xref: /openbmc/linux/drivers/net/ethernet/intel/igb/igb_main.c (revision 10c1d542c7e871865bca381842fd04a92d2b95ec)
1 /* Intel(R) Gigabit Ethernet Linux driver
2  * Copyright(c) 2007-2014 Intel Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, see <http://www.gnu.org/licenses/>.
15  *
16  * The full GNU General Public License is included in this distribution in
17  * the file called "COPYING".
18  *
19  * Contact Information:
20  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
21  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
22  */
23 
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/init.h>
29 #include <linux/bitops.h>
30 #include <linux/vmalloc.h>
31 #include <linux/pagemap.h>
32 #include <linux/netdevice.h>
33 #include <linux/ipv6.h>
34 #include <linux/slab.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <net/pkt_sched.h>
38 #include <linux/net_tstamp.h>
39 #include <linux/mii.h>
40 #include <linux/ethtool.h>
41 #include <linux/if.h>
42 #include <linux/if_vlan.h>
43 #include <linux/pci.h>
44 #include <linux/pci-aspm.h>
45 #include <linux/delay.h>
46 #include <linux/interrupt.h>
47 #include <linux/ip.h>
48 #include <linux/tcp.h>
49 #include <linux/sctp.h>
50 #include <linux/if_ether.h>
51 #include <linux/aer.h>
52 #include <linux/prefetch.h>
53 #include <linux/pm_runtime.h>
54 #include <linux/etherdevice.h>
55 #ifdef CONFIG_IGB_DCA
56 #include <linux/dca.h>
57 #endif
58 #include <linux/i2c.h>
59 #include "igb.h"
60 
61 #define MAJ 5
62 #define MIN 4
63 #define BUILD 0
64 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
65 __stringify(BUILD) "-k"
66 
67 enum queue_mode {
68 	QUEUE_MODE_STRICT_PRIORITY,
69 	QUEUE_MODE_STREAM_RESERVATION,
70 };
71 
72 enum tx_queue_prio {
73 	TX_QUEUE_PRIO_HIGH,
74 	TX_QUEUE_PRIO_LOW,
75 };
76 
77 char igb_driver_name[] = "igb";
78 char igb_driver_version[] = DRV_VERSION;
79 static const char igb_driver_string[] =
80 				"Intel(R) Gigabit Ethernet Network Driver";
81 static const char igb_copyright[] =
82 				"Copyright (c) 2007-2014 Intel Corporation.";
83 
84 static const struct e1000_info *igb_info_tbl[] = {
85 	[board_82575] = &e1000_82575_info,
86 };
87 
88 static const struct pci_device_id igb_pci_tbl[] = {
89 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
90 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
91 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
92 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
93 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
94 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
95 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
96 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
97 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
98 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
99 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
100 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
101 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
102 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
103 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
104 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
105 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
106 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
107 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
108 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
109 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
110 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
111 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
112 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
113 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
114 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
115 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
116 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
117 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
118 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
119 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
120 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
121 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
122 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
123 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
124 	/* required last entry */
125 	{0, }
126 };
127 
128 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
129 
130 static int igb_setup_all_tx_resources(struct igb_adapter *);
131 static int igb_setup_all_rx_resources(struct igb_adapter *);
132 static void igb_free_all_tx_resources(struct igb_adapter *);
133 static void igb_free_all_rx_resources(struct igb_adapter *);
134 static void igb_setup_mrqc(struct igb_adapter *);
135 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
136 static void igb_remove(struct pci_dev *pdev);
137 static int igb_sw_init(struct igb_adapter *);
138 int igb_open(struct net_device *);
139 int igb_close(struct net_device *);
140 static void igb_configure(struct igb_adapter *);
141 static void igb_configure_tx(struct igb_adapter *);
142 static void igb_configure_rx(struct igb_adapter *);
143 static void igb_clean_all_tx_rings(struct igb_adapter *);
144 static void igb_clean_all_rx_rings(struct igb_adapter *);
145 static void igb_clean_tx_ring(struct igb_ring *);
146 static void igb_clean_rx_ring(struct igb_ring *);
147 static void igb_set_rx_mode(struct net_device *);
148 static void igb_update_phy_info(struct timer_list *);
149 static void igb_watchdog(struct timer_list *);
150 static void igb_watchdog_task(struct work_struct *);
151 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
152 static void igb_get_stats64(struct net_device *dev,
153 			    struct rtnl_link_stats64 *stats);
154 static int igb_change_mtu(struct net_device *, int);
155 static int igb_set_mac(struct net_device *, void *);
156 static void igb_set_uta(struct igb_adapter *adapter, bool set);
157 static irqreturn_t igb_intr(int irq, void *);
158 static irqreturn_t igb_intr_msi(int irq, void *);
159 static irqreturn_t igb_msix_other(int irq, void *);
160 static irqreturn_t igb_msix_ring(int irq, void *);
161 #ifdef CONFIG_IGB_DCA
162 static void igb_update_dca(struct igb_q_vector *);
163 static void igb_setup_dca(struct igb_adapter *);
164 #endif /* CONFIG_IGB_DCA */
165 static int igb_poll(struct napi_struct *, int);
166 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
167 static int igb_clean_rx_irq(struct igb_q_vector *, int);
168 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
169 static void igb_tx_timeout(struct net_device *);
170 static void igb_reset_task(struct work_struct *);
171 static void igb_vlan_mode(struct net_device *netdev,
172 			  netdev_features_t features);
173 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
174 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
175 static void igb_restore_vlan(struct igb_adapter *);
176 static void igb_rar_set_index(struct igb_adapter *, u32);
177 static void igb_ping_all_vfs(struct igb_adapter *);
178 static void igb_msg_task(struct igb_adapter *);
179 static void igb_vmm_control(struct igb_adapter *);
180 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
181 static void igb_flush_mac_table(struct igb_adapter *);
182 static int igb_available_rars(struct igb_adapter *, u8);
183 static void igb_set_default_mac_filter(struct igb_adapter *);
184 static int igb_uc_sync(struct net_device *, const unsigned char *);
185 static int igb_uc_unsync(struct net_device *, const unsigned char *);
186 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
187 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
188 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
189 			       int vf, u16 vlan, u8 qos, __be16 vlan_proto);
190 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
191 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
192 				   bool setting);
193 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
194 				 struct ifla_vf_info *ivi);
195 static void igb_check_vf_rate_limit(struct igb_adapter *);
196 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
197 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
198 
199 #ifdef CONFIG_PCI_IOV
200 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
201 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
202 static int igb_disable_sriov(struct pci_dev *dev);
203 static int igb_pci_disable_sriov(struct pci_dev *dev);
204 #endif
205 
206 static int igb_suspend(struct device *);
207 static int igb_resume(struct device *);
208 static int igb_runtime_suspend(struct device *dev);
209 static int igb_runtime_resume(struct device *dev);
210 static int igb_runtime_idle(struct device *dev);
211 static const struct dev_pm_ops igb_pm_ops = {
212 	SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
213 	SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
214 			igb_runtime_idle)
215 };
216 static void igb_shutdown(struct pci_dev *);
217 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
218 #ifdef CONFIG_IGB_DCA
219 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
220 static struct notifier_block dca_notifier = {
221 	.notifier_call	= igb_notify_dca,
222 	.next		= NULL,
223 	.priority	= 0
224 };
225 #endif
226 #ifdef CONFIG_NET_POLL_CONTROLLER
227 /* for netdump / net console */
228 static void igb_netpoll(struct net_device *);
229 #endif
230 #ifdef CONFIG_PCI_IOV
231 static unsigned int max_vfs;
232 module_param(max_vfs, uint, 0);
233 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
234 #endif /* CONFIG_PCI_IOV */
235 
236 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
237 		     pci_channel_state_t);
238 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
239 static void igb_io_resume(struct pci_dev *);
240 
241 static const struct pci_error_handlers igb_err_handler = {
242 	.error_detected = igb_io_error_detected,
243 	.slot_reset = igb_io_slot_reset,
244 	.resume = igb_io_resume,
245 };
246 
247 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
248 
249 static struct pci_driver igb_driver = {
250 	.name     = igb_driver_name,
251 	.id_table = igb_pci_tbl,
252 	.probe    = igb_probe,
253 	.remove   = igb_remove,
254 #ifdef CONFIG_PM
255 	.driver.pm = &igb_pm_ops,
256 #endif
257 	.shutdown = igb_shutdown,
258 	.sriov_configure = igb_pci_sriov_configure,
259 	.err_handler = &igb_err_handler
260 };
261 
262 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
263 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
264 MODULE_LICENSE("GPL");
265 MODULE_VERSION(DRV_VERSION);
266 
267 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
268 static int debug = -1;
269 module_param(debug, int, 0);
270 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
271 
272 struct igb_reg_info {
273 	u32 ofs;
274 	char *name;
275 };
276 
277 static const struct igb_reg_info igb_reg_info_tbl[] = {
278 
279 	/* General Registers */
280 	{E1000_CTRL, "CTRL"},
281 	{E1000_STATUS, "STATUS"},
282 	{E1000_CTRL_EXT, "CTRL_EXT"},
283 
284 	/* Interrupt Registers */
285 	{E1000_ICR, "ICR"},
286 
287 	/* RX Registers */
288 	{E1000_RCTL, "RCTL"},
289 	{E1000_RDLEN(0), "RDLEN"},
290 	{E1000_RDH(0), "RDH"},
291 	{E1000_RDT(0), "RDT"},
292 	{E1000_RXDCTL(0), "RXDCTL"},
293 	{E1000_RDBAL(0), "RDBAL"},
294 	{E1000_RDBAH(0), "RDBAH"},
295 
296 	/* TX Registers */
297 	{E1000_TCTL, "TCTL"},
298 	{E1000_TDBAL(0), "TDBAL"},
299 	{E1000_TDBAH(0), "TDBAH"},
300 	{E1000_TDLEN(0), "TDLEN"},
301 	{E1000_TDH(0), "TDH"},
302 	{E1000_TDT(0), "TDT"},
303 	{E1000_TXDCTL(0), "TXDCTL"},
304 	{E1000_TDFH, "TDFH"},
305 	{E1000_TDFT, "TDFT"},
306 	{E1000_TDFHS, "TDFHS"},
307 	{E1000_TDFPC, "TDFPC"},
308 
309 	/* List Terminator */
310 	{}
311 };
312 
313 /* igb_regdump - register printout routine */
314 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
315 {
316 	int n = 0;
317 	char rname[16];
318 	u32 regs[8];
319 
320 	switch (reginfo->ofs) {
321 	case E1000_RDLEN(0):
322 		for (n = 0; n < 4; n++)
323 			regs[n] = rd32(E1000_RDLEN(n));
324 		break;
325 	case E1000_RDH(0):
326 		for (n = 0; n < 4; n++)
327 			regs[n] = rd32(E1000_RDH(n));
328 		break;
329 	case E1000_RDT(0):
330 		for (n = 0; n < 4; n++)
331 			regs[n] = rd32(E1000_RDT(n));
332 		break;
333 	case E1000_RXDCTL(0):
334 		for (n = 0; n < 4; n++)
335 			regs[n] = rd32(E1000_RXDCTL(n));
336 		break;
337 	case E1000_RDBAL(0):
338 		for (n = 0; n < 4; n++)
339 			regs[n] = rd32(E1000_RDBAL(n));
340 		break;
341 	case E1000_RDBAH(0):
342 		for (n = 0; n < 4; n++)
343 			regs[n] = rd32(E1000_RDBAH(n));
344 		break;
345 	case E1000_TDBAL(0):
346 		for (n = 0; n < 4; n++)
347 			regs[n] = rd32(E1000_RDBAL(n));
348 		break;
349 	case E1000_TDBAH(0):
350 		for (n = 0; n < 4; n++)
351 			regs[n] = rd32(E1000_TDBAH(n));
352 		break;
353 	case E1000_TDLEN(0):
354 		for (n = 0; n < 4; n++)
355 			regs[n] = rd32(E1000_TDLEN(n));
356 		break;
357 	case E1000_TDH(0):
358 		for (n = 0; n < 4; n++)
359 			regs[n] = rd32(E1000_TDH(n));
360 		break;
361 	case E1000_TDT(0):
362 		for (n = 0; n < 4; n++)
363 			regs[n] = rd32(E1000_TDT(n));
364 		break;
365 	case E1000_TXDCTL(0):
366 		for (n = 0; n < 4; n++)
367 			regs[n] = rd32(E1000_TXDCTL(n));
368 		break;
369 	default:
370 		pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
371 		return;
372 	}
373 
374 	snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
375 	pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
376 		regs[2], regs[3]);
377 }
378 
379 /* igb_dump - Print registers, Tx-rings and Rx-rings */
380 static void igb_dump(struct igb_adapter *adapter)
381 {
382 	struct net_device *netdev = adapter->netdev;
383 	struct e1000_hw *hw = &adapter->hw;
384 	struct igb_reg_info *reginfo;
385 	struct igb_ring *tx_ring;
386 	union e1000_adv_tx_desc *tx_desc;
387 	struct my_u0 { u64 a; u64 b; } *u0;
388 	struct igb_ring *rx_ring;
389 	union e1000_adv_rx_desc *rx_desc;
390 	u32 staterr;
391 	u16 i, n;
392 
393 	if (!netif_msg_hw(adapter))
394 		return;
395 
396 	/* Print netdevice Info */
397 	if (netdev) {
398 		dev_info(&adapter->pdev->dev, "Net device Info\n");
399 		pr_info("Device Name     state            trans_start\n");
400 		pr_info("%-15s %016lX %016lX\n", netdev->name,
401 			netdev->state, dev_trans_start(netdev));
402 	}
403 
404 	/* Print Registers */
405 	dev_info(&adapter->pdev->dev, "Register Dump\n");
406 	pr_info(" Register Name   Value\n");
407 	for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
408 	     reginfo->name; reginfo++) {
409 		igb_regdump(hw, reginfo);
410 	}
411 
412 	/* Print TX Ring Summary */
413 	if (!netdev || !netif_running(netdev))
414 		goto exit;
415 
416 	dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
417 	pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
418 	for (n = 0; n < adapter->num_tx_queues; n++) {
419 		struct igb_tx_buffer *buffer_info;
420 		tx_ring = adapter->tx_ring[n];
421 		buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
422 		pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
423 			n, tx_ring->next_to_use, tx_ring->next_to_clean,
424 			(u64)dma_unmap_addr(buffer_info, dma),
425 			dma_unmap_len(buffer_info, len),
426 			buffer_info->next_to_watch,
427 			(u64)buffer_info->time_stamp);
428 	}
429 
430 	/* Print TX Rings */
431 	if (!netif_msg_tx_done(adapter))
432 		goto rx_ring_summary;
433 
434 	dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
435 
436 	/* Transmit Descriptor Formats
437 	 *
438 	 * Advanced Transmit Descriptor
439 	 *   +--------------------------------------------------------------+
440 	 * 0 |         Buffer Address [63:0]                                |
441 	 *   +--------------------------------------------------------------+
442 	 * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
443 	 *   +--------------------------------------------------------------+
444 	 *   63      46 45    40 39 38 36 35 32 31   24             15       0
445 	 */
446 
447 	for (n = 0; n < adapter->num_tx_queues; n++) {
448 		tx_ring = adapter->tx_ring[n];
449 		pr_info("------------------------------------\n");
450 		pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
451 		pr_info("------------------------------------\n");
452 		pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
453 
454 		for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
455 			const char *next_desc;
456 			struct igb_tx_buffer *buffer_info;
457 			tx_desc = IGB_TX_DESC(tx_ring, i);
458 			buffer_info = &tx_ring->tx_buffer_info[i];
459 			u0 = (struct my_u0 *)tx_desc;
460 			if (i == tx_ring->next_to_use &&
461 			    i == tx_ring->next_to_clean)
462 				next_desc = " NTC/U";
463 			else if (i == tx_ring->next_to_use)
464 				next_desc = " NTU";
465 			else if (i == tx_ring->next_to_clean)
466 				next_desc = " NTC";
467 			else
468 				next_desc = "";
469 
470 			pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
471 				i, le64_to_cpu(u0->a),
472 				le64_to_cpu(u0->b),
473 				(u64)dma_unmap_addr(buffer_info, dma),
474 				dma_unmap_len(buffer_info, len),
475 				buffer_info->next_to_watch,
476 				(u64)buffer_info->time_stamp,
477 				buffer_info->skb, next_desc);
478 
479 			if (netif_msg_pktdata(adapter) && buffer_info->skb)
480 				print_hex_dump(KERN_INFO, "",
481 					DUMP_PREFIX_ADDRESS,
482 					16, 1, buffer_info->skb->data,
483 					dma_unmap_len(buffer_info, len),
484 					true);
485 		}
486 	}
487 
488 	/* Print RX Rings Summary */
489 rx_ring_summary:
490 	dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
491 	pr_info("Queue [NTU] [NTC]\n");
492 	for (n = 0; n < adapter->num_rx_queues; n++) {
493 		rx_ring = adapter->rx_ring[n];
494 		pr_info(" %5d %5X %5X\n",
495 			n, rx_ring->next_to_use, rx_ring->next_to_clean);
496 	}
497 
498 	/* Print RX Rings */
499 	if (!netif_msg_rx_status(adapter))
500 		goto exit;
501 
502 	dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
503 
504 	/* Advanced Receive Descriptor (Read) Format
505 	 *    63                                           1        0
506 	 *    +-----------------------------------------------------+
507 	 *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
508 	 *    +----------------------------------------------+------+
509 	 *  8 |       Header Buffer Address [63:1]           |  DD  |
510 	 *    +-----------------------------------------------------+
511 	 *
512 	 *
513 	 * Advanced Receive Descriptor (Write-Back) Format
514 	 *
515 	 *   63       48 47    32 31  30      21 20 17 16   4 3     0
516 	 *   +------------------------------------------------------+
517 	 * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
518 	 *   | Checksum   Ident  |   |           |    | Type | Type |
519 	 *   +------------------------------------------------------+
520 	 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
521 	 *   +------------------------------------------------------+
522 	 *   63       48 47    32 31            20 19               0
523 	 */
524 
525 	for (n = 0; n < adapter->num_rx_queues; n++) {
526 		rx_ring = adapter->rx_ring[n];
527 		pr_info("------------------------------------\n");
528 		pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
529 		pr_info("------------------------------------\n");
530 		pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
531 		pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
532 
533 		for (i = 0; i < rx_ring->count; i++) {
534 			const char *next_desc;
535 			struct igb_rx_buffer *buffer_info;
536 			buffer_info = &rx_ring->rx_buffer_info[i];
537 			rx_desc = IGB_RX_DESC(rx_ring, i);
538 			u0 = (struct my_u0 *)rx_desc;
539 			staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
540 
541 			if (i == rx_ring->next_to_use)
542 				next_desc = " NTU";
543 			else if (i == rx_ring->next_to_clean)
544 				next_desc = " NTC";
545 			else
546 				next_desc = "";
547 
548 			if (staterr & E1000_RXD_STAT_DD) {
549 				/* Descriptor Done */
550 				pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
551 					"RWB", i,
552 					le64_to_cpu(u0->a),
553 					le64_to_cpu(u0->b),
554 					next_desc);
555 			} else {
556 				pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
557 					"R  ", i,
558 					le64_to_cpu(u0->a),
559 					le64_to_cpu(u0->b),
560 					(u64)buffer_info->dma,
561 					next_desc);
562 
563 				if (netif_msg_pktdata(adapter) &&
564 				    buffer_info->dma && buffer_info->page) {
565 					print_hex_dump(KERN_INFO, "",
566 					  DUMP_PREFIX_ADDRESS,
567 					  16, 1,
568 					  page_address(buffer_info->page) +
569 						      buffer_info->page_offset,
570 					  igb_rx_bufsz(rx_ring), true);
571 				}
572 			}
573 		}
574 	}
575 
576 exit:
577 	return;
578 }
579 
580 /**
581  *  igb_get_i2c_data - Reads the I2C SDA data bit
582  *  @hw: pointer to hardware structure
583  *  @i2cctl: Current value of I2CCTL register
584  *
585  *  Returns the I2C data bit value
586  **/
587 static int igb_get_i2c_data(void *data)
588 {
589 	struct igb_adapter *adapter = (struct igb_adapter *)data;
590 	struct e1000_hw *hw = &adapter->hw;
591 	s32 i2cctl = rd32(E1000_I2CPARAMS);
592 
593 	return !!(i2cctl & E1000_I2C_DATA_IN);
594 }
595 
596 /**
597  *  igb_set_i2c_data - Sets the I2C data bit
598  *  @data: pointer to hardware structure
599  *  @state: I2C data value (0 or 1) to set
600  *
601  *  Sets the I2C data bit
602  **/
603 static void igb_set_i2c_data(void *data, int state)
604 {
605 	struct igb_adapter *adapter = (struct igb_adapter *)data;
606 	struct e1000_hw *hw = &adapter->hw;
607 	s32 i2cctl = rd32(E1000_I2CPARAMS);
608 
609 	if (state)
610 		i2cctl |= E1000_I2C_DATA_OUT;
611 	else
612 		i2cctl &= ~E1000_I2C_DATA_OUT;
613 
614 	i2cctl &= ~E1000_I2C_DATA_OE_N;
615 	i2cctl |= E1000_I2C_CLK_OE_N;
616 	wr32(E1000_I2CPARAMS, i2cctl);
617 	wrfl();
618 
619 }
620 
621 /**
622  *  igb_set_i2c_clk - Sets the I2C SCL clock
623  *  @data: pointer to hardware structure
624  *  @state: state to set clock
625  *
626  *  Sets the I2C clock line to state
627  **/
628 static void igb_set_i2c_clk(void *data, int state)
629 {
630 	struct igb_adapter *adapter = (struct igb_adapter *)data;
631 	struct e1000_hw *hw = &adapter->hw;
632 	s32 i2cctl = rd32(E1000_I2CPARAMS);
633 
634 	if (state) {
635 		i2cctl |= E1000_I2C_CLK_OUT;
636 		i2cctl &= ~E1000_I2C_CLK_OE_N;
637 	} else {
638 		i2cctl &= ~E1000_I2C_CLK_OUT;
639 		i2cctl &= ~E1000_I2C_CLK_OE_N;
640 	}
641 	wr32(E1000_I2CPARAMS, i2cctl);
642 	wrfl();
643 }
644 
645 /**
646  *  igb_get_i2c_clk - Gets the I2C SCL clock state
647  *  @data: pointer to hardware structure
648  *
649  *  Gets the I2C clock state
650  **/
651 static int igb_get_i2c_clk(void *data)
652 {
653 	struct igb_adapter *adapter = (struct igb_adapter *)data;
654 	struct e1000_hw *hw = &adapter->hw;
655 	s32 i2cctl = rd32(E1000_I2CPARAMS);
656 
657 	return !!(i2cctl & E1000_I2C_CLK_IN);
658 }
659 
660 static const struct i2c_algo_bit_data igb_i2c_algo = {
661 	.setsda		= igb_set_i2c_data,
662 	.setscl		= igb_set_i2c_clk,
663 	.getsda		= igb_get_i2c_data,
664 	.getscl		= igb_get_i2c_clk,
665 	.udelay		= 5,
666 	.timeout	= 20,
667 };
668 
669 /**
670  *  igb_get_hw_dev - return device
671  *  @hw: pointer to hardware structure
672  *
673  *  used by hardware layer to print debugging information
674  **/
675 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
676 {
677 	struct igb_adapter *adapter = hw->back;
678 	return adapter->netdev;
679 }
680 
681 /**
682  *  igb_init_module - Driver Registration Routine
683  *
684  *  igb_init_module is the first routine called when the driver is
685  *  loaded. All it does is register with the PCI subsystem.
686  **/
687 static int __init igb_init_module(void)
688 {
689 	int ret;
690 
691 	pr_info("%s - version %s\n",
692 	       igb_driver_string, igb_driver_version);
693 	pr_info("%s\n", igb_copyright);
694 
695 #ifdef CONFIG_IGB_DCA
696 	dca_register_notify(&dca_notifier);
697 #endif
698 	ret = pci_register_driver(&igb_driver);
699 	return ret;
700 }
701 
702 module_init(igb_init_module);
703 
704 /**
705  *  igb_exit_module - Driver Exit Cleanup Routine
706  *
707  *  igb_exit_module is called just before the driver is removed
708  *  from memory.
709  **/
710 static void __exit igb_exit_module(void)
711 {
712 #ifdef CONFIG_IGB_DCA
713 	dca_unregister_notify(&dca_notifier);
714 #endif
715 	pci_unregister_driver(&igb_driver);
716 }
717 
718 module_exit(igb_exit_module);
719 
720 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
721 /**
722  *  igb_cache_ring_register - Descriptor ring to register mapping
723  *  @adapter: board private structure to initialize
724  *
725  *  Once we know the feature-set enabled for the device, we'll cache
726  *  the register offset the descriptor ring is assigned to.
727  **/
728 static void igb_cache_ring_register(struct igb_adapter *adapter)
729 {
730 	int i = 0, j = 0;
731 	u32 rbase_offset = adapter->vfs_allocated_count;
732 
733 	switch (adapter->hw.mac.type) {
734 	case e1000_82576:
735 		/* The queues are allocated for virtualization such that VF 0
736 		 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
737 		 * In order to avoid collision we start at the first free queue
738 		 * and continue consuming queues in the same sequence
739 		 */
740 		if (adapter->vfs_allocated_count) {
741 			for (; i < adapter->rss_queues; i++)
742 				adapter->rx_ring[i]->reg_idx = rbase_offset +
743 							       Q_IDX_82576(i);
744 		}
745 		/* Fall through */
746 	case e1000_82575:
747 	case e1000_82580:
748 	case e1000_i350:
749 	case e1000_i354:
750 	case e1000_i210:
751 	case e1000_i211:
752 		/* Fall through */
753 	default:
754 		for (; i < adapter->num_rx_queues; i++)
755 			adapter->rx_ring[i]->reg_idx = rbase_offset + i;
756 		for (; j < adapter->num_tx_queues; j++)
757 			adapter->tx_ring[j]->reg_idx = rbase_offset + j;
758 		break;
759 	}
760 }
761 
762 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
763 {
764 	struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
765 	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
766 	u32 value = 0;
767 
768 	if (E1000_REMOVED(hw_addr))
769 		return ~value;
770 
771 	value = readl(&hw_addr[reg]);
772 
773 	/* reads should not return all F's */
774 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
775 		struct net_device *netdev = igb->netdev;
776 		hw->hw_addr = NULL;
777 		netif_device_detach(netdev);
778 		netdev_err(netdev, "PCIe link lost, device now detached\n");
779 	}
780 
781 	return value;
782 }
783 
784 /**
785  *  igb_write_ivar - configure ivar for given MSI-X vector
786  *  @hw: pointer to the HW structure
787  *  @msix_vector: vector number we are allocating to a given ring
788  *  @index: row index of IVAR register to write within IVAR table
789  *  @offset: column offset of in IVAR, should be multiple of 8
790  *
791  *  This function is intended to handle the writing of the IVAR register
792  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
793  *  each containing an cause allocation for an Rx and Tx ring, and a
794  *  variable number of rows depending on the number of queues supported.
795  **/
796 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
797 			   int index, int offset)
798 {
799 	u32 ivar = array_rd32(E1000_IVAR0, index);
800 
801 	/* clear any bits that are currently set */
802 	ivar &= ~((u32)0xFF << offset);
803 
804 	/* write vector and valid bit */
805 	ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
806 
807 	array_wr32(E1000_IVAR0, index, ivar);
808 }
809 
810 #define IGB_N0_QUEUE -1
811 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
812 {
813 	struct igb_adapter *adapter = q_vector->adapter;
814 	struct e1000_hw *hw = &adapter->hw;
815 	int rx_queue = IGB_N0_QUEUE;
816 	int tx_queue = IGB_N0_QUEUE;
817 	u32 msixbm = 0;
818 
819 	if (q_vector->rx.ring)
820 		rx_queue = q_vector->rx.ring->reg_idx;
821 	if (q_vector->tx.ring)
822 		tx_queue = q_vector->tx.ring->reg_idx;
823 
824 	switch (hw->mac.type) {
825 	case e1000_82575:
826 		/* The 82575 assigns vectors using a bitmask, which matches the
827 		 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
828 		 * or more queues to a vector, we write the appropriate bits
829 		 * into the MSIXBM register for that vector.
830 		 */
831 		if (rx_queue > IGB_N0_QUEUE)
832 			msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
833 		if (tx_queue > IGB_N0_QUEUE)
834 			msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
835 		if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
836 			msixbm |= E1000_EIMS_OTHER;
837 		array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
838 		q_vector->eims_value = msixbm;
839 		break;
840 	case e1000_82576:
841 		/* 82576 uses a table that essentially consists of 2 columns
842 		 * with 8 rows.  The ordering is column-major so we use the
843 		 * lower 3 bits as the row index, and the 4th bit as the
844 		 * column offset.
845 		 */
846 		if (rx_queue > IGB_N0_QUEUE)
847 			igb_write_ivar(hw, msix_vector,
848 				       rx_queue & 0x7,
849 				       (rx_queue & 0x8) << 1);
850 		if (tx_queue > IGB_N0_QUEUE)
851 			igb_write_ivar(hw, msix_vector,
852 				       tx_queue & 0x7,
853 				       ((tx_queue & 0x8) << 1) + 8);
854 		q_vector->eims_value = BIT(msix_vector);
855 		break;
856 	case e1000_82580:
857 	case e1000_i350:
858 	case e1000_i354:
859 	case e1000_i210:
860 	case e1000_i211:
861 		/* On 82580 and newer adapters the scheme is similar to 82576
862 		 * however instead of ordering column-major we have things
863 		 * ordered row-major.  So we traverse the table by using
864 		 * bit 0 as the column offset, and the remaining bits as the
865 		 * row index.
866 		 */
867 		if (rx_queue > IGB_N0_QUEUE)
868 			igb_write_ivar(hw, msix_vector,
869 				       rx_queue >> 1,
870 				       (rx_queue & 0x1) << 4);
871 		if (tx_queue > IGB_N0_QUEUE)
872 			igb_write_ivar(hw, msix_vector,
873 				       tx_queue >> 1,
874 				       ((tx_queue & 0x1) << 4) + 8);
875 		q_vector->eims_value = BIT(msix_vector);
876 		break;
877 	default:
878 		BUG();
879 		break;
880 	}
881 
882 	/* add q_vector eims value to global eims_enable_mask */
883 	adapter->eims_enable_mask |= q_vector->eims_value;
884 
885 	/* configure q_vector to set itr on first interrupt */
886 	q_vector->set_itr = 1;
887 }
888 
889 /**
890  *  igb_configure_msix - Configure MSI-X hardware
891  *  @adapter: board private structure to initialize
892  *
893  *  igb_configure_msix sets up the hardware to properly
894  *  generate MSI-X interrupts.
895  **/
896 static void igb_configure_msix(struct igb_adapter *adapter)
897 {
898 	u32 tmp;
899 	int i, vector = 0;
900 	struct e1000_hw *hw = &adapter->hw;
901 
902 	adapter->eims_enable_mask = 0;
903 
904 	/* set vector for other causes, i.e. link changes */
905 	switch (hw->mac.type) {
906 	case e1000_82575:
907 		tmp = rd32(E1000_CTRL_EXT);
908 		/* enable MSI-X PBA support*/
909 		tmp |= E1000_CTRL_EXT_PBA_CLR;
910 
911 		/* Auto-Mask interrupts upon ICR read. */
912 		tmp |= E1000_CTRL_EXT_EIAME;
913 		tmp |= E1000_CTRL_EXT_IRCA;
914 
915 		wr32(E1000_CTRL_EXT, tmp);
916 
917 		/* enable msix_other interrupt */
918 		array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
919 		adapter->eims_other = E1000_EIMS_OTHER;
920 
921 		break;
922 
923 	case e1000_82576:
924 	case e1000_82580:
925 	case e1000_i350:
926 	case e1000_i354:
927 	case e1000_i210:
928 	case e1000_i211:
929 		/* Turn on MSI-X capability first, or our settings
930 		 * won't stick.  And it will take days to debug.
931 		 */
932 		wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
933 		     E1000_GPIE_PBA | E1000_GPIE_EIAME |
934 		     E1000_GPIE_NSICR);
935 
936 		/* enable msix_other interrupt */
937 		adapter->eims_other = BIT(vector);
938 		tmp = (vector++ | E1000_IVAR_VALID) << 8;
939 
940 		wr32(E1000_IVAR_MISC, tmp);
941 		break;
942 	default:
943 		/* do nothing, since nothing else supports MSI-X */
944 		break;
945 	} /* switch (hw->mac.type) */
946 
947 	adapter->eims_enable_mask |= adapter->eims_other;
948 
949 	for (i = 0; i < adapter->num_q_vectors; i++)
950 		igb_assign_vector(adapter->q_vector[i], vector++);
951 
952 	wrfl();
953 }
954 
955 /**
956  *  igb_request_msix - Initialize MSI-X interrupts
957  *  @adapter: board private structure to initialize
958  *
959  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
960  *  kernel.
961  **/
962 static int igb_request_msix(struct igb_adapter *adapter)
963 {
964 	struct net_device *netdev = adapter->netdev;
965 	int i, err = 0, vector = 0, free_vector = 0;
966 
967 	err = request_irq(adapter->msix_entries[vector].vector,
968 			  igb_msix_other, 0, netdev->name, adapter);
969 	if (err)
970 		goto err_out;
971 
972 	for (i = 0; i < adapter->num_q_vectors; i++) {
973 		struct igb_q_vector *q_vector = adapter->q_vector[i];
974 
975 		vector++;
976 
977 		q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
978 
979 		if (q_vector->rx.ring && q_vector->tx.ring)
980 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
981 				q_vector->rx.ring->queue_index);
982 		else if (q_vector->tx.ring)
983 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
984 				q_vector->tx.ring->queue_index);
985 		else if (q_vector->rx.ring)
986 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
987 				q_vector->rx.ring->queue_index);
988 		else
989 			sprintf(q_vector->name, "%s-unused", netdev->name);
990 
991 		err = request_irq(adapter->msix_entries[vector].vector,
992 				  igb_msix_ring, 0, q_vector->name,
993 				  q_vector);
994 		if (err)
995 			goto err_free;
996 	}
997 
998 	igb_configure_msix(adapter);
999 	return 0;
1000 
1001 err_free:
1002 	/* free already assigned IRQs */
1003 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
1004 
1005 	vector--;
1006 	for (i = 0; i < vector; i++) {
1007 		free_irq(adapter->msix_entries[free_vector++].vector,
1008 			 adapter->q_vector[i]);
1009 	}
1010 err_out:
1011 	return err;
1012 }
1013 
1014 /**
1015  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
1016  *  @adapter: board private structure to initialize
1017  *  @v_idx: Index of vector to be freed
1018  *
1019  *  This function frees the memory allocated to the q_vector.
1020  **/
1021 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1022 {
1023 	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1024 
1025 	adapter->q_vector[v_idx] = NULL;
1026 
1027 	/* igb_get_stats64() might access the rings on this vector,
1028 	 * we must wait a grace period before freeing it.
1029 	 */
1030 	if (q_vector)
1031 		kfree_rcu(q_vector, rcu);
1032 }
1033 
1034 /**
1035  *  igb_reset_q_vector - Reset config for interrupt vector
1036  *  @adapter: board private structure to initialize
1037  *  @v_idx: Index of vector to be reset
1038  *
1039  *  If NAPI is enabled it will delete any references to the
1040  *  NAPI struct. This is preparation for igb_free_q_vector.
1041  **/
1042 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1043 {
1044 	struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1045 
1046 	/* Coming from igb_set_interrupt_capability, the vectors are not yet
1047 	 * allocated. So, q_vector is NULL so we should stop here.
1048 	 */
1049 	if (!q_vector)
1050 		return;
1051 
1052 	if (q_vector->tx.ring)
1053 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1054 
1055 	if (q_vector->rx.ring)
1056 		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1057 
1058 	netif_napi_del(&q_vector->napi);
1059 
1060 }
1061 
1062 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1063 {
1064 	int v_idx = adapter->num_q_vectors;
1065 
1066 	if (adapter->flags & IGB_FLAG_HAS_MSIX)
1067 		pci_disable_msix(adapter->pdev);
1068 	else if (adapter->flags & IGB_FLAG_HAS_MSI)
1069 		pci_disable_msi(adapter->pdev);
1070 
1071 	while (v_idx--)
1072 		igb_reset_q_vector(adapter, v_idx);
1073 }
1074 
1075 /**
1076  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1077  *  @adapter: board private structure to initialize
1078  *
1079  *  This function frees the memory allocated to the q_vectors.  In addition if
1080  *  NAPI is enabled it will delete any references to the NAPI struct prior
1081  *  to freeing the q_vector.
1082  **/
1083 static void igb_free_q_vectors(struct igb_adapter *adapter)
1084 {
1085 	int v_idx = adapter->num_q_vectors;
1086 
1087 	adapter->num_tx_queues = 0;
1088 	adapter->num_rx_queues = 0;
1089 	adapter->num_q_vectors = 0;
1090 
1091 	while (v_idx--) {
1092 		igb_reset_q_vector(adapter, v_idx);
1093 		igb_free_q_vector(adapter, v_idx);
1094 	}
1095 }
1096 
1097 /**
1098  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1099  *  @adapter: board private structure to initialize
1100  *
1101  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1102  *  MSI-X interrupts allocated.
1103  */
1104 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1105 {
1106 	igb_free_q_vectors(adapter);
1107 	igb_reset_interrupt_capability(adapter);
1108 }
1109 
1110 /**
1111  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1112  *  @adapter: board private structure to initialize
1113  *  @msix: boolean value of MSIX capability
1114  *
1115  *  Attempt to configure interrupts using the best available
1116  *  capabilities of the hardware and kernel.
