalloc(size, GFP_KERNEL);
1173 	if (!q_vector)
1174 		return -ENOMEM;
1175 
1176 	/* initialize NAPI */
1177 	netif_napi_add(adapter->netdev, &q_vector->napi,
1178 		       igb_poll, 64);
1179 
1180 	/* tie q_vector and adapter together */
1181 	adapter->q_vector[v_idx] = q_vector;
1182 	q_vector->adapter = adapter;
1183 
1184 	/* initialize work limits */
1185 	q_vector->tx.work_limit = adapter->tx_work_limit;
1186 
1187 	/* initialize ITR configuration */
1188 	q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1189 	q_vector->itr_val = IGB_START_ITR;
1190 
1191 	/* initialize pointer to rings */
1192 	ring = q_vector->ring;
1193 
1194 	/* intialize ITR */
1195 	if (rxr_count) {
1196 		/* rx or rx/tx vector */
1197 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1198 			q_vector->itr_val = adapter->rx_itr_setting;
1199 	} else {
1200 		/* tx only vector */
1201 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1202 			q_vector->itr_val = adapter->tx_itr_setting;
1203 	}
1204 
1205 	if (txr_count) {
1206 		/* assign generic ring traits */
1207 		ring->dev = &adapter->pdev->dev;
1208 		ring->netdev = adapter->netdev;
1209 
1210 		/* configure backlink on ring */
1211 		ring->q_vector = q_vector;
1212 
1213 		/* update q_vector Tx values */
1214 		igb_add_ring(ring, &q_vector->tx);
1215 
1216 		/* For 82575, context index must be unique per ring. */
1217 		if (adapter->hw.mac.type == e1000_82575)
1218 			set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1219 
1220 		/* apply Tx specific ring traits */
1221 		ring->count = adapter->tx_ring_count;
1222 		ring->queue_index = txr_idx;
1223 
1224 		/* assign ring to adapter */
1225 		adapter->tx_ring[txr_idx] = ring;
1226 
1227 		/* push pointer to next ring */
1228 		ring++;
1229 	}
1230 
1231 	if (rxr_count) {
1232 		/* assign generic ring traits */
1233 		ring->dev = &adapter->pdev->dev;
1234 		ring->netdev = adapter->netdev;
1235 
1236 		/* configure backlink on ring */
1237 		ring->q_vector = q_vector;
1238 
1239 		/* update q_vector Rx values */
1240 		igb_add_ring(ring, &q_vector->rx);
1241 
1242 		/* set flag indicating ring supports SCTP checksum offload */
1243 		if (adapter->hw.mac.type >= e1000_82576)
1244 			set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1245 
1246 		/*
1247 		 * On i350, i354, i210, and i211, loopback VLAN packets
1248 		 * have the tag byte-swapped.
1249 		 */
1250 		if (adapter->hw.mac.type >= e1000_i350)
1251 			set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1252 
1253 		/* apply Rx specific ring traits */
1254 		ring->count = adapter->rx_ring_count;
1255 		ring->queue_index = rxr_idx;
1256 
1257 		/* assign ring to adapter */
1258 		adapter->rx_ring[rxr_idx] = ring;
1259 	}
1260 
1261 	return 0;
1262 }
1263 
1264 
1265 /**
1266  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1267  *  @adapter: board private structure to initialize
1268  *
1269  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1270  *  return -ENOMEM.
1271  **/
1272 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1273 {
1274 	int q_vectors = adapter->num_q_vectors;
1275 	int rxr_remaining = adapter->num_rx_queues;
1276 	int txr_remaining = adapter->num_tx_queues;
1277 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1278 	int err;
1279 
1280 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
1281 		for (; rxr_remaining; v_idx++) {
1282 			err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1283 						 0, 0, 1, rxr_idx);
1284 
1285 			if (err)
1286 				goto err_out;
1287 
1288 			/* update counts and index */
1289 			rxr_remaining--;
1290 			rxr_idx++;
1291 		}
1292 	}
1293 
1294 	for (; v_idx < q_vectors; v_idx++) {
1295 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1296 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1297 		err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1298 					 tqpv, txr_idx, rqpv, rxr_idx);
1299 
1300 		if (err)
1301 			goto err_out;
1302 
1303 		/* update counts and index */
1304 		rxr_remaining -= rqpv;
1305 		txr_remaining -= tqpv;
1306 		rxr_idx++;
1307 		txr_idx++;
1308 	}
1309 
1310 	return 0;
1311 
1312 err_out:
1313 	adapter->num_tx_queues = 0;
1314 	adapter->num_rx_queues = 0;
1315 	adapter->num_q_vectors = 0;
1316 
1317 	while (v_idx--)
1318 		igb_free_q_vector(adapter, v_idx);
1319 
1320 	return -ENOMEM;
1321 }
1322 
1323 /**
1324  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1325  *  @adapter: board private structure to initialize
1326  *  @msix: boolean value of MSIX capability
1327  *
1328  *  This function initializes the interrupts and allocates all of the queues.
1329  **/
1330 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1331 {
1332 	struct pci_dev *pdev = adapter->pdev;
1333 	int err;
1334 
1335 	igb_set_interrupt_capability(adapter, msix);
1336 
1337 	err = igb_alloc_q_vectors(adapter);
1338 	if (err) {
1339 		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1340 		goto err_alloc_q_vectors;
1341 	}
1342 
1343 	igb_cache_ring_register(adapter);
1344 
1345 	return 0;
1346 
1347 err_alloc_q_vectors:
1348 	igb_reset_interrupt_capability(adapter);
1349 	return err;
1350 }
1351 
1352 /**
1353  *  igb_request_irq - initialize interrupts
1354  *  @adapter: board private structure to initialize
1355  *
1356  *  Attempts to configure interrupts using the best available
1357  *  capabilities of the hardware and kernel.
1358  **/
1359 static int igb_request_irq(struct igb_adapter *adapter)
1360 {
1361 	struct net_device *netdev = adapter->netdev;
1362 	struct pci_dev *pdev = adapter->pdev;
1363 	int err = 0;
1364 
1365 	if (adapter->msix_entries) {
1366 		err = igb_request_msix(adapter);
1367 		if (!err)
1368 			goto request_done;
1369 		/* fall back to MSI */
1370 		igb_free_all_tx_resources(adapter);
1371 		igb_free_all_rx_resources(adapter);
1372 
1373 		igb_clear_interrupt_scheme(adapter);
1374 		err = igb_init_interrupt_scheme(adapter, false);
1375 		if (err)
1376 			goto request_done;
1377 
1378 		igb_setup_all_tx_resources(adapter);
1379 		igb_setup_all_rx_resources(adapter);
1380 		igb_configure(adapter);
1381 	}
1382 
1383 	igb_assign_vector(adapter->q_vector[0], 0);
1384 
1385 	if (adapter->flags & IGB_FLAG_HAS_MSI) {
1386 		err = request_irq(pdev->irq, igb_intr_msi, 0,
1387 				  netdev->name, adapter);
1388 		if (!err)
1389 			goto request_done;
1390 
1391 		/* fall back to legacy interrupts */
1392 		igb_reset_interrupt_capability(adapter);
1393 		adapter->flags &= ~IGB_FLAG_HAS_MSI;
1394 	}
1395 
1396 	err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1397 			  netdev->name, adapter);
1398 
1399 	if (err)
1400 		dev_err(&pdev->dev, "Error %d getting interrupt\n",
1401 			err);
1402 
1403 request_done:
1404 	return err;
1405 }
1406 
1407 static void igb_free_irq(struct igb_adapter *adapter)
1408 {
1409 	if (adapter->msix_entries) {
1410 		int vector = 0, i;
1411 
1412 		free_irq(adapter->msix_entries[vector++].vector, adapter);
1413 
1414 		for (i = 0; i < adapter->num_q_vectors; i++)
1415 			free_irq(adapter->msix_entries[vector++].vector,
1416 				 adapter->q_vector[i]);
1417 	} else {
1418 		free_irq(adapter->pdev->irq, adapter);
1419 	}
1420 }
1421 
1422 /**
1423  *  igb_irq_disable - Mask off interrupt generation on the NIC
1424  *  @adapter: board private structure
1425  **/
1426 static void igb_irq_disable(struct igb_adapter *adapter)
1427 {
1428 	struct e1000_hw *hw = &adapter->hw;
1429 
1430 	/* we need to be careful when disabling interrupts.  The VFs are also
1431 	 * mapped into these registers and so clearing the bits can cause
1432 	 * issues on the VF drivers so we only need to clear what we set
1433 	 */
1434 	if (adapter->msix_entries) {
1435 		u32 regval = rd32(E1000_EIAM);
1436 		wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1437 		wr32(E1000_EIMC, adapter->eims_enable_mask);
1438 		regval = rd32(E1000_EIAC);
1439 		wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1440 	}
1441 
1442 	wr32(E1000_IAM, 0);
1443 	wr32(E1000_IMC, ~0);
1444 	wrfl();
1445 	if (adapter->msix_entries) {
1446 		int i;
1447 		for (i = 0; i < adapter->num_q_vectors; i++)
1448 			synchronize_irq(adapter->msix_entries[i].vector);
1449 	} else {
1450 		synchronize_irq(adapter->pdev->irq);
1451 	}
1452 }
1453 
1454 /**
1455  *  igb_irq_enable - Enable default interrupt generation settings
1456  *  @adapter: board private structure
1457  **/
1458 static void igb_irq_enable(struct igb_adapter *adapter)
1459 {
1460 	struct e1000_hw *hw = &adapter->hw;
1461 
1462 	if (adapter->msix_entries) {
1463 		u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1464 		u32 regval = rd32(E1000_EIAC);
1465 		wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1466 		regval = rd32(E1000_EIAM);
1467 		wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1468 		wr32(E1000_EIMS, adapter->eims_enable_mask);
1469 		if (adapter->vfs_allocated_count) {
1470 			wr32(E1000_MBVFIMR, 0xFF);
1471 			ims |= E1000_IMS_VMMB;
1472 		}
1473 		wr32(E1000_IMS, ims);
1474 	} else {
1475 		wr32(E1000_IMS, IMS_ENABLE_MASK |
1476 				E1000_IMS_DRSTA);
1477 		wr32(E1000_IAM, IMS_ENABLE_MASK |
1478 				E1000_IMS_DRSTA);
1479 	}
1480 }
1481 
1482 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1483 {
1484 	struct e1000_hw *hw = &adapter->hw;
1485 	u16 vid = adapter->hw.mng_cookie.vlan_id;
1486 	u16 old_vid = adapter->mng_vlan_id;
1487 
1488 	if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1489 		/* add VID to filter table */
1490 		igb_vfta_set(hw, vid, true);
1491 		adapter->mng_vlan_id = vid;
1492 	} else {
1493 		adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1494 	}
1495 
1496 	if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1497 	    (vid != old_vid) &&
1498 	    !test_bit(old_vid, adapter->active_vlans)) {
1499 		/* remove VID from filter table */
1500 		igb_vfta_set(hw, old_vid, false);
1501 	}
1502 }
1503 
1504 /**
1505  *  igb_release_hw_control - release control of the h/w to f/w
1506  *  @adapter: address of board private structure
1507  *
1508  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1509  *  For ASF and Pass Through versions of f/w this means that the
1510  *  driver is no longer loaded.
1511  **/
1512 static void igb_release_hw_control(struct igb_adapter *adapter)
1513 {
1514 	struct e1000_hw *hw = &adapter->hw;
1515 	u32 ctrl_ext;
1516 
1517 	/* Let firmware take over control of h/w */
1518 	ctrl_ext = rd32(E1000_CTRL_EXT);
1519 	wr32(E1000_CTRL_EXT,
1520 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1521 }
1522 
1523 /**
1524  *  igb_get_hw_control - get control of the h/w from f/w
1525  *  @adapter: address of board private structure
1526  *
1527  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1528  *  For ASF and Pass Through versions of f/w this means that
1529  *  the driver is loaded.
1530  **/
1531 static void igb_get_hw_control(struct igb_adapter *adapter)
1532 {
1533 	struct e1000_hw *hw = &adapter->hw;
1534 	u32 ctrl_ext;
1535 
1536 	/* Let firmware know the driver has taken over */
1537 	ctrl_ext = rd32(E1000_CTRL_EXT);
1538 	wr32(E1000_CTRL_EXT,
1539 			ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1540 }
1541 
1542 /**
1543  *  igb_configure - configure the hardware for RX and TX
1544  *  @adapter: private board structure
1545  **/
1546 static void igb_configure(struct igb_adapter *adapter)
1547 {
1548 	struct net_device *netdev = adapter->netdev;
1549 	int i;
1550 
1551 	igb_get_hw_control(adapter);
1552 	igb_set_rx_mode(netdev);
1553 
1554 	igb_restore_vlan(adapter);
1555 
1556 	igb_setup_tctl(adapter);
1557 	igb_setup_mrqc(adapter);
1558 	igb_setup_rctl(adapter);
1559 
1560 	igb_configure_tx(adapter);
1561 	igb_configure_rx(adapter);
1562 
1563 	igb_rx_fifo_flush_82575(&adapter->hw);
1564 
1565 	/* call igb_desc_unused which always leaves
1566 	 * at least 1 descriptor unused to make sure
1567 	 * next_to_use != next_to_clean
1568 	 */
1569 	for (i = 0; i < adapter->num_rx_queues; i++) {
1570 		struct igb_ring *ring = adapter->rx_ring[i];
1571 		igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1572 	}
1573 }
1574 
1575 /**
1576  *  igb_power_up_link - Power up the phy/serdes link
1577  *  @adapter: address of board private structure
1578  **/
1579 void igb_power_up_link(struct igb_adapter *adapter)
1580 {
1581 	igb_reset_phy(&adapter->hw);
1582 
1583 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1584 		igb_power_up_phy_copper(&adapter->hw);
1585 	else
1586 		igb_power_up_serdes_link_82575(&adapter->hw);
1587 }
1588 
1589 /**
1590  *  igb_power_down_link - Power down the phy/serdes link
1591  *  @adapter: address of board private structure
1592  */
1593 static void igb_power_down_link(struct igb_adapter *adapter)
1594 {
1595 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1596 		igb_power_down_phy_copper_82575(&adapter->hw);
1597 	else
1598 		igb_shutdown_serdes_link_82575(&adapter->hw);
1599 }
1600 
1601 /**
1602  *  igb_up - Open the interface and prepare it to handle traffic
1603  *  @adapter: board private structure
1604  **/
1605 int igb_up(struct igb_adapter *adapter)
1606 {
1607 	struct e1000_hw *hw = &adapter->hw;
1608 	int i;
1609 
1610 	/* hardware has been reset, we need to reload some things */
1611 	igb_configure(adapter);
1612 
1613 	clear_bit(__IGB_DOWN, &adapter->state);
1614 
1615 	for (i = 0; i < adapter->num_q_vectors; i++)
1616 		napi_enable(&(adapter->q_vector[i]->napi));
1617 
1618 	if (adapter->msix_entries)
1619 		igb_configure_msix(adapter);
1620 	else
1621 		igb_assign_vector(adapter->q_vector[0], 0);
1622 
1623 	/* Clear any pending interrupts. */
1624 	rd32(E1000_ICR);
1625 	igb_irq_enable(adapter);
1626 
1627 	/* notify VFs that reset has been completed */
1628 	if (adapter->vfs_allocated_count) {
1629 		u32 reg_data = rd32(E1000_CTRL_EXT);
1630 		reg_data |= E1000_CTRL_EXT_PFRSTD;
1631 		wr32(E1000_CTRL_EXT, reg_data);
1632 	}
1633 
1634 	netif_tx_start_all_queues(adapter->netdev);
1635 
1636 	/* start the watchdog. */
1637 	hw->mac.get_link_status = 1;
1638 	schedule_work(&adapter->watchdog_task);
1639 
1640 	return 0;
1641 }
1642 
1643 void igb_down(struct igb_adapter *adapter)
1644 {
1645 	struct net_device *netdev = adapter->netdev;
1646 	struct e1000_hw *hw = &adapter->hw;
1647 	u32 tctl, rctl;
1648 	int i;
1649 
1650 	/* signal that we're down so the interrupt handler does not
1651 	 * reschedule our watchdog timer
1652 	 */
1653 	set_bit(__IGB_DOWN, &adapter->state);
1654 
1655 	/* disable receives in the hardware */
1656 	rctl = rd32(E1000_RCTL);
1657 	wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1658 	/* flush and sleep below */
1659 
1660 	netif_tx_stop_all_queues(netdev);
1661 
1662 	/* disable transmits in the hardware */
1663 	tctl = rd32(E1000_TCTL);
1664 	tctl &= ~E1000_TCTL_EN;
1665 	wr32(E1000_TCTL, tctl);
1666 	/* flush both disables and wait for them to finish */
1667 	wrfl();
1668 	msleep(10);
1669 
1670 	igb_irq_disable(adapter);
1671 
1672 	for (i = 0; i < adapter->num_q_vectors; i++) {
1673 		napi_synchronize(&(adapter->q_vector[i]->napi));
1674 		napi_disable(&(adapter->q_vector[i]->napi));
1675 	}
1676 
1677 
1678 	del_timer_sync(&adapter->watchdog_timer);
1679 	del_timer_sync(&adapter->phy_info_timer);
1680 
1681 	netif_carrier_off(netdev);
1682 
1683 	/* record the stats before reset*/
1684 	spin_lock(&adapter->stats64_lock);
1685 	igb_update_stats(adapter, &adapter->stats64);
1686 	spin_unlock(&adapter->stats64_lock);
1687 
1688 	adapter->link_speed = 0;
1689 	adapter->link_duplex = 0;
1690 
1691 	if (!pci_channel_offline(adapter->pdev))
1692 		igb_reset(adapter);
1693 	igb_clean_all_tx_rings(adapter);
1694 	igb_clean_all_rx_rings(adapter);
1695 #ifdef CONFIG_IGB_DCA
1696 
1697 	/* since we reset the hardware DCA settings were cleared */
1698 	igb_setup_dca(adapter);
1699 #endif
1700 }
1701 
1702 void igb_reinit_locked(struct igb_adapter *adapter)
1703 {
1704 	WARN_ON(in_interrupt());
1705 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1706 		msleep(1);
1707 	igb_down(adapter);
1708 	igb_up(adapter);
1709 	clear_bit(__IGB_RESETTING, &adapter->state);
1710 }
1711 
1712 void igb_reset(struct igb_adapter *adapter)
1713 {
1714 	struct pci_dev *pdev = adapter->pdev;
1715 	struct e1000_hw *hw = &adapter->hw;
1716 	struct e1000_mac_info *mac = &hw->mac;
1717 	struct e1000_fc_info *fc = &hw->fc;
1718 	u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1719 
1720 	/* Repartition Pba for greater than 9k mtu
1721 	 * To take effect CTRL.RST is required.
1722 	 */
1723 	switch (mac->type) {
1724 	case e1000_i350:
1725 	case e1000_i354:
1726 	case e1000_82580:
1727 		pba = rd32(E1000_RXPBS);
1728 		pba = igb_rxpbs_adjust_82580(pba);
1729 		break;
1730 	case e1000_82576:
1731 		pba = rd32(E1000_RXPBS);
1732 		pba &= E1000_RXPBS_SIZE_MASK_82576;
1733 		break;
1734 	case e1000_82575:
1735 	case e1000_i210:
1736 	case e1000_i211:
1737 	default:
1738 		pba = E1000_PBA_34K;
1739 		break;
1740 	}
1741 
1742 	if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1743 	    (mac->type < e1000_82576)) {
1744 		/* adjust PBA for jumbo frames */
1745 		wr32(E1000_PBA, pba);
1746 
1747 		/* To maintain wire speed transmits, the Tx FIFO should be
1748 		 * large enough to accommodate two full transmit packets,
1749 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
1750 		 * the Rx FIFO should be large enough to accommodate at least
1751 		 * one full receive packet and is similarly rounded up and
1752 		 * expressed in KB.
1753 		 */
1754 		pba = rd32(E1000_PBA);
1755 		/* upper 16 bits has Tx packet buffer allocation size in KB */
1756 		tx_space = pba >> 16;
1757 		/* lower 16 bits has Rx packet buffer allocation size in KB */
1758 		pba &= 0xffff;
1759 		/* the Tx fifo also stores 16 bytes of information about the Tx
1760 		 * but don't include ethernet FCS because hardware appends it
1761 		 */
1762 		min_tx_space = (adapter->max_frame_size +
1763 				sizeof(union e1000_adv_tx_desc) -
1764 				ETH_FCS_LEN) * 2;
1765 		min_tx_space = ALIGN(min_tx_space, 1024);
1766 		min_tx_space >>= 10;
1767 		/* software strips receive CRC, so leave room for it */
1768 		min_rx_space = adapter->max_frame_size;
1769 		min_rx_space = ALIGN(min_rx_space, 1024);
1770 		min_rx_space >>= 10;
1771 
1772 		/* If current Tx allocation is less than the min Tx FIFO size,
1773 		 * and the min Tx FIFO size is less than the current Rx FIFO
1774 		 * allocation, take space away from current Rx allocation
1775 		 */
1776 		if (tx_space < min_tx_space &&
1777 		    ((min_tx_space - tx_space) < pba)) {
1778 			pba = pba - (min_tx_space - tx_space);
1779 
1780 			/* if short on Rx space, Rx wins and must trump Tx
1781 			 * adjustment
1782 			 */
1783 			if (pba < min_rx_space)
1784 				pba = min_rx_space;
1785 		}
1786 		wr32(E1000_PBA, pba);
1787 	}
1788 
1789 	/* flow control settings */
1790 	/* The high water mark must be low enough to fit one full frame
1791 	 * (or the size used for early receive) above it in the Rx FIFO.
1792 	 * Set it to the lower of:
1793 	 * - 90% of the Rx FIFO size, or
1794 	 * - the full Rx FIFO size minus one full frame
1795 	 */
1796 	hwm = min(((pba << 10) * 9 / 10),
1797 			((pba << 10) - 2 * adapter->max_frame_size));
1798 
1799 	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
1800 	fc->low_water = fc->high_water - 16;
1801 	fc->pause_time = 0xFFFF;
1802 	fc->send_xon = 1;
1803 	fc->current_mode = fc->requested_mode;
1804 
1805 	/* disable receive for all VFs and wait one second */
1806 	if (adapter->vfs_allocated_count) {
1807 		int i;
1808 		for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1809 			adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1810 
1811 		/* ping all the active vfs to let them know we are going down */
1812 		igb_ping_all_vfs(adapter);
1813 
1814 		/* disable transmits and receives */
1815 		wr32(E1000_VFRE, 0);
1816 		wr32(E1000_VFTE, 0);
1817 	}
1818 
1819 	/* Allow time for pending master requests to run */
1820 	hw->mac.ops.reset_hw(hw);
1821 	wr32(E1000_WUC, 0);
1822 
1823 	if (hw->mac.ops.init_hw(hw))
1824 		dev_err(&pdev->dev, "Hardware Error\n");
1825 
1826 	/* Flow control settings reset on hardware reset, so guarantee flow
1827 	 * control is off when forcing speed.
1828 	 */
1829 	if (!hw->mac.autoneg)
1830 		igb_force_mac_fc(hw);
1831 
1832 	igb_init_dmac(adapter, pba);
1833 #ifdef CONFIG_IGB_HWMON
1834 	/* Re-initialize the thermal sensor on i350 devices. */
1835 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
1836 		if (mac->type == e1000_i350 && hw->bus.func == 0) {
1837 			/* If present, re-initialize the external thermal sensor
1838 			 * interface.
1839 			 */
1840 			if (adapter->ets)
1841 				mac->ops.init_thermal_sensor_thresh(hw);
1842 		}
1843 	}
1844 #endif
1845 	if (!netif_running(adapter->netdev))
1846 		igb_power_down_link(adapter);
1847 
1848 	igb_update_mng_vlan(adapter);
1849 
1850 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1851 	wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1852 
1853 	/* Re-enable PTP, where applicable. */
1854 	igb_ptp_reset(adapter);
1855 
1856 	igb_get_phy_info(hw);
1857 }
1858 
1859 static netdev_features_t igb_fix_features(struct net_device *netdev,
1860 	netdev_features_t features)
1861 {
1862 	/* Since there is no support for separate Rx/Tx vlan accel
1863 	 * enable/disable make sure Tx flag is always in same state as Rx.
1864 	 */
1865 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
1866 		features |= NETIF_F_HW_VLAN_CTAG_TX;
1867 	else
1868 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1869 
1870 	return features;
1871 }
1872 
1873 static int igb_set_features(struct net_device *netdev,
1874 	netdev_features_t features)
1875 {
1876 	netdev_features_t changed = netdev->features ^ features;
1877 	struct igb_adapter *adapter = netdev_priv(netdev);
1878 
1879 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1880 		igb_vlan_mode(netdev, features);
1881 
1882 	if (!(changed & NETIF_F_RXALL))
1883 		return 0;
1884 
1885 	netdev->features = features;
1886 
1887 	if (netif_running(netdev))
1888 		igb_reinit_locked(adapter);
1889 	else
1890 		igb_reset(adapter);
1891 
1892 	return 0;
1893 }
1894 
1895 static const struct net_device_ops igb_netdev_ops = {
1896 	.ndo_open		= igb_open,
1897 	.ndo_stop		= igb_close,
1898 	.ndo_start_xmit		= igb_xmit_frame,
1899 	.ndo_get_stats64	= igb_get_stats64,
1900 	.ndo_set_rx_mode	= igb_set_rx_mode,
1901 	.ndo_set_mac_address	= igb_set_mac,
1902 	.ndo_change_mtu		= igb_change_mtu,
1903 	.ndo_do_ioctl		= igb_ioctl,
1904 	.ndo_tx_timeout		= igb_tx_timeout,
1905 	.ndo_validate_addr	= eth_validate_addr,
1906 	.ndo_vlan_rx_add_vid	= igb_vlan_rx_add_vid,
1907 	.ndo_vlan_rx_kill_vid	= igb_vlan_rx_kill_vid,
1908 	.ndo_set_vf_mac		= igb_ndo_set_vf_mac,
1909 	.ndo_set_vf_vlan	= igb_ndo_set_vf_vlan,
1910 	.ndo_set_vf_tx_rate	= igb_ndo_set_vf_bw,
1911 	.ndo_set_vf_spoofchk	= igb_ndo_set_vf_spoofchk,
1912 	.ndo_get_vf_config	= igb_ndo_get_vf_config,
1913 #ifdef CONFIG_NET_POLL_CONTROLLER
1914 	.ndo_poll_controller	= igb_netpoll,
1915 #endif
1916 	.ndo_fix_features	= igb_fix_features,
1917 	.ndo_set_features	= igb_set_features,
1918 };
1919 
1920 /**
1921  * igb_set_fw_version - Configure version string for ethtool
1922  * @adapter: adapter struct
1923  **/
1924 void igb_set_fw_version(struct igb_adapter *adapter)
1925 {
1926 	struct e1000_hw *hw = &adapter->hw;
1927 	struct e1000_fw_version fw;
1928 
1929 	igb_get_fw_version(hw, &fw);
1930 
1931 	switch (hw->mac.type) {
1932 	case e1000_i211:
1933 		snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1934 			 "%2d.%2d-%d",
1935 			 fw.invm_major, fw.invm_minor, fw.invm_img_type);
1936 		break;
1937 
1938 	default:
1939 		/* if option is rom valid, display its version too */
1940 		if (fw.or_valid) {
1941 			snprintf(adapter->fw_version,
1942 				 sizeof(adapter->fw_version),
1943 				 "%d.%d, 0x%08x, %d.%d.%d",
1944 				 fw.eep_major, fw.eep_minor, fw.etrack_id,
1945 				 fw.or_major, fw.or_build, fw.or_patch);
1946 		/* no option rom */
1947 		} else {
1948 			snprintf(adapter->fw_version,
1949 				 sizeof(adapter->fw_version),
1950 				 "%d.%d, 0x%08x",
1951 				 fw.eep_major, fw.eep_minor, fw.etrack_id);
1952 		}
1953 		break;
1954 	}
1955 	return;
1956 }
1957 
1958 /**
1959  *  igb_init_i2c - Init I2C interface
1960  *  @adapter: pointer to adapter structure
1961  **/
1962 static s32 igb_init_i2c(struct igb_adapter *adapter)
1963 {
1964 	s32 status = E1000_SUCCESS;
1965 
1966 	/* I2C interface supported on i350 devices */
1967 	if (adapter->hw.mac.type != e1000_i350)
1968 		return E1000_SUCCESS;
1969 
1970 	/* Initialize the i2c bus which is controlled by the registers.
