1 /*******************************************************************************
2 
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2015 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, see <http://www.gnu.org/licenses/>.
17 
18   The full GNU General Public License is included in this distribution in
19   the file called "COPYING".
20 
21   Contact Information:
22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 
25 *******************************************************************************/
26 
27 /******************************************************************************
28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
29 ******************************************************************************/
30 
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32 
33 #include <linux/types.h>
34 #include <linux/bitops.h>
35 #include <linux/module.h>
36 #include <linux/pci.h>
37 #include <linux/netdevice.h>
38 #include <linux/vmalloc.h>
39 #include <linux/string.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/tcp.h>
43 #include <linux/sctp.h>
44 #include <linux/ipv6.h>
45 #include <linux/slab.h>
46 #include <net/checksum.h>
47 #include <net/ip6_checksum.h>
48 #include <linux/ethtool.h>
49 #include <linux/if.h>
50 #include <linux/if_vlan.h>
51 #include <linux/prefetch.h>
52 
53 #include "ixgbevf.h"
54 
55 const char ixgbevf_driver_name[] = "ixgbevf";
56 static const char ixgbevf_driver_string[] =
57 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
58 
59 #define DRV_VERSION "3.2.2-k"
60 const char ixgbevf_driver_version[] = DRV_VERSION;
61 static char ixgbevf_copyright[] =
62 	"Copyright (c) 2009 - 2015 Intel Corporation.";
63 
64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
65 	[board_82599_vf]	= &ixgbevf_82599_vf_info,
66 	[board_82599_vf_hv]	= &ixgbevf_82599_vf_hv_info,
67 	[board_X540_vf]		= &ixgbevf_X540_vf_info,
68 	[board_X540_vf_hv]	= &ixgbevf_X540_vf_hv_info,
69 	[board_X550_vf]		= &ixgbevf_X550_vf_info,
70 	[board_X550_vf_hv]	= &ixgbevf_X550_vf_hv_info,
71 	[board_X550EM_x_vf]	= &ixgbevf_X550EM_x_vf_info,
72 	[board_X550EM_x_vf_hv]	= &ixgbevf_X550EM_x_vf_hv_info,
73 	[board_x550em_a_vf]	= &ixgbevf_x550em_a_vf_info,
74 };
75 
76 /* ixgbevf_pci_tbl - PCI Device ID Table
77  *
78  * Wildcard entries (PCI_ANY_ID) should come last
79  * Last entry must be all 0s
80  *
81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82  *   Class, Class Mask, private data (not used) }
83  */
84 static const struct pci_device_id ixgbevf_pci_tbl[] = {
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
94 	/* required last entry */
95 	{0, }
96 };
97 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
98 
99 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
100 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
101 MODULE_LICENSE("GPL");
102 MODULE_VERSION(DRV_VERSION);
103 
104 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
105 static int debug = -1;
106 module_param(debug, int, 0);
107 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
108 
109 static struct workqueue_struct *ixgbevf_wq;
110 
111 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
112 {
113 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
114 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
115 	    !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
116 		queue_work(ixgbevf_wq, &adapter->service_task);
117 }
118 
119 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
120 {
121 	BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
122 
123 	/* flush memory to make sure state is correct before next watchdog */
124 	smp_mb__before_atomic();
125 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
126 }
127 
128 /* forward decls */
129 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
130 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
131 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
132 
133 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
134 {
135 	struct ixgbevf_adapter *adapter = hw->back;
136 
137 	if (!hw->hw_addr)
138 		return;
139 	hw->hw_addr = NULL;
140 	dev_err(&adapter->pdev->dev, "Adapter removed\n");
141 	if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
142 		ixgbevf_service_event_schedule(adapter);
143 }
144 
145 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
146 {
147 	u32 value;
148 
149 	/* The following check not only optimizes a bit by not
150 	 * performing a read on the status register when the
151 	 * register just read was a status register read that
152 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
153 	 * potential recursion.
154 	 */
155 	if (reg == IXGBE_VFSTATUS) {
156 		ixgbevf_remove_adapter(hw);
157 		return;
158 	}
159 	value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
160 	if (value == IXGBE_FAILED_READ_REG)
161 		ixgbevf_remove_adapter(hw);
162 }
163 
164 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
165 {
166 	u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
167 	u32 value;
168 
169 	if (IXGBE_REMOVED(reg_addr))
170 		return IXGBE_FAILED_READ_REG;
171 	value = readl(reg_addr + reg);
172 	if (unlikely(value == IXGBE_FAILED_READ_REG))
173 		ixgbevf_check_remove(hw, reg);
174 	return value;
175 }
176 
177 /**
178  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
179  * @adapter: pointer to adapter struct
180  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
181  * @queue: queue to map the corresponding interrupt to
182  * @msix_vector: the vector to map to the corresponding queue
183  **/
184 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
185 			     u8 queue, u8 msix_vector)
186 {
187 	u32 ivar, index;
188 	struct ixgbe_hw *hw = &adapter->hw;
189 
190 	if (direction == -1) {
191 		/* other causes */
192 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
193 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
194 		ivar &= ~0xFF;
195 		ivar |= msix_vector;
196 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
197 	} else {
198 		/* Tx or Rx causes */
199 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
200 		index = ((16 * (queue & 1)) + (8 * direction));
201 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
202 		ivar &= ~(0xFF << index);
203 		ivar |= (msix_vector << index);
204 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
205 	}
206 }
207 
208 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
209 					struct ixgbevf_tx_buffer *tx_buffer)
210 {
211 	if (tx_buffer->skb) {
212 		dev_kfree_skb_any(tx_buffer->skb);
213 		if (dma_unmap_len(tx_buffer, len))
214 			dma_unmap_single(tx_ring->dev,
215 					 dma_unmap_addr(tx_buffer, dma),
216 					 dma_unmap_len(tx_buffer, len),
217 					 DMA_TO_DEVICE);
218 	} else if (dma_unmap_len(tx_buffer, len)) {
219 		dma_unmap_page(tx_ring->dev,
220 			       dma_unmap_addr(tx_buffer, dma),
221 			       dma_unmap_len(tx_buffer, len),
222 			       DMA_TO_DEVICE);
223 	}
224 	tx_buffer->next_to_watch = NULL;
225 	tx_buffer->skb = NULL;
226 	dma_unmap_len_set(tx_buffer, len, 0);
227 	/* tx_buffer must be completely set up in the transmit path */
228 }
229 
230 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
231 {
232 	return ring->stats.packets;
233 }
234 
235 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
236 {
237 	struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
238 	struct ixgbe_hw *hw = &adapter->hw;
239 
240 	u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
241 	u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
242 
243 	if (head != tail)
244 		return (head < tail) ?
245 			tail - head : (tail + ring->count - head);
246 
247 	return 0;
248 }
249 
250 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
251 {
252 	u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
253 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
254 	u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
255 
256 	clear_check_for_tx_hang(tx_ring);
257 
258 	/* Check for a hung queue, but be thorough. This verifies
259 	 * that a transmit has been completed since the previous
260 	 * check AND there is at least one packet pending. The
261 	 * ARMED bit is set to indicate a potential hang.
262 	 */
263 	if ((tx_done_old == tx_done) && tx_pending) {
264 		/* make sure it is true for two checks in a row */
265 		return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
266 					&tx_ring->state);
267 	}
268 	/* reset the countdown */
269 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
270 
271 	/* update completed stats and continue */
272 	tx_ring->tx_stats.tx_done_old = tx_done;
273 
274 	return false;
275 }
276 
277 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
278 {
279 	/* Do the reset outside of interrupt context */
280 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
281 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
282 		ixgbevf_service_event_schedule(adapter);
283 	}
284 }
285 
286 /**
287  * ixgbevf_tx_timeout - Respond to a Tx Hang
288  * @netdev: network interface device structure
289  **/
290 static void ixgbevf_tx_timeout(struct net_device *netdev)
291 {
292 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
293 
294 	ixgbevf_tx_timeout_reset(adapter);
295 }
296 
297 /**
298  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
299  * @q_vector: board private structure
300  * @tx_ring: tx ring to clean
301  * @napi_budget: Used to determine if we are in netpoll
302  **/
303 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
304 				 struct ixgbevf_ring *tx_ring, int napi_budget)
305 {
306 	struct ixgbevf_adapter *adapter = q_vector->adapter;
307 	struct ixgbevf_tx_buffer *tx_buffer;
308 	union ixgbe_adv_tx_desc *tx_desc;
309 	unsigned int total_bytes = 0, total_packets = 0;
310 	unsigned int budget = tx_ring->count / 2;
311 	unsigned int i = tx_ring->next_to_clean;
312 
313 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
314 		return true;
315 
316 	tx_buffer = &tx_ring->tx_buffer_info[i];
317 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
318 	i -= tx_ring->count;
319 
320 	do {
321 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
322 
323 		/* if next_to_watch is not set then there is no work pending */
324 		if (!eop_desc)
325 			break;
326 
327 		/* prevent any other reads prior to eop_desc */
328 		read_barrier_depends();
329 
330 		/* if DD is not set pending work has not been completed */
331 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
332 			break;
333 
334 		/* clear next_to_watch to prevent false hangs */
335 		tx_buffer->next_to_watch = NULL;
336 
337 		/* update the statistics for this packet */
338 		total_bytes += tx_buffer->bytecount;
339 		total_packets += tx_buffer->gso_segs;
340 
341 		/* free the skb */
342 		napi_consume_skb(tx_buffer->skb, napi_budget);
343 
344 		/* unmap skb header data */
345 		dma_unmap_single(tx_ring->dev,
346 				 dma_unmap_addr(tx_buffer, dma),
347 				 dma_unmap_len(tx_buffer, len),
348 				 DMA_TO_DEVICE);
349 
350 		/* clear tx_buffer data */
351 		tx_buffer->skb = NULL;
352 		dma_unmap_len_set(tx_buffer, len, 0);
353 
354 		/* unmap remaining buffers */
355 		while (tx_desc != eop_desc) {
356 			tx_buffer++;
357 			tx_desc++;
358 			i++;
359 			if (unlikely(!i)) {
360 				i -= tx_ring->count;
361 				tx_buffer = tx_ring->tx_buffer_info;
362 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
363 			}
364 
365 			/* unmap any remaining paged data */
366 			if (dma_unmap_len(tx_buffer, len)) {
367 				dma_unmap_page(tx_ring->dev,
368 					       dma_unmap_addr(tx_buffer, dma),
369 					       dma_unmap_len(tx_buffer, len),
370 					       DMA_TO_DEVICE);
371 				dma_unmap_len_set(tx_buffer, len, 0);
372 			}
373 		}
374 
375 		/* move us one more past the eop_desc for start of next pkt */
376 		tx_buffer++;
377 		tx_desc++;
378 		i++;
379 		if (unlikely(!i)) {
380 			i -= tx_ring->count;
381 			tx_buffer = tx_ring->tx_buffer_info;
382 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
383 		}
384 
385 		/* issue prefetch for next Tx descriptor */
386 		prefetch(tx_desc);
387 
388 		/* update budget accounting */
389 		budget--;
390 	} while (likely(budget));
391 
392 	i += tx_ring->count;
393 	tx_ring->next_to_clean = i;
394 	u64_stats_update_begin(&tx_ring->syncp);
395 	tx_ring->stats.bytes += total_bytes;
396 	tx_ring->stats.packets += total_packets;
397 	u64_stats_update_end(&tx_ring->syncp);
398 	q_vector->tx.total_bytes += total_bytes;
399 	q_vector->tx.total_packets += total_packets;
400 
401 	if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
402 		struct ixgbe_hw *hw = &adapter->hw;
403 		union ixgbe_adv_tx_desc *eop_desc;
404 
405 		eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
406 
407 		pr_err("Detected Tx Unit Hang\n"
408 		       "  Tx Queue             <%d>\n"
409 		       "  TDH, TDT             <%x>, <%x>\n"
410 		       "  next_to_use          <%x>\n"
411 		       "  next_to_clean        <%x>\n"
412 		       "tx_buffer_info[next_to_clean]\n"
413 		       "  next_to_watch        <%p>\n"
414 		       "  eop_desc->wb.status  <%x>\n"
415 		       "  time_stamp           <%lx>\n"
416 		       "  jiffies              <%lx>\n",
417 		       tx_ring->queue_index,
418 		       IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
419 		       IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
420 		       tx_ring->next_to_use, i,
421 		       eop_desc, (eop_desc ? eop_desc->wb.status : 0),
422 		       tx_ring->tx_buffer_info[i].time_stamp, jiffies);
423 
424 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
425 
426 		/* schedule immediate reset if we believe we hung */
427 		ixgbevf_tx_timeout_reset(adapter);
428 
429 		return true;
430 	}
431 
432 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
433 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
434 		     (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
435 		/* Make sure that anybody stopping the queue after this
436 		 * sees the new next_to_clean.
437 		 */
438 		smp_mb();
439 
440 		if (__netif_subqueue_stopped(tx_ring->netdev,
441 					     tx_ring->queue_index) &&
442 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
443 			netif_wake_subqueue(tx_ring->netdev,
444 					    tx_ring->queue_index);
445 			++tx_ring->tx_stats.restart_queue;
446 		}
447 	}
448 
449 	return !!budget;
450 }
451 
452 /**
453  * ixgbevf_rx_skb - Helper function to determine proper Rx method
454  * @q_vector: structure containing interrupt and ring information
455  * @skb: packet to send up
456  **/
457 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
458 			   struct sk_buff *skb)
459 {
460 #ifdef CONFIG_NET_RX_BUSY_POLL
461 	skb_mark_napi_id(skb, &q_vector->napi);
462 
463 	if (ixgbevf_qv_busy_polling(q_vector)) {
464 		netif_receive_skb(skb);
465 		/* exit early if we busy polled */
466 		return;
467 	}
468 #endif /* CONFIG_NET_RX_BUSY_POLL */
469 
470 	napi_gro_receive(&q_vector->napi, skb);
471 }
472 
473 #define IXGBE_RSS_L4_TYPES_MASK \
474 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
475 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
476 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
477 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
478 
479 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
480 				   union ixgbe_adv_rx_desc *rx_desc,
481 				   struct sk_buff *skb)
482 {
483 	u16 rss_type;
484 
485 	if (!(ring->netdev->features & NETIF_F_RXHASH))
486 		return;
487 
488 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
489 		   IXGBE_RXDADV_RSSTYPE_MASK;
490 
491 	if (!rss_type)
492 		return;
493 
494 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
495 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
496 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
497 }
498 
499 /**
500  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
501  * @ring: structure containig ring specific data
502  * @rx_desc: current Rx descriptor being processed
503  * @skb: skb currently being received and modified
504  **/
505 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
506 				       union ixgbe_adv_rx_desc *rx_desc,
507 				       struct sk_buff *skb)
508 {
509 	skb_checksum_none_assert(skb);
510 
511 	/* Rx csum disabled */
512 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
513 		return;
514 
515 	/* if IP and error */
516 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
517 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
518 		ring->rx_stats.csum_err++;
519 		return;
520 	}
521 
522 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
523 		return;
524 
525 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
526 		ring->rx_stats.csum_err++;
527 		return;
528 	}
529 
530 	/* It must be a TCP or UDP packet with a valid checksum */
531 	skb->ip_summed = CHECKSUM_UNNECESSARY;
532 }
533 
534 /**
535  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
536  * @rx_ring: rx descriptor ring packet is being transacted on
537  * @rx_desc: pointer to the EOP Rx descriptor
538  * @skb: pointer to current skb being populated
539  *
540  * This function checks the ring, descriptor, and packet information in
541  * order to populate the checksum, VLAN, protocol, and other fields within
542  * the skb.
543  **/
544 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
545 				       union ixgbe_adv_rx_desc *rx_desc,
546 				       struct sk_buff *skb)
547 {
548 	ixgbevf_rx_hash(rx_ring, rx_desc, skb);
549 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
550 
551 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
552 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
553 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
554 
555 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
556 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
557 	}
558 
559 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
560 }
561 
562 /**
563  * ixgbevf_is_non_eop - process handling of non-EOP buffers
564  * @rx_ring: Rx ring being processed
565  * @rx_desc: Rx descriptor for current buffer
566  * @skb: current socket buffer containing buffer in progress
567  *
568  * This function updates next to clean.  If the buffer is an EOP buffer
569  * this function exits returning false, otherwise it will place the
570  * sk_buff in the next buffer to be chained and return true indicating
571  * that this is in fact a non-EOP buffer.