1117  **/
1118 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1119 {
1120 	int err;
1121 	int numvecs, i;
1122 
1123 	if (!msix)
1124 		goto msi_only;
1125 	adapter->flags |= IGB_FLAG_HAS_MSIX;
1126 
1127 	/* Number of supported queues. */
1128 	adapter->num_rx_queues = adapter->rss_queues;
1129 	if (adapter->vfs_allocated_count)
1130 		adapter->num_tx_queues = 1;
1131 	else
1132 		adapter->num_tx_queues = adapter->rss_queues;
1133 
1134 	/* start with one vector for every Rx queue */
1135 	numvecs = adapter->num_rx_queues;
1136 
1137 	/* if Tx handler is separate add 1 for every Tx queue */
1138 	if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1139 		numvecs += adapter->num_tx_queues;
1140 
1141 	/* store the number of vectors reserved for queues */
1142 	adapter->num_q_vectors = numvecs;
1143 
1144 	/* add 1 vector for link status interrupts */
1145 	numvecs++;
1146 	for (i = 0; i < numvecs; i++)
1147 		adapter->msix_entries[i].entry = i;
1148 
1149 	err = pci_enable_msix_range(adapter->pdev,
1150 				    adapter->msix_entries,
1151 				    numvecs,
1152 				    numvecs);
1153 	if (err > 0)
1154 		return;
1155 
1156 	igb_reset_interrupt_capability(adapter);
1157 
1158 	/* If we can't do MSI-X, try MSI */
1159 msi_only:
1160 	adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1161 #ifdef CONFIG_PCI_IOV
1162 	/* disable SR-IOV for non MSI-X configurations */
1163 	if (adapter->vf_data) {
1164 		struct e1000_hw *hw = &adapter->hw;
1165 		/* disable iov and allow time for transactions to clear */
1166 		pci_disable_sriov(adapter->pdev);
1167 		msleep(500);
1168 
1169 		kfree(adapter->vf_mac_list);
1170 		adapter->vf_mac_list = NULL;
1171 		kfree(adapter->vf_data);
1172 		adapter->vf_data = NULL;
1173 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1174 		wrfl();
1175 		msleep(100);
1176 		dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1177 	}
1178 #endif
1179 	adapter->vfs_allocated_count = 0;
1180 	adapter->rss_queues = 1;
1181 	adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1182 	adapter->num_rx_queues = 1;
1183 	adapter->num_tx_queues = 1;
1184 	adapter->num_q_vectors = 1;
1185 	if (!pci_enable_msi(adapter->pdev))
1186 		adapter->flags |= IGB_FLAG_HAS_MSI;
1187 }
1188 
1189 static void igb_add_ring(struct igb_ring *ring,
1190 			 struct igb_ring_container *head)
1191 {
1192 	head->ring = ring;
1193 	head->count++;
1194 }
1195 
1196 /**
1197  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1198  *  @adapter: board private structure to initialize
1199  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1200  *  @v_idx: index of vector in adapter struct
1201  *  @txr_count: total number of Tx rings to allocate
1202  *  @txr_idx: index of first Tx ring to allocate
1203  *  @rxr_count: total number of Rx rings to allocate
1204  *  @rxr_idx: index of first Rx ring to allocate
1205  *
1206  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1207  **/
1208 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1209 			      int v_count, int v_idx,
1210 			      int txr_count, int txr_idx,
1211 			      int rxr_count, int rxr_idx)
1212 {
1213 	struct igb_q_vector *q_vector;
1214 	struct igb_ring *ring;
1215 	int ring_count, size;
1216 
1217 	/* igb only supports 1 Tx and/or 1 Rx queue per vector */
1218 	if (txr_count > 1 || rxr_count > 1)
1219 		return -ENOMEM;
1220 
1221 	ring_count = txr_count + rxr_count;
1222 	size = sizeof(struct igb_q_vector) +
1223 	       (sizeof(struct igb_ring) * ring_count);
1224 
1225 	/* allocate q_vector and rings */
1226 	q_vector = adapter->q_vector[v_idx];
1227 	if (!q_vector) {
1228 		q_vector = kzalloc(size, GFP_KERNEL);
1229 	} else if (size > ksize(q_vector)) {
1230 		kfree_rcu(q_vector, rcu);
1231 		q_vector = kzalloc(size, GFP_KERNEL);
1232 	} else {
1233 		memset(q_vector, 0, size);
1234 	}
1235 	if (!q_vector)
1236 		return -ENOMEM;
1237 
1238 	/* initialize NAPI */
1239 	netif_napi_add(adapter->netdev, &q_vector->napi,
1240 		       igb_poll, 64);
1241 
1242 	/* tie q_vector and adapter together */
1243 	adapter->q_vector[v_idx] = q_vector;
1244 	q_vector->adapter = adapter;
1245 
1246 	/* initialize work limits */
1247 	q_vector->tx.work_limit = adapter->tx_work_limit;
1248 
1249 	/* initialize ITR configuration */
1250 	q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1251 	q_vector->itr_val = IGB_START_ITR;
1252 
1253 	/* initialize pointer to rings */
1254 	ring = q_vector->ring;
1255 
1256 	/* intialize ITR */
1257 	if (rxr_count) {
1258 		/* rx or rx/tx vector */
1259 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1260 			q_vector->itr_val = adapter->rx_itr_setting;
1261 	} else {
1262 		/* tx only vector */
1263 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1264 			q_vector->itr_val = adapter->tx_itr_setting;
1265 	}
1266 
1267 	if (txr_count) {
1268 		/* assign generic ring traits */
1269 		ring->dev = &adapter->pdev->dev;
1270 		ring->netdev = adapter->netdev;
1271 
1272 		/* configure backlink on ring */
1273 		ring->q_vector = q_vector;
1274 
1275 		/* update q_vector Tx values */
1276 		igb_add_ring(ring, &q_vector->tx);
1277 
1278 		/* For 82575, context index must be unique per ring. */
1279 		if (adapter->hw.mac.type == e1000_82575)
1280 			set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1281 
1282 		/* apply Tx specific ring traits */
1283 		ring->count = adapter->tx_ring_count;
1284 		ring->queue_index = txr_idx;
1285 
1286 		ring->cbs_enable = false;
1287 		ring->idleslope = 0;
1288 		ring->sendslope = 0;
1289 		ring->hicredit = 0;
1290 		ring->locredit = 0;
1291 
1292 		u64_stats_init(&ring->tx_syncp);
1293 		u64_stats_init(&ring->tx_syncp2);
1294 
1295 		/* assign ring to adapter */
1296 		adapter->tx_ring[txr_idx] = ring;
1297 
1298 		/* push pointer to next ring */
1299 		ring++;
1300 	}
1301 
1302 	if (rxr_count) {
1303 		/* assign generic ring traits */
1304 		ring->dev = &adapter->pdev->dev;
1305 		ring->netdev = adapter->netdev;
1306 
1307 		/* configure backlink on ring */
1308 		ring->q_vector = q_vector;
1309 
1310 		/* update q_vector Rx values */
1311 		igb_add_ring(ring, &q_vector->rx);
1312 
1313 		/* set flag indicating ring supports SCTP checksum offload */
1314 		if (adapter->hw.mac.type >= e1000_82576)
1315 			set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1316 
1317 		/* On i350, i354, i210, and i211, loopback VLAN packets
1318 		 * have the tag byte-swapped.
1319 		 */
1320 		if (adapter->hw.mac.type >= e1000_i350)
1321 			set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1322 
1323 		/* apply Rx specific ring traits */
1324 		ring->count = adapter->rx_ring_count;
1325 		ring->queue_index = rxr_idx;
1326 
1327 		u64_stats_init(&ring->rx_syncp);
1328 
1329 		/* assign ring to adapter */
1330 		adapter->rx_ring[rxr_idx] = ring;
1331 	}
1332 
1333 	return 0;
1334 }
1335 
1336 
1337 /**
1338  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1339  *  @adapter: board private structure to initialize
1340  *
1341  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1342  *  return -ENOMEM.
1343  **/
1344 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1345 {
1346 	int q_vectors = adapter->num_q_vectors;
1347 	int rxr_remaining = adapter->num_rx_queues;
1348 	int txr_remaining = adapter->num_tx_queues;
1349 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1350 	int err;
1351 
1352 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
1353 		for (; rxr_remaining; v_idx++) {
1354 			err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1355 						 0, 0, 1, rxr_idx);
1356 
1357 			if (err)
1358 				goto err_out;
1359 
1360 			/* update counts and index */
1361 			rxr_remaining--;
1362 			rxr_idx++;
1363 		}
1364 	}
1365 
1366 	for (; v_idx < q_vectors; v_idx++) {
1367 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1368 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1369 
1370 		err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1371 					 tqpv, txr_idx, rqpv, rxr_idx);
1372 
1373 		if (err)
1374 			goto err_out;
1375 
1376 		/* update counts and index */
1377 		rxr_remaining -= rqpv;
1378 		txr_remaining -= tqpv;
1379 		rxr_idx++;
1380 		txr_idx++;
1381 	}
1382 
1383 	return 0;
1384 
1385 err_out:
1386 	adapter->num_tx_queues = 0;
1387 	adapter->num_rx_queues = 0;
1388 	adapter->num_q_vectors = 0;
1389 
1390 	while (v_idx--)
1391 		igb_free_q_vector(adapter, v_idx);
1392 
1393 	return -ENOMEM;
1394 }
1395 
1396 /**
1397  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1398  *  @adapter: board private structure to initialize
1399  *  @msix: boolean value of MSIX capability
1400  *
1401  *  This function initializes the interrupts and allocates all of the queues.
1402  **/
1403 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1404 {
1405 	struct pci_dev *pdev = adapter->pdev;
1406 	int err;
1407 
1408 	igb_set_interrupt_capability(adapter, msix);
1409 
1410 	err = igb_alloc_q_vectors(adapter);
1411 	if (err) {
1412 		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1413 		goto err_alloc_q_vectors;
1414 	}
1415 
1416 	igb_cache_ring_register(adapter);
1417 
1418 	return 0;
1419 
1420 err_alloc_q_vectors:
1421 	igb_reset_interrupt_capability(adapter);
1422 	return err;
1423 }
1424 
1425 /**
1426  *  igb_request_irq - initialize interrupts
1427  *  @adapter: board private structure to initialize
1428  *
1429  *  Attempts to configure interrupts using the best available
1430  *  capabilities of the hardware and kernel.
1431  **/
1432 static int igb_request_irq(struct igb_adapter *adapter)
1433 {
1434 	struct net_device *netdev = adapter->netdev;
1435 	struct pci_dev *pdev = adapter->pdev;
1436 	int err = 0;
1437 
1438 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1439 		err = igb_request_msix(adapter);
1440 		if (!err)
1441 			goto request_done;
1442 		/* fall back to MSI */
1443 		igb_free_all_tx_resources(adapter);
1444 		igb_free_all_rx_resources(adapter);
1445 
1446 		igb_clear_interrupt_scheme(adapter);
1447 		err = igb_init_interrupt_scheme(adapter, false);
1448 		if (err)
1449 			goto request_done;
1450 
1451 		igb_setup_all_tx_resources(adapter);
1452 		igb_setup_all_rx_resources(adapter);
1453 		igb_configure(adapter);
1454 	}
1455 
1456 	igb_assign_vector(adapter->q_vector[0], 0);
1457 
1458 	if (adapter->flags & IGB_FLAG_HAS_MSI) {
1459 		err = request_irq(pdev->irq, igb_intr_msi, 0,
1460 				  netdev->name, adapter);
1461 		if (!err)
1462 			goto request_done;
1463 
1464 		/* fall back to legacy interrupts */
1465 		igb_reset_interrupt_capability(adapter);
1466 		adapter->flags &= ~IGB_FLAG_HAS_MSI;
1467 	}
1468 
1469 	err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1470 			  netdev->name, adapter);
1471 
1472 	if (err)
1473 		dev_err(&pdev->dev, "Error %d getting interrupt\n",
1474 			err);
1475 
1476 request_done:
1477 	return err;
1478 }
1479 
1480 static void igb_free_irq(struct igb_adapter *adapter)
1481 {
1482 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1483 		int vector = 0, i;
1484 
1485 		free_irq(adapter->msix_entries[vector++].vector, adapter);
1486 
1487 		for (i = 0; i < adapter->num_q_vectors; i++)
1488 			free_irq(adapter->msix_entries[vector++].vector,
1489 				 adapter->q_vector[i]);
1490 	} else {
1491 		free_irq(adapter->pdev->irq, adapter);
1492 	}
1493 }
1494 
1495 /**
1496  *  igb_irq_disable - Mask off interrupt generation on the NIC
1497  *  @adapter: board private structure
1498  **/
1499 static void igb_irq_disable(struct igb_adapter *adapter)
1500 {
1501 	struct e1000_hw *hw = &adapter->hw;
1502 
1503 	/* we need to be careful when disabling interrupts.  The VFs are also
1504 	 * mapped into these registers and so clearing the bits can cause
1505 	 * issues on the VF drivers so we only need to clear what we set
1506 	 */
1507 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1508 		u32 regval = rd32(E1000_EIAM);
1509 
1510 		wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1511 		wr32(E1000_EIMC, adapter->eims_enable_mask);
1512 		regval = rd32(E1000_EIAC);
1513 		wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1514 	}
1515 
1516 	wr32(E1000_IAM, 0);
1517 	wr32(E1000_IMC, ~0);
1518 	wrfl();
1519 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1520 		int i;
1521 
1522 		for (i = 0; i < adapter->num_q_vectors; i++)
1523 			synchronize_irq(adapter->msix_entries[i].vector);
1524 	} else {
1525 		synchronize_irq(adapter->pdev->irq);
1526 	}
1527 }
1528 
1529 /**
1530  *  igb_irq_enable - Enable default interrupt generation settings
1531  *  @adapter: board private structure
1532  **/
1533 static void igb_irq_enable(struct igb_adapter *adapter)
1534 {
1535 	struct e1000_hw *hw = &adapter->hw;
1536 
1537 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1538 		u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1539 		u32 regval = rd32(E1000_EIAC);
1540 
1541 		wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1542 		regval = rd32(E1000_EIAM);
1543 		wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1544 		wr32(E1000_EIMS, adapter->eims_enable_mask);
1545 		if (adapter->vfs_allocated_count) {
1546 			wr32(E1000_MBVFIMR, 0xFF);
1547 			ims |= E1000_IMS_VMMB;
1548 		}
1549 		wr32(E1000_IMS, ims);
1550 	} else {
1551 		wr32(E1000_IMS, IMS_ENABLE_MASK |
1552 				E1000_IMS_DRSTA);
1553 		wr32(E1000_IAM, IMS_ENABLE_MASK |
1554 				E1000_IMS_DRSTA);
1555 	}
1556 }
1557 
1558 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1559 {
1560 	struct e1000_hw *hw = &adapter->hw;
1561 	u16 pf_id = adapter->vfs_allocated_count;
1562 	u16 vid = adapter->hw.mng_cookie.vlan_id;
1563 	u16 old_vid = adapter->mng_vlan_id;
1564 
1565 	if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1566 		/* add VID to filter table */
1567 		igb_vfta_set(hw, vid, pf_id, true, true);
1568 		adapter->mng_vlan_id = vid;
1569 	} else {
1570 		adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1571 	}
1572 
1573 	if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1574 	    (vid != old_vid) &&
1575 	    !test_bit(old_vid, adapter->active_vlans)) {
1576 		/* remove VID from filter table */
1577 		igb_vfta_set(hw, vid, pf_id, false, true);
1578 	}
1579 }
1580 
1581 /**
1582  *  igb_release_hw_control - release control of the h/w to f/w
1583  *  @adapter: address of board private structure
1584  *
1585  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1586  *  For ASF and Pass Through versions of f/w this means that the
1587  *  driver is no longer loaded.
1588  **/
1589 static void igb_release_hw_control(struct igb_adapter *adapter)
1590 {
1591 	struct e1000_hw *hw = &adapter->hw;
1592 	u32 ctrl_ext;
1593 
1594 	/* Let firmware take over control of h/w */
1595 	ctrl_ext = rd32(E1000_CTRL_EXT);
1596 	wr32(E1000_CTRL_EXT,
1597 			ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1598 }
1599 
1600 /**
1601  *  igb_get_hw_control - get control of the h/w from f/w
1602  *  @adapter: address of board private structure
1603  *
1604  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1605  *  For ASF and Pass Through versions of f/w this means that
1606  *  the driver is loaded.
1607  **/
1608 static void igb_get_hw_control(struct igb_adapter *adapter)
1609 {
1610 	struct e1000_hw *hw = &adapter->hw;
1611 	u32 ctrl_ext;
1612 
1613 	/* Let firmware know the driver has taken over */
1614 	ctrl_ext = rd32(E1000_CTRL_EXT);
1615 	wr32(E1000_CTRL_EXT,
1616 			ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1617 }
1618 
1619 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1620 {
1621 	struct net_device *netdev = adapter->netdev;
1622 	struct e1000_hw *hw = &adapter->hw;
1623 
1624 	WARN_ON(hw->mac.type != e1000_i210);
1625 
1626 	if (enable)
1627 		adapter->flags |= IGB_FLAG_FQTSS;
1628 	else
1629 		adapter->flags &= ~IGB_FLAG_FQTSS;
1630 
1631 	if (netif_running(netdev))
1632 		schedule_work(&adapter->reset_task);
1633 }
1634 
1635 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1636 {
1637 	return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1638 }
1639 
1640 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1641 				   enum tx_queue_prio prio)
1642 {
1643 	u32 val;
1644 
1645 	WARN_ON(hw->mac.type != e1000_i210);
1646 	WARN_ON(queue < 0 || queue > 4);
1647 
1648 	val = rd32(E1000_I210_TXDCTL(queue));
1649 
1650 	if (prio == TX_QUEUE_PRIO_HIGH)
1651 		val |= E1000_TXDCTL_PRIORITY;
1652 	else
1653 		val &= ~E1000_TXDCTL_PRIORITY;
1654 
1655 	wr32(E1000_I210_TXDCTL(queue), val);
1656 }
1657 
1658 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1659 {
1660 	u32 val;
1661 
1662 	WARN_ON(hw->mac.type != e1000_i210);
1663 	WARN_ON(queue < 0 || queue > 1);
1664 
1665 	val = rd32(E1000_I210_TQAVCC(queue));
1666 
1667 	if (mode == QUEUE_MODE_STREAM_RESERVATION)
1668 		val |= E1000_TQAVCC_QUEUEMODE;
1669 	else
1670 		val &= ~E1000_TQAVCC_QUEUEMODE;
1671 
1672 	wr32(E1000_I210_TQAVCC(queue), val);
1673 }
1674 
1675 /**
1676  *  igb_configure_cbs - Configure Credit-Based Shaper (CBS)
1677  *  @adapter: pointer to adapter struct
1678  *  @queue: queue number
1679  *  @enable: true = enable CBS, false = disable CBS
1680  *  @idleslope: idleSlope in kbps
1681  *  @sendslope: sendSlope in kbps
1682  *  @hicredit: hiCredit in bytes
1683  *  @locredit: loCredit in bytes
1684  *
1685  *  Configure CBS for a given hardware queue. When disabling, idleslope,
1686  *  sendslope, hicredit, locredit arguments are ignored. Returns 0 if
1687  *  success. Negative otherwise.
1688  **/
1689 static void igb_configure_cbs(struct igb_adapter *adapter, int queue,
1690 			      bool enable, int idleslope, int sendslope,
1691 			      int hicredit, int locredit)
1692 {
1693 	struct net_device *netdev = adapter->netdev;
1694 	struct e1000_hw *hw = &adapter->hw;
1695 	u32 tqavcc;
1696 	u16 value;
1697 
1698 	WARN_ON(hw->mac.type != e1000_i210);
1699 	WARN_ON(queue < 0 || queue > 1);
1700 
1701 	if (enable) {
1702 		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1703 		set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1704 
1705 		/* According to i210 datasheet section 7.2.7.7, we should set
1706 		 * the 'idleSlope' field from TQAVCC register following the
1707 		 * equation:
1708 		 *
1709 		 * For 100 Mbps link speed:
1710 		 *
1711 		 *     value = BW * 0x7735 * 0.2                          (E1)
1712 		 *
1713 		 * For 1000Mbps link speed:
1714 		 *
1715 		 *     value = BW * 0x7735 * 2                            (E2)
1716 		 *
1717 		 * E1 and E2 can be merged into one equation as shown below.
1718 		 * Note that 'link-speed' is in Mbps.
1719 		 *
1720 		 *     value = BW * 0x7735 * 2 * link-speed
1721 		 *                           --------------               (E3)
1722 		 *                                1000
1723 		 *
1724 		 * 'BW' is the percentage bandwidth out of full link speed
1725 		 * which can be found with the following equation. Note that
1726 		 * idleSlope here is the parameter from this function which
1727 		 * is in kbps.
1728 		 *
1729 		 *     BW =     idleSlope
1730 		 *          -----------------                             (E4)
1731 		 *          link-speed * 1000
1732 		 *
1733 		 * That said, we can come up with a generic equation to
1734 		 * calculate the value we should set it TQAVCC register by
1735 		 * replacing 'BW' in E3 by E4. The resulting equation is:
1736 		 *
1737 		 * value =     idleSlope     * 0x7735 * 2 * link-speed
1738 		 *         -----------------            --------------    (E5)
1739 		 *         link-speed * 1000                 1000
1740 		 *
1741 		 * 'link-speed' is present in both sides of the fraction so
1742 		 * it is canceled out. The final equation is the following:
1743 		 *
1744 		 *     value = idleSlope * 61034
1745 		 *             -----------------                          (E6)
1746 		 *                  1000000
1747 		 *
1748 		 * NOTE: For i210, given the above, we can see that idleslope
1749 		 *       is represented in 16.38431 kbps units by the value at
1750 		 *       the TQAVCC register (1Gbps / 61034), which reduces
1751 		 *       the granularity for idleslope increments.
1752 		 *       For instance, if you want to configure a 2576kbps
1753 		 *       idleslope, the value to be written on the register
1754 		 *       would have to be 157.23. If rounded down, you end
1755 		 *       up with less bandwidth available than originally
1756 		 *       required (~2572 kbps). If rounded up, you end up
1757 		 *       with a higher bandwidth (~2589 kbps). Below the
1758 		 *       approach we take is to always round up the
1759 		 *       calculated value, so the resulting bandwidth might
1760 		 *       be slightly higher for some configurations.
1761 		 */
1762 		value = DIV_ROUND_UP_ULL(idleslope * 61034ULL, 1000000);
1763 
1764 		tqavcc = rd32(E1000_I210_TQAVCC(queue));
1765 		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1766 		tqavcc |= value;
1767 		wr32(E1000_I210_TQAVCC(queue), tqavcc);
1768 
1769 		wr32(E1000_I210_TQAVHC(queue), 0x80000000 + hicredit * 0x7735);
1770 	} else {
1771 		set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1772 		set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1773 
1774 		/* Set idleSlope to zero. */
1775 		tqavcc = rd32(E1000_I210_TQAVCC(queue));
1776 		tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1777 		wr32(E1000_I210_TQAVCC(queue), tqavcc);
1778 
1779 		/* Set hiCredit to zero. */
1780 		wr32(E1000_I210_TQAVHC(queue), 0);
1781 	}
1782 
1783 	/* XXX: In i210 controller the sendSlope and loCredit parameters from
1784 	 * CBS are not configurable by software so we don't do any 'controller
1785 	 * configuration' in respect to these parameters.
1786 	 */
1787 
1788 	netdev_dbg(netdev, "CBS %s: queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1789 		   (enable) ? "enabled" : "disabled", queue,
1790 		   idleslope, sendslope, hicredit, locredit);
1791 }
1792 
1793 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1794 			       bool enable, int idleslope, int sendslope,
1795 			       int hicredit, int locredit)
1796 {
1797 	struct igb_ring *ring;
1798 
1799 	if (queue < 0 || queue > adapter->num_tx_queues)
1800 		return -EINVAL;
1801 
1802 	ring = adapter->tx_ring[queue];
1803 
1804 	ring->cbs_enable = enable;
1805 	ring->idleslope = idleslope;
1806 	ring->sendslope = sendslope;
1807 	ring->hicredit = hicredit;
1808 	ring->locredit = locredit;
1809 
1810 	return 0;
1811 }
1812 
1813 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1814 {
1815 	struct igb_ring *ring;
1816 	int i;
1817 
1818 	for (i = 0; i < adapter->num_tx_queues; i++) {
1819 		ring = adapter->tx_ring[i];
1820 
1821 		if (ring->cbs_enable)
1822 			return true;
1823 	}
1824 
1825 	return false;
1826 }
1827 
1828 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1829 {
1830 	struct net_device *netdev = adapter->netdev;
1831 	struct e1000_hw *hw = &adapter->hw;
1832 	u32 val;
1833 
1834 	/* Only i210 controller supports changing the transmission mode. */
1835 	if (hw->mac.type != e1000_i210)
1836 		return;
1837 
1838 	if (is_fqtss_enabled(adapter)) {
1839 		int i, max_queue;
1840 
1841 		/* Configure TQAVCTRL register: set transmit mode to 'Qav',
1842 		 * set data fetch arbitration to 'round robin' and set data
1843 		 * transfer arbitration to 'credit shaper algorithm.
1844 		 */
1845 		val = rd32(E1000_I210_TQAVCTRL);
1846 		val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_DATATRANARB;
1847 		val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1848 		wr32(E1000_I210_TQAVCTRL, val);
1849 
1850 		/* Configure Tx and Rx packet buffers sizes as described in
1851 		 * i210 datasheet section 7.2.7.7.
1852 		 */
1853 		val = rd32(E1000_TXPBS);
1854 		val &= ~I210_TXPBSIZE_MASK;
1855 		val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB |
1856 			I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB;
1857 		wr32(E1000_TXPBS, val);
1858 
1859 		val = rd32(E1000_RXPBS);
1860 		val &= ~I210_RXPBSIZE_MASK;
1861 		val |= I210_RXPBSIZE_PB_32KB;
1862 		wr32(E1000_RXPBS, val);
1863 
1864 		/* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1865 		 * register should not exceed the buffer size programmed in
1866 		 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1867 		 * so according to the datasheet we should set MAX_TPKT_SIZE to
1868 		 * 4kB / 64.
1869 		 *
1870 		 * However, when we do so, no frame from queue 2 and 3 are
1871 		 * transmitted.  It seems the MAX_TPKT_SIZE should not be great
1872 		 * or _equal_ to the buffer size programmed in TXPBS. For this
1873 		 * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1874 		 */
1875 		val = (4096 - 1) / 64;
1876 		wr32(E1000_I210_DTXMXPKTSZ, val);
1877 
1878 		/* Since FQTSS mode is enabled, apply any CBS configuration
1879 		 * previously set. If no previous CBS configuration has been
1880 		 * done, then the initial configuration is applied, which means
1881 		 * CBS is disabled.
1882 		 */
1883 		max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1884 			    adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1885 
1886 		for (i = 0; i < max_queue; i++) {
1887 			struct igb_ring *ring = adapter->tx_ring[i];
1888 
1889 			igb_configure_cbs(adapter, i, ring->cbs_enable,
1890 					  ring->idleslope, ring->sendslope,
1891 					  ring->hicredit, ring->locredit);
1892 		}
1893 	} else {
1894 		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1895 		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1896 		wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1897 
1898 		val = rd32(E1000_I210_TQAVCTRL);
1899 		/* According to Section 8.12.21, the other flags we've set when
1900 		 * enabling FQTSS are not relevant when disabling FQTSS so we
1901 		 * don't set they here.
1902 		 */
1903 		val &= ~E1000_TQAVCTRL_XMIT_MODE;
1904 		wr32(E1000_I210_TQAVCTRL, val);
1905 	}
1906 
1907 	netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1908 		   "enabled" : "disabled");
1909 }
1910 
1911 /**
1912  *  igb_configure - configure the hardware for RX and TX
1913  *  @adapter: private board structure
1914  **/
1915 static void igb_configure(struct igb_adapter *adapter)
1916 {
1917 	struct net_device *netdev = adapter->netdev;
1918 	int i;
1919 
1920 	igb_get_hw_control(adapter);
1921 	igb_set_rx_mode(netdev);
1922 	igb_setup_tx_mode(adapter);
1923 
1924 	igb_restore_vlan(adapter);
1925 
1926 	igb_setup_tctl(adapter);
1927 	igb_setup_mrqc(adapter);
1928 	igb_setup_rctl(adapter);
1929 
1930 	igb_nfc_filter_restore(adapter);
1931 	igb_configure_tx(adapter);
1932 	igb_configure_rx(adapter);
1933 
1934 	igb_rx_fifo_flush_82575(&adapter->hw);
1935 
1936 	/* call igb_desc_unused which always leaves
1937 	 * at least 1 descriptor unused to make sure
1938 	 * next_to_use != next_to_clean
1939 	 */
1940 	for (i = 0; i < adapter->num_rx_queues; i++) {
1941 		struct igb_ring *ring = adapter->rx_ring[i];
1942 		igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1943 	}
1944 }
1945 
1946 /**
1947  *  igb_power_up_link - Power up the phy/serdes link
1948  *  @adapter: address of board private structure
1949  **/
1950 void igb_power_up_link(struct igb_adapter *adapter)
1951 {
1952 	igb_reset_phy(&adapter->hw);
1953 
1954 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1955 		igb_power_up_phy_copper(&adapter->hw);
1956 	else
1957 		igb_power_up_serdes_link_82575(&adapter->hw);
1958 
1959 	igb_setup_link(&adapter->hw);
1960 }
1961 
1962 /**
1963  *  igb_power_down_link - Power down the phy/serdes link
1964  *  @adapter: address of board private structure
1965  */
1966 static void igb_power_down_link(struct igb_adapter *adapter)
1967 {
1968 	if (adapter->hw.phy.media_type == e1000_media_type_copper)
1969 		igb_power_down_phy_copper_82575(&adapter->hw);
1970 	else
1971 		igb_shutdown_serdes_link_82575(&adapter->hw);
1972 }
1973 
1974 /**
1975  * Detect and switch function for Media Auto Sense
1976  * @adapter: address of the board private structure
1977  **/
1978 static void igb_check_swap_media(struct igb_adapter *adapter)
1979 {
1980 	struct e1000_hw *hw = &adapter->hw;
1981 	u32 ctrl_ext, connsw;
1982 	bool swap_now = false;
1983 
1984 	ctrl_ext = rd32(E1000_CTRL_EXT);
1985 	connsw = rd32(E1000_CONNSW);
1986 
1987 	/* need to live swap if current media is copper and we have fiber/serdes
1988 	 * to go to.
1989 	 */
1990 
1991 	if ((hw->phy.media_type == e1000_media_type_copper) &&
1992 	    (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1993 		swap_now = true;
1994 	} else if (!(connsw & E1000_CONNSW_SERDESD)) {
1995 		/* copper signal takes time to appear */
1996 		if (adapter->copper_tries < 4) {
1997 			adapter->copper_tries++;
1998 			connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1999 			wr32(E1000_CONNSW, connsw);
2000 			return;
2001 		} else {
2002 			adapter->copper_tries = 0;
2003 			if ((connsw & E1000_CONNSW_PHYSD) &&
2004 			    (!(connsw & E1000_CONNSW_PHY_PDN))) {
2005 				swap_now = true;
2006 				connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2007 				wr32(E1000_CONNSW, connsw);
2008 			}
2009 		}
2010 	}
2011 
2012 	if (!swap_now)
2013 		return;
2014 
2015 	switch (hw->phy.media_type) {
2016 	case e1000_media_type_copper:
2017 		netdev_info(adapter->netdev,
2018 			"MAS: changing media to fiber/serdes\n");
2019 		ctrl_ext |=
2020 			E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2021 		adapter->flags |= IGB_FLAG_MEDIA_RESET;
2022 		adapter->copper_tries = 0;
2023 		break;
2024 	case e1000_media_type_internal_serdes:
2025 	case e1000_media_type_fiber:
2026 		netdev_info(adapter->netdev,
2027 			"MAS: changing media to copper\n");
2028 		ctrl_ext &=
2029 			~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2030 		adapter->flags |= IGB_FLAG_MEDIA_RESET;
2031 		break;
2032 	default:
2033 		/* shouldn't get here during regular operation */
2034 		netdev_err(adapter->netdev,
2035 			"AMS: Invalid media type found, returning\n");
2036 		break;
2037 	}
2038 	wr32(E1000_CTRL_EXT, ctrl_ext);
2039 }
2040 
2041 /**
2042  *  igb_up - Open the interface and prepare it to handle traffic
2043  *  @adapter: board private structure
2044  **/
2045 int igb_up(struct igb_adapter *adapter)
2046 {
2047 	struct e1000_hw *hw = &adapter->hw;
2048 	int i;
2049 
2050 	/* hardware has been reset, we need to reload some things */
2051 	igb_configure(adapter);
2052 
2053 	clear_bit(__IGB_DOWN, &adapter->state);
2054 
2055 	for (i = 0; i < adapter->num_q_vectors; i++)
2056 		napi_enable(&(adapter->q_vector[i]->napi));
2057 
2058 	if (adapter->flags & IGB_FLAG_HAS_MSIX)
2059 		igb_configure_msix(adapter);
2060 	else
2061 		igb_assign_vector(adapter->q_vector[0], 0);
2062 
2063 	/* Clear any pending interrupts. */
2064 	rd32(E1000_ICR);
2065 	igb_irq_enable(adapter);
2066 
2067 	/* notify VFs that reset has been completed */
2068 	if (adapter->vfs_allocated_count) {
2069 		u32 reg_data = rd32(E1000_CTRL_EXT);
2070 
2071 		reg_data |= E1000_CTRL_EXT_PFRSTD;
2072 		wr32(E1000_CTRL_EXT, reg_data);
2073 	}
2074 
2075 	netif_tx_start_all_queues(adapter->netdev);
2076 
2077 	/* start the watchdog. */
2078 	hw->mac.get_link_status = 1;
2079 	schedule_work(&adapter->watchdog_task);
2080 
2081 	if ((adapter->flags & IGB_FLAG_EEE) &&
2082 	    (!hw->dev_spec._82575.eee_disable))
2083 		adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2084 
2085 	return 0;
2086 }
2087 
2088 void igb_down(struct igb_adapter *adapter)
2089 {
2090 	struct net_device *netdev = adapter->netdev;
2091 	struct e1000_hw *hw = &adapter->hw;
2092 	u32 tctl, rctl;
2093 	int i;
2094 
2095 	/* signal that we're down so the interrupt handler does not
2096 	 * reschedule our watchdog timer
2097 	 */
2098 	set_bit(__IGB_DOWN, &adapter->state);
2099 
2100 	/* disable receives in the hardware */
2101 	rctl = rd32(E1000_RCTL);
2102 	wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2103 	/* flush and sleep below */
2104 
2105 	igb_nfc_filter_exit(adapter);
2106 
2107 	netif_carrier_off(netdev);
2108 	netif_tx_stop_all_queues(netdev);
2109 
2110 	/* disable transmits in the hardware */
2111 	tctl = rd32(E1000_TCTL);
2112 	tctl &= ~E1000_TCTL_EN;
2113 	wr32(E1000_TCTL, tctl);
2114 	/* flush both disables and wait for them to finish */
2115 	wrfl();
2116 	usleep_range(10000, 11000);
2117 
2118 	igb_irq_disable(adapter);
2119 
2120 	adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2121 
2122 	for (i = 0; i < adapter->num_q_vectors; i++) {
2123 		if (adapter->q_vector[i]) {
2124 			napi_synchronize(&adapter->q_vector[i]->napi);
2125 			napi_disable(&adapter->q_vector[i]->napi);
2126 		}
2127 	}
2128 
2129 	del_timer_sync(&adapter->watchdog_timer);
2130 	del_timer_sync(&adapter->phy_info_timer);
2131 
2132 	/* record the stats before reset*/
2133 	spin_lock(&adapter->stats64_lock);
2134 	igb_update_stats(adapter);
2135 	spin_unlock(&adapter->stats64_lock);
2136 
2137 	adapter->link_speed = 0;
2138 	adapter->link_duplex = 0;
2139 
2140 	if (!pci_channel_offline(adapter->pdev))
2141 		igb_reset(adapter);
2142 
2143 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
2144 	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2145 
2146 	igb_clean_all_tx_rings(adapter);
2147 	igb_clean_all_rx_rings(adapter);
2148 #ifdef CONFIG_IGB_DCA
2149 
2150 	/* since we reset the hardware DCA settings were cleared */
2151 	igb_setup_dca(adapter);
2152 #endif
2153 }
2154 
2155 void igb_reinit_locked(struct igb_adapter *adapter)
2156 {
2157 	WARN_ON(in_interrupt());
2158 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2159 		usleep_range(1000, 2000);
2160 	igb_down(adapter);
2161 	igb_up(adapter);
2162 	clear_bit(__IGB_RESETTING, &adapter->state);
2163 }
2164 
2165 /** igb_enable_mas - Media Autosense re-enable after swap
2166  *
2167  * @adapter: adapter struct
2168  **/
2169 static void igb_enable_mas(struct igb_adapter *adapter)
2170 {
2171 	struct e1000_hw *hw = &adapter->hw;
2172 	u32 connsw = rd32(E1000_CONNSW);
2173 
2174 	/* configure for SerDes media detect */
2175 	if ((hw->phy.media_type == e1000_media_type_copper) &&
2176 	    (!(connsw & E1000_CONNSW_SERDESD))) {
2177 		connsw |= E1000_CONNSW_ENRGSRC;
2178 		connsw |= E1000_CONNSW_AUTOSENSE_EN;
2179 		wr32(E1000_CONNSW, connsw);
2180 		wrfl();
2181 	}
2182 }
2183 
2184 void igb_reset(struct igb_adapter *adapter)
2185 {
2186 	struct pci_dev *pdev = adapter->pdev;
2187 	struct e1000_hw *hw = &adapter->hw;
2188 	struct e1000_mac_info *mac = &hw->mac;
2189 	struct e1000_fc_info *fc = &hw->fc;
2190 	u32 pba, hwm;
2191 
2192 	/* Repartition Pba for greater than 9k mtu
2193 	 * To take effect CTRL.RST is required.
2194 	 */
2195 	switch (mac->type) {
2196 	case e1000_i350:
2197 	case e1000_i354:
2198 	case e1000_82580:
2199 		pba = rd32(E1000_RXPBS);
2200 		pba = igb_rxpbs_adjust_82580(pba);
2201 		break;
2202 	case e1000_82576:
2203 		pba = rd32(E1000_RXPBS);
2204 		pba &= E1000_RXPBS_SIZE_MASK_82576;
2205 		break;
2206 	case e1000_82575:
2207 	case e1000_i210:
2208 	case e1000_i211:
2209 	default:
2210 		pba = E1000_PBA_34K;
2211 		break;
2212 	}
2213 
2214 	if (mac->type == e1000_82575) {
2215 		u32 min_rx_space, min_tx_space, needed_tx_space;
2216 
2217 		/* write Rx PBA so that hardware can report correct Tx PBA */
2218 		wr32(E1000_PBA, pba);
2219 
2220 		/* To maintain wire speed transmits, the Tx FIFO should be
2221 		 * large enough to accommodate two full transmit packets,
2222 		 * rounded up to the next 1KB and expressed in KB.  Likewise,
2223 		 * the Rx FIFO should be large enough to accommodate at least
2224 		 * one full receive packet and is similarly rounded up and
2225 		 * expressed in KB.
2226 		 */
2227 		min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2228 
2229 		/* The Tx FIFO also stores 16 bytes of information about the Tx
2230 		 * but don't include Ethernet FCS because hardware appends it.
2231 		 * We only need to round down to the nearest 512 byte block
2232 		 * count since the value we care about is 2 frames, not 1.
2233 		 */
2234 		min_tx_space = adapter->max_frame_size;
2235 		min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2236 		min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2237 
2238 		/* upper 16 bits has Tx packet buffer allocation size in KB */
2239 		needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2240 
2241 		/* If current Tx allocation is less than the min Tx FIFO size,
2242 		 * and the min Tx FIFO size is less than the current Rx FIFO
2243 		 * allocation, take space away from current Rx allocation.
2244 		 */
2245 		if (needed_tx_space < pba) {
2246 			pba -= needed_tx_space;
2247 
2248 			/* if short on Rx space, Rx wins and must trump Tx
2249 			 * adjustment
2250 			 */
2251 			if (pba < min_rx_space)
2252 				pba = min_rx_space;
2253 		}
2254 
2255 		/* adjust PBA for jumbo frames */
2256 		wr32(E1000_PBA, pba);
2257 	}
2258 
2259 	/* flow control settings
2260 	 * The high water mark must be low enough to fit one full frame
2261 	 * after transmitting the pause frame.  As such we must have enough
2262 	 * space to allow for us to complete our current transmit and then
2263 	 * receive the frame that is in progress from the link partner.