1971 	 * This bus will use the i2c_algo_bit structue that implements
1972 	 * the protocol through toggling of the 4 bits in the register.
1973 	 */
1974 	adapter->i2c_adap.owner = THIS_MODULE;
1975 	adapter->i2c_algo = igb_i2c_algo;
1976 	adapter->i2c_algo.data = adapter;
1977 	adapter->i2c_adap.algo_data = &adapter->i2c_algo;
1978 	adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
1979 	strlcpy(adapter->i2c_adap.name, "igb BB",
1980 		sizeof(adapter->i2c_adap.name));
1981 	status = i2c_bit_add_bus(&adapter->i2c_adap);
1982 	return status;
1983 }
1984 
1985 /**
1986  *  igb_probe - Device Initialization Routine
1987  *  @pdev: PCI device information struct
1988  *  @ent: entry in igb_pci_tbl
1989  *
1990  *  Returns 0 on success, negative on failure
1991  *
1992  *  igb_probe initializes an adapter identified by a pci_dev structure.
1993  *  The OS initialization, configuring of the adapter private structure,
1994  *  and a hardware reset occur.
1995  **/
1996 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1997 {
1998 	struct net_device *netdev;
1999 	struct igb_adapter *adapter;
2000 	struct e1000_hw *hw;
2001 	u16 eeprom_data = 0;
2002 	s32 ret_val;
2003 	static int global_quad_port_a; /* global quad port a indication */
2004 	const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2005 	unsigned long mmio_start, mmio_len;
2006 	int err, pci_using_dac;
2007 	u8 part_str[E1000_PBANUM_LENGTH];
2008 
2009 	/* Catch broken hardware that put the wrong VF device ID in
2010 	 * the PCIe SR-IOV capability.
2011 	 */
2012 	if (pdev->is_virtfn) {
2013 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2014 			pci_name(pdev), pdev->vendor, pdev->device);
2015 		return -EINVAL;
2016 	}
2017 
2018 	err = pci_enable_device_mem(pdev);
2019 	if (err)
2020 		return err;
2021 
2022 	pci_using_dac = 0;
2023 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2024 	if (!err) {
2025 		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2026 		if (!err)
2027 			pci_using_dac = 1;
2028 	} else {
2029 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2030 		if (err) {
2031 			err = dma_set_coherent_mask(&pdev->dev,
2032 						    DMA_BIT_MASK(32));
2033 			if (err) {
2034 				dev_err(&pdev->dev,
2035 					"No usable DMA configuration, aborting\n");
2036 				goto err_dma;
2037 			}
2038 		}
2039 	}
2040 
2041 	err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2042 					   IORESOURCE_MEM),
2043 					   igb_driver_name);
2044 	if (err)
2045 		goto err_pci_reg;
2046 
2047 	pci_enable_pcie_error_reporting(pdev);
2048 
2049 	pci_set_master(pdev);
2050 	pci_save_state(pdev);
2051 
2052 	err = -ENOMEM;
2053 	netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2054 				   IGB_MAX_TX_QUEUES);
2055 	if (!netdev)
2056 		goto err_alloc_etherdev;
2057 
2058 	SET_NETDEV_DEV(netdev, &pdev->dev);
2059 
2060 	pci_set_drvdata(pdev, netdev);
2061 	adapter = netdev_priv(netdev);
2062 	adapter->netdev = netdev;
2063 	adapter->pdev = pdev;
2064 	hw = &adapter->hw;
2065 	hw->back = adapter;
2066 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2067 
2068 	mmio_start = pci_resource_start(pdev, 0);
2069 	mmio_len = pci_resource_len(pdev, 0);
2070 
2071 	err = -EIO;
2072 	hw->hw_addr = ioremap(mmio_start, mmio_len);
2073 	if (!hw->hw_addr)
2074 		goto err_ioremap;
2075 
2076 	netdev->netdev_ops = &igb_netdev_ops;
2077 	igb_set_ethtool_ops(netdev);
2078 	netdev->watchdog_timeo = 5 * HZ;
2079 
2080 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2081 
2082 	netdev->mem_start = mmio_start;
2083 	netdev->mem_end = mmio_start + mmio_len;
2084 
2085 	/* PCI config space info */
2086 	hw->vendor_id = pdev->vendor;
2087 	hw->device_id = pdev->device;
2088 	hw->revision_id = pdev->revision;
2089 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2090 	hw->subsystem_device_id = pdev->subsystem_device;
2091 
2092 	/* Copy the default MAC, PHY and NVM function pointers */
2093 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2094 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2095 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2096 	/* Initialize skew-specific constants */
2097 	err = ei->get_invariants(hw);
2098 	if (err)
2099 		goto err_sw_init;
2100 
2101 	/* setup the private structure */
2102 	err = igb_sw_init(adapter);
2103 	if (err)
2104 		goto err_sw_init;
2105 
2106 	igb_get_bus_info_pcie(hw);
2107 
2108 	hw->phy.autoneg_wait_to_complete = false;
2109 
2110 	/* Copper options */
2111 	if (hw->phy.media_type == e1000_media_type_copper) {
2112 		hw->phy.mdix = AUTO_ALL_MODES;
2113 		hw->phy.disable_polarity_correction = false;
2114 		hw->phy.ms_type = e1000_ms_hw_default;
2115 	}
2116 
2117 	if (igb_check_reset_block(hw))
2118 		dev_info(&pdev->dev,
2119 			"PHY reset is blocked due to SOL/IDER session.\n");
2120 
2121 	/* features is initialized to 0 in allocation, it might have bits
2122 	 * set by igb_sw_init so we should use an or instead of an
2123 	 * assignment.
2124 	 */
2125 	netdev->features |= NETIF_F_SG |
2126 			    NETIF_F_IP_CSUM |
2127 			    NETIF_F_IPV6_CSUM |
2128 			    NETIF_F_TSO |
2129 			    NETIF_F_TSO6 |
2130 			    NETIF_F_RXHASH |
2131 			    NETIF_F_RXCSUM |
2132 			    NETIF_F_HW_VLAN_CTAG_RX |
2133 			    NETIF_F_HW_VLAN_CTAG_TX;
2134 
2135 	/* copy netdev features into list of user selectable features */
2136 	netdev->hw_features |= netdev->features;
2137 	netdev->hw_features |= NETIF_F_RXALL;
2138 
2139 	/* set this bit last since it cannot be part of hw_features */
2140 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2141 
2142 	netdev->vlan_features |= NETIF_F_TSO |
2143 				 NETIF_F_TSO6 |
2144 				 NETIF_F_IP_CSUM |
2145 				 NETIF_F_IPV6_CSUM |
2146 				 NETIF_F_SG;
2147 
2148 	netdev->priv_flags |= IFF_SUPP_NOFCS;
2149 
2150 	if (pci_using_dac) {
2151 		netdev->features |= NETIF_F_HIGHDMA;
2152 		netdev->vlan_features |= NETIF_F_HIGHDMA;
2153 	}
2154 
2155 	if (hw->mac.type >= e1000_82576) {
2156 		netdev->hw_features |= NETIF_F_SCTP_CSUM;
2157 		netdev->features |= NETIF_F_SCTP_CSUM;
2158 	}
2159 
2160 	netdev->priv_flags |= IFF_UNICAST_FLT;
2161 
2162 	adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2163 
2164 	/* before reading the NVM, reset the controller to put the device in a
2165 	 * known good starting state
2166 	 */
2167 	hw->mac.ops.reset_hw(hw);
2168 
2169 	/* make sure the NVM is good , i211 parts have special NVM that
2170 	 * doesn't contain a checksum
2171 	 */
2172 	if (hw->mac.type != e1000_i211) {
2173 		if (hw->nvm.ops.validate(hw) < 0) {
2174 			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2175 			err = -EIO;
2176 			goto err_eeprom;
2177 		}
2178 	}
2179 
2180 	/* copy the MAC address out of the NVM */
2181 	if (hw->mac.ops.read_mac_addr(hw))
2182 		dev_err(&pdev->dev, "NVM Read Error\n");
2183 
2184 	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2185 
2186 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2187 		dev_err(&pdev->dev, "Invalid MAC Address\n");
2188 		err = -EIO;
2189 		goto err_eeprom;
2190 	}
2191 
2192 	/* get firmware version for ethtool -i */
2193 	igb_set_fw_version(adapter);
2194 
2195 	setup_timer(&adapter->watchdog_timer, igb_watchdog,
2196 		    (unsigned long) adapter);
2197 	setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2198 		    (unsigned long) adapter);
2199 
2200 	INIT_WORK(&adapter->reset_task, igb_reset_task);
2201 	INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2202 
2203 	/* Initialize link properties that are user-changeable */
2204 	adapter->fc_autoneg = true;
2205 	hw->mac.autoneg = true;
2206 	hw->phy.autoneg_advertised = 0x2f;
2207 
2208 	hw->fc.requested_mode = e1000_fc_default;
2209 	hw->fc.current_mode = e1000_fc_default;
2210 
2211 	igb_validate_mdi_setting(hw);
2212 
2213 	/* By default, support wake on port A */
2214 	if (hw->bus.func == 0)
2215 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2216 
2217 	/* Check the NVM for wake support on non-port A ports */
2218 	if (hw->mac.type >= e1000_82580)
2219 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2220 				 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2221 				 &eeprom_data);
2222 	else if (hw->bus.func == 1)
2223 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2224 
2225 	if (eeprom_data & IGB_EEPROM_APME)
2226 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2227 
2228 	/* now that we have the eeprom settings, apply the special cases where
2229 	 * the eeprom may be wrong or the board simply won't support wake on
2230 	 * lan on a particular port
2231 	 */
2232 	switch (pdev->device) {
2233 	case E1000_DEV_ID_82575GB_QUAD_COPPER:
2234 		adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2235 		break;
2236 	case E1000_DEV_ID_82575EB_FIBER_SERDES:
2237 	case E1000_DEV_ID_82576_FIBER:
2238 	case E1000_DEV_ID_82576_SERDES:
2239 		/* Wake events only supported on port A for dual fiber
2240 		 * regardless of eeprom setting
2241 		 */
2242 		if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2243 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2244 		break;
2245 	case E1000_DEV_ID_82576_QUAD_COPPER:
2246 	case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2247 		/* if quad port adapter, disable WoL on all but port A */
2248 		if (global_quad_port_a != 0)
2249 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2250 		else
2251 			adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2252 		/* Reset for multiple quad port adapters */
2253 		if (++global_quad_port_a == 4)
2254 			global_quad_port_a = 0;
2255 		break;
2256 	default:
2257 		/* If the device can't wake, don't set software support */
2258 		if (!device_can_wakeup(&adapter->pdev->dev))
2259 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2260 	}
2261 
2262 	/* initialize the wol settings based on the eeprom settings */
2263 	if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2264 		adapter->wol |= E1000_WUFC_MAG;
2265 
2266 	/* Some vendors want WoL disabled by default, but still supported */
2267 	if ((hw->mac.type == e1000_i350) &&
2268 	    (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2269 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2270 		adapter->wol = 0;
2271 	}
2272 
2273 	device_set_wakeup_enable(&adapter->pdev->dev,
2274 				 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2275 
2276 	/* reset the hardware with the new settings */
2277 	igb_reset(adapter);
2278 
2279 	/* Init the I2C interface */
2280 	err = igb_init_i2c(adapter);
2281 	if (err) {
2282 		dev_err(&pdev->dev, "failed to init i2c interface\n");
2283 		goto err_eeprom;
2284 	}
2285 
2286 	/* let the f/w know that the h/w is now under the control of the
2287 	 * driver. */
2288 	igb_get_hw_control(adapter);
2289 
2290 	strcpy(netdev->name, "eth%d");
2291 	err = register_netdev(netdev);
2292 	if (err)
2293 		goto err_register;
2294 
2295 	/* carrier off reporting is important to ethtool even BEFORE open */
2296 	netif_carrier_off(netdev);
2297 
2298 #ifdef CONFIG_IGB_DCA
2299 	if (dca_add_requester(&pdev->dev) == 0) {
2300 		adapter->flags |= IGB_FLAG_DCA_ENABLED;
2301 		dev_info(&pdev->dev, "DCA enabled\n");
2302 		igb_setup_dca(adapter);
2303 	}
2304 
2305 #endif
2306 #ifdef CONFIG_IGB_HWMON
2307 	/* Initialize the thermal sensor on i350 devices. */
2308 	if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2309 		u16 ets_word;
2310 
2311 		/* Read the NVM to determine if this i350 device supports an
2312 		 * external thermal sensor.
2313 		 */
2314 		hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2315 		if (ets_word != 0x0000 && ets_word != 0xFFFF)
2316 			adapter->ets = true;
2317 		else
2318 			adapter->ets = false;
2319 		if (igb_sysfs_init(adapter))
2320 			dev_err(&pdev->dev,
2321 				"failed to allocate sysfs resources\n");
2322 	} else {
2323 		adapter->ets = false;
2324 	}
2325 #endif
2326 	/* do hw tstamp init after resetting */
2327 	igb_ptp_init(adapter);
2328 
2329 	dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2330 	/* print bus type/speed/width info, not applicable to i354 */
2331 	if (hw->mac.type != e1000_i354) {
2332 		dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2333 			 netdev->name,
2334 			 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2335 			  (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2336 			   "unknown"),
2337 			 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2338 			  "Width x4" :
2339 			  (hw->bus.width == e1000_bus_width_pcie_x2) ?
2340 			  "Width x2" :
2341 			  (hw->bus.width == e1000_bus_width_pcie_x1) ?
2342 			  "Width x1" : "unknown"), netdev->dev_addr);
2343 	}
2344 
2345 	ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2346 	if (ret_val)
2347 		strcpy(part_str, "Unknown");
2348 	dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2349 	dev_info(&pdev->dev,
2350 		"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2351 		adapter->msix_entries ? "MSI-X" :
2352 		(adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2353 		adapter->num_rx_queues, adapter->num_tx_queues);
2354 	switch (hw->mac.type) {
2355 	case e1000_i350:
2356 	case e1000_i210:
2357 	case e1000_i211:
2358 		igb_set_eee_i350(hw);
2359 		break;
2360 	case e1000_i354:
2361 		if (hw->phy.media_type == e1000_media_type_copper) {
2362 			if ((rd32(E1000_CTRL_EXT) &
2363 			    E1000_CTRL_EXT_LINK_MODE_SGMII))
2364 				igb_set_eee_i354(hw);
2365 		}
2366 		break;
2367 	default:
2368 		break;
2369 	}
2370 
2371 	pm_runtime_put_noidle(&pdev->dev);
2372 	return 0;
2373 
2374 err_register:
2375 	igb_release_hw_control(adapter);
2376 	memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2377 err_eeprom:
2378 	if (!igb_check_reset_block(hw))
2379 		igb_reset_phy(hw);
2380 
2381 	if (hw->flash_address)
2382 		iounmap(hw->flash_address);
2383 err_sw_init:
2384 	igb_clear_interrupt_scheme(adapter);
2385 	iounmap(hw->hw_addr);
2386 err_ioremap:
2387 	free_netdev(netdev);
2388 err_alloc_etherdev:
2389 	pci_release_selected_regions(pdev,
2390 				     pci_select_bars(pdev, IORESOURCE_MEM));
2391 err_pci_reg:
2392 err_dma:
2393 	pci_disable_device(pdev);
2394 	return err;
2395 }
2396 
2397 #ifdef CONFIG_PCI_IOV
2398 static int  igb_disable_sriov(struct pci_dev *pdev)
2399 {
2400 	struct net_device *netdev = pci_get_drvdata(pdev);
2401 	struct igb_adapter *adapter = netdev_priv(netdev);
2402 	struct e1000_hw *hw = &adapter->hw;
2403 
2404 	/* reclaim resources allocated to VFs */
2405 	if (adapter->vf_data) {
2406 		/* disable iov and allow time for transactions to clear */
2407 		if (pci_vfs_assigned(pdev)) {
2408 			dev_warn(&pdev->dev,
2409 				 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2410 			return -EPERM;
2411 		} else {
2412 			pci_disable_sriov(pdev);
2413 			msleep(500);
2414 		}
2415 
2416 		kfree(adapter->vf_data);
2417 		adapter->vf_data = NULL;
2418 		adapter->vfs_allocated_count = 0;
2419 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2420 		wrfl();
2421 		msleep(100);
2422 		dev_info(&pdev->dev, "IOV Disabled\n");
2423 
2424 		/* Re-enable DMA Coalescing flag since IOV is turned off */
2425 		adapter->flags |= IGB_FLAG_DMAC;
2426 	}
2427 
2428 	return 0;
2429 }
2430 
2431 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2432 {
2433 	struct net_device *netdev = pci_get_drvdata(pdev);
2434 	struct igb_adapter *adapter = netdev_priv(netdev);
2435 	int old_vfs = pci_num_vf(pdev);
2436 	int err = 0;
2437 	int i;
2438 
2439 	if (!num_vfs)
2440 		goto out;
2441 	else if (old_vfs && old_vfs == num_vfs)
2442 		goto out;
2443 	else if (old_vfs && old_vfs != num_vfs)
2444 		err = igb_disable_sriov(pdev);
2445 
2446 	if (err)
2447 		goto out;
2448 
2449 	if (num_vfs > 7) {
2450 		err = -EPERM;
2451 		goto out;
2452 	}
2453 
2454 	adapter->vfs_allocated_count = num_vfs;
2455 
2456 	adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2457 				sizeof(struct vf_data_storage), GFP_KERNEL);
2458 
2459 	/* if allocation failed then we do not support SR-IOV */
2460 	if (!adapter->vf_data) {
2461 		adapter->vfs_allocated_count = 0;
2462 		dev_err(&pdev->dev,
2463 			"Unable to allocate memory for VF Data Storage\n");
2464 		err = -ENOMEM;
2465 		goto out;
2466 	}
2467 
2468 	err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2469 	if (err)
2470 		goto err_out;
2471 
2472 	dev_info(&pdev->dev, "%d VFs allocated\n",
2473 		 adapter->vfs_allocated_count);
2474 	for (i = 0; i < adapter->vfs_allocated_count; i++)
2475 		igb_vf_configure(adapter, i);
2476 
2477 	/* DMA Coalescing is not supported in IOV mode. */
2478 	adapter->flags &= ~IGB_FLAG_DMAC;
2479 	goto out;
2480 
2481 err_out:
2482 	kfree(adapter->vf_data);
2483 	adapter->vf_data = NULL;
2484 	adapter->vfs_allocated_count = 0;
2485 out:
2486 	return err;
2487 }
2488 
2489 #endif
2490 /**
2491  *  igb_remove_i2c - Cleanup  I2C interface
2492  *  @adapter: pointer to adapter structure
2493  **/
2494 static void igb_remove_i2c(struct igb_adapter *adapter)
2495 {
2496 	/* free the adapter bus structure */
2497 	i2c_del_adapter(&adapter->i2c_adap);
2498 }
2499 
2500 /**
2501  *  igb_remove - Device Removal Routine
2502  *  @pdev: PCI device information struct
2503  *
2504  *  igb_remove is called by the PCI subsystem to alert the driver
2505  *  that it should release a PCI device.  The could be caused by a
2506  *  Hot-Plug event, or because the driver is going to be removed from
2507  *  memory.
2508  **/
2509 static void igb_remove(struct pci_dev *pdev)
2510 {
2511 	struct net_device *netdev = pci_get_drvdata(pdev);
2512 	struct igb_adapter *adapter = netdev_priv(netdev);
2513 	struct e1000_hw *hw = &adapter->hw;
2514 
2515 	pm_runtime_get_noresume(&pdev->dev);
2516 #ifdef CONFIG_IGB_HWMON
2517 	igb_sysfs_exit(adapter);
2518 #endif
2519 	igb_remove_i2c(adapter);
2520 	igb_ptp_stop(adapter);
2521 	/* The watchdog timer may be rescheduled, so explicitly
2522 	 * disable watchdog from being rescheduled.
2523 	 */
2524 	set_bit(__IGB_DOWN, &adapter->state);
2525 	del_timer_sync(&adapter->watchdog_timer);
2526 	del_timer_sync(&adapter->phy_info_timer);
2527 
2528 	cancel_work_sync(&adapter->reset_task);
2529 	cancel_work_sync(&adapter->watchdog_task);
2530 
2531 #ifdef CONFIG_IGB_DCA
2532 	if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2533 		dev_info(&pdev->dev, "DCA disabled\n");
2534 		dca_remove_requester(&pdev->dev);
2535 		adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2536 		wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2537 	}
2538 #endif
2539 
2540 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
2541 	 * would have already happened in close and is redundant.
2542 	 */
2543 	igb_release_hw_control(adapter);
2544 
2545 	unregister_netdev(netdev);
2546 
2547 	igb_clear_interrupt_scheme(adapter);
2548 
2549 #ifdef CONFIG_PCI_IOV
2550 	igb_disable_sriov(pdev);
2551 #endif
2552 
2553 	iounmap(hw->hw_addr);
2554 	if (hw->flash_address)
2555 		iounmap(hw->flash_address);
2556 	pci_release_selected_regions(pdev,
2557 				     pci_select_bars(pdev, IORESOURCE_MEM));
2558 
2559 	kfree(adapter->shadow_vfta);
2560 	free_netdev(netdev);
2561 
2562 	pci_disable_pcie_error_reporting(pdev);
2563 
2564 	pci_disable_device(pdev);
2565 }
2566 
2567 /**
2568  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2569  *  @adapter: board private structure to initialize
2570  *
2571  *  This function initializes the vf specific data storage and then attempts to
2572  *  allocate the VFs.  The reason for ordering it this way is because it is much
2573  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
2574  *  the memory for the VFs.
2575  **/
2576 static void igb_probe_vfs(struct igb_adapter *adapter)
2577 {
2578 #ifdef CONFIG_PCI_IOV
2579 	struct pci_dev *pdev = adapter->pdev;
2580 	struct e1000_hw *hw = &adapter->hw;
2581 
2582 	/* Virtualization features not supported on i210 family. */
2583 	if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2584 		return;
2585 
2586 	pci_sriov_set_totalvfs(pdev, 7);
2587 	igb_enable_sriov(pdev, max_vfs);
2588 
2589 #endif /* CONFIG_PCI_IOV */
2590 }
2591 
2592 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2593 {
2594 	struct e1000_hw *hw = &adapter->hw;
2595 	u32 max_rss_queues;
2596 
2597 	/* Determine the maximum number of RSS queues supported. */
2598 	switch (hw->mac.type) {
2599 	case e1000_i211:
2600 		max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2601 		break;
2602 	case e1000_82575:
2603 	case e1000_i210:
2604 		max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2605 		break;
2606 	case e1000_i350:
2607 		/* I350 cannot do RSS and SR-IOV at the same time */
2608 		if (!!adapter->vfs_allocated_count) {
2609 			max_rss_queues = 1;
2610 			break;
2611 		}
2612 		/* fall through */
2613 	case e1000_82576:
2614 		if (!!adapter->vfs_allocated_count) {
2615 			max_rss_queues = 2;
2616 			break;
2617 		}
2618 		/* fall through */
2619 	case e1000_82580:
2620 	case e1000_i354:
2621 	default:
2622 		max_rss_queues = IGB_MAX_RX_QUEUES;
2623 		break;
2624 	}
2625 
2626 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2627 
2628 	/* Determine if we need to pair queues. */
2629 	switch (hw->mac.type) {
2630 	case e1000_82575:
2631 	case e1000_i211:
2632 		/* Device supports enough interrupts without queue pairing. */
2633 		break;
2634 	case e1000_82576:
2635 		/* If VFs are going to be allocated with RSS queues then we
2636 		 * should pair the queues in order to conserve interrupts due
2637 		 * to limited supply.
2638 		 */
2639 		if ((adapter->rss_queues > 1) &&
2640 		    (adapter->vfs_allocated_count > 6))
2641 			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2642 		/* fall through */
2643 	case e1000_82580:
2644 	case e1000_i350:
2645 	case e1000_i354:
2646 	case e1000_i210:
2647 	default:
2648 		/* If rss_queues > half of max_rss_queues, pair the queues in
2649 		 * order to conserve interrupts due to limited supply.
2650 		 */
2651 		if (adapter->rss_queues > (max_rss_queues / 2))
2652 			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2653 		break;
2654 	}
2655 }
2656 
2657 /**
2658  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
2659  *  @adapter: board private structure to initialize
2660  *
2661  *  igb_sw_init initializes the Adapter private data structure.
2662  *  Fields are initialized based on PCI device information and
2663  *  OS network device settings (MTU size).
2664  **/
2665 static int igb_sw_init(struct igb_adapter *adapter)
2666 {
2667 	struct e1000_hw *hw = &adapter->hw;
2668 	struct net_device *netdev = adapter->netdev;
2669 	struct pci_dev *pdev = adapter->pdev;
2670 
2671 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2672 
2673 	/* set default ring sizes */
2674 	adapter->tx_ring_count = IGB_DEFAULT_TXD;
2675 	adapter->rx_ring_count = IGB_DEFAULT_RXD;
2676 
2677 	/* set default ITR values */
2678 	adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2679 	adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2680 
2681 	/* set default work limits */
2682 	adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2683 
2684 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2685 				  VLAN_HLEN;
2686 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2687 
2688 	spin_lock_init(&adapter->stats64_lock);
2689 #ifdef CONFIG_PCI_IOV
2690 	switch (hw->mac.type) {
2691 	case e1000_82576:
2692 	case e1000_i350:
2693 		if (max_vfs > 7) {
2694 			dev_warn(&pdev->dev,
2695 				 "Maximum of 7 VFs per PF, using max\n");
2696 			max_vfs = adapter->vfs_allocated_count = 7;
2697 		} else
2698 			adapter->vfs_allocated_count = max_vfs;
2699 		if (adapter->vfs_allocated_count)
2700 			dev_warn(&pdev->dev,
2701 				 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2702 		break;
2703 	default:
2704 		break;
2705 	}
2706 #endif /* CONFIG_PCI_IOV */
2707 
2708 	igb_init_queue_configuration(adapter);
2709 
2710 	/* Setup and initialize a copy of the hw vlan table array */
2711 	adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2712 				       GFP_ATOMIC);
2713 
2714 	/* This call may decrease the number of queues */
2715 	if (igb_init_interrupt_scheme(adapter, true)) {
2716 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2717 		return -ENOMEM;
2718 	}
2719 
2720 	igb_probe_vfs(adapter);
2721 
2722 	/* Explicitly disable IRQ since the NIC can be in any state. */
2723 	igb_irq_disable(adapter);
2724 
2725 	if (hw->mac.type >= e1000_i350)
2726 		adapter->flags &= ~IGB_FLAG_DMAC;
2727 
2728 	set_bit(__IGB_DOWN, &adapter->state);
2729 	return 0;
2730 }
2731 
2732 /**
2733  *  igb_open - Called when a network interface is made active
2734  *  @netdev: network interface device structure
2735  *
2736  *  Returns 0 on success, negative value on failure
2737  *
2738  *  The open entry point is called when a network interface is made
2739  *  active by the system (IFF_UP).  At this point all resources needed
2740  *  for transmit and receive operations are allocated, the interrupt
2741  *  handler is registered with the OS, the watchdog timer is started,
2742  *  and the stack is notified that the interface is ready.
2743  **/
2744 static int __igb_open(struct net_device *netdev, bool resuming)
2745 {
2746 	struct igb_adapter *adapter = netdev_priv(netdev);
2747 	struct e1000_hw *hw = &adapter->hw;
2748 	struct pci_dev *pdev = adapter->pdev;
2749 	int err;
2750 	int i;
2751 
2752 	/* disallow open during test */
2753 	if (test_bit(__IGB_TESTING, &adapter->state)) {
2754 		WARN_ON(resuming);
2755 		return -EBUSY;
2756 	}
2757 
2758 	if (!resuming)
2759 		pm_runtime_get_sync(&pdev->dev);
2760 
2761 	netif_carrier_off(netdev);
2762 
2763 	/* allocate transmit descriptors */
2764 	err = igb_setup_all_tx_resources(adapter);
2765 	if (err)
2766 		goto err_setup_tx;
2767 
2768 	/* allocate receive descriptors */
2769 	err = igb_setup_all_rx_resources(adapter);
2770 	if (err)
2771 		goto err_setup_rx;
2772 
2773 	igb_power_up_link(adapter);
2774 
2775 	/* before we allocate an interrupt, we must be ready to handle it.
2776 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2777 	 * as soon as we call pci_request_irq, so we have to setup our
2778 	 * clean_rx handler before we do so.