572  **/
573 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
574 			       union ixgbe_adv_rx_desc *rx_desc)
575 {
576 	u32 ntc = rx_ring->next_to_clean + 1;
577 
578 	/* fetch, update, and store next to clean */
579 	ntc = (ntc < rx_ring->count) ? ntc : 0;
580 	rx_ring->next_to_clean = ntc;
581 
582 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
583 
584 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
585 		return false;
586 
587 	return true;
588 }
589 
590 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
591 				      struct ixgbevf_rx_buffer *bi)
592 {
593 	struct page *page = bi->page;
594 	dma_addr_t dma = bi->dma;
595 
596 	/* since we are recycling buffers we should seldom need to alloc */
597 	if (likely(page))
598 		return true;
599 
600 	/* alloc new page for storage */
601 	page = dev_alloc_page();
602 	if (unlikely(!page)) {
603 		rx_ring->rx_stats.alloc_rx_page_failed++;
604 		return false;
605 	}
606 
607 	/* map page for use */
608 	dma = dma_map_page(rx_ring->dev, page, 0,
609 			   PAGE_SIZE, DMA_FROM_DEVICE);
610 
611 	/* if mapping failed free memory back to system since
612 	 * there isn't much point in holding memory we can't use
613 	 */
614 	if (dma_mapping_error(rx_ring->dev, dma)) {
615 		__free_page(page);
616 
617 		rx_ring->rx_stats.alloc_rx_buff_failed++;
618 		return false;
619 	}
620 
621 	bi->dma = dma;
622 	bi->page = page;
623 	bi->page_offset = 0;
624 
625 	return true;
626 }
627 
628 /**
629  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
630  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
631  * @cleaned_count: number of buffers to replace
632  **/
633 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
634 				     u16 cleaned_count)
635 {
636 	union ixgbe_adv_rx_desc *rx_desc;
637 	struct ixgbevf_rx_buffer *bi;
638 	unsigned int i = rx_ring->next_to_use;
639 
640 	/* nothing to do or no valid netdev defined */
641 	if (!cleaned_count || !rx_ring->netdev)
642 		return;
643 
644 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
645 	bi = &rx_ring->rx_buffer_info[i];
646 	i -= rx_ring->count;
647 
648 	do {
649 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
650 			break;
651 
652 		/* Refresh the desc even if pkt_addr didn't change
653 		 * because each write-back erases this info.
654 		 */
655 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
656 
657 		rx_desc++;
658 		bi++;
659 		i++;
660 		if (unlikely(!i)) {
661 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
662 			bi = rx_ring->rx_buffer_info;
663 			i -= rx_ring->count;
664 		}
665 
666 		/* clear the hdr_addr for the next_to_use descriptor */
667 		rx_desc->read.hdr_addr = 0;
668 
669 		cleaned_count--;
670 	} while (cleaned_count);
671 
672 	i += rx_ring->count;
673 
674 	if (rx_ring->next_to_use != i) {
675 		/* record the next descriptor to use */
676 		rx_ring->next_to_use = i;
677 
678 		/* update next to alloc since we have filled the ring */
679 		rx_ring->next_to_alloc = i;
680 
681 		/* Force memory writes to complete before letting h/w
682 		 * know there are new descriptors to fetch.  (Only
683 		 * applicable for weak-ordered memory model archs,
684 		 * such as IA-64).
685 		 */
686 		wmb();
687 		ixgbevf_write_tail(rx_ring, i);
688 	}
689 }
690 
691 /**
692  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
693  * @rx_ring: rx descriptor ring packet is being transacted on
694  * @rx_desc: pointer to the EOP Rx descriptor
695  * @skb: pointer to current skb being fixed
696  *
697  * Check for corrupted packet headers caused by senders on the local L2
698  * embedded NIC switch not setting up their Tx Descriptors right.  These
699  * should be very rare.
700  *
701  * Also address the case where we are pulling data in on pages only
702  * and as such no data is present in the skb header.
703  *
704  * In addition if skb is not at least 60 bytes we need to pad it so that
705  * it is large enough to qualify as a valid Ethernet frame.
706  *
707  * Returns true if an error was encountered and skb was freed.
708  **/
709 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
710 				    union ixgbe_adv_rx_desc *rx_desc,
711 				    struct sk_buff *skb)
712 {
713 	/* verify that the packet does not have any known errors */
714 	if (unlikely(ixgbevf_test_staterr(rx_desc,
715 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
716 		struct net_device *netdev = rx_ring->netdev;
717 
718 		if (!(netdev->features & NETIF_F_RXALL)) {
719 			dev_kfree_skb_any(skb);
720 			return true;
721 		}
722 	}
723 
724 	/* if eth_skb_pad returns an error the skb was freed */
725 	if (eth_skb_pad(skb))
726 		return true;
727 
728 	return false;
729 }
730 
731 /**
732  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
733  * @rx_ring: rx descriptor ring to store buffers on
734  * @old_buff: donor buffer to have page reused
735  *
736  * Synchronizes page for reuse by the adapter
737  **/
738 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
739 				  struct ixgbevf_rx_buffer *old_buff)
740 {
741 	struct ixgbevf_rx_buffer *new_buff;
742 	u16 nta = rx_ring->next_to_alloc;
743 
744 	new_buff = &rx_ring->rx_buffer_info[nta];
745 
746 	/* update, and store next to alloc */
747 	nta++;
748 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
749 
750 	/* transfer page from old buffer to new buffer */
751 	new_buff->page = old_buff->page;
752 	new_buff->dma = old_buff->dma;
753 	new_buff->page_offset = old_buff->page_offset;
754 
755 	/* sync the buffer for use by the device */
756 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
757 					 new_buff->page_offset,
758 					 IXGBEVF_RX_BUFSZ,
759 					 DMA_FROM_DEVICE);
760 }
761 
762 static inline bool ixgbevf_page_is_reserved(struct page *page)
763 {
764 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
765 }
766 
767 /**
768  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
769  * @rx_ring: rx descriptor ring to transact packets on
770  * @rx_buffer: buffer containing page to add
771  * @rx_desc: descriptor containing length of buffer written by hardware
772  * @skb: sk_buff to place the data into
773  *
774  * This function will add the data contained in rx_buffer->page to the skb.
775  * This is done either through a direct copy if the data in the buffer is
776  * less than the skb header size, otherwise it will just attach the page as
777  * a frag to the skb.
778  *
779  * The function will then update the page offset if necessary and return
780  * true if the buffer can be reused by the adapter.
781  **/
782 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
783 				struct ixgbevf_rx_buffer *rx_buffer,
784 				union ixgbe_adv_rx_desc *rx_desc,
785 				struct sk_buff *skb)
786 {
787 	struct page *page = rx_buffer->page;
788 	unsigned char *va = page_address(page) + rx_buffer->page_offset;
789 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
790 #if (PAGE_SIZE < 8192)
791 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
792 #else
793 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
794 #endif
795 	unsigned int pull_len;
796 
797 	if (unlikely(skb_is_nonlinear(skb)))
798 		goto add_tail_frag;
799 
800 	if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
801 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
802 
803 		/* page is not reserved, we can reuse buffer as is */
804 		if (likely(!ixgbevf_page_is_reserved(page)))
805 			return true;
806 
807 		/* this page cannot be reused so discard it */
808 		put_page(page);
809 		return false;
810 	}
811 
812 	/* we need the header to contain the greater of either ETH_HLEN or
813 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
814 	 */
815 	pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
816 
817 	/* align pull length to size of long to optimize memcpy performance */
818 	memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
819 
820 	/* update all of the pointers */
821 	va += pull_len;
822 	size -= pull_len;
823 
824 add_tail_frag:
825 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
826 			(unsigned long)va & ~PAGE_MASK, size, truesize);
827 
828 	/* avoid re-using remote pages */
829 	if (unlikely(ixgbevf_page_is_reserved(page)))
830 		return false;
831 
832 #if (PAGE_SIZE < 8192)
833 	/* if we are only owner of page we can reuse it */
834 	if (unlikely(page_count(page) != 1))
835 		return false;
836 
837 	/* flip page offset to other buffer */
838 	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
839 
840 #else
841 	/* move offset up to the next cache line */
842 	rx_buffer->page_offset += truesize;
843 
844 	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
845 		return false;
846 
847 #endif
848 	/* Even if we own the page, we are not allowed to use atomic_set()
849 	 * This would break get_page_unless_zero() users.
850 	 */
851 	page_ref_inc(page);
852 
853 	return true;
854 }
855 
856 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
857 					       union ixgbe_adv_rx_desc *rx_desc,
858 					       struct sk_buff *skb)
859 {
860 	struct ixgbevf_rx_buffer *rx_buffer;
861 	struct page *page;
862 
863 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
864 	page = rx_buffer->page;
865 	prefetchw(page);
866 
867 	if (likely(!skb)) {
868 		void *page_addr = page_address(page) +
869 				  rx_buffer->page_offset;
870 
871 		/* prefetch first cache line of first page */
872 		prefetch(page_addr);
873 #if L1_CACHE_BYTES < 128
874 		prefetch(page_addr + L1_CACHE_BYTES);
875 #endif
876 
877 		/* allocate a skb to store the frags */
878 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
879 						IXGBEVF_RX_HDR_SIZE);
880 		if (unlikely(!skb)) {
881 			rx_ring->rx_stats.alloc_rx_buff_failed++;
882 			return NULL;
883 		}
884 
885 		/* we will be copying header into skb->data in
886 		 * pskb_may_pull so it is in our interest to prefetch
887 		 * it now to avoid a possible cache miss
888 		 */
889 		prefetchw(skb->data);
890 	}
891 
892 	/* we are reusing so sync this buffer for CPU use */
893 	dma_sync_single_range_for_cpu(rx_ring->dev,
894 				      rx_buffer->dma,
895 				      rx_buffer->page_offset,
896 				      IXGBEVF_RX_BUFSZ,
897 				      DMA_FROM_DEVICE);
898 
899 	/* pull page into skb */
900 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
901 		/* hand second half of page back to the ring */
902 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
903 	} else {
904 		/* we are not reusing the buffer so unmap it */
905 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
906 			       PAGE_SIZE, DMA_FROM_DEVICE);
907 	}
908 
909 	/* clear contents of buffer_info */
910 	rx_buffer->dma = 0;
911 	rx_buffer->page = NULL;
912 
913 	return skb;
914 }
915 
916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
917 					     u32 qmask)
918 {
919 	struct ixgbe_hw *hw = &adapter->hw;
920 
921 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
922 }
923 
924 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
925 				struct ixgbevf_ring *rx_ring,
926 				int budget)
927 {
928 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
929 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
930 	struct sk_buff *skb = rx_ring->skb;
931 
932 	while (likely(total_rx_packets < budget)) {
933 		union ixgbe_adv_rx_desc *rx_desc;
934 
935 		/* return some buffers to hardware, one at a time is too slow */
936 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
937 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
938 			cleaned_count = 0;
939 		}
940 
941 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
942 
943 		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
944 			break;
945 
946 		/* This memory barrier is needed to keep us from reading
947 		 * any other fields out of the rx_desc until we know the
948 		 * RXD_STAT_DD bit is set
949 		 */
950 		rmb();
951 
952 		/* retrieve a buffer from the ring */
953 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
954 
955 		/* exit if we failed to retrieve a buffer */
956 		if (!skb)
957 			break;
958 
959 		cleaned_count++;
960 
961 		/* fetch next buffer in frame if non-eop */
962 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
963 			continue;
964 
965 		/* verify the packet layout is correct */
966 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
967 			skb = NULL;
968 			continue;
969 		}
970 
971 		/* probably a little skewed due to removing CRC */
972 		total_rx_bytes += skb->len;
973 
974 		/* Workaround hardware that can't do proper VEPA multicast
975 		 * source pruning.
976 		 */
977 		if ((skb->pkt_type == PACKET_BROADCAST ||
978 		     skb->pkt_type == PACKET_MULTICAST) &&
979 		    ether_addr_equal(rx_ring->netdev->dev_addr,
980 				     eth_hdr(skb)->h_source)) {
981 			dev_kfree_skb_irq(skb);
982 			continue;
983 		}
984 
985 		/* populate checksum, VLAN, and protocol */
986 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
987 
988 		ixgbevf_rx_skb(q_vector, skb);
989 
990 		/* reset skb pointer */
991 		skb = NULL;
992 
993 		/* update budget accounting */
994 		total_rx_packets++;
995 	}
996 
997 	/* place incomplete frames back on ring for completion */
998 	rx_ring->skb = skb;
999 
1000 	u64_stats_update_begin(&rx_ring->syncp);
1001 	rx_ring->stats.packets += total_rx_packets;
1002 	rx_ring->stats.bytes += total_rx_bytes;
1003 	u64_stats_update_end(&rx_ring->syncp);
1004 	q_vector->rx.total_packets += total_rx_packets;
1005 	q_vector->rx.total_bytes += total_rx_bytes;
1006 
1007 	return total_rx_packets;
1008 }
1009 
1010 /**
1011  * ixgbevf_poll - NAPI polling calback
1012  * @napi: napi struct with our devices info in it
1013  * @budget: amount of work driver is allowed to do this pass, in packets
1014  *
1015  * This function will clean more than one or more rings associated with a
1016  * q_vector.
1017  **/
1018 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1019 {
1020 	struct ixgbevf_q_vector *q_vector =
1021 		container_of(napi, struct ixgbevf_q_vector, napi);
1022 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1023 	struct ixgbevf_ring *ring;
1024 	int per_ring_budget, work_done = 0;
1025 	bool clean_complete = true;
1026 
1027 	ixgbevf_for_each_ring(ring, q_vector->tx) {
1028 		if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1029 			clean_complete = false;
1030 	}
1031 
1032 	if (budget <= 0)
1033 		return budget;
1034 #ifdef CONFIG_NET_RX_BUSY_POLL
1035 	if (!ixgbevf_qv_lock_napi(q_vector))
1036 		return budget;
1037 #endif
1038 
1039 	/* attempt to distribute budget to each queue fairly, but don't allow
1040 	 * the budget to go below 1 because we'll exit polling
1041 	 */
1042 	if (q_vector->rx.count > 1)
1043 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1044 	else
1045 		per_ring_budget = budget;
1046 
1047 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1048 		int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1049 						   per_ring_budget);
1050 		work_done += cleaned;
1051 		if (cleaned >= per_ring_budget)
1052 			clean_complete = false;
1053 	}
1054 
1055 #ifdef CONFIG_NET_RX_BUSY_POLL
1056 	ixgbevf_qv_unlock_napi(q_vector);
1057 #endif
1058 
1059 	/* If all work not completed, return budget and keep polling */
1060 	if (!clean_complete)
1061 		return budget;
1062 	/* all work done, exit the polling mode */
1063 	napi_complete_done(napi, work_done);
1064 	if (adapter->rx_itr_setting == 1)
1065 		ixgbevf_set_itr(q_vector);
1066 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1067 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1068 		ixgbevf_irq_enable_queues(adapter,
1069 					  BIT(q_vector->v_idx));
1070 
1071 	return 0;
1072 }
1073 
1074 /**
1075  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1076  * @q_vector: structure containing interrupt and ring information
1077  **/
1078 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1079 {
1080 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1081 	struct ixgbe_hw *hw = &adapter->hw;
1082 	int v_idx = q_vector->v_idx;
1083 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1084 
1085 	/* set the WDIS bit to not clear the timer bits and cause an
1086 	 * immediate assertion of the interrupt
1087 	 */
1088 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1089 
1090 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1091 }
1092 
1093 #ifdef CONFIG_NET_RX_BUSY_POLL
1094 /* must be called with local_bh_disable()d */
1095 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1096 {
1097 	struct ixgbevf_q_vector *q_vector =
1098 			container_of(napi, struct ixgbevf_q_vector, napi);
1099 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1100 	struct ixgbevf_ring  *ring;
1101 	int found = 0;
1102 
1103 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1104 		return LL_FLUSH_FAILED;
1105 
1106 	if (!ixgbevf_qv_lock_poll(q_vector))
1107 		return LL_FLUSH_BUSY;
1108 
1109 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1110 		found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1111 #ifdef BP_EXTENDED_STATS
1112 		if (found)
1113 			ring->stats.cleaned += found;
1114 		else
1115 			ring->stats.misses++;
1116 #endif
1117 		if (found)
1118 			break;
1119 	}
1120 
1121 	ixgbevf_qv_unlock_poll(q_vector);
1122 
1123 	return found;
1124 }
1125 #endif /* CONFIG_NET_RX_BUSY_POLL */
1126 
1127 /**
1128  * ixgbevf_configure_msix - Configure MSI-X hardware
1129  * @adapter: board private structure
1130  *
1131  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1132  * interrupts.