2264 	 * Set it to:
2265 	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2266 	 */
2267 	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2268 
2269 	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
2270 	fc->low_water = fc->high_water - 16;
2271 	fc->pause_time = 0xFFFF;
2272 	fc->send_xon = 1;
2273 	fc->current_mode = fc->requested_mode;
2274 
2275 	/* disable receive for all VFs and wait one second */
2276 	if (adapter->vfs_allocated_count) {
2277 		int i;
2278 
2279 		for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2280 			adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2281 
2282 		/* ping all the active vfs to let them know we are going down */
2283 		igb_ping_all_vfs(adapter);
2284 
2285 		/* disable transmits and receives */
2286 		wr32(E1000_VFRE, 0);
2287 		wr32(E1000_VFTE, 0);
2288 	}
2289 
2290 	/* Allow time for pending master requests to run */
2291 	hw->mac.ops.reset_hw(hw);
2292 	wr32(E1000_WUC, 0);
2293 
2294 	if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2295 		/* need to resetup here after media swap */
2296 		adapter->ei.get_invariants(hw);
2297 		adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2298 	}
2299 	if ((mac->type == e1000_82575) &&
2300 	    (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2301 		igb_enable_mas(adapter);
2302 	}
2303 	if (hw->mac.ops.init_hw(hw))
2304 		dev_err(&pdev->dev, "Hardware Error\n");
2305 
2306 	/* RAR registers were cleared during init_hw, clear mac table */
2307 	igb_flush_mac_table(adapter);
2308 	__dev_uc_unsync(adapter->netdev, NULL);
2309 
2310 	/* Recover default RAR entry */
2311 	igb_set_default_mac_filter(adapter);
2312 
2313 	/* Flow control settings reset on hardware reset, so guarantee flow
2314 	 * control is off when forcing speed.
2315 	 */
2316 	if (!hw->mac.autoneg)
2317 		igb_force_mac_fc(hw);
2318 
2319 	igb_init_dmac(adapter, pba);
2320 #ifdef CONFIG_IGB_HWMON
2321 	/* Re-initialize the thermal sensor on i350 devices. */
2322 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
2323 		if (mac->type == e1000_i350 && hw->bus.func == 0) {
2324 			/* If present, re-initialize the external thermal sensor
2325 			 * interface.
2326 			 */
2327 			if (adapter->ets)
2328 				mac->ops.init_thermal_sensor_thresh(hw);
2329 		}
2330 	}
2331 #endif
2332 	/* Re-establish EEE setting */
2333 	if (hw->phy.media_type == e1000_media_type_copper) {
2334 		switch (mac->type) {
2335 		case e1000_i350:
2336 		case e1000_i210:
2337 		case e1000_i211:
2338 			igb_set_eee_i350(hw, true, true);
2339 			break;
2340 		case e1000_i354:
2341 			igb_set_eee_i354(hw, true, true);
2342 			break;
2343 		default:
2344 			break;
2345 		}
2346 	}
2347 	if (!netif_running(adapter->netdev))
2348 		igb_power_down_link(adapter);
2349 
2350 	igb_update_mng_vlan(adapter);
2351 
2352 	/* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2353 	wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2354 
2355 	/* Re-enable PTP, where applicable. */
2356 	if (adapter->ptp_flags & IGB_PTP_ENABLED)
2357 		igb_ptp_reset(adapter);
2358 
2359 	igb_get_phy_info(hw);
2360 }
2361 
2362 static netdev_features_t igb_fix_features(struct net_device *netdev,
2363 	netdev_features_t features)
2364 {
2365 	/* Since there is no support for separate Rx/Tx vlan accel
2366 	 * enable/disable make sure Tx flag is always in same state as Rx.
2367 	 */
2368 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2369 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2370 	else
2371 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2372 
2373 	return features;
2374 }
2375 
2376 static int igb_set_features(struct net_device *netdev,
2377 	netdev_features_t features)
2378 {
2379 	netdev_features_t changed = netdev->features ^ features;
2380 	struct igb_adapter *adapter = netdev_priv(netdev);
2381 
2382 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2383 		igb_vlan_mode(netdev, features);
2384 
2385 	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2386 		return 0;
2387 
2388 	if (!(features & NETIF_F_NTUPLE)) {
2389 		struct hlist_node *node2;
2390 		struct igb_nfc_filter *rule;
2391 
2392 		spin_lock(&adapter->nfc_lock);
2393 		hlist_for_each_entry_safe(rule, node2,
2394 					  &adapter->nfc_filter_list, nfc_node) {
2395 			igb_erase_filter(adapter, rule);
2396 			hlist_del(&rule->nfc_node);
2397 			kfree(rule);
2398 		}
2399 		spin_unlock(&adapter->nfc_lock);
2400 		adapter->nfc_filter_count = 0;
2401 	}
2402 
2403 	netdev->features = features;
2404 
2405 	if (netif_running(netdev))
2406 		igb_reinit_locked(adapter);
2407 	else
2408 		igb_reset(adapter);
2409 
2410 	return 0;
2411 }
2412 
2413 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2414 			   struct net_device *dev,
2415 			   const unsigned char *addr, u16 vid,
2416 			   u16 flags)
2417 {
2418 	/* guarantee we can provide a unique filter for the unicast address */
2419 	if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2420 		struct igb_adapter *adapter = netdev_priv(dev);
2421 		int vfn = adapter->vfs_allocated_count;
2422 
2423 		if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2424 			return -ENOMEM;
2425 	}
2426 
2427 	return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2428 }
2429 
2430 #define IGB_MAX_MAC_HDR_LEN	127
2431 #define IGB_MAX_NETWORK_HDR_LEN	511
2432 
2433 static netdev_features_t
2434 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2435 		   netdev_features_t features)
2436 {
2437 	unsigned int network_hdr_len, mac_hdr_len;
2438 
2439 	/* Make certain the headers can be described by a context descriptor */
2440 	mac_hdr_len = skb_network_header(skb) - skb->data;
2441 	if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2442 		return features & ~(NETIF_F_HW_CSUM |
2443 				    NETIF_F_SCTP_CRC |
2444 				    NETIF_F_HW_VLAN_CTAG_TX |
2445 				    NETIF_F_TSO |
2446 				    NETIF_F_TSO6);
2447 
2448 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2449 	if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2450 		return features & ~(NETIF_F_HW_CSUM |
2451 				    NETIF_F_SCTP_CRC |
2452 				    NETIF_F_TSO |
2453 				    NETIF_F_TSO6);
2454 
2455 	/* We can only support IPV4 TSO in tunnels if we can mangle the
2456 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2457 	 */
2458 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2459 		features &= ~NETIF_F_TSO;
2460 
2461 	return features;
2462 }
2463 
2464 static int igb_offload_cbs(struct igb_adapter *adapter,
2465 			   struct tc_cbs_qopt_offload *qopt)
2466 {
2467 	struct e1000_hw *hw = &adapter->hw;
2468 	int err;
2469 
2470 	/* CBS offloading is only supported by i210 controller. */
2471 	if (hw->mac.type != e1000_i210)
2472 		return -EOPNOTSUPP;
2473 
2474 	/* CBS offloading is only supported by queue 0 and queue 1. */
2475 	if (qopt->queue < 0 || qopt->queue > 1)
2476 		return -EINVAL;
2477 
2478 	err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2479 				  qopt->idleslope, qopt->sendslope,
2480 				  qopt->hicredit, qopt->locredit);
2481 	if (err)
2482 		return err;
2483 
2484 	if (is_fqtss_enabled(adapter)) {
2485 		igb_configure_cbs(adapter, qopt->queue, qopt->enable,
2486 				  qopt->idleslope, qopt->sendslope,
2487 				  qopt->hicredit, qopt->locredit);
2488 
2489 		if (!is_any_cbs_enabled(adapter))
2490 			enable_fqtss(adapter, false);
2491 
2492 	} else {
2493 		enable_fqtss(adapter, true);
2494 	}
2495 
2496 	return 0;
2497 }
2498 
2499 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2500 			void *type_data)
2501 {
2502 	struct igb_adapter *adapter = netdev_priv(dev);
2503 
2504 	switch (type) {
2505 	case TC_SETUP_QDISC_CBS:
2506 		return igb_offload_cbs(adapter, type_data);
2507 
2508 	default:
2509 		return -EOPNOTSUPP;
2510 	}
2511 }
2512 
2513 static const struct net_device_ops igb_netdev_ops = {
2514 	.ndo_open		= igb_open,
2515 	.ndo_stop		= igb_close,
2516 	.ndo_start_xmit		= igb_xmit_frame,
2517 	.ndo_get_stats64	= igb_get_stats64,
2518 	.ndo_set_rx_mode	= igb_set_rx_mode,
2519 	.ndo_set_mac_address	= igb_set_mac,
2520 	.ndo_change_mtu		= igb_change_mtu,
2521 	.ndo_do_ioctl		= igb_ioctl,
2522 	.ndo_tx_timeout		= igb_tx_timeout,
2523 	.ndo_validate_addr	= eth_validate_addr,
2524 	.ndo_vlan_rx_add_vid	= igb_vlan_rx_add_vid,
2525 	.ndo_vlan_rx_kill_vid	= igb_vlan_rx_kill_vid,
2526 	.ndo_set_vf_mac		= igb_ndo_set_vf_mac,
2527 	.ndo_set_vf_vlan	= igb_ndo_set_vf_vlan,
2528 	.ndo_set_vf_rate	= igb_ndo_set_vf_bw,
2529 	.ndo_set_vf_spoofchk	= igb_ndo_set_vf_spoofchk,
2530 	.ndo_get_vf_config	= igb_ndo_get_vf_config,
2531 #ifdef CONFIG_NET_POLL_CONTROLLER
2532 	.ndo_poll_controller	= igb_netpoll,
2533 #endif
2534 	.ndo_fix_features	= igb_fix_features,
2535 	.ndo_set_features	= igb_set_features,
2536 	.ndo_fdb_add		= igb_ndo_fdb_add,
2537 	.ndo_features_check	= igb_features_check,
2538 	.ndo_setup_tc		= igb_setup_tc,
2539 };
2540 
2541 /**
2542  * igb_set_fw_version - Configure version string for ethtool
2543  * @adapter: adapter struct
2544  **/
2545 void igb_set_fw_version(struct igb_adapter *adapter)
2546 {
2547 	struct e1000_hw *hw = &adapter->hw;
2548 	struct e1000_fw_version fw;
2549 
2550 	igb_get_fw_version(hw, &fw);
2551 
2552 	switch (hw->mac.type) {
2553 	case e1000_i210:
2554 	case e1000_i211:
2555 		if (!(igb_get_flash_presence_i210(hw))) {
2556 			snprintf(adapter->fw_version,
2557 				 sizeof(adapter->fw_version),
2558 				 "%2d.%2d-%d",
2559 				 fw.invm_major, fw.invm_minor,
2560 				 fw.invm_img_type);
2561 			break;
2562 		}
2563 		/* fall through */
2564 	default:
2565 		/* if option is rom valid, display its version too */
2566 		if (fw.or_valid) {
2567 			snprintf(adapter->fw_version,
2568 				 sizeof(adapter->fw_version),
2569 				 "%d.%d, 0x%08x, %d.%d.%d",
2570 				 fw.eep_major, fw.eep_minor, fw.etrack_id,
2571 				 fw.or_major, fw.or_build, fw.or_patch);
2572 		/* no option rom */
2573 		} else if (fw.etrack_id != 0X0000) {
2574 			snprintf(adapter->fw_version,
2575 			    sizeof(adapter->fw_version),
2576 			    "%d.%d, 0x%08x",
2577 			    fw.eep_major, fw.eep_minor, fw.etrack_id);
2578 		} else {
2579 		snprintf(adapter->fw_version,
2580 		    sizeof(adapter->fw_version),
2581 		    "%d.%d.%d",
2582 		    fw.eep_major, fw.eep_minor, fw.eep_build);
2583 		}
2584 		break;
2585 	}
2586 }
2587 
2588 /**
2589  * igb_init_mas - init Media Autosense feature if enabled in the NVM
2590  *
2591  * @adapter: adapter struct
2592  **/
2593 static void igb_init_mas(struct igb_adapter *adapter)
2594 {
2595 	struct e1000_hw *hw = &adapter->hw;
2596 	u16 eeprom_data;
2597 
2598 	hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2599 	switch (hw->bus.func) {
2600 	case E1000_FUNC_0:
2601 		if (eeprom_data & IGB_MAS_ENABLE_0) {
2602 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2603 			netdev_info(adapter->netdev,
2604 				"MAS: Enabling Media Autosense for port %d\n",
2605 				hw->bus.func);
2606 		}
2607 		break;
2608 	case E1000_FUNC_1:
2609 		if (eeprom_data & IGB_MAS_ENABLE_1) {
2610 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2611 			netdev_info(adapter->netdev,
2612 				"MAS: Enabling Media Autosense for port %d\n",
2613 				hw->bus.func);
2614 		}
2615 		break;
2616 	case E1000_FUNC_2:
2617 		if (eeprom_data & IGB_MAS_ENABLE_2) {
2618 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2619 			netdev_info(adapter->netdev,
2620 				"MAS: Enabling Media Autosense for port %d\n",
2621 				hw->bus.func);
2622 		}
2623 		break;
2624 	case E1000_FUNC_3:
2625 		if (eeprom_data & IGB_MAS_ENABLE_3) {
2626 			adapter->flags |= IGB_FLAG_MAS_ENABLE;
2627 			netdev_info(adapter->netdev,
2628 				"MAS: Enabling Media Autosense for port %d\n",
2629 				hw->bus.func);
2630 		}
2631 		break;
2632 	default:
2633 		/* Shouldn't get here */
2634 		netdev_err(adapter->netdev,
2635 			"MAS: Invalid port configuration, returning\n");
2636 		break;
2637 	}
2638 }
2639 
2640 /**
2641  *  igb_init_i2c - Init I2C interface
2642  *  @adapter: pointer to adapter structure
2643  **/
2644 static s32 igb_init_i2c(struct igb_adapter *adapter)
2645 {
2646 	s32 status = 0;
2647 
2648 	/* I2C interface supported on i350 devices */
2649 	if (adapter->hw.mac.type != e1000_i350)
2650 		return 0;
2651 
2652 	/* Initialize the i2c bus which is controlled by the registers.
2653 	 * This bus will use the i2c_algo_bit structue that implements
2654 	 * the protocol through toggling of the 4 bits in the register.
2655 	 */
2656 	adapter->i2c_adap.owner = THIS_MODULE;
2657 	adapter->i2c_algo = igb_i2c_algo;
2658 	adapter->i2c_algo.data = adapter;
2659 	adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2660 	adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2661 	strlcpy(adapter->i2c_adap.name, "igb BB",
2662 		sizeof(adapter->i2c_adap.name));
2663 	status = i2c_bit_add_bus(&adapter->i2c_adap);
2664 	return status;
2665 }
2666 
2667 /**
2668  *  igb_probe - Device Initialization Routine
2669  *  @pdev: PCI device information struct
2670  *  @ent: entry in igb_pci_tbl
2671  *
2672  *  Returns 0 on success, negative on failure
2673  *
2674  *  igb_probe initializes an adapter identified by a pci_dev structure.
2675  *  The OS initialization, configuring of the adapter private structure,
2676  *  and a hardware reset occur.
2677  **/
2678 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2679 {
2680 	struct net_device *netdev;
2681 	struct igb_adapter *adapter;
2682 	struct e1000_hw *hw;
2683 	u16 eeprom_data = 0;
2684 	s32 ret_val;
2685 	static int global_quad_port_a; /* global quad port a indication */
2686 	const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2687 	int err, pci_using_dac;
2688 	u8 part_str[E1000_PBANUM_LENGTH];
2689 
2690 	/* Catch broken hardware that put the wrong VF device ID in
2691 	 * the PCIe SR-IOV capability.
2692 	 */
2693 	if (pdev->is_virtfn) {
2694 		WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2695 			pci_name(pdev), pdev->vendor, pdev->device);
2696 		return -EINVAL;
2697 	}
2698 
2699 	err = pci_enable_device_mem(pdev);
2700 	if (err)
2701 		return err;
2702 
2703 	pci_using_dac = 0;
2704 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2705 	if (!err) {
2706 		pci_using_dac = 1;
2707 	} else {
2708 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2709 		if (err) {
2710 			dev_err(&pdev->dev,
2711 				"No usable DMA configuration, aborting\n");
2712 			goto err_dma;
2713 		}
2714 	}
2715 
2716 	err = pci_request_mem_regions(pdev, igb_driver_name);
2717 	if (err)
2718 		goto err_pci_reg;
2719 
2720 	pci_enable_pcie_error_reporting(pdev);
2721 
2722 	pci_set_master(pdev);
2723 	pci_save_state(pdev);
2724 
2725 	err = -ENOMEM;
2726 	netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2727 				   IGB_MAX_TX_QUEUES);
2728 	if (!netdev)
2729 		goto err_alloc_etherdev;
2730 
2731 	SET_NETDEV_DEV(netdev, &pdev->dev);
2732 
2733 	pci_set_drvdata(pdev, netdev);
2734 	adapter = netdev_priv(netdev);
2735 	adapter->netdev = netdev;
2736 	adapter->pdev = pdev;
2737 	hw = &adapter->hw;
2738 	hw->back = adapter;
2739 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2740 
2741 	err = -EIO;
2742 	adapter->io_addr = pci_iomap(pdev, 0, 0);
2743 	if (!adapter->io_addr)
2744 		goto err_ioremap;
2745 	/* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
2746 	hw->hw_addr = adapter->io_addr;
2747 
2748 	netdev->netdev_ops = &igb_netdev_ops;
2749 	igb_set_ethtool_ops(netdev);
2750 	netdev->watchdog_timeo = 5 * HZ;
2751 
2752 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2753 
2754 	netdev->mem_start = pci_resource_start(pdev, 0);
2755 	netdev->mem_end = pci_resource_end(pdev, 0);
2756 
2757 	/* PCI config space info */
2758 	hw->vendor_id = pdev->vendor;
2759 	hw->device_id = pdev->device;
2760 	hw->revision_id = pdev->revision;
2761 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2762 	hw->subsystem_device_id = pdev->subsystem_device;
2763 
2764 	/* Copy the default MAC, PHY and NVM function pointers */
2765 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2766 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2767 	memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2768 	/* Initialize skew-specific constants */
2769 	err = ei->get_invariants(hw);
2770 	if (err)
2771 		goto err_sw_init;
2772 
2773 	/* setup the private structure */
2774 	err = igb_sw_init(adapter);
2775 	if (err)
2776 		goto err_sw_init;
2777 
2778 	igb_get_bus_info_pcie(hw);
2779 
2780 	hw->phy.autoneg_wait_to_complete = false;
2781 
2782 	/* Copper options */
2783 	if (hw->phy.media_type == e1000_media_type_copper) {
2784 		hw->phy.mdix = AUTO_ALL_MODES;
2785 		hw->phy.disable_polarity_correction = false;
2786 		hw->phy.ms_type = e1000_ms_hw_default;
2787 	}
2788 
2789 	if (igb_check_reset_block(hw))
2790 		dev_info(&pdev->dev,
2791 			"PHY reset is blocked due to SOL/IDER session.\n");
2792 
2793 	/* features is initialized to 0 in allocation, it might have bits
2794 	 * set by igb_sw_init so we should use an or instead of an
2795 	 * assignment.
2796 	 */
2797 	netdev->features |= NETIF_F_SG |
2798 			    NETIF_F_TSO |
2799 			    NETIF_F_TSO6 |
2800 			    NETIF_F_RXHASH |
2801 			    NETIF_F_RXCSUM |
2802 			    NETIF_F_HW_CSUM;
2803 
2804 	if (hw->mac.type >= e1000_82576)
2805 		netdev->features |= NETIF_F_SCTP_CRC;
2806 
2807 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2808 				  NETIF_F_GSO_GRE_CSUM | \
2809 				  NETIF_F_GSO_IPXIP4 | \
2810 				  NETIF_F_GSO_IPXIP6 | \
2811 				  NETIF_F_GSO_UDP_TUNNEL | \
2812 				  NETIF_F_GSO_UDP_TUNNEL_CSUM)
2813 
2814 	netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
2815 	netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
2816 
2817 	/* copy netdev features into list of user selectable features */
2818 	netdev->hw_features |= netdev->features |
2819 			       NETIF_F_HW_VLAN_CTAG_RX |
2820 			       NETIF_F_HW_VLAN_CTAG_TX |
2821 			       NETIF_F_RXALL;
2822 
2823 	if (hw->mac.type >= e1000_i350)
2824 		netdev->hw_features |= NETIF_F_NTUPLE;
2825 
2826 	if (pci_using_dac)
2827 		netdev->features |= NETIF_F_HIGHDMA;
2828 
2829 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2830 	netdev->mpls_features |= NETIF_F_HW_CSUM;
2831 	netdev->hw_enc_features |= netdev->vlan_features;
2832 
2833 	/* set this bit last since it cannot be part of vlan_features */
2834 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2835 			    NETIF_F_HW_VLAN_CTAG_RX |
2836 			    NETIF_F_HW_VLAN_CTAG_TX;
2837 
2838 	netdev->priv_flags |= IFF_SUPP_NOFCS;
2839 
2840 	netdev->priv_flags |= IFF_UNICAST_FLT;
2841 
2842 	/* MTU range: 68 - 9216 */
2843 	netdev->min_mtu = ETH_MIN_MTU;
2844 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2845 
2846 	adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2847 
2848 	/* before reading the NVM, reset the controller to put the device in a
2849 	 * known good starting state
2850 	 */
2851 	hw->mac.ops.reset_hw(hw);
2852 
2853 	/* make sure the NVM is good , i211/i210 parts can have special NVM
2854 	 * that doesn't contain a checksum
2855 	 */
2856 	switch (hw->mac.type) {
2857 	case e1000_i210:
2858 	case e1000_i211:
2859 		if (igb_get_flash_presence_i210(hw)) {
2860 			if (hw->nvm.ops.validate(hw) < 0) {
2861 				dev_err(&pdev->dev,
2862 					"The NVM Checksum Is Not Valid\n");
2863 				err = -EIO;
2864 				goto err_eeprom;
2865 			}
2866 		}
2867 		break;
2868 	default:
2869 		if (hw->nvm.ops.validate(hw) < 0) {
2870 			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2871 			err = -EIO;
2872 			goto err_eeprom;
2873 		}
2874 		break;
2875 	}
2876 
2877 	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
2878 		/* copy the MAC address out of the NVM */
2879 		if (hw->mac.ops.read_mac_addr(hw))
2880 			dev_err(&pdev->dev, "NVM Read Error\n");
2881 	}
2882 
2883 	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2884 
2885 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2886 		dev_err(&pdev->dev, "Invalid MAC Address\n");
2887 		err = -EIO;
2888 		goto err_eeprom;
2889 	}
2890 
2891 	igb_set_default_mac_filter(adapter);
2892 
2893 	/* get firmware version for ethtool -i */
2894 	igb_set_fw_version(adapter);
2895 
2896 	/* configure RXPBSIZE and TXPBSIZE */
2897 	if (hw->mac.type == e1000_i210) {
2898 		wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
2899 		wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
2900 	}
2901 
2902 	timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
2903 	timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
2904 
2905 	INIT_WORK(&adapter->reset_task, igb_reset_task);
2906 	INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2907 
2908 	/* Initialize link properties that are user-changeable */
2909 	adapter->fc_autoneg = true;
2910 	hw->mac.autoneg = true;
2911 	hw->phy.autoneg_advertised = 0x2f;
2912 
2913 	hw->fc.requested_mode = e1000_fc_default;
2914 	hw->fc.current_mode = e1000_fc_default;
2915 
2916 	igb_validate_mdi_setting(hw);
2917 
2918 	/* By default, support wake on port A */
2919 	if (hw->bus.func == 0)
2920 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2921 
2922 	/* Check the NVM for wake support on non-port A ports */
2923 	if (hw->mac.type >= e1000_82580)
2924 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2925 				 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2926 				 &eeprom_data);
2927 	else if (hw->bus.func == 1)
2928 		hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2929 
2930 	if (eeprom_data & IGB_EEPROM_APME)
2931 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2932 
2933 	/* now that we have the eeprom settings, apply the special cases where
2934 	 * the eeprom may be wrong or the board simply won't support wake on
2935 	 * lan on a particular port
2936 	 */
2937 	switch (pdev->device) {
2938 	case E1000_DEV_ID_82575GB_QUAD_COPPER:
2939 		adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2940 		break;
2941 	case E1000_DEV_ID_82575EB_FIBER_SERDES:
2942 	case E1000_DEV_ID_82576_FIBER:
2943 	case E1000_DEV_ID_82576_SERDES:
2944 		/* Wake events only supported on port A for dual fiber
2945 		 * regardless of eeprom setting
2946 		 */
2947 		if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2948 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2949 		break;
2950 	case E1000_DEV_ID_82576_QUAD_COPPER:
2951 	case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2952 		/* if quad port adapter, disable WoL on all but port A */
2953 		if (global_quad_port_a != 0)
2954 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2955 		else
2956 			adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2957 		/* Reset for multiple quad port adapters */
2958 		if (++global_quad_port_a == 4)
2959 			global_quad_port_a = 0;
2960 		break;
2961 	default:
2962 		/* If the device can't wake, don't set software support */
2963 		if (!device_can_wakeup(&adapter->pdev->dev))
2964 			adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2965 	}
2966 
2967 	/* initialize the wol settings based on the eeprom settings */
2968 	if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2969 		adapter->wol |= E1000_WUFC_MAG;
2970 
2971 	/* Some vendors want WoL disabled by default, but still supported */
2972 	if ((hw->mac.type == e1000_i350) &&
2973 	    (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2974 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2975 		adapter->wol = 0;
2976 	}
2977 
2978 	/* Some vendors want the ability to Use the EEPROM setting as
2979 	 * enable/disable only, and not for capability
2980 	 */
2981 	if (((hw->mac.type == e1000_i350) ||
2982 	     (hw->mac.type == e1000_i354)) &&
2983 	    (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
2984 		adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2985 		adapter->wol = 0;
2986 	}
2987 	if (hw->mac.type == e1000_i350) {
2988 		if (((pdev->subsystem_device == 0x5001) ||
2989 		     (pdev->subsystem_device == 0x5002)) &&
2990 				(hw->bus.func == 0)) {
2991 			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2992 			adapter->wol = 0;
2993 		}
2994 		if (pdev->subsystem_device == 0x1F52)
2995 			adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2996 	}
2997 
2998 	device_set_wakeup_enable(&adapter->pdev->dev,
2999 				 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3000 
3001 	/* reset the hardware with the new settings */
3002 	igb_reset(adapter);
3003 
3004 	/* Init the I2C interface */
3005 	err = igb_init_i2c(adapter);
3006 	if (err) {
3007 		dev_err(&pdev->dev, "failed to init i2c interface\n");
3008 		goto err_eeprom;
3009 	}
3010 
3011 	/* let the f/w know that the h/w is now under the control of the
3012 	 * driver.
3013 	 */
3014 	igb_get_hw_control(adapter);
3015 
3016 	strcpy(netdev->name, "eth%d");
3017 	err = register_netdev(netdev);
3018 	if (err)
3019 		goto err_register;
3020 
3021 	/* carrier off reporting is important to ethtool even BEFORE open */
3022 	netif_carrier_off(netdev);
3023 
3024 #ifdef CONFIG_IGB_DCA
3025 	if (dca_add_requester(&pdev->dev) == 0) {
3026 		adapter->flags |= IGB_FLAG_DCA_ENABLED;
3027 		dev_info(&pdev->dev, "DCA enabled\n");
3028 		igb_setup_dca(adapter);
3029 	}
3030 
3031 #endif
3032 #ifdef CONFIG_IGB_HWMON
3033 	/* Initialize the thermal sensor on i350 devices. */
3034 	if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3035 		u16 ets_word;
3036 
3037 		/* Read the NVM to determine if this i350 device supports an
3038 		 * external thermal sensor.
3039 		 */
3040 		hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3041 		if (ets_word != 0x0000 && ets_word != 0xFFFF)
3042 			adapter->ets = true;
3043 		else
3044 			adapter->ets = false;
3045 		if (igb_sysfs_init(adapter))
3046 			dev_err(&pdev->dev,
3047 				"failed to allocate sysfs resources\n");
3048 	} else {
3049 		adapter->ets = false;
3050 	}
3051 #endif
3052 	/* Check if Media Autosense is enabled */
3053 	adapter->ei = *ei;
3054 	if (hw->dev_spec._82575.mas_capable)
3055 		igb_init_mas(adapter);
3056 
3057 	/* do hw tstamp init after resetting */
3058 	igb_ptp_init(adapter);
3059 
3060 	dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3061 	/* print bus type/speed/width info, not applicable to i354 */
3062 	if (hw->mac.type != e1000_i354) {
3063 		dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3064 			 netdev->name,
3065 			 ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3066 			  (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3067 			   "unknown"),
3068 			 ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3069 			  "Width x4" :
3070 			  (hw->bus.width == e1000_bus_width_pcie_x2) ?
3071 			  "Width x2" :
3072 			  (hw->bus.width == e1000_bus_width_pcie_x1) ?
3073 			  "Width x1" : "unknown"), netdev->dev_addr);
3074 	}
3075 
3076 	if ((hw->mac.type >= e1000_i210 ||
3077 	     igb_get_flash_presence_i210(hw))) {
3078 		ret_val = igb_read_part_string(hw, part_str,
3079 					       E1000_PBANUM_LENGTH);
3080 	} else {
3081 		ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3082 	}
3083 
3084 	if (ret_val)
3085 		strcpy(part_str, "Unknown");
3086 	dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3087 	dev_info(&pdev->dev,
3088 		"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3089 		(adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3090 		(adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3091 		adapter->num_rx_queues, adapter->num_tx_queues);
3092 	if (hw->phy.media_type == e1000_media_type_copper) {
3093 		switch (hw->mac.type) {
3094 		case e1000_i350:
3095 		case e1000_i210:
3096 		case e1000_i211:
3097 			/* Enable EEE for internal copper PHY devices */
3098 			err = igb_set_eee_i350(hw, true, true);
3099 			if ((!err) &&
3100 			    (!hw->dev_spec._82575.eee_disable)) {
3101 				adapter->eee_advert =
3102 					MDIO_EEE_100TX | MDIO_EEE_1000T;
3103 				adapter->flags |= IGB_FLAG_EEE;
3104 			}
3105 			break;
3106 		case e1000_i354:
3107 			if ((rd32(E1000_CTRL_EXT) &
3108 			    E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3109 				err = igb_set_eee_i354(hw, true, true);
3110 				if ((!err) &&
3111 					(!hw->dev_spec._82575.eee_disable)) {
3112 					adapter->eee_advert =
3113 					   MDIO_EEE_100TX | MDIO_EEE_1000T;
3114 					adapter->flags |= IGB_FLAG_EEE;
3115 				}
3116 			}
3117 			break;
3118 		default:
3119 			break;
3120 		}
3121 	}
3122 	pm_runtime_put_noidle(&pdev->dev);
3123 	return 0;
3124 
3125 err_register:
3126 	igb_release_hw_control(adapter);
3127 	memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3128 err_eeprom:
3129 	if (!igb_check_reset_block(hw))
3130 		igb_reset_phy(hw);
3131 
3132 	if (hw->flash_address)
3133 		iounmap(hw->flash_address);
3134 err_sw_init:
3135 	kfree(adapter->mac_table);
3136 	kfree(adapter->shadow_vfta);
3137 	igb_clear_interrupt_scheme(adapter);
3138 #ifdef CONFIG_PCI_IOV
3139 	igb_disable_sriov(pdev);
3140 #endif
3141 	pci_iounmap(pdev, adapter->io_addr);
3142 err_ioremap:
3143 	free_netdev(netdev);
3144 err_alloc_etherdev:
3145 	pci_release_mem_regions(pdev);
3146 err_pci_reg:
3147 err_dma:
3148 	pci_disable_device(pdev);
3149 	return err;
3150 }
3151 
3152 #ifdef CONFIG_PCI_IOV
3153 static int igb_disable_sriov(struct pci_dev *pdev)
3154 {
3155 	struct net_device *netdev = pci_get_drvdata(pdev);
3156 	struct igb_adapter *adapter = netdev_priv(netdev);
3157 	struct e1000_hw *hw = &adapter->hw;
3158 
3159 	/* reclaim resources allocated to VFs */
3160 	if (adapter->vf_data) {
3161 		/* disable iov and allow time for transactions to clear */
3162 		if (pci_vfs_assigned(pdev)) {
3163 			dev_warn(&pdev->dev,
3164 				 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3165 			return -EPERM;
3166 		} else {
3167 			pci_disable_sriov(pdev);
3168 			msleep(500);
3169 		}
3170 
3171 		kfree(adapter->vf_mac_list);
3172 		adapter->vf_mac_list = NULL;
3173 		kfree(adapter->vf_data);
3174 		adapter->vf_data = NULL;
3175 		adapter->vfs_allocated_count = 0;
3176 		wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3177 		wrfl();
3178 		msleep(100);
3179 		dev_info(&pdev->dev, "IOV Disabled\n");
3180 
3181 		/* Re-enable DMA Coalescing flag since IOV is turned off */
3182 		adapter->flags |= IGB_FLAG_DMAC;
3183 	}
3184 
3185 	return 0;
3186 }
3187 
3188 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3189 {
3190 	struct net_device *netdev = pci_get_drvdata(pdev);
3191 	struct igb_adapter *adapter = netdev_priv(netdev);
3192 	int old_vfs = pci_num_vf(pdev);
3193 	struct vf_mac_filter *mac_list;
3194 	int err = 0;
3195 	int num_vf_mac_filters, i;
3196 
3197 	if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3198 		err = -EPERM;
3199 		goto out;
3200 	}
3201 	if (!num_vfs)
3202 		goto out;
3203 
3204 	if (old_vfs) {
3205 		dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3206 			 old_vfs, max_vfs);
3207 		adapter->vfs_allocated_count = old_vfs;
3208 	} else
3209 		adapter->vfs_allocated_count = num_vfs;
3210 
3211 	adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3212 				sizeof(struct vf_data_storage), GFP_KERNEL);
3213 
3214 	/* if allocation failed then we do not support SR-IOV */
3215 	if (!adapter->vf_data) {
3216 		adapter->vfs_allocated_count = 0;
3217 		err = -ENOMEM;
3218 		goto out;
3219 	}
3220 
3221 	/* Due to the limited number of RAR entries calculate potential
3222 	 * number of MAC filters available for the VFs. Reserve entries
3223 	 * for PF default MAC, PF MAC filters and at least one RAR entry
3224 	 * for each VF for VF MAC.
3225 	 */
3226 	num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3227 			     (1 + IGB_PF_MAC_FILTERS_RESERVED +
3228 			      adapter->vfs_allocated_count);
3229 
3230 	adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3231 				       sizeof(struct vf_mac_filter),
3232 				       GFP_KERNEL);
3233 
3234 	mac_list = adapter->vf_mac_list;
3235 	INIT_LIST_HEAD(&adapter->vf_macs.l);
3236 
3237 	if (adapter->vf_mac_list) {
3238 		/* Initialize list of VF MAC filters */
3239 		for (i = 0; i < num_vf_mac_filters; i++) {
3240 			mac_list->vf = -1;
3241 			mac_list->free = true;
3242 			list_add(&mac_list->l, &adapter->vf_macs.l);
3243 			mac_list++;
3244 		}
3245 	} else {
3246 		/* If we could not allocate memory for the VF MAC filters
3247 		 * we can continue without this feature but warn user.
3248 		 */
3249 		dev_err(&pdev->dev,
3250 			"Unable to allocate memory for VF MAC filter list\n");
3251 	}
3252 
3253 	/* only call pci_enable_sriov() if no VFs are allocated already */
3254 	if (!old_vfs) {
3255 		err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3256 		if (err)
3257 			goto err_out;
3258 	}
3259 	dev_info(&pdev->dev, "%d VFs allocated\n",
3260 		 adapter->vfs_allocated_count);
3261 	for (i = 0; i < adapter->vfs_allocated_count; i++)
3262 		igb_vf_configure(adapter, i);
3263 
3264 	/* DMA Coalescing is not supported in IOV mode. */
3265 	adapter->flags &= ~IGB_FLAG_DMAC;
3266 	goto out;
3267 
3268 err_out:
3269 	kfree(adapter->vf_mac_list);
3270 	adapter->vf_mac_list = NULL;
3271 	kfree(adapter->vf_data);
3272 	adapter->vf_data = NULL;
3273 	adapter->vfs_allocated_count = 0;
3274 out:
3275 	return err;
3276 }
3277 
3278 #endif
3279 /**
3280  *  igb_remove_i2c - Cleanup  I2C interface
3281  *  @adapter: pointer to adapter structure
3282  **/
3283 static void igb_remove_i2c(struct igb_adapter *adapter)
3284 {
3285 	/* free the adapter bus structure */
3286 	i2c_del_adapter(&adapter->i2c_adap);
3287 }
3288 
3289 /**
3290  *  igb_remove - Device Removal Routine
3291  *  @pdev: PCI device information struct
3292  *
3293  *  igb_remove is called by the PCI subsystem to alert the driver
3294  *  that it should release a PCI device.  The could be caused by a
3295  *  Hot-Plug event, or because the driver is going to be removed from
3296  *  memory.
3297  **/
3298 static void igb_remove(struct pci_dev *pdev)
3299 {
3300 	struct net_device *netdev = pci_get_drvdata(pdev);
3301 	struct igb_adapter *adapter = netdev_priv(netdev);
3302 	struct e1000_hw *hw = &adapter->hw;
3303 
3304 	pm_runtime_get_noresume(&pdev->dev);
3305 #ifdef CONFIG_IGB_HWMON
3306 	igb_sysfs_exit(adapter);
3307 #endif
3308 	igb_remove_i2c(adapter);
3309 	igb_ptp_stop(adapter);
3310 	/* The watchdog timer may be rescheduled, so explicitly
3311 	 * disable watchdog from being rescheduled.
3312 	 */
3313 	set_bit(__IGB_DOWN, &adapter->state);
3314 	del_timer_sync(&adapter->watchdog_timer);
3315 	del_timer_sync(&adapter->phy_info_timer);
3316 
3317 	cancel_work_sync(&adapter->reset_task);
3318 	cancel_work_sync(&adapter->watchdog_task);
3319 
3320 #ifdef CONFIG_IGB_DCA
3321 	if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3322 		dev_info(&pdev->dev, "DCA disabled\n");
3323 		dca_remove_requester(&pdev->dev);
3324 		adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3325 		wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3326 	}
3327 #endif
3328 
3329 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
3330 	 * would have already happened in close and is redundant.
3331 	 */
3332 	igb_release_hw_control(adapter);
3333 
3334 #ifdef CONFIG_PCI_IOV
3335 	igb_disable_sriov(pdev);
3336 #endif
3337 
3338 	unregister_netdev(netdev);
3339 
3340 	igb_clear_interrupt_scheme(adapter);
3341 
3342 	pci_iounmap(pdev, adapter->io_addr);
3343 	if (hw->flash_address)
3344 		iounmap(hw->flash_address);
3345 	pci_release_mem_regions(pdev);
3346 
3347 	kfree(adapter->mac_table);
3348 	kfree(adapter->shadow_vfta);
3349 	free_netdev(netdev);
3350 
3351 	pci_disable_pcie_error_reporting(pdev);
3352 
3353 	pci_disable_device(pdev);
3354 }
3355 
3356 /**
3357  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3358  *  @adapter: board private structure to initialize
3359  *
3360  *  This function initializes the vf specific data storage and then attempts to
3361  *  allocate the VFs.  The reason for ordering it this way is because it is much
3362  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3363  *  the memory for the VFs.