2779 	 */
2780 	igb_configure(adapter);
2781 
2782 	err = igb_request_irq(adapter);
2783 	if (err)
2784 		goto err_req_irq;
2785 
2786 	/* Notify the stack of the actual queue counts. */
2787 	err = netif_set_real_num_tx_queues(adapter->netdev,
2788 					   adapter->num_tx_queues);
2789 	if (err)
2790 		goto err_set_queues;
2791 
2792 	err = netif_set_real_num_rx_queues(adapter->netdev,
2793 					   adapter->num_rx_queues);
2794 	if (err)
2795 		goto err_set_queues;
2796 
2797 	/* From here on the code is the same as igb_up() */
2798 	clear_bit(__IGB_DOWN, &adapter->state);
2799 
2800 	for (i = 0; i < adapter->num_q_vectors; i++)
2801 		napi_enable(&(adapter->q_vector[i]->napi));
2802 
2803 	/* Clear any pending interrupts. */
2804 	rd32(E1000_ICR);
2805 
2806 	igb_irq_enable(adapter);
2807 
2808 	/* notify VFs that reset has been completed */
2809 	if (adapter->vfs_allocated_count) {
2810 		u32 reg_data = rd32(E1000_CTRL_EXT);
2811 		reg_data |= E1000_CTRL_EXT_PFRSTD;
2812 		wr32(E1000_CTRL_EXT, reg_data);
2813 	}
2814 
2815 	netif_tx_start_all_queues(netdev);
2816 
2817 	if (!resuming)
2818 		pm_runtime_put(&pdev->dev);
2819 
2820 	/* start the watchdog. */
2821 	hw->mac.get_link_status = 1;
2822 	schedule_work(&adapter->watchdog_task);
2823 
2824 	return 0;
2825 
2826 err_set_queues:
2827 	igb_free_irq(adapter);
2828 err_req_irq:
2829 	igb_release_hw_control(adapter);
2830 	igb_power_down_link(adapter);
2831 	igb_free_all_rx_resources(adapter);
2832 err_setup_rx:
2833 	igb_free_all_tx_resources(adapter);
2834 err_setup_tx:
2835 	igb_reset(adapter);
2836 	if (!resuming)
2837 		pm_runtime_put(&pdev->dev);
2838 
2839 	return err;
2840 }
2841 
2842 static int igb_open(struct net_device *netdev)
2843 {
2844 	return __igb_open(netdev, false);
2845 }
2846 
2847 /**
2848  *  igb_close - Disables a network interface
2849  *  @netdev: network interface device structure
2850  *
2851  *  Returns 0, this is not allowed to fail
2852  *
2853  *  The close entry point is called when an interface is de-activated
2854  *  by the OS.  The hardware is still under the driver's control, but
2855  *  needs to be disabled.  A global MAC reset is issued to stop the
2856  *  hardware, and all transmit and receive resources are freed.
2857  **/
2858 static int __igb_close(struct net_device *netdev, bool suspending)
2859 {
2860 	struct igb_adapter *adapter = netdev_priv(netdev);
2861 	struct pci_dev *pdev = adapter->pdev;
2862 
2863 	WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2864 
2865 	if (!suspending)
2866 		pm_runtime_get_sync(&pdev->dev);
2867 
2868 	igb_down(adapter);
2869 	igb_free_irq(adapter);
2870 
2871 	igb_free_all_tx_resources(adapter);
2872 	igb_free_all_rx_resources(adapter);
2873 
2874 	if (!suspending)
2875 		pm_runtime_put_sync(&pdev->dev);
2876 	return 0;
2877 }
2878 
2879 static int igb_close(struct net_device *netdev)
2880 {
2881 	return __igb_close(netdev, false);
2882 }
2883 
2884 /**
2885  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
2886  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
2887  *
2888  *  Return 0 on success, negative on failure
2889  **/
2890 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2891 {
2892 	struct device *dev = tx_ring->dev;
2893 	int size;
2894 
2895 	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2896 
2897 	tx_ring->tx_buffer_info = vzalloc(size);
2898 	if (!tx_ring->tx_buffer_info)
2899 		goto err;
2900 
2901 	/* round up to nearest 4K */
2902 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2903 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2904 
2905 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2906 					   &tx_ring->dma, GFP_KERNEL);
2907 	if (!tx_ring->desc)
2908 		goto err;
2909 
2910 	tx_ring->next_to_use = 0;
2911 	tx_ring->next_to_clean = 0;
2912 
2913 	return 0;
2914 
2915 err:
2916 	vfree(tx_ring->tx_buffer_info);
2917 	tx_ring->tx_buffer_info = NULL;
2918 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2919 	return -ENOMEM;
2920 }
2921 
2922 /**
2923  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
2924  *				 (Descriptors) for all queues
2925  *  @adapter: board private structure
2926  *
2927  *  Return 0 on success, negative on failure
2928  **/
2929 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2930 {
2931 	struct pci_dev *pdev = adapter->pdev;
2932 	int i, err = 0;
2933 
2934 	for (i = 0; i < adapter->num_tx_queues; i++) {
2935 		err = igb_setup_tx_resources(adapter->tx_ring[i]);
2936 		if (err) {
2937 			dev_err(&pdev->dev,
2938 				"Allocation for Tx Queue %u failed\n", i);
2939 			for (i--; i >= 0; i--)
2940 				igb_free_tx_resources(adapter->tx_ring[i]);
2941 			break;
2942 		}
2943 	}
2944 
2945 	return err;
2946 }
2947 
2948 /**
2949  *  igb_setup_tctl - configure the transmit control registers
2950  *  @adapter: Board private structure
2951  **/
2952 void igb_setup_tctl(struct igb_adapter *adapter)
2953 {
2954 	struct e1000_hw *hw = &adapter->hw;
2955 	u32 tctl;
2956 
2957 	/* disable queue 0 which is enabled by default on 82575 and 82576 */
2958 	wr32(E1000_TXDCTL(0), 0);
2959 
2960 	/* Program the Transmit Control Register */
2961 	tctl = rd32(E1000_TCTL);
2962 	tctl &= ~E1000_TCTL_CT;
2963 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2964 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2965 
2966 	igb_config_collision_dist(hw);
2967 
2968 	/* Enable transmits */
2969 	tctl |= E1000_TCTL_EN;
2970 
2971 	wr32(E1000_TCTL, tctl);
2972 }
2973 
2974 /**
2975  *  igb_configure_tx_ring - Configure transmit ring after Reset
2976  *  @adapter: board private structure
2977  *  @ring: tx ring to configure
2978  *
2979  *  Configure a transmit ring after a reset.
2980  **/
2981 void igb_configure_tx_ring(struct igb_adapter *adapter,
2982                            struct igb_ring *ring)
2983 {
2984 	struct e1000_hw *hw = &adapter->hw;
2985 	u32 txdctl = 0;
2986 	u64 tdba = ring->dma;
2987 	int reg_idx = ring->reg_idx;
2988 
2989 	/* disable the queue */
2990 	wr32(E1000_TXDCTL(reg_idx), 0);
2991 	wrfl();
2992 	mdelay(10);
2993 
2994 	wr32(E1000_TDLEN(reg_idx),
2995 	     ring->count * sizeof(union e1000_adv_tx_desc));
2996 	wr32(E1000_TDBAL(reg_idx),
2997 	     tdba & 0x00000000ffffffffULL);
2998 	wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2999 
3000 	ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3001 	wr32(E1000_TDH(reg_idx), 0);
3002 	writel(0, ring->tail);
3003 
3004 	txdctl |= IGB_TX_PTHRESH;
3005 	txdctl |= IGB_TX_HTHRESH << 8;
3006 	txdctl |= IGB_TX_WTHRESH << 16;
3007 
3008 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3009 	wr32(E1000_TXDCTL(reg_idx), txdctl);
3010 }
3011 
3012 /**
3013  *  igb_configure_tx - Configure transmit Unit after Reset
3014  *  @adapter: board private structure
3015  *
3016  *  Configure the Tx unit of the MAC after a reset.
3017  **/
3018 static void igb_configure_tx(struct igb_adapter *adapter)
3019 {
3020 	int i;
3021 
3022 	for (i = 0; i < adapter->num_tx_queues; i++)
3023 		igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3024 }
3025 
3026 /**
3027  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3028  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3029  *
3030  *  Returns 0 on success, negative on failure
3031  **/
3032 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3033 {
3034 	struct device *dev = rx_ring->dev;
3035 	int size;
3036 
3037 	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3038 
3039 	rx_ring->rx_buffer_info = vzalloc(size);
3040 	if (!rx_ring->rx_buffer_info)
3041 		goto err;
3042 
3043 	/* Round up to nearest 4K */
3044 	rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3045 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3046 
3047 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3048 					   &rx_ring->dma, GFP_KERNEL);
3049 	if (!rx_ring->desc)
3050 		goto err;
3051 
3052 	rx_ring->next_to_alloc = 0;
3053 	rx_ring->next_to_clean = 0;
3054 	rx_ring->next_to_use = 0;
3055 
3056 	return 0;
3057 
3058 err:
3059 	vfree(rx_ring->rx_buffer_info);
3060 	rx_ring->rx_buffer_info = NULL;
3061 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3062 	return -ENOMEM;
3063 }
3064 
3065 /**
3066  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3067  *				 (Descriptors) for all queues
3068  *  @adapter: board private structure
3069  *
3070  *  Return 0 on success, negative on failure
3071  **/
3072 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3073 {
3074 	struct pci_dev *pdev = adapter->pdev;
3075 	int i, err = 0;
3076 
3077 	for (i = 0; i < adapter->num_rx_queues; i++) {
3078 		err = igb_setup_rx_resources(adapter->rx_ring[i]);
3079 		if (err) {
3080 			dev_err(&pdev->dev,
3081 				"Allocation for Rx Queue %u failed\n", i);
3082 			for (i--; i >= 0; i--)
3083 				igb_free_rx_resources(adapter->rx_ring[i]);
3084 			break;
3085 		}
3086 	}
3087 
3088 	return err;
3089 }
3090 
3091 /**
3092  *  igb_setup_mrqc - configure the multiple receive queue control registers
3093  *  @adapter: Board private structure
3094  **/
3095 static void igb_setup_mrqc(struct igb_adapter *adapter)
3096 {
3097 	struct e1000_hw *hw = &adapter->hw;
3098 	u32 mrqc, rxcsum;
3099 	u32 j, num_rx_queues, shift = 0;
3100 	static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3101 					0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3102 					0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3103 					0xFA01ACBE };
3104 
3105 	/* Fill out hash function seeds */
3106 	for (j = 0; j < 10; j++)
3107 		wr32(E1000_RSSRK(j), rsskey[j]);
3108 
3109 	num_rx_queues = adapter->rss_queues;
3110 
3111 	switch (hw->mac.type) {
3112 	case e1000_82575:
3113 		shift = 6;
3114 		break;
3115 	case e1000_82576:
3116 		/* 82576 supports 2 RSS queues for SR-IOV */
3117 		if (adapter->vfs_allocated_count) {
3118 			shift = 3;
3119 			num_rx_queues = 2;
3120 		}
3121 		break;
3122 	default:
3123 		break;
3124 	}
3125 
3126 	/* Populate the indirection table 4 entries at a time.  To do this
3127 	 * we are generating the results for n and n+2 and then interleaving
3128 	 * those with the results with n+1 and n+3.
3129 	 */
3130 	for (j = 0; j < 32; j++) {
3131 		/* first pass generates n and n+2 */
3132 		u32 base = ((j * 0x00040004) + 0x00020000) * num_rx_queues;
3133 		u32 reta = (base & 0x07800780) >> (7 - shift);
3134 
3135 		/* second pass generates n+1 and n+3 */
3136 		base += 0x00010001 * num_rx_queues;
3137 		reta |= (base & 0x07800780) << (1 + shift);
3138 
3139 		wr32(E1000_RETA(j), reta);
3140 	}
3141 
3142 	/* Disable raw packet checksumming so that RSS hash is placed in
3143 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3144 	 * offloads as they are enabled by default
3145 	 */
3146 	rxcsum = rd32(E1000_RXCSUM);
3147 	rxcsum |= E1000_RXCSUM_PCSD;
3148 
3149 	if (adapter->hw.mac.type >= e1000_82576)
3150 		/* Enable Receive Checksum Offload for SCTP */
3151 		rxcsum |= E1000_RXCSUM_CRCOFL;
3152 
3153 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
3154 	wr32(E1000_RXCSUM, rxcsum);
3155 
3156 	/* Generate RSS hash based on packet types, TCP/UDP
3157 	 * port numbers and/or IPv4/v6 src and dst addresses
3158 	 */
3159 	mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3160 	       E1000_MRQC_RSS_FIELD_IPV4_TCP |
3161 	       E1000_MRQC_RSS_FIELD_IPV6 |
3162 	       E1000_MRQC_RSS_FIELD_IPV6_TCP |
3163 	       E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3164 
3165 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3166 		mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3167 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3168 		mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3169 
3170 	/* If VMDq is enabled then we set the appropriate mode for that, else
3171 	 * we default to RSS so that an RSS hash is calculated per packet even
3172 	 * if we are only using one queue
3173 	 */
3174 	if (adapter->vfs_allocated_count) {
3175 		if (hw->mac.type > e1000_82575) {
3176 			/* Set the default pool for the PF's first queue */
3177 			u32 vtctl = rd32(E1000_VT_CTL);
3178 			vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3179 				   E1000_VT_CTL_DISABLE_DEF_POOL);
3180 			vtctl |= adapter->vfs_allocated_count <<
3181 				E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3182 			wr32(E1000_VT_CTL, vtctl);
3183 		}
3184 		if (adapter->rss_queues > 1)
3185 			mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3186 		else
3187 			mrqc |= E1000_MRQC_ENABLE_VMDQ;
3188 	} else {
3189 		if (hw->mac.type != e1000_i211)
3190 			mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3191 	}
3192 	igb_vmm_control(adapter);
3193 
3194 	wr32(E1000_MRQC, mrqc);
3195 }
3196 
3197 /**
3198  *  igb_setup_rctl - configure the receive control registers
3199  *  @adapter: Board private structure
3200  **/
3201 void igb_setup_rctl(struct igb_adapter *adapter)
3202 {
3203 	struct e1000_hw *hw = &adapter->hw;
3204 	u32 rctl;
3205 
3206 	rctl = rd32(E1000_RCTL);
3207 
3208 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3209 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3210 
3211 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3212 		(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3213 
3214 	/* enable stripping of CRC. It's unlikely this will break BMC
3215 	 * redirection as it did with e1000. Newer features require
3216 	 * that the HW strips the CRC.
3217 	 */
3218 	rctl |= E1000_RCTL_SECRC;
3219 
3220 	/* disable store bad packets and clear size bits. */
3221 	rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3222 
3223 	/* enable LPE to prevent packets larger than max_frame_size */
3224 	rctl |= E1000_RCTL_LPE;
3225 
3226 	/* disable queue 0 to prevent tail write w/o re-config */
3227 	wr32(E1000_RXDCTL(0), 0);
3228 
3229 	/* Attention!!!  For SR-IOV PF driver operations you must enable
3230 	 * queue drop for all VF and PF queues to prevent head of line blocking
3231 	 * if an un-trusted VF does not provide descriptors to hardware.
3232 	 */
3233 	if (adapter->vfs_allocated_count) {
3234 		/* set all queue drop enable bits */
3235 		wr32(E1000_QDE, ALL_QUEUES);
3236 	}
3237 
3238 	/* This is useful for sniffing bad packets. */
3239 	if (adapter->netdev->features & NETIF_F_RXALL) {
3240 		/* UPE and MPE will be handled by normal PROMISC logic
3241 		 * in e1000e_set_rx_mode
3242 		 */
3243 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3244 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
3245 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3246 
3247 		rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3248 			  E1000_RCTL_DPF | /* Allow filtered pause */
3249 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3250 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3251 		 * and that breaks VLANs.
3252 		 */
3253 	}
3254 
3255 	wr32(E1000_RCTL, rctl);
3256 }
3257 
3258 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3259                                    int vfn)
3260 {
3261 	struct e1000_hw *hw = &adapter->hw;
3262 	u32 vmolr;
3263 
3264 	/* if it isn't the PF check to see if VFs are enabled and
3265 	 * increase the size to support vlan tags
3266 	 */
3267 	if (vfn < adapter->vfs_allocated_count &&
3268 	    adapter->vf_data[vfn].vlans_enabled)
3269 		size += VLAN_TAG_SIZE;
3270 
3271 	vmolr = rd32(E1000_VMOLR(vfn));
3272 	vmolr &= ~E1000_VMOLR_RLPML_MASK;
3273 	vmolr |= size | E1000_VMOLR_LPE;
3274 	wr32(E1000_VMOLR(vfn), vmolr);
3275 
3276 	return 0;
3277 }
3278 
3279 /**
3280  *  igb_rlpml_set - set maximum receive packet size
3281  *  @adapter: board private structure
3282  *
3283  *  Configure maximum receivable packet size.
3284  **/
3285 static void igb_rlpml_set(struct igb_adapter *adapter)
3286 {
3287 	u32 max_frame_size = adapter->max_frame_size;
3288 	struct e1000_hw *hw = &adapter->hw;
3289 	u16 pf_id = adapter->vfs_allocated_count;
3290 
3291 	if (pf_id) {
3292 		igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3293 		/* If we're in VMDQ or SR-IOV mode, then set global RLPML
3294 		 * to our max jumbo frame size, in case we need to enable
3295 		 * jumbo frames on one of the rings later.
3296 		 * This will not pass over-length frames into the default
3297 		 * queue because it's gated by the VMOLR.RLPML.
3298 		 */
3299 		max_frame_size = MAX_JUMBO_FRAME_SIZE;
3300 	}
3301 
3302 	wr32(E1000_RLPML, max_frame_size);
3303 }
3304 
3305 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3306 				 int vfn, bool aupe)
3307 {
3308 	struct e1000_hw *hw = &adapter->hw;
3309 	u32 vmolr;
3310 
3311 	/* This register exists only on 82576 and newer so if we are older then
3312 	 * we should exit and do nothing
3313 	 */
3314 	if (hw->mac.type < e1000_82576)
3315 		return;
3316 
3317 	vmolr = rd32(E1000_VMOLR(vfn));
3318 	vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3319 	if (aupe)
3320 		vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3321 	else
3322 		vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3323 
3324 	/* clear all bits that might not be set */
3325 	vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3326 
3327 	if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3328 		vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3329 	/* for VMDq only allow the VFs and pool 0 to accept broadcast and
3330 	 * multicast packets
3331 	 */
3332 	if (vfn <= adapter->vfs_allocated_count)
3333 		vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3334 
3335 	wr32(E1000_VMOLR(vfn), vmolr);
3336 }
3337 
3338 /**
3339  *  igb_configure_rx_ring - Configure a receive ring after Reset
3340  *  @adapter: board private structure
3341  *  @ring: receive ring to be configured
3342  *
3343  *  Configure the Rx unit of the MAC after a reset.
3344  **/
3345 void igb_configure_rx_ring(struct igb_adapter *adapter,
3346 			   struct igb_ring *ring)
3347 {
3348 	struct e1000_hw *hw = &adapter->hw;
3349 	u64 rdba = ring->dma;
3350 	int reg_idx = ring->reg_idx;
3351 	u32 srrctl = 0, rxdctl = 0;
3352 
3353 	/* disable the queue */
3354 	wr32(E1000_RXDCTL(reg_idx), 0);
3355 
3356 	/* Set DMA base address registers */
3357 	wr32(E1000_RDBAL(reg_idx),
3358 	     rdba & 0x00000000ffffffffULL);
3359 	wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3360 	wr32(E1000_RDLEN(reg_idx),
3361 	     ring->count * sizeof(union e1000_adv_rx_desc));
3362 
3363 	/* initialize head and tail */
3364 	ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3365 	wr32(E1000_RDH(reg_idx), 0);
3366 	writel(0, ring->tail);
3367 
3368 	/* set descriptor configuration */
3369 	srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3370 	srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3371 	srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3372 	if (hw->mac.type >= e1000_82580)
3373 		srrctl |= E1000_SRRCTL_TIMESTAMP;
3374 	/* Only set Drop Enable if we are supporting multiple queues */
3375 	if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3376 		srrctl |= E1000_SRRCTL_DROP_EN;
3377 
3378 	wr32(E1000_SRRCTL(reg_idx), srrctl);
3379 
3380 	/* set filtering for VMDQ pools */
3381 	igb_set_vmolr(adapter, reg_idx & 0x7, true);
3382 
3383 	rxdctl |= IGB_RX_PTHRESH;
3384 	rxdctl |= IGB_RX_HTHRESH << 8;
3385 	rxdctl |= IGB_RX_WTHRESH << 16;
3386 
3387 	/* enable receive descriptor fetching */
3388 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3389 	wr32(E1000_RXDCTL(reg_idx), rxdctl);
3390 }
3391 
3392 /**
3393  *  igb_configure_rx - Configure receive Unit after Reset
3394  *  @adapter: board private structure
3395  *
3396  *  Configure the Rx unit of the MAC after a reset.
3397  **/
3398 static void igb_configure_rx(struct igb_adapter *adapter)
3399 {
3400 	int i;
3401 
3402 	/* set UTA to appropriate mode */
3403 	igb_set_uta(adapter);
3404 
3405 	/* set the correct pool for the PF default MAC address in entry 0 */
3406 	igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3407 			 adapter->vfs_allocated_count);
3408 
3409 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
3410 	 * the Base and Length of the Rx Descriptor Ring
3411 	 */
3412 	for (i = 0; i < adapter->num_rx_queues; i++)
3413 		igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3414 }
3415 
3416 /**
3417  *  igb_free_tx_resources - Free Tx Resources per Queue
3418  *  @tx_ring: Tx descriptor ring for a specific queue
3419  *
3420  *  Free all transmit software resources
3421  **/
3422 void igb_free_tx_resources(struct igb_ring *tx_ring)
3423 {
3424 	igb_clean_tx_ring(tx_ring);
3425 
3426 	vfree(tx_ring->tx_buffer_info);
3427 	tx_ring->tx_buffer_info = NULL;
3428 
3429 	/* if not set, then don't free */
3430 	if (!tx_ring->desc)
3431 		return;
3432 
3433 	dma_free_coherent(tx_ring->dev, tx_ring->size,
3434 			  tx_ring->desc, tx_ring->dma);
3435 
3436 	tx_ring->desc = NULL;
3437 }
3438 
3439 /**
3440  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
3441  *  @adapter: board private structure
3442  *
3443  *  Free all transmit software resources
3444  **/
3445 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3446 {
3447 	int i;
3448 
3449 	for (i = 0; i < adapter->num_tx_queues; i++)
3450 		igb_free_tx_resources(adapter->tx_ring[i]);
3451 }
3452 
3453 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3454 				    struct igb_tx_buffer *tx_buffer)
3455 {
3456 	if (tx_buffer->skb) {
3457 		dev_kfree_skb_any(tx_buffer->skb);
3458 		if (dma_unmap_len(tx_buffer, len))
3459 			dma_unmap_single(ring->dev,
3460 					 dma_unmap_addr(tx_buffer, dma),
3461 					 dma_unmap_len(tx_buffer, len),
3462 					 DMA_TO_DEVICE);
3463 	} else if (dma_unmap_len(tx_buffer, len)) {
3464 		dma_unmap_page(ring->dev,
3465 			       dma_unmap_addr(tx_buffer, dma),
3466 			       dma_unmap_len(tx_buffer, len),
3467 			       DMA_TO_DEVICE);
3468 	}
3469 	tx_buffer->next_to_watch = NULL;
3470 	tx_buffer->skb = NULL;
3471 	dma_unmap_len_set(tx_buffer, len, 0);
3472 	/* buffer_info must be completely set up in the transmit path */
3473 }
3474 
3475 /**
3476  *  igb_clean_tx_ring - Free Tx Buffers
3477  *  @tx_ring: ring to be cleaned
3478  **/
3479 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3480 {
3481 	struct igb_tx_buffer *buffer_info;
3482 	unsigned long size;
3483 	u16 i;
3484 
3485 	if (!tx_ring->tx_buffer_info)
3486 		return;
3487 	/* Free all the Tx ring sk_buffs */
3488 
3489 	for (i = 0; i < tx_ring->count; i++) {
3490 		buffer_info = &tx_ring->tx_buffer_info[i];
3491 		igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3492 	}
3493 
3494 	netdev_tx_reset_queue(txring_txq(tx_ring));
3495 
3496 	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3497 	memset(tx_ring->tx_buffer_info, 0, size);
3498 
3499 	/* Zero out the descriptor ring */
3500 	memset(tx_ring->desc, 0, tx_ring->size);
3501 
3502 	tx_ring->next_to_use = 0;
3503 	tx_ring->next_to_clean = 0;
3504 }
3505 
3506 /**
3507  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
3508  *  @adapter: board private structure
3509  **/
3510 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3511 {
3512 	int i;
3513 
3514 	for (i = 0; i < adapter->num_tx_queues; i++)
3515 		igb_clean_tx_ring(adapter->tx_ring[i]);
3516 }
3517 
3518 /**
3519  *  igb_free_rx_resources - Free Rx Resources
3520  *  @rx_ring: ring to clean the resources from
3521  *
3522  *  Free all receive software resources
3523  **/
3524 void igb_free_rx_resources(struct igb_ring *rx_ring)
3525 {
3526 	igb_clean_rx_ring(rx_ring);
3527 
3528 	vfree(rx_ring->rx_buffer_info);
3529 	rx_ring->rx_buffer_info = NULL;
3530 
3531 	/* if not set, then don't free */
3532 	if (!rx_ring->desc)
3533 		return;
3534 
3535 	dma_free_coherent(rx_ring->dev, rx_ring->size,
3536 			  rx_ring->desc, rx_ring->dma);
3537 
3538 	rx_ring->desc = NULL;
3539 }
3540 
3541 /**
3542  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
3543  *  @adapter: board private structure
3544  *
3545  *  Free all receive software resources
3546  **/
3547 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3548 {
3549 	int i;
3550 
3551 	for (i = 0; i < adapter->num_rx_queues; i++)
3552 		igb_free_rx_resources(adapter->rx_ring[i]);
3553 }
3554 
3555 /**
3556  *  igb_clean_rx_ring - Free Rx Buffers per Queue
3557  *  @rx_ring: ring to free buffers from
3558  **/
3559 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3560 {
3561 	unsigned long size;
3562 	u16 i;
3563 
3564 	if (rx_ring->skb)
3565 		dev_kfree_skb(rx_ring->skb);
3566 	rx_ring->skb = NULL;
3567 
3568 	if (!rx_ring->rx_buffer_info)
3569 		return;
3570 
3571 	/* Free all the Rx ring sk_buffs */
3572 	for (i = 0; i < rx_ring->count; i++) {
3573 		struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3574 
3575 		if (!buffer_info->page)
3576 			continue;
3577 
3578 		dma_unmap_page(rx_ring->dev,
3579 			       buffer_info->dma,
3580 			       PAGE_SIZE,
3581 			       DMA_FROM_DEVICE);
3582 		__free_page(buffer_info->page);
3583 
3584 		buffer_info->page = NULL;
3585 	}
3586 
3587 	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3588 	memset(rx_ring->rx_buffer_info, 0, size);
3589 
3590 	/* Zero out the descriptor ring */
3591 	memset(rx_ring->desc, 0, rx_ring->size);
3592 
3593 	rx_ring->next_to_alloc = 0;
3594 	rx_ring->next_to_clean = 0;
3595 	rx_ring->next_to_use = 0;
3596 }
3597 
3598 /**
3599  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
3600  *  @adapter: board private structure
3601  **/
3602 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3603 {
3604 	int i;
3605 
3606 	for (i = 0; i < adapter->num_rx_queues; i++)
3607 		igb_clean_rx_ring(adapter->rx_ring[i]);
3608 }
3609 
3610 /**
3611  *  igb_set_mac - Change the Ethernet Address of the NIC
3612  *  @netdev: network interface device structure
3613  *  @p: pointer to an address structure
3614  *
3615  *  Returns 0 on success, negative on failure
3616  **/
3617 static int igb_set_mac(struct net_device *netdev, void *p)
3618 {
3619 	struct igb_adapter *adapter = netdev_priv(netdev);
3620 	struct e1000_hw *hw = &adapter->hw;
3621 	struct sockaddr *addr = p;
3622 
3623 	if (!is_valid_ether_addr(addr->sa_data))
3624 		return -EADDRNOTAVAIL;
3625 
3626 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3627 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3628 
3629 	/* set the correct pool for the new PF MAC address in entry 0 */
3630 	igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3631 			 adapter->vfs_allocated_count);
3632 
3633 	return 0;
3634 }
3635 
3636 /**
3637  *  igb_write_mc_addr_list - write multicast addresses to MTA
3638  *  @netdev: network interface device structure
3639  *
3640  *  Writes multicast address list to the MTA hash table.