1133  **/
1134 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1135 {
1136 	struct ixgbevf_q_vector *q_vector;
1137 	int q_vectors, v_idx;
1138 
1139 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1140 	adapter->eims_enable_mask = 0;
1141 
1142 	/* Populate the IVAR table and set the ITR values to the
1143 	 * corresponding register.
1144 	 */
1145 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1146 		struct ixgbevf_ring *ring;
1147 
1148 		q_vector = adapter->q_vector[v_idx];
1149 
1150 		ixgbevf_for_each_ring(ring, q_vector->rx)
1151 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1152 
1153 		ixgbevf_for_each_ring(ring, q_vector->tx)
1154 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1155 
1156 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1157 			/* Tx only vector */
1158 			if (adapter->tx_itr_setting == 1)
1159 				q_vector->itr = IXGBE_12K_ITR;
1160 			else
1161 				q_vector->itr = adapter->tx_itr_setting;
1162 		} else {
1163 			/* Rx or Rx/Tx vector */
1164 			if (adapter->rx_itr_setting == 1)
1165 				q_vector->itr = IXGBE_20K_ITR;
1166 			else
1167 				q_vector->itr = adapter->rx_itr_setting;
1168 		}
1169 
1170 		/* add q_vector eims value to global eims_enable_mask */
1171 		adapter->eims_enable_mask |= BIT(v_idx);
1172 
1173 		ixgbevf_write_eitr(q_vector);
1174 	}
1175 
1176 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1177 	/* setup eims_other and add value to global eims_enable_mask */
1178 	adapter->eims_other = BIT(v_idx);
1179 	adapter->eims_enable_mask |= adapter->eims_other;
1180 }
1181 
1182 enum latency_range {
1183 	lowest_latency = 0,
1184 	low_latency = 1,
1185 	bulk_latency = 2,
1186 	latency_invalid = 255
1187 };
1188 
1189 /**
1190  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1191  * @q_vector: structure containing interrupt and ring information
1192  * @ring_container: structure containing ring performance data
1193  *
1194  * Stores a new ITR value based on packets and byte
1195  * counts during the last interrupt.  The advantage of per interrupt
1196  * computation is faster updates and more accurate ITR for the current
1197  * traffic pattern.  Constants in this function were computed
1198  * based on theoretical maximum wire speed and thresholds were set based
1199  * on testing data as well as attempting to minimize response time
1200  * while increasing bulk throughput.
1201  **/
1202 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1203 			       struct ixgbevf_ring_container *ring_container)
1204 {
1205 	int bytes = ring_container->total_bytes;
1206 	int packets = ring_container->total_packets;
1207 	u32 timepassed_us;
1208 	u64 bytes_perint;
1209 	u8 itr_setting = ring_container->itr;
1210 
1211 	if (packets == 0)
1212 		return;
1213 
1214 	/* simple throttle rate management
1215 	 *    0-20MB/s lowest (100000 ints/s)
1216 	 *   20-100MB/s low   (20000 ints/s)
1217 	 *  100-1249MB/s bulk (12000 ints/s)
1218 	 */
1219 	/* what was last interrupt timeslice? */
1220 	timepassed_us = q_vector->itr >> 2;
1221 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1222 
1223 	switch (itr_setting) {
1224 	case lowest_latency:
1225 		if (bytes_perint > 10)
1226 			itr_setting = low_latency;
1227 		break;
1228 	case low_latency:
1229 		if (bytes_perint > 20)
1230 			itr_setting = bulk_latency;
1231 		else if (bytes_perint <= 10)
1232 			itr_setting = lowest_latency;
1233 		break;
1234 	case bulk_latency:
1235 		if (bytes_perint <= 20)
1236 			itr_setting = low_latency;
1237 		break;
1238 	}
1239 
1240 	/* clear work counters since we have the values we need */
1241 	ring_container->total_bytes = 0;
1242 	ring_container->total_packets = 0;
1243 
1244 	/* write updated itr to ring container */
1245 	ring_container->itr = itr_setting;
1246 }
1247 
1248 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1249 {
1250 	u32 new_itr = q_vector->itr;
1251 	u8 current_itr;
1252 
1253 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1254 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1255 
1256 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1257 
1258 	switch (current_itr) {
1259 	/* counts and packets in update_itr are dependent on these numbers */
1260 	case lowest_latency:
1261 		new_itr = IXGBE_100K_ITR;
1262 		break;
1263 	case low_latency:
1264 		new_itr = IXGBE_20K_ITR;
1265 		break;
1266 	case bulk_latency:
1267 		new_itr = IXGBE_12K_ITR;
1268 		break;
1269 	default:
1270 		break;
1271 	}
1272 
1273 	if (new_itr != q_vector->itr) {
1274 		/* do an exponential smoothing */
1275 		new_itr = (10 * new_itr * q_vector->itr) /
1276 			  ((9 * new_itr) + q_vector->itr);
1277 
1278 		/* save the algorithm value here */
1279 		q_vector->itr = new_itr;
1280 
1281 		ixgbevf_write_eitr(q_vector);
1282 	}
1283 }
1284 
1285 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1286 {
1287 	struct ixgbevf_adapter *adapter = data;
1288 	struct ixgbe_hw *hw = &adapter->hw;
1289 
1290 	hw->mac.get_link_status = 1;
1291 
1292 	ixgbevf_service_event_schedule(adapter);
1293 
1294 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1295 
1296 	return IRQ_HANDLED;
1297 }
1298 
1299 /**
1300  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1301  * @irq: unused
1302  * @data: pointer to our q_vector struct for this interrupt vector
1303  **/
1304 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1305 {
1306 	struct ixgbevf_q_vector *q_vector = data;
1307 
1308 	/* EIAM disabled interrupts (on this vector) for us */
1309 	if (q_vector->rx.ring || q_vector->tx.ring)
1310 		napi_schedule_irqoff(&q_vector->napi);
1311 
1312 	return IRQ_HANDLED;
1313 }
1314 
1315 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1316 				     int r_idx)
1317 {
1318 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1319 
1320 	a->rx_ring[r_idx]->next = q_vector->rx.ring;
1321 	q_vector->rx.ring = a->rx_ring[r_idx];
1322 	q_vector->rx.count++;
1323 }
1324 
1325 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1326 				     int t_idx)
1327 {
1328 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1329 
1330 	a->tx_ring[t_idx]->next = q_vector->tx.ring;
1331 	q_vector->tx.ring = a->tx_ring[t_idx];
1332 	q_vector->tx.count++;
1333 }
1334 
1335 /**
1336  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1337  * @adapter: board private structure to initialize
1338  *
1339  * This function maps descriptor rings to the queue-specific vectors
1340  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1341  * one vector per ring/queue, but on a constrained vector budget, we
1342  * group the rings as "efficiently" as possible.  You would add new
1343  * mapping configurations in here.
1344  **/
1345 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1346 {
1347 	int q_vectors;
1348 	int v_start = 0;
1349 	int rxr_idx = 0, txr_idx = 0;
1350 	int rxr_remaining = adapter->num_rx_queues;
1351 	int txr_remaining = adapter->num_tx_queues;
1352 	int i, j;
1353 	int rqpv, tqpv;
1354 
1355 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1356 
1357 	/* The ideal configuration...
1358 	 * We have enough vectors to map one per queue.
1359 	 */
1360 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1361 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1362 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1363 
1364 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1365 			map_vector_to_txq(adapter, v_start, txr_idx);
1366 		return 0;
1367 	}
1368 
1369 	/* If we don't have enough vectors for a 1-to-1
1370 	 * mapping, we'll have to group them so there are
1371 	 * multiple queues per vector.
1372 	 */
1373 	/* Re-adjusting *qpv takes care of the remainder. */
1374 	for (i = v_start; i < q_vectors; i++) {
1375 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1376 		for (j = 0; j < rqpv; j++) {
1377 			map_vector_to_rxq(adapter, i, rxr_idx);
1378 			rxr_idx++;
1379 			rxr_remaining--;
1380 		}
1381 	}
1382 	for (i = v_start; i < q_vectors; i++) {
1383 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1384 		for (j = 0; j < tqpv; j++) {
1385 			map_vector_to_txq(adapter, i, txr_idx);
1386 			txr_idx++;
1387 			txr_remaining--;
1388 		}
1389 	}
1390 
1391 	return 0;
1392 }
1393 
1394 /**
1395  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1396  * @adapter: board private structure
1397  *
1398  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1399  * interrupts from the kernel.
1400  **/
1401 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1402 {
1403 	struct net_device *netdev = adapter->netdev;
1404 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1405 	int vector, err;
1406 	int ri = 0, ti = 0;
1407 
1408 	for (vector = 0; vector < q_vectors; vector++) {
1409 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1410 		struct msix_entry *entry = &adapter->msix_entries[vector];
1411 
1412 		if (q_vector->tx.ring && q_vector->rx.ring) {
1413 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1414 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
1415 			ti++;
1416 		} else if (q_vector->rx.ring) {
1417 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1418 				 "%s-%s-%d", netdev->name, "rx", ri++);
1419 		} else if (q_vector->tx.ring) {
1420 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1421 				 "%s-%s-%d", netdev->name, "tx", ti++);
1422 		} else {
1423 			/* skip this unused q_vector */
1424 			continue;
1425 		}
1426 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1427 				  q_vector->name, q_vector);
1428 		if (err) {
1429 			hw_dbg(&adapter->hw,
1430 			       "request_irq failed for MSIX interrupt Error: %d\n",
1431 			       err);
1432 			goto free_queue_irqs;
1433 		}
1434 	}
1435 
1436 	err = request_irq(adapter->msix_entries[vector].vector,
1437 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1438 	if (err) {
1439 		hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1440 		       err);
1441 		goto free_queue_irqs;
1442 	}
1443 
1444 	return 0;
1445 
1446 free_queue_irqs:
1447 	while (vector) {
1448 		vector--;
1449 		free_irq(adapter->msix_entries[vector].vector,
1450 			 adapter->q_vector[vector]);
1451 	}
1452 	/* This failure is non-recoverable - it indicates the system is
1453 	 * out of MSIX vector resources and the VF driver cannot run
1454 	 * without them.  Set the number of msix vectors to zero
1455 	 * indicating that not enough can be allocated.  The error
1456 	 * will be returned to the user indicating device open failed.
1457 	 * Any further attempts to force the driver to open will also
1458 	 * fail.  The only way to recover is to unload the driver and
1459 	 * reload it again.  If the system has recovered some MSIX
1460 	 * vectors then it may succeed.
1461 	 */
1462 	adapter->num_msix_vectors = 0;
1463 	return err;
1464 }
1465 
1466 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1467 {
1468 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1469 
1470 	for (i = 0; i < q_vectors; i++) {
1471 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1472 
1473 		q_vector->rx.ring = NULL;
1474 		q_vector->tx.ring = NULL;
1475 		q_vector->rx.count = 0;
1476 		q_vector->tx.count = 0;
1477 	}
1478 }
1479 
1480 /**
1481  * ixgbevf_request_irq - initialize interrupts
1482  * @adapter: board private structure
1483  *
1484  * Attempts to configure interrupts using the best available
1485  * capabilities of the hardware and kernel.
1486  **/
1487 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1488 {
1489 	int err = ixgbevf_request_msix_irqs(adapter);
1490 
1491 	if (err)
1492 		hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1493 
1494 	return err;
1495 }
1496 
1497 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1498 {
1499 	int i, q_vectors;
1500 
1501 	if (!adapter->msix_entries)
1502 		return;
1503 
1504 	q_vectors = adapter->num_msix_vectors;
1505 	i = q_vectors - 1;
1506 
1507 	free_irq(adapter->msix_entries[i].vector, adapter);
1508 	i--;
1509 
1510 	for (; i >= 0; i--) {
1511 		/* free only the irqs that were actually requested */
1512 		if (!adapter->q_vector[i]->rx.ring &&
1513 		    !adapter->q_vector[i]->tx.ring)
1514 			continue;
1515 
1516 		free_irq(adapter->msix_entries[i].vector,
1517 			 adapter->q_vector[i]);
1518 	}
1519 
1520 	ixgbevf_reset_q_vectors(adapter);
1521 }
1522 
1523 /**
1524  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1525  * @adapter: board private structure
1526  **/
1527 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1528 {
1529 	struct ixgbe_hw *hw = &adapter->hw;
1530 	int i;
1531 
1532 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1533 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1534 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1535 
1536 	IXGBE_WRITE_FLUSH(hw);
1537 
1538 	for (i = 0; i < adapter->num_msix_vectors; i++)
1539 		synchronize_irq(adapter->msix_entries[i].vector);
1540 }
1541 
1542 /**
1543  * ixgbevf_irq_enable - Enable default interrupt generation settings
1544  * @adapter: board private structure
1545  **/
1546 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1547 {
1548 	struct ixgbe_hw *hw = &adapter->hw;
1549 
1550 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1551 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1552 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1553 }
1554 
1555 /**
1556  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1557  * @adapter: board private structure
1558  * @ring: structure containing ring specific data
1559  *
1560  * Configure the Tx descriptor ring after a reset.
1561  **/
1562 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1563 				      struct ixgbevf_ring *ring)
1564 {
1565 	struct ixgbe_hw *hw = &adapter->hw;
1566 	u64 tdba = ring->dma;
1567 	int wait_loop = 10;
1568 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1569 	u8 reg_idx = ring->reg_idx;
1570 
1571 	/* disable queue to avoid issues while updating state */
1572 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1573 	IXGBE_WRITE_FLUSH(hw);
1574 
1575 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1576 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1577 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1578 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1579 
1580 	/* disable head writeback */
1581 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1582 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1583 
1584 	/* enable relaxed ordering */
1585 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1586 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1587 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1588 
1589 	/* reset head and tail pointers */
1590 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1591 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1592 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1593 
1594 	/* reset ntu and ntc to place SW in sync with hardwdare */
1595 	ring->next_to_clean = 0;
1596 	ring->next_to_use = 0;
1597 
1598 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1599 	 * to or less than the number of on chip descriptors, which is
1600 	 * currently 40.
1601 	 */
1602 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1603 
1604 	/* Setting PTHRESH to 32 both improves performance */
1605 	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1606 		   32;           /* PTHRESH = 32 */
1607 
1608 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1609 
1610 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1611 
1612 	/* poll to verify queue is enabled */
1613 	do {
1614 		usleep_range(1000, 2000);
1615 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1616 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1617 	if (!wait_loop)
1618 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1619 }
1620 
1621 /**
1622  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1623  * @adapter: board private structure
1624  *
1625  * Configure the Tx unit of the MAC after a reset.