3364  **/
3365 static void igb_probe_vfs(struct igb_adapter *adapter)
3366 {
3367 #ifdef CONFIG_PCI_IOV
3368 	struct pci_dev *pdev = adapter->pdev;
3369 	struct e1000_hw *hw = &adapter->hw;
3370 
3371 	/* Virtualization features not supported on i210 family. */
3372 	if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3373 		return;
3374 
3375 	/* Of the below we really only want the effect of getting
3376 	 * IGB_FLAG_HAS_MSIX set (if available), without which
3377 	 * igb_enable_sriov() has no effect.
3378 	 */
3379 	igb_set_interrupt_capability(adapter, true);
3380 	igb_reset_interrupt_capability(adapter);
3381 
3382 	pci_sriov_set_totalvfs(pdev, 7);
3383 	igb_enable_sriov(pdev, max_vfs);
3384 
3385 #endif /* CONFIG_PCI_IOV */
3386 }
3387 
3388 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3389 {
3390 	struct e1000_hw *hw = &adapter->hw;
3391 	unsigned int max_rss_queues;
3392 
3393 	/* Determine the maximum number of RSS queues supported. */
3394 	switch (hw->mac.type) {
3395 	case e1000_i211:
3396 		max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3397 		break;
3398 	case e1000_82575:
3399 	case e1000_i210:
3400 		max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3401 		break;
3402 	case e1000_i350:
3403 		/* I350 cannot do RSS and SR-IOV at the same time */
3404 		if (!!adapter->vfs_allocated_count) {
3405 			max_rss_queues = 1;
3406 			break;
3407 		}
3408 		/* fall through */
3409 	case e1000_82576:
3410 		if (!!adapter->vfs_allocated_count) {
3411 			max_rss_queues = 2;
3412 			break;
3413 		}
3414 		/* fall through */
3415 	case e1000_82580:
3416 	case e1000_i354:
3417 	default:
3418 		max_rss_queues = IGB_MAX_RX_QUEUES;
3419 		break;
3420 	}
3421 
3422 	return max_rss_queues;
3423 }
3424 
3425 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3426 {
3427 	u32 max_rss_queues;
3428 
3429 	max_rss_queues = igb_get_max_rss_queues(adapter);
3430 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3431 
3432 	igb_set_flag_queue_pairs(adapter, max_rss_queues);
3433 }
3434 
3435 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3436 			      const u32 max_rss_queues)
3437 {
3438 	struct e1000_hw *hw = &adapter->hw;
3439 
3440 	/* Determine if we need to pair queues. */
3441 	switch (hw->mac.type) {
3442 	case e1000_82575:
3443 	case e1000_i211:
3444 		/* Device supports enough interrupts without queue pairing. */
3445 		break;
3446 	case e1000_82576:
3447 	case e1000_82580:
3448 	case e1000_i350:
3449 	case e1000_i354:
3450 	case e1000_i210:
3451 	default:
3452 		/* If rss_queues > half of max_rss_queues, pair the queues in
3453 		 * order to conserve interrupts due to limited supply.
3454 		 */
3455 		if (adapter->rss_queues > (max_rss_queues / 2))
3456 			adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3457 		else
3458 			adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3459 		break;
3460 	}
3461 }
3462 
3463 /**
3464  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3465  *  @adapter: board private structure to initialize
3466  *
3467  *  igb_sw_init initializes the Adapter private data structure.
3468  *  Fields are initialized based on PCI device information and
3469  *  OS network device settings (MTU size).
3470  **/
3471 static int igb_sw_init(struct igb_adapter *adapter)
3472 {
3473 	struct e1000_hw *hw = &adapter->hw;
3474 	struct net_device *netdev = adapter->netdev;
3475 	struct pci_dev *pdev = adapter->pdev;
3476 
3477 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3478 
3479 	/* set default ring sizes */
3480 	adapter->tx_ring_count = IGB_DEFAULT_TXD;
3481 	adapter->rx_ring_count = IGB_DEFAULT_RXD;
3482 
3483 	/* set default ITR values */
3484 	adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3485 	adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3486 
3487 	/* set default work limits */
3488 	adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3489 
3490 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3491 				  VLAN_HLEN;
3492 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3493 
3494 	spin_lock_init(&adapter->nfc_lock);
3495 	spin_lock_init(&adapter->stats64_lock);
3496 #ifdef CONFIG_PCI_IOV
3497 	switch (hw->mac.type) {
3498 	case e1000_82576:
3499 	case e1000_i350:
3500 		if (max_vfs > 7) {
3501 			dev_warn(&pdev->dev,
3502 				 "Maximum of 7 VFs per PF, using max\n");
3503 			max_vfs = adapter->vfs_allocated_count = 7;
3504 		} else
3505 			adapter->vfs_allocated_count = max_vfs;
3506 		if (adapter->vfs_allocated_count)
3507 			dev_warn(&pdev->dev,
3508 				 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3509 		break;
3510 	default:
3511 		break;
3512 	}
3513 #endif /* CONFIG_PCI_IOV */
3514 
3515 	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
3516 	adapter->flags |= IGB_FLAG_HAS_MSIX;
3517 
3518 	adapter->mac_table = kzalloc(sizeof(struct igb_mac_addr) *
3519 				     hw->mac.rar_entry_count, GFP_ATOMIC);
3520 	if (!adapter->mac_table)
3521 		return -ENOMEM;
3522 
3523 	igb_probe_vfs(adapter);
3524 
3525 	igb_init_queue_configuration(adapter);
3526 
3527 	/* Setup and initialize a copy of the hw vlan table array */
3528 	adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3529 				       GFP_ATOMIC);
3530 	if (!adapter->shadow_vfta)
3531 		return -ENOMEM;
3532 
3533 	/* This call may decrease the number of queues */
3534 	if (igb_init_interrupt_scheme(adapter, true)) {
3535 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3536 		return -ENOMEM;
3537 	}
3538 
3539 	/* Explicitly disable IRQ since the NIC can be in any state. */
3540 	igb_irq_disable(adapter);
3541 
3542 	if (hw->mac.type >= e1000_i350)
3543 		adapter->flags &= ~IGB_FLAG_DMAC;
3544 
3545 	set_bit(__IGB_DOWN, &adapter->state);
3546 	return 0;
3547 }
3548 
3549 /**
3550  *  igb_open - Called when a network interface is made active
3551  *  @netdev: network interface device structure
3552  *
3553  *  Returns 0 on success, negative value on failure
3554  *
3555  *  The open entry point is called when a network interface is made
3556  *  active by the system (IFF_UP).  At this point all resources needed
3557  *  for transmit and receive operations are allocated, the interrupt
3558  *  handler is registered with the OS, the watchdog timer is started,
3559  *  and the stack is notified that the interface is ready.
3560  **/
3561 static int __igb_open(struct net_device *netdev, bool resuming)
3562 {
3563 	struct igb_adapter *adapter = netdev_priv(netdev);
3564 	struct e1000_hw *hw = &adapter->hw;
3565 	struct pci_dev *pdev = adapter->pdev;
3566 	int err;
3567 	int i;
3568 
3569 	/* disallow open during test */
3570 	if (test_bit(__IGB_TESTING, &adapter->state)) {
3571 		WARN_ON(resuming);
3572 		return -EBUSY;
3573 	}
3574 
3575 	if (!resuming)
3576 		pm_runtime_get_sync(&pdev->dev);
3577 
3578 	netif_carrier_off(netdev);
3579 
3580 	/* allocate transmit descriptors */
3581 	err = igb_setup_all_tx_resources(adapter);
3582 	if (err)
3583 		goto err_setup_tx;
3584 
3585 	/* allocate receive descriptors */
3586 	err = igb_setup_all_rx_resources(adapter);
3587 	if (err)
3588 		goto err_setup_rx;
3589 
3590 	igb_power_up_link(adapter);
3591 
3592 	/* before we allocate an interrupt, we must be ready to handle it.
3593 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3594 	 * as soon as we call pci_request_irq, so we have to setup our
3595 	 * clean_rx handler before we do so.
3596 	 */
3597 	igb_configure(adapter);
3598 
3599 	err = igb_request_irq(adapter);
3600 	if (err)
3601 		goto err_req_irq;
3602 
3603 	/* Notify the stack of the actual queue counts. */
3604 	err = netif_set_real_num_tx_queues(adapter->netdev,
3605 					   adapter->num_tx_queues);
3606 	if (err)
3607 		goto err_set_queues;
3608 
3609 	err = netif_set_real_num_rx_queues(adapter->netdev,
3610 					   adapter->num_rx_queues);
3611 	if (err)
3612 		goto err_set_queues;
3613 
3614 	/* From here on the code is the same as igb_up() */
3615 	clear_bit(__IGB_DOWN, &adapter->state);
3616 
3617 	for (i = 0; i < adapter->num_q_vectors; i++)
3618 		napi_enable(&(adapter->q_vector[i]->napi));
3619 
3620 	/* Clear any pending interrupts. */
3621 	rd32(E1000_ICR);
3622 
3623 	igb_irq_enable(adapter);
3624 
3625 	/* notify VFs that reset has been completed */
3626 	if (adapter->vfs_allocated_count) {
3627 		u32 reg_data = rd32(E1000_CTRL_EXT);
3628 
3629 		reg_data |= E1000_CTRL_EXT_PFRSTD;
3630 		wr32(E1000_CTRL_EXT, reg_data);
3631 	}
3632 
3633 	netif_tx_start_all_queues(netdev);
3634 
3635 	if (!resuming)
3636 		pm_runtime_put(&pdev->dev);
3637 
3638 	/* start the watchdog. */
3639 	hw->mac.get_link_status = 1;
3640 	schedule_work(&adapter->watchdog_task);
3641 
3642 	return 0;
3643 
3644 err_set_queues:
3645 	igb_free_irq(adapter);
3646 err_req_irq:
3647 	igb_release_hw_control(adapter);
3648 	igb_power_down_link(adapter);
3649 	igb_free_all_rx_resources(adapter);
3650 err_setup_rx:
3651 	igb_free_all_tx_resources(adapter);
3652 err_setup_tx:
3653 	igb_reset(adapter);
3654 	if (!resuming)
3655 		pm_runtime_put(&pdev->dev);
3656 
3657 	return err;
3658 }
3659 
3660 int igb_open(struct net_device *netdev)
3661 {
3662 	return __igb_open(netdev, false);
3663 }
3664 
3665 /**
3666  *  igb_close - Disables a network interface
3667  *  @netdev: network interface device structure
3668  *
3669  *  Returns 0, this is not allowed to fail
3670  *
3671  *  The close entry point is called when an interface is de-activated
3672  *  by the OS.  The hardware is still under the driver's control, but
3673  *  needs to be disabled.  A global MAC reset is issued to stop the
3674  *  hardware, and all transmit and receive resources are freed.
3675  **/
3676 static int __igb_close(struct net_device *netdev, bool suspending)
3677 {
3678 	struct igb_adapter *adapter = netdev_priv(netdev);
3679 	struct pci_dev *pdev = adapter->pdev;
3680 
3681 	WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3682 
3683 	if (!suspending)
3684 		pm_runtime_get_sync(&pdev->dev);
3685 
3686 	igb_down(adapter);
3687 	igb_free_irq(adapter);
3688 
3689 	igb_free_all_tx_resources(adapter);
3690 	igb_free_all_rx_resources(adapter);
3691 
3692 	if (!suspending)
3693 		pm_runtime_put_sync(&pdev->dev);
3694 	return 0;
3695 }
3696 
3697 int igb_close(struct net_device *netdev)
3698 {
3699 	if (netif_device_present(netdev) || netdev->dismantle)
3700 		return __igb_close(netdev, false);
3701 	return 0;
3702 }
3703 
3704 /**
3705  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
3706  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
3707  *
3708  *  Return 0 on success, negative on failure
3709  **/
3710 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3711 {
3712 	struct device *dev = tx_ring->dev;
3713 	int size;
3714 
3715 	size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3716 
3717 	tx_ring->tx_buffer_info = vmalloc(size);
3718 	if (!tx_ring->tx_buffer_info)
3719 		goto err;
3720 
3721 	/* round up to nearest 4K */
3722 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3723 	tx_ring->size = ALIGN(tx_ring->size, 4096);
3724 
3725 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3726 					   &tx_ring->dma, GFP_KERNEL);
3727 	if (!tx_ring->desc)
3728 		goto err;
3729 
3730 	tx_ring->next_to_use = 0;
3731 	tx_ring->next_to_clean = 0;
3732 
3733 	return 0;
3734 
3735 err:
3736 	vfree(tx_ring->tx_buffer_info);
3737 	tx_ring->tx_buffer_info = NULL;
3738 	dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3739 	return -ENOMEM;
3740 }
3741 
3742 /**
3743  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
3744  *				 (Descriptors) for all queues
3745  *  @adapter: board private structure
3746  *
3747  *  Return 0 on success, negative on failure
3748  **/
3749 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3750 {
3751 	struct pci_dev *pdev = adapter->pdev;
3752 	int i, err = 0;
3753 
3754 	for (i = 0; i < adapter->num_tx_queues; i++) {
3755 		err = igb_setup_tx_resources(adapter->tx_ring[i]);
3756 		if (err) {
3757 			dev_err(&pdev->dev,
3758 				"Allocation for Tx Queue %u failed\n", i);
3759 			for (i--; i >= 0; i--)
3760 				igb_free_tx_resources(adapter->tx_ring[i]);
3761 			break;
3762 		}
3763 	}
3764 
3765 	return err;
3766 }
3767 
3768 /**
3769  *  igb_setup_tctl - configure the transmit control registers
3770  *  @adapter: Board private structure
3771  **/
3772 void igb_setup_tctl(struct igb_adapter *adapter)
3773 {
3774 	struct e1000_hw *hw = &adapter->hw;
3775 	u32 tctl;
3776 
3777 	/* disable queue 0 which is enabled by default on 82575 and 82576 */
3778 	wr32(E1000_TXDCTL(0), 0);
3779 
3780 	/* Program the Transmit Control Register */
3781 	tctl = rd32(E1000_TCTL);
3782 	tctl &= ~E1000_TCTL_CT;
3783 	tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3784 		(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3785 
3786 	igb_config_collision_dist(hw);
3787 
3788 	/* Enable transmits */
3789 	tctl |= E1000_TCTL_EN;
3790 
3791 	wr32(E1000_TCTL, tctl);
3792 }
3793 
3794 /**
3795  *  igb_configure_tx_ring - Configure transmit ring after Reset
3796  *  @adapter: board private structure
3797  *  @ring: tx ring to configure
3798  *
3799  *  Configure a transmit ring after a reset.
3800  **/
3801 void igb_configure_tx_ring(struct igb_adapter *adapter,
3802 			   struct igb_ring *ring)
3803 {
3804 	struct e1000_hw *hw = &adapter->hw;
3805 	u32 txdctl = 0;
3806 	u64 tdba = ring->dma;
3807 	int reg_idx = ring->reg_idx;
3808 
3809 	/* disable the queue */
3810 	wr32(E1000_TXDCTL(reg_idx), 0);
3811 	wrfl();
3812 	mdelay(10);
3813 
3814 	wr32(E1000_TDLEN(reg_idx),
3815 	     ring->count * sizeof(union e1000_adv_tx_desc));
3816 	wr32(E1000_TDBAL(reg_idx),
3817 	     tdba & 0x00000000ffffffffULL);
3818 	wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3819 
3820 	ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
3821 	wr32(E1000_TDH(reg_idx), 0);
3822 	writel(0, ring->tail);
3823 
3824 	txdctl |= IGB_TX_PTHRESH;
3825 	txdctl |= IGB_TX_HTHRESH << 8;
3826 	txdctl |= IGB_TX_WTHRESH << 16;
3827 
3828 	/* reinitialize tx_buffer_info */
3829 	memset(ring->tx_buffer_info, 0,
3830 	       sizeof(struct igb_tx_buffer) * ring->count);
3831 
3832 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3833 	wr32(E1000_TXDCTL(reg_idx), txdctl);
3834 }
3835 
3836 /**
3837  *  igb_configure_tx - Configure transmit Unit after Reset
3838  *  @adapter: board private structure
3839  *
3840  *  Configure the Tx unit of the MAC after a reset.
3841  **/
3842 static void igb_configure_tx(struct igb_adapter *adapter)
3843 {
3844 	int i;
3845 
3846 	for (i = 0; i < adapter->num_tx_queues; i++)
3847 		igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3848 }
3849 
3850 /**
3851  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3852  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3853  *
3854  *  Returns 0 on success, negative on failure
3855  **/
3856 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3857 {
3858 	struct device *dev = rx_ring->dev;
3859 	int size;
3860 
3861 	size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3862 
3863 	rx_ring->rx_buffer_info = vmalloc(size);
3864 	if (!rx_ring->rx_buffer_info)
3865 		goto err;
3866 
3867 	/* Round up to nearest 4K */
3868 	rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3869 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3870 
3871 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3872 					   &rx_ring->dma, GFP_KERNEL);
3873 	if (!rx_ring->desc)
3874 		goto err;
3875 
3876 	rx_ring->next_to_alloc = 0;
3877 	rx_ring->next_to_clean = 0;
3878 	rx_ring->next_to_use = 0;
3879 
3880 	return 0;
3881 
3882 err:
3883 	vfree(rx_ring->rx_buffer_info);
3884 	rx_ring->rx_buffer_info = NULL;
3885 	dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3886 	return -ENOMEM;
3887 }
3888 
3889 /**
3890  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3891  *				 (Descriptors) for all queues
3892  *  @adapter: board private structure
3893  *
3894  *  Return 0 on success, negative on failure
3895  **/
3896 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3897 {
3898 	struct pci_dev *pdev = adapter->pdev;
3899 	int i, err = 0;
3900 
3901 	for (i = 0; i < adapter->num_rx_queues; i++) {
3902 		err = igb_setup_rx_resources(adapter->rx_ring[i]);
3903 		if (err) {
3904 			dev_err(&pdev->dev,
3905 				"Allocation for Rx Queue %u failed\n", i);
3906 			for (i--; i >= 0; i--)
3907 				igb_free_rx_resources(adapter->rx_ring[i]);
3908 			break;
3909 		}
3910 	}
3911 
3912 	return err;
3913 }
3914 
3915 /**
3916  *  igb_setup_mrqc - configure the multiple receive queue control registers
3917  *  @adapter: Board private structure
3918  **/
3919 static void igb_setup_mrqc(struct igb_adapter *adapter)
3920 {
3921 	struct e1000_hw *hw = &adapter->hw;
3922 	u32 mrqc, rxcsum;
3923 	u32 j, num_rx_queues;
3924 	u32 rss_key[10];
3925 
3926 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
3927 	for (j = 0; j < 10; j++)
3928 		wr32(E1000_RSSRK(j), rss_key[j]);
3929 
3930 	num_rx_queues = adapter->rss_queues;
3931 
3932 	switch (hw->mac.type) {
3933 	case e1000_82576:
3934 		/* 82576 supports 2 RSS queues for SR-IOV */
3935 		if (adapter->vfs_allocated_count)
3936 			num_rx_queues = 2;
3937 		break;
3938 	default:
3939 		break;
3940 	}
3941 
3942 	if (adapter->rss_indir_tbl_init != num_rx_queues) {
3943 		for (j = 0; j < IGB_RETA_SIZE; j++)
3944 			adapter->rss_indir_tbl[j] =
3945 			(j * num_rx_queues) / IGB_RETA_SIZE;
3946 		adapter->rss_indir_tbl_init = num_rx_queues;
3947 	}
3948 	igb_write_rss_indir_tbl(adapter);
3949 
3950 	/* Disable raw packet checksumming so that RSS hash is placed in
3951 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3952 	 * offloads as they are enabled by default
3953 	 */
3954 	rxcsum = rd32(E1000_RXCSUM);
3955 	rxcsum |= E1000_RXCSUM_PCSD;
3956 
3957 	if (adapter->hw.mac.type >= e1000_82576)
3958 		/* Enable Receive Checksum Offload for SCTP */
3959 		rxcsum |= E1000_RXCSUM_CRCOFL;
3960 
3961 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
3962 	wr32(E1000_RXCSUM, rxcsum);
3963 
3964 	/* Generate RSS hash based on packet types, TCP/UDP
3965 	 * port numbers and/or IPv4/v6 src and dst addresses
3966 	 */
3967 	mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3968 	       E1000_MRQC_RSS_FIELD_IPV4_TCP |
3969 	       E1000_MRQC_RSS_FIELD_IPV6 |
3970 	       E1000_MRQC_RSS_FIELD_IPV6_TCP |
3971 	       E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3972 
3973 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3974 		mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3975 	if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3976 		mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3977 
3978 	/* If VMDq is enabled then we set the appropriate mode for that, else
3979 	 * we default to RSS so that an RSS hash is calculated per packet even
3980 	 * if we are only using one queue
3981 	 */
3982 	if (adapter->vfs_allocated_count) {
3983 		if (hw->mac.type > e1000_82575) {
3984 			/* Set the default pool for the PF's first queue */
3985 			u32 vtctl = rd32(E1000_VT_CTL);
3986 
3987 			vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3988 				   E1000_VT_CTL_DISABLE_DEF_POOL);
3989 			vtctl |= adapter->vfs_allocated_count <<
3990 				E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3991 			wr32(E1000_VT_CTL, vtctl);
3992 		}
3993 		if (adapter->rss_queues > 1)
3994 			mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
3995 		else
3996 			mrqc |= E1000_MRQC_ENABLE_VMDQ;
3997 	} else {
3998 		if (hw->mac.type != e1000_i211)
3999 			mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4000 	}
4001 	igb_vmm_control(adapter);
4002 
4003 	wr32(E1000_MRQC, mrqc);
4004 }
4005 
4006 /**
4007  *  igb_setup_rctl - configure the receive control registers
4008  *  @adapter: Board private structure
4009  **/
4010 void igb_setup_rctl(struct igb_adapter *adapter)
4011 {
4012 	struct e1000_hw *hw = &adapter->hw;
4013 	u32 rctl;
4014 
4015 	rctl = rd32(E1000_RCTL);
4016 
4017 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4018 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4019 
4020 	rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4021 		(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4022 
4023 	/* enable stripping of CRC. It's unlikely this will break BMC
4024 	 * redirection as it did with e1000. Newer features require
4025 	 * that the HW strips the CRC.
4026 	 */
4027 	rctl |= E1000_RCTL_SECRC;
4028 
4029 	/* disable store bad packets and clear size bits. */
4030 	rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4031 
4032 	/* enable LPE to allow for reception of jumbo frames */
4033 	rctl |= E1000_RCTL_LPE;
4034 
4035 	/* disable queue 0 to prevent tail write w/o re-config */
4036 	wr32(E1000_RXDCTL(0), 0);
4037 
4038 	/* Attention!!!  For SR-IOV PF driver operations you must enable
4039 	 * queue drop for all VF and PF queues to prevent head of line blocking
4040 	 * if an un-trusted VF does not provide descriptors to hardware.
4041 	 */
4042 	if (adapter->vfs_allocated_count) {
4043 		/* set all queue drop enable bits */
4044 		wr32(E1000_QDE, ALL_QUEUES);
4045 	}
4046 
4047 	/* This is useful for sniffing bad packets. */
4048 	if (adapter->netdev->features & NETIF_F_RXALL) {
4049 		/* UPE and MPE will be handled by normal PROMISC logic
4050 		 * in e1000e_set_rx_mode
4051 		 */
4052 		rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4053 			 E1000_RCTL_BAM | /* RX All Bcast Pkts */
4054 			 E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4055 
4056 		rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4057 			  E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4058 		/* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4059 		 * and that breaks VLANs.
4060 		 */
4061 	}
4062 
4063 	wr32(E1000_RCTL, rctl);
4064 }
4065 
4066 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4067 				   int vfn)
4068 {
4069 	struct e1000_hw *hw = &adapter->hw;
4070 	u32 vmolr;
4071 
4072 	if (size > MAX_JUMBO_FRAME_SIZE)
4073 		size = MAX_JUMBO_FRAME_SIZE;
4074 
4075 	vmolr = rd32(E1000_VMOLR(vfn));
4076 	vmolr &= ~E1000_VMOLR_RLPML_MASK;
4077 	vmolr |= size | E1000_VMOLR_LPE;
4078 	wr32(E1000_VMOLR(vfn), vmolr);
4079 
4080 	return 0;
4081 }
4082 
4083 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4084 					 int vfn, bool enable)
4085 {
4086 	struct e1000_hw *hw = &adapter->hw;
4087 	u32 val, reg;
4088 
4089 	if (hw->mac.type < e1000_82576)
4090 		return;
4091 
4092 	if (hw->mac.type == e1000_i350)
4093 		reg = E1000_DVMOLR(vfn);
4094 	else
4095 		reg = E1000_VMOLR(vfn);
4096 
4097 	val = rd32(reg);
4098 	if (enable)
4099 		val |= E1000_VMOLR_STRVLAN;
4100 	else
4101 		val &= ~(E1000_VMOLR_STRVLAN);
4102 	wr32(reg, val);
4103 }
4104 
4105 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4106 				 int vfn, bool aupe)
4107 {
4108 	struct e1000_hw *hw = &adapter->hw;
4109 	u32 vmolr;
4110 
4111 	/* This register exists only on 82576 and newer so if we are older then
4112 	 * we should exit and do nothing
4113 	 */
4114 	if (hw->mac.type < e1000_82576)
4115 		return;
4116 
4117 	vmolr = rd32(E1000_VMOLR(vfn));
4118 	if (aupe)
4119 		vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4120 	else
4121 		vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4122 
4123 	/* clear all bits that might not be set */
4124 	vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4125 
4126 	if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4127 		vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4128 	/* for VMDq only allow the VFs and pool 0 to accept broadcast and
4129 	 * multicast packets
4130 	 */
4131 	if (vfn <= adapter->vfs_allocated_count)
4132 		vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4133 
4134 	wr32(E1000_VMOLR(vfn), vmolr);
4135 }
4136 
4137 /**
4138  *  igb_configure_rx_ring - Configure a receive ring after Reset
4139  *  @adapter: board private structure
4140  *  @ring: receive ring to be configured
4141  *
4142  *  Configure the Rx unit of the MAC after a reset.
4143  **/
4144 void igb_configure_rx_ring(struct igb_adapter *adapter,
4145 			   struct igb_ring *ring)
4146 {
4147 	struct e1000_hw *hw = &adapter->hw;
4148 	union e1000_adv_rx_desc *rx_desc;
4149 	u64 rdba = ring->dma;
4150 	int reg_idx = ring->reg_idx;
4151 	u32 srrctl = 0, rxdctl = 0;
4152 
4153 	/* disable the queue */
4154 	wr32(E1000_RXDCTL(reg_idx), 0);
4155 
4156 	/* Set DMA base address registers */
4157 	wr32(E1000_RDBAL(reg_idx),
4158 	     rdba & 0x00000000ffffffffULL);
4159 	wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4160 	wr32(E1000_RDLEN(reg_idx),
4161 	     ring->count * sizeof(union e1000_adv_rx_desc));
4162 
4163 	/* initialize head and tail */
4164 	ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4165 	wr32(E1000_RDH(reg_idx), 0);
4166 	writel(0, ring->tail);
4167 
4168 	/* set descriptor configuration */
4169 	srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4170 	if (ring_uses_large_buffer(ring))
4171 		srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4172 	else
4173 		srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4174 	srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4175 	if (hw->mac.type >= e1000_82580)
4176 		srrctl |= E1000_SRRCTL_TIMESTAMP;
4177 	/* Only set Drop Enable if we are supporting multiple queues */
4178 	if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
4179 		srrctl |= E1000_SRRCTL_DROP_EN;
4180 
4181 	wr32(E1000_SRRCTL(reg_idx), srrctl);
4182 
4183 	/* set filtering for VMDQ pools */
4184 	igb_set_vmolr(adapter, reg_idx & 0x7, true);
4185 
4186 	rxdctl |= IGB_RX_PTHRESH;
4187 	rxdctl |= IGB_RX_HTHRESH << 8;
4188 	rxdctl |= IGB_RX_WTHRESH << 16;
4189 
4190 	/* initialize rx_buffer_info */
4191 	memset(ring->rx_buffer_info, 0,
4192 	       sizeof(struct igb_rx_buffer) * ring->count);
4193 
4194 	/* initialize Rx descriptor 0 */
4195 	rx_desc = IGB_RX_DESC(ring, 0);
4196 	rx_desc->wb.upper.length = 0;
4197 
4198 	/* enable receive descriptor fetching */
4199 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4200 	wr32(E1000_RXDCTL(reg_idx), rxdctl);
4201 }
4202 
4203 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4204 				  struct igb_ring *rx_ring)
4205 {
4206 	/* set build_skb and buffer size flags */
4207 	clear_ring_build_skb_enabled(rx_ring);
4208 	clear_ring_uses_large_buffer(rx_ring);
4209 
4210 	if (adapter->flags & IGB_FLAG_RX_LEGACY)
4211 		return;
4212 
4213 	set_ring_build_skb_enabled(rx_ring);
4214 
4215 #if (PAGE_SIZE < 8192)
4216 	if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4217 		return;
4218 
4219 	set_ring_uses_large_buffer(rx_ring);
4220 #endif
4221 }
4222 
4223 /**
4224  *  igb_configure_rx - Configure receive Unit after Reset
4225  *  @adapter: board private structure
4226  *
4227  *  Configure the Rx unit of the MAC after a reset.
4228  **/
4229 static void igb_configure_rx(struct igb_adapter *adapter)
4230 {
4231 	int i;
4232 
4233 	/* set the correct pool for the PF default MAC address in entry 0 */
4234 	igb_set_default_mac_filter(adapter);
4235 
4236 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
4237 	 * the Base and Length of the Rx Descriptor Ring
4238 	 */
4239 	for (i = 0; i < adapter->num_rx_queues; i++) {
4240 		struct igb_ring *rx_ring = adapter->rx_ring[i];
4241 
4242 		igb_set_rx_buffer_len(adapter, rx_ring);
4243 		igb_configure_rx_ring(adapter, rx_ring);
4244 	}
4245 }
4246 
4247 /**
4248  *  igb_free_tx_resources - Free Tx Resources per Queue
4249  *  @tx_ring: Tx descriptor ring for a specific queue
4250  *
4251  *  Free all transmit software resources
4252  **/
4253 void igb_free_tx_resources(struct igb_ring *tx_ring)
4254 {
4255 	igb_clean_tx_ring(tx_ring);
4256 
4257 	vfree(tx_ring->tx_buffer_info);
4258 	tx_ring->tx_buffer_info = NULL;
4259 
4260 	/* if not set, then don't free */
4261 	if (!tx_ring->desc)
4262 		return;
4263 
4264 	dma_free_coherent(tx_ring->dev, tx_ring->size,
4265 			  tx_ring->desc, tx_ring->dma);
4266 
4267 	tx_ring->desc = NULL;
4268 }
4269 
4270 /**
4271  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
4272  *  @adapter: board private structure
4273  *
4274  *  Free all transmit software resources
4275  **/
4276 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4277 {
4278 	int i;
4279 
4280 	for (i = 0; i < adapter->num_tx_queues; i++)
4281 		if (adapter->tx_ring[i])
4282 			igb_free_tx_resources(adapter->tx_ring[i]);
4283 }
4284 
4285 /**
4286  *  igb_clean_tx_ring - Free Tx Buffers
4287  *  @tx_ring: ring to be cleaned
4288  **/
4289 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4290 {
4291 	u16 i = tx_ring->next_to_clean;
4292 	struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4293 
4294 	while (i != tx_ring->next_to_use) {
4295 		union e1000_adv_tx_desc *eop_desc, *tx_desc;
4296 
4297 		/* Free all the Tx ring sk_buffs */
4298 		dev_kfree_skb_any(tx_buffer->skb);
4299 
4300 		/* unmap skb header data */
4301 		dma_unmap_single(tx_ring->dev,
4302 				 dma_unmap_addr(tx_buffer, dma),
4303 				 dma_unmap_len(tx_buffer, len),
4304 				 DMA_TO_DEVICE);
4305 
4306 		/* check for eop_desc to determine the end of the packet */
4307 		eop_desc = tx_buffer->next_to_watch;
4308 		tx_desc = IGB_TX_DESC(tx_ring, i);
4309 
4310 		/* unmap remaining buffers */
4311 		while (tx_desc != eop_desc) {
4312 			tx_buffer++;
4313 			tx_desc++;
4314 			i++;
4315 			if (unlikely(i == tx_ring->count)) {
4316 				i = 0;
4317 				tx_buffer = tx_ring->tx_buffer_info;
4318 				tx_desc = IGB_TX_DESC(tx_ring, 0);
4319 			}
4320 
4321 			/* unmap any remaining paged data */
4322 			if (dma_unmap_len(tx_buffer, len))
4323 				dma_unmap_page(tx_ring->dev,
4324 					       dma_unmap_addr(tx_buffer, dma),
4325 					       dma_unmap_len(tx_buffer, len),
4326 					       DMA_TO_DEVICE);
4327 		}
4328 
4329 		/* move us one more past the eop_desc for start of next pkt */
4330 		tx_buffer++;
4331 		i++;
4332 		if (unlikely(i == tx_ring->count)) {
4333 			i = 0;
4334 			tx_buffer = tx_ring->tx_buffer_info;
4335 		}
4336 	}
4337 
4338 	/* reset BQL for queue */
4339 	netdev_tx_reset_queue(txring_txq(tx_ring));
4340 
4341 	/* reset next_to_use and next_to_clean */
4342 	tx_ring->next_to_use = 0;
4343 	tx_ring->next_to_clean = 0;
4344 }
4345 
4346 /**
4347  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
4348  *  @adapter: board private structure
4349  **/
4350 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4351 {
4352 	int i;
4353 
4354 	for (i = 0; i < adapter->num_tx_queues; i++)
4355 		if (adapter->tx_ring[i])
4356 			igb_clean_tx_ring(adapter->tx_ring[i]);
4357 }
4358 
4359 /**
4360  *  igb_free_rx_resources - Free Rx Resources
4361  *  @rx_ring: ring to clean the resources from
4362  *
4363  *  Free all receive software resources
4364  **/
4365 void igb_free_rx_resources(struct igb_ring *rx_ring)
4366 {
4367 	igb_clean_rx_ring(rx_ring);
4368 
4369 	vfree(rx_ring->rx_buffer_info);
4370 	rx_ring->rx_buffer_info = NULL;
4371 
4372 	/* if not set, then don't free */
4373 	if (!rx_ring->desc)
4374 		return;
4375 
4376 	dma_free_coherent(rx_ring->dev, rx_ring->size,
4377 			  rx_ring->desc, rx_ring->dma);
4378 
4379 	rx_ring->desc = NULL;
4380 }
4381 
4382 /**
4383  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4384  *  @adapter: board private structure
4385  *
4386  *  Free all receive software resources
4387  **/
4388 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4389 {
4390 	int i;
4391 
4392 	for (i = 0; i < adapter->num_rx_queues; i++)
4393 		if (adapter->rx_ring[i])
4394 			igb_free_rx_resources(adapter->rx_ring[i]);
4395 }
4396 
4397 /**
4398  *  igb_clean_rx_ring - Free Rx Buffers per Queue
4399  *  @rx_ring: ring to free buffers from
4400  **/
4401 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4402 {
4403 	u16 i = rx_ring->next_to_clean;
4404 
4405 	if (rx_ring->skb)
4406 		dev_kfree_skb(rx_ring->skb);
4407 	rx_ring->skb = NULL;
4408 
4409 	/* Free all the Rx ring sk_buffs */
4410 	while (i != rx_ring->next_to_alloc) {
4411 		struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4412 
4413 		/* Invalidate cache lines that may have been written to by
4414 		 * device so that we avoid corrupting memory.
4415 		 */
4416 		dma_sync_single_range_for_cpu(rx_ring->dev,
4417 					      buffer_info->dma,
4418 					      buffer_info->page_offset,
4419 					      igb_rx_bufsz(rx_ring),
4420 					      DMA_FROM_DEVICE);
4421 
4422 		/* free resources associated with mapping */
4423 		dma_unmap_page_attrs(rx_ring->dev,
4424 				     buffer_info->dma,
4425 				     igb_rx_pg_size(rx_ring),
4426 				     DMA_FROM_DEVICE,
4427 				     IGB_RX_DMA_ATTR);
4428 		__page_frag_cache_drain(buffer_info->page,
4429 					buffer_info->pagecnt_bias);
4430 
4431 		i++;
4432 		if (i == rx_ring->count)
4433 			i = 0;
4434 	}
4435 
4436 	rx_ring->next_to_alloc = 0;
4437 	rx_ring->next_to_clean = 0;
4438 	rx_ring->next_to_use = 0;
4439 }
4440 
4441 /**
4442  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4443  *  @adapter: board private structure
4444  **/
4445 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4446 {
4447 	int i;
4448 
4449 	for (i = 0; i < adapter->num_rx_queues; i++)
4450 		if (adapter->rx_ring[i])
4451 			igb_clean_rx_ring(adapter->rx_ring[i]);
4452 }
4453 
4454 /**
4455  *  igb_set_mac - Change the Ethernet Address of the NIC
4456  *  @netdev: network interface device structure
4457  *  @p: pointer to an address structure
4458  *
4459  *  Returns 0 on success, negative on failure
4460  **/
4461 static int igb_set_mac(struct net_device *netdev, void *p)
4462 {
4463 	struct igb_adapter *adapter = netdev_priv(netdev);
4464 	struct e1000_hw *hw = &adapter->hw;
4465 	struct sockaddr *addr = p;
4466 
4467 	if (!is_valid_ether_addr(addr->sa_data))
4468 		return -EADDRNOTAVAIL;
4469 
4470 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4471 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4472 
4473 	/* set the correct pool for the new PF MAC address in entry 0 */
4474 	igb_set_default_mac_filter(adapter);
4475 
4476 	return 0;
4477 }
4478 
4479 /**
4480  *  igb_write_mc_addr_list - write multicast addresses to MTA
4481  *  @netdev: network interface device structure
4482  *
4483  *  Writes multicast address list to the MTA hash table.