3641  *  Returns: -ENOMEM on failure
3642  *           0 on no addresses written
3643  *           X on writing X addresses to MTA
3644  **/
3645 static int igb_write_mc_addr_list(struct net_device *netdev)
3646 {
3647 	struct igb_adapter *adapter = netdev_priv(netdev);
3648 	struct e1000_hw *hw = &adapter->hw;
3649 	struct netdev_hw_addr *ha;
3650 	u8  *mta_list;
3651 	int i;
3652 
3653 	if (netdev_mc_empty(netdev)) {
3654 		/* nothing to program, so clear mc list */
3655 		igb_update_mc_addr_list(hw, NULL, 0);
3656 		igb_restore_vf_multicasts(adapter);
3657 		return 0;
3658 	}
3659 
3660 	mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3661 	if (!mta_list)
3662 		return -ENOMEM;
3663 
3664 	/* The shared function expects a packed array of only addresses. */
3665 	i = 0;
3666 	netdev_for_each_mc_addr(ha, netdev)
3667 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3668 
3669 	igb_update_mc_addr_list(hw, mta_list, i);
3670 	kfree(mta_list);
3671 
3672 	return netdev_mc_count(netdev);
3673 }
3674 
3675 /**
3676  *  igb_write_uc_addr_list - write unicast addresses to RAR table
3677  *  @netdev: network interface device structure
3678  *
3679  *  Writes unicast address list to the RAR table.
3680  *  Returns: -ENOMEM on failure/insufficient address space
3681  *           0 on no addresses written
3682  *           X on writing X addresses to the RAR table
3683  **/
3684 static int igb_write_uc_addr_list(struct net_device *netdev)
3685 {
3686 	struct igb_adapter *adapter = netdev_priv(netdev);
3687 	struct e1000_hw *hw = &adapter->hw;
3688 	unsigned int vfn = adapter->vfs_allocated_count;
3689 	unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3690 	int count = 0;
3691 
3692 	/* return ENOMEM indicating insufficient memory for addresses */
3693 	if (netdev_uc_count(netdev) > rar_entries)
3694 		return -ENOMEM;
3695 
3696 	if (!netdev_uc_empty(netdev) && rar_entries) {
3697 		struct netdev_hw_addr *ha;
3698 
3699 		netdev_for_each_uc_addr(ha, netdev) {
3700 			if (!rar_entries)
3701 				break;
3702 			igb_rar_set_qsel(adapter, ha->addr,
3703 					 rar_entries--,
3704 					 vfn);
3705 			count++;
3706 		}
3707 	}
3708 	/* write the addresses in reverse order to avoid write combining */
3709 	for (; rar_entries > 0 ; rar_entries--) {
3710 		wr32(E1000_RAH(rar_entries), 0);
3711 		wr32(E1000_RAL(rar_entries), 0);
3712 	}
3713 	wrfl();
3714 
3715 	return count;
3716 }
3717 
3718 /**
3719  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3720  *  @netdev: network interface device structure
3721  *
3722  *  The set_rx_mode entry point is called whenever the unicast or multicast
3723  *  address lists or the network interface flags are updated.  This routine is
3724  *  responsible for configuring the hardware for proper unicast, multicast,
3725  *  promiscuous mode, and all-multi behavior.
3726  **/
3727 static void igb_set_rx_mode(struct net_device *netdev)
3728 {
3729 	struct igb_adapter *adapter = netdev_priv(netdev);
3730 	struct e1000_hw *hw = &adapter->hw;
3731 	unsigned int vfn = adapter->vfs_allocated_count;
3732 	u32 rctl, vmolr = 0;
3733 	int count;
3734 
3735 	/* Check for Promiscuous and All Multicast modes */
3736 	rctl = rd32(E1000_RCTL);
3737 
3738 	/* clear the effected bits */
3739 	rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3740 
3741 	if (netdev->flags & IFF_PROMISC) {
3742 		u32 mrqc = rd32(E1000_MRQC);
3743 		/* retain VLAN HW filtering if in VT mode */
3744 		if (mrqc & E1000_MRQC_ENABLE_VMDQ)
3745 			rctl |= E1000_RCTL_VFE;
3746 		rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3747 		vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3748 	} else {
3749 		if (netdev->flags & IFF_ALLMULTI) {
3750 			rctl |= E1000_RCTL_MPE;
3751 			vmolr |= E1000_VMOLR_MPME;
3752 		} else {
3753 			/* Write addresses to the MTA, if the attempt fails
3754 			 * then we should just turn on promiscuous mode so
3755 			 * that we can at least receive multicast traffic
3756 			 */
3757 			count = igb_write_mc_addr_list(netdev);
3758 			if (count < 0) {
3759 				rctl |= E1000_RCTL_MPE;
3760 				vmolr |= E1000_VMOLR_MPME;
3761 			} else if (count) {
3762 				vmolr |= E1000_VMOLR_ROMPE;
3763 			}
3764 		}
3765 		/* Write addresses to available RAR registers, if there is not
3766 		 * sufficient space to store all the addresses then enable
3767 		 * unicast promiscuous mode
3768 		 */
3769 		count = igb_write_uc_addr_list(netdev);
3770 		if (count < 0) {
3771 			rctl |= E1000_RCTL_UPE;
3772 			vmolr |= E1000_VMOLR_ROPE;
3773 		}
3774 		rctl |= E1000_RCTL_VFE;
3775 	}
3776 	wr32(E1000_RCTL, rctl);
3777 
3778 	/* In order to support SR-IOV and eventually VMDq it is necessary to set
3779 	 * the VMOLR to enable the appropriate modes.  Without this workaround
3780 	 * we will have issues with VLAN tag stripping not being done for frames
3781 	 * that are only arriving because we are the default pool
3782 	 */
3783 	if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3784 		return;
3785 
3786 	vmolr |= rd32(E1000_VMOLR(vfn)) &
3787 		 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3788 	wr32(E1000_VMOLR(vfn), vmolr);
3789 	igb_restore_vf_multicasts(adapter);
3790 }
3791 
3792 static void igb_check_wvbr(struct igb_adapter *adapter)
3793 {
3794 	struct e1000_hw *hw = &adapter->hw;
3795 	u32 wvbr = 0;
3796 
3797 	switch (hw->mac.type) {
3798 	case e1000_82576:
3799 	case e1000_i350:
3800 		if (!(wvbr = rd32(E1000_WVBR)))
3801 			return;
3802 		break;
3803 	default:
3804 		break;
3805 	}
3806 
3807 	adapter->wvbr |= wvbr;
3808 }
3809 
3810 #define IGB_STAGGERED_QUEUE_OFFSET 8
3811 
3812 static void igb_spoof_check(struct igb_adapter *adapter)
3813 {
3814 	int j;
3815 
3816 	if (!adapter->wvbr)
3817 		return;
3818 
3819 	for(j = 0; j < adapter->vfs_allocated_count; j++) {
3820 		if (adapter->wvbr & (1 << j) ||
3821 		    adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3822 			dev_warn(&adapter->pdev->dev,
3823 				"Spoof event(s) detected on VF %d\n", j);
3824 			adapter->wvbr &=
3825 				~((1 << j) |
3826 				  (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3827 		}
3828 	}
3829 }
3830 
3831 /* Need to wait a few seconds after link up to get diagnostic information from
3832  * the phy
3833  */
3834 static void igb_update_phy_info(unsigned long data)
3835 {
3836 	struct igb_adapter *adapter = (struct igb_adapter *) data;
3837 	igb_get_phy_info(&adapter->hw);
3838 }
3839 
3840 /**
3841  *  igb_has_link - check shared code for link and determine up/down
3842  *  @adapter: pointer to driver private info
3843  **/
3844 bool igb_has_link(struct igb_adapter *adapter)
3845 {
3846 	struct e1000_hw *hw = &adapter->hw;
3847 	bool link_active = false;
3848 	s32 ret_val = 0;
3849 
3850 	/* get_link_status is set on LSC (link status) interrupt or
3851 	 * rx sequence error interrupt.  get_link_status will stay
3852 	 * false until the e1000_check_for_link establishes link
3853 	 * for copper adapters ONLY
3854 	 */
3855 	switch (hw->phy.media_type) {
3856 	case e1000_media_type_copper:
3857 		if (hw->mac.get_link_status) {
3858 			ret_val = hw->mac.ops.check_for_link(hw);
3859 			link_active = !hw->mac.get_link_status;
3860 		} else {
3861 			link_active = true;
3862 		}
3863 		break;
3864 	case e1000_media_type_internal_serdes:
3865 		ret_val = hw->mac.ops.check_for_link(hw);
3866 		link_active = hw->mac.serdes_has_link;
3867 		break;
3868 	default:
3869 	case e1000_media_type_unknown:
3870 		break;
3871 	}
3872 
3873 	return link_active;
3874 }
3875 
3876 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3877 {
3878 	bool ret = false;
3879 	u32 ctrl_ext, thstat;
3880 
3881 	/* check for thermal sensor event on i350 copper only */
3882 	if (hw->mac.type == e1000_i350) {
3883 		thstat = rd32(E1000_THSTAT);
3884 		ctrl_ext = rd32(E1000_CTRL_EXT);
3885 
3886 		if ((hw->phy.media_type == e1000_media_type_copper) &&
3887 		    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
3888 			ret = !!(thstat & event);
3889 	}
3890 
3891 	return ret;
3892 }
3893 
3894 /**
3895  *  igb_watchdog - Timer Call-back
3896  *  @data: pointer to adapter cast into an unsigned long
3897  **/
3898 static void igb_watchdog(unsigned long data)
3899 {
3900 	struct igb_adapter *adapter = (struct igb_adapter *)data;
3901 	/* Do the rest outside of interrupt context */
3902 	schedule_work(&adapter->watchdog_task);
3903 }
3904 
3905 static void igb_watchdog_task(struct work_struct *work)
3906 {
3907 	struct igb_adapter *adapter = container_of(work,
3908 						   struct igb_adapter,
3909 						   watchdog_task);
3910 	struct e1000_hw *hw = &adapter->hw;
3911 	struct e1000_phy_info *phy = &hw->phy;
3912 	struct net_device *netdev = adapter->netdev;
3913 	u32 link;
3914 	int i;
3915 
3916 	link = igb_has_link(adapter);
3917 	if (link) {
3918 		/* Cancel scheduled suspend requests. */
3919 		pm_runtime_resume(netdev->dev.parent);
3920 
3921 		if (!netif_carrier_ok(netdev)) {
3922 			u32 ctrl;
3923 			hw->mac.ops.get_speed_and_duplex(hw,
3924 							 &adapter->link_speed,
3925 							 &adapter->link_duplex);
3926 
3927 			ctrl = rd32(E1000_CTRL);
3928 			/* Links status message must follow this format */
3929 			printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3930 			       "Duplex, Flow Control: %s\n",
3931 			       netdev->name,
3932 			       adapter->link_speed,
3933 			       adapter->link_duplex == FULL_DUPLEX ?
3934 			       "Full" : "Half",
3935 			       (ctrl & E1000_CTRL_TFCE) &&
3936 			       (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3937 			       (ctrl & E1000_CTRL_RFCE) ?  "RX" :
3938 			       (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
3939 
3940 			/* check if SmartSpeed worked */
3941 			igb_check_downshift(hw);
3942 			if (phy->speed_downgraded)
3943 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
3944 
3945 			/* check for thermal sensor event */
3946 			if (igb_thermal_sensor_event(hw,
3947 			    E1000_THSTAT_LINK_THROTTLE)) {
3948 				netdev_info(netdev, "The network adapter link "
3949 					    "speed was downshifted because it "
3950 					    "overheated\n");
3951 			}
3952 
3953 			/* adjust timeout factor according to speed/duplex */
3954 			adapter->tx_timeout_factor = 1;
3955 			switch (adapter->link_speed) {
3956 			case SPEED_10:
3957 				adapter->tx_timeout_factor = 14;
3958 				break;
3959 			case SPEED_100:
3960 				/* maybe add some timeout factor ? */
3961 				break;
3962 			}
3963 
3964 			netif_carrier_on(netdev);
3965 
3966 			igb_ping_all_vfs(adapter);
3967 			igb_check_vf_rate_limit(adapter);
3968 
3969 			/* link state has changed, schedule phy info update */
3970 			if (!test_bit(__IGB_DOWN, &adapter->state))
3971 				mod_timer(&adapter->phy_info_timer,
3972 					  round_jiffies(jiffies + 2 * HZ));
3973 		}
3974 	} else {
3975 		if (netif_carrier_ok(netdev)) {
3976 			adapter->link_speed = 0;
3977 			adapter->link_duplex = 0;
3978 
3979 			/* check for thermal sensor event */
3980 			if (igb_thermal_sensor_event(hw,
3981 			    E1000_THSTAT_PWR_DOWN)) {
3982 				netdev_err(netdev, "The network adapter was "
3983 					   "stopped because it overheated\n");
3984 			}
3985 
3986 			/* Links status message must follow this format */
3987 			printk(KERN_INFO "igb: %s NIC Link is Down\n",
3988 			       netdev->name);
3989 			netif_carrier_off(netdev);
3990 
3991 			igb_ping_all_vfs(adapter);
3992 
3993 			/* link state has changed, schedule phy info update */
3994 			if (!test_bit(__IGB_DOWN, &adapter->state))
3995 				mod_timer(&adapter->phy_info_timer,
3996 					  round_jiffies(jiffies + 2 * HZ));
3997 
3998 			pm_schedule_suspend(netdev->dev.parent,
3999 					    MSEC_PER_SEC * 5);
4000 		}
4001 	}
4002 
4003 	spin_lock(&adapter->stats64_lock);
4004 	igb_update_stats(adapter, &adapter->stats64);
4005 	spin_unlock(&adapter->stats64_lock);
4006 
4007 	for (i = 0; i < adapter->num_tx_queues; i++) {
4008 		struct igb_ring *tx_ring = adapter->tx_ring[i];
4009 		if (!netif_carrier_ok(netdev)) {
4010 			/* We've lost link, so the controller stops DMA,
4011 			 * but we've got queued Tx work that's never going
4012 			 * to get done, so reset controller to flush Tx.
4013 			 * (Do the reset outside of interrupt context).
4014 			 */
4015 			if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4016 				adapter->tx_timeout_count++;
4017 				schedule_work(&adapter->reset_task);
4018 				/* return immediately since reset is imminent */
4019 				return;
4020 			}
4021 		}
4022 
4023 		/* Force detection of hung controller every watchdog period */
4024 		set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4025 	}
4026 
4027 	/* Cause software interrupt to ensure Rx ring is cleaned */
4028 	if (adapter->msix_entries) {
4029 		u32 eics = 0;
4030 		for (i = 0; i < adapter->num_q_vectors; i++)
4031 			eics |= adapter->q_vector[i]->eims_value;
4032 		wr32(E1000_EICS, eics);
4033 	} else {
4034 		wr32(E1000_ICS, E1000_ICS_RXDMT0);
4035 	}
4036 
4037 	igb_spoof_check(adapter);
4038 	igb_ptp_rx_hang(adapter);
4039 
4040 	/* Reset the timer */
4041 	if (!test_bit(__IGB_DOWN, &adapter->state))
4042 		mod_timer(&adapter->watchdog_timer,
4043 			  round_jiffies(jiffies + 2 * HZ));
4044 }
4045 
4046 enum latency_range {
4047 	lowest_latency = 0,
4048 	low_latency = 1,
4049 	bulk_latency = 2,
4050 	latency_invalid = 255
4051 };
4052 
4053 /**
4054  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
4055  *  @q_vector: pointer to q_vector
4056  *
4057  *  Stores a new ITR value based on strictly on packet size.  This
4058  *  algorithm is less sophisticated than that used in igb_update_itr,
4059  *  due to the difficulty of synchronizing statistics across multiple
4060  *  receive rings.  The divisors and thresholds used by this function
4061  *  were determined based on theoretical maximum wire speed and testing
4062  *  data, in order to minimize response time while increasing bulk
4063  *  throughput.
4064  *  This functionality is controlled by the InterruptThrottleRate module
4065  *  parameter (see igb_param.c)
4066  *  NOTE:  This function is called only when operating in a multiqueue
4067  *         receive environment.
4068  **/
4069 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4070 {
4071 	int new_val = q_vector->itr_val;
4072 	int avg_wire_size = 0;
4073 	struct igb_adapter *adapter = q_vector->adapter;
4074 	unsigned int packets;
4075 
4076 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
4077 	 * ints/sec - ITR timer value of 120 ticks.
4078 	 */
4079 	if (adapter->link_speed != SPEED_1000) {
4080 		new_val = IGB_4K_ITR;
4081 		goto set_itr_val;
4082 	}
4083 
4084 	packets = q_vector->rx.total_packets;
4085 	if (packets)
4086 		avg_wire_size = q_vector->rx.total_bytes / packets;
4087 
4088 	packets = q_vector->tx.total_packets;
4089 	if (packets)
4090 		avg_wire_size = max_t(u32, avg_wire_size,
4091 				      q_vector->tx.total_bytes / packets);
4092 
4093 	/* if avg_wire_size isn't set no work was done */
4094 	if (!avg_wire_size)
4095 		goto clear_counts;
4096 
4097 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
4098 	avg_wire_size += 24;
4099 
4100 	/* Don't starve jumbo frames */
4101 	avg_wire_size = min(avg_wire_size, 3000);
4102 
4103 	/* Give a little boost to mid-size frames */
4104 	if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4105 		new_val = avg_wire_size / 3;
4106 	else
4107 		new_val = avg_wire_size / 2;
4108 
4109 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4110 	if (new_val < IGB_20K_ITR &&
4111 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4112 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4113 		new_val = IGB_20K_ITR;
4114 
4115 set_itr_val:
4116 	if (new_val != q_vector->itr_val) {
4117 		q_vector->itr_val = new_val;
4118 		q_vector->set_itr = 1;
4119 	}
4120 clear_counts:
4121 	q_vector->rx.total_bytes = 0;
4122 	q_vector->rx.total_packets = 0;
4123 	q_vector->tx.total_bytes = 0;
4124 	q_vector->tx.total_packets = 0;
4125 }
4126 
4127 /**
4128  *  igb_update_itr - update the dynamic ITR value based on statistics
4129  *  @q_vector: pointer to q_vector
4130  *  @ring_container: ring info to update the itr for
4131  *
4132  *  Stores a new ITR value based on packets and byte
4133  *  counts during the last interrupt.  The advantage of per interrupt
4134  *  computation is faster updates and more accurate ITR for the current
4135  *  traffic pattern.  Constants in this function were computed
4136  *  based on theoretical maximum wire speed and thresholds were set based
4137  *  on testing data as well as attempting to minimize response time
4138  *  while increasing bulk throughput.
4139  *  this functionality is controlled by the InterruptThrottleRate module
4140  *  parameter (see igb_param.c)
4141  *  NOTE:  These calculations are only valid when operating in a single-
4142  *         queue environment.
4143  **/
4144 static void igb_update_itr(struct igb_q_vector *q_vector,
4145 			   struct igb_ring_container *ring_container)
4146 {
4147 	unsigned int packets = ring_container->total_packets;
4148 	unsigned int bytes = ring_container->total_bytes;
4149 	u8 itrval = ring_container->itr;
4150 
4151 	/* no packets, exit with status unchanged */
4152 	if (packets == 0)
4153 		return;
4154 
4155 	switch (itrval) {
4156 	case lowest_latency:
4157 		/* handle TSO and jumbo frames */
4158 		if (bytes/packets > 8000)
4159 			itrval = bulk_latency;
4160 		else if ((packets < 5) && (bytes > 512))
4161 			itrval = low_latency;
4162 		break;
4163 	case low_latency:  /* 50 usec aka 20000 ints/s */
4164 		if (bytes > 10000) {
4165 			/* this if handles the TSO accounting */
4166 			if (bytes/packets > 8000) {
4167 				itrval = bulk_latency;
4168 			} else if ((packets < 10) || ((bytes/packets) > 1200)) {
4169 				itrval = bulk_latency;
4170 			} else if ((packets > 35)) {
4171 				itrval = lowest_latency;
4172 			}
4173 		} else if (bytes/packets > 2000) {
4174 			itrval = bulk_latency;
4175 		} else if (packets <= 2 && bytes < 512) {
4176 			itrval = lowest_latency;
4177 		}
4178 		break;
4179 	case bulk_latency: /* 250 usec aka 4000 ints/s */
4180 		if (bytes > 25000) {
4181 			if (packets > 35)
4182 				itrval = low_latency;
4183 		} else if (bytes < 1500) {
4184 			itrval = low_latency;
4185 		}
4186 		break;
4187 	}
4188 
4189 	/* clear work counters since we have the values we need */
4190 	ring_container->total_bytes = 0;
4191 	ring_container->total_packets = 0;
4192 
4193 	/* write updated itr to ring container */
4194 	ring_container->itr = itrval;
4195 }
4196 
4197 static void igb_set_itr(struct igb_q_vector *q_vector)
4198 {
4199 	struct igb_adapter *adapter = q_vector->adapter;
4200 	u32 new_itr = q_vector->itr_val;
4201 	u8 current_itr = 0;
4202 
4203 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4204 	if (adapter->link_speed != SPEED_1000) {
4205 		current_itr = 0;
4206 		new_itr = IGB_4K_ITR;
4207 		goto set_itr_now;
4208 	}
4209 
4210 	igb_update_itr(q_vector, &q_vector->tx);
4211 	igb_update_itr(q_vector, &q_vector->rx);
4212 
4213 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4214 
4215 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4216 	if (current_itr == lowest_latency &&
4217 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4218 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4219 		current_itr = low_latency;
4220 
4221 	switch (current_itr) {
4222 	/* counts and packets in update_itr are dependent on these numbers */
4223 	case lowest_latency:
4224 		new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4225 		break;
4226 	case low_latency:
4227 		new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4228 		break;
4229 	case bulk_latency:
4230 		new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
4231 		break;
4232 	default:
4233 		break;
4234 	}
4235 
4236 set_itr_now:
4237 	if (new_itr != q_vector->itr_val) {
4238 		/* this attempts to bias the interrupt rate towards Bulk
4239 		 * by adding intermediate steps when interrupt rate is
4240 		 * increasing
4241 		 */
4242 		new_itr = new_itr > q_vector->itr_val ?
4243 			  max((new_itr * q_vector->itr_val) /
4244 			  (new_itr + (q_vector->itr_val >> 2)),
4245 			  new_itr) : new_itr;
4246 		/* Don't write the value here; it resets the adapter's
4247 		 * internal timer, and causes us to delay far longer than
4248 		 * we should between interrupts.  Instead, we write the ITR
4249 		 * value at the beginning of the next interrupt so the timing
4250 		 * ends up being correct.
4251 		 */
4252 		q_vector->itr_val = new_itr;
4253 		q_vector->set_itr = 1;
4254 	}
4255 }
4256 
4257 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4258 			    u32 type_tucmd, u32 mss_l4len_idx)
4259 {
4260 	struct e1000_adv_tx_context_desc *context_desc;
4261 	u16 i = tx_ring->next_to_use;
4262 
4263 	context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4264 
4265 	i++;
4266 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4267 
4268 	/* set bits to identify this as an advanced context descriptor */
4269 	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4270 
4271 	/* For 82575, context index must be unique per ring. */
4272 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4273 		mss_l4len_idx |= tx_ring->reg_idx << 4;
4274 
4275 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
4276 	context_desc->seqnum_seed	= 0;
4277 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
4278 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
4279 }
4280 
4281 static int igb_tso(struct igb_ring *tx_ring,
4282 		   struct igb_tx_buffer *first,
4283 		   u8 *hdr_len)
4284 {
4285 	struct sk_buff *skb = first->skb;
4286 	u32 vlan_macip_lens, type_tucmd;
4287 	u32 mss_l4len_idx, l4len;
4288 
4289 	if (skb->ip_summed != CHECKSUM_PARTIAL)
4290 		return 0;
4291 
4292 	if (!skb_is_gso(skb))
4293 		return 0;
4294 
4295 	if (skb_header_cloned(skb)) {
4296 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4297 		if (err)
4298 			return err;
4299 	}
4300 
4301 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4302 	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4303 
4304 	if (first->protocol == __constant_htons(ETH_P_IP)) {
4305 		struct iphdr *iph = ip_hdr(skb);
4306 		iph->tot_len = 0;
4307 		iph->check = 0;
4308 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4309 							 iph->daddr, 0,
4310 							 IPPROTO_TCP,
4311 							 0);
4312 		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4313 		first->tx_flags |= IGB_TX_FLAGS_TSO |
4314 				   IGB_TX_FLAGS_CSUM |
4315 				   IGB_TX_FLAGS_IPV4;
4316 	} else if (skb_is_gso_v6(skb)) {
4317 		ipv6_hdr(skb)->payload_len = 0;
4318 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4319 						       &ipv6_hdr(skb)->daddr,
4320 						       0, IPPROTO_TCP, 0);
4321 		first->tx_flags |= IGB_TX_FLAGS_TSO |
4322 				   IGB_TX_FLAGS_CSUM;
4323 	}
4324 
4325 	/* compute header lengths */
4326 	l4len = tcp_hdrlen(skb);
4327 	*hdr_len = skb_transport_offset(skb) + l4len;
4328 
4329 	/* update gso size and bytecount with header size */
4330 	first->gso_segs = skb_shinfo(skb)->gso_segs;
4331 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
4332 
4333 	/* MSS L4LEN IDX */
4334 	mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4335 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4336 
4337 	/* VLAN MACLEN IPLEN */
4338 	vlan_macip_lens = skb_network_header_len(skb);
4339 	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4340 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4341 
4342 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4343 
4344 	return 1;
4345 }
4346 
4347 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4348 {
4349 	struct sk_buff *skb = first->skb;
4350 	u32 vlan_macip_lens = 0;
4351 	u32 mss_l4len_idx = 0;
4352 	u32 type_tucmd = 0;
4353 
4354 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
4355 		if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4356 			return;
4357 	} else {
4358 		u8 l4_hdr = 0;
4359 		switch (first->protocol) {
4360 		case __constant_htons(ETH_P_IP):
4361 			vlan_macip_lens |= skb_network_header_len(skb);
4362 			type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4363 			l4_hdr = ip_hdr(skb)->protocol;
4364 			break;
4365 		case __constant_htons(ETH_P_IPV6):
4366 			vlan_macip_lens |= skb_network_header_len(skb);
4367 			l4_hdr = ipv6_hdr(skb)->nexthdr;
4368 			break;
4369 		default:
4370 			if (unlikely(net_ratelimit())) {
4371 				dev_warn(tx_ring->dev,
4372 					 "partial checksum but proto=%x!\n",
4373 					 first->protocol);
4374 			}
4375 			break;
4376 		}
4377 
4378 		switch (l4_hdr) {
4379 		case IPPROTO_TCP:
4380 			type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4381 			mss_l4len_idx = tcp_hdrlen(skb) <<
4382 					E1000_ADVTXD_L4LEN_SHIFT;
4383 			break;
4384 		case IPPROTO_SCTP:
4385 			type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4386 			mss_l4len_idx = sizeof(struct sctphdr) <<
4387 					E1000_ADVTXD_L4LEN_SHIFT;
4388 			break;
4389 		case IPPROTO_UDP:
4390 			mss_l4len_idx = sizeof(struct udphdr) <<
4391 					E1000_ADVTXD_L4LEN_SHIFT;
4392 			break;
4393 		default:
4394 			if (unlikely(net_ratelimit())) {
4395 				dev_warn(tx_ring->dev,
4396 					 "partial checksum but l4 proto=%x!\n",
4397 					 l4_hdr);
4398 			}
4399 			break;
4400 		}
4401 
4402 		/* update TX checksum flag */
4403 		first->tx_flags |= IGB_TX_FLAGS_CSUM;
4404 	}
4405 
4406 	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4407 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4408 
4409 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4410 }
4411 
4412 #define IGB_SET_FLAG(_input, _flag, _result) \
4413 	((_flag <= _result) ? \
4414 	 ((u32)(_input & _flag) * (_result / _flag)) : \
4415 	 ((u32)(_input & _flag) / (_flag / _result)))
4416 
4417 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4418 {
4419 	/* set type for advanced descriptor with frame checksum insertion */
4420 	u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4421 		       E1000_ADVTXD_DCMD_DEXT |
4422 		       E1000_ADVTXD_DCMD_IFCS;
4423 
4424 	/* set HW vlan bit if vlan is present */
4425 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4426 				 (E1000_ADVTXD_DCMD_VLE));
4427 
4428 	/* set segmentation bits for TSO */
4429 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4430 				 (E1000_ADVTXD_DCMD_TSE));
4431 
4432 	/* set timestamp bit if present */
4433 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4434 				 (E1000_ADVTXD_MAC_TSTAMP));
4435 
4436 	/* insert frame checksum */
4437 	cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4438 
4439 	return cmd_type;
4440 }
4441 
4442 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4443 				 union e1000_adv_tx_desc *tx_desc,
4444 				 u32 tx_flags, unsigned int paylen)
4445 {
4446 	u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4447 
4448 	/* 82575 requires a unique index per ring */
4449 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4450 		olinfo_status |= tx_ring->reg_idx << 4;
4451 
4452 	/* insert L4 checksum */
4453 	olinfo_status |= IGB_SET_FLAG(tx_flags,
4454 				      IGB_TX_FLAGS_CSUM,
4455 				      (E1000_TXD_POPTS_TXSM << 8));
4456 
4457 	/* insert IPv4 checksum */
4458 	olinfo_status |= IGB_SET_FLAG(tx_flags,
4459 				      IGB_TX_FLAGS_IPV4,
4460 				      (E1000_TXD_POPTS_IXSM << 8));
4461 
4462 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4463 }
4464 
4465 static void igb_tx_map(struct igb_ring *tx_ring,
4466 		       struct igb_tx_buffer *first,
4467 		       const u8 hdr_len)
4468 {
4469 	struct sk_buff *skb = first->skb;
4470 	struct igb_tx_buffer *tx_buffer;
4471 	union e1000_adv_tx_desc *tx_desc;
4472 	struct skb_frag_struct *frag;
4473 	dma_addr_t dma;
4474 	unsigned int data_len, size;
4475 	u32 tx_flags = first->tx_flags;
4476 	u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4477 	u16 i = tx_ring->next_to_use;
4478 
4479 	tx_desc = IGB_TX_DESC(tx_ring, i);
4480 
4481 	igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4482 
4483 	size = skb_headlen(skb);
4484 	data_len = skb->data_len;
4485 
4486 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4487 
4488 	tx_buffer = first;
4489 
4490 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4491 		if (dma_mapping_error(tx_ring->dev, dma))
4492 			goto dma_error;
4493 
4494 		/* record length, and DMA address */
4495 		dma_unmap_len_set(tx_buffer, len, size);
4496 		dma_unmap_addr_set(tx_buffer, dma, dma);
4497 
4498 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
4499 
4500 		while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4501 			tx_desc->read.cmd_type_len =
4502 				cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4503 
4504 			i++;
4505 			tx_desc++;
4506 			if (i == tx_ring->count) {
4507 				tx_desc = IGB_TX_DESC(tx_ring, 0);
4508 				i = 0;
4509 			}
4510 			tx_desc->read.olinfo_status = 0;
4511 
4512 			dma += IGB_MAX_DATA_PER_TXD;
4513 			size -= IGB_MAX_DATA_PER_TXD;
4514 
4515 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
4516 		}
4517 
4518 		if (likely(!data_len))
4519 			break;
4520 
4521 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4522 
4523 		i++;
4524 		tx_desc++;
4525 		if (i == tx_ring->count) {
4526 			tx_desc = IGB_TX_DESC(tx_ring, 0);
4527 			i = 0;
4528 		}
4529 		tx_desc->read.olinfo_status = 0;
4530 
4531 		size = skb_frag_size(frag);
4532 		data_len -= size;
4533 
4534 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4535 				       size, DMA_TO_DEVICE);
4536 
4537 		tx_buffer = &tx_ring->tx_buffer_info[i];
4538 	}
4539 
4540 	/* write last descriptor with RS and EOP bits */
4541 	cmd_type |= size | IGB_TXD_DCMD;
4542 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4543 
4544 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4545 
4546 	/* set the timestamp */
4547 	first->time_stamp = jiffies;
4548 
4549 	/* Force memory writes to complete before letting h/w know there
4550 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
4551 	 * memory model archs, such as IA-64).
