xref: /linux/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c (revision a508da6cc0093171833efb8376b00473f24221b9)
1 /*******************************************************************************
2 
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21 
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 
26 *******************************************************************************/
27 
28 
29 /******************************************************************************
30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
32 
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/ipv6.h>
46 #include <linux/slab.h>
47 #include <net/checksum.h>
48 #include <net/ip6_checksum.h>
49 #include <linux/ethtool.h>
50 #include <linux/if.h>
51 #include <linux/if_vlan.h>
52 #include <linux/prefetch.h>
53 
54 #include "ixgbevf.h"
55 
56 const char ixgbevf_driver_name[] = "ixgbevf";
57 static const char ixgbevf_driver_string[] =
58 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
59 
60 #define DRV_VERSION "2.6.0-k"
61 const char ixgbevf_driver_version[] = DRV_VERSION;
62 static char ixgbevf_copyright[] =
63 	"Copyright (c) 2009 - 2012 Intel Corporation.";
64 
65 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
66 	[board_82599_vf] = &ixgbevf_82599_vf_info,
67 	[board_X540_vf]  = &ixgbevf_X540_vf_info,
68 };
69 
70 /* ixgbevf_pci_tbl - PCI Device ID Table
71  *
72  * Wildcard entries (PCI_ANY_ID) should come last
73  * Last entry must be all 0s
74  *
75  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
76  *   Class, Class Mask, private data (not used) }
77  */
78 static struct pci_device_id ixgbevf_pci_tbl[] = {
79 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
80 	board_82599_vf},
81 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
82 	board_X540_vf},
83 
84 	/* required last entry */
85 	{0, }
86 };
87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
88 
89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
90 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
91 MODULE_LICENSE("GPL");
92 MODULE_VERSION(DRV_VERSION);
93 
94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
95 static int debug = -1;
96 module_param(debug, int, 0);
97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
98 
99 /* forward decls */
100 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector);
101 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
102 			       u32 itr_reg);
103 
104 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
105 					   struct ixgbevf_ring *rx_ring,
106 					   u32 val)
107 {
108 	/*
109 	 * Force memory writes to complete before letting h/w
110 	 * know there are new descriptors to fetch.  (Only
111 	 * applicable for weak-ordered memory model archs,
112 	 * such as IA-64).
113 	 */
114 	wmb();
115 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
116 }
117 
118 /*
119  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
120  * @adapter: pointer to adapter struct
121  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
122  * @queue: queue to map the corresponding interrupt to
123  * @msix_vector: the vector to map to the corresponding queue
124  *
125  */
126 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
127 			     u8 queue, u8 msix_vector)
128 {
129 	u32 ivar, index;
130 	struct ixgbe_hw *hw = &adapter->hw;
131 	if (direction == -1) {
132 		/* other causes */
133 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
134 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
135 		ivar &= ~0xFF;
136 		ivar |= msix_vector;
137 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
138 	} else {
139 		/* tx or rx causes */
140 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
141 		index = ((16 * (queue & 1)) + (8 * direction));
142 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
143 		ivar &= ~(0xFF << index);
144 		ivar |= (msix_vector << index);
145 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
146 	}
147 }
148 
149 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_adapter *adapter,
150 					       struct ixgbevf_tx_buffer
151 					       *tx_buffer_info)
152 {
153 	if (tx_buffer_info->dma) {
154 		if (tx_buffer_info->mapped_as_page)
155 			dma_unmap_page(&adapter->pdev->dev,
156 				       tx_buffer_info->dma,
157 				       tx_buffer_info->length,
158 				       DMA_TO_DEVICE);
159 		else
160 			dma_unmap_single(&adapter->pdev->dev,
161 					 tx_buffer_info->dma,
162 					 tx_buffer_info->length,
163 					 DMA_TO_DEVICE);
164 		tx_buffer_info->dma = 0;
165 	}
166 	if (tx_buffer_info->skb) {
167 		dev_kfree_skb_any(tx_buffer_info->skb);
168 		tx_buffer_info->skb = NULL;
169 	}
170 	tx_buffer_info->time_stamp = 0;
171 	/* tx_buffer_info must be completely set up in the transmit path */
172 }
173 
174 #define IXGBE_MAX_TXD_PWR	14
175 #define IXGBE_MAX_DATA_PER_TXD	(1 << IXGBE_MAX_TXD_PWR)
176 
177 /* Tx Descriptors needed, worst case */
178 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