4484  *  Returns: -ENOMEM on failure
4485  *           0 on no addresses written
4486  *           X on writing X addresses to MTA
4487  **/
4488 static int igb_write_mc_addr_list(struct net_device *netdev)
4489 {
4490 	struct igb_adapter *adapter = netdev_priv(netdev);
4491 	struct e1000_hw *hw = &adapter->hw;
4492 	struct netdev_hw_addr *ha;
4493 	u8  *mta_list;
4494 	int i;
4495 
4496 	if (netdev_mc_empty(netdev)) {
4497 		/* nothing to program, so clear mc list */
4498 		igb_update_mc_addr_list(hw, NULL, 0);
4499 		igb_restore_vf_multicasts(adapter);
4500 		return 0;
4501 	}
4502 
4503 	mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
4504 	if (!mta_list)
4505 		return -ENOMEM;
4506 
4507 	/* The shared function expects a packed array of only addresses. */
4508 	i = 0;
4509 	netdev_for_each_mc_addr(ha, netdev)
4510 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
4511 
4512 	igb_update_mc_addr_list(hw, mta_list, i);
4513 	kfree(mta_list);
4514 
4515 	return netdev_mc_count(netdev);
4516 }
4517 
4518 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4519 {
4520 	struct e1000_hw *hw = &adapter->hw;
4521 	u32 i, pf_id;
4522 
4523 	switch (hw->mac.type) {
4524 	case e1000_i210:
4525 	case e1000_i211:
4526 	case e1000_i350:
4527 		/* VLAN filtering needed for VLAN prio filter */
4528 		if (adapter->netdev->features & NETIF_F_NTUPLE)
4529 			break;
4530 		/* fall through */
4531 	case e1000_82576:
4532 	case e1000_82580:
4533 	case e1000_i354:
4534 		/* VLAN filtering needed for pool filtering */
4535 		if (adapter->vfs_allocated_count)
4536 			break;
4537 		/* fall through */
4538 	default:
4539 		return 1;
4540 	}
4541 
4542 	/* We are already in VLAN promisc, nothing to do */
4543 	if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4544 		return 0;
4545 
4546 	if (!adapter->vfs_allocated_count)
4547 		goto set_vfta;
4548 
4549 	/* Add PF to all active pools */
4550 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4551 
4552 	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4553 		u32 vlvf = rd32(E1000_VLVF(i));
4554 
4555 		vlvf |= BIT(pf_id);
4556 		wr32(E1000_VLVF(i), vlvf);
4557 	}
4558 
4559 set_vfta:
4560 	/* Set all bits in the VLAN filter table array */
4561 	for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4562 		hw->mac.ops.write_vfta(hw, i, ~0U);
4563 
4564 	/* Set flag so we don't redo unnecessary work */
4565 	adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4566 
4567 	return 0;
4568 }
4569 
4570 #define VFTA_BLOCK_SIZE 8
4571 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4572 {
4573 	struct e1000_hw *hw = &adapter->hw;
4574 	u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4575 	u32 vid_start = vfta_offset * 32;
4576 	u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4577 	u32 i, vid, word, bits, pf_id;
4578 
4579 	/* guarantee that we don't scrub out management VLAN */
4580 	vid = adapter->mng_vlan_id;
4581 	if (vid >= vid_start && vid < vid_end)
4582 		vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4583 
4584 	if (!adapter->vfs_allocated_count)
4585 		goto set_vfta;
4586 
4587 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4588 
4589 	for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4590 		u32 vlvf = rd32(E1000_VLVF(i));
4591 
4592 		/* pull VLAN ID from VLVF */
4593 		vid = vlvf & VLAN_VID_MASK;
4594 
4595 		/* only concern ourselves with a certain range */
4596 		if (vid < vid_start || vid >= vid_end)
4597 			continue;
4598 
4599 		if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4600 			/* record VLAN ID in VFTA */
4601 			vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4602 
4603 			/* if PF is part of this then continue */
4604 			if (test_bit(vid, adapter->active_vlans))
4605 				continue;
4606 		}
4607 
4608 		/* remove PF from the pool */
4609 		bits = ~BIT(pf_id);
4610 		bits &= rd32(E1000_VLVF(i));
4611 		wr32(E1000_VLVF(i), bits);
4612 	}
4613 
4614 set_vfta:
4615 	/* extract values from active_vlans and write back to VFTA */
4616 	for (i = VFTA_BLOCK_SIZE; i--;) {
4617 		vid = (vfta_offset + i) * 32;
4618 		word = vid / BITS_PER_LONG;
4619 		bits = vid % BITS_PER_LONG;
4620 
4621 		vfta[i] |= adapter->active_vlans[word] >> bits;
4622 
4623 		hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4624 	}
4625 }
4626 
4627 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4628 {
4629 	u32 i;
4630 
4631 	/* We are not in VLAN promisc, nothing to do */
4632 	if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4633 		return;
4634 
4635 	/* Set flag so we don't redo unnecessary work */
4636 	adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4637 
4638 	for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4639 		igb_scrub_vfta(adapter, i);
4640 }
4641 
4642 /**
4643  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4644  *  @netdev: network interface device structure
4645  *
4646  *  The set_rx_mode entry point is called whenever the unicast or multicast
4647  *  address lists or the network interface flags are updated.  This routine is
4648  *  responsible for configuring the hardware for proper unicast, multicast,
4649  *  promiscuous mode, and all-multi behavior.
4650  **/
4651 static void igb_set_rx_mode(struct net_device *netdev)
4652 {
4653 	struct igb_adapter *adapter = netdev_priv(netdev);
4654 	struct e1000_hw *hw = &adapter->hw;
4655 	unsigned int vfn = adapter->vfs_allocated_count;
4656 	u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
4657 	int count;
4658 
4659 	/* Check for Promiscuous and All Multicast modes */
4660 	if (netdev->flags & IFF_PROMISC) {
4661 		rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
4662 		vmolr |= E1000_VMOLR_MPME;
4663 
4664 		/* enable use of UTA filter to force packets to default pool */
4665 		if (hw->mac.type == e1000_82576)
4666 			vmolr |= E1000_VMOLR_ROPE;
4667 	} else {
4668 		if (netdev->flags & IFF_ALLMULTI) {
4669 			rctl |= E1000_RCTL_MPE;
4670 			vmolr |= E1000_VMOLR_MPME;
4671 		} else {
4672 			/* Write addresses to the MTA, if the attempt fails
4673 			 * then we should just turn on promiscuous mode so
4674 			 * that we can at least receive multicast traffic
4675 			 */
4676 			count = igb_write_mc_addr_list(netdev);
4677 			if (count < 0) {
4678 				rctl |= E1000_RCTL_MPE;
4679 				vmolr |= E1000_VMOLR_MPME;
4680 			} else if (count) {
4681 				vmolr |= E1000_VMOLR_ROMPE;
4682 			}
4683 		}
4684 	}
4685 
4686 	/* Write addresses to available RAR registers, if there is not
4687 	 * sufficient space to store all the addresses then enable
4688 	 * unicast promiscuous mode
4689 	 */
4690 	if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
4691 		rctl |= E1000_RCTL_UPE;
4692 		vmolr |= E1000_VMOLR_ROPE;
4693 	}
4694 
4695 	/* enable VLAN filtering by default */
4696 	rctl |= E1000_RCTL_VFE;
4697 
4698 	/* disable VLAN filtering for modes that require it */
4699 	if ((netdev->flags & IFF_PROMISC) ||
4700 	    (netdev->features & NETIF_F_RXALL)) {
4701 		/* if we fail to set all rules then just clear VFE */
4702 		if (igb_vlan_promisc_enable(adapter))
4703 			rctl &= ~E1000_RCTL_VFE;
4704 	} else {
4705 		igb_vlan_promisc_disable(adapter);
4706 	}
4707 
4708 	/* update state of unicast, multicast, and VLAN filtering modes */
4709 	rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
4710 				     E1000_RCTL_VFE);
4711 	wr32(E1000_RCTL, rctl);
4712 
4713 #if (PAGE_SIZE < 8192)
4714 	if (!adapter->vfs_allocated_count) {
4715 		if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4716 			rlpml = IGB_MAX_FRAME_BUILD_SKB;
4717 	}
4718 #endif
4719 	wr32(E1000_RLPML, rlpml);
4720 
4721 	/* In order to support SR-IOV and eventually VMDq it is necessary to set
4722 	 * the VMOLR to enable the appropriate modes.  Without this workaround
4723 	 * we will have issues with VLAN tag stripping not being done for frames
4724 	 * that are only arriving because we are the default pool
4725 	 */
4726 	if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4727 		return;
4728 
4729 	/* set UTA to appropriate mode */
4730 	igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
4731 
4732 	vmolr |= rd32(E1000_VMOLR(vfn)) &
4733 		 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
4734 
4735 	/* enable Rx jumbo frames, restrict as needed to support build_skb */
4736 	vmolr &= ~E1000_VMOLR_RLPML_MASK;
4737 #if (PAGE_SIZE < 8192)
4738 	if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4739 		vmolr |= IGB_MAX_FRAME_BUILD_SKB;
4740 	else
4741 #endif
4742 		vmolr |= MAX_JUMBO_FRAME_SIZE;
4743 	vmolr |= E1000_VMOLR_LPE;
4744 
4745 	wr32(E1000_VMOLR(vfn), vmolr);
4746 
4747 	igb_restore_vf_multicasts(adapter);
4748 }
4749 
4750 static void igb_check_wvbr(struct igb_adapter *adapter)
4751 {
4752 	struct e1000_hw *hw = &adapter->hw;
4753 	u32 wvbr = 0;
4754 
4755 	switch (hw->mac.type) {
4756 	case e1000_82576:
4757 	case e1000_i350:
4758 		wvbr = rd32(E1000_WVBR);
4759 		if (!wvbr)
4760 			return;
4761 		break;
4762 	default:
4763 		break;
4764 	}
4765 
4766 	adapter->wvbr |= wvbr;
4767 }
4768 
4769 #define IGB_STAGGERED_QUEUE_OFFSET 8
4770 
4771 static void igb_spoof_check(struct igb_adapter *adapter)
4772 {
4773 	int j;
4774 
4775 	if (!adapter->wvbr)
4776 		return;
4777 
4778 	for (j = 0; j < adapter->vfs_allocated_count; j++) {
4779 		if (adapter->wvbr & BIT(j) ||
4780 		    adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
4781 			dev_warn(&adapter->pdev->dev,
4782 				"Spoof event(s) detected on VF %d\n", j);
4783 			adapter->wvbr &=
4784 				~(BIT(j) |
4785 				  BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
4786 		}
4787 	}
4788 }
4789 
4790 /* Need to wait a few seconds after link up to get diagnostic information from
4791  * the phy
4792  */
4793 static void igb_update_phy_info(struct timer_list *t)
4794 {
4795 	struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
4796 	igb_get_phy_info(&adapter->hw);
4797 }
4798 
4799 /**
4800  *  igb_has_link - check shared code for link and determine up/down
4801  *  @adapter: pointer to driver private info
4802  **/
4803 bool igb_has_link(struct igb_adapter *adapter)
4804 {
4805 	struct e1000_hw *hw = &adapter->hw;
4806 	bool link_active = false;
4807 
4808 	/* get_link_status is set on LSC (link status) interrupt or
4809 	 * rx sequence error interrupt.  get_link_status will stay
4810 	 * false until the e1000_check_for_link establishes link
4811 	 * for copper adapters ONLY
4812 	 */
4813 	switch (hw->phy.media_type) {
4814 	case e1000_media_type_copper:
4815 		if (!hw->mac.get_link_status)
4816 			return true;
4817 	case e1000_media_type_internal_serdes:
4818 		hw->mac.ops.check_for_link(hw);
4819 		link_active = !hw->mac.get_link_status;
4820 		break;
4821 	default:
4822 	case e1000_media_type_unknown:
4823 		break;
4824 	}
4825 
4826 	if (((hw->mac.type == e1000_i210) ||
4827 	     (hw->mac.type == e1000_i211)) &&
4828 	     (hw->phy.id == I210_I_PHY_ID)) {
4829 		if (!netif_carrier_ok(adapter->netdev)) {
4830 			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4831 		} else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4832 			adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4833 			adapter->link_check_timeout = jiffies;
4834 		}
4835 	}
4836 
4837 	return link_active;
4838 }
4839 
4840 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4841 {
4842 	bool ret = false;
4843 	u32 ctrl_ext, thstat;
4844 
4845 	/* check for thermal sensor event on i350 copper only */
4846 	if (hw->mac.type == e1000_i350) {
4847 		thstat = rd32(E1000_THSTAT);
4848 		ctrl_ext = rd32(E1000_CTRL_EXT);
4849 
4850 		if ((hw->phy.media_type == e1000_media_type_copper) &&
4851 		    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4852 			ret = !!(thstat & event);
4853 	}
4854 
4855 	return ret;
4856 }
4857 
4858 /**
4859  *  igb_check_lvmmc - check for malformed packets received
4860  *  and indicated in LVMMC register
4861  *  @adapter: pointer to adapter
4862  **/
4863 static void igb_check_lvmmc(struct igb_adapter *adapter)
4864 {
4865 	struct e1000_hw *hw = &adapter->hw;
4866 	u32 lvmmc;
4867 
4868 	lvmmc = rd32(E1000_LVMMC);
4869 	if (lvmmc) {
4870 		if (unlikely(net_ratelimit())) {
4871 			netdev_warn(adapter->netdev,
4872 				    "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4873 				    lvmmc);
4874 		}
4875 	}
4876 }
4877 
4878 /**
4879  *  igb_watchdog - Timer Call-back
4880  *  @data: pointer to adapter cast into an unsigned long
4881  **/
4882 static void igb_watchdog(struct timer_list *t)
4883 {
4884 	struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
4885 	/* Do the rest outside of interrupt context */
4886 	schedule_work(&adapter->watchdog_task);
4887 }
4888 
4889 static void igb_watchdog_task(struct work_struct *work)
4890 {
4891 	struct igb_adapter *adapter = container_of(work,
4892 						   struct igb_adapter,
4893 						   watchdog_task);
4894 	struct e1000_hw *hw = &adapter->hw;
4895 	struct e1000_phy_info *phy = &hw->phy;
4896 	struct net_device *netdev = adapter->netdev;
4897 	u32 link;
4898 	int i;
4899 	u32 connsw;
4900 	u16 phy_data, retry_count = 20;
4901 
4902 	link = igb_has_link(adapter);
4903 
4904 	if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4905 		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4906 			adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4907 		else
4908 			link = false;
4909 	}
4910 
4911 	/* Force link down if we have fiber to swap to */
4912 	if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4913 		if (hw->phy.media_type == e1000_media_type_copper) {
4914 			connsw = rd32(E1000_CONNSW);
4915 			if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4916 				link = 0;
4917 		}
4918 	}
4919 	if (link) {
4920 		/* Perform a reset if the media type changed. */
4921 		if (hw->dev_spec._82575.media_changed) {
4922 			hw->dev_spec._82575.media_changed = false;
4923 			adapter->flags |= IGB_FLAG_MEDIA_RESET;
4924 			igb_reset(adapter);
4925 		}
4926 		/* Cancel scheduled suspend requests. */
4927 		pm_runtime_resume(netdev->dev.parent);
4928 
4929 		if (!netif_carrier_ok(netdev)) {
4930 			u32 ctrl;
4931 
4932 			hw->mac.ops.get_speed_and_duplex(hw,
4933 							 &adapter->link_speed,
4934 							 &adapter->link_duplex);
4935 
4936 			ctrl = rd32(E1000_CTRL);
4937 			/* Links status message must follow this format */
4938 			netdev_info(netdev,
4939 			       "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4940 			       netdev->name,
4941 			       adapter->link_speed,
4942 			       adapter->link_duplex == FULL_DUPLEX ?
4943 			       "Full" : "Half",
4944 			       (ctrl & E1000_CTRL_TFCE) &&
4945 			       (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4946 			       (ctrl & E1000_CTRL_RFCE) ?  "RX" :
4947 			       (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
4948 
4949 			/* disable EEE if enabled */
4950 			if ((adapter->flags & IGB_FLAG_EEE) &&
4951 				(adapter->link_duplex == HALF_DUPLEX)) {
4952 				dev_info(&adapter->pdev->dev,
4953 				"EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4954 				adapter->hw.dev_spec._82575.eee_disable = true;
4955 				adapter->flags &= ~IGB_FLAG_EEE;
4956 			}
4957 
4958 			/* check if SmartSpeed worked */
4959 			igb_check_downshift(hw);
4960 			if (phy->speed_downgraded)
4961 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4962 
4963 			/* check for thermal sensor event */
4964 			if (igb_thermal_sensor_event(hw,
4965 			    E1000_THSTAT_LINK_THROTTLE))
4966 				netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
4967 
4968 			/* adjust timeout factor according to speed/duplex */
4969 			adapter->tx_timeout_factor = 1;
4970 			switch (adapter->link_speed) {
4971 			case SPEED_10:
4972 				adapter->tx_timeout_factor = 14;
4973 				break;
4974 			case SPEED_100:
4975 				/* maybe add some timeout factor ? */
4976 				break;
4977 			}
4978 
4979 			if (adapter->link_speed != SPEED_1000)
4980 				goto no_wait;
4981 
4982 			/* wait for Remote receiver status OK */
4983 retry_read_status:
4984 			if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
4985 					      &phy_data)) {
4986 				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
4987 				    retry_count) {
4988 					msleep(100);
4989 					retry_count--;
4990 					goto retry_read_status;
4991 				} else if (!retry_count) {
4992 					dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
4993 				}
4994 			} else {
4995 				dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
4996 			}
4997 no_wait:
4998 			netif_carrier_on(netdev);
4999 
5000 			igb_ping_all_vfs(adapter);
5001 			igb_check_vf_rate_limit(adapter);
5002 
5003 			/* link state has changed, schedule phy info update */
5004 			if (!test_bit(__IGB_DOWN, &adapter->state))
5005 				mod_timer(&adapter->phy_info_timer,
5006 					  round_jiffies(jiffies + 2 * HZ));
5007 		}
5008 	} else {
5009 		if (netif_carrier_ok(netdev)) {
5010 			adapter->link_speed = 0;
5011 			adapter->link_duplex = 0;
5012 
5013 			/* check for thermal sensor event */
5014 			if (igb_thermal_sensor_event(hw,
5015 			    E1000_THSTAT_PWR_DOWN)) {
5016 				netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5017 			}
5018 
5019 			/* Links status message must follow this format */
5020 			netdev_info(netdev, "igb: %s NIC Link is Down\n",
5021 			       netdev->name);
5022 			netif_carrier_off(netdev);
5023 
5024 			igb_ping_all_vfs(adapter);
5025 
5026 			/* link state has changed, schedule phy info update */
5027 			if (!test_bit(__IGB_DOWN, &adapter->state))
5028 				mod_timer(&adapter->phy_info_timer,
5029 					  round_jiffies(jiffies + 2 * HZ));
5030 
5031 			/* link is down, time to check for alternate media */
5032 			if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5033 				igb_check_swap_media(adapter);
5034 				if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5035 					schedule_work(&adapter->reset_task);
5036 					/* return immediately */
5037 					return;
5038 				}
5039 			}
5040 			pm_schedule_suspend(netdev->dev.parent,
5041 					    MSEC_PER_SEC * 5);
5042 
5043 		/* also check for alternate media here */
5044 		} else if (!netif_carrier_ok(netdev) &&
5045 			   (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5046 			igb_check_swap_media(adapter);
5047 			if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5048 				schedule_work(&adapter->reset_task);
5049 				/* return immediately */
5050 				return;
5051 			}
5052 		}
5053 	}
5054 
5055 	spin_lock(&adapter->stats64_lock);
5056 	igb_update_stats(adapter);
5057 	spin_unlock(&adapter->stats64_lock);
5058 
5059 	for (i = 0; i < adapter->num_tx_queues; i++) {
5060 		struct igb_ring *tx_ring = adapter->tx_ring[i];
5061 		if (!netif_carrier_ok(netdev)) {
5062 			/* We've lost link, so the controller stops DMA,
5063 			 * but we've got queued Tx work that's never going
5064 			 * to get done, so reset controller to flush Tx.
5065 			 * (Do the reset outside of interrupt context).
5066 			 */
5067 			if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5068 				adapter->tx_timeout_count++;
5069 				schedule_work(&adapter->reset_task);
5070 				/* return immediately since reset is imminent */
5071 				return;
5072 			}
5073 		}
5074 
5075 		/* Force detection of hung controller every watchdog period */
5076 		set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5077 	}
5078 
5079 	/* Cause software interrupt to ensure Rx ring is cleaned */
5080 	if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5081 		u32 eics = 0;
5082 
5083 		for (i = 0; i < adapter->num_q_vectors; i++)
5084 			eics |= adapter->q_vector[i]->eims_value;
5085 		wr32(E1000_EICS, eics);
5086 	} else {
5087 		wr32(E1000_ICS, E1000_ICS_RXDMT0);
5088 	}
5089 
5090 	igb_spoof_check(adapter);
5091 	igb_ptp_rx_hang(adapter);
5092 	igb_ptp_tx_hang(adapter);
5093 
5094 	/* Check LVMMC register on i350/i354 only */
5095 	if ((adapter->hw.mac.type == e1000_i350) ||
5096 	    (adapter->hw.mac.type == e1000_i354))
5097 		igb_check_lvmmc(adapter);
5098 
5099 	/* Reset the timer */
5100 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
5101 		if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5102 			mod_timer(&adapter->watchdog_timer,
5103 				  round_jiffies(jiffies +  HZ));
5104 		else
5105 			mod_timer(&adapter->watchdog_timer,
5106 				  round_jiffies(jiffies + 2 * HZ));
5107 	}
5108 }
5109 
5110 enum latency_range {
5111 	lowest_latency = 0,
5112 	low_latency = 1,
5113 	bulk_latency = 2,
5114 	latency_invalid = 255
5115 };
5116 
5117 /**
5118  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
5119  *  @q_vector: pointer to q_vector
5120  *
5121  *  Stores a new ITR value based on strictly on packet size.  This
5122  *  algorithm is less sophisticated than that used in igb_update_itr,
5123  *  due to the difficulty of synchronizing statistics across multiple
5124  *  receive rings.  The divisors and thresholds used by this function
5125  *  were determined based on theoretical maximum wire speed and testing
5126  *  data, in order to minimize response time while increasing bulk
5127  *  throughput.
5128  *  This functionality is controlled by ethtool's coalescing settings.
5129  *  NOTE:  This function is called only when operating in a multiqueue
5130  *         receive environment.
5131  **/
5132 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5133 {
5134 	int new_val = q_vector->itr_val;
5135 	int avg_wire_size = 0;
5136 	struct igb_adapter *adapter = q_vector->adapter;
5137 	unsigned int packets;
5138 
5139 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
5140 	 * ints/sec - ITR timer value of 120 ticks.
5141 	 */
5142 	if (adapter->link_speed != SPEED_1000) {
5143 		new_val = IGB_4K_ITR;
5144 		goto set_itr_val;
5145 	}
5146 
5147 	packets = q_vector->rx.total_packets;
5148 	if (packets)
5149 		avg_wire_size = q_vector->rx.total_bytes / packets;
5150 
5151 	packets = q_vector->tx.total_packets;
5152 	if (packets)
5153 		avg_wire_size = max_t(u32, avg_wire_size,
5154 				      q_vector->tx.total_bytes / packets);
5155 
5156 	/* if avg_wire_size isn't set no work was done */
5157 	if (!avg_wire_size)
5158 		goto clear_counts;
5159 
5160 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
5161 	avg_wire_size += 24;
5162 
5163 	/* Don't starve jumbo frames */
5164 	avg_wire_size = min(avg_wire_size, 3000);
5165 
5166 	/* Give a little boost to mid-size frames */
5167 	if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5168 		new_val = avg_wire_size / 3;
5169 	else
5170 		new_val = avg_wire_size / 2;
5171 
5172 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
5173 	if (new_val < IGB_20K_ITR &&
5174 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5175 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5176 		new_val = IGB_20K_ITR;
5177 
5178 set_itr_val:
5179 	if (new_val != q_vector->itr_val) {
5180 		q_vector->itr_val = new_val;
5181 		q_vector->set_itr = 1;
5182 	}
5183 clear_counts:
5184 	q_vector->rx.total_bytes = 0;
5185 	q_vector->rx.total_packets = 0;
5186 	q_vector->tx.total_bytes = 0;
5187 	q_vector->tx.total_packets = 0;
5188 }
5189 
5190 /**
5191  *  igb_update_itr - update the dynamic ITR value based on statistics
5192  *  @q_vector: pointer to q_vector
5193  *  @ring_container: ring info to update the itr for
5194  *
5195  *  Stores a new ITR value based on packets and byte
5196  *  counts during the last interrupt.  The advantage of per interrupt
5197  *  computation is faster updates and more accurate ITR for the current
5198  *  traffic pattern.  Constants in this function were computed
5199  *  based on theoretical maximum wire speed and thresholds were set based
5200  *  on testing data as well as attempting to minimize response time
5201  *  while increasing bulk throughput.
5202  *  This functionality is controlled by ethtool's coalescing settings.
5203  *  NOTE:  These calculations are only valid when operating in a single-
5204  *         queue environment.
5205  **/
5206 static void igb_update_itr(struct igb_q_vector *q_vector,
5207 			   struct igb_ring_container *ring_container)
5208 {
5209 	unsigned int packets = ring_container->total_packets;
5210 	unsigned int bytes = ring_container->total_bytes;
5211 	u8 itrval = ring_container->itr;
5212 
5213 	/* no packets, exit with status unchanged */
5214 	if (packets == 0)
5215 		return;
5216 
5217 	switch (itrval) {
5218 	case lowest_latency:
5219 		/* handle TSO and jumbo frames */
5220 		if (bytes/packets > 8000)
5221 			itrval = bulk_latency;
5222 		else if ((packets < 5) && (bytes > 512))
5223 			itrval = low_latency;
5224 		break;
5225 	case low_latency:  /* 50 usec aka 20000 ints/s */
5226 		if (bytes > 10000) {
5227 			/* this if handles the TSO accounting */
5228 			if (bytes/packets > 8000)
5229 				itrval = bulk_latency;
5230 			else if ((packets < 10) || ((bytes/packets) > 1200))
5231 				itrval = bulk_latency;
5232 			else if ((packets > 35))
5233 				itrval = lowest_latency;
5234 		} else if (bytes/packets > 2000) {
5235 			itrval = bulk_latency;
5236 		} else if (packets <= 2 && bytes < 512) {
5237 			itrval = lowest_latency;
5238 		}
5239 		break;
5240 	case bulk_latency: /* 250 usec aka 4000 ints/s */
5241 		if (bytes > 25000) {
5242 			if (packets > 35)
5243 				itrval = low_latency;
5244 		} else if (bytes < 1500) {
5245 			itrval = low_latency;
5246 		}
5247 		break;
5248 	}
5249 
5250 	/* clear work counters since we have the values we need */
5251 	ring_container->total_bytes = 0;
5252 	ring_container->total_packets = 0;
5253 
5254 	/* write updated itr to ring container */
5255 	ring_container->itr = itrval;
5256 }
5257 
5258 static void igb_set_itr(struct igb_q_vector *q_vector)
5259 {
5260 	struct igb_adapter *adapter = q_vector->adapter;
5261 	u32 new_itr = q_vector->itr_val;
5262 	u8 current_itr = 0;
5263 
5264 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5265 	if (adapter->link_speed != SPEED_1000) {
5266 		current_itr = 0;
5267 		new_itr = IGB_4K_ITR;
5268 		goto set_itr_now;
5269 	}
5270 
5271 	igb_update_itr(q_vector, &q_vector->tx);
5272 	igb_update_itr(q_vector, &q_vector->rx);
5273 
5274 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5275 
5276 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
5277 	if (current_itr == lowest_latency &&
5278 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5279 	     (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5280 		current_itr = low_latency;
5281 
5282 	switch (current_itr) {
5283 	/* counts and packets in update_itr are dependent on these numbers */
5284 	case lowest_latency:
5285 		new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5286 		break;
5287 	case low_latency:
5288 		new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5289 		break;
5290 	case bulk_latency:
5291 		new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
5292 		break;
5293 	default:
5294 		break;
5295 	}
5296 
5297 set_itr_now:
5298 	if (new_itr != q_vector->itr_val) {
5299 		/* this attempts to bias the interrupt rate towards Bulk
5300 		 * by adding intermediate steps when interrupt rate is
5301 		 * increasing
5302 		 */
5303 		new_itr = new_itr > q_vector->itr_val ?
5304 			  max((new_itr * q_vector->itr_val) /
5305 			  (new_itr + (q_vector->itr_val >> 2)),
5306 			  new_itr) : new_itr;
5307 		/* Don't write the value here; it resets the adapter's
5308 		 * internal timer, and causes us to delay far longer than
5309 		 * we should between interrupts.  Instead, we write the ITR
5310 		 * value at the beginning of the next interrupt so the timing
5311 		 * ends up being correct.
5312 		 */
5313 		q_vector->itr_val = new_itr;
5314 		q_vector->set_itr = 1;
5315 	}
5316 }
5317 
5318 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
5319 			    u32 type_tucmd, u32 mss_l4len_idx)
5320 {
5321 	struct e1000_adv_tx_context_desc *context_desc;
5322 	u16 i = tx_ring->next_to_use;
5323 
5324 	context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5325 
5326 	i++;
5327 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5328 
5329 	/* set bits to identify this as an advanced context descriptor */
5330 	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5331 
5332 	/* For 82575, context index must be unique per ring. */
5333 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5334 		mss_l4len_idx |= tx_ring->reg_idx << 4;
5335 
5336 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
5337 	context_desc->seqnum_seed	= 0;
5338 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
5339 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
5340 }
5341 
5342 static int igb_tso(struct igb_ring *tx_ring,
5343 		   struct igb_tx_buffer *first,
5344 		   u8 *hdr_len)
5345 {
5346 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5347 	struct sk_buff *skb = first->skb;
5348 	union {
5349 		struct iphdr *v4;
5350 		struct ipv6hdr *v6;
5351 		unsigned char *hdr;
5352 	} ip;
5353 	union {
5354 		struct tcphdr *tcp;
5355 		unsigned char *hdr;
5356 	} l4;
5357 	u32 paylen, l4_offset;
5358 	int err;
5359 
5360 	if (skb->ip_summed != CHECKSUM_PARTIAL)
5361 		return 0;
5362 
5363 	if (!skb_is_gso(skb))
5364 		return 0;
5365 
5366 	err = skb_cow_head(skb, 0);
5367 	if (err < 0)
5368 		return err;
5369 
5370 	ip.hdr = skb_network_header(skb);
5371 	l4.hdr = skb_checksum_start(skb);
5372 
5373 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5374 	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5375 
5376 	/* initialize outer IP header fields */
5377 	if (ip.v4->version == 4) {
5378 		unsigned char *csum_start = skb_checksum_start(skb);
5379 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5380 
5381 		/* IP header will have to cancel out any data that
5382 		 * is not a part of the outer IP header
5383 		 */
5384 		ip.v4->check = csum_fold(csum_partial(trans_start,
5385 						      csum_start - trans_start,
5386 						      0));
5387 		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5388 
5389 		ip.v4->tot_len = 0;
5390 		first->tx_flags |= IGB_TX_FLAGS_TSO |
5391 				   IGB_TX_FLAGS_CSUM |
5392 				   IGB_TX_FLAGS_IPV4;
5393 	} else {
5394 		ip.v6->payload_len = 0;
5395 		first->tx_flags |= IGB_TX_FLAGS_TSO |
5396 				   IGB_TX_FLAGS_CSUM;
5397 	}
5398 
5399 	/* determine offset of inner transport header */
5400 	l4_offset = l4.hdr - skb->data;
5401 
5402 	/* compute length of segmentation header */
5403 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
5404 
5405 	/* remove payload length from inner checksum */
5406 	paylen = skb->len - l4_offset;
5407 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
5408 
5409 	/* update gso size and bytecount with header size */
5410 	first->gso_segs = skb_shinfo(skb)->gso_segs;
5411 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
5412 
5413 	/* MSS L4LEN IDX */
5414 	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5415 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5416 
5417 	/* VLAN MACLEN IPLEN */
5418 	vlan_macip_lens = l4.hdr - ip.hdr;
5419 	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5420 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5421 
5422 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
5423 
5424 	return 1;
5425 }
5426 
5427 static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
5428 {
5429 	unsigned int offset = 0;
5430 
5431 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
5432 
5433 	return offset == skb_checksum_start_offset(skb);
5434 }
5435 
5436 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5437 {
5438 	struct sk_buff *skb = first->skb;
5439 	u32 vlan_macip_lens = 0;
5440 	u32 type_tucmd = 0;
5441 
5442 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
5443 csum_failed:
5444 		if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
5445 			return;
5446 		goto no_csum;
5447 	}
5448 
5449 	switch (skb->csum_offset) {
5450 	case offsetof(struct tcphdr, check):
5451 		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5452 		/* fall through */
5453 	case offsetof(struct udphdr, check):
5454 		break;
5455 	case offsetof(struct sctphdr, checksum):
5456 		/* validate that this is actually an SCTP request */
5457 		if (((first->protocol == htons(ETH_P_IP)) &&
5458 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
5459 		    ((first->protocol == htons(ETH_P_IPV6)) &&
5460 		     igb_ipv6_csum_is_sctp(skb))) {
5461 			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5462 			break;
5463 		}
5464 	default:
5465 		skb_checksum_help(skb);
5466 		goto csum_failed;
5467 	}
5468 
5469 	/* update TX checksum flag */
5470 	first->tx_flags |= IGB_TX_FLAGS_CSUM;
5471 	vlan_macip_lens = skb_checksum_start_offset(skb) -
5472 			  skb_network_offset(skb);
5473 no_csum:
5474 	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5475 	vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5476 
5477 	igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
5478 }
5479 
5480 #define IGB_SET_FLAG(_input, _flag, _result) \
5481 	((_flag <= _result) ? \
5482 	 ((u32)(_input & _flag) * (_result / _flag)) : \
5483 	 ((u32)(_input & _flag) / (_flag / _result)))
5484 
5485 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
5486 {
5487 	/* set type for advanced descriptor with frame checksum insertion */
5488 	u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
5489 		       E1000_ADVTXD_DCMD_DEXT |
5490 		       E1000_ADVTXD_DCMD_IFCS;
5491 
5492 	/* set HW vlan bit if vlan is present */
5493 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
5494 				 (E1000_ADVTXD_DCMD_VLE));
5495 
5496 	/* set segmentation bits for TSO */
5497 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
5498 				 (E1000_ADVTXD_DCMD_TSE));
5499 
5500 	/* set timestamp bit if present */
5501 	cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
5502 				 (E1000_ADVTXD_MAC_TSTAMP));
5503 
5504 	/* insert frame checksum */
5505 	cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
5506 
5507 	return cmd_type;
5508 }
5509 
5510 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
5511 				 union e1000_adv_tx_desc *tx_desc,
5512 				 u32 tx_flags, unsigned int paylen)
5513 {
5514 	u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
5515 
5516 	/* 82575 requires a unique index per ring */
5517 	if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5518 		olinfo_status |= tx_ring->reg_idx << 4;
5519 
5520 	/* insert L4 checksum */
5521 	olinfo_status |= IGB_SET_FLAG(tx_flags,
5522 				      IGB_TX_FLAGS_CSUM,
5523 				      (E1000_TXD_POPTS_TXSM << 8));
5524 
5525 	/* insert IPv4 checksum */
5526 	olinfo_status |= IGB_SET_FLAG(tx_flags,
5527 				      IGB_TX_FLAGS_IPV4,
5528 				      (E1000_TXD_POPTS_IXSM << 8));
5529 
5530 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5531 }
5532 
5533 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5534 {
5535 	struct net_device *netdev = tx_ring->netdev;
5536 
5537 	netif_stop_subqueue(netdev, tx_ring->queue_index);
5538 
5539 	/* Herbert's original patch had:
5540 	 *  smp_mb__after_netif_stop_queue();
5541 	 * but since that doesn't exist yet, just open code it.
5542 	 */
5543 	smp_mb();
5544 
5545 	/* We need to check again in a case another CPU has just
5546 	 * made room available.
5547 	 */
5548 	if (igb_desc_unused(tx_ring) < size)
5549 		return -EBUSY;
5550 
5551 	/* A reprieve! */
5552 	netif_wake_subqueue(netdev, tx_ring->queue_index);
5553 
5554 	u64_stats_update_begin(&tx_ring->tx_syncp2);
5555 	tx_ring->tx_stats.restart_queue2++;
5556 	u64_stats_update_end(&tx_ring->tx_syncp2);
5557 
5558 	return 0;
5559 }
5560 
5561 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5562 {
5563 	if (igb_desc_unused(tx_ring) >= size)
5564 		return 0;
5565 	return __igb_maybe_stop_tx(tx_ring, size);
5566 }
5567 
5568 static int igb_tx_map(struct igb_ring *tx_ring,
5569 		      struct igb_tx_buffer *first,
5570 		      const u8 hdr_len)
5571 {
5572 	struct sk_buff *skb = first->skb;
5573 	struct igb_tx_buffer *tx_buffer;
5574 	union e1000_adv_tx_desc *tx_desc;
5575 	struct skb_frag_struct *frag;
5576 	dma_addr_t dma;
5577 	unsigned int data_len, size;
5578 	u32 tx_flags = first->tx_flags;
5579 	u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5580 	u16 i = tx_ring->next_to_use;
5581 
5582 	tx_desc = IGB_TX_DESC(tx_ring, i);
5583 
5584 	igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5585 
5586 	size = skb_headlen(skb);
5587 	data_len = skb->data_len;
5588 
5589 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5590 
5591 	tx_buffer = first;
5592 
5593 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5594 		if (dma_mapping_error(tx_ring->dev, dma))
5595 			goto dma_error;
5596 
5597 		/* record length, and DMA address */
5598 		dma_unmap_len_set(tx_buffer, len, size);
5599 		dma_unmap_addr_set(tx_buffer, dma, dma);
5600 
5601 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
5602 
5603 		while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5604 			tx_desc->read.cmd_type_len =
5605 				cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5606 
5607 			i++;
5608 			tx_desc++;
5609 			if (i == tx_ring->count) {
5610 				tx_desc = IGB_TX_DESC(tx_ring, 0);
5611 				i = 0;
5612 			}
5613 			tx_desc->read.olinfo_status = 0;
5614 
5615 			dma += IGB_MAX_DATA_PER_TXD;
5616 			size -= IGB_MAX_DATA_PER_TXD;
5617 
5618 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
5619 		}
5620 
5621 		if (likely(!data_len))
5622 			break;
5623 
5624 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
5625 
5626 		i++;
5627 		tx_desc++;
5628 		if (i == tx_ring->count) {
5629 			tx_desc = IGB_TX_DESC(tx_ring, 0);
5630 			i = 0;
5631 		}
5632 		tx_desc->read.olinfo_status = 0;
5633 
5634 		size = skb_frag_size(frag);
5635 		data_len -= size;
5636 
5637 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
5638 				       size, DMA_TO_DEVICE);
5639 
5640 		tx_buffer = &tx_ring->tx_buffer_info[i];
5641 	}
5642 
5643 	/* write last descriptor with RS and EOP bits */
5644 	cmd_type |= size | IGB_TXD_DCMD;
5645 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
5646 
5647 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5648 
5649 	/* set the timestamp */
5650 	first->time_stamp = jiffies;
5651 
5652 	/* Force memory writes to complete before letting h/w know there
5653 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
5654 	 * memory model archs, such as IA-64).
5655 	 *
5656 	 * We also need this memory barrier to make certain all of the
5657 	 * status bits have been updated before next_to_watch is written.