4552 	 *
4553 	 * We also need this memory barrier to make certain all of the
4554 	 * status bits have been updated before next_to_watch is written.
4555 	 */
4556 	wmb();
4557 
4558 	/* set next_to_watch value indicating a packet is present */
4559 	first->next_to_watch = tx_desc;
4560 
4561 	i++;
4562 	if (i == tx_ring->count)
4563 		i = 0;
4564 
4565 	tx_ring->next_to_use = i;
4566 
4567 	writel(i, tx_ring->tail);
4568 
4569 	/* we need this if more than one processor can write to our tail
4570 	 * at a time, it synchronizes IO on IA64/Altix systems
4571 	 */
4572 	mmiowb();
4573 
4574 	return;
4575 
4576 dma_error:
4577 	dev_err(tx_ring->dev, "TX DMA map failed\n");
4578 
4579 	/* clear dma mappings for failed tx_buffer_info map */
4580 	for (;;) {
4581 		tx_buffer = &tx_ring->tx_buffer_info[i];
4582 		igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4583 		if (tx_buffer == first)
4584 			break;
4585 		if (i == 0)
4586 			i = tx_ring->count;
4587 		i--;
4588 	}
4589 
4590 	tx_ring->next_to_use = i;
4591 }
4592 
4593 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4594 {
4595 	struct net_device *netdev = tx_ring->netdev;
4596 
4597 	netif_stop_subqueue(netdev, tx_ring->queue_index);
4598 
4599 	/* Herbert's original patch had:
4600 	 *  smp_mb__after_netif_stop_queue();
4601 	 * but since that doesn't exist yet, just open code it.
4602 	 */
4603 	smp_mb();
4604 
4605 	/* We need to check again in a case another CPU has just
4606 	 * made room available.
4607 	 */
4608 	if (igb_desc_unused(tx_ring) < size)
4609 		return -EBUSY;
4610 
4611 	/* A reprieve! */
4612 	netif_wake_subqueue(netdev, tx_ring->queue_index);
4613 
4614 	u64_stats_update_begin(&tx_ring->tx_syncp2);
4615 	tx_ring->tx_stats.restart_queue2++;
4616 	u64_stats_update_end(&tx_ring->tx_syncp2);
4617 
4618 	return 0;
4619 }
4620 
4621 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4622 {
4623 	if (igb_desc_unused(tx_ring) >= size)
4624 		return 0;
4625 	return __igb_maybe_stop_tx(tx_ring, size);
4626 }
4627 
4628 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4629 				struct igb_ring *tx_ring)
4630 {
4631 	struct igb_tx_buffer *first;
4632 	int tso;
4633 	u32 tx_flags = 0;
4634 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
4635 	__be16 protocol = vlan_get_protocol(skb);
4636 	u8 hdr_len = 0;
4637 
4638 	/* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4639 	 *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4640 	 *       + 2 desc gap to keep tail from touching head,
4641 	 *       + 1 desc for context descriptor,
4642 	 * otherwise try next time
4643 	 */
4644 	if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4645 		unsigned short f;
4646 		for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4647 			count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4648 	} else {
4649 		count += skb_shinfo(skb)->nr_frags;
4650 	}
4651 
4652 	if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4653 		/* this is a hard error */
4654 		return NETDEV_TX_BUSY;
4655 	}
4656 
4657 	/* record the location of the first descriptor for this packet */
4658 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4659 	first->skb = skb;
4660 	first->bytecount = skb->len;
4661 	first->gso_segs = 1;
4662 
4663 	skb_tx_timestamp(skb);
4664 
4665 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4666 		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4667 
4668 		if (!(adapter->ptp_tx_skb)) {
4669 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4670 			tx_flags |= IGB_TX_FLAGS_TSTAMP;
4671 
4672 			adapter->ptp_tx_skb = skb_get(skb);
4673 			adapter->ptp_tx_start = jiffies;
4674 			if (adapter->hw.mac.type == e1000_82576)
4675 				schedule_work(&adapter->ptp_tx_work);
4676 		}
4677 	}
4678 
4679 	if (vlan_tx_tag_present(skb)) {
4680 		tx_flags |= IGB_TX_FLAGS_VLAN;
4681 		tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4682 	}
4683 
4684 	/* record initial flags and protocol */
4685 	first->tx_flags = tx_flags;
4686 	first->protocol = protocol;
4687 
4688 	tso = igb_tso(tx_ring, first, &hdr_len);
4689 	if (tso < 0)
4690 		goto out_drop;
4691 	else if (!tso)
4692 		igb_tx_csum(tx_ring, first);
4693 
4694 	igb_tx_map(tx_ring, first, hdr_len);
4695 
4696 	/* Make sure there is space in the ring for the next send. */
4697 	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4698 
4699 	return NETDEV_TX_OK;
4700 
4701 out_drop:
4702 	igb_unmap_and_free_tx_resource(tx_ring, first);
4703 
4704 	return NETDEV_TX_OK;
4705 }
4706 
4707 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4708 						    struct sk_buff *skb)
4709 {
4710 	unsigned int r_idx = skb->queue_mapping;
4711 
4712 	if (r_idx >= adapter->num_tx_queues)
4713 		r_idx = r_idx % adapter->num_tx_queues;
4714 
4715 	return adapter->tx_ring[r_idx];
4716 }
4717 
4718 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4719 				  struct net_device *netdev)
4720 {
4721 	struct igb_adapter *adapter = netdev_priv(netdev);
4722 
4723 	if (test_bit(__IGB_DOWN, &adapter->state)) {
4724 		dev_kfree_skb_any(skb);
4725 		return NETDEV_TX_OK;
4726 	}
4727 
4728 	if (skb->len <= 0) {
4729 		dev_kfree_skb_any(skb);
4730 		return NETDEV_TX_OK;
4731 	}
4732 
4733 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4734 	 * in order to meet this minimum size requirement.
4735 	 */
4736 	if (unlikely(skb->len < 17)) {
4737 		if (skb_pad(skb, 17 - skb->len))
4738 			return NETDEV_TX_OK;
4739 		skb->len = 17;
4740 		skb_set_tail_pointer(skb, 17);
4741 	}
4742 
4743 	return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4744 }
4745 
4746 /**
4747  *  igb_tx_timeout - Respond to a Tx Hang
4748  *  @netdev: network interface device structure
4749  **/
4750 static void igb_tx_timeout(struct net_device *netdev)
4751 {
4752 	struct igb_adapter *adapter = netdev_priv(netdev);
4753 	struct e1000_hw *hw = &adapter->hw;
4754 
4755 	/* Do the reset outside of interrupt context */
4756 	adapter->tx_timeout_count++;
4757 
4758 	if (hw->mac.type >= e1000_82580)
4759 		hw->dev_spec._82575.global_device_reset = true;
4760 
4761 	schedule_work(&adapter->reset_task);
4762 	wr32(E1000_EICS,
4763 	     (adapter->eims_enable_mask & ~adapter->eims_other));
4764 }
4765 
4766 static void igb_reset_task(struct work_struct *work)
4767 {
4768 	struct igb_adapter *adapter;
4769 	adapter = container_of(work, struct igb_adapter, reset_task);
4770 
4771 	igb_dump(adapter);
4772 	netdev_err(adapter->netdev, "Reset adapter\n");
4773 	igb_reinit_locked(adapter);
4774 }
4775 
4776 /**
4777  *  igb_get_stats64 - Get System Network Statistics
4778  *  @netdev: network interface device structure
4779  *  @stats: rtnl_link_stats64 pointer
4780  **/
4781 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4782 						struct rtnl_link_stats64 *stats)
4783 {
4784 	struct igb_adapter *adapter = netdev_priv(netdev);
4785 
4786 	spin_lock(&adapter->stats64_lock);
4787 	igb_update_stats(adapter, &adapter->stats64);
4788 	memcpy(stats, &adapter->stats64, sizeof(*stats));
4789 	spin_unlock(&adapter->stats64_lock);
4790 
4791 	return stats;
4792 }
4793 
4794 /**
4795  *  igb_change_mtu - Change the Maximum Transfer Unit
4796  *  @netdev: network interface device structure
4797  *  @new_mtu: new value for maximum frame size
4798  *
4799  *  Returns 0 on success, negative on failure
4800  **/
4801 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4802 {
4803 	struct igb_adapter *adapter = netdev_priv(netdev);
4804 	struct pci_dev *pdev = adapter->pdev;
4805 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4806 
4807 	if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4808 		dev_err(&pdev->dev, "Invalid MTU setting\n");
4809 		return -EINVAL;
4810 	}
4811 
4812 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4813 	if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4814 		dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4815 		return -EINVAL;
4816 	}
4817 
4818 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4819 		msleep(1);
4820 
4821 	/* igb_down has a dependency on max_frame_size */
4822 	adapter->max_frame_size = max_frame;
4823 
4824 	if (netif_running(netdev))
4825 		igb_down(adapter);
4826 
4827 	dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4828 		 netdev->mtu, new_mtu);
4829 	netdev->mtu = new_mtu;
4830 
4831 	if (netif_running(netdev))
4832 		igb_up(adapter);
4833 	else
4834 		igb_reset(adapter);
4835 
4836 	clear_bit(__IGB_RESETTING, &adapter->state);
4837 
4838 	return 0;
4839 }
4840 
4841 /**
4842  *  igb_update_stats - Update the board statistics counters
4843  *  @adapter: board private structure
4844  **/
4845 void igb_update_stats(struct igb_adapter *adapter,
4846 		      struct rtnl_link_stats64 *net_stats)
4847 {
4848 	struct e1000_hw *hw = &adapter->hw;
4849 	struct pci_dev *pdev = adapter->pdev;
4850 	u32 reg, mpc;
4851 	u16 phy_tmp;
4852 	int i;
4853 	u64 bytes, packets;
4854 	unsigned int start;
4855 	u64 _bytes, _packets;
4856 
4857 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4858 
4859 	/* Prevent stats update while adapter is being reset, or if the pci
4860 	 * connection is down.
4861 	 */
4862 	if (adapter->link_speed == 0)
4863 		return;
4864 	if (pci_channel_offline(pdev))
4865 		return;
4866 
4867 	bytes = 0;
4868 	packets = 0;
4869 	for (i = 0; i < adapter->num_rx_queues; i++) {
4870 		u32 rqdpc = rd32(E1000_RQDPC(i));
4871 		struct igb_ring *ring = adapter->rx_ring[i];
4872 
4873 		if (rqdpc) {
4874 			ring->rx_stats.drops += rqdpc;
4875 			net_stats->rx_fifo_errors += rqdpc;
4876 		}
4877 
4878 		do {
4879 			start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4880 			_bytes = ring->rx_stats.bytes;
4881 			_packets = ring->rx_stats.packets;
4882 		} while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4883 		bytes += _bytes;
4884 		packets += _packets;
4885 	}
4886 
4887 	net_stats->rx_bytes = bytes;
4888 	net_stats->rx_packets = packets;
4889 
4890 	bytes = 0;
4891 	packets = 0;
4892 	for (i = 0; i < adapter->num_tx_queues; i++) {
4893 		struct igb_ring *ring = adapter->tx_ring[i];
4894 		do {
4895 			start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4896 			_bytes = ring->tx_stats.bytes;
4897 			_packets = ring->tx_stats.packets;
4898 		} while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4899 		bytes += _bytes;
4900 		packets += _packets;
4901 	}
4902 	net_stats->tx_bytes = bytes;
4903 	net_stats->tx_packets = packets;
4904 
4905 	/* read stats registers */
4906 	adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4907 	adapter->stats.gprc += rd32(E1000_GPRC);
4908 	adapter->stats.gorc += rd32(E1000_GORCL);
4909 	rd32(E1000_GORCH); /* clear GORCL */
4910 	adapter->stats.bprc += rd32(E1000_BPRC);
4911 	adapter->stats.mprc += rd32(E1000_MPRC);
4912 	adapter->stats.roc += rd32(E1000_ROC);
4913 
4914 	adapter->stats.prc64 += rd32(E1000_PRC64);
4915 	adapter->stats.prc127 += rd32(E1000_PRC127);
4916 	adapter->stats.prc255 += rd32(E1000_PRC255);
4917 	adapter->stats.prc511 += rd32(E1000_PRC511);
4918 	adapter->stats.prc1023 += rd32(E1000_PRC1023);
4919 	adapter->stats.prc1522 += rd32(E1000_PRC1522);
4920 	adapter->stats.symerrs += rd32(E1000_SYMERRS);
4921 	adapter->stats.sec += rd32(E1000_SEC);
4922 
4923 	mpc = rd32(E1000_MPC);
4924 	adapter->stats.mpc += mpc;
4925 	net_stats->rx_fifo_errors += mpc;
4926 	adapter->stats.scc += rd32(E1000_SCC);
4927 	adapter->stats.ecol += rd32(E1000_ECOL);
4928 	adapter->stats.mcc += rd32(E1000_MCC);
4929 	adapter->stats.latecol += rd32(E1000_LATECOL);
4930 	adapter->stats.dc += rd32(E1000_DC);
4931 	adapter->stats.rlec += rd32(E1000_RLEC);
4932 	adapter->stats.xonrxc += rd32(E1000_XONRXC);
4933 	adapter->stats.xontxc += rd32(E1000_XONTXC);
4934 	adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4935 	adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4936 	adapter->stats.fcruc += rd32(E1000_FCRUC);
4937 	adapter->stats.gptc += rd32(E1000_GPTC);
4938 	adapter->stats.gotc += rd32(E1000_GOTCL);
4939 	rd32(E1000_GOTCH); /* clear GOTCL */
4940 	adapter->stats.rnbc += rd32(E1000_RNBC);
4941 	adapter->stats.ruc += rd32(E1000_RUC);
4942 	adapter->stats.rfc += rd32(E1000_RFC);
4943 	adapter->stats.rjc += rd32(E1000_RJC);
4944 	adapter->stats.tor += rd32(E1000_TORH);
4945 	adapter->stats.tot += rd32(E1000_TOTH);
4946 	adapter->stats.tpr += rd32(E1000_TPR);
4947 
4948 	adapter->stats.ptc64 += rd32(E1000_PTC64);
4949 	adapter->stats.ptc127 += rd32(E1000_PTC127);
4950 	adapter->stats.ptc255 += rd32(E1000_PTC255);
4951 	adapter->stats.ptc511 += rd32(E1000_PTC511);
4952 	adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4953 	adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4954 
4955 	adapter->stats.mptc += rd32(E1000_MPTC);
4956 	adapter->stats.bptc += rd32(E1000_BPTC);
4957 
4958 	adapter->stats.tpt += rd32(E1000_TPT);
4959 	adapter->stats.colc += rd32(E1000_COLC);
4960 
4961 	adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4962 	/* read internal phy specific stats */
4963 	reg = rd32(E1000_CTRL_EXT);
4964 	if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4965 		adapter->stats.rxerrc += rd32(E1000_RXERRC);
4966 
4967 		/* this stat has invalid values on i210/i211 */
4968 		if ((hw->mac.type != e1000_i210) &&
4969 		    (hw->mac.type != e1000_i211))
4970 			adapter->stats.tncrs += rd32(E1000_TNCRS);
4971 	}
4972 
4973 	adapter->stats.tsctc += rd32(E1000_TSCTC);
4974 	adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4975 
4976 	adapter->stats.iac += rd32(E1000_IAC);
4977 	adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4978 	adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4979 	adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4980 	adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4981 	adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4982 	adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4983 	adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4984 	adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4985 
4986 	/* Fill out the OS statistics structure */
4987 	net_stats->multicast = adapter->stats.mprc;
4988 	net_stats->collisions = adapter->stats.colc;
4989 
4990 	/* Rx Errors */
4991 
4992 	/* RLEC on some newer hardware can be incorrect so build
4993 	 * our own version based on RUC and ROC
4994 	 */
4995 	net_stats->rx_errors = adapter->stats.rxerrc +
4996 		adapter->stats.crcerrs + adapter->stats.algnerrc +
4997 		adapter->stats.ruc + adapter->stats.roc +
4998 		adapter->stats.cexterr;
4999 	net_stats->rx_length_errors = adapter->stats.ruc +
5000 				      adapter->stats.roc;
5001 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
5002 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
5003 	net_stats->rx_missed_errors = adapter->stats.mpc;
5004 
5005 	/* Tx Errors */
5006 	net_stats->tx_errors = adapter->stats.ecol +
5007 			       adapter->stats.latecol;
5008 	net_stats->tx_aborted_errors = adapter->stats.ecol;
5009 	net_stats->tx_window_errors = adapter->stats.latecol;
5010 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
5011 
5012 	/* Tx Dropped needs to be maintained elsewhere */
5013 
5014 	/* Phy Stats */
5015 	if (hw->phy.media_type == e1000_media_type_copper) {
5016 		if ((adapter->link_speed == SPEED_1000) &&
5017 		   (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5018 			phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5019 			adapter->phy_stats.idle_errors += phy_tmp;
5020 		}
5021 	}
5022 
5023 	/* Management Stats */
5024 	adapter->stats.mgptc += rd32(E1000_MGTPTC);
5025 	adapter->stats.mgprc += rd32(E1000_MGTPRC);
5026 	adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5027 
5028 	/* OS2BMC Stats */
5029 	reg = rd32(E1000_MANC);
5030 	if (reg & E1000_MANC_EN_BMC2OS) {
5031 		adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5032 		adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5033 		adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5034 		adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5035 	}
5036 }
5037 
5038 static irqreturn_t igb_msix_other(int irq, void *data)
5039 {
5040 	struct igb_adapter *adapter = data;
5041 	struct e1000_hw *hw = &adapter->hw;
5042 	u32 icr = rd32(E1000_ICR);
5043 	/* reading ICR causes bit 31 of EICR to be cleared */
5044 
5045 	if (icr & E1000_ICR_DRSTA)
5046 		schedule_work(&adapter->reset_task);
5047 
5048 	if (icr & E1000_ICR_DOUTSYNC) {
5049 		/* HW is reporting DMA is out of sync */
5050 		adapter->stats.doosync++;
5051 		/* The DMA Out of Sync is also indication of a spoof event
5052 		 * in IOV mode. Check the Wrong VM Behavior register to
5053 		 * see if it is really a spoof event.
5054 		 */
5055 		igb_check_wvbr(adapter);
5056 	}
5057 
5058 	/* Check for a mailbox event */
5059 	if (icr & E1000_ICR_VMMB)
5060 		igb_msg_task(adapter);
5061 
5062 	if (icr & E1000_ICR_LSC) {
5063 		hw->mac.get_link_status = 1;
5064 		/* guard against interrupt when we're going down */
5065 		if (!test_bit(__IGB_DOWN, &adapter->state))
5066 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5067 	}
5068 
5069 	if (icr & E1000_ICR_TS) {
5070 		u32 tsicr = rd32(E1000_TSICR);
5071 
5072 		if (tsicr & E1000_TSICR_TXTS) {
5073 			/* acknowledge the interrupt */
5074 			wr32(E1000_TSICR, E1000_TSICR_TXTS);
5075 			/* retrieve hardware timestamp */
5076 			schedule_work(&adapter->ptp_tx_work);
5077 		}
5078 	}
5079 
5080 	wr32(E1000_EIMS, adapter->eims_other);
5081 
5082 	return IRQ_HANDLED;
5083 }
5084 
5085 static void igb_write_itr(struct igb_q_vector *q_vector)
5086 {
5087 	struct igb_adapter *adapter = q_vector->adapter;
5088 	u32 itr_val = q_vector->itr_val & 0x7FFC;
5089 
5090 	if (!q_vector->set_itr)
5091 		return;
5092 
5093 	if (!itr_val)
5094 		itr_val = 0x4;
5095 
5096 	if (adapter->hw.mac.type == e1000_82575)
5097 		itr_val |= itr_val << 16;
5098 	else
5099 		itr_val |= E1000_EITR_CNT_IGNR;
5100 
5101 	writel(itr_val, q_vector->itr_register);
5102 	q_vector->set_itr = 0;
5103 }
5104 
5105 static irqreturn_t igb_msix_ring(int irq, void *data)
5106 {
5107 	struct igb_q_vector *q_vector = data;
5108 
5109 	/* Write the ITR value calculated from the previous interrupt. */
5110 	igb_write_itr(q_vector);
5111 
5112 	napi_schedule(&q_vector->napi);
5113 
5114 	return IRQ_HANDLED;
5115 }
5116 
5117 #ifdef CONFIG_IGB_DCA
5118 static void igb_update_tx_dca(struct igb_adapter *adapter,
5119 			      struct igb_ring *tx_ring,
5120 			      int cpu)
5121 {
5122 	struct e1000_hw *hw = &adapter->hw;
5123 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5124 
5125 	if (hw->mac.type != e1000_82575)
5126 		txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5127 
5128 	/* We can enable relaxed ordering for reads, but not writes when
5129 	 * DCA is enabled.  This is due to a known issue in some chipsets
5130 	 * which will cause the DCA tag to be cleared.
5131 	 */
5132 	txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5133 		  E1000_DCA_TXCTRL_DATA_RRO_EN |
5134 		  E1000_DCA_TXCTRL_DESC_DCA_EN;
5135 
5136 	wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5137 }
5138 
5139 static void igb_update_rx_dca(struct igb_adapter *adapter,
5140 			      struct igb_ring *rx_ring,
5141 			      int cpu)
5142 {
5143 	struct e1000_hw *hw = &adapter->hw;
5144 	u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5145 
5146 	if (hw->mac.type != e1000_82575)
5147 		rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5148 
5149 	/* We can enable relaxed ordering for reads, but not writes when
5150 	 * DCA is enabled.  This is due to a known issue in some chipsets
5151 	 * which will cause the DCA tag to be cleared.