1626  **/
1627 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1628 {
1629 	u32 i;
1630 
1631 	/* Setup the HW Tx Head and Tail descriptor pointers */
1632 	for (i = 0; i < adapter->num_tx_queues; i++)
1633 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1634 }
1635 
1636 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1637 
1638 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1639 {
1640 	struct ixgbe_hw *hw = &adapter->hw;
1641 	u32 srrctl;
1642 
1643 	srrctl = IXGBE_SRRCTL_DROP_EN;
1644 
1645 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1646 	srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1647 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1648 
1649 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1650 }
1651 
1652 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1653 {
1654 	struct ixgbe_hw *hw = &adapter->hw;
1655 
1656 	/* PSRTYPE must be initialized in 82599 */
1657 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1658 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1659 		      IXGBE_PSRTYPE_L2HDR;
1660 
1661 	if (adapter->num_rx_queues > 1)
1662 		psrtype |= BIT(29);
1663 
1664 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1665 }
1666 
1667 #define IXGBEVF_MAX_RX_DESC_POLL 10
1668 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1669 				     struct ixgbevf_ring *ring)
1670 {
1671 	struct ixgbe_hw *hw = &adapter->hw;
1672 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1673 	u32 rxdctl;
1674 	u8 reg_idx = ring->reg_idx;
1675 
1676 	if (IXGBE_REMOVED(hw->hw_addr))
1677 		return;
1678 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1679 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1680 
1681 	/* write value back with RXDCTL.ENABLE bit cleared */
1682 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1683 
1684 	/* the hardware may take up to 100us to really disable the Rx queue */
1685 	do {
1686 		udelay(10);
1687 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1688 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1689 
1690 	if (!wait_loop)
1691 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1692 		       reg_idx);
1693 }
1694 
1695 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1696 					 struct ixgbevf_ring *ring)
1697 {
1698 	struct ixgbe_hw *hw = &adapter->hw;
1699 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1700 	u32 rxdctl;
1701 	u8 reg_idx = ring->reg_idx;
1702 
1703 	if (IXGBE_REMOVED(hw->hw_addr))
1704 		return;
1705 	do {
1706 		usleep_range(1000, 2000);
1707 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1708 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1709 
1710 	if (!wait_loop)
1711 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1712 		       reg_idx);
1713 }
1714 
1715 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1716 {
1717 	struct ixgbe_hw *hw = &adapter->hw;
1718 	u32 vfmrqc = 0, vfreta = 0;
1719 	u16 rss_i = adapter->num_rx_queues;
1720 	u8 i, j;
1721 
1722 	/* Fill out hash function seeds */
1723 	netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1724 	for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1725 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1726 
1727 	for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1728 		if (j == rss_i)
1729 			j = 0;
1730 
1731 		adapter->rss_indir_tbl[i] = j;
1732 
1733 		vfreta |= j << (i & 0x3) * 8;
1734 		if ((i & 3) == 3) {
1735 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1736 			vfreta = 0;
1737 		}
1738 	}
1739 
1740 	/* Perform hash on these packet types */
1741 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1742 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1743 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1744 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1745 
1746 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1747 
1748 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1749 }
1750 
1751 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1752 				      struct ixgbevf_ring *ring)
1753 {
1754 	struct ixgbe_hw *hw = &adapter->hw;
1755 	u64 rdba = ring->dma;
1756 	u32 rxdctl;
1757 	u8 reg_idx = ring->reg_idx;
1758 
1759 	/* disable queue to avoid issues while updating state */
1760 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1761 	ixgbevf_disable_rx_queue(adapter, ring);
1762 
1763 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1764 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1765 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1766 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1767 
1768 #ifndef CONFIG_SPARC
1769 	/* enable relaxed ordering */
1770 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1771 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1772 #else
1773 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1774 			IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1775 			IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1776 #endif
1777 
1778 	/* reset head and tail pointers */
1779 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1780 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1781 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1782 
1783 	/* reset ntu and ntc to place SW in sync with hardwdare */
1784 	ring->next_to_clean = 0;
1785 	ring->next_to_use = 0;
1786 	ring->next_to_alloc = 0;
1787 
1788 	ixgbevf_configure_srrctl(adapter, reg_idx);
1789 
1790 	/* allow any size packet since we can handle overflow */
1791 	rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1792 
1793 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1794 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1795 
1796 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1797 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1798 }
1799 
1800 /**
1801  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1802  * @adapter: board private structure
1803  *
1804  * Configure the Rx unit of the MAC after a reset.
1805  **/
1806 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1807 {
1808 	struct ixgbe_hw *hw = &adapter->hw;
1809 	struct net_device *netdev = adapter->netdev;
1810 	int i, ret;
1811 
1812 	ixgbevf_setup_psrtype(adapter);
1813 	if (hw->mac.type >= ixgbe_mac_X550_vf)
1814 		ixgbevf_setup_vfmrqc(adapter);
1815 
1816 	spin_lock_bh(&adapter->mbx_lock);
1817 	/* notify the PF of our intent to use this size of frame */
1818 	ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1819 	spin_unlock_bh(&adapter->mbx_lock);
1820 	if (ret)
1821 		dev_err(&adapter->pdev->dev,
1822 			"Failed to set MTU at %d\n", netdev->mtu);
1823 
1824 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1825 	 * the Base and Length of the Rx Descriptor Ring
1826 	 */
1827 	for (i = 0; i < adapter->num_rx_queues; i++)
1828 		ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1829 }
1830 
1831 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1832 				   __be16 proto, u16 vid)
1833 {
1834 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1835 	struct ixgbe_hw *hw = &adapter->hw;
1836 	int err;
1837 
1838 	spin_lock_bh(&adapter->mbx_lock);
1839 
1840 	/* add VID to filter table */
1841 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1842 
1843 	spin_unlock_bh(&adapter->mbx_lock);
1844 
1845 	/* translate error return types so error makes sense */
1846 	if (err == IXGBE_ERR_MBX)
1847 		return -EIO;
1848 
1849 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
1850 		return -EACCES;
1851 
1852 	set_bit(vid, adapter->active_vlans);
1853 
1854 	return err;
1855 }
1856 
1857 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1858 				    __be16 proto, u16 vid)
1859 {
1860 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1861 	struct ixgbe_hw *hw = &adapter->hw;
1862 	int err;
1863 
1864 	spin_lock_bh(&adapter->mbx_lock);
1865 
1866 	/* remove VID from filter table */
1867 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1868 
1869 	spin_unlock_bh(&adapter->mbx_lock);
1870 
1871 	clear_bit(vid, adapter->active_vlans);
1872 
1873 	return err;
1874 }
1875 
1876 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1877 {
1878 	u16 vid;
1879 
1880 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1881 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
1882 					htons(ETH_P_8021Q), vid);
1883 }
1884 
1885 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1886 {
1887 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1888 	struct ixgbe_hw *hw = &adapter->hw;
1889 	int count = 0;
1890 
1891 	if ((netdev_uc_count(netdev)) > 10) {
1892 		pr_err("Too many unicast filters - No Space\n");
1893 		return -ENOSPC;
1894 	}
1895 
1896 	if (!netdev_uc_empty(netdev)) {
1897 		struct netdev_hw_addr *ha;
1898 
1899 		netdev_for_each_uc_addr(ha, netdev) {
1900 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1901 			udelay(200);
1902 		}
1903 	} else {
1904 		/* If the list is empty then send message to PF driver to
1905 		 * clear all MAC VLANs on this VF.
1906 		 */
1907 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1908 	}
1909 
1910 	return count;
1911 }
1912 
1913 /**
1914  * ixgbevf_set_rx_mode - Multicast and unicast set
1915  * @netdev: network interface device structure
1916  *
1917  * The set_rx_method entry point is called whenever the multicast address
1918  * list, unicast address list or the network interface flags are updated.
1919  * This routine is responsible for configuring the hardware for proper
1920  * multicast mode and configuring requested unicast filters.
1921  **/
1922 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1923 {
1924 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1925 	struct ixgbe_hw *hw = &adapter->hw;
1926 	unsigned int flags = netdev->flags;
1927 	int xcast_mode;
1928 
1929 	xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1930 		     (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1931 		     IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1932 
1933 	spin_lock_bh(&adapter->mbx_lock);
1934 
1935 	hw->mac.ops.update_xcast_mode(hw, xcast_mode);
1936 
1937 	/* reprogram multicast list */
1938 	hw->mac.ops.update_mc_addr_list(hw, netdev);
1939 
1940 	ixgbevf_write_uc_addr_list(netdev);
1941 
1942 	spin_unlock_bh(&adapter->mbx_lock);
1943 }
1944 
1945 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1946 {
1947 	int q_idx;
1948 	struct ixgbevf_q_vector *q_vector;
1949 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1950 
1951 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1952 		q_vector = adapter->q_vector[q_idx];
1953 #ifdef CONFIG_NET_RX_BUSY_POLL
1954 		ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1955 #endif
1956 		napi_enable(&q_vector->napi);
1957 	}
1958 }
1959 
1960 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1961 {
1962 	int q_idx;
1963 	struct ixgbevf_q_vector *q_vector;
1964 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1965 
1966 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1967 		q_vector = adapter->q_vector[q_idx];
1968 		napi_disable(&q_vector->napi);
1969 #ifdef CONFIG_NET_RX_BUSY_POLL
1970 		while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1971 			pr_info("QV %d locked\n", q_idx);
1972 			usleep_range(1000, 20000);
1973 		}
1974 #endif /* CONFIG_NET_RX_BUSY_POLL */
1975 	}
1976 }
1977 
1978 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1979 {
1980 	struct ixgbe_hw *hw = &adapter->hw;
1981 	unsigned int def_q = 0;
1982 	unsigned int num_tcs = 0;
1983 	unsigned int num_rx_queues = adapter->num_rx_queues;
1984 	unsigned int num_tx_queues = adapter->num_tx_queues;
1985 	int err;
1986 
1987 	spin_lock_bh(&adapter->mbx_lock);
1988 
1989 	/* fetch queue configuration from the PF */
1990 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1991 
1992 	spin_unlock_bh(&adapter->mbx_lock);
1993 
1994 	if (err)
1995 		return err;
1996 
1997 	if (num_tcs > 1) {
1998 		/* we need only one Tx queue */
1999 		num_tx_queues = 1;
2000 
2001 		/* update default Tx ring register index */
2002 		adapter->tx_ring[0]->reg_idx = def_q;
2003 
2004 		/* we need as many queues as traffic classes */
2005 		num_rx_queues = num_tcs;
2006 	}
2007 
2008 	/* if we have a bad config abort request queue reset */
2009 	if ((adapter->num_rx_queues != num_rx_queues) ||
2010 	    (adapter->num_tx_queues != num_tx_queues)) {
2011 		/* force mailbox timeout to prevent further messages */
2012 		hw->mbx.timeout = 0;
2013 
2014 		/* wait for watchdog to come around and bail us out */
2015 		set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2016 	}
2017 
2018 	return 0;
2019 }
2020 
2021 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2022 {
2023 	ixgbevf_configure_dcb(adapter);
2024 
2025 	ixgbevf_set_rx_mode(adapter->netdev);
2026 
2027 	ixgbevf_restore_vlan(adapter);
2028 
2029 	ixgbevf_configure_tx(adapter);
2030 	ixgbevf_configure_rx(adapter);
2031 }
2032 
2033 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2034 {
2035 	/* Only save pre-reset stats if there are some */
2036 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2037 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2038 			adapter->stats.base_vfgprc;
2039 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2040 			adapter->stats.base_vfgptc;
2041 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2042 			adapter->stats.base_vfgorc;
2043 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2044 			adapter->stats.base_vfgotc;
2045 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2046 			adapter->stats.base_vfmprc;
2047 	}
2048 }
2049 
2050 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2051 {
2052 	struct ixgbe_hw *hw = &adapter->hw;
2053 
2054 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2055 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2056 	adapter->stats.last_vfgorc |=
2057 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2058 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2059 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2060 	adapter->stats.last_vfgotc |=
2061 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2062 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2063 
2064 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2065 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2066 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2067 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2068 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2069 }
2070 
2071 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2072 {
2073 	struct ixgbe_hw *hw = &adapter->hw;
2074 	int api[] = { ixgbe_mbox_api_12,
2075 		      ixgbe_mbox_api_11,
2076 		      ixgbe_mbox_api_10,
2077 		      ixgbe_mbox_api_unknown };
2078 	int err, idx = 0;
2079 
2080 	spin_lock_bh(&adapter->mbx_lock);
2081 
2082 	while (api[idx] != ixgbe_mbox_api_unknown) {
2083 		err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2084 		if (!err)
2085 			break;
2086 		idx++;
2087 	}
2088 
2089 	spin_unlock_bh(&adapter->mbx_lock);
2090 }
2091 
2092 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2093 {
2094 	struct net_device *netdev = adapter->netdev;
2095 	struct ixgbe_hw *hw = &adapter->hw;
2096 
2097 	ixgbevf_configure_msix(adapter);
2098 
2099 	spin_lock_bh(&adapter->mbx_lock);
2100 
2101 	if (is_valid_ether_addr(hw->mac.addr))
2102 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2103 	else
2104 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2105 
2106 	spin_unlock_bh(&adapter->mbx_lock);
2107 
2108 	smp_mb__before_atomic();
2109 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2110 	ixgbevf_napi_enable_all(adapter);
2111 
2112 	/* clear any pending interrupts, may auto mask */
2113 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2114 	ixgbevf_irq_enable(adapter);
2115 
2116 	/* enable transmits */
2117 	netif_tx_start_all_queues(netdev);
2118 
2119 	ixgbevf_save_reset_stats(adapter);
2120 	ixgbevf_init_last_counter_stats(adapter);
2121 
2122 	hw->mac.get_link_status = 1;
2123 	mod_timer(&adapter->service_timer, jiffies);
2124 }
2125 
2126 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2127 {
2128 	ixgbevf_configure(adapter);
2129 
2130 	ixgbevf_up_complete(adapter);
2131 }
2132 
2133 /**
2134  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2135  * @rx_ring: ring to free buffers from
2136  **/
2137 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2138 {
2139 	struct device *dev = rx_ring->dev;
2140 	unsigned long size;
2141 	unsigned int i;
2142 
2143 	/* Free Rx ring sk_buff */
2144 	if (rx_ring->skb) {
2145 		dev_kfree_skb(rx_ring->skb);
2146 		rx_ring->skb = NULL;
2147 	}
2148 
2149 	/* ring already cleared, nothing to do */
2150 	if (!rx_ring->rx_buffer_info)
2151 		return;
2152 
2153 	/* Free all the Rx ring pages */
2154 	for (i = 0; i < rx_ring->count; i++) {
2155 		struct ixgbevf_rx_buffer *rx_buffer;
2156 
2157 		rx_buffer = &rx_ring->rx_buffer_info[i];
2158 		if (rx_buffer->dma)
2159 			dma_unmap_page(dev, rx_buffer->dma,
2160 				       PAGE_SIZE, DMA_FROM_DEVICE);
2161 		rx_buffer->dma = 0;
2162 		if (rx_buffer->page)
2163 			__free_page(rx_buffer->page);
2164 		rx_buffer->page = NULL;
2165 	}
2166 
2167 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2168 	memset(rx_ring->rx_buffer_info, 0, size);
2169 
2170 	/* Zero out the descriptor ring */
2171 	memset(rx_ring->desc, 0, rx_ring->size);
2172 }
2173 
2174 /**
2175  * ixgbevf_clean_tx_ring - Free Tx Buffers
2176  * @tx_ring: ring to be cleaned
2177  **/
2178 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2179 {
2180 	struct ixgbevf_tx_buffer *tx_buffer_info;
2181 	unsigned long size;
2182 	unsigned int i;
2183 
2184 	if (!tx_ring->tx_buffer_info)
2185 		return;
2186 
2187 	/* Free all the Tx ring sk_buffs */
2188 	for (i = 0; i < tx_ring->count; i++) {
2189 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2190 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2191 	}
2192 
2193 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2194 	memset(tx_ring->tx_buffer_info, 0, size);
2195 
2196 	memset(tx_ring->desc, 0, tx_ring->size);
2197 }
2198 
2199 /**
2200  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2201  * @adapter: board private structure
2202  **/
2203 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2204 {
2205 	int i;
2206 
2207 	for (i = 0; i < adapter->num_rx_queues; i++)
2208 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2209 }
2210 
2211 /**
2212  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2213  * @adapter: board private structure
2214  **/
2215 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2216 {
2217 	int i;
2218 
2219 	for (i = 0; i < adapter->num_tx_queues; i++)
2220 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2221 }
2222 
2223 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2224 {
2225 	struct net_device *netdev = adapter->netdev;
2226 	struct ixgbe_hw *hw = &adapter->hw;
2227 	int i;
2228 
2229 	/* signal that we are down to the interrupt handler */
2230 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2231 		return; /* do nothing if already down */
2232 
2233 	/* disable all enabled Rx queues */
2234 	for (i = 0; i < adapter->num_rx_queues; i++)
2235 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2236 
2237 	usleep_range(10000, 20000);
2238 
2239 	netif_tx_stop_all_queues(netdev);
2240 
2241 	/* call carrier off first to avoid false dev_watchdog timeouts */
2242 	netif_carrier_off(netdev);
2243 	netif_tx_disable(netdev);
2244 
2245 	ixgbevf_irq_disable(adapter);
2246 
2247 	ixgbevf_napi_disable_all(adapter);
2248 
2249 	del_timer_sync(&adapter->service_timer);
2250 
2251 	/* disable transmits in the hardware now that interrupts are off */
2252 	for (i = 0; i < adapter->num_tx_queues; i++) {
2253 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2254 
2255 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2256 				IXGBE_TXDCTL_SWFLSH);
2257 	}
2258 
2259 	if (!pci_channel_offline(adapter->pdev))
2260 		ixgbevf_reset(adapter);
2261 
2262 	ixgbevf_clean_all_tx_rings(adapter);
2263 	ixgbevf_clean_all_rx_rings(adapter);
2264 }
2265 
2266 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2267 {
2268 	WARN_ON(in_interrupt());
2269 
2270 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2271 		msleep(1);
2272 
2273 	ixgbevf_down(adapter);
2274 	ixgbevf_up(adapter);
2275 
2276 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2277 }
2278 
2279 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2280 {
2281 	struct ixgbe_hw *hw = &adapter->hw;
2282 	struct net_device *netdev = adapter->netdev;
2283 
2284 	if (hw->mac.ops.reset_hw(hw)) {
2285 		hw_dbg(hw, "PF still resetting\n");
2286 	} else {
2287 		hw->mac.ops.init_hw(hw);
2288 		ixgbevf_negotiate_api(adapter);
2289 	}
2290 
2291 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2292 		ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2293 		ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2294 	}
2295 
2296 	adapter->last_reset = jiffies;
2297 }
2298 
2299 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2300 					int vectors)
2301 {
2302 	int vector_threshold;
2303 
2304 	/* We'll want at least 2 (vector_threshold):
2305 	 * 1) TxQ[0] + RxQ[0] handler
2306 	 * 2) Other (Link Status Change, etc.)