5658 	 */
5659 	wmb();
5660 
5661 	/* set next_to_watch value indicating a packet is present */
5662 	first->next_to_watch = tx_desc;
5663 
5664 	i++;
5665 	if (i == tx_ring->count)
5666 		i = 0;
5667 
5668 	tx_ring->next_to_use = i;
5669 
5670 	/* Make sure there is space in the ring for the next send. */
5671 	igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
5672 
5673 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
5674 		writel(i, tx_ring->tail);
5675 
5676 		/* we need this if more than one processor can write to our tail
5677 		 * at a time, it synchronizes IO on IA64/Altix systems
5678 		 */
5679 		mmiowb();
5680 	}
5681 	return 0;
5682 
5683 dma_error:
5684 	dev_err(tx_ring->dev, "TX DMA map failed\n");
5685 	tx_buffer = &tx_ring->tx_buffer_info[i];
5686 
5687 	/* clear dma mappings for failed tx_buffer_info map */
5688 	while (tx_buffer != first) {
5689 		if (dma_unmap_len(tx_buffer, len))
5690 			dma_unmap_page(tx_ring->dev,
5691 				       dma_unmap_addr(tx_buffer, dma),
5692 				       dma_unmap_len(tx_buffer, len),
5693 				       DMA_TO_DEVICE);
5694 		dma_unmap_len_set(tx_buffer, len, 0);
5695 
5696 		if (i-- == 0)
5697 			i += tx_ring->count;
5698 		tx_buffer = &tx_ring->tx_buffer_info[i];
5699 	}
5700 
5701 	if (dma_unmap_len(tx_buffer, len))
5702 		dma_unmap_single(tx_ring->dev,
5703 				 dma_unmap_addr(tx_buffer, dma),
5704 				 dma_unmap_len(tx_buffer, len),
5705 				 DMA_TO_DEVICE);
5706 	dma_unmap_len_set(tx_buffer, len, 0);
5707 
5708 	dev_kfree_skb_any(tx_buffer->skb);
5709 	tx_buffer->skb = NULL;
5710 
5711 	tx_ring->next_to_use = i;
5712 
5713 	return -1;
5714 }
5715 
5716 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
5717 				struct igb_ring *tx_ring)
5718 {
5719 	struct igb_tx_buffer *first;
5720 	int tso;
5721 	u32 tx_flags = 0;
5722 	unsigned short f;
5723 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
5724 	__be16 protocol = vlan_get_protocol(skb);
5725 	u8 hdr_len = 0;
5726 
5727 	/* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
5728 	 *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
5729 	 *       + 2 desc gap to keep tail from touching head,
5730 	 *       + 1 desc for context descriptor,
5731 	 * otherwise try next time
5732 	 */
5733 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5734 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5735 
5736 	if (igb_maybe_stop_tx(tx_ring, count + 3)) {
5737 		/* this is a hard error */
5738 		return NETDEV_TX_BUSY;
5739 	}
5740 
5741 	/* record the location of the first descriptor for this packet */
5742 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
5743 	first->skb = skb;
5744 	first->bytecount = skb->len;
5745 	first->gso_segs = 1;
5746 
5747 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
5748 		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5749 
5750 		if (adapter->tstamp_config.tx_type & HWTSTAMP_TX_ON &&
5751 		    !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
5752 					   &adapter->state)) {
5753 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5754 			tx_flags |= IGB_TX_FLAGS_TSTAMP;
5755 
5756 			adapter->ptp_tx_skb = skb_get(skb);
5757 			adapter->ptp_tx_start = jiffies;
5758 			if (adapter->hw.mac.type == e1000_82576)
5759 				schedule_work(&adapter->ptp_tx_work);
5760 		} else {
5761 			adapter->tx_hwtstamp_skipped++;
5762 		}
5763 	}
5764 
5765 	skb_tx_timestamp(skb);
5766 
5767 	if (skb_vlan_tag_present(skb)) {
5768 		tx_flags |= IGB_TX_FLAGS_VLAN;
5769 		tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5770 	}
5771 
5772 	/* record initial flags and protocol */
5773 	first->tx_flags = tx_flags;
5774 	first->protocol = protocol;
5775 
5776 	tso = igb_tso(tx_ring, first, &hdr_len);
5777 	if (tso < 0)
5778 		goto out_drop;
5779 	else if (!tso)
5780 		igb_tx_csum(tx_ring, first);
5781 
5782 	if (igb_tx_map(tx_ring, first, hdr_len))
5783 		goto cleanup_tx_tstamp;
5784 
5785 	return NETDEV_TX_OK;
5786 
5787 out_drop:
5788 	dev_kfree_skb_any(first->skb);
5789 	first->skb = NULL;
5790 cleanup_tx_tstamp:
5791 	if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
5792 		struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5793 
5794 		dev_kfree_skb_any(adapter->ptp_tx_skb);
5795 		adapter->ptp_tx_skb = NULL;
5796 		if (adapter->hw.mac.type == e1000_82576)
5797 			cancel_work_sync(&adapter->ptp_tx_work);
5798 		clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
5799 	}
5800 
5801 	return NETDEV_TX_OK;
5802 }
5803 
5804 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5805 						    struct sk_buff *skb)
5806 {
5807 	unsigned int r_idx = skb->queue_mapping;
5808 
5809 	if (r_idx >= adapter->num_tx_queues)
5810 		r_idx = r_idx % adapter->num_tx_queues;
5811 
5812 	return adapter->tx_ring[r_idx];
5813 }
5814 
5815 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5816 				  struct net_device *netdev)
5817 {
5818 	struct igb_adapter *adapter = netdev_priv(netdev);
5819 
5820 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5821 	 * in order to meet this minimum size requirement.
5822 	 */
5823 	if (skb_put_padto(skb, 17))
5824 		return NETDEV_TX_OK;
5825 
5826 	return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5827 }
5828 
5829 /**
5830  *  igb_tx_timeout - Respond to a Tx Hang
5831  *  @netdev: network interface device structure
5832  **/
5833 static void igb_tx_timeout(struct net_device *netdev)
5834 {
5835 	struct igb_adapter *adapter = netdev_priv(netdev);
5836 	struct e1000_hw *hw = &adapter->hw;
5837 
5838 	/* Do the reset outside of interrupt context */
5839 	adapter->tx_timeout_count++;
5840 
5841 	if (hw->mac.type >= e1000_82580)
5842 		hw->dev_spec._82575.global_device_reset = true;
5843 
5844 	schedule_work(&adapter->reset_task);
5845 	wr32(E1000_EICS,
5846 	     (adapter->eims_enable_mask & ~adapter->eims_other));
5847 }
5848 
5849 static void igb_reset_task(struct work_struct *work)
5850 {
5851 	struct igb_adapter *adapter;
5852 	adapter = container_of(work, struct igb_adapter, reset_task);
5853 
5854 	igb_dump(adapter);
5855 	netdev_err(adapter->netdev, "Reset adapter\n");
5856 	igb_reinit_locked(adapter);
5857 }
5858 
5859 /**
5860  *  igb_get_stats64 - Get System Network Statistics
5861  *  @netdev: network interface device structure
5862  *  @stats: rtnl_link_stats64 pointer
5863  **/
5864 static void igb_get_stats64(struct net_device *netdev,
5865 			    struct rtnl_link_stats64 *stats)
5866 {
5867 	struct igb_adapter *adapter = netdev_priv(netdev);
5868 
5869 	spin_lock(&adapter->stats64_lock);
5870 	igb_update_stats(adapter);
5871 	memcpy(stats, &adapter->stats64, sizeof(*stats));
5872 	spin_unlock(&adapter->stats64_lock);
5873 }
5874 
5875 /**
5876  *  igb_change_mtu - Change the Maximum Transfer Unit
5877  *  @netdev: network interface device structure
5878  *  @new_mtu: new value for maximum frame size
5879  *
5880  *  Returns 0 on success, negative on failure
5881  **/
5882 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5883 {
5884 	struct igb_adapter *adapter = netdev_priv(netdev);
5885 	struct pci_dev *pdev = adapter->pdev;
5886 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5887 
5888 	/* adjust max frame to be at least the size of a standard frame */
5889 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5890 		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5891 
5892 	while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5893 		usleep_range(1000, 2000);
5894 
5895 	/* igb_down has a dependency on max_frame_size */
5896 	adapter->max_frame_size = max_frame;
5897 
5898 	if (netif_running(netdev))
5899 		igb_down(adapter);
5900 
5901 	dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5902 		 netdev->mtu, new_mtu);
5903 	netdev->mtu = new_mtu;
5904 
5905 	if (netif_running(netdev))
5906 		igb_up(adapter);
5907 	else
5908 		igb_reset(adapter);
5909 
5910 	clear_bit(__IGB_RESETTING, &adapter->state);
5911 
5912 	return 0;
5913 }
5914 
5915 /**
5916  *  igb_update_stats - Update the board statistics counters
5917  *  @adapter: board private structure
5918  **/
5919 void igb_update_stats(struct igb_adapter *adapter)
5920 {
5921 	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
5922 	struct e1000_hw *hw = &adapter->hw;
5923 	struct pci_dev *pdev = adapter->pdev;
5924 	u32 reg, mpc;
5925 	int i;
5926 	u64 bytes, packets;
5927 	unsigned int start;
5928 	u64 _bytes, _packets;
5929 
5930 	/* Prevent stats update while adapter is being reset, or if the pci
5931 	 * connection is down.
5932 	 */
5933 	if (adapter->link_speed == 0)
5934 		return;
5935 	if (pci_channel_offline(pdev))
5936 		return;
5937 
5938 	bytes = 0;
5939 	packets = 0;
5940 
5941 	rcu_read_lock();
5942 	for (i = 0; i < adapter->num_rx_queues; i++) {
5943 		struct igb_ring *ring = adapter->rx_ring[i];
5944 		u32 rqdpc = rd32(E1000_RQDPC(i));
5945 		if (hw->mac.type >= e1000_i210)
5946 			wr32(E1000_RQDPC(i), 0);
5947 
5948 		if (rqdpc) {
5949 			ring->rx_stats.drops += rqdpc;
5950 			net_stats->rx_fifo_errors += rqdpc;
5951 		}
5952 
5953 		do {
5954 			start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5955 			_bytes = ring->rx_stats.bytes;
5956 			_packets = ring->rx_stats.packets;
5957 		} while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5958 		bytes += _bytes;
5959 		packets += _packets;
5960 	}
5961 
5962 	net_stats->rx_bytes = bytes;
5963 	net_stats->rx_packets = packets;
5964 
5965 	bytes = 0;
5966 	packets = 0;
5967 	for (i = 0; i < adapter->num_tx_queues; i++) {
5968 		struct igb_ring *ring = adapter->tx_ring[i];
5969 		do {
5970 			start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5971 			_bytes = ring->tx_stats.bytes;
5972 			_packets = ring->tx_stats.packets;
5973 		} while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5974 		bytes += _bytes;
5975 		packets += _packets;
5976 	}
5977 	net_stats->tx_bytes = bytes;
5978 	net_stats->tx_packets = packets;
5979 	rcu_read_unlock();
5980 
5981 	/* read stats registers */
5982 	adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5983 	adapter->stats.gprc += rd32(E1000_GPRC);
5984 	adapter->stats.gorc += rd32(E1000_GORCL);
5985 	rd32(E1000_GORCH); /* clear GORCL */
5986 	adapter->stats.bprc += rd32(E1000_BPRC);
5987 	adapter->stats.mprc += rd32(E1000_MPRC);
5988 	adapter->stats.roc += rd32(E1000_ROC);
5989 
5990 	adapter->stats.prc64 += rd32(E1000_PRC64);
5991 	adapter->stats.prc127 += rd32(E1000_PRC127);
5992 	adapter->stats.prc255 += rd32(E1000_PRC255);
5993 	adapter->stats.prc511 += rd32(E1000_PRC511);
5994 	adapter->stats.prc1023 += rd32(E1000_PRC1023);
5995 	adapter->stats.prc1522 += rd32(E1000_PRC1522);
5996 	adapter->stats.symerrs += rd32(E1000_SYMERRS);
5997 	adapter->stats.sec += rd32(E1000_SEC);
5998 
5999 	mpc = rd32(E1000_MPC);
6000 	adapter->stats.mpc += mpc;
6001 	net_stats->rx_fifo_errors += mpc;
6002 	adapter->stats.scc += rd32(E1000_SCC);
6003 	adapter->stats.ecol += rd32(E1000_ECOL);
6004 	adapter->stats.mcc += rd32(E1000_MCC);
6005 	adapter->stats.latecol += rd32(E1000_LATECOL);
6006 	adapter->stats.dc += rd32(E1000_DC);
6007 	adapter->stats.rlec += rd32(E1000_RLEC);
6008 	adapter->stats.xonrxc += rd32(E1000_XONRXC);
6009 	adapter->stats.xontxc += rd32(E1000_XONTXC);
6010 	adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6011 	adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6012 	adapter->stats.fcruc += rd32(E1000_FCRUC);
6013 	adapter->stats.gptc += rd32(E1000_GPTC);
6014 	adapter->stats.gotc += rd32(E1000_GOTCL);
6015 	rd32(E1000_GOTCH); /* clear GOTCL */
6016 	adapter->stats.rnbc += rd32(E1000_RNBC);
6017 	adapter->stats.ruc += rd32(E1000_RUC);
6018 	adapter->stats.rfc += rd32(E1000_RFC);
6019 	adapter->stats.rjc += rd32(E1000_RJC);
6020 	adapter->stats.tor += rd32(E1000_TORH);
6021 	adapter->stats.tot += rd32(E1000_TOTH);
6022 	adapter->stats.tpr += rd32(E1000_TPR);
6023 
6024 	adapter->stats.ptc64 += rd32(E1000_PTC64);
6025 	adapter->stats.ptc127 += rd32(E1000_PTC127);
6026 	adapter->stats.ptc255 += rd32(E1000_PTC255);
6027 	adapter->stats.ptc511 += rd32(E1000_PTC511);
6028 	adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6029 	adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6030 
6031 	adapter->stats.mptc += rd32(E1000_MPTC);
6032 	adapter->stats.bptc += rd32(E1000_BPTC);
6033 
6034 	adapter->stats.tpt += rd32(E1000_TPT);
6035 	adapter->stats.colc += rd32(E1000_COLC);
6036 
6037 	adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6038 	/* read internal phy specific stats */
6039 	reg = rd32(E1000_CTRL_EXT);
6040 	if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6041 		adapter->stats.rxerrc += rd32(E1000_RXERRC);
6042 
6043 		/* this stat has invalid values on i210/i211 */
6044 		if ((hw->mac.type != e1000_i210) &&
6045 		    (hw->mac.type != e1000_i211))
6046 			adapter->stats.tncrs += rd32(E1000_TNCRS);
6047 	}
6048 
6049 	adapter->stats.tsctc += rd32(E1000_TSCTC);
6050 	adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6051 
6052 	adapter->stats.iac += rd32(E1000_IAC);
6053 	adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6054 	adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6055 	adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6056 	adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6057 	adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6058 	adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6059 	adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6060 	adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6061 
6062 	/* Fill out the OS statistics structure */
6063 	net_stats->multicast = adapter->stats.mprc;
6064 	net_stats->collisions = adapter->stats.colc;
6065 
6066 	/* Rx Errors */
6067 
6068 	/* RLEC on some newer hardware can be incorrect so build
6069 	 * our own version based on RUC and ROC
6070 	 */
6071 	net_stats->rx_errors = adapter->stats.rxerrc +
6072 		adapter->stats.crcerrs + adapter->stats.algnerrc +
6073 		adapter->stats.ruc + adapter->stats.roc +
6074 		adapter->stats.cexterr;
6075 	net_stats->rx_length_errors = adapter->stats.ruc +
6076 				      adapter->stats.roc;
6077 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
6078 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
6079 	net_stats->rx_missed_errors = adapter->stats.mpc;
6080 
6081 	/* Tx Errors */
6082 	net_stats->tx_errors = adapter->stats.ecol +
6083 			       adapter->stats.latecol;
6084 	net_stats->tx_aborted_errors = adapter->stats.ecol;
6085 	net_stats->tx_window_errors = adapter->stats.latecol;
6086 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
6087 
6088 	/* Tx Dropped needs to be maintained elsewhere */
6089 
6090 	/* Management Stats */
6091 	adapter->stats.mgptc += rd32(E1000_MGTPTC);
6092 	adapter->stats.mgprc += rd32(E1000_MGTPRC);
6093 	adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6094 
6095 	/* OS2BMC Stats */
6096 	reg = rd32(E1000_MANC);
6097 	if (reg & E1000_MANC_EN_BMC2OS) {
6098 		adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6099 		adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6100 		adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6101 		adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6102 	}
6103 }
6104 
6105 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6106 {
6107 	struct e1000_hw *hw = &adapter->hw;
6108 	struct ptp_clock_event event;
6109 	struct timespec64 ts;
6110 	u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
6111 
6112 	if (tsicr & TSINTR_SYS_WRAP) {
6113 		event.type = PTP_CLOCK_PPS;
6114 		if (adapter->ptp_caps.pps)
6115 			ptp_clock_event(adapter->ptp_clock, &event);
6116 		ack |= TSINTR_SYS_WRAP;
6117 	}
6118 
6119 	if (tsicr & E1000_TSICR_TXTS) {
6120 		/* retrieve hardware timestamp */
6121 		schedule_work(&adapter->ptp_tx_work);
6122 		ack |= E1000_TSICR_TXTS;
6123 	}
6124 
6125 	if (tsicr & TSINTR_TT0) {
6126 		spin_lock(&adapter->tmreg_lock);
6127 		ts = timespec64_add(adapter->perout[0].start,
6128 				    adapter->perout[0].period);
6129 		/* u32 conversion of tv_sec is safe until y2106 */
6130 		wr32(E1000_TRGTTIML0, ts.tv_nsec);
6131 		wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
6132 		tsauxc = rd32(E1000_TSAUXC);
6133 		tsauxc |= TSAUXC_EN_TT0;
6134 		wr32(E1000_TSAUXC, tsauxc);
6135 		adapter->perout[0].start = ts;
6136 		spin_unlock(&adapter->tmreg_lock);
6137 		ack |= TSINTR_TT0;
6138 	}
6139 
6140 	if (tsicr & TSINTR_TT1) {
6141 		spin_lock(&adapter->tmreg_lock);
6142 		ts = timespec64_add(adapter->perout[1].start,
6143 				    adapter->perout[1].period);
6144 		wr32(E1000_TRGTTIML1, ts.tv_nsec);
6145 		wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
6146 		tsauxc = rd32(E1000_TSAUXC);
6147 		tsauxc |= TSAUXC_EN_TT1;
6148 		wr32(E1000_TSAUXC, tsauxc);
6149 		adapter->perout[1].start = ts;
6150 		spin_unlock(&adapter->tmreg_lock);
6151 		ack |= TSINTR_TT1;
6152 	}
6153 
6154 	if (tsicr & TSINTR_AUTT0) {
6155 		nsec = rd32(E1000_AUXSTMPL0);
6156 		sec  = rd32(E1000_AUXSTMPH0);
6157 		event.type = PTP_CLOCK_EXTTS;
6158 		event.index = 0;
6159 		event.timestamp = sec * 1000000000ULL + nsec;
6160 		ptp_clock_event(adapter->ptp_clock, &event);
6161 		ack |= TSINTR_AUTT0;
6162 	}
6163 
6164 	if (tsicr & TSINTR_AUTT1) {
6165 		nsec = rd32(E1000_AUXSTMPL1);
6166 		sec  = rd32(E1000_AUXSTMPH1);
6167 		event.type = PTP_CLOCK_EXTTS;
6168 		event.index = 1;
6169 		event.timestamp = sec * 1000000000ULL + nsec;
6170 		ptp_clock_event(adapter->ptp_clock, &event);
6171 		ack |= TSINTR_AUTT1;
6172 	}
6173 
6174 	/* acknowledge the interrupts */
6175 	wr32(E1000_TSICR, ack);
6176 }
6177 
6178 static irqreturn_t igb_msix_other(int irq, void *data)
6179 {
6180 	struct igb_adapter *adapter = data;
6181 	struct e1000_hw *hw = &adapter->hw;
6182 	u32 icr = rd32(E1000_ICR);
6183 	/* reading ICR causes bit 31 of EICR to be cleared */
6184 
6185 	if (icr & E1000_ICR_DRSTA)
6186 		schedule_work(&adapter->reset_task);
6187 
6188 	if (icr & E1000_ICR_DOUTSYNC) {
6189 		/* HW is reporting DMA is out of sync */
6190 		adapter->stats.doosync++;
6191 		/* The DMA Out of Sync is also indication of a spoof event
6192 		 * in IOV mode. Check the Wrong VM Behavior register to
6193 		 * see if it is really a spoof event.
6194 		 */
6195 		igb_check_wvbr(adapter);
6196 	}
6197 
6198 	/* Check for a mailbox event */
6199 	if (icr & E1000_ICR_VMMB)
6200 		igb_msg_task(adapter);
6201 
6202 	if (icr & E1000_ICR_LSC) {
6203 		hw->mac.get_link_status = 1;
6204 		/* guard against interrupt when we're going down */
6205 		if (!test_bit(__IGB_DOWN, &adapter->state))
6206 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
6207 	}
6208 
6209 	if (icr & E1000_ICR_TS)
6210 		igb_tsync_interrupt(adapter);
6211 
6212 	wr32(E1000_EIMS, adapter->eims_other);
6213 
6214 	return IRQ_HANDLED;
6215 }
6216 
6217 static void igb_write_itr(struct igb_q_vector *q_vector)
6218 {
6219 	struct igb_adapter *adapter = q_vector->adapter;
6220 	u32 itr_val = q_vector->itr_val & 0x7FFC;
6221 
6222 	if (!q_vector->set_itr)
6223 		return;
6224 
6225 	if (!itr_val)
6226 		itr_val = 0x4;
6227 
6228 	if (adapter->hw.mac.type == e1000_82575)
6229 		itr_val |= itr_val << 16;
6230 	else
6231 		itr_val |= E1000_EITR_CNT_IGNR;
6232 
6233 	writel(itr_val, q_vector->itr_register);
6234 	q_vector->set_itr = 0;
6235 }
6236 
6237 static irqreturn_t igb_msix_ring(int irq, void *data)
6238 {
6239 	struct igb_q_vector *q_vector = data;
6240 
6241 	/* Write the ITR value calculated from the previous interrupt. */
6242 	igb_write_itr(q_vector);
6243 
6244 	napi_schedule(&q_vector->napi);
6245 
6246 	return IRQ_HANDLED;
6247 }
6248 
6249 #ifdef CONFIG_IGB_DCA
6250 static void igb_update_tx_dca(struct igb_adapter *adapter,
6251 			      struct igb_ring *tx_ring,
6252 			      int cpu)
6253 {
6254 	struct e1000_hw *hw = &adapter->hw;
6255 	u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
6256 
6257 	if (hw->mac.type != e1000_82575)
6258 		txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
6259 
6260 	/* We can enable relaxed ordering for reads, but not writes when
6261 	 * DCA is enabled.  This is due to a known issue in some chipsets
6262 	 * which will cause the DCA tag to be cleared.
6263 	 */
6264 	txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
6265 		  E1000_DCA_TXCTRL_DATA_RRO_EN |
6266 		  E1000_DCA_TXCTRL_DESC_DCA_EN;
6267 
6268 	wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
6269 }
6270 
6271 static void igb_update_rx_dca(struct igb_adapter *adapter,
6272 			      struct igb_ring *rx_ring,
6273 			      int cpu)
6274 {
6275 	struct e1000_hw *hw = &adapter->hw;
6276 	u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
6277 
6278 	if (hw->mac.type != e1000_82575)
6279 		rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
6280 
6281 	/* We can enable relaxed ordering for reads, but not writes when
6282 	 * DCA is enabled.  This is due to a known issue in some chipsets
6283 	 * which will cause the DCA tag to be cleared.
6284 	 */
6285 	rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
6286 		  E1000_DCA_RXCTRL_DESC_DCA_EN;
6287 
6288 	wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
6289 }
6290 
6291 static void igb_update_dca(struct igb_q_vector *q_vector)
6292 {
6293 	struct igb_adapter *adapter = q_vector->adapter;
6294 	int cpu = get_cpu();
6295 
6296 	if (q_vector->cpu == cpu)
6297 		goto out_no_update;
6298 
6299 	if (q_vector->tx.ring)
6300 		igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
6301 
6302 	if (q_vector->rx.ring)
6303 		igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
6304 
6305 	q_vector->cpu = cpu;
6306 out_no_update:
6307 	put_cpu();
6308 }
6309 
6310 static void igb_setup_dca(struct igb_adapter *adapter)
6311 {
6312 	struct e1000_hw *hw = &adapter->hw;
6313 	int i;
6314 
6315 	if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
6316 		return;
6317 
6318 	/* Always use CB2 mode, difference is masked in the CB driver. */
6319 	wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
6320 
6321 	for (i = 0; i < adapter->num_q_vectors; i++) {
6322 		adapter->q_vector[i]->cpu = -1;
6323 		igb_update_dca(adapter->q_vector[i]);
6324 	}
6325 }
6326 
6327 static int __igb_notify_dca(struct device *dev, void *data)
6328 {
6329 	struct net_device *netdev = dev_get_drvdata(dev);
6330 	struct igb_adapter *adapter = netdev_priv(netdev);
6331 	struct pci_dev *pdev = adapter->pdev;
6332 	struct e1000_hw *hw = &adapter->hw;
6333 	unsigned long event = *(unsigned long *)data;
6334 
6335 	switch (event) {
6336 	case DCA_PROVIDER_ADD:
6337 		/* if already enabled, don't do it again */
6338 		if (adapter->flags & IGB_FLAG_DCA_ENABLED)
6339 			break;
6340 		if (dca_add_requester(dev) == 0) {
6341 			adapter->flags |= IGB_FLAG_DCA_ENABLED;
6342 			dev_info(&pdev->dev, "DCA enabled\n");
6343 			igb_setup_dca(adapter);
6344 			break;
6345 		}
6346 		/* Fall Through since DCA is disabled. */
6347 	case DCA_PROVIDER_REMOVE:
6348 		if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
6349 			/* without this a class_device is left
6350 			 * hanging around in the sysfs model
6351 			 */
6352 			dca_remove_requester(dev);
6353 			dev_info(&pdev->dev, "DCA disabled\n");
6354 			adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
6355 			wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
6356 		}
6357 		break;
6358 	}
6359 
6360 	return 0;
6361 }
6362 
6363 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
6364 			  void *p)
6365 {
6366 	int ret_val;
6367 
6368 	ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6369 					 __igb_notify_dca);
6370 
6371 	return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6372 }
6373 #endif /* CONFIG_IGB_DCA */
6374 
6375 #ifdef CONFIG_PCI_IOV
6376 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6377 {
6378 	unsigned char mac_addr[ETH_ALEN];
6379 
6380 	eth_zero_addr(mac_addr);
6381 	igb_set_vf_mac(adapter, vf, mac_addr);
6382 
6383 	/* By default spoof check is enabled for all VFs */
6384 	adapter->vf_data[vf].spoofchk_enabled = true;
6385 
6386 	return 0;
6387 }
6388 
6389 #endif
6390 static void igb_ping_all_vfs(struct igb_adapter *adapter)
6391 {
6392 	struct e1000_hw *hw = &adapter->hw;
6393 	u32 ping;
6394 	int i;
6395 
6396 	for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
6397 		ping = E1000_PF_CONTROL_MSG;
6398 		if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
6399 			ping |= E1000_VT_MSGTYPE_CTS;
6400 		igb_write_mbx(hw, &ping, 1, i);
6401 	}
6402 }
6403 
6404 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6405 {
6406 	struct e1000_hw *hw = &adapter->hw;
6407 	u32 vmolr = rd32(E1000_VMOLR(vf));
6408 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6409 
6410 	vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
6411 			    IGB_VF_FLAG_MULTI_PROMISC);
6412 	vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6413 
6414 	if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
6415 		vmolr |= E1000_VMOLR_MPME;
6416 		vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
6417 		*msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
6418 	} else {
6419 		/* if we have hashes and we are clearing a multicast promisc
6420 		 * flag we need to write the hashes to the MTA as this step
6421 		 * was previously skipped
6422 		 */
6423 		if (vf_data->num_vf_mc_hashes > 30) {
6424 			vmolr |= E1000_VMOLR_MPME;
6425 		} else if (vf_data->num_vf_mc_hashes) {
6426 			int j;
6427 
6428 			vmolr |= E1000_VMOLR_ROMPE;
6429 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6430 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6431 		}
6432 	}
6433 
6434 	wr32(E1000_VMOLR(vf), vmolr);
6435 
6436 	/* there are flags left unprocessed, likely not supported */
6437 	if (*msgbuf & E1000_VT_MSGINFO_MASK)
6438 		return -EINVAL;
6439 
6440 	return 0;
6441 }
6442 
6443 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
6444 				  u32 *msgbuf, u32 vf)
6445 {
6446 	int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6447 	u16 *hash_list = (u16 *)&msgbuf[1];
6448 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6449 	int i;
6450 
6451 	/* salt away the number of multicast addresses assigned
6452 	 * to this VF for later use to restore when the PF multi cast
6453 	 * list changes
6454 	 */
6455 	vf_data->num_vf_mc_hashes = n;
6456 
6457 	/* only up to 30 hash values supported */
6458 	if (n > 30)
6459 		n = 30;
6460 
6461 	/* store the hashes for later use */
6462 	for (i = 0; i < n; i++)
6463 		vf_data->vf_mc_hashes[i] = hash_list[i];
6464 
6465 	/* Flush and reset the mta with the new values */
6466 	igb_set_rx_mode(adapter->netdev);
6467 
6468 	return 0;
6469 }
6470 
6471 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
6472 {
6473 	struct e1000_hw *hw = &adapter->hw;
6474 	struct vf_data_storage *vf_data;
6475 	int i, j;
6476 
6477 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
6478 		u32 vmolr = rd32(E1000_VMOLR(i));
6479 
6480 		vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6481 
6482 		vf_data = &adapter->vf_data[i];
6483 
6484 		if ((vf_data->num_vf_mc_hashes > 30) ||
6485 		    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
6486 			vmolr |= E1000_VMOLR_MPME;
6487 		} else if (vf_data->num_vf_mc_hashes) {
6488 			vmolr |= E1000_VMOLR_ROMPE;
6489 			for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6490 				igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6491 		}
6492 		wr32(E1000_VMOLR(i), vmolr);
6493 	}
6494 }
6495 
6496 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
6497 {
6498 	struct e1000_hw *hw = &adapter->hw;
6499 	u32 pool_mask, vlvf_mask, i;
6500 
6501 	/* create mask for VF and other pools */
6502 	pool_mask = E1000_VLVF_POOLSEL_MASK;
6503 	vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
6504 
6505 	/* drop PF from pool bits */
6506 	pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
6507 			     adapter->vfs_allocated_count);
6508 
6509 	/* Find the vlan filter for this id */
6510 	for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
6511 		u32 vlvf = rd32(E1000_VLVF(i));
6512 		u32 vfta_mask, vid, vfta;
6513 
6514 		/* remove the vf from the pool */
6515 		if (!(vlvf & vlvf_mask))
6516 			continue;
6517 
6518 		/* clear out bit from VLVF */
6519 		vlvf ^= vlvf_mask;
6520 
6521 		/* if other pools are present, just remove ourselves */
6522 		if (vlvf & pool_mask)
6523 			goto update_vlvfb;
6524 
6525 		/* if PF is present, leave VFTA */
6526 		if (vlvf & E1000_VLVF_POOLSEL_MASK)
6527 			goto update_vlvf;
6528 
6529 		vid = vlvf & E1000_VLVF_VLANID_MASK;
6530 		vfta_mask = BIT(vid % 32);
6531 
6532 		/* clear bit from VFTA */
6533 		vfta = adapter->shadow_vfta[vid / 32];
6534 		if (vfta & vfta_mask)
6535 			hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
6536 update_vlvf:
6537 		/* clear pool selection enable */
6538 		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6539 			vlvf &= E1000_VLVF_POOLSEL_MASK;
6540 		else
6541 			vlvf = 0;
6542 update_vlvfb:
6543 		/* clear pool bits */
6544 		wr32(E1000_VLVF(i), vlvf);
6545 	}
6546 }
6547 
6548 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6549 {
6550 	u32 vlvf;
6551 	int idx;
6552 
6553 	/* short cut the special case */
6554 	if (vlan == 0)
6555 		return 0;
6556 
6557 	/* Search for the VLAN id in the VLVF entries */
6558 	for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6559 		vlvf = rd32(E1000_VLVF(idx));
6560 		if ((vlvf & VLAN_VID_MASK) == vlan)
6561 			break;
6562 	}
6563 
6564 	return idx;
6565 }
6566 
6567 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6568 {
6569 	struct e1000_hw *hw = &adapter->hw;
6570 	u32 bits, pf_id;
6571 	int idx;
6572 
6573 	idx = igb_find_vlvf_entry(hw, vid);
6574 	if (!idx)
6575 		return;
6576 
6577 	/* See if any other pools are set for this VLAN filter
6578 	 * entry other than the PF.
6579 	 */
6580 	pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6581 	bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
6582 	bits &= rd32(E1000_VLVF(idx));
6583 
6584 	/* Disable the filter so this falls into the default pool. */
6585 	if (!bits) {
6586 		if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6587 			wr32(E1000_VLVF(idx), BIT(pf_id));
6588 		else
6589 			wr32(E1000_VLVF(idx), 0);
6590 	}
6591 }
6592 
6593 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6594 			   bool add, u32 vf)
6595 {
6596 	int pf_id = adapter->vfs_allocated_count;
6597 	struct e1000_hw *hw = &adapter->hw;
6598 	int err;
6599 
6600 	/* If VLAN overlaps with one the PF is currently monitoring make
6601 	 * sure that we are able to allocate a VLVF entry.  This may be
6602 	 * redundant but it guarantees PF will maintain visibility to
6603 	 * the VLAN.
6604 	 */
6605 	if (add && test_bit(vid, adapter->active_vlans)) {
6606 		err = igb_vfta_set(hw, vid, pf_id, true, false);
6607 		if (err)
6608 			return err;
6609 	}
6610 
6611 	err = igb_vfta_set(hw, vid, vf, add, false);
6612 
6613 	if (add && !err)
6614 		return err;
6615 
6616 	/* If we failed to add the VF VLAN or we are removing the VF VLAN
6617 	 * we may need to drop the PF pool bit in order to allow us to free
6618 	 * up the VLVF resources.
6619 	 */
6620 	if (test_bit(vid, adapter->active_vlans) ||
6621 	    (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6622 		igb_update_pf_vlvf(adapter, vid);
6623 
6624 	return err;
6625 }
6626 
6627 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
6628 {
6629 	struct e1000_hw *hw = &adapter->hw;
6630 
6631 	if (vid)
6632 		wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
6633 	else
6634 		wr32(E1000_VMVIR(vf), 0);
6635 }
6636 
6637 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
6638 				u16 vlan, u8 qos)
6639 {
6640 	int err;
6641 
6642 	err = igb_set_vf_vlan(adapter, vlan, true, vf);
6643 	if (err)
6644 		return err;
6645 
6646 	igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
6647 	igb_set_vmolr(adapter, vf, !vlan);
6648 
6649 	/* revoke access to previous VLAN */
6650 	if (vlan != adapter->vf_data[vf].pf_vlan)
6651 		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6652 				false, vf);
6653 
6654 	adapter->vf_data[vf].pf_vlan = vlan;
6655 	adapter->vf_data[vf].pf_qos = qos;
6656 	igb_set_vf_vlan_strip(adapter, vf, true);
6657 	dev_info(&adapter->pdev->dev,
6658 		 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
6659 	if (test_bit(__IGB_DOWN, &adapter->state)) {
6660 		dev_warn(&adapter->pdev->dev,
6661 			 "The VF VLAN has been set, but the PF device is not up.\n");
6662 		dev_warn(&adapter->pdev->dev,
6663 			 "Bring the PF device up before attempting to use the VF device.\n");
6664 	}
6665 
6666 	return err;
6667 }
6668 
6669 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
6670 {
6671 	/* Restore tagless access via VLAN 0 */
6672 	igb_set_vf_vlan(adapter, 0, true, vf);
6673 
6674 	igb_set_vmvir(adapter, 0, vf);
6675 	igb_set_vmolr(adapter, vf, true);
6676 
6677 	/* Remove any PF assigned VLAN */
6678 	if (adapter->vf_data[vf].pf_vlan)
6679 		igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6680 				false, vf);
6681 
6682 	adapter->vf_data[vf].pf_vlan = 0;
6683 	adapter->vf_data[vf].pf_qos = 0;
6684 	igb_set_vf_vlan_strip(adapter, vf, false);
6685 
6686 	return 0;
6687 }
6688 
6689 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
6690 			       u16 vlan, u8 qos, __be16 vlan_proto)
6691 {
6692 	struct igb_adapter *adapter = netdev_priv(netdev);
6693 
6694 	if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
6695 		return -EINVAL;
6696 
6697 	if (vlan_proto != htons(ETH_P_8021Q))
6698 		return -EPROTONOSUPPORT;
6699 
6700 	return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
6701 			       igb_disable_port_vlan(adapter, vf);
6702 }
6703 
6704 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6705 {
6706 	int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6707 	int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
6708 	int ret;
6709 
6710 	if (adapter->vf_data[vf].pf_vlan)
6711 		return -1;
6712 
6713 	/* VLAN 0 is a special case, don't allow it to be removed */
6714 	if (!vid && !add)
6715 		return 0;
6716 
6717 	ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
6718 	if (!ret)
6719 		igb_set_vf_vlan_strip(adapter, vf, !!vid);
6720 	return ret;
6721 }
6722 
6723 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
6724 {
6725 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6726 
6727 	/* clear flags - except flag that indicates PF has set the MAC */
6728 	vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
6729 	vf_data->last_nack = jiffies;
6730 
6731 	/* reset vlans for device */
6732 	igb_clear_vf_vfta(adapter, vf);
6733 	igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
6734 	igb_set_vmvir(adapter, vf_data->pf_vlan |
6735 			       (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
6736 	igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
6737 	igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
6738 
6739 	/* reset multicast table array for vf */
6740 	adapter->vf_data[vf].num_vf_mc_hashes = 0;
6741 
6742 	/* Flush and reset the mta with the new values */
6743 	igb_set_rx_mode(adapter->netdev);
6744 }
6745 
6746 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
6747 {
6748 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6749 
6750 	/* clear mac address as we were hotplug removed/added */
6751 	if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
6752 		eth_zero_addr(vf_mac);
6753 
6754 	/* process remaining reset events */
6755 	igb_vf_reset(adapter, vf);
6756 }
6757 
6758 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
6759 {
6760 	struct e1000_hw *hw = &adapter->hw;
6761 	unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6762 	u32 reg, msgbuf[3];
6763 	u8 *addr = (u8 *)(&msgbuf[1]);
6764 
6765 	/* process all the same items cleared in a function level reset */
6766 	igb_vf_reset(adapter, vf);
6767 
6768 	/* set vf mac address */
6769 	igb_set_vf_mac(adapter, vf, vf_mac);
6770 
6771 	/* enable transmit and receive for vf */
6772 	reg = rd32(E1000_VFTE);
6773 	wr32(E1000_VFTE, reg | BIT(vf));
6774 	reg = rd32(E1000_VFRE);
6775 	wr32(E1000_VFRE, reg | BIT(vf));
6776 
6777 	adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
6778 
6779 	/* reply to reset with ack and vf mac address */
6780 	if (!is_zero_ether_addr(vf_mac)) {
6781 		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
6782 		memcpy(addr, vf_mac, ETH_ALEN);
6783 	} else {
6784 		msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
6785 	}
6786 	igb_write_mbx(hw, msgbuf, 3, vf);
6787 }
6788 
6789 static void igb_flush_mac_table(struct igb_adapter *adapter)
6790 {
6791 	struct e1000_hw *hw = &adapter->hw;
6792 	int i;
6793 
6794 	for (i = 0; i < hw->mac.rar_entry_count; i++) {
6795 		adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6796 		memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6797 		adapter->mac_table[i].queue = 0;
6798 		igb_rar_set_index(adapter, i);
6799 	}
6800 }
6801 
6802 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
6803 {
6804 	struct e1000_hw *hw = &adapter->hw;
6805 	/* do not count rar entries reserved for VFs MAC addresses */
6806 	int rar_entries = hw->mac.rar_entry_count -
6807 			  adapter->vfs_allocated_count;
6808 	int i, count = 0;
6809 
6810 	for (i = 0; i < rar_entries; i++) {
6811 		/* do not count default entries */
6812 		if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
6813 			continue;
6814 
6815 		/* do not count "in use" entries for different queues */
6816 		if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
6817 		    (adapter->mac_table[i].queue != queue))
6818 			continue;
6819 
6820 		count++;
6821 	}
6822 
6823 	return count;
6824 }
6825 
6826 /* Set default MAC address for the PF in the first RAR entry */
6827 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
6828 {
6829 	struct igb_mac_addr *mac_table = &adapter->mac_table[0];
6830 
6831 	ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
6832 	mac_table->queue = adapter->vfs_allocated_count;
6833 	mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
6834 
6835 	igb_rar_set_index(adapter, 0);
6836 }
6837 
6838 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6839 			      const u8 queue)
6840 {
6841 	struct e1000_hw *hw = &adapter->hw;
6842 	int rar_entries = hw->mac.rar_entry_count -
6843 			  adapter->vfs_allocated_count;
6844 	int i;
6845 
6846 	if (is_zero_ether_addr(addr))
6847 		return -EINVAL;
6848 
6849 	/* Search for the first empty entry in the MAC table.