5152 	 */
5153 	rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5154 		  E1000_DCA_RXCTRL_DESC_DCA_EN;
5155 
5156 	wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5157 }
5158 
5159 static void igb_update_dca(struct igb_q_vector *q_vector)
5160 {
5161 	struct igb_adapter *adapter = q_vector->adapter;
5162 	int cpu = get_cpu();
5163 
5164 	if (q_vector->cpu == cpu)
5165 		goto out_no_update;
5166 
5167 	if (q_vector->tx.ring)
5168 		igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5169 
5170 	if (q_vector->rx.ring)
5171 		igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5172 
5173 	q_vector->cpu = cpu;
5174 out_no_update:
5175 	put_cpu();
5176 }
5177 
5178 static void igb_setup_dca(struct igb_adapter *adapter)
5179 {
5180 	struct e1000_hw *hw = &adapter->hw;
5181 	int i;
5182 
5183 	if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5184 		return;
5185 
5186 	/* Always use CB2 mode, difference is masked in the CB driver. */
5187 	wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5188 
5189 	for (i = 0; i < adapter->num_q_vectors; i++) {
5190 		adapter->q_vector[i]->cpu = -1;
5191 		igb_update_dca(adapter->q_vector[i]);
5192 	}
5193 }
5194 
5195 static int __igb_notify_dca(struct device *dev, void *data)
5196 {
5197 	struct net_device *netdev = dev_get_drvdata(dev);
5198 	struct igb_adapter *adapter = netdev_priv(netdev);
5199 	struct pci_dev *pdev = adapter->pdev;
5200 	struct e1000_hw *hw = &adapter->hw;
5201 	unsigned long event = *(unsigned long *)data;
5202 
5203 	switch (event) {
5204 	case DCA_PROVIDER_ADD:
5205 		/* if already enabled, don't do it again */
5206 		if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5207 			break;
5208 		if (dca_add_requester(dev) == 0) {
5209 			adapter->flags |= IGB_FLAG_DCA_ENABLED;
5210 			dev_info(&pdev->dev, "DCA enabled\n");
5211 			igb_setup_dca(adapter);
5212 			break;
5213 		}
5214 		/* Fall Through since DCA is disabled. */
5215 	case DCA_PROVIDER_REMOVE:
5216 		if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5217 			/* without this a class_device is left
5218 			 * hanging around in the sysfs model
5219 			 */
5220 			dca_remove_requester(dev);
5221 			dev_info(&pdev->dev, "DCA disabled\n");
5222 			adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5223 			wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5224 		}
5225 		break;
5226 	}
5227 
5228 	return 0;
5229 }
5230 
5231 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5232 			  void *p)
5233 {
5234 	int ret_val;
5235 
5236 	ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5237 					 __igb_notify_dca);
5238 
5239 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5240 }
5241 #endif /* CONFIG_IGB_DCA */
5242 
5243 #ifdef CONFIG_PCI_IOV
5244 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5245 {
5246 	unsigned char mac_addr[ETH_ALEN];
5247 
5248 	eth_zero_addr(mac_addr);
5249 	igb_set_vf_mac(adapter, vf, mac_addr);
5250 
5251 	/* By default spoof check is enabled for all VFs */
5252 	adapter->vf_data[vf].spoofchk_enabled = true;
5253 
5254 	return 0;
5255 }
5256 
5257 #endif
5258 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5259 {
5260 	struct e1000_hw *hw = &adapter->hw;
5261 	u32 ping;
5262 	int i;
5263 
5264 	for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5265 		ping = E1000_PF_CONTROL_MSG;
5266 		if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5267 			ping |= E1000_VT_MSGTYPE_CTS;
5268 		igb_write_mbx(hw, &ping, 1, i);
5269 	}
5270 }
5271 
5272 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5273 {
5274 	struct e1000_hw *hw = &adapter->hw;
5275 	u32 vmolr = rd32(E1000_VMOLR(vf));
5276 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5277 
5278 	vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5279 			    IGB_VF_FLAG_MULTI_PROMISC);
5280 	vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5281 
5282 	if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5283 		vmolr |= E1000_VMOLR_MPME;
5284 		vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5285 		*msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5286 	} else {
5287 		/* if we have hashes and we are clearing a multicast promisc
5288 		 * flag we need to write the hashes to the MTA as this step
5289 		 * was previously skipped
5290 		 */
5291 		if (vf_data->num_vf_mc_hashes > 30) {
5292 			vmolr |= E1000_VMOLR_MPME;
5293 		} else if (vf_data->num_vf_mc_hashes) {
5294 			int j;
5295 			vmolr |= E1000_VMOLR_ROMPE;
5296 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5297 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5298 		}
5299 	}
5300 
5301 	wr32(E1000_VMOLR(vf), vmolr);
5302 
5303 	/* there are flags left unprocessed, likely not supported */
5304 	if (*msgbuf & E1000_VT_MSGINFO_MASK)
5305 		return -EINVAL;
5306 
5307 	return 0;
5308 }
5309 
5310 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5311 				  u32 *msgbuf, u32 vf)
5312 {
5313 	int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5314 	u16 *hash_list = (u16 *)&msgbuf[1];
5315 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5316 	int i;
5317 
5318 	/* salt away the number of multicast addresses assigned
5319 	 * to this VF for later use to restore when the PF multi cast
5320 	 * list changes
5321 	 */
5322 	vf_data->num_vf_mc_hashes = n;
5323 
5324 	/* only up to 30 hash values supported */
5325 	if (n > 30)
5326 		n = 30;
5327 
5328 	/* store the hashes for later use */
5329 	for (i = 0; i < n; i++)
5330 		vf_data->vf_mc_hashes[i] = hash_list[i];
5331 
5332 	/* Flush and reset the mta with the new values */
5333 	igb_set_rx_mode(adapter->netdev);
5334 
5335 	return 0;
5336 }
5337 
5338 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5339 {
5340 	struct e1000_hw *hw = &adapter->hw;
5341 	struct vf_data_storage *vf_data;
5342 	int i, j;
5343 
5344 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
5345 		u32 vmolr = rd32(E1000_VMOLR(i));
5346 		vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5347 
5348 		vf_data = &adapter->vf_data[i];
5349 
5350 		if ((vf_data->num_vf_mc_hashes > 30) ||
5351 		    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5352 			vmolr |= E1000_VMOLR_MPME;
5353 		} else if (vf_data->num_vf_mc_hashes) {
5354 			vmolr |= E1000_VMOLR_ROMPE;
5355 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5356 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5357 		}
5358 		wr32(E1000_VMOLR(i), vmolr);
5359 	}
5360 }
5361 
5362 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5363 {
5364 	struct e1000_hw *hw = &adapter->hw;
5365 	u32 pool_mask, reg, vid;
5366 	int i;
5367 
5368 	pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5369 
5370 	/* Find the vlan filter for this id */
5371 	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5372 		reg = rd32(E1000_VLVF(i));
5373 
5374 		/* remove the vf from the pool */
5375 		reg &= ~pool_mask;
5376 
5377 		/* if pool is empty then remove entry from vfta */
5378 		if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5379 		    (reg & E1000_VLVF_VLANID_ENABLE)) {
5380 			reg = 0;
5381 			vid = reg & E1000_VLVF_VLANID_MASK;
5382 			igb_vfta_set(hw, vid, false);
5383 		}
5384 
5385 		wr32(E1000_VLVF(i), reg);
5386 	}
5387 
5388 	adapter->vf_data[vf].vlans_enabled = 0;
5389 }
5390 
5391 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5392 {
5393 	struct e1000_hw *hw = &adapter->hw;
5394 	u32 reg, i;
5395 
5396 	/* The vlvf table only exists on 82576 hardware and newer */
5397 	if (hw->mac.type < e1000_82576)
5398 		return -1;
5399 
5400 	/* we only need to do this if VMDq is enabled */
5401 	if (!adapter->vfs_allocated_count)
5402 		return -1;
5403 
5404 	/* Find the vlan filter for this id */
5405 	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5406 		reg = rd32(E1000_VLVF(i));
5407 		if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5408 		    vid == (reg & E1000_VLVF_VLANID_MASK))
5409 			break;
5410 	}
5411 
5412 	if (add) {
5413 		if (i == E1000_VLVF_ARRAY_SIZE) {
5414 			/* Did not find a matching VLAN ID entry that was
5415 			 * enabled.  Search for a free filter entry, i.e.
5416 			 * one without the enable bit set
5417 			 */
5418 			for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5419 				reg = rd32(E1000_VLVF(i));
5420 				if (!(reg & E1000_VLVF_VLANID_ENABLE))
5421 					break;
5422 			}
5423 		}
5424 		if (i < E1000_VLVF_ARRAY_SIZE) {
5425 			/* Found an enabled/available entry */
5426 			reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5427 
5428 			/* if !enabled we need to set this up in vfta */
5429 			if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5430 				/* add VID to filter table */
5431 				igb_vfta_set(hw, vid, true);
5432 				reg |= E1000_VLVF_VLANID_ENABLE;
5433 			}
5434 			reg &= ~E1000_VLVF_VLANID_MASK;
5435 			reg |= vid;
5436 			wr32(E1000_VLVF(i), reg);
5437 
5438 			/* do not modify RLPML for PF devices */
5439 			if (vf >= adapter->vfs_allocated_count)
5440 				return 0;
5441 
5442 			if (!adapter->vf_data[vf].vlans_enabled) {
5443 				u32 size;
5444 				reg = rd32(E1000_VMOLR(vf));
5445 				size = reg & E1000_VMOLR_RLPML_MASK;
5446 				size += 4;
5447 				reg &= ~E1000_VMOLR_RLPML_MASK;
5448 				reg |= size;
5449 				wr32(E1000_VMOLR(vf), reg);
5450 			}
5451 
5452 			adapter->vf_data[vf].vlans_enabled++;
5453 		}
5454 	} else {
5455 		if (i < E1000_VLVF_ARRAY_SIZE) {
5456 			/* remove vf from the pool */
5457 			reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5458 			/* if pool is empty then remove entry from vfta */
5459 			if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5460 				reg = 0;
5461 				igb_vfta_set(hw, vid, false);
5462 			}
5463 			wr32(E1000_VLVF(i), reg);
5464 
5465 			/* do not modify RLPML for PF devices */
5466 			if (vf >= adapter->vfs_allocated_count)
5467 				return 0;
5468 
5469 			adapter->vf_data[vf].vlans_enabled--;
5470 			if (!adapter->vf_data[vf].vlans_enabled) {
5471 				u32 size;
5472 				reg = rd32(E1000_VMOLR(vf));
5473 				size = reg & E1000_VMOLR_RLPML_MASK;
5474 				size -= 4;
5475 				reg &= ~E1000_VMOLR_RLPML_MASK;
5476 				reg |= size;
5477 				wr32(E1000_VMOLR(vf), reg);
5478 			}
5479 		}
5480 	}
5481 	return 0;
5482 }
5483 
5484 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5485 {
5486 	struct e1000_hw *hw = &adapter->hw;
5487 
5488 	if (vid)
5489 		wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5490 	else
5491 		wr32(E1000_VMVIR(vf), 0);
5492 }
5493 
5494 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5495 			       int vf, u16 vlan, u8 qos)
5496 {
5497 	int err = 0;
5498 	struct igb_adapter *adapter = netdev_priv(netdev);
5499 
5500 	if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5501 		return -EINVAL;
5502 	if (vlan || qos) {
5503 		err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5504 		if (err)
5505 			goto out;
5506 		igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5507 		igb_set_vmolr(adapter, vf, !vlan);
5508 		adapter->vf_data[vf].pf_vlan = vlan;
5509 		adapter->vf_data[vf].pf_qos = qos;
5510 		dev_info(&adapter->pdev->dev,
5511 			 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5512 		if (test_bit(__IGB_DOWN, &adapter->state)) {
5513 			dev_warn(&adapter->pdev->dev,
5514 				 "The VF VLAN has been set, but the PF device is not up.\n");
5515 			dev_warn(&adapter->pdev->dev,
5516 				 "Bring the PF device up before attempting to use the VF device.\n");
5517 		}
5518 	} else {
5519 		igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5520 			     false, vf);
5521 		igb_set_vmvir(adapter, vlan, vf);
5522 		igb_set_vmolr(adapter, vf, true);
5523 		adapter->vf_data[vf].pf_vlan = 0;
5524 		adapter->vf_data[vf].pf_qos = 0;
5525 	}
5526 out:
5527 	return err;
5528 }
5529 
5530 static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5531 {
5532 	struct e1000_hw *hw = &adapter->hw;
5533 	int i;
5534 	u32 reg;
5535 
5536 	/* Find the vlan filter for this id */
5537 	for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5538 		reg = rd32(E1000_VLVF(i));
5539 		if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5540 		    vid == (reg & E1000_VLVF_VLANID_MASK))
5541 			break;
5542 	}
5543 
5544 	if (i >= E1000_VLVF_ARRAY_SIZE)
5545 		i = -1;
5546 
5547 	return i;
5548 }
5549 
5550 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5551 {
5552 	struct e1000_hw *hw = &adapter->hw;
5553 	int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5554 	int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5555 	int err = 0;
5556 
5557 	/* If in promiscuous mode we need to make sure the PF also has
5558 	 * the VLAN filter set.
5559 	 */
5560 	if (add && (adapter->netdev->flags & IFF_PROMISC))
5561 		err = igb_vlvf_set(adapter, vid, add,
5562 				   adapter->vfs_allocated_count);
5563 	if (err)
5564 		goto out;
5565 
5566 	err = igb_vlvf_set(adapter, vid, add, vf);
5567 
5568 	if (err)
5569 		goto out;
5570 
5571 	/* Go through all the checks to see if the VLAN filter should
5572 	 * be wiped completely.
5573 	 */
5574 	if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5575 		u32 vlvf, bits;
5576 
5577 		int regndx = igb_find_vlvf_entry(adapter, vid);
5578 		if (regndx < 0)
5579 			goto out;
5580 		/* See if any other pools are set for this VLAN filter
5581 		 * entry other than the PF.
5582 		 */
5583 		vlvf = bits = rd32(E1000_VLVF(regndx));
5584 		bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5585 			      adapter->vfs_allocated_count);
5586 		/* If the filter was removed then ensure PF pool bit
5587 		 * is cleared if the PF only added itself to the pool
5588 		 * because the PF is in promiscuous mode.
5589 		 */
5590 		if ((vlvf & VLAN_VID_MASK) == vid &&
5591 		    !test_bit(vid, adapter->active_vlans) &&
5592 		    !bits)
5593 			igb_vlvf_set(adapter, vid, add,
5594 				     adapter->vfs_allocated_count);
5595 	}
5596 
5597 out:
5598 	return err;
5599 }
5600 
5601 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5602 {
5603 	/* clear flags - except flag that indicates PF has set the MAC */
5604 	adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5605 	adapter->vf_data[vf].last_nack = jiffies;
5606 
5607 	/* reset offloads to defaults */
5608 	igb_set_vmolr(adapter, vf, true);
5609 
5610 	/* reset vlans for device */
5611 	igb_clear_vf_vfta(adapter, vf);
5612 	if (adapter->vf_data[vf].pf_vlan)
5613 		igb_ndo_set_vf_vlan(adapter->netdev, vf,
5614 				    adapter->vf_data[vf].pf_vlan,
5615 				    adapter->vf_data[vf].pf_qos);
5616 	else
5617 		igb_clear_vf_vfta(adapter, vf);
5618 
5619 	/* reset multicast table array for vf */
5620 	adapter->vf_data[vf].num_vf_mc_hashes = 0;
5621 
5622 	/* Flush and reset the mta with the new values */
5623 	igb_set_rx_mode(adapter->netdev);
5624 }
5625 
5626 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5627 {
5628 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5629 
5630 	/* clear mac address as we were hotplug removed/added */
5631 	if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5632 		eth_zero_addr(vf_mac);
5633 
5634 	/* process remaining reset events */
5635 	igb_vf_reset(adapter, vf);
5636 }
5637 
5638 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5639 {
5640 	struct e1000_hw *hw = &adapter->hw;
5641 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5642 	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5643 	u32 reg, msgbuf[3];
5644 	u8 *addr = (u8 *)(&msgbuf[1]);
5645 
5646 	/* process all the same items cleared in a function level reset */
5647 	igb_vf_reset(adapter, vf);
5648 
5649 	/* set vf mac address */
5650 	igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5651 
5652 	/* enable transmit and receive for vf */
5653 	reg = rd32(E1000_VFTE);
5654 	wr32(E1000_VFTE, reg | (1 << vf));
5655 	reg = rd32(E1000_VFRE);
5656 	wr32(E1000_VFRE, reg | (1 << vf));
5657 
5658 	adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5659 
5660 	/* reply to reset with ack and vf mac address */
5661 	msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5662 	memcpy(addr, vf_mac, 6);
5663 	igb_write_mbx(hw, msgbuf, 3, vf);
5664 }
5665 
5666 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5667 {
5668 	/* The VF MAC Address is stored in a packed array of bytes
5669 	 * starting at the second 32 bit word of the msg array
5670 	 */
5671 	unsigned char *addr = (char *)&msg[1];
5672 	int err = -1;
5673 
5674 	if (is_valid_ether_addr(addr))
5675 		err = igb_set_vf_mac(adapter, vf, addr);
5676 
5677 	return err;
5678 }
5679 
5680 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5681 {
5682 	struct e1000_hw *hw = &adapter->hw;
5683 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5684 	u32 msg = E1000_VT_MSGTYPE_NACK;
5685 
5686 	/* if device isn't clear to send it shouldn't be reading either */
5687 	if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5688 	    time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5689 		igb_write_mbx(hw, &msg, 1, vf);
5690 		vf_data->last_nack = jiffies;
5691 	}
5692 }
5693 
5694 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5695 {
5696 	struct pci_dev *pdev = adapter->pdev;
5697 	u32 msgbuf[E1000_VFMAILBOX_SIZE];
5698 	struct e1000_hw *hw = &adapter->hw;
5699 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5700 	s32 retval;
5701 
5702 	retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5703 
5704 	if (retval) {
5705 		/* if receive failed revoke VF CTS stats and restart init */
5706 		dev_err(&pdev->dev, "Error receiving message from VF\n");
5707 		vf_data->flags &= ~IGB_VF_FLAG_CTS;
5708 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5709 			return;
5710 		goto out;
5711 	}
5712 
5713 	/* this is a message we already processed, do nothing */
5714 	if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5715 		return;
5716 
5717 	/* until the vf completes a reset it should not be
5718 	 * allowed to start any configuration.
5719 	 */
5720 	if (msgbuf[0] == E1000_VF_RESET) {
5721 		igb_vf_reset_msg(adapter, vf);
5722 		return;
5723 	}
5724 
5725 	if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5726 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5727 			return;
5728 		retval = -1;
5729 		goto out;
5730 	}
5731 
5732 	switch ((msgbuf[0] & 0xFFFF)) {
5733 	case E1000_VF_SET_MAC_ADDR:
5734 		retval = -EINVAL;
5735 		if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5736 			retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5737 		else
5738 			dev_warn(&pdev->dev,
5739 				 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5740 				 vf);
5741 		break;
5742 	case E1000_VF_SET_PROMISC:
5743 		retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5744 		break;
5745 	case E1000_VF_SET_MULTICAST:
5746 		retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5747 		break;
5748 	case E1000_VF_SET_LPE:
5749 		retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5750 		break;
5751 	case E1000_VF_SET_VLAN:
5752 		retval = -1;
5753 		if (vf_data->pf_vlan)
5754 			dev_warn(&pdev->dev,
5755 				 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5756 				 vf);
5757 		else
5758 			retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5759 		break;
5760 	default:
5761 		dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5762 		retval = -1;
5763 		break;
5764 	}
5765 
5766 	msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5767 out:
5768 	/* notify the VF of the results of what it sent us */
5769 	if (retval)
5770 		msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5771 	else
5772 		msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5773 
5774 	igb_write_mbx(hw, msgbuf, 1, vf);
5775 }
5776 
5777 static void igb_msg_task(struct igb_adapter *adapter)
5778 {
5779 	struct e1000_hw *hw = &adapter->hw;
5780 	u32 vf;
5781 
5782 	for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5783 		/* process any reset requests */
5784 		if (!igb_check_for_rst(hw, vf))
5785 			igb_vf_reset_event(adapter, vf);
5786 
5787 		/* process any messages pending */
5788 		if (!igb_check_for_msg(hw, vf))
5789 			igb_rcv_msg_from_vf(adapter, vf);
5790 
5791 		/* process any acks */
5792 		if (!igb_check_for_ack(hw, vf))
5793 			igb_rcv_ack_from_vf(adapter, vf);
5794 	}
5795 }
5796 
5797 /**
5798  *  igb_set_uta - Set unicast filter table address
5799  *  @adapter: board private structure
5800  *
5801  *  The unicast table address is a register array of 32-bit registers.
5802  *  The table is meant to be used in a way similar to how the MTA is used
5803  *  however due to certain limitations in the hardware it is necessary to
5804  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5805  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
5806  **/
5807 static void igb_set_uta(struct igb_adapter *adapter)
5808 {
5809 	struct e1000_hw *hw = &adapter->hw;
5810 	int i;
5811 
5812 	/* The UTA table only exists on 82576 hardware and newer */
5813 	if (hw->mac.type < e1000_82576)
5814 		return;
5815 
5816 	/* we only need to do this if VMDq is enabled */
5817 	if (!adapter->vfs_allocated_count)
5818 		return;
5819 
5820 	for (i = 0; i < hw->mac.uta_reg_count; i++)
5821 		array_wr32(E1000_UTA, i, ~0);
5822 }
5823 
5824 /**
5825  *  igb_intr_msi - Interrupt Handler
5826  *  @irq: interrupt number
5827  *  @data: pointer to a network interface device structure
5828  **/
5829 static irqreturn_t igb_intr_msi(int irq, void *data)
5830 {
5831 	struct igb_adapter *adapter = data;
5832 	struct igb_q_vector *q_vector = adapter->q_vector[0];
5833 	struct e1000_hw *hw = &adapter->hw;
5834 	/* read ICR disables interrupts using IAM */
5835 	u32 icr = rd32(E1000_ICR);
5836 
5837 	igb_write_itr(q_vector);
5838 
5839 	if (icr & E1000_ICR_DRSTA)
5840 		schedule_work(&adapter->reset_task);
5841 
5842 	if (icr & E1000_ICR_DOUTSYNC) {
5843 		/* HW is reporting DMA is out of sync */
5844 		adapter->stats.doosync++;
5845 	}
5846 
5847 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5848 		hw->mac.get_link_status = 1;
5849 		if (!test_bit(__IGB_DOWN, &adapter->state))
5850 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5851 	}
5852 
5853 	if (icr & E1000_ICR_TS) {
5854 		u32 tsicr = rd32(E1000_TSICR);
5855 
5856 		if (tsicr & E1000_TSICR_TXTS) {
5857 			/* acknowledge the interrupt */
5858 			wr32(E1000_TSICR, E1000_TSICR_TXTS);
5859 			/* retrieve hardware timestamp */
5860 			schedule_work(&adapter->ptp_tx_work);
5861 		}
5862 	}
5863 
5864 	napi_schedule(&q_vector->napi);
5865 
5866 	return IRQ_HANDLED;
5867 }
5868 
5869 /**
5870  *  igb_intr - Legacy Interrupt Handler
5871  *  @irq: interrupt number
5872  *  @data: pointer to a network interface device structure
5873  **/
5874 static irqreturn_t igb_intr(int irq, void *data)
5875 {
5876 	struct igb_adapter *adapter = data;
5877 	struct igb_q_vector *q_vector = adapter->q_vector[0];
5878 	struct e1000_hw *hw = &adapter->hw;
5879 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5880 	 * need for the IMC write
5881 	 */
5882 	u32 icr = rd32(E1000_ICR);
5883 
5884 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5885 	 * not set, then the adapter didn't send an interrupt
5886 	 */
5887 	if (!(icr & E1000_ICR_INT_ASSERTED))
5888 		return IRQ_NONE;
5889 
5890 	igb_write_itr(q_vector);
5891 
5892 	if (icr & E1000_ICR_DRSTA)
5893 		schedule_work(&adapter->reset_task);
5894 
5895 	if (icr & E1000_ICR_DOUTSYNC) {
5896 		/* HW is reporting DMA is out of sync */
5897 		adapter->stats.doosync++;
5898 	}
5899 
5900 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5901 		hw->mac.get_link_status = 1;
5902 		/* guard against interrupt when we're going down */
5903 		if (!test_bit(__IGB_DOWN, &adapter->state))
5904 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5905 	}
5906 
5907 	if (icr & E1000_ICR_TS) {
5908 		u32 tsicr = rd32(E1000_TSICR);
5909 
5910 		if (tsicr & E1000_TSICR_TXTS) {
5911 			/* acknowledge the interrupt */
5912 			wr32(E1000_TSICR, E1000_TSICR_TXTS);
5913 			/* retrieve hardware timestamp */
5914 			schedule_work(&adapter->ptp_tx_work);
5915 		}
5916 	}
5917 
5918 	napi_schedule(&q_vector->napi);
5919 
5920 	return IRQ_HANDLED;
5921 }
5922 
5923 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5924 {
5925 	struct igb_adapter *adapter = q_vector->adapter;
5926 	struct e1000_hw *hw = &adapter->hw;
5927 
5928 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5929 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5930 		if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5931 			igb_set_itr(q_vector);
5932 		else
5933 			igb_update_ring_itr(q_vector);
5934 	}
5935 
5936 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
5937 		if (adapter->msix_entries)
5938 			wr32(E1000_EIMS, q_vector->eims_value);
5939 		else
5940 			igb_irq_enable(adapter);
5941 	}
5942 }
5943 
5944 /**
5945  *  igb_poll - NAPI Rx polling callback
5946  *  @napi: napi polling structure
5947  *  @budget: count of how many packets we should handle
5948  **/
5949 static int igb_poll(struct napi_struct *napi, int budget)
5950 {
5951 	struct igb_q_vector *q_vector = container_of(napi,
5952 						     struct igb_q_vector,
5953 						     napi);
5954 	bool clean_complete = true;
5955 
5956 #ifdef CONFIG_IGB_DCA
5957 	if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5958 		igb_update_dca(q_vector);
5959 #endif
5960 	if (q_vector->tx.ring)
5961 		clean_complete = igb_clean_tx_irq(q_vector);
5962 
5963 	if (q_vector->rx.ring)
5964 		clean_complete &= igb_clean_rx_irq(q_vector, budget);
5965 
5966 	/* If all work not completed, return budget and keep polling */
5967 	if (!clean_complete)
5968 		return budget;
5969 
5970 	/* If not enough Rx work done, exit the polling mode */
5971 	napi_complete(napi);
5972 	igb_ring_irq_enable(q_vector);
5973 
5974 	return 0;
5975 }
5976 
5977 /**
5978  *  igb_clean_tx_irq - Reclaim resources after transmit completes
5979  *  @q_vector: pointer to q_vector containing needed info
5980  *
5981  *  returns true if ring is completely cleaned
5982  **/
5983 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5984 {
5985 	struct igb_adapter *adapter = q_vector->adapter;
5986 	struct igb_ring *tx_ring = q_vector->tx.ring;
5987 	struct igb_tx_buffer *tx_buffer;
5988 	union e1000_adv_tx_desc *tx_desc;
5989 	unsigned int total_bytes = 0, total_packets = 0;
5990 	unsigned int budget = q_vector->tx.work_limit;
5991 	unsigned int i = tx_ring->next_to_clean;
5992 
5993 	if (test_bit(__IGB_DOWN, &adapter->state))
5994 		return true;
5995 
5996 	tx_buffer = &tx_ring->tx_buffer_info[i];
5997 	tx_desc = IGB_TX_DESC(tx_ring, i);
5998 	i -= tx_ring->count;
5999 
6000 	do {
6001 		union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6002 
6003 		/* if next_to_watch is not set then there is no work pending */
6004 		if (!eop_desc)
6005 			break;
6006 
6007 		/* prevent any other reads prior to eop_desc */
6008 		read_barrier_depends();
6009 
6010 		/* if DD is not set pending work has not been completed */
6011 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6012 			break;
6013 
6014 		/* clear next_to_watch to prevent false hangs */
6015 		tx_buffer->next_to_watch = NULL;
6016 
6017 		/* update the statistics for this packet */
6018 		total_bytes += tx_buffer->bytecount;
6019 		total_packets += tx_buffer->gso_segs;
6020 
6021 		/* free the skb */
6022 		dev_kfree_skb_any(tx_buffer->skb);
6023 
6024 		/* unmap skb header data */
6025 		dma_unmap_single(tx_ring->dev,
6026 				 dma_unmap_addr(tx_buffer, dma),
6027 				 dma_unmap_len(tx_buffer, len),
6028 				 DMA_TO_DEVICE);
6029 
6030 		/* clear tx_buffer data */
6031 		tx_buffer->skb = NULL;
6032 		dma_unmap_len_set(tx_buffer, len, 0);
6033 
6034 		/* clear last DMA location and unmap remaining buffers */
6035 		while (tx_desc != eop_desc) {
6036 			tx_buffer++;
6037 			tx_desc++;
6038 			i++;
6039 			if (unlikely(!i)) {
6040 				i -= tx_ring->count;
6041 				tx_buffer = tx_ring->tx_buffer_info;
6042 				tx_desc = IGB_TX_DESC(tx_ring, 0);
6043 			}
6044 
6045 			/* unmap any remaining paged data */
6046 			if (dma_unmap_len(tx_buffer, len)) {
6047 				dma_unmap_page(tx_ring->dev,
6048 					       dma_unmap_addr(tx_buffer, dma),
6049 					       dma_unmap_len(tx_buffer, len),
6050 					       DMA_TO_DEVICE);
6051 				dma_unmap_len_set(tx_buffer, len, 0);
6052 			}
6053 		}
6054 
6055 		/* move us one more past the eop_desc for start of next pkt */
6056 		tx_buffer++;
6057 		tx_desc++;
6058 		i++;
6059 		if (unlikely(!i)) {
6060 			i -= tx_ring->count;
6061 			tx_buffer = tx_ring->tx_buffer_info;
6062 			tx_desc = IGB_TX_DESC(tx_ring, 0);
6063 		}
6064 
6065 		/* issue prefetch for next Tx descriptor */
6066 		prefetch(tx_desc);
6067 
6068 		/* update budget accounting */
6069 		budget--;
6070 	} while (likely(budget));
6071 
6072 	netdev_tx_completed_queue(txring_txq(tx_ring),
6073 				  total_packets, total_bytes);
6074 	i += tx_ring->count;
6075 	tx_ring->next_to_clean = i;
6076 	u64_stats_update_begin(&tx_ring->tx_syncp);
6077 	tx_ring->tx_stats.bytes += total_bytes;
6078 	tx_ring->tx_stats.packets += total_packets;
6079 	u64_stats_update_end(&tx_ring->tx_syncp);
6080 	q_vector->tx.total_bytes += total_bytes;
6081 	q_vector->tx.total_packets += total_packets;
6082 
6083 	if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6084 		struct e1000_hw *hw = &adapter->hw;
6085 
6086 		/* Detect a transmit hang in hardware, this serializes the
6087 		 * check with the clearing of time_stamp and movement of i
6088 		 */
6089 		clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6090 		if (tx_buffer->next_to_watch &&
6091 		    time_after(jiffies, tx_buffer->time_stamp +
6092 			       (adapter->tx_timeout_factor * HZ)) &&
6093 		    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6094 
6095 			/* detected Tx unit hang */
6096 			dev_err(tx_ring->dev,
6097 				"Detected Tx Unit Hang\n"
6098 				"  Tx Queue             <%d>\n"
6099 				"  TDH                  <%x>\n"
6100 				"  TDT                  <%x>\n"
6101 				"  next_to_use          <%x>\n"
6102 				"  next_to_clean        <%x>\n"
6103 				"buffer_info[next_to_clean]\n"
6104 				"  time_stamp           <%lx>\n"
6105 				"  next_to_watch        <%p>\n"
6106 				"  jiffies              <%lx>\n"
6107 				"  desc.status          <%x>\n",
6108 				tx_ring->queue_index,
6109 				rd32(E1000_TDH(tx_ring->reg_idx)),
6110 				readl(tx_ring->tail),
6111 				tx_ring->next_to_use,
6112 				tx_ring->next_to_clean,
6113 				tx_buffer->time_stamp,
6114 				tx_buffer->next_to_watch,
6115 				jiffies,
6116 				tx_buffer->next_to_watch->wb.status);
6117 			netif_stop_subqueue(tx_ring->netdev,
6118 					    tx_ring->queue_index);
6119 
6120 			/* we are about to reset, no point in enabling stuff */
6121 			return true;
6122 		}
6123 	}
6124 
6125 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6126 	if (unlikely(total_packets &&
6127 	    netif_carrier_ok(tx_ring->netdev) &&
6128 	    igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6129 		/* Make sure that anybody stopping the queue after this
6130 		 * sees the new next_to_clean.