2307 	 */
2308 	vector_threshold = MIN_MSIX_COUNT;
2309 
2310 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2311 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2312 	 * Right now, we simply care about how many we'll get; we'll
2313 	 * set them up later while requesting irq's.
2314 	 */
2315 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2316 					vector_threshold, vectors);
2317 
2318 	if (vectors < 0) {
2319 		dev_err(&adapter->pdev->dev,
2320 			"Unable to allocate MSI-X interrupts\n");
2321 		kfree(adapter->msix_entries);
2322 		adapter->msix_entries = NULL;
2323 		return vectors;
2324 	}
2325 
2326 	/* Adjust for only the vectors we'll use, which is minimum
2327 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2328 	 * vectors we were allocated.
2329 	 */
2330 	adapter->num_msix_vectors = vectors;
2331 
2332 	return 0;
2333 }
2334 
2335 /**
2336  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2337  * @adapter: board private structure to initialize
2338  *
2339  * This is the top level queue allocation routine.  The order here is very
2340  * important, starting with the "most" number of features turned on at once,
2341  * and ending with the smallest set of features.  This way large combinations
2342  * can be allocated if they're turned on, and smaller combinations are the
2343  * fallthrough conditions.
2344  *
2345  **/
2346 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2347 {
2348 	struct ixgbe_hw *hw = &adapter->hw;
2349 	unsigned int def_q = 0;
2350 	unsigned int num_tcs = 0;
2351 	int err;
2352 
2353 	/* Start with base case */
2354 	adapter->num_rx_queues = 1;
2355 	adapter->num_tx_queues = 1;
2356 
2357 	spin_lock_bh(&adapter->mbx_lock);
2358 
2359 	/* fetch queue configuration from the PF */
2360 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2361 
2362 	spin_unlock_bh(&adapter->mbx_lock);
2363 
2364 	if (err)
2365 		return;
2366 
2367 	/* we need as many queues as traffic classes */
2368 	if (num_tcs > 1) {
2369 		adapter->num_rx_queues = num_tcs;
2370 	} else {
2371 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2372 
2373 		switch (hw->api_version) {
2374 		case ixgbe_mbox_api_11:
2375 		case ixgbe_mbox_api_12:
2376 			adapter->num_rx_queues = rss;
2377 			adapter->num_tx_queues = rss;
2378 		default:
2379 			break;
2380 		}
2381 	}
2382 }
2383 
2384 /**
2385  * ixgbevf_alloc_queues - Allocate memory for all rings
2386  * @adapter: board private structure to initialize
2387  *
2388  * We allocate one ring per queue at run-time since we don't know the
2389  * number of queues at compile-time.  The polling_netdev array is
2390  * intended for Multiqueue, but should work fine with a single queue.
2391  **/
2392 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2393 {
2394 	struct ixgbevf_ring *ring;
2395 	int rx = 0, tx = 0;
2396 
2397 	for (; tx < adapter->num_tx_queues; tx++) {
2398 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2399 		if (!ring)
2400 			goto err_allocation;
2401 
2402 		ring->dev = &adapter->pdev->dev;
2403 		ring->netdev = adapter->netdev;
2404 		ring->count = adapter->tx_ring_count;
2405 		ring->queue_index = tx;
2406 		ring->reg_idx = tx;
2407 
2408 		adapter->tx_ring[tx] = ring;
2409 	}
2410 
2411 	for (; rx < adapter->num_rx_queues; rx++) {
2412 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2413 		if (!ring)
2414 			goto err_allocation;
2415 
2416 		ring->dev = &adapter->pdev->dev;
2417 		ring->netdev = adapter->netdev;
2418 
2419 		ring->count = adapter->rx_ring_count;
2420 		ring->queue_index = rx;
2421 		ring->reg_idx = rx;
2422 
2423 		adapter->rx_ring[rx] = ring;
2424 	}
2425 
2426 	return 0;
2427 
2428 err_allocation:
2429 	while (tx) {
2430 		kfree(adapter->tx_ring[--tx]);
2431 		adapter->tx_ring[tx] = NULL;
2432 	}
2433 
2434 	while (rx) {
2435 		kfree(adapter->rx_ring[--rx]);
2436 		adapter->rx_ring[rx] = NULL;
2437 	}
2438 	return -ENOMEM;
2439 }
2440 
2441 /**
2442  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2443  * @adapter: board private structure to initialize
2444  *
2445  * Attempt to configure the interrupts using the best available
2446  * capabilities of the hardware and the kernel.
2447  **/
2448 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2449 {
2450 	struct net_device *netdev = adapter->netdev;
2451 	int err;
2452 	int vector, v_budget;
2453 
2454 	/* It's easy to be greedy for MSI-X vectors, but it really
2455 	 * doesn't do us much good if we have a lot more vectors
2456 	 * than CPU's.  So let's be conservative and only ask for
2457 	 * (roughly) the same number of vectors as there are CPU's.
2458 	 * The default is to use pairs of vectors.
2459 	 */
2460 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2461 	v_budget = min_t(int, v_budget, num_online_cpus());
2462 	v_budget += NON_Q_VECTORS;
2463 
2464 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2465 	 * mean we disable MSI-X capabilities of the adapter.
2466 	 */
2467 	adapter->msix_entries = kcalloc(v_budget,
2468 					sizeof(struct msix_entry), GFP_KERNEL);
2469 	if (!adapter->msix_entries)
2470 		return -ENOMEM;
2471 
2472 	for (vector = 0; vector < v_budget; vector++)
2473 		adapter->msix_entries[vector].entry = vector;
2474 
2475 	err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2476 	if (err)
2477 		return err;
2478 
2479 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2480 	if (err)
2481 		return err;
2482 
2483 	return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2484 }
2485 
2486 /**
2487  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2488  * @adapter: board private structure to initialize
2489  *
2490  * We allocate one q_vector per queue interrupt.  If allocation fails we
2491  * return -ENOMEM.
2492  **/
2493 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2494 {
2495 	int q_idx, num_q_vectors;
2496 	struct ixgbevf_q_vector *q_vector;
2497 
2498 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2499 
2500 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2501 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2502 		if (!q_vector)
2503 			goto err_out;
2504 		q_vector->adapter = adapter;
2505 		q_vector->v_idx = q_idx;
2506 		netif_napi_add(adapter->netdev, &q_vector->napi,
2507 			       ixgbevf_poll, 64);
2508 		adapter->q_vector[q_idx] = q_vector;
2509 	}
2510 
2511 	return 0;
2512 
2513 err_out:
2514 	while (q_idx) {
2515 		q_idx--;
2516 		q_vector = adapter->q_vector[q_idx];
2517 #ifdef CONFIG_NET_RX_BUSY_POLL
2518 		napi_hash_del(&q_vector->napi);
2519 #endif
2520 		netif_napi_del(&q_vector->napi);
2521 		kfree(q_vector);
2522 		adapter->q_vector[q_idx] = NULL;
2523 	}
2524 	return -ENOMEM;
2525 }
2526 
2527 /**
2528  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2529  * @adapter: board private structure to initialize
2530  *
2531  * This function frees the memory allocated to the q_vectors.  In addition if
2532  * NAPI is enabled it will delete any references to the NAPI struct prior
2533  * to freeing the q_vector.
2534  **/
2535 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2536 {
2537 	int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2538 
2539 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2540 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2541 
2542 		adapter->q_vector[q_idx] = NULL;
2543 #ifdef CONFIG_NET_RX_BUSY_POLL
2544 		napi_hash_del(&q_vector->napi);
2545 #endif
2546 		netif_napi_del(&q_vector->napi);
2547 		kfree(q_vector);
2548 	}
2549 }
2550 
2551 /**
2552  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2553  * @adapter: board private structure
2554  *
2555  **/
2556 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2557 {
2558 	if (!adapter->msix_entries)
2559 		return;
2560 
2561 	pci_disable_msix(adapter->pdev);
2562 	kfree(adapter->msix_entries);
2563 	adapter->msix_entries = NULL;
2564 }
2565 
2566 /**
2567  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2568  * @adapter: board private structure to initialize
2569  *
2570  **/
2571 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2572 {
2573 	int err;
2574 
2575 	/* Number of supported queues */
2576 	ixgbevf_set_num_queues(adapter);
2577 
2578 	err = ixgbevf_set_interrupt_capability(adapter);
2579 	if (err) {
2580 		hw_dbg(&adapter->hw,
2581 		       "Unable to setup interrupt capabilities\n");
2582 		goto err_set_interrupt;
2583 	}
2584 
2585 	err = ixgbevf_alloc_q_vectors(adapter);
2586 	if (err) {
2587 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2588 		goto err_alloc_q_vectors;
2589 	}
2590 
2591 	err = ixgbevf_alloc_queues(adapter);
2592 	if (err) {
2593 		pr_err("Unable to allocate memory for queues\n");
2594 		goto err_alloc_queues;
2595 	}
2596 
2597 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2598 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2599 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2600 
2601 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2602 
2603 	return 0;
2604 err_alloc_queues:
2605 	ixgbevf_free_q_vectors(adapter);
2606 err_alloc_q_vectors:
2607 	ixgbevf_reset_interrupt_capability(adapter);
2608 err_set_interrupt:
2609 	return err;
2610 }
2611 
2612 /**
2613  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2614  * @adapter: board private structure to clear interrupt scheme on
2615  *
2616  * We go through and clear interrupt specific resources and reset the structure
2617  * to pre-load conditions
2618  **/
2619 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2620 {
2621 	int i;
2622 
2623 	for (i = 0; i < adapter->num_tx_queues; i++) {
2624 		kfree(adapter->tx_ring[i]);
2625 		adapter->tx_ring[i] = NULL;
2626 	}
2627 	for (i = 0; i < adapter->num_rx_queues; i++) {
2628 		kfree(adapter->rx_ring[i]);
2629 		adapter->rx_ring[i] = NULL;
2630 	}
2631 
2632 	adapter->num_tx_queues = 0;
2633 	adapter->num_rx_queues = 0;
2634 
2635 	ixgbevf_free_q_vectors(adapter);
2636 	ixgbevf_reset_interrupt_capability(adapter);
2637 }
2638 
2639 /**
2640  * ixgbevf_sw_init - Initialize general software structures
2641  * @adapter: board private structure to initialize
2642  *
2643  * ixgbevf_sw_init initializes the Adapter private data structure.
2644  * Fields are initialized based on PCI device information and
2645  * OS network device settings (MTU size).
2646  **/
2647 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2648 {
2649 	struct ixgbe_hw *hw = &adapter->hw;
2650 	struct pci_dev *pdev = adapter->pdev;
2651 	struct net_device *netdev = adapter->netdev;
2652 	int err;
2653 
2654 	/* PCI config space info */
2655 	hw->vendor_id = pdev->vendor;
2656 	hw->device_id = pdev->device;
2657 	hw->revision_id = pdev->revision;
2658 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2659 	hw->subsystem_device_id = pdev->subsystem_device;
2660 
2661 	hw->mbx.ops.init_params(hw);
2662 
2663 	/* assume legacy case in which PF would only give VF 2 queues */
2664 	hw->mac.max_tx_queues = 2;
2665 	hw->mac.max_rx_queues = 2;
2666 
2667 	/* lock to protect mailbox accesses */
2668 	spin_lock_init(&adapter->mbx_lock);
2669 
2670 	err = hw->mac.ops.reset_hw(hw);
2671 	if (err) {
2672 		dev_info(&pdev->dev,
2673 			 "PF still in reset state.  Is the PF interface up?\n");
2674 	} else {
2675 		err = hw->mac.ops.init_hw(hw);
2676 		if (err) {
2677 			pr_err("init_shared_code failed: %d\n", err);
2678 			goto out;
2679 		}
2680 		ixgbevf_negotiate_api(adapter);
2681 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2682 		if (err)
2683 			dev_info(&pdev->dev, "Error reading MAC address\n");
2684 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2685 			dev_info(&pdev->dev,
2686 				 "MAC address not assigned by administrator.\n");
2687 		ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2688 	}
2689 
2690 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2691 		dev_info(&pdev->dev, "Assigning random MAC address\n");
2692 		eth_hw_addr_random(netdev);
2693 		ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2694 		ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2695 	}
2696 
2697 	/* Enable dynamic interrupt throttling rates */
2698 	adapter->rx_itr_setting = 1;
2699 	adapter->tx_itr_setting = 1;
2700 
2701 	/* set default ring sizes */
2702 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2703 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2704 
2705 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2706 	return 0;
2707 
2708 out:
2709 	return err;
2710 }
2711 
2712 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2713 	{							\
2714 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2715 		if (current_counter < last_counter)		\
2716 			counter += 0x100000000LL;		\
2717 		last_counter = current_counter;			\
2718 		counter &= 0xFFFFFFFF00000000LL;		\
2719 		counter |= current_counter;			\
2720 	}
2721 
2722 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2723 	{								 \
2724 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2725 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2726 		u64 current_counter = (current_counter_msb << 32) |	 \
2727 			current_counter_lsb;				 \
2728 		if (current_counter < last_counter)			 \
2729 			counter += 0x1000000000LL;			 \
2730 		last_counter = current_counter;				 \
2731 		counter &= 0xFFFFFFF000000000LL;			 \
2732 		counter |= current_counter;				 \
2733 	}
2734 /**
2735  * ixgbevf_update_stats - Update the board statistics counters.