6850 	 * Do not touch entries at the end of the table reserved for the VF MAC
6851 	 * addresses.
6852 	 */
6853 	for (i = 0; i < rar_entries; i++) {
6854 		if (adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE)
6855 			continue;
6856 
6857 		ether_addr_copy(adapter->mac_table[i].addr, addr);
6858 		adapter->mac_table[i].queue = queue;
6859 		adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE;
6860 
6861 		igb_rar_set_index(adapter, i);
6862 		return i;
6863 	}
6864 
6865 	return -ENOSPC;
6866 }
6867 
6868 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6869 			      const u8 queue)
6870 {
6871 	struct e1000_hw *hw = &adapter->hw;
6872 	int rar_entries = hw->mac.rar_entry_count -
6873 			  adapter->vfs_allocated_count;
6874 	int i;
6875 
6876 	if (is_zero_ether_addr(addr))
6877 		return -EINVAL;
6878 
6879 	/* Search for matching entry in the MAC table based on given address
6880 	 * and queue. Do not touch entries at the end of the table reserved
6881 	 * for the VF MAC addresses.
6882 	 */
6883 	for (i = 0; i < rar_entries; i++) {
6884 		if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
6885 			continue;
6886 		if (adapter->mac_table[i].queue != queue)
6887 			continue;
6888 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
6889 			continue;
6890 
6891 		adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6892 		memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6893 		adapter->mac_table[i].queue = 0;
6894 
6895 		igb_rar_set_index(adapter, i);
6896 		return 0;
6897 	}
6898 
6899 	return -ENOENT;
6900 }
6901 
6902 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
6903 {
6904 	struct igb_adapter *adapter = netdev_priv(netdev);
6905 	int ret;
6906 
6907 	ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6908 
6909 	return min_t(int, ret, 0);
6910 }
6911 
6912 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
6913 {
6914 	struct igb_adapter *adapter = netdev_priv(netdev);
6915 
6916 	igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6917 
6918 	return 0;
6919 }
6920 
6921 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
6922 				 const u32 info, const u8 *addr)
6923 {
6924 	struct pci_dev *pdev = adapter->pdev;
6925 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6926 	struct list_head *pos;
6927 	struct vf_mac_filter *entry = NULL;
6928 	int ret = 0;
6929 
6930 	switch (info) {
6931 	case E1000_VF_MAC_FILTER_CLR:
6932 		/* remove all unicast MAC filters related to the current VF */
6933 		list_for_each(pos, &adapter->vf_macs.l) {
6934 			entry = list_entry(pos, struct vf_mac_filter, l);
6935 			if (entry->vf == vf) {
6936 				entry->vf = -1;
6937 				entry->free = true;
6938 				igb_del_mac_filter(adapter, entry->vf_mac, vf);
6939 			}
6940 		}
6941 		break;
6942 	case E1000_VF_MAC_FILTER_ADD:
6943 		if (vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) {
6944 			dev_warn(&pdev->dev,
6945 				 "VF %d requested MAC filter but is administratively denied\n",
6946 				 vf);
6947 			return -EINVAL;
6948 		}
6949 
6950 		if (!is_valid_ether_addr(addr)) {
6951 			dev_warn(&pdev->dev,
6952 				 "VF %d attempted to set invalid MAC filter\n",
6953 				 vf);
6954 			return -EINVAL;
6955 		}
6956 
6957 		/* try to find empty slot in the list */
6958 		list_for_each(pos, &adapter->vf_macs.l) {
6959 			entry = list_entry(pos, struct vf_mac_filter, l);
6960 			if (entry->free)
6961 				break;
6962 		}
6963 
6964 		if (entry && entry->free) {
6965 			entry->free = false;
6966 			entry->vf = vf;
6967 			ether_addr_copy(entry->vf_mac, addr);
6968 
6969 			ret = igb_add_mac_filter(adapter, addr, vf);
6970 			ret = min_t(int, ret, 0);
6971 		} else {
6972 			ret = -ENOSPC;
6973 		}
6974 
6975 		if (ret == -ENOSPC)
6976 			dev_warn(&pdev->dev,
6977 				 "VF %d has requested MAC filter but there is no space for it\n",
6978 				 vf);
6979 		break;
6980 	default:
6981 		ret = -EINVAL;
6982 		break;
6983 	}
6984 
6985 	return ret;
6986 }
6987 
6988 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
6989 {
6990 	struct pci_dev *pdev = adapter->pdev;
6991 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6992 	u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
6993 
6994 	/* The VF MAC Address is stored in a packed array of bytes
6995 	 * starting at the second 32 bit word of the msg array
6996 	 */
6997 	unsigned char *addr = (unsigned char *)&msg[1];
6998 	int ret = 0;
6999 
7000 	if (!info) {
7001 		if (vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) {
7002 			dev_warn(&pdev->dev,
7003 				 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7004 				 vf);
7005 			return -EINVAL;
7006 		}
7007 
7008 		if (!is_valid_ether_addr(addr)) {
7009 			dev_warn(&pdev->dev,
7010 				 "VF %d attempted to set invalid MAC\n",
7011 				 vf);
7012 			return -EINVAL;
7013 		}
7014 
7015 		ret = igb_set_vf_mac(adapter, vf, addr);
7016 	} else {
7017 		ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7018 	}
7019 
7020 	return ret;
7021 }
7022 
7023 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7024 {
7025 	struct e1000_hw *hw = &adapter->hw;
7026 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7027 	u32 msg = E1000_VT_MSGTYPE_NACK;
7028 
7029 	/* if device isn't clear to send it shouldn't be reading either */
7030 	if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7031 	    time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7032 		igb_write_mbx(hw, &msg, 1, vf);
7033 		vf_data->last_nack = jiffies;
7034 	}
7035 }
7036 
7037 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7038 {
7039 	struct pci_dev *pdev = adapter->pdev;
7040 	u32 msgbuf[E1000_VFMAILBOX_SIZE];
7041 	struct e1000_hw *hw = &adapter->hw;
7042 	struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7043 	s32 retval;
7044 
7045 	retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7046 
7047 	if (retval) {
7048 		/* if receive failed revoke VF CTS stats and restart init */
7049 		dev_err(&pdev->dev, "Error receiving message from VF\n");
7050 		vf_data->flags &= ~IGB_VF_FLAG_CTS;
7051 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7052 			goto unlock;
7053 		goto out;
7054 	}
7055 
7056 	/* this is a message we already processed, do nothing */
7057 	if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7058 		goto unlock;
7059 
7060 	/* until the vf completes a reset it should not be
7061 	 * allowed to start any configuration.
7062 	 */
7063 	if (msgbuf[0] == E1000_VF_RESET) {
7064 		/* unlocks mailbox */
7065 		igb_vf_reset_msg(adapter, vf);
7066 		return;
7067 	}
7068 
7069 	if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7070 		if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7071 			goto unlock;
7072 		retval = -1;
7073 		goto out;
7074 	}
7075 
7076 	switch ((msgbuf[0] & 0xFFFF)) {
7077 	case E1000_VF_SET_MAC_ADDR:
7078 		retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7079 		break;
7080 	case E1000_VF_SET_PROMISC:
7081 		retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7082 		break;
7083 	case E1000_VF_SET_MULTICAST:
7084 		retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7085 		break;
7086 	case E1000_VF_SET_LPE:
7087 		retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7088 		break;
7089 	case E1000_VF_SET_VLAN:
7090 		retval = -1;
7091 		if (vf_data->pf_vlan)
7092 			dev_warn(&pdev->dev,
7093 				 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7094 				 vf);
7095 		else
7096 			retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7097 		break;
7098 	default:
7099 		dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7100 		retval = -1;
7101 		break;
7102 	}
7103 
7104 	msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7105 out:
7106 	/* notify the VF of the results of what it sent us */
7107 	if (retval)
7108 		msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7109 	else
7110 		msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7111 
7112 	/* unlocks mailbox */
7113 	igb_write_mbx(hw, msgbuf, 1, vf);
7114 	return;
7115 
7116 unlock:
7117 	igb_unlock_mbx(hw, vf);
7118 }
7119 
7120 static void igb_msg_task(struct igb_adapter *adapter)
7121 {
7122 	struct e1000_hw *hw = &adapter->hw;
7123 	u32 vf;
7124 
7125 	for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7126 		/* process any reset requests */
7127 		if (!igb_check_for_rst(hw, vf))
7128 			igb_vf_reset_event(adapter, vf);
7129 
7130 		/* process any messages pending */
7131 		if (!igb_check_for_msg(hw, vf))
7132 			igb_rcv_msg_from_vf(adapter, vf);
7133 
7134 		/* process any acks */
7135 		if (!igb_check_for_ack(hw, vf))
7136 			igb_rcv_ack_from_vf(adapter, vf);
7137 	}
7138 }
7139 
7140 /**
7141  *  igb_set_uta - Set unicast filter table address
7142  *  @adapter: board private structure
7143  *  @set: boolean indicating if we are setting or clearing bits
7144  *
7145  *  The unicast table address is a register array of 32-bit registers.
7146  *  The table is meant to be used in a way similar to how the MTA is used
7147  *  however due to certain limitations in the hardware it is necessary to
7148  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7149  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
7150  **/
7151 static void igb_set_uta(struct igb_adapter *adapter, bool set)
7152 {
7153 	struct e1000_hw *hw = &adapter->hw;
7154 	u32 uta = set ? ~0 : 0;
7155 	int i;
7156 
7157 	/* we only need to do this if VMDq is enabled */
7158 	if (!adapter->vfs_allocated_count)
7159 		return;
7160 
7161 	for (i = hw->mac.uta_reg_count; i--;)
7162 		array_wr32(E1000_UTA, i, uta);
7163 }
7164 
7165 /**
7166  *  igb_intr_msi - Interrupt Handler
7167  *  @irq: interrupt number
7168  *  @data: pointer to a network interface device structure
7169  **/
7170 static irqreturn_t igb_intr_msi(int irq, void *data)
7171 {
7172 	struct igb_adapter *adapter = data;
7173 	struct igb_q_vector *q_vector = adapter->q_vector[0];
7174 	struct e1000_hw *hw = &adapter->hw;
7175 	/* read ICR disables interrupts using IAM */
7176 	u32 icr = rd32(E1000_ICR);
7177 
7178 	igb_write_itr(q_vector);
7179 
7180 	if (icr & E1000_ICR_DRSTA)
7181 		schedule_work(&adapter->reset_task);
7182 
7183 	if (icr & E1000_ICR_DOUTSYNC) {
7184 		/* HW is reporting DMA is out of sync */
7185 		adapter->stats.doosync++;
7186 	}
7187 
7188 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7189 		hw->mac.get_link_status = 1;
7190 		if (!test_bit(__IGB_DOWN, &adapter->state))
7191 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
7192 	}
7193 
7194 	if (icr & E1000_ICR_TS)
7195 		igb_tsync_interrupt(adapter);
7196 
7197 	napi_schedule(&q_vector->napi);
7198 
7199 	return IRQ_HANDLED;
7200 }
7201 
7202 /**
7203  *  igb_intr - Legacy Interrupt Handler
7204  *  @irq: interrupt number
7205  *  @data: pointer to a network interface device structure
7206  **/
7207 static irqreturn_t igb_intr(int irq, void *data)
7208 {
7209 	struct igb_adapter *adapter = data;
7210 	struct igb_q_vector *q_vector = adapter->q_vector[0];
7211 	struct e1000_hw *hw = &adapter->hw;
7212 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
7213 	 * need for the IMC write
7214 	 */
7215 	u32 icr = rd32(E1000_ICR);
7216 
7217 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
7218 	 * not set, then the adapter didn't send an interrupt
7219 	 */
7220 	if (!(icr & E1000_ICR_INT_ASSERTED))
7221 		return IRQ_NONE;
7222 
7223 	igb_write_itr(q_vector);
7224 
7225 	if (icr & E1000_ICR_DRSTA)
7226 		schedule_work(&adapter->reset_task);
7227 
7228 	if (icr & E1000_ICR_DOUTSYNC) {
7229 		/* HW is reporting DMA is out of sync */
7230 		adapter->stats.doosync++;
7231 	}
7232 
7233 	if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7234 		hw->mac.get_link_status = 1;
7235 		/* guard against interrupt when we're going down */
7236 		if (!test_bit(__IGB_DOWN, &adapter->state))
7237 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
7238 	}
7239 
7240 	if (icr & E1000_ICR_TS)
7241 		igb_tsync_interrupt(adapter);
7242 
7243 	napi_schedule(&q_vector->napi);
7244 
7245 	return IRQ_HANDLED;
7246 }
7247 
7248 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
7249 {
7250 	struct igb_adapter *adapter = q_vector->adapter;
7251 	struct e1000_hw *hw = &adapter->hw;
7252 
7253 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
7254 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
7255 		if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
7256 			igb_set_itr(q_vector);
7257 		else
7258 			igb_update_ring_itr(q_vector);
7259 	}
7260 
7261 	if (!test_bit(__IGB_DOWN, &adapter->state)) {
7262 		if (adapter->flags & IGB_FLAG_HAS_MSIX)
7263 			wr32(E1000_EIMS, q_vector->eims_value);
7264 		else
7265 			igb_irq_enable(adapter);
7266 	}
7267 }
7268 
7269 /**
7270  *  igb_poll - NAPI Rx polling callback
7271  *  @napi: napi polling structure
7272  *  @budget: count of how many packets we should handle
7273  **/
7274 static int igb_poll(struct napi_struct *napi, int budget)
7275 {
7276 	struct igb_q_vector *q_vector = container_of(napi,
7277 						     struct igb_q_vector,
7278 						     napi);
7279 	bool clean_complete = true;
7280 	int work_done = 0;
7281 
7282 #ifdef CONFIG_IGB_DCA
7283 	if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
7284 		igb_update_dca(q_vector);
7285 #endif
7286 	if (q_vector->tx.ring)
7287 		clean_complete = igb_clean_tx_irq(q_vector, budget);
7288 
7289 	if (q_vector->rx.ring) {
7290 		int cleaned = igb_clean_rx_irq(q_vector, budget);
7291 
7292 		work_done += cleaned;
7293 		if (cleaned >= budget)
7294 			clean_complete = false;
7295 	}
7296 
7297 	/* If all work not completed, return budget and keep polling */
7298 	if (!clean_complete)
7299 		return budget;
7300 
7301 	/* If not enough Rx work done, exit the polling mode */
7302 	napi_complete_done(napi, work_done);
7303 	igb_ring_irq_enable(q_vector);
7304 
7305 	return 0;
7306 }
7307 
7308 /**
7309  *  igb_clean_tx_irq - Reclaim resources after transmit completes
7310  *  @q_vector: pointer to q_vector containing needed info
7311  *  @napi_budget: Used to determine if we are in netpoll
7312  *
7313  *  returns true if ring is completely cleaned
7314  **/
7315 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
7316 {
7317 	struct igb_adapter *adapter = q_vector->adapter;
7318 	struct igb_ring *tx_ring = q_vector->tx.ring;
7319 	struct igb_tx_buffer *tx_buffer;
7320 	union e1000_adv_tx_desc *tx_desc;
7321 	unsigned int total_bytes = 0, total_packets = 0;
7322 	unsigned int budget = q_vector->tx.work_limit;
7323 	unsigned int i = tx_ring->next_to_clean;
7324 
7325 	if (test_bit(__IGB_DOWN, &adapter->state))
7326 		return true;
7327 
7328 	tx_buffer = &tx_ring->tx_buffer_info[i];
7329 	tx_desc = IGB_TX_DESC(tx_ring, i);
7330 	i -= tx_ring->count;
7331 
7332 	do {
7333 		union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
7334 
7335 		/* if next_to_watch is not set then there is no work pending */
7336 		if (!eop_desc)
7337 			break;
7338 
7339 		/* prevent any other reads prior to eop_desc */
7340 		smp_rmb();
7341 
7342 		/* if DD is not set pending work has not been completed */
7343 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
7344 			break;
7345 
7346 		/* clear next_to_watch to prevent false hangs */
7347 		tx_buffer->next_to_watch = NULL;
7348 
7349 		/* update the statistics for this packet */
7350 		total_bytes += tx_buffer->bytecount;
7351 		total_packets += tx_buffer->gso_segs;
7352 
7353 		/* free the skb */
7354 		napi_consume_skb(tx_buffer->skb, napi_budget);
7355 
7356 		/* unmap skb header data */
7357 		dma_unmap_single(tx_ring->dev,
7358 				 dma_unmap_addr(tx_buffer, dma),
7359 				 dma_unmap_len(tx_buffer, len),
7360 				 DMA_TO_DEVICE);
7361 
7362 		/* clear tx_buffer data */
7363 		dma_unmap_len_set(tx_buffer, len, 0);
7364 
7365 		/* clear last DMA location and unmap remaining buffers */
7366 		while (tx_desc != eop_desc) {
7367 			tx_buffer++;
7368 			tx_desc++;
7369 			i++;
7370 			if (unlikely(!i)) {
7371 				i -= tx_ring->count;
7372 				tx_buffer = tx_ring->tx_buffer_info;
7373 				tx_desc = IGB_TX_DESC(tx_ring, 0);
7374 			}
7375 
7376 			/* unmap any remaining paged data */
7377 			if (dma_unmap_len(tx_buffer, len)) {
7378 				dma_unmap_page(tx_ring->dev,
7379 					       dma_unmap_addr(tx_buffer, dma),
7380 					       dma_unmap_len(tx_buffer, len),
7381 					       DMA_TO_DEVICE);
7382 				dma_unmap_len_set(tx_buffer, len, 0);
7383 			}
7384 		}
7385 
7386 		/* move us one more past the eop_desc for start of next pkt */
7387 		tx_buffer++;
7388 		tx_desc++;
7389 		i++;
7390 		if (unlikely(!i)) {
7391 			i -= tx_ring->count;
7392 			tx_buffer = tx_ring->tx_buffer_info;
7393 			tx_desc = IGB_TX_DESC(tx_ring, 0);
7394 		}
7395 
7396 		/* issue prefetch for next Tx descriptor */
7397 		prefetch(tx_desc);
7398 
7399 		/* update budget accounting */
7400 		budget--;
7401 	} while (likely(budget));
7402 
7403 	netdev_tx_completed_queue(txring_txq(tx_ring),
7404 				  total_packets, total_bytes);
7405 	i += tx_ring->count;
7406 	tx_ring->next_to_clean = i;
7407 	u64_stats_update_begin(&tx_ring->tx_syncp);
7408 	tx_ring->tx_stats.bytes += total_bytes;
7409 	tx_ring->tx_stats.packets += total_packets;
7410 	u64_stats_update_end(&tx_ring->tx_syncp);
7411 	q_vector->tx.total_bytes += total_bytes;
7412 	q_vector->tx.total_packets += total_packets;
7413 
7414 	if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
7415 		struct e1000_hw *hw = &adapter->hw;
7416 
7417 		/* Detect a transmit hang in hardware, this serializes the
7418 		 * check with the clearing of time_stamp and movement of i
7419 		 */
7420 		clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
7421 		if (tx_buffer->next_to_watch &&
7422 		    time_after(jiffies, tx_buffer->time_stamp +
7423 			       (adapter->tx_timeout_factor * HZ)) &&
7424 		    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
7425 
7426 			/* detected Tx unit hang */
7427 			dev_err(tx_ring->dev,
7428 				"Detected Tx Unit Hang\n"
7429 				"  Tx Queue             <%d>\n"
7430 				"  TDH                  <%x>\n"
7431 				"  TDT                  <%x>\n"
7432 				"  next_to_use          <%x>\n"
7433 				"  next_to_clean        <%x>\n"
7434 				"buffer_info[next_to_clean]\n"
7435 				"  time_stamp           <%lx>\n"
7436 				"  next_to_watch        <%p>\n"
7437 				"  jiffies              <%lx>\n"
7438 				"  desc.status          <%x>\n",
7439 				tx_ring->queue_index,
7440 				rd32(E1000_TDH(tx_ring->reg_idx)),
7441 				readl(tx_ring->tail),
7442 				tx_ring->next_to_use,
7443 				tx_ring->next_to_clean,
7444 				tx_buffer->time_stamp,
7445 				tx_buffer->next_to_watch,
7446 				jiffies,
7447 				tx_buffer->next_to_watch->wb.status);
7448 			netif_stop_subqueue(tx_ring->netdev,
7449 					    tx_ring->queue_index);
7450 
7451 			/* we are about to reset, no point in enabling stuff */
7452 			return true;
7453 		}
7454 	}
7455 
7456 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
7457 	if (unlikely(total_packets &&
7458 	    netif_carrier_ok(tx_ring->netdev) &&
7459 	    igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
7460 		/* Make sure that anybody stopping the queue after this
7461 		 * sees the new next_to_clean.
7462 		 */
7463 		smp_mb();
7464 		if (__netif_subqueue_stopped(tx_ring->netdev,
7465 					     tx_ring->queue_index) &&
7466 		    !(test_bit(__IGB_DOWN, &adapter->state))) {
7467 			netif_wake_subqueue(tx_ring->netdev,
7468 					    tx_ring->queue_index);
7469 
7470 			u64_stats_update_begin(&tx_ring->tx_syncp);
7471 			tx_ring->tx_stats.restart_queue++;
7472 			u64_stats_update_end(&tx_ring->tx_syncp);
7473 		}
7474 	}
7475 
7476 	return !!budget;
7477 }
7478 
7479 /**
7480  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
7481  *  @rx_ring: rx descriptor ring to store buffers on
7482  *  @old_buff: donor buffer to have page reused
7483  *
7484  *  Synchronizes page for reuse by the adapter
7485  **/
7486 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
7487 			      struct igb_rx_buffer *old_buff)
7488 {
7489 	struct igb_rx_buffer *new_buff;
7490 	u16 nta = rx_ring->next_to_alloc;
7491 
7492 	new_buff = &rx_ring->rx_buffer_info[nta];
7493 
7494 	/* update, and store next to alloc */
7495 	nta++;
7496 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
7497 
7498 	/* Transfer page from old buffer to new buffer.
7499 	 * Move each member individually to avoid possible store
7500 	 * forwarding stalls.
7501 	 */
7502 	new_buff->dma		= old_buff->dma;
7503 	new_buff->page		= old_buff->page;
7504 	new_buff->page_offset	= old_buff->page_offset;
7505 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
7506 }
7507 
7508 static inline bool igb_page_is_reserved(struct page *page)
7509 {
7510 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
7511 }
7512 
7513 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer)
7514 {
7515 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
7516 	struct page *page = rx_buffer->page;
7517 
7518 	/* avoid re-using remote pages */
7519 	if (unlikely(igb_page_is_reserved(page)))
7520 		return false;
7521 
7522 #if (PAGE_SIZE < 8192)
7523 	/* if we are only owner of page we can reuse it */
7524 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
7525 		return false;
7526 #else
7527 #define IGB_LAST_OFFSET \
7528 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
7529 
7530 	if (rx_buffer->page_offset > IGB_LAST_OFFSET)
7531 		return false;
7532 #endif
7533 
7534 	/* If we have drained the page fragment pool we need to update
7535 	 * the pagecnt_bias and page count so that we fully restock the
7536 	 * number of references the driver holds.
7537 	 */
7538 	if (unlikely(!pagecnt_bias)) {
7539 		page_ref_add(page, USHRT_MAX);
7540 		rx_buffer->pagecnt_bias = USHRT_MAX;
7541 	}
7542 
7543 	return true;
7544 }
7545 
7546 /**
7547  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
7548  *  @rx_ring: rx descriptor ring to transact packets on
7549  *  @rx_buffer: buffer containing page to add
7550  *  @skb: sk_buff to place the data into
7551  *  @size: size of buffer to be added
7552  *
7553  *  This function will add the data contained in rx_buffer->page to the skb.
7554  **/
7555 static void igb_add_rx_frag(struct igb_ring *rx_ring,
7556 			    struct igb_rx_buffer *rx_buffer,
7557 			    struct sk_buff *skb,
7558 			    unsigned int size)
7559 {
7560 #if (PAGE_SIZE < 8192)
7561 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7562 #else
7563 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
7564 				SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
7565 				SKB_DATA_ALIGN(size);
7566 #endif
7567 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
7568 			rx_buffer->page_offset, size, truesize);
7569 #if (PAGE_SIZE < 8192)
7570 	rx_buffer->page_offset ^= truesize;
7571 #else
7572 	rx_buffer->page_offset += truesize;
7573 #endif
7574 }
7575 
7576 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
7577 					 struct igb_rx_buffer *rx_buffer,
7578 					 union e1000_adv_rx_desc *rx_desc,
7579 					 unsigned int size)
7580 {
7581 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7582 #if (PAGE_SIZE < 8192)
7583 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7584 #else
7585 	unsigned int truesize = SKB_DATA_ALIGN(size);
7586 #endif
7587 	unsigned int headlen;
7588 	struct sk_buff *skb;
7589 
7590 	/* prefetch first cache line of first page */
7591 	prefetch(va);
7592 #if L1_CACHE_BYTES < 128
7593 	prefetch(va + L1_CACHE_BYTES);
7594 #endif
7595 
7596 	/* allocate a skb to store the frags */
7597 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
7598 	if (unlikely(!skb))
7599 		return NULL;
7600 
7601 	if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
7602 		igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
7603 		va += IGB_TS_HDR_LEN;
7604 		size -= IGB_TS_HDR_LEN;
7605 	}
7606 
7607 	/* Determine available headroom for copy */
7608 	headlen = size;
7609 	if (headlen > IGB_RX_HDR_LEN)
7610 		headlen = eth_get_headlen(va, IGB_RX_HDR_LEN);
7611 
7612 	/* align pull length to size of long to optimize memcpy performance */
7613 	memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
7614 
7615 	/* update all of the pointers */
7616 	size -= headlen;
7617 	if (size) {
7618 		skb_add_rx_frag(skb, 0, rx_buffer->page,
7619 				(va + headlen) - page_address(rx_buffer->page),
7620 				size, truesize);
7621 #if (PAGE_SIZE < 8192)
7622 		rx_buffer->page_offset ^= truesize;
7623 #else
7624 		rx_buffer->page_offset += truesize;
7625 #endif
7626 	} else {
7627 		rx_buffer->pagecnt_bias++;
7628 	}
7629 
7630 	return skb;
7631 }
7632 
7633 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
7634 				     struct igb_rx_buffer *rx_buffer,
7635 				     union e1000_adv_rx_desc *rx_desc,
7636 				     unsigned int size)
7637 {
7638 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7639 #if (PAGE_SIZE < 8192)
7640 	unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7641 #else
7642 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
7643 				SKB_DATA_ALIGN(IGB_SKB_PAD + size);
7644 #endif
7645 	struct sk_buff *skb;
7646 
7647 	/* prefetch first cache line of first page */
7648 	prefetch(va);
7649 #if L1_CACHE_BYTES < 128
7650 	prefetch(va + L1_CACHE_BYTES);
7651 #endif
7652 
7653 	/* build an skb around the page buffer */
7654 	skb = build_skb(va - IGB_SKB_PAD, truesize);
7655 	if (unlikely(!skb))
7656 		return NULL;
7657 
7658 	/* update pointers within the skb to store the data */
7659 	skb_reserve(skb, IGB_SKB_PAD);
7660 	__skb_put(skb, size);
7661 
7662 	/* pull timestamp out of packet data */
7663 	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
7664 		igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
7665 		__skb_pull(skb, IGB_TS_HDR_LEN);
7666 	}
7667 
7668 	/* update buffer offset */
7669 #if (PAGE_SIZE < 8192)
7670 	rx_buffer->page_offset ^= truesize;
7671 #else
7672 	rx_buffer->page_offset += truesize;
7673 #endif
7674 
7675 	return skb;
7676 }
7677 
7678 static inline void igb_rx_checksum(struct igb_ring *ring,
7679 				   union e1000_adv_rx_desc *rx_desc,
7680 				   struct sk_buff *skb)
7681 {
7682 	skb_checksum_none_assert(skb);
7683 
7684 	/* Ignore Checksum bit is set */
7685 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
7686 		return;
7687 
7688 	/* Rx checksum disabled via ethtool */
7689 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
7690 		return;
7691 
7692 	/* TCP/UDP checksum error bit is set */
7693 	if (igb_test_staterr(rx_desc,
7694 			     E1000_RXDEXT_STATERR_TCPE |
7695 			     E1000_RXDEXT_STATERR_IPE)) {
7696 		/* work around errata with sctp packets where the TCPE aka
7697 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
7698 		 * packets, (aka let the stack check the crc32c)
7699 		 */
7700 		if (!((skb->len == 60) &&
7701 		      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
7702 			u64_stats_update_begin(&ring->rx_syncp);
7703 			ring->rx_stats.csum_err++;
7704 			u64_stats_update_end(&ring->rx_syncp);
7705 		}
7706 		/* let the stack verify checksum errors */
7707 		return;
7708 	}
7709 	/* It must be a TCP or UDP packet with a valid checksum */
7710 	if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
7711 				      E1000_RXD_STAT_UDPCS))
7712 		skb->ip_summed = CHECKSUM_UNNECESSARY;
7713 
7714 	dev_dbg(ring->dev, "cksum success: bits %08X\n",
7715 		le32_to_cpu(rx_desc->wb.upper.status_error));
7716 }
7717 
7718 static inline void igb_rx_hash(struct igb_ring *ring,
7719 			       union e1000_adv_rx_desc *rx_desc,
7720 			       struct sk_buff *skb)
7721 {
7722 	if (ring->netdev->features & NETIF_F_RXHASH)
7723 		skb_set_hash(skb,
7724 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
7725 			     PKT_HASH_TYPE_L3);
7726 }
7727 
7728 /**
7729  *  igb_is_non_eop - process handling of non-EOP buffers
7730  *  @rx_ring: Rx ring being processed
7731  *  @rx_desc: Rx descriptor for current buffer
7732  *  @skb: current socket buffer containing buffer in progress
7733  *
7734  *  This function updates next to clean.  If the buffer is an EOP buffer
7735  *  this function exits returning false, otherwise it will place the
7736  *  sk_buff in the next buffer to be chained and return true indicating
7737  *  that this is in fact a non-EOP buffer.
7738  **/
7739 static bool igb_is_non_eop(struct igb_ring *rx_ring,
7740 			   union e1000_adv_rx_desc *rx_desc)
7741 {
7742 	u32 ntc = rx_ring->next_to_clean + 1;
7743 
7744 	/* fetch, update, and store next to clean */
7745 	ntc = (ntc < rx_ring->count) ? ntc : 0;
7746 	rx_ring->next_to_clean = ntc;
7747 
7748 	prefetch(IGB_RX_DESC(rx_ring, ntc));
7749 
7750 	if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
7751 		return false;
7752 
7753 	return true;
7754 }
7755 
7756 /**
7757  *  igb_cleanup_headers - Correct corrupted or empty headers
7758  *  @rx_ring: rx descriptor ring packet is being transacted on
7759  *  @rx_desc: pointer to the EOP Rx descriptor
7760  *  @skb: pointer to current skb being fixed
7761  *
7762  *  Address the case where we are pulling data in on pages only
7763  *  and as such no data is present in the skb header.
7764  *
7765  *  In addition if skb is not at least 60 bytes we need to pad it so that
7766  *  it is large enough to qualify as a valid Ethernet frame.
7767  *
7768  *  Returns true if an error was encountered and skb was freed.
7769  **/
7770 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
7771 				union e1000_adv_rx_desc *rx_desc,
7772 				struct sk_buff *skb)
7773 {
7774 	if (unlikely((igb_test_staterr(rx_desc,
7775 				       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
7776 		struct net_device *netdev = rx_ring->netdev;
7777 		if (!(netdev->features & NETIF_F_RXALL)) {
7778 			dev_kfree_skb_any(skb);
7779 			return true;
7780 		}
7781 	}
7782 
7783 	/* if eth_skb_pad returns an error the skb was freed */
7784 	if (eth_skb_pad(skb))
7785 		return true;
7786 
7787 	return false;
7788 }
7789 
7790 /**
7791  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
7792  *  @rx_ring: rx descriptor ring packet is being transacted on
7793  *  @rx_desc: pointer to the EOP Rx descriptor
7794  *  @skb: pointer to current skb being populated
7795  *
7796  *  This function checks the ring, descriptor, and packet information in
7797  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
7798  *  other fields within the skb.
7799  **/
7800 static void igb_process_skb_fields(struct igb_ring *rx_ring,
7801 				   union e1000_adv_rx_desc *rx_desc,
7802 				   struct sk_buff *skb)
7803 {
7804 	struct net_device *dev = rx_ring->netdev;
7805 
7806 	igb_rx_hash(rx_ring, rx_desc, skb);
7807 
7808 	igb_rx_checksum(rx_ring, rx_desc, skb);
7809 
7810 	if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
7811 	    !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
7812 		igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
7813 
7814 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
7815 	    igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
7816 		u16 vid;
7817 
7818 		if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
7819 		    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
7820 			vid = be16_to_cpu(rx_desc->wb.upper.vlan);
7821 		else
7822 			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
7823 
7824 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
7825 	}
7826 
7827 	skb_record_rx_queue(skb, rx_ring->queue_index);
7828 
7829 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
7830 }
7831 
7832 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
7833 					       const unsigned int size)
7834 {
7835 	struct igb_rx_buffer *rx_buffer;
7836 
7837 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
7838 	prefetchw(rx_buffer->page);
7839 
7840 	/* we are reusing so sync this buffer for CPU use */
7841 	dma_sync_single_range_for_cpu(rx_ring->dev,
7842 				      rx_buffer->dma,
7843 				      rx_buffer->page_offset,
7844 				      size,
7845 				      DMA_FROM_DEVICE);
7846 
7847 	rx_buffer->pagecnt_bias--;
7848 
7849 	return rx_buffer;
7850 }
7851 
7852 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
7853 			      struct igb_rx_buffer *rx_buffer)
7854 {
7855 	if (igb_can_reuse_rx_page(rx_buffer)) {
7856 		/* hand second half of page back to the ring */
7857 		igb_reuse_rx_page(rx_ring, rx_buffer);
7858 	} else {
7859 		/* We are not reusing the buffer so unmap it and free
7860 		 * any references we are holding to it
7861 		 */
7862 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
7863 				     igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
7864 				     IGB_RX_DMA_ATTR);
7865 		__page_frag_cache_drain(rx_buffer->page,
7866 					rx_buffer->pagecnt_bias);
7867 	}
7868 
7869 	/* clear contents of rx_buffer */
7870 	rx_buffer->page = NULL;
7871 }
7872 
7873 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
7874 {
7875 	struct igb_ring *rx_ring = q_vector->rx.ring;
7876 	struct sk_buff *skb = rx_ring->skb;
7877 	unsigned int total_bytes = 0, total_packets = 0;
7878 	u16 cleaned_count = igb_desc_unused(rx_ring);
7879 
7880 	while (likely(total_packets < budget)) {
7881 		union e1000_adv_rx_desc *rx_desc;
7882 		struct igb_rx_buffer *rx_buffer;
7883 		unsigned int size;
7884 
7885 		/* return some buffers to hardware, one at a time is too slow */
7886 		if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
7887 			igb_alloc_rx_buffers(rx_ring, cleaned_count);
7888 			cleaned_count = 0;
7889 		}
7890 
7891 		rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
7892 		size = le16_to_cpu(rx_desc->wb.upper.length);
7893 		if (!size)
7894 			break;
7895 
7896 		/* This memory barrier is needed to keep us from reading
7897 		 * any other fields out of the rx_desc until we know the
7898 		 * descriptor has been written back
7899 		 */
7900 		dma_rmb();
7901 
7902 		rx_buffer = igb_get_rx_buffer(rx_ring, size);
7903 
7904 		/* retrieve a buffer from the ring */
7905 		if (skb)
7906 			igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
7907 		else if (ring_uses_build_skb(rx_ring))
7908 			skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size);
7909 		else
7910 			skb = igb_construct_skb(rx_ring, rx_buffer,
7911 						rx_desc, size);
7912 
7913 		/* exit if we failed to retrieve a buffer */
7914 		if (!skb) {
7915 			rx_ring->rx_stats.alloc_failed++;
7916 			rx_buffer->pagecnt_bias++;
7917 			break;
7918 		}
7919 
7920 		igb_put_rx_buffer(rx_ring, rx_buffer);
7921 		cleaned_count++;
7922 
7923 		/* fetch next buffer in frame if non-eop */
7924 		if (igb_is_non_eop(rx_ring, rx_desc))
7925 			continue;
7926 
7927 		/* verify the packet layout is correct */
7928 		if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
7929 			skb = NULL;
7930 			continue;
7931 		}
7932 
7933 		/* probably a little skewed due to removing CRC */
7934 		total_bytes += skb->len;
7935 
7936 		/* populate checksum, timestamp, VLAN, and protocol */
7937 		igb_process_skb_fields(rx_ring, rx_desc, skb);
7938 
7939 		napi_gro_receive(&q_vector->napi, skb);
7940 
7941 		/* reset skb pointer */
7942 		skb = NULL;
7943 
7944 		/* update budget accounting */
7945 		total_packets++;
7946 	}
7947 
7948 	/* place incomplete frames back on ring for completion */
7949 	rx_ring->skb = skb;
7950 
7951 	u64_stats_update_begin(&rx_ring->rx_syncp);
7952 	rx_ring->rx_stats.packets += total_packets;
7953 	rx_ring->rx_stats.bytes += total_bytes;
7954 	u64_stats_update_end(&rx_ring->rx_syncp);
7955 	q_vector->rx.total_packets += total_packets;
7956 	q_vector->rx.total_bytes += total_bytes;
7957 
7958 	if (cleaned_count)
7959 		igb_alloc_rx_buffers(rx_ring, cleaned_count);
7960 
7961 	return total_packets;
7962 }
7963 
7964 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
7965 {
7966 	return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
7967 }
7968 
7969 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
7970 				  struct igb_rx_buffer *bi)
7971 {
7972 	struct page *page = bi->page;
7973 	dma_addr_t dma;
7974 
7975 	/* since we are recycling buffers we should seldom need to alloc */
7976 	if (likely(page))
7977 		return true;
7978 
7979 	/* alloc new page for storage */
7980 	page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
7981 	if (unlikely(!page)) {
7982 		rx_ring->rx_stats.alloc_failed++;
7983 		return false;
7984 	}
7985 
7986 	/* map page for use */
7987 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
7988 				 igb_rx_pg_size(rx_ring),
7989 				 DMA_FROM_DEVICE,
7990 				 IGB_RX_DMA_ATTR);
7991 
7992 	/* if mapping failed free memory back to system since
7993 	 * there isn't much point in holding memory we can't use
7994 	 */
7995 	if (dma_mapping_error(rx_ring->dev, dma)) {
7996 		__free_pages(page, igb_rx_pg_order(rx_ring));
7997 
7998 		rx_ring->rx_stats.alloc_failed++;
7999 		return false;
8000 	}
8001 
8002 	bi->dma = dma;
8003 	bi->page = page;
8004 	bi->page_offset = igb_rx_offset(rx_ring);
8005 	bi->pagecnt_bias = 1;
8006 
8007 	return true;
8008 }
8009 
8010 /**
8011  *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
8012  *  @adapter: address of board private structure
8013  **/
8014 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
8015 {
8016 	union e1000_adv_rx_desc *rx_desc;
8017 	struct igb_rx_buffer *bi;
8018 	u16 i = rx_ring->next_to_use;
8019 	u16 bufsz;
8020 
8021 	/* nothing to do */
8022 	if (!cleaned_count)
8023 		return;
8024 
8025 	rx_desc = IGB_RX_DESC(rx_ring, i);
8026 	bi = &rx_ring->rx_buffer_info[i];
8027 	i -= rx_ring->count;
8028 
8029 	bufsz = igb_rx_bufsz(rx_ring);
8030 
8031 	do {
8032 		if (!igb_alloc_mapped_page(rx_ring, bi))
8033 			break;
8034 
8035 		/* sync the buffer for use by the device */
8036 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
8037 						 bi->page_offset, bufsz,
8038 						 DMA_FROM_DEVICE);
8039 
8040 		/* Refresh the desc even if buffer_addrs didn't change
8041 		 * because each write-back erases this info.