6131 		 */
6132 		smp_mb();
6133 		if (__netif_subqueue_stopped(tx_ring->netdev,
6134 					     tx_ring->queue_index) &&
6135 		    !(test_bit(__IGB_DOWN, &adapter->state))) {
6136 			netif_wake_subqueue(tx_ring->netdev,
6137 					    tx_ring->queue_index);
6138 
6139 			u64_stats_update_begin(&tx_ring->tx_syncp);
6140 			tx_ring->tx_stats.restart_queue++;
6141 			u64_stats_update_end(&tx_ring->tx_syncp);
6142 		}
6143 	}
6144 
6145 	return !!budget;
6146 }
6147 
6148 /**
6149  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
6150  *  @rx_ring: rx descriptor ring to store buffers on
6151  *  @old_buff: donor buffer to have page reused
6152  *
6153  *  Synchronizes page for reuse by the adapter
6154  **/
6155 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6156 			      struct igb_rx_buffer *old_buff)
6157 {
6158 	struct igb_rx_buffer *new_buff;
6159 	u16 nta = rx_ring->next_to_alloc;
6160 
6161 	new_buff = &rx_ring->rx_buffer_info[nta];
6162 
6163 	/* update, and store next to alloc */
6164 	nta++;
6165 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6166 
6167 	/* transfer page from old buffer to new buffer */
6168 	memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6169 
6170 	/* sync the buffer for use by the device */
6171 	dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6172 					 old_buff->page_offset,
6173 					 IGB_RX_BUFSZ,
6174 					 DMA_FROM_DEVICE);
6175 }
6176 
6177 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6178 				  struct page *page,
6179 				  unsigned int truesize)
6180 {
6181 	/* avoid re-using remote pages */
6182 	if (unlikely(page_to_nid(page) != numa_node_id()))
6183 		return false;
6184 
6185 #if (PAGE_SIZE < 8192)
6186 	/* if we are only owner of page we can reuse it */
6187 	if (unlikely(page_count(page) != 1))
6188 		return false;
6189 
6190 	/* flip page offset to other buffer */
6191 	rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6192 
6193 	/* since we are the only owner of the page and we need to
6194 	 * increment it, just set the value to 2 in order to avoid
6195 	 * an unnecessary locked operation
6196 	 */
6197 	atomic_set(&page->_count, 2);
6198 #else
6199 	/* move offset up to the next cache line */
6200 	rx_buffer->page_offset += truesize;
6201 
6202 	if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6203 		return false;
6204 
6205 	/* bump ref count on page before it is given to the stack */
6206 	get_page(page);
6207 #endif
6208 
6209 	return true;
6210 }
6211 
6212 /**
6213  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6214  *  @rx_ring: rx descriptor ring to transact packets on
6215  *  @rx_buffer: buffer containing page to add
6216  *  @rx_desc: descriptor containing length of buffer written by hardware
6217  *  @skb: sk_buff to place the data into
6218  *
6219  *  This function will add the data contained in rx_buffer->page to the skb.
6220  *  This is done either through a direct copy if the data in the buffer is
6221  *  less than the skb header size, otherwise it will just attach the page as
6222  *  a frag to the skb.
6223  *
6224  *  The function will then update the page offset if necessary and return
6225  *  true if the buffer can be reused by the adapter.
6226  **/
6227 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6228 			    struct igb_rx_buffer *rx_buffer,
6229 			    union e1000_adv_rx_desc *rx_desc,
6230 			    struct sk_buff *skb)
6231 {
6232 	struct page *page = rx_buffer->page;
6233 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6234 #if (PAGE_SIZE < 8192)
6235 	unsigned int truesize = IGB_RX_BUFSZ;
6236 #else
6237 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6238 #endif
6239 
6240 	if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6241 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
6242 
6243 		if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6244 			igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6245 			va += IGB_TS_HDR_LEN;
6246 			size -= IGB_TS_HDR_LEN;
6247 		}
6248 
6249 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6250 
6251 		/* we can reuse buffer as-is, just make sure it is local */
6252 		if (likely(page_to_nid(page) == numa_node_id()))
6253 			return true;
6254 
6255 		/* this page cannot be reused so discard it */
6256 		put_page(page);
6257 		return false;
6258 	}
6259 
6260 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6261 			rx_buffer->page_offset, size, truesize);
6262 
6263 	return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6264 }
6265 
6266 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6267 					   union e1000_adv_rx_desc *rx_desc,
6268 					   struct sk_buff *skb)
6269 {
6270 	struct igb_rx_buffer *rx_buffer;
6271 	struct page *page;
6272 
6273 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6274 
6275 	page = rx_buffer->page;
6276 	prefetchw(page);
6277 
6278 	if (likely(!skb)) {
6279 		void *page_addr = page_address(page) +
6280 				  rx_buffer->page_offset;
6281 
6282 		/* prefetch first cache line of first page */
6283 		prefetch(page_addr);
6284 #if L1_CACHE_BYTES < 128
6285 		prefetch(page_addr + L1_CACHE_BYTES);
6286 #endif
6287 
6288 		/* allocate a skb to store the frags */
6289 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6290 						IGB_RX_HDR_LEN);
6291 		if (unlikely(!skb)) {
6292 			rx_ring->rx_stats.alloc_failed++;
6293 			return NULL;
6294 		}
6295 
6296 		/* we will be copying header into skb->data in
6297 		 * pskb_may_pull so it is in our interest to prefetch
6298 		 * it now to avoid a possible cache miss
6299 		 */
6300 		prefetchw(skb->data);
6301 	}
6302 
6303 	/* we are reusing so sync this buffer for CPU use */
6304 	dma_sync_single_range_for_cpu(rx_ring->dev,
6305 				      rx_buffer->dma,
6306 				      rx_buffer->page_offset,
6307 				      IGB_RX_BUFSZ,
6308 				      DMA_FROM_DEVICE);
6309 
6310 	/* pull page into skb */
6311 	if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6312 		/* hand second half of page back to the ring */
6313 		igb_reuse_rx_page(rx_ring, rx_buffer);
6314 	} else {
6315 		/* we are not reusing the buffer so unmap it */
6316 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6317 			       PAGE_SIZE, DMA_FROM_DEVICE);
6318 	}
6319 
6320 	/* clear contents of rx_buffer */
6321 	rx_buffer->page = NULL;
6322 
6323 	return skb;
6324 }
6325 
6326 static inline void igb_rx_checksum(struct igb_ring *ring,
6327 				   union e1000_adv_rx_desc *rx_desc,
6328 				   struct sk_buff *skb)
6329 {
6330 	skb_checksum_none_assert(skb);
6331 
6332 	/* Ignore Checksum bit is set */
6333 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6334 		return;
6335 
6336 	/* Rx checksum disabled via ethtool */
6337 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
6338 		return;
6339 
6340 	/* TCP/UDP checksum error bit is set */
6341 	if (igb_test_staterr(rx_desc,
6342 			     E1000_RXDEXT_STATERR_TCPE |
6343 			     E1000_RXDEXT_STATERR_IPE)) {
6344 		/* work around errata with sctp packets where the TCPE aka
6345 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6346 		 * packets, (aka let the stack check the crc32c)
6347 		 */
6348 		if (!((skb->len == 60) &&
6349 		      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6350 			u64_stats_update_begin(&ring->rx_syncp);
6351 			ring->rx_stats.csum_err++;
6352 			u64_stats_update_end(&ring->rx_syncp);
6353 		}
6354 		/* let the stack verify checksum errors */
6355 		return;
6356 	}
6357 	/* It must be a TCP or UDP packet with a valid checksum */
6358 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6359 				      E1000_RXD_STAT_UDPCS))
6360 		skb->ip_summed = CHECKSUM_UNNECESSARY;
6361 
6362 	dev_dbg(ring->dev, "cksum success: bits %08X\n",
6363 		le32_to_cpu(rx_desc->wb.upper.status_error));
6364 }
6365 
6366 static inline void igb_rx_hash(struct igb_ring *ring,
6367 			       union e1000_adv_rx_desc *rx_desc,
6368 			       struct sk_buff *skb)
6369 {
6370 	if (ring->netdev->features & NETIF_F_RXHASH)
6371 		skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6372 }
6373 
6374 /**
6375  *  igb_is_non_eop - process handling of non-EOP buffers
6376  *  @rx_ring: Rx ring being processed
6377  *  @rx_desc: Rx descriptor for current buffer
6378  *  @skb: current socket buffer containing buffer in progress
6379  *
6380  *  This function updates next to clean.  If the buffer is an EOP buffer
6381  *  this function exits returning false, otherwise it will place the
6382  *  sk_buff in the next buffer to be chained and return true indicating
6383  *  that this is in fact a non-EOP buffer.
6384  **/
6385 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6386 			   union e1000_adv_rx_desc *rx_desc)
6387 {
6388 	u32 ntc = rx_ring->next_to_clean + 1;
6389 
6390 	/* fetch, update, and store next to clean */
6391 	ntc = (ntc < rx_ring->count) ? ntc : 0;
6392 	rx_ring->next_to_clean = ntc;
6393 
6394 	prefetch(IGB_RX_DESC(rx_ring, ntc));
6395 
6396 	if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6397 		return false;
6398 
6399 	return true;
6400 }
6401 
6402 /**
6403  *  igb_get_headlen - determine size of header for LRO/GRO
6404  *  @data: pointer to the start of the headers
6405  *  @max_len: total length of section to find headers in
6406  *
6407  *  This function is meant to determine the length of headers that will
6408  *  be recognized by hardware for LRO, and GRO offloads.  The main
6409  *  motivation of doing this is to only perform one pull for IPv4 TCP
6410  *  packets so that we can do basic things like calculating the gso_size
6411  *  based on the average data per packet.
6412  **/
6413 static unsigned int igb_get_headlen(unsigned char *data,
6414 				    unsigned int max_len)
6415 {
6416 	union {
6417 		unsigned char *network;
6418 		/* l2 headers */
6419 		struct ethhdr *eth;
6420 		struct vlan_hdr *vlan;
6421 		/* l3 headers */
6422 		struct iphdr *ipv4;
6423 		struct ipv6hdr *ipv6;
6424 	} hdr;
6425 	__be16 protocol;
6426 	u8 nexthdr = 0;	/* default to not TCP */
6427 	u8 hlen;
6428 
6429 	/* this should never happen, but better safe than sorry */
6430 	if (max_len < ETH_HLEN)
6431 		return max_len;
6432 
6433 	/* initialize network frame pointer */
6434 	hdr.network = data;
6435 
6436 	/* set first protocol and move network header forward */
6437 	protocol = hdr.eth->h_proto;
6438 	hdr.network += ETH_HLEN;
6439 
6440 	/* handle any vlan tag if present */
6441 	if (protocol == __constant_htons(ETH_P_8021Q)) {
6442 		if ((hdr.network - data) > (max_len - VLAN_HLEN))
6443 			return max_len;
6444 
6445 		protocol = hdr.vlan->h_vlan_encapsulated_proto;
6446 		hdr.network += VLAN_HLEN;
6447 	}
6448 
6449 	/* handle L3 protocols */
6450 	if (protocol == __constant_htons(ETH_P_IP)) {
6451 		if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6452 			return max_len;
6453 
6454 		/* access ihl as a u8 to avoid unaligned access on ia64 */
6455 		hlen = (hdr.network[0] & 0x0F) << 2;
6456 
6457 		/* verify hlen meets minimum size requirements */
6458 		if (hlen < sizeof(struct iphdr))
6459 			return hdr.network - data;
6460 
6461 		/* record next protocol if header is present */
6462 		if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6463 			nexthdr = hdr.ipv4->protocol;
6464 	} else if (protocol == __constant_htons(ETH_P_IPV6)) {
6465 		if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6466 			return max_len;
6467 
6468 		/* record next protocol */
6469 		nexthdr = hdr.ipv6->nexthdr;
6470 		hlen = sizeof(struct ipv6hdr);
6471 	} else {
6472 		return hdr.network - data;
6473 	}
6474 
6475 	/* relocate pointer to start of L4 header */
6476 	hdr.network += hlen;
6477 
6478 	/* finally sort out TCP */
6479 	if (nexthdr == IPPROTO_TCP) {
6480 		if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6481 			return max_len;
6482 
6483 		/* access doff as a u8 to avoid unaligned access on ia64 */
6484 		hlen = (hdr.network[12] & 0xF0) >> 2;
6485 
6486 		/* verify hlen meets minimum size requirements */
6487 		if (hlen < sizeof(struct tcphdr))
6488 			return hdr.network - data;
6489 
6490 		hdr.network += hlen;
6491 	} else if (nexthdr == IPPROTO_UDP) {
6492 		if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6493 			return max_len;
6494 
6495 		hdr.network += sizeof(struct udphdr);
6496 	}
6497 
6498 	/* If everything has gone correctly hdr.network should be the
6499 	 * data section of the packet and will be the end of the header.
6500 	 * If not then it probably represents the end of the last recognized
6501 	 * header.
6502 	 */
6503 	if ((hdr.network - data) < max_len)
6504 		return hdr.network - data;
6505 	else
6506 		return max_len;
6507 }
6508 
6509 /**
6510  *  igb_pull_tail - igb specific version of skb_pull_tail
6511  *  @rx_ring: rx descriptor ring packet is being transacted on
6512  *  @rx_desc: pointer to the EOP Rx descriptor
6513  *  @skb: pointer to current skb being adjusted
6514  *
6515  *  This function is an igb specific version of __pskb_pull_tail.  The
6516  *  main difference between this version and the original function is that
6517  *  this function can make several assumptions about the state of things
6518  *  that allow for significant optimizations versus the standard function.
6519  *  As a result we can do things like drop a frag and maintain an accurate
6520  *  truesize for the skb.
6521  */
6522 static void igb_pull_tail(struct igb_ring *rx_ring,
6523 			  union e1000_adv_rx_desc *rx_desc,
6524 			  struct sk_buff *skb)
6525 {
6526 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6527 	unsigned char *va;
6528 	unsigned int pull_len;
6529 
6530 	/* it is valid to use page_address instead of kmap since we are
6531 	 * working with pages allocated out of the lomem pool per
6532 	 * alloc_page(GFP_ATOMIC)
6533 	 */
6534 	va = skb_frag_address(frag);
6535 
6536 	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6537 		/* retrieve timestamp from buffer */
6538 		igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6539 
6540 		/* update pointers to remove timestamp header */
6541 		skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6542 		frag->page_offset += IGB_TS_HDR_LEN;
6543 		skb->data_len -= IGB_TS_HDR_LEN;
6544 		skb->len -= IGB_TS_HDR_LEN;
6545 
6546 		/* move va to start of packet data */
6547 		va += IGB_TS_HDR_LEN;
6548 	}
6549 
6550 	/* we need the header to contain the greater of either ETH_HLEN or
6551 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
6552 	 */
6553 	pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6554 
6555 	/* align pull length to size of long to optimize memcpy performance */
6556 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6557 
6558 	/* update all of the pointers */
6559 	skb_frag_size_sub(frag, pull_len);
6560 	frag->page_offset += pull_len;
6561 	skb->data_len -= pull_len;
6562 	skb->tail += pull_len;
6563 }
6564 
6565 /**
6566  *  igb_cleanup_headers - Correct corrupted or empty headers
6567  *  @rx_ring: rx descriptor ring packet is being transacted on
6568  *  @rx_desc: pointer to the EOP Rx descriptor
6569  *  @skb: pointer to current skb being fixed
6570  *
6571  *  Address the case where we are pulling data in on pages only
6572  *  and as such no data is present in the skb header.
6573  *
6574  *  In addition if skb is not at least 60 bytes we need to pad it so that
6575  *  it is large enough to qualify as a valid Ethernet frame.
6576  *
6577  *  Returns true if an error was encountered and skb was freed.
6578  **/
6579 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6580 				union e1000_adv_rx_desc *rx_desc,
6581 				struct sk_buff *skb)
6582 {
6583 	if (unlikely((igb_test_staterr(rx_desc,
6584 				       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6585 		struct net_device *netdev = rx_ring->netdev;
6586 		if (!(netdev->features & NETIF_F_RXALL)) {
6587 			dev_kfree_skb_any(skb);
6588 			return true;
6589 		}
6590 	}
6591 
6592 	/* place header in linear portion of buffer */
6593 	if (skb_is_nonlinear(skb))
6594 		igb_pull_tail(rx_ring, rx_desc, skb);
6595 
6596 	/* if skb_pad returns an error the skb was freed */
6597 	if (unlikely(skb->len < 60)) {
6598 		int pad_len = 60 - skb->len;
6599 
6600 		if (skb_pad(skb, pad_len))
6601 			return true;
6602 		__skb_put(skb, pad_len);
6603 	}
6604 
6605 	return false;
6606 }
6607 
6608 /**
6609  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
6610  *  @rx_ring: rx descriptor ring packet is being transacted on
6611  *  @rx_desc: pointer to the EOP Rx descriptor
6612  *  @skb: pointer to current skb being populated
6613  *
6614  *  This function checks the ring, descriptor, and packet information in
6615  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
6616  *  other fields within the skb.
6617  **/
6618 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6619 				   union e1000_adv_rx_desc *rx_desc,
6620 				   struct sk_buff *skb)
6621 {
6622 	struct net_device *dev = rx_ring->netdev;
6623 
6624 	igb_rx_hash(rx_ring, rx_desc, skb);
6625 
6626 	igb_rx_checksum(rx_ring, rx_desc, skb);
6627 
6628 	igb_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
6629 
6630 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6631 	    igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6632 		u16 vid;
6633 		if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6634 		    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6635 			vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6636 		else
6637 			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6638 
6639 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6640 	}
6641 
6642 	skb_record_rx_queue(skb, rx_ring->queue_index);
6643 
6644 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6645 }
6646 
6647 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6648 {
6649 	struct igb_ring *rx_ring = q_vector->rx.ring;
6650 	struct sk_buff *skb = rx_ring->skb;
6651 	unsigned int total_bytes = 0, total_packets = 0;
6652 	u16 cleaned_count = igb_desc_unused(rx_ring);
6653 
6654 	do {
6655 		union e1000_adv_rx_desc *rx_desc;
6656 
6657 		/* return some buffers to hardware, one at a time is too slow */
6658 		if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6659 			igb_alloc_rx_buffers(rx_ring, cleaned_count);
6660 			cleaned_count = 0;
6661 		}
6662 
6663 		rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6664 
6665 		if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6666 			break;
6667 
6668 		/* This memory barrier is needed to keep us from reading
6669 		 * any other fields out of the rx_desc until we know the
6670 		 * RXD_STAT_DD bit is set
6671 		 */
6672 		rmb();
6673 
6674 		/* retrieve a buffer from the ring */
6675 		skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6676 
6677 		/* exit if we failed to retrieve a buffer */
6678 		if (!skb)
6679 			break;
6680 
6681 		cleaned_count++;
6682 
6683 		/* fetch next buffer in frame if non-eop */
6684 		if (igb_is_non_eop(rx_ring, rx_desc))
6685 			continue;
6686 
6687 		/* verify the packet layout is correct */
6688 		if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6689 			skb = NULL;
6690 			continue;
6691 		}
6692 
6693 		/* probably a little skewed due to removing CRC */
6694 		total_bytes += skb->len;
6695 
6696 		/* populate checksum, timestamp, VLAN, and protocol */
6697 		igb_process_skb_fields(rx_ring, rx_desc, skb);
6698 
6699 		napi_gro_receive(&q_vector->napi, skb);
6700 
6701 		/* reset skb pointer */
6702 		skb = NULL;
6703 
6704 		/* update budget accounting */
6705 		total_packets++;
6706 	} while (likely(total_packets < budget));
6707 
6708 	/* place incomplete frames back on ring for completion */
6709 	rx_ring->skb = skb;
6710 
6711 	u64_stats_update_begin(&rx_ring->rx_syncp);
6712 	rx_ring->rx_stats.packets += total_packets;
6713 	rx_ring->rx_stats.bytes += total_bytes;
6714 	u64_stats_update_end(&rx_ring->rx_syncp);
6715 	q_vector->rx.total_packets += total_packets;
6716 	q_vector->rx.total_bytes += total_bytes;
6717 
6718 	if (cleaned_count)
6719 		igb_alloc_rx_buffers(rx_ring, cleaned_count);
6720 
6721 	return (total_packets < budget);
6722 }
6723 
6724 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6725 				  struct igb_rx_buffer *bi)
6726 {
6727 	struct page *page = bi->page;
6728 	dma_addr_t dma;
6729 
6730 	/* since we are recycling buffers we should seldom need to alloc */
6731 	if (likely(page))
6732 		return true;
6733 
6734 	/* alloc new page for storage */
6735 	page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6736 	if (unlikely(!page)) {
6737 		rx_ring->rx_stats.alloc_failed++;
6738 		return false;
6739 	}
6740 
6741 	/* map page for use */
6742 	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6743 
6744 	/* if mapping failed free memory back to system since
6745 	 * there isn't much point in holding memory we can't use
6746 	 */
6747 	if (dma_mapping_error(rx_ring->dev, dma)) {
6748 		__free_page(page);
6749 
6750 		rx_ring->rx_stats.alloc_failed++;
6751 		return false;
6752 	}
6753 
6754 	bi->dma = dma;
6755 	bi->page = page;
6756 	bi->page_offset = 0;
6757 
6758 	return true;
6759 }
6760 
6761 /**
6762  *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
6763  *  @adapter: address of board private structure
6764  **/
6765 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6766 {
6767 	union e1000_adv_rx_desc *rx_desc;
6768 	struct igb_rx_buffer *bi;
6769 	u16 i = rx_ring->next_to_use;
6770 
6771 	/* nothing to do */
6772 	if (!cleaned_count)
6773 		return;
6774 
6775 	rx_desc = IGB_RX_DESC(rx_ring, i);
6776 	bi = &rx_ring->rx_buffer_info[i];
6777 	i -= rx_ring->count;
6778 
6779 	do {
6780 		if (!igb_alloc_mapped_page(rx_ring, bi))
6781 			break;
6782 
6783 		/* Refresh the desc even if buffer_addrs didn't change
6784 		 * because each write-back erases this info.
6785 		 */
6786 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
6787 
6788 		rx_desc++;
6789 		bi++;
6790 		i++;
6791 		if (unlikely(!i)) {
6792 			rx_desc = IGB_RX_DESC(rx_ring, 0);
6793 			bi = rx_ring->rx_buffer_info;
6794 			i -= rx_ring->count;
6795 		}
6796 
6797 		/* clear the hdr_addr for the next_to_use descriptor */
6798 		rx_desc->read.hdr_addr = 0;
6799 
6800 		cleaned_count--;
6801 	} while (cleaned_count);
6802 
6803 	i += rx_ring->count;
6804 
6805 	if (rx_ring->next_to_use != i) {
6806 		/* record the next descriptor to use */
6807 		rx_ring->next_to_use = i;
6808 
6809 		/* update next to alloc since we have filled the ring */
6810 		rx_ring->next_to_alloc = i;
6811 
6812 		/* Force memory writes to complete before letting h/w
6813 		 * know there are new descriptors to fetch.  (Only
6814 		 * applicable for weak-ordered memory model archs,
6815 		 * such as IA-64).