2736  * @adapter: board private structure
2737  **/
2738 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2739 {
2740 	struct ixgbe_hw *hw = &adapter->hw;
2741 	int i;
2742 
2743 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2744 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2745 		return;
2746 
2747 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2748 				adapter->stats.vfgprc);
2749 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2750 				adapter->stats.vfgptc);
2751 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2752 				adapter->stats.last_vfgorc,
2753 				adapter->stats.vfgorc);
2754 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2755 				adapter->stats.last_vfgotc,
2756 				adapter->stats.vfgotc);
2757 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2758 				adapter->stats.vfmprc);
2759 
2760 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2761 		adapter->hw_csum_rx_error +=
2762 			adapter->rx_ring[i]->hw_csum_rx_error;
2763 		adapter->rx_ring[i]->hw_csum_rx_error = 0;
2764 	}
2765 }
2766 
2767 /**
2768  * ixgbevf_service_timer - Timer Call-back
2769  * @data: pointer to adapter cast into an unsigned long
2770  **/
2771 static void ixgbevf_service_timer(unsigned long data)
2772 {
2773 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2774 
2775 	/* Reset the timer */
2776 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2777 
2778 	ixgbevf_service_event_schedule(adapter);
2779 }
2780 
2781 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2782 {
2783 	if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
2784 		return;
2785 
2786 	/* If we're already down or resetting, just bail */
2787 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2788 	    test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2789 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2790 		return;
2791 
2792 	adapter->tx_timeout_count++;
2793 
2794 	rtnl_lock();
2795 	ixgbevf_reinit_locked(adapter);
2796 	rtnl_unlock();
2797 }
2798 
2799 /**
2800  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2801  * @adapter: pointer to the device adapter structure
2802  *
2803  * This function serves two purposes.  First it strobes the interrupt lines
2804  * in order to make certain interrupts are occurring.  Secondly it sets the
2805  * bits needed to check for TX hangs.  As a result we should immediately
2806  * determine if a hang has occurred.
2807  **/
2808 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2809 {
2810 	struct ixgbe_hw *hw = &adapter->hw;
2811 	u32 eics = 0;
2812 	int i;
2813 
2814 	/* If we're down or resetting, just bail */
2815 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2816 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2817 		return;
2818 
2819 	/* Force detection of hung controller */
2820 	if (netif_carrier_ok(adapter->netdev)) {
2821 		for (i = 0; i < adapter->num_tx_queues; i++)
2822 			set_check_for_tx_hang(adapter->tx_ring[i]);
2823 	}
2824 
2825 	/* get one bit for every active Tx/Rx interrupt vector */
2826 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2827 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2828 
2829 		if (qv->rx.ring || qv->tx.ring)
2830 			eics |= BIT(i);
2831 	}
2832 
2833 	/* Cause software interrupt to ensure rings are cleaned */
2834 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2835 }
2836 
2837 /**
2838  * ixgbevf_watchdog_update_link - update the link status
2839  * @adapter: pointer to the device adapter structure
2840  **/
2841 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2842 {
2843 	struct ixgbe_hw *hw = &adapter->hw;
2844 	u32 link_speed = adapter->link_speed;
2845 	bool link_up = adapter->link_up;
2846 	s32 err;
2847 
2848 	spin_lock_bh(&adapter->mbx_lock);
2849 
2850 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2851 
2852 	spin_unlock_bh(&adapter->mbx_lock);
2853 
2854 	/* if check for link returns error we will need to reset */
2855 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2856 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
2857 		link_up = false;
2858 	}
2859 
2860 	adapter->link_up = link_up;
2861 	adapter->link_speed = link_speed;
2862 }
2863 
2864 /**
2865  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2866  *				 print link up message
2867  * @adapter: pointer to the device adapter structure
2868  **/
2869 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2870 {
2871 	struct net_device *netdev = adapter->netdev;
2872 
2873 	/* only continue if link was previously down */
2874 	if (netif_carrier_ok(netdev))
2875 		return;
2876 
2877 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2878 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2879 		 "10 Gbps" :
2880 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2881 		 "1 Gbps" :
2882 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2883 		 "100 Mbps" :
2884 		 "unknown speed");
2885 
2886 	netif_carrier_on(netdev);
2887 }
2888 
2889 /**
2890  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2891  *				   print link down message
2892  * @adapter: pointer to the adapter structure
2893  **/
2894 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2895 {
2896 	struct net_device *netdev = adapter->netdev;
2897 
2898 	adapter->link_speed = 0;
2899 
2900 	/* only continue if link was up previously */
2901 	if (!netif_carrier_ok(netdev))
2902 		return;
2903 
2904 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2905 
2906 	netif_carrier_off(netdev);
2907 }
2908 
2909 /**
2910  * ixgbevf_watchdog_subtask - worker thread to bring link up
2911  * @work: pointer to work_struct containing our data
2912  **/
2913 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2914 {
2915 	/* if interface is down do nothing */
2916 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2917 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2918 		return;
2919 
2920 	ixgbevf_watchdog_update_link(adapter);
2921 
2922 	if (adapter->link_up)
2923 		ixgbevf_watchdog_link_is_up(adapter);
2924 	else
2925 		ixgbevf_watchdog_link_is_down(adapter);
2926 
2927 	ixgbevf_update_stats(adapter);
2928 }
2929 
2930 /**
2931  * ixgbevf_service_task - manages and runs subtasks
2932  * @work: pointer to work_struct containing our data
2933  **/
2934 static void ixgbevf_service_task(struct work_struct *work)
2935 {
2936 	struct ixgbevf_adapter *adapter = container_of(work,
2937 						       struct ixgbevf_adapter,
2938 						       service_task);
2939 	struct ixgbe_hw *hw = &adapter->hw;
2940 
2941 	if (IXGBE_REMOVED(hw->hw_addr)) {
2942 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2943 			rtnl_lock();
2944 			ixgbevf_down(adapter);
2945 			rtnl_unlock();
2946 		}
2947 		return;
2948 	}
2949 
2950 	ixgbevf_queue_reset_subtask(adapter);
2951 	ixgbevf_reset_subtask(adapter);
2952 	ixgbevf_watchdog_subtask(adapter);
2953 	ixgbevf_check_hang_subtask(adapter);
2954 
2955 	ixgbevf_service_event_complete(adapter);
2956 }
2957 
2958 /**
2959  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2960  * @tx_ring: Tx descriptor ring for a specific queue
2961  *
2962  * Free all transmit software resources
2963  **/
2964 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2965 {
2966 	ixgbevf_clean_tx_ring(tx_ring);
2967 
2968 	vfree(tx_ring->tx_buffer_info);
2969 	tx_ring->tx_buffer_info = NULL;
2970 
2971 	/* if not set, then don't free */
2972 	if (!tx_ring->desc)
2973 		return;
2974 
2975 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2976 			  tx_ring->dma);
2977 
2978 	tx_ring->desc = NULL;
2979 }
2980 
2981 /**
2982  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2983  * @adapter: board private structure
2984  *
2985  * Free all transmit software resources
2986  **/
2987 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2988 {
2989 	int i;
2990 
2991 	for (i = 0; i < adapter->num_tx_queues; i++)
2992 		if (adapter->tx_ring[i]->desc)
2993 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2994 }
2995 
2996 /**
2997  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2998  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2999  *
3000  * Return 0 on success, negative on failure
3001  **/
3002 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3003 {
3004 	struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3005 	int size;
3006 
3007 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3008 	tx_ring->tx_buffer_info = vzalloc(size);
3009 	if (!tx_ring->tx_buffer_info)
3010 		goto err;
3011 
3012 	/* round up to nearest 4K */
3013 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3014 	tx_ring->size = ALIGN(tx_ring->size, 4096);
3015 
3016 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3017 					   &tx_ring->dma, GFP_KERNEL);
3018 	if (!tx_ring->desc)
3019 		goto err;
3020 
3021 	return 0;
3022 
3023 err:
3024 	vfree(tx_ring->tx_buffer_info);
3025 	tx_ring->tx_buffer_info = NULL;
3026 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3027 	return -ENOMEM;
3028 }
3029 
3030 /**
3031  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3032  * @adapter: board private structure
3033  *
3034  * If this function returns with an error, then it's possible one or
3035  * more of the rings is populated (while the rest are not).  It is the
3036  * callers duty to clean those orphaned rings.
3037  *
3038  * Return 0 on success, negative on failure
3039  **/
3040 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3041 {
3042 	int i, err = 0;
3043 
3044 	for (i = 0; i < adapter->num_tx_queues; i++) {
3045 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3046 		if (!err)
3047 			continue;
3048 		hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3049 		break;
3050 	}
3051 
3052 	return err;
3053 }
3054 
3055 /**
3056  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3057  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3058  *
3059  * Returns 0 on success, negative on failure
3060  **/
3061 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3062 {
3063 	int size;
3064 
3065 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3066 	rx_ring->rx_buffer_info = vzalloc(size);
3067 	if (!rx_ring->rx_buffer_info)
3068 		goto err;
3069 
3070 	/* Round up to nearest 4K */
3071 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3072 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3073 
3074 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3075 					   &rx_ring->dma, GFP_KERNEL);
3076 
3077 	if (!rx_ring->desc)
3078 		goto err;
3079 
3080 	return 0;
3081 err:
3082 	vfree(rx_ring->rx_buffer_info);
3083 	rx_ring->rx_buffer_info = NULL;
3084 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3085 	return -ENOMEM;
3086 }
3087 
3088 /**
3089  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3090  * @adapter: board private structure
3091  *
3092  * If this function returns with an error, then it's possible one or
3093  * more of the rings is populated (while the rest are not).  It is the
3094  * callers duty to clean those orphaned rings.
3095  *
3096  * Return 0 on success, negative on failure
3097  **/
3098 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3099 {
3100 	int i, err = 0;
3101 
3102 	for (i = 0; i < adapter->num_rx_queues; i++) {
3103 		err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3104 		if (!err)
3105 			continue;
3106 		hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3107 		break;
3108 	}
3109 	return err;
3110 }
3111 
3112 /**
3113  * ixgbevf_free_rx_resources - Free Rx Resources
3114  * @rx_ring: ring to clean the resources from
3115  *
3116  * Free all receive software resources
3117  **/
3118 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3119 {
3120 	ixgbevf_clean_rx_ring(rx_ring);
3121 
3122 	vfree(rx_ring->rx_buffer_info);
3123 	rx_ring->rx_buffer_info = NULL;
3124 
3125 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3126 			  rx_ring->dma);
3127 
3128 	rx_ring->desc = NULL;
3129 }
3130 
3131 /**
3132  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3133  * @adapter: board private structure
3134  *
3135  * Free all receive software resources
3136  **/
3137 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3138 {
3139 	int i;
3140 
3141 	for (i = 0; i < adapter->num_rx_queues; i++)
3142 		if (adapter->rx_ring[i]->desc)
3143 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3144 }
3145 
3146 /**
3147  * ixgbevf_open - Called when a network interface is made active
3148  * @netdev: network interface device structure
3149  *
3150  * Returns 0 on success, negative value on failure
3151  *
3152  * The open entry point is called when a network interface is made
3153  * active by the system (IFF_UP).  At this point all resources needed
3154  * for transmit and receive operations are allocated, the interrupt
3155  * handler is registered with the OS, the watchdog timer is started,
3156  * and the stack is notified that the interface is ready.
3157  **/
3158 int ixgbevf_open(struct net_device *netdev)
3159 {
3160 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3161 	struct ixgbe_hw *hw = &adapter->hw;
3162 	int err;
3163 
3164 	/* A previous failure to open the device because of a lack of
3165 	 * available MSIX vector resources may have reset the number
3166 	 * of msix vectors variable to zero.  The only way to recover
3167 	 * is to unload/reload the driver and hope that the system has
3168 	 * been able to recover some MSIX vector resources.
3169 	 */
3170 	if (!adapter->num_msix_vectors)
3171 		return -ENOMEM;
3172 
3173 	if (hw->adapter_stopped) {
3174 		ixgbevf_reset(adapter);
3175 		/* if adapter is still stopped then PF isn't up and
3176 		 * the VF can't start.
3177 		 */
3178 		if (hw->adapter_stopped) {
3179 			err = IXGBE_ERR_MBX;
3180 			pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3181 			goto err_setup_reset;
3182 		}
3183 	}
3184 
3185 	/* disallow open during test */
3186 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3187 		return -EBUSY;
3188 
3189 	netif_carrier_off(netdev);
3190 
3191 	/* allocate transmit descriptors */
3192 	err = ixgbevf_setup_all_tx_resources(adapter);
3193 	if (err)
3194 		goto err_setup_tx;
3195 
3196 	/* allocate receive descriptors */
3197 	err = ixgbevf_setup_all_rx_resources(adapter);
3198 	if (err)
3199 		goto err_setup_rx;
3200 
3201 	ixgbevf_configure(adapter);
3202 
3203 	/* Map the Tx/Rx rings to the vectors we were allotted.
3204 	 * if request_irq will be called in this function map_rings
3205 	 * must be called *before* up_complete
3206 	 */
3207 	ixgbevf_map_rings_to_vectors(adapter);
3208 
3209 	err = ixgbevf_request_irq(adapter);
3210 	if (err)
3211 		goto err_req_irq;
3212 
3213 	ixgbevf_up_complete(adapter);
3214 
3215 	return 0;
3216 
3217 err_req_irq:
3218 	ixgbevf_down(adapter);
3219 err_setup_rx:
3220 	ixgbevf_free_all_rx_resources(adapter);
3221 err_setup_tx:
3222 	ixgbevf_free_all_tx_resources(adapter);
3223 	ixgbevf_reset(adapter);
3224 
3225 err_setup_reset:
3226 
3227 	return err;
3228 }
3229 
3230 /**
3231  * ixgbevf_close - Disables a network interface
3232  * @netdev: network interface device structure
3233  *
3234  * Returns 0, this is not allowed to fail
3235  *
3236  * The close entry point is called when an interface is de-activated
3237  * by the OS.  The hardware is still under the drivers control, but
3238  * needs to be disabled.  A global MAC reset is issued to stop the
3239  * hardware, and all transmit and receive resources are freed.
3240  **/
3241 int ixgbevf_close(struct net_device *netdev)
3242 {
3243 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3244 
3245 	ixgbevf_down(adapter);
3246 	ixgbevf_free_irq(adapter);
3247 
3248 	ixgbevf_free_all_tx_resources(adapter);
3249 	ixgbevf_free_all_rx_resources(adapter);
3250 
3251 	return 0;
3252 }
3253 
3254 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3255 {
3256 	struct net_device *dev = adapter->netdev;
3257 
3258 	if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3259 				&adapter->state))
3260 		return;
3261 
3262 	/* if interface is down do nothing */
3263 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3264 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3265 		return;
3266 
3267 	/* Hardware has to reinitialize queues and interrupts to
3268 	 * match packet buffer alignment. Unfortunately, the
3269 	 * hardware is not flexible enough to do this dynamically.