8042 		 */
8043 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
8044 
8045 		rx_desc++;
8046 		bi++;
8047 		i++;
8048 		if (unlikely(!i)) {
8049 			rx_desc = IGB_RX_DESC(rx_ring, 0);
8050 			bi = rx_ring->rx_buffer_info;
8051 			i -= rx_ring->count;
8052 		}
8053 
8054 		/* clear the length for the next_to_use descriptor */
8055 		rx_desc->wb.upper.length = 0;
8056 
8057 		cleaned_count--;
8058 	} while (cleaned_count);
8059 
8060 	i += rx_ring->count;
8061 
8062 	if (rx_ring->next_to_use != i) {
8063 		/* record the next descriptor to use */
8064 		rx_ring->next_to_use = i;
8065 
8066 		/* update next to alloc since we have filled the ring */
8067 		rx_ring->next_to_alloc = i;
8068 
8069 		/* Force memory writes to complete before letting h/w
8070 		 * know there are new descriptors to fetch.  (Only
8071 		 * applicable for weak-ordered memory model archs,
8072 		 * such as IA-64).
8073 		 */
8074 		wmb();
8075 		writel(i, rx_ring->tail);
8076 	}
8077 }
8078 
8079 /**
8080  * igb_mii_ioctl -
8081  * @netdev:
8082  * @ifreq:
8083  * @cmd:
8084  **/
8085 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8086 {
8087 	struct igb_adapter *adapter = netdev_priv(netdev);
8088 	struct mii_ioctl_data *data = if_mii(ifr);
8089 
8090 	if (adapter->hw.phy.media_type != e1000_media_type_copper)
8091 		return -EOPNOTSUPP;
8092 
8093 	switch (cmd) {
8094 	case SIOCGMIIPHY:
8095 		data->phy_id = adapter->hw.phy.addr;
8096 		break;
8097 	case SIOCGMIIREG:
8098 		if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
8099 				     &data->val_out))
8100 			return -EIO;
8101 		break;
8102 	case SIOCSMIIREG:
8103 	default:
8104 		return -EOPNOTSUPP;
8105 	}
8106 	return 0;
8107 }
8108 
8109 /**
8110  * igb_ioctl -
8111  * @netdev:
8112  * @ifreq:
8113  * @cmd:
8114  **/
8115 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8116 {
8117 	switch (cmd) {
8118 	case SIOCGMIIPHY:
8119 	case SIOCGMIIREG:
8120 	case SIOCSMIIREG:
8121 		return igb_mii_ioctl(netdev, ifr, cmd);
8122 	case SIOCGHWTSTAMP:
8123 		return igb_ptp_get_ts_config(netdev, ifr);
8124 	case SIOCSHWTSTAMP:
8125 		return igb_ptp_set_ts_config(netdev, ifr);
8126 	default:
8127 		return -EOPNOTSUPP;
8128 	}
8129 }
8130 
8131 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8132 {
8133 	struct igb_adapter *adapter = hw->back;
8134 
8135 	pci_read_config_word(adapter->pdev, reg, value);
8136 }
8137 
8138 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8139 {
8140 	struct igb_adapter *adapter = hw->back;
8141 
8142 	pci_write_config_word(adapter->pdev, reg, *value);
8143 }
8144 
8145 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8146 {
8147 	struct igb_adapter *adapter = hw->back;
8148 
8149 	if (pcie_capability_read_word(adapter->pdev, reg, value))
8150 		return -E1000_ERR_CONFIG;
8151 
8152 	return 0;
8153 }
8154 
8155 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8156 {
8157 	struct igb_adapter *adapter = hw->back;
8158 
8159 	if (pcie_capability_write_word(adapter->pdev, reg, *value))
8160 		return -E1000_ERR_CONFIG;
8161 
8162 	return 0;
8163 }
8164 
8165 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
8166 {
8167 	struct igb_adapter *adapter = netdev_priv(netdev);
8168 	struct e1000_hw *hw = &adapter->hw;
8169 	u32 ctrl, rctl;
8170 	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
8171 
8172 	if (enable) {
8173 		/* enable VLAN tag insert/strip */
8174 		ctrl = rd32(E1000_CTRL);
8175 		ctrl |= E1000_CTRL_VME;
8176 		wr32(E1000_CTRL, ctrl);
8177 
8178 		/* Disable CFI check */
8179 		rctl = rd32(E1000_RCTL);
8180 		rctl &= ~E1000_RCTL_CFIEN;
8181 		wr32(E1000_RCTL, rctl);
8182 	} else {
8183 		/* disable VLAN tag insert/strip */
8184 		ctrl = rd32(E1000_CTRL);
8185 		ctrl &= ~E1000_CTRL_VME;
8186 		wr32(E1000_CTRL, ctrl);
8187 	}
8188 
8189 	igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
8190 }
8191 
8192 static int igb_vlan_rx_add_vid(struct net_device *netdev,
8193 			       __be16 proto, u16 vid)
8194 {
8195 	struct igb_adapter *adapter = netdev_priv(netdev);
8196 	struct e1000_hw *hw = &adapter->hw;
8197 	int pf_id = adapter->vfs_allocated_count;
8198 
8199 	/* add the filter since PF can receive vlans w/o entry in vlvf */
8200 	if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8201 		igb_vfta_set(hw, vid, pf_id, true, !!vid);
8202 
8203 	set_bit(vid, adapter->active_vlans);
8204 
8205 	return 0;
8206 }
8207 
8208 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
8209 				__be16 proto, u16 vid)
8210 {
8211 	struct igb_adapter *adapter = netdev_priv(netdev);
8212 	int pf_id = adapter->vfs_allocated_count;
8213 	struct e1000_hw *hw = &adapter->hw;
8214 
8215 	/* remove VID from filter table */
8216 	if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8217 		igb_vfta_set(hw, vid, pf_id, false, true);
8218 
8219 	clear_bit(vid, adapter->active_vlans);
8220 
8221 	return 0;
8222 }
8223 
8224 static void igb_restore_vlan(struct igb_adapter *adapter)
8225 {
8226 	u16 vid = 1;
8227 
8228 	igb_vlan_mode(adapter->netdev, adapter->netdev->features);
8229 	igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
8230 
8231 	for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
8232 		igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
8233 }
8234 
8235 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
8236 {
8237 	struct pci_dev *pdev = adapter->pdev;
8238 	struct e1000_mac_info *mac = &adapter->hw.mac;
8239 
8240 	mac->autoneg = 0;
8241 
8242 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
8243 	 * for the switch() below to work
8244 	 */
8245 	if ((spd & 1) || (dplx & ~1))
8246 		goto err_inval;
8247 
8248 	/* Fiber NIC's only allow 1000 gbps Full duplex
8249 	 * and 100Mbps Full duplex for 100baseFx sfp
8250 	 */
8251 	if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
8252 		switch (spd + dplx) {
8253 		case SPEED_10 + DUPLEX_HALF:
8254 		case SPEED_10 + DUPLEX_FULL:
8255 		case SPEED_100 + DUPLEX_HALF:
8256 			goto err_inval;
8257 		default:
8258 			break;
8259 		}
8260 	}
8261 
8262 	switch (spd + dplx) {
8263 	case SPEED_10 + DUPLEX_HALF:
8264 		mac->forced_speed_duplex = ADVERTISE_10_HALF;
8265 		break;
8266 	case SPEED_10 + DUPLEX_FULL:
8267 		mac->forced_speed_duplex = ADVERTISE_10_FULL;
8268 		break;
8269 	case SPEED_100 + DUPLEX_HALF:
8270 		mac->forced_speed_duplex = ADVERTISE_100_HALF;
8271 		break;
8272 	case SPEED_100 + DUPLEX_FULL:
8273 		mac->forced_speed_duplex = ADVERTISE_100_FULL;
8274 		break;
8275 	case SPEED_1000 + DUPLEX_FULL:
8276 		mac->autoneg = 1;
8277 		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
8278 		break;
8279 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
8280 	default:
8281 		goto err_inval;
8282 	}
8283 
8284 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
8285 	adapter->hw.phy.mdix = AUTO_ALL_MODES;
8286 
8287 	return 0;
8288 
8289 err_inval:
8290 	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
8291 	return -EINVAL;
8292 }
8293 
8294 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
8295 			  bool runtime)
8296 {
8297 	struct net_device *netdev = pci_get_drvdata(pdev);
8298 	struct igb_adapter *adapter = netdev_priv(netdev);
8299 	struct e1000_hw *hw = &adapter->hw;
8300 	u32 ctrl, rctl, status;
8301 	u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
8302 #ifdef CONFIG_PM
8303 	int retval = 0;
8304 #endif
8305 
8306 	rtnl_lock();
8307 	netif_device_detach(netdev);
8308 
8309 	if (netif_running(netdev))
8310 		__igb_close(netdev, true);
8311 
8312 	igb_ptp_suspend(adapter);
8313 
8314 	igb_clear_interrupt_scheme(adapter);
8315 	rtnl_unlock();
8316 
8317 #ifdef CONFIG_PM
8318 	retval = pci_save_state(pdev);
8319 	if (retval)
8320 		return retval;
8321 #endif
8322 
8323 	status = rd32(E1000_STATUS);
8324 	if (status & E1000_STATUS_LU)
8325 		wufc &= ~E1000_WUFC_LNKC;
8326 
8327 	if (wufc) {
8328 		igb_setup_rctl(adapter);
8329 		igb_set_rx_mode(netdev);
8330 
8331 		/* turn on all-multi mode if wake on multicast is enabled */
8332 		if (wufc & E1000_WUFC_MC) {
8333 			rctl = rd32(E1000_RCTL);
8334 			rctl |= E1000_RCTL_MPE;
8335 			wr32(E1000_RCTL, rctl);
8336 		}
8337 
8338 		ctrl = rd32(E1000_CTRL);
8339 		/* advertise wake from D3Cold */
8340 		#define E1000_CTRL_ADVD3WUC 0x00100000
8341 		/* phy power management enable */
8342 		#define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
8343 		ctrl |= E1000_CTRL_ADVD3WUC;
8344 		wr32(E1000_CTRL, ctrl);
8345 
8346 		/* Allow time for pending master requests to run */
8347 		igb_disable_pcie_master(hw);
8348 
8349 		wr32(E1000_WUC, E1000_WUC_PME_EN);
8350 		wr32(E1000_WUFC, wufc);
8351 	} else {
8352 		wr32(E1000_WUC, 0);
8353 		wr32(E1000_WUFC, 0);
8354 	}
8355 
8356 	*enable_wake = wufc || adapter->en_mng_pt;
8357 	if (!*enable_wake)
8358 		igb_power_down_link(adapter);
8359 	else
8360 		igb_power_up_link(adapter);
8361 
8362 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
8363 	 * would have already happened in close and is redundant.
8364 	 */
8365 	igb_release_hw_control(adapter);
8366 
8367 	pci_disable_device(pdev);
8368 
8369 	return 0;
8370 }
8371 
8372 static void igb_deliver_wake_packet(struct net_device *netdev)
8373 {
8374 	struct igb_adapter *adapter = netdev_priv(netdev);
8375 	struct e1000_hw *hw = &adapter->hw;
8376 	struct sk_buff *skb;
8377 	u32 wupl;
8378 
8379 	wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
8380 
8381 	/* WUPM stores only the first 128 bytes of the wake packet.
8382 	 * Read the packet only if we have the whole thing.
8383 	 */
8384 	if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
8385 		return;
8386 
8387 	skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
8388 	if (!skb)
8389 		return;
8390 
8391 	skb_put(skb, wupl);
8392 
8393 	/* Ensure reads are 32-bit aligned */
8394 	wupl = roundup(wupl, 4);
8395 
8396 	memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
8397 
8398 	skb->protocol = eth_type_trans(skb, netdev);
8399 	netif_rx(skb);
8400 }
8401 
8402 static int __maybe_unused igb_suspend(struct device *dev)
8403 {
8404 	int retval;
8405 	bool wake;
8406 	struct pci_dev *pdev = to_pci_dev(dev);
8407 
8408 	retval = __igb_shutdown(pdev, &wake, 0);
8409 	if (retval)
8410 		return retval;
8411 
8412 	if (wake) {
8413 		pci_prepare_to_sleep(pdev);
8414 	} else {
8415 		pci_wake_from_d3(pdev, false);
8416 		pci_set_power_state(pdev, PCI_D3hot);
8417 	}
8418 
8419 	return 0;
8420 }
8421 
8422 static int __maybe_unused igb_resume(struct device *dev)
8423 {
8424 	struct pci_dev *pdev = to_pci_dev(dev);
8425 	struct net_device *netdev = pci_get_drvdata(pdev);
8426 	struct igb_adapter *adapter = netdev_priv(netdev);
8427 	struct e1000_hw *hw = &adapter->hw;
8428 	u32 err, val;
8429 
8430 	pci_set_power_state(pdev, PCI_D0);
8431 	pci_restore_state(pdev);
8432 	pci_save_state(pdev);
8433 
8434 	if (!pci_device_is_present(pdev))
8435 		return -ENODEV;
8436 	err = pci_enable_device_mem(pdev);
8437 	if (err) {
8438 		dev_err(&pdev->dev,
8439 			"igb: Cannot enable PCI device from suspend\n");
8440 		return err;
8441 	}
8442 	pci_set_master(pdev);
8443 
8444 	pci_enable_wake(pdev, PCI_D3hot, 0);
8445 	pci_enable_wake(pdev, PCI_D3cold, 0);
8446 
8447 	if (igb_init_interrupt_scheme(adapter, true)) {
8448 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8449 		return -ENOMEM;
8450 	}
8451 
8452 	igb_reset(adapter);
8453 
8454 	/* let the f/w know that the h/w is now under the control of the
8455 	 * driver.
8456 	 */
8457 	igb_get_hw_control(adapter);
8458 
8459 	val = rd32(E1000_WUS);
8460 	if (val & WAKE_PKT_WUS)
8461 		igb_deliver_wake_packet(netdev);
8462 
8463 	wr32(E1000_WUS, ~0);
8464 
8465 	rtnl_lock();
8466 	if (!err && netif_running(netdev))
8467 		err = __igb_open(netdev, true);
8468 
8469 	if (!err)
8470 		netif_device_attach(netdev);
8471 	rtnl_unlock();
8472 
8473 	return err;
8474 }
8475 
8476 static int __maybe_unused igb_runtime_idle(struct device *dev)
8477 {
8478 	struct pci_dev *pdev = to_pci_dev(dev);
8479 	struct net_device *netdev = pci_get_drvdata(pdev);
8480 	struct igb_adapter *adapter = netdev_priv(netdev);
8481 
8482 	if (!igb_has_link(adapter))
8483 		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
8484 
8485 	return -EBUSY;
8486 }
8487 
8488 static int __maybe_unused igb_runtime_suspend(struct device *dev)
8489 {
8490 	struct pci_dev *pdev = to_pci_dev(dev);
8491 	int retval;
8492 	bool wake;
8493 
8494 	retval = __igb_shutdown(pdev, &wake, 1);
8495 	if (retval)
8496 		return retval;
8497 
8498 	if (wake) {
8499 		pci_prepare_to_sleep(pdev);
8500 	} else {
8501 		pci_wake_from_d3(pdev, false);
8502 		pci_set_power_state(pdev, PCI_D3hot);
8503 	}
8504 
8505 	return 0;
8506 }
8507 
8508 static int __maybe_unused igb_runtime_resume(struct device *dev)
8509 {
8510 	return igb_resume(dev);
8511 }
8512 
8513 static void igb_shutdown(struct pci_dev *pdev)
8514 {
8515 	bool wake;
8516 
8517 	__igb_shutdown(pdev, &wake, 0);
8518 
8519 	if (system_state == SYSTEM_POWER_OFF) {
8520 		pci_wake_from_d3(pdev, wake);
8521 		pci_set_power_state(pdev, PCI_D3hot);
8522 	}
8523 }
8524 
8525 #ifdef CONFIG_PCI_IOV
8526 static int igb_sriov_reinit(struct pci_dev *dev)
8527 {
8528 	struct net_device *netdev = pci_get_drvdata(dev);
8529 	struct igb_adapter *adapter = netdev_priv(netdev);
8530 	struct pci_dev *pdev = adapter->pdev;
8531 
8532 	rtnl_lock();
8533 
8534 	if (netif_running(netdev))
8535 		igb_close(netdev);
8536 	else
8537 		igb_reset(adapter);
8538 
8539 	igb_clear_interrupt_scheme(adapter);
8540 
8541 	igb_init_queue_configuration(adapter);
8542 
8543 	if (igb_init_interrupt_scheme(adapter, true)) {
8544 		rtnl_unlock();
8545 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8546 		return -ENOMEM;
8547 	}
8548 
8549 	if (netif_running(netdev))
8550 		igb_open(netdev);
8551 
8552 	rtnl_unlock();
8553 
8554 	return 0;
8555 }
8556 
8557 static int igb_pci_disable_sriov(struct pci_dev *dev)
8558 {
8559 	int err = igb_disable_sriov(dev);
8560 
8561 	if (!err)
8562 		err = igb_sriov_reinit(dev);
8563 
8564 	return err;
8565 }
8566 
8567 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
8568 {
8569 	int err = igb_enable_sriov(dev, num_vfs);
8570 
8571 	if (err)
8572 		goto out;
8573 
8574 	err = igb_sriov_reinit(dev);
8575 	if (!err)
8576 		return num_vfs;
8577 
8578 out:
8579 	return err;
8580 }
8581 
8582 #endif
8583 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
8584 {
8585 #ifdef CONFIG_PCI_IOV
8586 	if (num_vfs == 0)
8587 		return igb_pci_disable_sriov(dev);
8588 	else
8589 		return igb_pci_enable_sriov(dev, num_vfs);
8590 #endif
8591 	return 0;
8592 }
8593 
8594 #ifdef CONFIG_NET_POLL_CONTROLLER
8595 /* Polling 'interrupt' - used by things like netconsole to send skbs
8596  * without having to re-enable interrupts. It's not called while
8597  * the interrupt routine is executing.
8598  */
8599 static void igb_netpoll(struct net_device *netdev)
8600 {
8601 	struct igb_adapter *adapter = netdev_priv(netdev);
8602 	struct e1000_hw *hw = &adapter->hw;
8603 	struct igb_q_vector *q_vector;
8604 	int i;
8605 
8606 	for (i = 0; i < adapter->num_q_vectors; i++) {
8607 		q_vector = adapter->q_vector[i];
8608 		if (adapter->flags & IGB_FLAG_HAS_MSIX)
8609 			wr32(E1000_EIMC, q_vector->eims_value);
8610 		else
8611 			igb_irq_disable(adapter);
8612 		napi_schedule(&q_vector->napi);
8613 	}
8614 }
8615 #endif /* CONFIG_NET_POLL_CONTROLLER */
8616 
8617 /**
8618  *  igb_io_error_detected - called when PCI error is detected
8619  *  @pdev: Pointer to PCI device
8620  *  @state: The current pci connection state
8621  *
8622  *  This function is called after a PCI bus error affecting
8623  *  this device has been detected.
8624  **/
8625 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
8626 					      pci_channel_state_t state)
8627 {
8628 	struct net_device *netdev = pci_get_drvdata(pdev);
8629 	struct igb_adapter *adapter = netdev_priv(netdev);
8630 
8631 	netif_device_detach(netdev);
8632 
8633 	if (state == pci_channel_io_perm_failure)
8634 		return PCI_ERS_RESULT_DISCONNECT;
8635 
8636 	if (netif_running(netdev))
8637 		igb_down(adapter);
8638 	pci_disable_device(pdev);
8639 
8640 	/* Request a slot slot reset. */
8641 	return PCI_ERS_RESULT_NEED_RESET;
8642 }
8643 
8644 /**
8645  *  igb_io_slot_reset - called after the pci bus has been reset.
8646  *  @pdev: Pointer to PCI device
8647  *
8648  *  Restart the card from scratch, as if from a cold-boot. Implementation
8649  *  resembles the first-half of the igb_resume routine.
8650  **/
8651 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
8652 {
8653 	struct net_device *netdev = pci_get_drvdata(pdev);
8654 	struct igb_adapter *adapter = netdev_priv(netdev);
8655 	struct e1000_hw *hw = &adapter->hw;
8656 	pci_ers_result_t result;
8657 	int err;
8658 
8659 	if (pci_enable_device_mem(pdev)) {
8660 		dev_err(&pdev->dev,
8661 			"Cannot re-enable PCI device after reset.\n");
8662 		result = PCI_ERS_RESULT_DISCONNECT;
8663 	} else {
8664 		pci_set_master(pdev);
8665 		pci_restore_state(pdev);
8666 		pci_save_state(pdev);
8667 
8668 		pci_enable_wake(pdev, PCI_D3hot, 0);
8669 		pci_enable_wake(pdev, PCI_D3cold, 0);
8670 
8671 		/* In case of PCI error, adapter lose its HW address
8672 		 * so we should re-assign it here.
8673 		 */
8674 		hw->hw_addr = adapter->io_addr;
8675 
8676 		igb_reset(adapter);
8677 		wr32(E1000_WUS, ~0);
8678 		result = PCI_ERS_RESULT_RECOVERED;
8679 	}
8680 
8681 	err = pci_cleanup_aer_uncorrect_error_status(pdev);
8682 	if (err) {
8683 		dev_err(&pdev->dev,
8684 			"pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
8685 			err);
8686 		/* non-fatal, continue */
8687 	}
8688 
8689 	return result;
8690 }
8691 
8692 /**
8693  *  igb_io_resume - called when traffic can start flowing again.
8694  *  @pdev: Pointer to PCI device
8695  *
8696  *  This callback is called when the error recovery driver tells us that
8697  *  its OK to resume normal operation. Implementation resembles the
8698  *  second-half of the igb_resume routine.
8699  */
8700 static void igb_io_resume(struct pci_dev *pdev)
8701 {
8702 	struct net_device *netdev = pci_get_drvdata(pdev);
8703 	struct igb_adapter *adapter = netdev_priv(netdev);
8704 
8705 	if (netif_running(netdev)) {
8706 		if (igb_up(adapter)) {
8707 			dev_err(&pdev->dev, "igb_up failed after reset\n");
8708 			return;
8709 		}
8710 	}
8711 
8712 	netif_device_attach(netdev);
8713 
8714 	/* let the f/w know that the h/w is now under the control of the
8715 	 * driver.
8716 	 */
8717 	igb_get_hw_control(adapter);
8718 }
8719 
8720 /**
8721  *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
8722  *  @adapter: Pointer to adapter structure
8723  *  @index: Index of the RAR entry which need to be synced with MAC table
8724  **/
8725 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
8726 {
8727 	struct e1000_hw *hw = &adapter->hw;
8728 	u32 rar_low, rar_high;
8729 	u8 *addr = adapter->mac_table[index].addr;
8730 
8731 	/* HW expects these to be in network order when they are plugged
8732 	 * into the registers which are little endian.  In order to guarantee
8733 	 * that ordering we need to do an leXX_to_cpup here in order to be
8734 	 * ready for the byteswap that occurs with writel
8735 	 */
8736 	rar_low = le32_to_cpup((__le32 *)(addr));
8737 	rar_high = le16_to_cpup((__le16 *)(addr + 4));
8738 
8739 	/* Indicate to hardware the Address is Valid. */
8740 	if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
8741 		if (is_valid_ether_addr(addr))
8742 			rar_high |= E1000_RAH_AV;
8743 
8744 		if (hw->mac.type == e1000_82575)
8745 			rar_high |= E1000_RAH_POOL_1 *
8746 				    adapter->mac_table[index].queue;
8747 		else
8748 			rar_high |= E1000_RAH_POOL_1 <<
8749 				    adapter->mac_table[index].queue;
8750 	}
8751 
8752 	wr32(E1000_RAL(index), rar_low);
8753 	wrfl();
8754 	wr32(E1000_RAH(index), rar_high);
8755 	wrfl();
8756 }
8757 
8758 static int igb_set_vf_mac(struct igb_adapter *adapter,
8759 			  int vf, unsigned char *mac_addr)
8760 {
8761 	struct e1000_hw *hw = &adapter->hw;
8762 	/* VF MAC addresses start at end of receive addresses and moves
8763 	 * towards the first, as a result a collision should not be possible
8764 	 */
8765 	int rar_entry = hw->mac.rar_entry_count - (vf + 1);
8766 	unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
8767 
8768 	ether_addr_copy(vf_mac_addr, mac_addr);
8769 	ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
8770 	adapter->mac_table[rar_entry].queue = vf;
8771 	adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
8772 	igb_rar_set_index(adapter, rar_entry);
8773 
8774 	return 0;
8775 }
8776 
8777 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
8778 {
8779 	struct igb_adapter *adapter = netdev_priv(netdev);
8780 
8781 	if (vf >= adapter->vfs_allocated_count)
8782 		return -EINVAL;
8783 
8784 	/* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
8785 	 * flag and allows to overwrite the MAC via VF netdev.  This
8786 	 * is necessary to allow libvirt a way to restore the original
8787 	 * MAC after unbinding vfio-pci and reloading igbvf after shutting
8788 	 * down a VM.
8789 	 */
8790 	if (is_zero_ether_addr(mac)) {
8791 		adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
8792 		dev_info(&adapter->pdev->dev,
8793 			 "remove administratively set MAC on VF %d\n",
8794 			 vf);
8795 	} else if (is_valid_ether_addr(mac)) {
8796 		adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
8797 		dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
8798 			 mac, vf);
8799 		dev_info(&adapter->pdev->dev,
8800 			 "Reload the VF driver to make this change effective.");
8801 		/* Generate additional warning if PF is down */
8802 		if (test_bit(__IGB_DOWN, &adapter->state)) {
8803 			dev_warn(&adapter->pdev->dev,
8804 				 "The VF MAC address has been set, but the PF device is not up.\n");
8805 			dev_warn(&adapter->pdev->dev,
8806 				 "Bring the PF device up before attempting to use the VF device.\n");
8807 		}
8808 	} else {
8809 		return -EINVAL;
8810 	}
8811 	return igb_set_vf_mac(adapter, vf, mac);
8812 }
8813 
8814 static int igb_link_mbps(int internal_link_speed)
8815 {
8816 	switch (internal_link_speed) {
8817 	case SPEED_100:
8818 		return 100;
8819 	case SPEED_1000:
8820 		return 1000;
8821 	default:
8822 		return 0;
8823 	}
8824 }
8825 
8826 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
8827 				  int link_speed)
8828 {
8829 	int rf_dec, rf_int;
8830 	u32 bcnrc_val;
8831 
8832 	if (tx_rate != 0) {
8833 		/* Calculate the rate factor values to set */
8834 		rf_int = link_speed / tx_rate;
8835 		rf_dec = (link_speed - (rf_int * tx_rate));
8836 		rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
8837 			 tx_rate;
8838 
8839 		bcnrc_val = E1000_RTTBCNRC_RS_ENA;
8840 		bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
8841 			      E1000_RTTBCNRC_RF_INT_MASK);
8842 		bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
8843 	} else {
8844 		bcnrc_val = 0;
8845 	}
8846 
8847 	wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
8848 	/* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
8849 	 * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
8850 	 */
8851 	wr32(E1000_RTTBCNRM, 0x14);
8852 	wr32(E1000_RTTBCNRC, bcnrc_val);
8853 }
8854 
8855 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
8856 {
8857 	int actual_link_speed, i;
8858 	bool reset_rate = false;
8859 
8860 	/* VF TX rate limit was not set or not supported */
8861 	if ((adapter->vf_rate_link_speed == 0) ||
8862 	    (adapter->hw.mac.type != e1000_82576))
8863 		return;
8864 
8865 	actual_link_speed = igb_link_mbps(adapter->link_speed);
8866 	if (actual_link_speed != adapter->vf_rate_link_speed) {
8867 		reset_rate = true;
8868 		adapter->vf_rate_link_speed = 0;
8869 		dev_info(&adapter->pdev->dev,
8870 			 "Link speed has been changed. VF Transmit rate is disabled\n");
8871 	}
8872 
8873 	for (i = 0; i < adapter->vfs_allocated_count; i++) {
8874 		if (reset_rate)
8875 			adapter->vf_data[i].tx_rate = 0;
8876 
8877 		igb_set_vf_rate_limit(&adapter->hw, i,
8878 				      adapter->vf_data[i].tx_rate,
8879 				      actual_link_speed);
8880 	}
8881 }
8882 
8883 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
8884 			     int min_tx_rate, int max_tx_rate)
8885 {
8886 	struct igb_adapter *adapter = netdev_priv(netdev);
8887 	struct e1000_hw *hw = &adapter->hw;
8888 	int actual_link_speed;
8889 
8890 	if (hw->mac.type != e1000_82576)
8891 		return -EOPNOTSUPP;
8892 
8893 	if (min_tx_rate)
8894 		return -EINVAL;
8895 
8896 	actual_link_speed = igb_link_mbps(adapter->link_speed);
8897 	if ((vf >= adapter->vfs_allocated_count) ||
8898 	    (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
8899 	    (max_tx_rate < 0) ||
8900 	    (max_tx_rate > actual_link_speed))
8901 		return -EINVAL;
8902 
8903 	adapter->vf_rate_link_speed = actual_link_speed;
8904 	adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
8905 	igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
8906 
8907 	return 0;
8908 }
8909 
8910 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
8911 				   bool setting)
8912 {
8913 	struct igb_adapter *adapter = netdev_priv(netdev);
8914 	struct e1000_hw *hw = &adapter->hw;
8915 	u32 reg_val, reg_offset;
8916 
8917 	if (!adapter->vfs_allocated_count)
8918 		return -EOPNOTSUPP;
8919 
8920 	if (vf >= adapter->vfs_allocated_count)
8921 		return -EINVAL;
8922 
8923 	reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
8924 	reg_val = rd32(reg_offset);
8925 	if (setting)
8926 		reg_val |= (BIT(vf) |
8927 			    BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8928 	else
8929 		reg_val &= ~(BIT(vf) |
8930 			     BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8931 	wr32(reg_offset, reg_val);
8932 
8933 	adapter->vf_data[vf].spoofchk_enabled = setting;
8934 	return 0;
8935 }
8936 
8937 static int igb_ndo_get_vf_config(struct net_device *netdev,
8938 				 int vf, struct ifla_vf_info *ivi)
8939 {
8940 	struct igb_adapter *adapter = netdev_priv(netdev);
8941 	if (vf >= adapter->vfs_allocated_count)
8942 		return -EINVAL;
8943 	ivi->vf = vf;
8944 	memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
8945 	ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
8946 	ivi->min_tx_rate = 0;
8947 	ivi->vlan = adapter->vf_data[vf].pf_vlan;
8948 	ivi->qos = adapter->vf_data[vf].pf_qos;
8949 	ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
8950 	return 0;
8951 }
8952 
8953 static void igb_vmm_control(struct igb_adapter *adapter)
8954 {
8955 	struct e1000_hw *hw = &adapter->hw;
8956 	u32 reg;
8957 
8958 	switch (hw->mac.type) {
8959 	case e1000_82575:
8960 	case e1000_i210:
8961 	case e1000_i211:
8962 	case e1000_i354:
8963 	default:
8964 		/* replication is not supported for 82575 */
8965 		return;
8966 	case e1000_82576:
8967 		/* notify HW that the MAC is adding vlan tags */
8968 		reg = rd32(E1000_DTXCTL);
8969 		reg |= E1000_DTXCTL_VLAN_ADDED;
8970 		wr32(E1000_DTXCTL, reg);
8971 		/* Fall through */
8972 	case e1000_82580:
8973 		/* enable replication vlan tag stripping */
8974 		reg = rd32(E1000_RPLOLR);
8975 		reg |= E1000_RPLOLR_STRVLAN;
8976 		wr32(E1000_RPLOLR, reg);
8977 		/* Fall through */
8978 	case e1000_i350:
8979 		/* none of the above registers are supported by i350 */
8980 		break;
8981 	}
8982 
8983 	if (adapter->vfs_allocated_count) {
8984 		igb_vmdq_set_loopback_pf(hw, true);
8985 		igb_vmdq_set_replication_pf(hw, true);
8986 		igb_vmdq_set_anti_spoofing_pf(hw, true,
8987 					      adapter->vfs_allocated_count);
8988 	} else {
8989 		igb_vmdq_set_loopback_pf(hw, false);
8990 		igb_vmdq_set_replication_pf(hw, false);
8991 	}
8992 }
8993 
8994 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
8995 {
8996 	struct e1000_hw *hw = &adapter->hw;
8997 	u32 dmac_thr;
8998 	u16 hwm;
8999 
9000 	if (hw->mac.type > e1000_82580) {
9001 		if (adapter->flags & IGB_FLAG_DMAC) {
9002 			u32 reg;
9003 
9004 			/* force threshold to 0. */
9005 			wr32(E1000_DMCTXTH, 0);
9006 
9007 			/* DMA Coalescing high water mark needs to be greater
9008 			 * than the Rx threshold. Set hwm to PBA - max frame
9009 			 * size in 16B units, capping it at PBA - 6KB.
9010 			 */
9011 			hwm = 64 * (pba - 6);
9012 			reg = rd32(E1000_FCRTC);
9013 			reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9014 			reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9015 				& E1000_FCRTC_RTH_COAL_MASK);
9016 			wr32(E1000_FCRTC, reg);
9017 
9018 			/* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9019 			 * frame size, capping it at PBA - 10KB.
9020 			 */
9021 			dmac_thr = pba - 10;
9022 			reg = rd32(E1000_DMACR);
9023 			reg &= ~E1000_DMACR_DMACTHR_MASK;
9024 			reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9025 				& E1000_DMACR_DMACTHR_MASK);
9026 
9027 			/* transition to L0x or L1 if available..*/
9028 			reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9029 
9030 			/* watchdog timer= +-1000 usec in 32usec intervals */
9031 			reg |= (1000 >> 5);
9032 
9033 			/* Disable BMC-to-OS Watchdog Enable */
9034 			if (hw->mac.type != e1000_i354)
9035 				reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9036 
9037 			wr32(E1000_DMACR, reg);
9038 
9039 			/* no lower threshold to disable
9040 			 * coalescing(smart fifb)-UTRESH=0
9041 			 */
9042 			wr32(E1000_DMCRTRH, 0);
9043 
9044 			reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9045 
9046 			wr32(E1000_DMCTLX, reg);
9047 
9048 			/* free space in tx packet buffer to wake from
9049 			 * DMA coal
9050 			 */
9051 			wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
9052 			     (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
9053 
9054 			/* make low power state decision controlled
9055 			 * by DMA coal
9056 			 */
9057 			reg = rd32(E1000_PCIEMISC);
9058 			reg &= ~E1000_PCIEMISC_LX_DECISION;
9059 			wr32(E1000_PCIEMISC, reg);
9060 		} /* endif adapter->dmac is not disabled */
9061 	} else if (hw->mac.type == e1000_82580) {
9062 		u32 reg = rd32(E1000_PCIEMISC);
9063 
9064 		wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
9065 		wr32(E1000_DMACR, 0);
9066 	}
9067 }
9068 
9069 /**
9070  *  igb_read_i2c_byte - Reads 8 bit word over I2C
9071  *  @hw: pointer to hardware structure
9072  *  @byte_offset: byte offset to read
9073  *  @dev_addr: device address
9074  *  @data: value read
9075  *
9076  *  Performs byte read operation over I2C interface at
9077  *  a specified device address.
9078  **/
9079 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9080 		      u8 dev_addr, u8 *data)
9081 {
9082 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9083 	struct i2c_client *this_client = adapter->i2c_client;
9084 	s32 status;
9085 	u16 swfw_mask = 0;
9086 
9087 	if (!this_client)
9088 		return E1000_ERR_I2C;
9089 
9090 	swfw_mask = E1000_SWFW_PHY0_SM;
9091 
9092 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9093 		return E1000_ERR_SWFW_SYNC;
9094 
9095 	status = i2c_smbus_read_byte_data(this_client, byte_offset);
9096 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9097 
9098 	if (status < 0)
9099 		return E1000_ERR_I2C;
9100 	else {
9101 		*data = status;
9102 		return 0;
9103 	}
9104 }
9105 
9106 /**
9107  *  igb_write_i2c_byte - Writes 8 bit word over I2C
9108  *  @hw: pointer to hardware structure
9109  *  @byte_offset: byte offset to write
9110  *  @dev_addr: device address
9111  *  @data: value to write
9112  *
9113  *  Performs byte write operation over I2C interface at
9114  *  a specified device address.
9115  **/
9116 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9117 		       u8 dev_addr, u8 data)
9118 {
9119 	struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9120 	struct i2c_client *this_client = adapter->i2c_client;
9121 	s32 status;
9122 	u16 swfw_mask = E1000_SWFW_PHY0_SM;
9123 
9124 	if (!this_client)
9125 		return E1000_ERR_I2C;
9126 
9127 	if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9128 		return E1000_ERR_SWFW_SYNC;
9129 	status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
9130 	hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9131 
9132 	if (status)
9133 		return E1000_ERR_I2C;
9134 	else
9135 		return 0;
9136 
9137 }
9138 
9139 int igb_reinit_queues(struct igb_adapter *adapter)
9140 {
9141 	struct net_device *netdev = adapter->netdev;
9142 	struct pci_dev *pdev = adapter->pdev;
9143 	int err = 0;
9144 
9145 	if (netif_running(netdev))
9146 		igb_close(netdev);
9147 
9148 	igb_reset_interrupt_capability(adapter);
9149 
9150 	if (igb_init_interrupt_scheme(adapter, true)) {
9151 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9152 		return -ENOMEM;
9153 	}
9154 
9155 	if (netif_running(netdev))
9156 		err = igb_open(netdev);
9157 
9158 	return err;
9159 }
9160 
9161 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
9162 {
9163 	struct igb_nfc_filter *rule;
9164 
9165 	spin_lock(&adapter->nfc_lock);
9166 
9167 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9168 		igb_erase_filter(adapter, rule);
9169 
9170 	spin_unlock(&adapter->nfc_lock);
9171 }
9172 
9173 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
9174 {
9175 	struct igb_nfc_filter *rule;
9176 
9177 	spin_lock(&adapter->nfc_lock);
9178 
9179 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9180 		igb_add_filter(adapter, rule);
9181 
9182 	spin_unlock(&adapter->nfc_lock);
9183 }
9184 /* igb_main.c */
9185