6816 		 */
6817 		wmb();
6818 		writel(i, rx_ring->tail);
6819 	}
6820 }
6821 
6822 /**
6823  * igb_mii_ioctl -
6824  * @netdev:
6825  * @ifreq:
6826  * @cmd:
6827  **/
6828 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6829 {
6830 	struct igb_adapter *adapter = netdev_priv(netdev);
6831 	struct mii_ioctl_data *data = if_mii(ifr);
6832 
6833 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
6834 		return -EOPNOTSUPP;
6835 
6836 	switch (cmd) {
6837 	case SIOCGMIIPHY:
6838 		data->phy_id = adapter->hw.phy.addr;
6839 		break;
6840 	case SIOCGMIIREG:
6841 		if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6842 		                     &data->val_out))
6843 			return -EIO;
6844 		break;
6845 	case SIOCSMIIREG:
6846 	default:
6847 		return -EOPNOTSUPP;
6848 	}
6849 	return 0;
6850 }
6851 
6852 /**
6853  * igb_ioctl -
6854  * @netdev:
6855  * @ifreq:
6856  * @cmd:
6857  **/
6858 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6859 {
6860 	switch (cmd) {
6861 	case SIOCGMIIPHY:
6862 	case SIOCGMIIREG:
6863 	case SIOCSMIIREG:
6864 		return igb_mii_ioctl(netdev, ifr, cmd);
6865 	case SIOCSHWTSTAMP:
6866 		return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6867 	default:
6868 		return -EOPNOTSUPP;
6869 	}
6870 }
6871 
6872 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6873 {
6874 	struct igb_adapter *adapter = hw->back;
6875 
6876 	if (pcie_capability_read_word(adapter->pdev, reg, value))
6877 		return -E1000_ERR_CONFIG;
6878 
6879 	return 0;
6880 }
6881 
6882 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6883 {
6884 	struct igb_adapter *adapter = hw->back;
6885 
6886 	if (pcie_capability_write_word(adapter->pdev, reg, *value))
6887 		return -E1000_ERR_CONFIG;
6888 
6889 	return 0;
6890 }
6891 
6892 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6893 {
6894 	struct igb_adapter *adapter = netdev_priv(netdev);
6895 	struct e1000_hw *hw = &adapter->hw;
6896 	u32 ctrl, rctl;
6897 	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
6898 
6899 	if (enable) {
6900 		/* enable VLAN tag insert/strip */
6901 		ctrl = rd32(E1000_CTRL);
6902 		ctrl |= E1000_CTRL_VME;
6903 		wr32(E1000_CTRL, ctrl);
6904 
6905 		/* Disable CFI check */
6906 		rctl = rd32(E1000_RCTL);
6907 		rctl &= ~E1000_RCTL_CFIEN;
6908 		wr32(E1000_RCTL, rctl);
6909 	} else {
6910 		/* disable VLAN tag insert/strip */
6911 		ctrl = rd32(E1000_CTRL);
6912 		ctrl &= ~E1000_CTRL_VME;
6913 		wr32(E1000_CTRL, ctrl);
6914 	}
6915 
6916 	igb_rlpml_set(adapter);
6917 }
6918 
6919 static int igb_vlan_rx_add_vid(struct net_device *netdev,
6920 			       __be16 proto, u16 vid)
6921 {
6922 	struct igb_adapter *adapter = netdev_priv(netdev);
6923 	struct e1000_hw *hw = &adapter->hw;
6924 	int pf_id = adapter->vfs_allocated_count;
6925 
6926 	/* attempt to add filter to vlvf array */
6927 	igb_vlvf_set(adapter, vid, true, pf_id);
6928 
6929 	/* add the filter since PF can receive vlans w/o entry in vlvf */
6930 	igb_vfta_set(hw, vid, true);
6931 
6932 	set_bit(vid, adapter->active_vlans);
6933 
6934 	return 0;
6935 }
6936 
6937 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
6938 				__be16 proto, u16 vid)
6939 {
6940 	struct igb_adapter *adapter = netdev_priv(netdev);
6941 	struct e1000_hw *hw = &adapter->hw;
6942 	int pf_id = adapter->vfs_allocated_count;
6943 	s32 err;
6944 
6945 	/* remove vlan from VLVF table array */
6946 	err = igb_vlvf_set(adapter, vid, false, pf_id);
6947 
6948 	/* if vid was not present in VLVF just remove it from table */
6949 	if (err)
6950 		igb_vfta_set(hw, vid, false);
6951 
6952 	clear_bit(vid, adapter->active_vlans);
6953 
6954 	return 0;
6955 }
6956 
6957 static void igb_restore_vlan(struct igb_adapter *adapter)
6958 {
6959 	u16 vid;
6960 
6961 	igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6962 
6963 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6964 		igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
6965 }
6966 
6967 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6968 {
6969 	struct pci_dev *pdev = adapter->pdev;
6970 	struct e1000_mac_info *mac = &adapter->hw.mac;
6971 
6972 	mac->autoneg = 0;
6973 
6974 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
6975 	 * for the switch() below to work
6976 	 */
6977 	if ((spd & 1) || (dplx & ~1))
6978 		goto err_inval;
6979 
6980 	/* Fiber NIC's only allow 1000 gbps Full duplex
6981 	 * and 100Mbps Full duplex for 100baseFx sfp
6982 	 */
6983 	if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
6984 		switch (spd + dplx) {
6985 		case SPEED_10 + DUPLEX_HALF:
6986 		case SPEED_10 + DUPLEX_FULL:
6987 		case SPEED_100 + DUPLEX_HALF:
6988 			goto err_inval;
6989 		default:
6990 			break;
6991 		}
6992 	}
6993 
6994 	switch (spd + dplx) {
6995 	case SPEED_10 + DUPLEX_HALF:
6996 		mac->forced_speed_duplex = ADVERTISE_10_HALF;
6997 		break;
6998 	case SPEED_10 + DUPLEX_FULL:
6999 		mac->forced_speed_duplex = ADVERTISE_10_FULL;
7000 		break;
7001 	case SPEED_100 + DUPLEX_HALF:
7002 		mac->forced_speed_duplex = ADVERTISE_100_HALF;
7003 		break;
7004 	case SPEED_100 + DUPLEX_FULL:
7005 		mac->forced_speed_duplex = ADVERTISE_100_FULL;
7006 		break;
7007 	case SPEED_1000 + DUPLEX_FULL:
7008 		mac->autoneg = 1;
7009 		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7010 		break;
7011 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
7012 	default:
7013 		goto err_inval;
7014 	}
7015 
7016 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7017 	adapter->hw.phy.mdix = AUTO_ALL_MODES;
7018 
7019 	return 0;
7020 
7021 err_inval:
7022 	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7023 	return -EINVAL;
7024 }
7025 
7026 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7027 			  bool runtime)
7028 {
7029 	struct net_device *netdev = pci_get_drvdata(pdev);
7030 	struct igb_adapter *adapter = netdev_priv(netdev);
7031 	struct e1000_hw *hw = &adapter->hw;
7032 	u32 ctrl, rctl, status;
7033 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7034 #ifdef CONFIG_PM
7035 	int retval = 0;
7036 #endif
7037 
7038 	netif_device_detach(netdev);
7039 
7040 	if (netif_running(netdev))
7041 		__igb_close(netdev, true);
7042 
7043 	igb_clear_interrupt_scheme(adapter);
7044 
7045 #ifdef CONFIG_PM
7046 	retval = pci_save_state(pdev);
7047 	if (retval)
7048 		return retval;
7049 #endif
7050 
7051 	status = rd32(E1000_STATUS);
7052 	if (status & E1000_STATUS_LU)
7053 		wufc &= ~E1000_WUFC_LNKC;
7054 
7055 	if (wufc) {
7056 		igb_setup_rctl(adapter);
7057 		igb_set_rx_mode(netdev);
7058 
7059 		/* turn on all-multi mode if wake on multicast is enabled */
7060 		if (wufc & E1000_WUFC_MC) {
7061 			rctl = rd32(E1000_RCTL);
7062 			rctl |= E1000_RCTL_MPE;
7063 			wr32(E1000_RCTL, rctl);
7064 		}
7065 
7066 		ctrl = rd32(E1000_CTRL);
7067 		/* advertise wake from D3Cold */
7068 		#define E1000_CTRL_ADVD3WUC 0x00100000
7069 		/* phy power management enable */
7070 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7071 		ctrl |= E1000_CTRL_ADVD3WUC;
7072 		wr32(E1000_CTRL, ctrl);
7073 
7074 		/* Allow time for pending master requests to run */
7075 		igb_disable_pcie_master(hw);
7076 
7077 		wr32(E1000_WUC, E1000_WUC_PME_EN);
7078 		wr32(E1000_WUFC, wufc);
7079 	} else {
7080 		wr32(E1000_WUC, 0);
7081 		wr32(E1000_WUFC, 0);
7082 	}
7083 
7084 	*enable_wake = wufc || adapter->en_mng_pt;
7085 	if (!*enable_wake)
7086 		igb_power_down_link(adapter);
7087 	else
7088 		igb_power_up_link(adapter);
7089 
7090 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7091 	 * would have already happened in close and is redundant.
7092 	 */
7093 	igb_release_hw_control(adapter);
7094 
7095 	pci_disable_device(pdev);
7096 
7097 	return 0;
7098 }
7099 
7100 #ifdef CONFIG_PM
7101 #ifdef CONFIG_PM_SLEEP
7102 static int igb_suspend(struct device *dev)
7103 {
7104 	int retval;
7105 	bool wake;
7106 	struct pci_dev *pdev = to_pci_dev(dev);
7107 
7108 	retval = __igb_shutdown(pdev, &wake, 0);
7109 	if (retval)
7110 		return retval;
7111 
7112 	if (wake) {
7113 		pci_prepare_to_sleep(pdev);
7114 	} else {
7115 		pci_wake_from_d3(pdev, false);
7116 		pci_set_power_state(pdev, PCI_D3hot);
7117 	}
7118 
7119 	return 0;
7120 }
7121 #endif /* CONFIG_PM_SLEEP */
7122 
7123 static int igb_resume(struct device *dev)
7124 {
7125 	struct pci_dev *pdev = to_pci_dev(dev);
7126 	struct net_device *netdev = pci_get_drvdata(pdev);
7127 	struct igb_adapter *adapter = netdev_priv(netdev);
7128 	struct e1000_hw *hw = &adapter->hw;
7129 	u32 err;
7130 
7131 	pci_set_power_state(pdev, PCI_D0);
7132 	pci_restore_state(pdev);
7133 	pci_save_state(pdev);
7134 
7135 	err = pci_enable_device_mem(pdev);
7136 	if (err) {
7137 		dev_err(&pdev->dev,
7138 			"igb: Cannot enable PCI device from suspend\n");
7139 		return err;
7140 	}
7141 	pci_set_master(pdev);
7142 
7143 	pci_enable_wake(pdev, PCI_D3hot, 0);
7144 	pci_enable_wake(pdev, PCI_D3cold, 0);
7145 
7146 	if (igb_init_interrupt_scheme(adapter, true)) {
7147 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7148 		return -ENOMEM;
7149 	}
7150 
7151 	igb_reset(adapter);
7152 
7153 	/* let the f/w know that the h/w is now under the control of the
7154 	 * driver.
7155 	 */
7156 	igb_get_hw_control(adapter);
7157 
7158 	wr32(E1000_WUS, ~0);
7159 
7160 	if (netdev->flags & IFF_UP) {
7161 		rtnl_lock();
7162 		err = __igb_open(netdev, true);
7163 		rtnl_unlock();
7164 		if (err)
7165 			return err;
7166 	}
7167 
7168 	netif_device_attach(netdev);
7169 	return 0;
7170 }
7171 
7172 #ifdef CONFIG_PM_RUNTIME
7173 static int igb_runtime_idle(struct device *dev)
7174 {
7175 	struct pci_dev *pdev = to_pci_dev(dev);
7176 	struct net_device *netdev = pci_get_drvdata(pdev);
7177 	struct igb_adapter *adapter = netdev_priv(netdev);
7178 
7179 	if (!igb_has_link(adapter))
7180 		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7181 
7182 	return -EBUSY;
7183 }
7184 
7185 static int igb_runtime_suspend(struct device *dev)
7186 {
7187 	struct pci_dev *pdev = to_pci_dev(dev);
7188 	int retval;
7189 	bool wake;
7190 
7191 	retval = __igb_shutdown(pdev, &wake, 1);
7192 	if (retval)
7193 		return retval;
7194 
7195 	if (wake) {
7196 		pci_prepare_to_sleep(pdev);
7197 	} else {
7198 		pci_wake_from_d3(pdev, false);
7199 		pci_set_power_state(pdev, PCI_D3hot);
7200 	}
7201 
7202 	return 0;
7203 }
7204 
7205 static int igb_runtime_resume(struct device *dev)
7206 {
7207 	return igb_resume(dev);
7208 }
7209 #endif /* CONFIG_PM_RUNTIME */
7210 #endif
7211 
7212 static void igb_shutdown(struct pci_dev *pdev)
7213 {
7214 	bool wake;
7215 
7216 	__igb_shutdown(pdev, &wake, 0);
7217 
7218 	if (system_state == SYSTEM_POWER_OFF) {
7219 		pci_wake_from_d3(pdev, wake);
7220 		pci_set_power_state(pdev, PCI_D3hot);
7221 	}
7222 }
7223 
7224 #ifdef CONFIG_PCI_IOV
7225 static int igb_sriov_reinit(struct pci_dev *dev)
7226 {
7227 	struct net_device *netdev = pci_get_drvdata(dev);
7228 	struct igb_adapter *adapter = netdev_priv(netdev);
7229 	struct pci_dev *pdev = adapter->pdev;
7230 
7231 	rtnl_lock();
7232 
7233 	if (netif_running(netdev))
7234 		igb_close(netdev);
7235 
7236 	igb_clear_interrupt_scheme(adapter);
7237 
7238 	igb_init_queue_configuration(adapter);
7239 
7240 	if (igb_init_interrupt_scheme(adapter, true)) {
7241 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7242 		return -ENOMEM;
7243 	}
7244 
7245 	if (netif_running(netdev))
7246 		igb_open(netdev);
7247 
7248 	rtnl_unlock();
7249 
7250 	return 0;
7251 }
7252 
7253 static int igb_pci_disable_sriov(struct pci_dev *dev)
7254 {
7255 	int err = igb_disable_sriov(dev);
7256 
7257 	if (!err)
7258 		err = igb_sriov_reinit(dev);
7259 
7260 	return err;
7261 }
7262 
7263 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7264 {
7265 	int err = igb_enable_sriov(dev, num_vfs);
7266 
7267 	if (err)
7268 		goto out;
7269 
7270 	err = igb_sriov_reinit(dev);
7271 	if (!err)
7272 		return num_vfs;
7273 
7274 out:
7275 	return err;
7276 }
7277 
7278 #endif
7279 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7280 {
7281 #ifdef CONFIG_PCI_IOV
7282 	if (num_vfs == 0)
7283 		return igb_pci_disable_sriov(dev);
7284 	else
7285 		return igb_pci_enable_sriov(dev, num_vfs);
7286 #endif
7287 	return 0;
7288 }
7289 
7290 #ifdef CONFIG_NET_POLL_CONTROLLER
7291 /* Polling 'interrupt' - used by things like netconsole to send skbs
7292  * without having to re-enable interrupts. It's not called while
7293  * the interrupt routine is executing.
7294  */
7295 static void igb_netpoll(struct net_device *netdev)
7296 {
7297 	struct igb_adapter *adapter = netdev_priv(netdev);
7298 	struct e1000_hw *hw = &adapter->hw;
7299 	struct igb_q_vector *q_vector;
7300 	int i;
7301 
7302 	for (i = 0; i < adapter->num_q_vectors; i++) {
7303 		q_vector = adapter->q_vector[i];
7304 		if (adapter->msix_entries)
7305 			wr32(E1000_EIMC, q_vector->eims_value);
7306 		else
7307 			igb_irq_disable(adapter);
7308 		napi_schedule(&q_vector->napi);
7309 	}
7310 }
7311 #endif /* CONFIG_NET_POLL_CONTROLLER */
7312 
7313 /**
7314  *  igb_io_error_detected - called when PCI error is detected
7315  *  @pdev: Pointer to PCI device
7316  *  @state: The current pci connection state
7317  *
7318  *  This function is called after a PCI bus error affecting
7319  *  this device has been detected.
7320  **/
7321 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7322 					      pci_channel_state_t state)
7323 {
7324 	struct net_device *netdev = pci_get_drvdata(pdev);
7325 	struct igb_adapter *adapter = netdev_priv(netdev);
7326 
7327 	netif_device_detach(netdev);
7328 
7329 	if (state == pci_channel_io_perm_failure)
7330 		return PCI_ERS_RESULT_DISCONNECT;
7331 
7332 	if (netif_running(netdev))
7333 		igb_down(adapter);
7334 	pci_disable_device(pdev);
7335 
7336 	/* Request a slot slot reset. */
7337 	return PCI_ERS_RESULT_NEED_RESET;
7338 }
7339 
7340 /**
7341  *  igb_io_slot_reset - called after the pci bus has been reset.
7342  *  @pdev: Pointer to PCI device
7343  *
7344  *  Restart the card from scratch, as if from a cold-boot. Implementation
7345  *  resembles the first-half of the igb_resume routine.
7346  **/
7347 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7348 {
7349 	struct net_device *netdev = pci_get_drvdata(pdev);
7350 	struct igb_adapter *adapter = netdev_priv(netdev);
7351 	struct e1000_hw *hw = &adapter->hw;
7352 	pci_ers_result_t result;
7353 	int err;
7354 
7355 	if (pci_enable_device_mem(pdev)) {
7356 		dev_err(&pdev->dev,
7357 			"Cannot re-enable PCI device after reset.\n");
7358 		result = PCI_ERS_RESULT_DISCONNECT;
7359 	} else {
7360 		pci_set_master(pdev);
7361 		pci_restore_state(pdev);
7362 		pci_save_state(pdev);
7363 
7364 		pci_enable_wake(pdev, PCI_D3hot, 0);
7365 		pci_enable_wake(pdev, PCI_D3cold, 0);
7366 
7367 		igb_reset(adapter);
7368 		wr32(E1000_WUS, ~0);
7369 		result = PCI_ERS_RESULT_RECOVERED;
7370 	}
7371 
7372 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
7373 	if (err) {
7374 		dev_err(&pdev->dev,
7375 			"pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7376 			err);
7377 		/* non-fatal, continue */
7378 	}
7379 
7380 	return result;
7381 }
7382 
7383 /**
7384  *  igb_io_resume - called when traffic can start flowing again.
7385  *  @pdev: Pointer to PCI device
7386  *
7387  *  This callback is called when the error recovery driver tells us that
7388  *  its OK to resume normal operation. Implementation resembles the
7389  *  second-half of the igb_resume routine.
7390  */
7391 static void igb_io_resume(struct pci_dev *pdev)
7392 {
7393 	struct net_device *netdev = pci_get_drvdata(pdev);
7394 	struct igb_adapter *adapter = netdev_priv(netdev);
7395 
7396 	if (netif_running(netdev)) {
7397 		if (igb_up(adapter)) {
7398 			dev_err(&pdev->dev, "igb_up failed after reset\n");
7399 			return;
7400 		}
7401 	}
7402 
7403 	netif_device_attach(netdev);
7404 
7405 	/* let the f/w know that the h/w is now under the control of the
7406 	 * driver.
7407 	 */
7408 	igb_get_hw_control(adapter);
7409 }
7410 
7411 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7412 			     u8 qsel)
7413 {
7414 	u32 rar_low, rar_high;
7415 	struct e1000_hw *hw = &adapter->hw;
7416 
7417 	/* HW expects these in little endian so we reverse the byte order
7418 	 * from network order (big endian) to little endian
7419 	 */
7420 	rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7421 		   ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7422 	rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7423 
7424 	/* Indicate to hardware the Address is Valid. */
7425 	rar_high |= E1000_RAH_AV;
7426 
7427 	if (hw->mac.type == e1000_82575)
7428 		rar_high |= E1000_RAH_POOL_1 * qsel;
7429 	else
7430 		rar_high |= E1000_RAH_POOL_1 << qsel;
7431 
7432 	wr32(E1000_RAL(index), rar_low);
7433 	wrfl();
7434 	wr32(E1000_RAH(index), rar_high);
7435 	wrfl();
7436 }
7437 
7438 static int igb_set_vf_mac(struct igb_adapter *adapter,
7439 			  int vf, unsigned char *mac_addr)
7440 {
7441 	struct e1000_hw *hw = &adapter->hw;
7442 	/* VF MAC addresses start at end of receive addresses and moves
7443 	 * towards the first, as a result a collision should not be possible
7444 	 */
7445 	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7446 
7447 	memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7448 
7449 	igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7450 
7451 	return 0;
7452 }
7453 
7454 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7455 {
7456 	struct igb_adapter *adapter = netdev_priv(netdev);
7457 	if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7458 		return -EINVAL;
7459 	adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7460 	dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7461 	dev_info(&adapter->pdev->dev,
7462 		 "Reload the VF driver to make this change effective.");
7463 	if (test_bit(__IGB_DOWN, &adapter->state)) {
7464 		dev_warn(&adapter->pdev->dev,
7465 			 "The VF MAC address has been set, but the PF device is not up.\n");
7466 		dev_warn(&adapter->pdev->dev,
7467 			 "Bring the PF device up before attempting to use the VF device.\n");
7468 	}
7469 	return igb_set_vf_mac(adapter, vf, mac);
7470 }
7471 
7472 static int igb_link_mbps(int internal_link_speed)
7473 {
7474 	switch (internal_link_speed) {
7475 	case SPEED_100:
7476 		return 100;
7477 	case SPEED_1000:
7478 		return 1000;
7479 	default:
7480 		return 0;
7481 	}
7482 }
7483 
7484 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7485 				  int link_speed)
7486 {
7487 	int rf_dec, rf_int;
7488 	u32 bcnrc_val;
7489 
7490 	if (tx_rate != 0) {
7491 		/* Calculate the rate factor values to set */
7492 		rf_int = link_speed / tx_rate;
7493 		rf_dec = (link_speed - (rf_int * tx_rate));
7494 		rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7495 			 tx_rate;
7496 
7497 		bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7498 		bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7499 			      E1000_RTTBCNRC_RF_INT_MASK);
7500 		bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7501 	} else {
7502 		bcnrc_val = 0;
7503 	}
7504 
7505 	wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7506 	/* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7507 	 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7508 	 */
7509 	wr32(E1000_RTTBCNRM, 0x14);
7510 	wr32(E1000_RTTBCNRC, bcnrc_val);
7511 }
7512 
7513 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7514 {
7515 	int actual_link_speed, i;
7516 	bool reset_rate = false;
7517 
7518 	/* VF TX rate limit was not set or not supported */
7519 	if ((adapter->vf_rate_link_speed == 0) ||
7520 	    (adapter->hw.mac.type != e1000_82576))
7521 		return;
7522 
7523 	actual_link_speed = igb_link_mbps(adapter->link_speed);
7524 	if (actual_link_speed != adapter->vf_rate_link_speed) {
7525 		reset_rate = true;
7526 		adapter->vf_rate_link_speed = 0;
7527 		dev_info(&adapter->pdev->dev,
7528 			 "Link speed has been changed. VF Transmit rate is disabled\n");
7529 	}
7530 
7531 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
7532 		if (reset_rate)
7533 			adapter->vf_data[i].tx_rate = 0;
7534 
7535 		igb_set_vf_rate_limit(&adapter->hw, i,
7536 				      adapter->vf_data[i].tx_rate,
7537 				      actual_link_speed);
7538 	}
7539 }
7540 
7541 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7542 {
7543 	struct igb_adapter *adapter = netdev_priv(netdev);
7544 	struct e1000_hw *hw = &adapter->hw;
7545 	int actual_link_speed;
7546 
7547 	if (hw->mac.type != e1000_82576)
7548 		return -EOPNOTSUPP;
7549 
7550 	actual_link_speed = igb_link_mbps(adapter->link_speed);
7551 	if ((vf >= adapter->vfs_allocated_count) ||
7552 	    (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7553 	    (tx_rate < 0) || (tx_rate > actual_link_speed))
7554 		return -EINVAL;
7555 
7556 	adapter->vf_rate_link_speed = actual_link_speed;
7557 	adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7558 	igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7559 
7560 	return 0;
7561 }
7562 
7563 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7564 				   bool setting)
7565 {
7566 	struct igb_adapter *adapter = netdev_priv(netdev);
7567 	struct e1000_hw *hw = &adapter->hw;
7568 	u32 reg_val, reg_offset;
7569 
7570 	if (!adapter->vfs_allocated_count)
7571 		return -EOPNOTSUPP;
7572 
7573 	if (vf >= adapter->vfs_allocated_count)
7574 		return -EINVAL;
7575 
7576 	reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7577 	reg_val = rd32(reg_offset);
7578 	if (setting)
7579 		reg_val |= ((1 << vf) |
7580 			    (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7581 	else
7582 		reg_val &= ~((1 << vf) |
7583 			     (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7584 	wr32(reg_offset, reg_val);
7585 
7586 	adapter->vf_data[vf].spoofchk_enabled = setting;
7587 	return E1000_SUCCESS;
7588 }
7589 
7590 static int igb_ndo_get_vf_config(struct net_device *netdev,
7591 				 int vf, struct ifla_vf_info *ivi)
7592 {
7593 	struct igb_adapter *adapter = netdev_priv(netdev);
7594 	if (vf >= adapter->vfs_allocated_count)
7595 		return -EINVAL;
7596 	ivi->vf = vf;
7597 	memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7598 	ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7599 	ivi->vlan = adapter->vf_data[vf].pf_vlan;
7600 	ivi->qos = adapter->vf_data[vf].pf_qos;
7601 	ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7602 	return 0;
7603 }
7604 
7605 static void igb_vmm_control(struct igb_adapter *adapter)
7606 {
7607 	struct e1000_hw *hw = &adapter->hw;
7608 	u32 reg;
7609 
7610 	switch (hw->mac.type) {
7611 	case e1000_82575:
7612 	case e1000_i210:
7613 	case e1000_i211:
7614 	case e1000_i354:
7615 	default:
7616 		/* replication is not supported for 82575 */
7617 		return;
7618 	case e1000_82576:
7619 		/* notify HW that the MAC is adding vlan tags */
7620 		reg = rd32(E1000_DTXCTL);
7621 		reg |= E1000_DTXCTL_VLAN_ADDED;
7622 		wr32(E1000_DTXCTL, reg);
7623 	case e1000_82580:
7624 		/* enable replication vlan tag stripping */
7625 		reg = rd32(E1000_RPLOLR);
7626 		reg |= E1000_RPLOLR_STRVLAN;
7627 		wr32(E1000_RPLOLR, reg);
7628 	case e1000_i350:
7629 		/* none of the above registers are supported by i350 */
7630 		break;
7631 	}
7632 
7633 	if (adapter->vfs_allocated_count) {
7634 		igb_vmdq_set_loopback_pf(hw, true);
7635 		igb_vmdq_set_replication_pf(hw, true);
7636 		igb_vmdq_set_anti_spoofing_pf(hw, true,
7637 					      adapter->vfs_allocated_count);
7638 	} else {
7639 		igb_vmdq_set_loopback_pf(hw, false);
7640 		igb_vmdq_set_replication_pf(hw, false);
7641 	}
7642 }
7643 
7644 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7645 {
7646 	struct e1000_hw *hw = &adapter->hw;
7647 	u32 dmac_thr;
7648 	u16 hwm;
7649 
7650 	if (hw->mac.type > e1000_82580) {
7651 		if (adapter->flags & IGB_FLAG_DMAC) {
7652 			u32 reg;
7653 
7654 			/* force threshold to 0. */
7655 			wr32(E1000_DMCTXTH, 0);
7656 
7657 			/* DMA Coalescing high water mark needs to be greater
7658 			 * than the Rx threshold. Set hwm to PBA - max frame
7659 			 * size in 16B units, capping it at PBA - 6KB.
7660 			 */
7661 			hwm = 64 * pba - adapter->max_frame_size / 16;
7662 			if (hwm < 64 * (pba - 6))
7663 				hwm = 64 * (pba - 6);
7664 			reg = rd32(E1000_FCRTC);
7665 			reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7666 			reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7667 				& E1000_FCRTC_RTH_COAL_MASK);
7668 			wr32(E1000_FCRTC, reg);
7669 
7670 			/* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7671 			 * frame size, capping it at PBA - 10KB.
7672 			 */
7673 			dmac_thr = pba - adapter->max_frame_size / 512;
7674 			if (dmac_thr < pba - 10)
7675 				dmac_thr = pba - 10;
7676 			reg = rd32(E1000_DMACR);
7677 			reg &= ~E1000_DMACR_DMACTHR_MASK;
7678 			reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7679 				& E1000_DMACR_DMACTHR_MASK);
7680 
7681 			/* transition to L0x or L1 if available..*/
7682 			reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7683 
7684 			/* watchdog timer= +-1000 usec in 32usec intervals */
7685 			reg |= (1000 >> 5);
7686 
7687 			/* Disable BMC-to-OS Watchdog Enable */
7688 			if (hw->mac.type != e1000_i354)
7689 				reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7690 
7691 			wr32(E1000_DMACR, reg);
7692 
7693 			/* no lower threshold to disable
7694 			 * coalescing(smart fifb)-UTRESH=0
7695 			 */
7696 			wr32(E1000_DMCRTRH, 0);
7697 
7698 			reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7699 
7700 			wr32(E1000_DMCTLX, reg);
7701 
7702 			/* free space in tx packet buffer to wake from
7703 			 * DMA coal
7704 			 */
7705 			wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7706 			     (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7707 
7708 			/* make low power state decision controlled
7709 			 * by DMA coal
7710 			 */
7711 			reg = rd32(E1000_PCIEMISC);
7712 			reg &= ~E1000_PCIEMISC_LX_DECISION;
7713 			wr32(E1000_PCIEMISC, reg);
7714 		} /* endif adapter->dmac is not disabled */
7715 	} else if (hw->mac.type == e1000_82580) {
7716 		u32 reg = rd32(E1000_PCIEMISC);
7717 		wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7718 		wr32(E1000_DMACR, 0);
7719 	}
7720 }
7721 
7722 /**
7723  *  igb_read_i2c_byte - Reads 8 bit word over I2C
7724  *  @hw: pointer to hardware structure
7725  *  @byte_offset: byte offset to read
7726  *  @dev_addr: device address
7727  *  @data: value read
7728  *
7729  *  Performs byte read operation over I2C interface at
7730  *  a specified device address.
7731  **/
7732 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7733 		      u8 dev_addr, u8 *data)
7734 {
7735 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7736 	struct i2c_client *this_client = adapter->i2c_client;
7737 	s32 status;
7738 	u16 swfw_mask = 0;
7739 
7740 	if (!this_client)
7741 		return E1000_ERR_I2C;
7742 
7743 	swfw_mask = E1000_SWFW_PHY0_SM;
7744 
7745 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7746 	    != E1000_SUCCESS)
7747 		return E1000_ERR_SWFW_SYNC;
7748 
7749 	status = i2c_smbus_read_byte_data(this_client, byte_offset);
7750 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7751 
7752 	if (status < 0)
7753 		return E1000_ERR_I2C;
7754 	else {
7755 		*data = status;
7756 		return E1000_SUCCESS;
7757 	}
7758 }
7759 
7760 /**
7761  *  igb_write_i2c_byte - Writes 8 bit word over I2C
7762  *  @hw: pointer to hardware structure
7763  *  @byte_offset: byte offset to write
7764  *  @dev_addr: device address
7765  *  @data: value to write
7766  *
7767  *  Performs byte write operation over I2C interface at
7768  *  a specified device address.
7769  **/
7770 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7771 		       u8 dev_addr, u8 data)
7772 {
7773 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7774 	struct i2c_client *this_client = adapter->i2c_client;
7775 	s32 status;
7776 	u16 swfw_mask = E1000_SWFW_PHY0_SM;
7777 
7778 	if (!this_client)
7779 		return E1000_ERR_I2C;
7780 
7781 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
7782 		return E1000_ERR_SWFW_SYNC;
7783 	status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
7784 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7785 
7786 	if (status)
7787 		return E1000_ERR_I2C;
7788 	else
7789 		return E1000_SUCCESS;
7790 
7791 }
7792 /* igb_main.c */
7793