3270 	 */
3271 	if (netif_running(dev))
3272 		ixgbevf_close(dev);
3273 
3274 	ixgbevf_clear_interrupt_scheme(adapter);
3275 	ixgbevf_init_interrupt_scheme(adapter);
3276 
3277 	if (netif_running(dev))
3278 		ixgbevf_open(dev);
3279 }
3280 
3281 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3282 				u32 vlan_macip_lens, u32 type_tucmd,
3283 				u32 mss_l4len_idx)
3284 {
3285 	struct ixgbe_adv_tx_context_desc *context_desc;
3286 	u16 i = tx_ring->next_to_use;
3287 
3288 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3289 
3290 	i++;
3291 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3292 
3293 	/* set bits to identify this as an advanced context descriptor */
3294 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3295 
3296 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3297 	context_desc->seqnum_seed	= 0;
3298 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3299 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3300 }
3301 
3302 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3303 		       struct ixgbevf_tx_buffer *first,
3304 		       u8 *hdr_len)
3305 {
3306 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3307 	struct sk_buff *skb = first->skb;
3308 	union {
3309 		struct iphdr *v4;
3310 		struct ipv6hdr *v6;
3311 		unsigned char *hdr;
3312 	} ip;
3313 	union {
3314 		struct tcphdr *tcp;
3315 		unsigned char *hdr;
3316 	} l4;
3317 	u32 paylen, l4_offset;
3318 	int err;
3319 
3320 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3321 		return 0;
3322 
3323 	if (!skb_is_gso(skb))
3324 		return 0;
3325 
3326 	err = skb_cow_head(skb, 0);
3327 	if (err < 0)
3328 		return err;
3329 
3330 	ip.hdr = skb_network_header(skb);
3331 	l4.hdr = skb_checksum_start(skb);
3332 
3333 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3334 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3335 
3336 	/* initialize outer IP header fields */
3337 	if (ip.v4->version == 4) {
3338 		unsigned char *csum_start = skb_checksum_start(skb);
3339 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3340 
3341 		/* IP header will have to cancel out any data that
3342 		 * is not a part of the outer IP header
3343 		 */
3344 		ip.v4->check = csum_fold(csum_partial(trans_start,
3345 						      csum_start - trans_start,
3346 						      0));
3347 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3348 
3349 		ip.v4->tot_len = 0;
3350 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3351 				   IXGBE_TX_FLAGS_CSUM |
3352 				   IXGBE_TX_FLAGS_IPV4;
3353 	} else {
3354 		ip.v6->payload_len = 0;
3355 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3356 				   IXGBE_TX_FLAGS_CSUM;
3357 	}
3358 
3359 	/* determine offset of inner transport header */
3360 	l4_offset = l4.hdr - skb->data;
3361 
3362 	/* compute length of segmentation header */
3363 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
3364 
3365 	/* remove payload length from inner checksum */
3366 	paylen = skb->len - l4_offset;
3367 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3368 
3369 	/* update gso size and bytecount with header size */
3370 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3371 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3372 
3373 	/* mss_l4len_id: use 1 as index for TSO */
3374 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3375 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3376 	mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3377 
3378 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3379 	vlan_macip_lens = l4.hdr - ip.hdr;
3380 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3381 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3382 
3383 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3384 			    type_tucmd, mss_l4len_idx);
3385 
3386 	return 1;
3387 }
3388 
3389 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3390 {
3391 	unsigned int offset = 0;
3392 
3393 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3394 
3395 	return offset == skb_checksum_start_offset(skb);
3396 }
3397 
3398 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3399 			    struct ixgbevf_tx_buffer *first)
3400 {
3401 	struct sk_buff *skb = first->skb;
3402 	u32 vlan_macip_lens = 0;
3403 	u32 type_tucmd = 0;
3404 
3405 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3406 		goto no_csum;
3407 
3408 	switch (skb->csum_offset) {
3409 	case offsetof(struct tcphdr, check):
3410 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3411 		/* fall through */
3412 	case offsetof(struct udphdr, check):
3413 		break;
3414 	case offsetof(struct sctphdr, checksum):
3415 		/* validate that this is actually an SCTP request */
3416 		if (((first->protocol == htons(ETH_P_IP)) &&
3417 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3418 		    ((first->protocol == htons(ETH_P_IPV6)) &&
3419 		     ixgbevf_ipv6_csum_is_sctp(skb))) {
3420 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3421 			break;
3422 		}
3423 		/* fall through */
3424 	default:
3425 		skb_checksum_help(skb);
3426 		goto no_csum;
3427 	}
3428 	/* update TX checksum flag */
3429 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3430 	vlan_macip_lens = skb_checksum_start_offset(skb) -
3431 			  skb_network_offset(skb);
3432 no_csum:
3433 	/* vlan_macip_lens: MACLEN, VLAN tag */
3434 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3435 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3436 
3437 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3438 }
3439 
3440 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3441 {
3442 	/* set type for advanced descriptor with frame checksum insertion */
3443 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3444 				      IXGBE_ADVTXD_DCMD_IFCS |
3445 				      IXGBE_ADVTXD_DCMD_DEXT);
3446 
3447 	/* set HW VLAN bit if VLAN is present */
3448 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3449 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3450 
3451 	/* set segmentation enable bits for TSO/FSO */
3452 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3453 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3454 
3455 	return cmd_type;
3456 }
3457 
3458 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3459 				     u32 tx_flags, unsigned int paylen)
3460 {
3461 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3462 
3463 	/* enable L4 checksum for TSO and TX checksum offload */
3464 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3465 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3466 
3467 	/* enble IPv4 checksum for TSO */
3468 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3469 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3470 
3471 	/* use index 1 context for TSO/FSO/FCOE */
3472 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3473 		olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3474 
3475 	/* Check Context must be set if Tx switch is enabled, which it
3476 	 * always is for case where virtual functions are running
3477 	 */
3478 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3479 
3480 	tx_desc->read.olinfo_status = olinfo_status;
3481 }
3482 
3483 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3484 			   struct ixgbevf_tx_buffer *first,
3485 			   const u8 hdr_len)
3486 {
3487 	dma_addr_t dma;
3488 	struct sk_buff *skb = first->skb;
3489 	struct ixgbevf_tx_buffer *tx_buffer;
3490 	union ixgbe_adv_tx_desc *tx_desc;
3491 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3492 	unsigned int data_len = skb->data_len;
3493 	unsigned int size = skb_headlen(skb);
3494 	unsigned int paylen = skb->len - hdr_len;
3495 	u32 tx_flags = first->tx_flags;
3496 	__le32 cmd_type;
3497 	u16 i = tx_ring->next_to_use;
3498 
3499 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3500 
3501 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3502 	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3503 
3504 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3505 	if (dma_mapping_error(tx_ring->dev, dma))
3506 		goto dma_error;
3507 
3508 	/* record length, and DMA address */
3509 	dma_unmap_len_set(first, len, size);
3510 	dma_unmap_addr_set(first, dma, dma);
3511 
3512 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
3513 
3514 	for (;;) {
3515 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3516 			tx_desc->read.cmd_type_len =
3517 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3518 
3519 			i++;
3520 			tx_desc++;
3521 			if (i == tx_ring->count) {
3522 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3523 				i = 0;
3524 			}
3525 
3526 			dma += IXGBE_MAX_DATA_PER_TXD;
3527 			size -= IXGBE_MAX_DATA_PER_TXD;
3528 
3529 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3530 			tx_desc->read.olinfo_status = 0;
3531 		}
3532 
3533 		if (likely(!data_len))
3534 			break;
3535 
3536 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3537 
3538 		i++;
3539 		tx_desc++;
3540 		if (i == tx_ring->count) {
3541 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3542 			i = 0;
3543 		}
3544 
3545 		size = skb_frag_size(frag);
3546 		data_len -= size;
3547 
3548 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3549 				       DMA_TO_DEVICE);
3550 		if (dma_mapping_error(tx_ring->dev, dma))
3551 			goto dma_error;
3552 
3553 		tx_buffer = &tx_ring->tx_buffer_info[i];
3554 		dma_unmap_len_set(tx_buffer, len, size);
3555 		dma_unmap_addr_set(tx_buffer, dma, dma);
3556 
3557 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3558 		tx_desc->read.olinfo_status = 0;
3559 
3560 		frag++;
3561 	}
3562 
3563 	/* write last descriptor with RS and EOP bits */
3564 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3565 	tx_desc->read.cmd_type_len = cmd_type;
3566 
3567 	/* set the timestamp */
3568 	first->time_stamp = jiffies;
3569 
3570 	/* Force memory writes to complete before letting h/w know there
3571 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
3572 	 * memory model archs, such as IA-64).
3573 	 *
3574 	 * We also need this memory barrier (wmb) to make certain all of the
3575 	 * status bits have been updated before next_to_watch is written.
3576 	 */
3577 	wmb();
3578 
3579 	/* set next_to_watch value indicating a packet is present */
3580 	first->next_to_watch = tx_desc;
3581 
3582 	i++;
3583 	if (i == tx_ring->count)
3584 		i = 0;
3585 
3586 	tx_ring->next_to_use = i;
3587 
3588 	/* notify HW of packet */
3589 	ixgbevf_write_tail(tx_ring, i);
3590 
3591 	return;
3592 dma_error:
3593 	dev_err(tx_ring->dev, "TX DMA map failed\n");
3594 
3595 	/* clear dma mappings for failed tx_buffer_info map */
3596 	for (;;) {
3597 		tx_buffer = &tx_ring->tx_buffer_info[i];
3598 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3599 		if (tx_buffer == first)
3600 			break;
3601 		if (i == 0)
3602 			i = tx_ring->count;
3603 		i--;
3604 	}
3605 
3606 	tx_ring->next_to_use = i;
3607 }
3608 
3609 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3610 {
3611 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3612 	/* Herbert's original patch had:
3613 	 *  smp_mb__after_netif_stop_queue();
3614 	 * but since that doesn't exist yet, just open code it.
3615 	 */
3616 	smp_mb();
3617 
3618 	/* We need to check again in a case another CPU has just
3619 	 * made room available.
3620 	 */
3621 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
3622 		return -EBUSY;
3623 
3624 	/* A reprieve! - use start_queue because it doesn't call schedule */
3625 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3626 	++tx_ring->tx_stats.restart_queue;
3627 
3628 	return 0;
3629 }
3630 
3631 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3632 {
3633 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3634 		return 0;
3635 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
3636 }
3637 
3638 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3639 {
3640 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3641 	struct ixgbevf_tx_buffer *first;
3642 	struct ixgbevf_ring *tx_ring;
3643 	int tso;
3644 	u32 tx_flags = 0;
3645 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
3646 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3647 	unsigned short f;
3648 #endif
3649 	u8 hdr_len = 0;
3650 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3651 
3652 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3653 		dev_kfree_skb_any(skb);
3654 		return NETDEV_TX_OK;
3655 	}
3656 
3657 	tx_ring = adapter->tx_ring[skb->queue_mapping];
3658 
3659 	/* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3660 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3661 	 *       + 2 desc gap to keep tail from touching head,
3662 	 *       + 1 desc for context descriptor,
3663 	 * otherwise try next time
3664 	 */
3665 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3666 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3667 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3668 #else
3669 	count += skb_shinfo(skb)->nr_frags;
3670 #endif
3671 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3672 		tx_ring->tx_stats.tx_busy++;
3673 		return NETDEV_TX_BUSY;
3674 	}
3675 
3676 	/* record the location of the first descriptor for this packet */
3677 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3678 	first->skb = skb;
3679 	first->bytecount = skb->len;
3680 	first->gso_segs = 1;
3681 
3682 	if (skb_vlan_tag_present(skb)) {
3683 		tx_flags |= skb_vlan_tag_get(skb);
3684 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3685 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3686 	}
3687 
3688 	/* record initial flags and protocol */
3689 	first->tx_flags = tx_flags;
3690 	first->protocol = vlan_get_protocol(skb);
3691 
3692 	tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3693 	if (tso < 0)
3694 		goto out_drop;
3695 	else if (!tso)
3696 		ixgbevf_tx_csum(tx_ring, first);
3697 
3698 	ixgbevf_tx_map(tx_ring, first, hdr_len);
3699 
3700 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3701 
3702 	return NETDEV_TX_OK;
3703 
3704 out_drop:
3705 	dev_kfree_skb_any(first->skb);
3706 	first->skb = NULL;
3707 
3708 	return NETDEV_TX_OK;
3709 }
3710 
3711 /**
3712  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3713  * @netdev: network interface device structure
3714  * @p: pointer to an address structure
3715  *
3716  * Returns 0 on success, negative on failure
3717  **/
3718 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3719 {
3720 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3721 	struct ixgbe_hw *hw = &adapter->hw;
3722 	struct sockaddr *addr = p;
3723 	int err;
3724 
3725 	if (!is_valid_ether_addr(addr->sa_data))
3726 		return -EADDRNOTAVAIL;
3727 
3728 	spin_lock_bh(&adapter->mbx_lock);
3729 
3730 	err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
3731 
3732 	spin_unlock_bh(&adapter->mbx_lock);
3733 
3734 	if (err)
3735 		return -EPERM;
3736 
3737 	ether_addr_copy(hw->mac.addr, addr->sa_data);
3738 	ether_addr_copy(netdev->dev_addr, addr->sa_data);
3739 
3740 	return 0;
3741 }
3742 
3743 /**
3744  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3745  * @netdev: network interface device structure
3746  * @new_mtu: new value for maximum frame size
3747  *
3748  * Returns 0 on success, negative on failure
3749  **/
3750 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3751 {
3752 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3753 	struct ixgbe_hw *hw = &adapter->hw;
3754 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3755 	int ret;
3756 
3757 	spin_lock_bh(&adapter->mbx_lock);
3758 	/* notify the PF of our intent to use this size of frame */
3759 	ret = hw->mac.ops.set_rlpml(hw, max_frame);
3760 	spin_unlock_bh(&adapter->mbx_lock);
3761 	if (ret)
3762 		return -EINVAL;
3763 
3764 	hw_dbg(hw, "changing MTU from %d to %d\n",
3765 	       netdev->mtu, new_mtu);
3766 
3767 	/* must set new MTU before calling down or up */
3768 	netdev->mtu = new_mtu;
3769 
3770 	return 0;
3771 }
3772 
3773 #ifdef CONFIG_NET_POLL_CONTROLLER
3774 /* Polling 'interrupt' - used by things like netconsole to send skbs
3775  * without having to re-enable interrupts. It's not called while
3776  * the interrupt routine is executing.
3777  */
3778 static void ixgbevf_netpoll(struct net_device *netdev)
3779 {
3780 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3781 	int i;
3782 
3783 	/* if interface is down do nothing */
3784 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3785 		return;
3786 	for (i = 0; i < adapter->num_rx_queues; i++)
3787 		ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3788 }
3789 #endif /* CONFIG_NET_POLL_CONTROLLER */
3790 
3791 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3792 {
3793 	struct net_device *netdev = pci_get_drvdata(pdev);
3794 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3795 #ifdef CONFIG_PM
3796 	int retval = 0;
3797 #endif
3798 
3799 	netif_device_detach(netdev);
3800 
3801 	if (netif_running(netdev)) {
3802 		rtnl_lock();
3803 		ixgbevf_down(adapter);
3804 		ixgbevf_free_irq(adapter);
3805 		ixgbevf_free_all_tx_resources(adapter);
3806 		ixgbevf_free_all_rx_resources(adapter);
3807 		ixgbevf_clear_interrupt_scheme(adapter);
3808 		rtnl_unlock();
3809 	}
3810 
3811 #ifdef CONFIG_PM
3812 	retval = pci_save_state(pdev);
3813 	if (retval)
3814 		return retval;
3815 
3816 #endif
3817 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3818 		pci_disable_device(pdev);
3819 
3820 	return 0;
3821 }
3822 
3823 #ifdef CONFIG_PM
3824 static int ixgbevf_resume(struct pci_dev *pdev)
3825 {
3826 	struct net_device *netdev = pci_get_drvdata(pdev);
3827 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3828 	u32 err;
3829 
3830 	pci_restore_state(pdev);
3831 	/* pci_restore_state clears dev->state_saved so call
3832 	 * pci_save_state to restore it.
3833 	 */
3834 	pci_save_state(pdev);
3835 
3836 	err = pci_enable_device_mem(pdev);
3837 	if (err) {
3838 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3839 		return err;
3840 	}
3841 	smp_mb__before_atomic();
3842 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3843 	pci_set_master(pdev);
3844 
3845 	ixgbevf_reset(adapter);
3846 
3847 	rtnl_lock();
3848 	err = ixgbevf_init_interrupt_scheme(adapter);
3849 	rtnl_unlock();
3850 	if (err) {
3851 		dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3852 		return err;
3853 	}
3854 
3855 	if (netif_running(netdev)) {
3856 		err = ixgbevf_open(netdev);
3857 		if (err)
3858 			return err;
3859 	}
3860 
3861 	netif_device_attach(netdev);
3862 
3863 	return err;
3864 }
3865 
3866 #endif /* CONFIG_PM */
3867 static void ixgbevf_shutdown(struct pci_dev *pdev)
3868 {
3869 	ixgbevf_suspend(pdev, PMSG_SUSPEND);
3870 }
3871 
3872 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3873 						struct rtnl_link_stats64 *stats)
3874 {
3875 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3876 	unsigned int start;
3877 	u64 bytes, packets;
3878 	const struct ixgbevf_ring *ring;
3879 	int i;
3880 
3881 	ixgbevf_update_stats(adapter);
3882 
3883 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3884 
3885 	for (i = 0; i < adapter->num_rx_queues; i++) {
3886 		ring = adapter->rx_ring[i];
3887 		do {
3888 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3889 			bytes = ring->stats.bytes;
3890 			packets = ring->stats.packets;
3891 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3892 		stats->rx_bytes += bytes;
3893 		stats->rx_packets += packets;
3894 	}
3895 
3896 	for (i = 0; i < adapter->num_tx_queues; i++) {
3897 		ring = adapter->tx_ring[i];
3898 		do {
3899 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3900 			bytes = ring->stats.bytes;
3901 			packets = ring->stats.packets;
3902 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3903 		stats->tx_bytes += bytes;
3904 		stats->tx_packets += packets;
3905 	}
3906 
3907 	return stats;
3908 }
3909 
3910 #define IXGBEVF_MAX_MAC_HDR_LEN		127
3911 #define IXGBEVF_MAX_NETWORK_HDR_LEN	511
3912 
3913 static netdev_features_t
3914 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
3915 		       netdev_features_t features)
3916 {
3917 	unsigned int network_hdr_len, mac_hdr_len;
3918 
3919 	/* Make certain the headers can be described by a context descriptor */
3920 	mac_hdr_len = skb_network_header(skb) - skb->data;
3921 	if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
3922 		return features & ~(NETIF_F_HW_CSUM |
3923 				    NETIF_F_SCTP_CRC |
3924 				    NETIF_F_HW_VLAN_CTAG_TX |
3925 				    NETIF_F_TSO |
3926 				    NETIF_F_TSO6);
3927 
3928 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3929 	if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
3930 		return features & ~(NETIF_F_HW_CSUM |
3931 				    NETIF_F_SCTP_CRC |
3932 				    NETIF_F_TSO |
3933 				    NETIF_F_TSO6);
3934 
3935 	/* We can only support IPV4 TSO in tunnels if we can mangle the
3936 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
3937 	 */
3938 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3939 		features &= ~NETIF_F_TSO;
3940 
3941 	return features;
3942 }
3943 
3944 static const struct net_device_ops ixgbevf_netdev_ops = {
3945 	.ndo_open		= ixgbevf_open,
3946 	.ndo_stop		= ixgbevf_close,
3947 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3948 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3949 	.ndo_get_stats64	= ixgbevf_get_stats,
3950 	.ndo_validate_addr	= eth_validate_addr,
3951 	.ndo_set_mac_address	= ixgbevf_set_mac,
3952 	.ndo_change_mtu		= ixgbevf_change_mtu,
3953 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3954 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3955 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3956 #ifdef CONFIG_NET_RX_BUSY_POLL
3957 	.ndo_busy_poll		= ixgbevf_busy_poll_recv,
3958 #endif
3959 #ifdef CONFIG_NET_POLL_CONTROLLER
3960 	.ndo_poll_controller	= ixgbevf_netpoll,
3961 #endif
3962 	.ndo_features_check	= ixgbevf_features_check,
3963 };
3964 
3965 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3966 {
3967 	dev->netdev_ops = &ixgbevf_netdev_ops;
3968 	ixgbevf_set_ethtool_ops(dev);
3969 	dev->watchdog_timeo = 5 * HZ;
3970 }
3971 
3972 /**
3973  * ixgbevf_probe - Device Initialization Routine
3974  * @pdev: PCI device information struct
3975  * @ent: entry in ixgbevf_pci_tbl
3976  *
3977  * Returns 0 on success, negative on failure
3978  *
3979  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3980  * The OS initialization, configuring of the adapter private structure,
3981  * and a hardware reset occur.
3982  **/
3983 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3984 {
3985 	struct net_device *netdev;
3986 	struct ixgbevf_adapter *adapter = NULL;
3987 	struct ixgbe_hw *hw = NULL;
3988 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3989 	int err, pci_using_dac;
3990 	bool disable_dev = false;
3991 
3992 	err = pci_enable_device(pdev);
3993 	if (err)
3994 		return err;
3995 
3996 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3997 		pci_using_dac = 1;
3998 	} else {
3999 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4000 		if (err) {
4001 			dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4002 			goto err_dma;
4003 		}
4004 		pci_using_dac = 0;
4005 	}
4006 
4007 	err = pci_request_regions(pdev, ixgbevf_driver_name);
4008 	if (err) {
4009 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4010 		goto err_pci_reg;
4011 	}
4012 
4013 	pci_set_master(pdev);
4014 
4015 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4016 				   MAX_TX_QUEUES);
4017 	if (!netdev) {
4018 		err = -ENOMEM;
4019 		goto err_alloc_etherdev;
4020 	}
4021 
4022 	SET_NETDEV_DEV(netdev, &pdev->dev);
4023 
4024 	adapter = netdev_priv(netdev);
4025 
4026 	adapter->netdev = netdev;
4027 	adapter->pdev = pdev;
4028 	hw = &adapter->hw;
4029 	hw->back = adapter;
4030 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4031 
4032 	/* call save state here in standalone driver because it relies on
4033 	 * adapter struct to exist, and needs to call netdev_priv
4034 	 */
4035 	pci_save_state(pdev);
4036 
4037 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4038 			      pci_resource_len(pdev, 0));
4039 	adapter->io_addr = hw->hw_addr;
4040 	if (!hw->hw_addr) {
4041 		err = -EIO;
4042 		goto err_ioremap;
4043 	}
4044 
4045 	ixgbevf_assign_netdev_ops(netdev);
4046 
4047 	/* Setup HW API */
4048 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4049 	hw->mac.type  = ii->mac;
4050 
4051 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4052 	       sizeof(struct ixgbe_mbx_operations));
4053 
4054 	/* setup the private structure */
4055 	err = ixgbevf_sw_init(adapter);
4056 	if (err)
4057 		goto err_sw_init;
4058 
4059 	/* The HW MAC address was set and/or determined in sw_init */
4060 	if (!is_valid_ether_addr(netdev->dev_addr)) {
4061 		pr_err("invalid MAC address\n");
4062 		err = -EIO;
4063 		goto err_sw_init;
4064 	}
4065 
4066 	netdev->hw_features = NETIF_F_SG |
4067 			      NETIF_F_TSO |
4068 			      NETIF_F_TSO6 |
4069 			      NETIF_F_RXCSUM |
4070 			      NETIF_F_HW_CSUM |
4071 			      NETIF_F_SCTP_CRC;
4072 
4073 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4074 				      NETIF_F_GSO_GRE_CSUM | \
4075 				      NETIF_F_GSO_IPXIP4 | \
4076 				      NETIF_F_GSO_IPXIP6 | \
4077 				      NETIF_F_GSO_UDP_TUNNEL | \
4078 				      NETIF_F_GSO_UDP_TUNNEL_CSUM)
4079 
4080 	netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4081 	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4082 			       IXGBEVF_GSO_PARTIAL_FEATURES;
4083 
4084 	netdev->features = netdev->hw_features;
4085 
4086 	if (pci_using_dac)
4087 		netdev->features |= NETIF_F_HIGHDMA;
4088 
4089 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4090 	netdev->mpls_features |= NETIF_F_HW_CSUM;
4091 	netdev->hw_enc_features |= netdev->vlan_features;
4092 
4093 	/* set this bit last since it cannot be part of vlan_features */
4094 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4095 			    NETIF_F_HW_VLAN_CTAG_RX |
4096 			    NETIF_F_HW_VLAN_CTAG_TX;
4097 
4098 	netdev->priv_flags |= IFF_UNICAST_FLT;
4099 
4100 	/* MTU range: 68 - 1504 or 9710 */
4101 	netdev->min_mtu = ETH_MIN_MTU;
4102 	switch (adapter->hw.api_version) {
4103 	case ixgbe_mbox_api_11:
4104 	case ixgbe_mbox_api_12:
4105 		netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4106 				  (ETH_HLEN + ETH_FCS_LEN);
4107 		break;
4108 	default:
4109 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4110 			netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4111 					  (ETH_HLEN + ETH_FCS_LEN);
4112 		else
4113 			netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4114 		break;
4115 	}
4116 
4117 	if (IXGBE_REMOVED(hw->hw_addr)) {
4118 		err = -EIO;
4119 		goto err_sw_init;
4120 	}
4121 
4122 	setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4123 		    (unsigned long)adapter);
4124 
4125 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4126 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4127 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4128 
4129 	err = ixgbevf_init_interrupt_scheme(adapter);
4130 	if (err)
4131 		goto err_sw_init;
4132 
4133 	strcpy(netdev->name, "eth%d");
4134 
4135 	err = register_netdev(netdev);
4136 	if (err)
4137 		goto err_register;
4138 
4139 	pci_set_drvdata(pdev, netdev);
4140 	netif_carrier_off(netdev);
4141 
4142 	ixgbevf_init_last_counter_stats(adapter);
4143 
4144 	/* print the VF info */
4145 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4146 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4147 
4148 	switch (hw->mac.type) {
4149 	case ixgbe_mac_X550_vf:
4150 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4151 		break;
4152 	case ixgbe_mac_X540_vf:
4153 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4154 		break;
4155 	case ixgbe_mac_82599_vf:
4156 	default:
4157 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4158 		break;
4159 	}
4160 
4161 	return 0;
4162 
4163 err_register:
4164 	ixgbevf_clear_interrupt_scheme(adapter);
4165 err_sw_init:
4166 	ixgbevf_reset_interrupt_capability(adapter);
4167 	iounmap(adapter->io_addr);
4168 err_ioremap:
4169 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4170 	free_netdev(netdev);
4171 err_alloc_etherdev:
4172 	pci_release_regions(pdev);
4173 err_pci_reg:
4174 err_dma:
4175 	if (!adapter || disable_dev)
4176 		pci_disable_device(pdev);
4177 	return err;
4178 }
4179 
4180 /**
4181  * ixgbevf_remove - Device Removal Routine
4182  * @pdev: PCI device information struct
4183  *
4184  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4185  * that it should release a PCI device.  The could be caused by a
4186  * Hot-Plug event, or because the driver is going to be removed from
4187  * memory.
4188  **/
4189 static void ixgbevf_remove(struct pci_dev *pdev)
4190 {
4191 	struct net_device *netdev = pci_get_drvdata(pdev);
4192 	struct ixgbevf_adapter *adapter;
4193 	bool disable_dev;
4194 
4195 	if (!netdev)
4196 		return;
4197 
4198 	adapter = netdev_priv(netdev);
4199 
4200 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4201 	cancel_work_sync(&adapter->service_task);
4202 
4203 	if (netdev->reg_state == NETREG_REGISTERED)
4204 		unregister_netdev(netdev);
4205 
4206 	ixgbevf_clear_interrupt_scheme(adapter);
4207 	ixgbevf_reset_interrupt_capability(adapter);
4208 
4209 	iounmap(adapter->io_addr);
4210 	pci_release_regions(pdev);
4211 
4212 	hw_dbg(&adapter->hw, "Remove complete\n");
4213 
4214 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4215 	free_netdev(netdev);
4216 
4217 	if (disable_dev)
4218 		pci_disable_device(pdev);
4219 }
4220 
4221 /**
4222  * ixgbevf_io_error_detected - called when PCI error is detected
4223  * @pdev: Pointer to PCI device
4224  * @state: The current pci connection state
4225  *
4226  * This function is called after a PCI bus error affecting
4227  * this device has been detected.
4228  **/
4229 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4230 						  pci_channel_state_t state)
4231 {
4232 	struct net_device *netdev = pci_get_drvdata(pdev);
4233 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4234 
4235 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4236 		return PCI_ERS_RESULT_DISCONNECT;
4237 
4238 	rtnl_lock();
4239 	netif_device_detach(netdev);
4240 
4241 	if (state == pci_channel_io_perm_failure) {
4242 		rtnl_unlock();
4243 		return PCI_ERS_RESULT_DISCONNECT;
4244 	}
4245 
4246 	if (netif_running(netdev))
4247 		ixgbevf_down(adapter);
4248 
4249 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4250 		pci_disable_device(pdev);
4251 	rtnl_unlock();
4252 
4253 	/* Request a slot slot reset. */
4254 	return PCI_ERS_RESULT_NEED_RESET;
4255 }
4256 
4257 /**
4258  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4259  * @pdev: Pointer to PCI device
4260  *
4261  * Restart the card from scratch, as if from a cold-boot. Implementation
4262  * resembles the first-half of the ixgbevf_resume routine.
4263  **/
4264 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4265 {
4266 	struct net_device *netdev = pci_get_drvdata(pdev);
4267 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4268 
4269 	if (pci_enable_device_mem(pdev)) {
4270 		dev_err(&pdev->dev,
4271 			"Cannot re-enable PCI device after reset.\n");
4272 		return PCI_ERS_RESULT_DISCONNECT;
4273 	}
4274 
4275 	smp_mb__before_atomic();
4276 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4277 	pci_set_master(pdev);
4278 
4279 	ixgbevf_reset(adapter);
4280 
4281 	return PCI_ERS_RESULT_RECOVERED;
4282 }
4283 
4284 /**
4285  * ixgbevf_io_resume - called when traffic can start flowing again.
4286  * @pdev: Pointer to PCI device
4287  *
4288  * This callback is called when the error recovery driver tells us that
4289  * its OK to resume normal operation. Implementation resembles the
4290  * second-half of the ixgbevf_resume routine.
4291  **/
4292 static void ixgbevf_io_resume(struct pci_dev *pdev)
4293 {
4294 	struct net_device *netdev = pci_get_drvdata(pdev);
4295 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4296 
4297 	if (netif_running(netdev))
4298 		ixgbevf_up(adapter);
4299 
4300 	netif_device_attach(netdev);
4301 }
4302 
4303 /* PCI Error Recovery (ERS) */
4304 static const struct pci_error_handlers ixgbevf_err_handler = {
4305 	.error_detected = ixgbevf_io_error_detected,
4306 	.slot_reset = ixgbevf_io_slot_reset,
4307 	.resume = ixgbevf_io_resume,
4308 };
4309 
4310 static struct pci_driver ixgbevf_driver = {
4311 	.name		= ixgbevf_driver_name,
4312 	.id_table	= ixgbevf_pci_tbl,
4313 	.probe		= ixgbevf_probe,
4314 	.remove		= ixgbevf_remove,
4315 #ifdef CONFIG_PM
4316 	/* Power Management Hooks */
4317 	.suspend	= ixgbevf_suspend,
4318 	.resume		= ixgbevf_resume,
4319 #endif
4320 	.shutdown	= ixgbevf_shutdown,
4321 	.err_handler	= &ixgbevf_err_handler
4322 };
4323 
4324 /**
4325  * ixgbevf_init_module - Driver Registration Routine
4326  *
4327  * ixgbevf_init_module is the first routine called when the driver is
4328  * loaded. All it does is register with the PCI subsystem.
4329  **/
4330 static int __init ixgbevf_init_module(void)
4331 {
4332 	pr_info("%s - version %s\n", ixgbevf_driver_string,
4333 		ixgbevf_driver_version);
4334 
4335 	pr_info("%s\n", ixgbevf_copyright);
4336 	ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4337 	if (!ixgbevf_wq) {
4338 		pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4339 		return -ENOMEM;
4340 	}
4341 
4342 	return pci_register_driver(&ixgbevf_driver);
4343 }
4344 
4345 module_init(ixgbevf_init_module);
4346 
4347 /**
4348  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4349  *
4350  * ixgbevf_exit_module is called just before the driver is removed
4351  * from memory.
4352  **/
4353 static void __exit ixgbevf_exit_module(void)
4354 {
4355 	pci_unregister_driver(&ixgbevf_driver);
4356 	if (ixgbevf_wq) {
4357 		destroy_workqueue(ixgbevf_wq);
4358 		ixgbevf_wq = NULL;
4359 	}
4360 }
4361 
4362 #ifdef DEBUG
4363 /**
4364  * ixgbevf_get_hw_dev_name - return device name string
4365  * used by hardware layer to print debugging information
4366  **/
4367 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4368 {
4369 	struct ixgbevf_adapter *adapter = hw->back;
4370 
4371 	return adapter->netdev->name;
4372 }
4373 
4374 #endif
4375 module_exit(ixgbevf_exit_module);
4376 
4377 /* ixgbevf_main.c */
4378