179 			 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
180 #ifdef MAX_SKB_FRAGS
181 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
182 	MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1)      /* for context */
183 #else
184 #define DESC_NEEDED TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD)
185 #endif
186 
187 static void ixgbevf_tx_timeout(struct net_device *netdev);
188 
189 /**
190  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
191  * @adapter: board private structure
192  * @tx_ring: tx ring to clean
193  **/
194 static bool ixgbevf_clean_tx_irq(struct ixgbevf_adapter *adapter,
195 				 struct ixgbevf_ring *tx_ring)
196 {
197 	struct net_device *netdev = adapter->netdev;
198 	struct ixgbe_hw *hw = &adapter->hw;
199 	union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
200 	struct ixgbevf_tx_buffer *tx_buffer_info;
201 	unsigned int i, eop, count = 0;
202 	unsigned int total_bytes = 0, total_packets = 0;
203 
204 	i = tx_ring->next_to_clean;
205 	eop = tx_ring->tx_buffer_info[i].next_to_watch;
206 	eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
207 
208 	while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
209 	       (count < tx_ring->work_limit)) {
210 		bool cleaned = false;
211 		rmb(); /* read buffer_info after eop_desc */
212 		/* eop could change between read and DD-check */
213 		if (unlikely(eop != tx_ring->tx_buffer_info[i].next_to_watch))
214 			goto cont_loop;
215 		for ( ; !cleaned; count++) {
216 			struct sk_buff *skb;
217 			tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
218 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
219 			cleaned = (i == eop);
220 			skb = tx_buffer_info->skb;
221 
222 			if (cleaned && skb) {
223 				unsigned int segs, bytecount;
224 
225 				/* gso_segs is currently only valid for tcp */
226 				segs = skb_shinfo(skb)->gso_segs ?: 1;
227 				/* multiply data chunks by size of headers */
228 				bytecount = ((segs - 1) * skb_headlen(skb)) +
229 					    skb->len;
230 				total_packets += segs;
231 				total_bytes += bytecount;
232 			}
233 
234 			ixgbevf_unmap_and_free_tx_resource(adapter,
235 							   tx_buffer_info);
236 
237 			tx_desc->wb.status = 0;
238 
239 			i++;
240 			if (i == tx_ring->count)
241 				i = 0;
242 		}
243 
244 cont_loop:
245 		eop = tx_ring->tx_buffer_info[i].next_to_watch;
246 		eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
247 	}
248 
249 	tx_ring->next_to_clean = i;
250 
251 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
252 	if (unlikely(count && netif_carrier_ok(netdev) &&
253 		     (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
254 		/* Make sure that anybody stopping the queue after this
255 		 * sees the new next_to_clean.
256 		 */
257 		smp_mb();
258 #ifdef HAVE_TX_MQ
259 		if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
260 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
261 			netif_wake_subqueue(netdev, tx_ring->queue_index);
262 			++adapter->restart_queue;
263 		}
264 #else
265 		if (netif_queue_stopped(netdev) &&
266 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
267 			netif_wake_queue(netdev);
268 			++adapter->restart_queue;
269 		}
270 #endif
271 	}
272 
273 	/* re-arm the interrupt */
274 	if ((count >= tx_ring->work_limit) &&
275 	    (!test_bit(__IXGBEVF_DOWN, &adapter->state))) {
276 		IXGBE_WRITE_REG(hw, IXGBE_VTEICS, tx_ring->v_idx);
277 	}
278 
279 	u64_stats_update_begin(&tx_ring->syncp);
280 	tx_ring->total_bytes += total_bytes;
281 	tx_ring->total_packets += total_packets;
282 	u64_stats_update_end(&tx_ring->syncp);
283 
284 	return count < tx_ring->work_limit;
285 }
286 
287 /**
288  * ixgbevf_receive_skb - Send a completed packet up the stack
289  * @q_vector: structure containing interrupt and ring information
290  * @skb: packet to send up
291  * @status: hardware indication of status of receive
292  * @rx_ring: rx descriptor ring (for a specific queue) to setup
293  * @rx_desc: rx descriptor
294  **/
295 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
296 				struct sk_buff *skb, u8 status,
297 				struct ixgbevf_ring *ring,
298 				union ixgbe_adv_rx_desc *rx_desc)
299 {
300 	struct ixgbevf_adapter *adapter = q_vector->adapter;
301 	bool is_vlan = (status & IXGBE_RXD_STAT_VP);
302 	u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
303 
304 	if (is_vlan && test_bit(tag, adapter->active_vlans))
305 		__vlan_hwaccel_put_tag(skb, tag);
306 
307 	if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
308 			napi_gro_receive(&q_vector->napi, skb);
309 	else
310 			netif_rx(skb);
311 }
312 
313 /**
314  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
315  * @adapter: address of board private structure
316  * @status_err: hardware indication of status of receive
317  * @skb: skb currently being received and modified
318  **/
319 static inline void ixgbevf_rx_checksum(struct ixgbevf_adapter *adapter,
320 				       u32 status_err, struct sk_buff *skb)
321 {
322 	skb_checksum_none_assert(skb);
323 
324 	/* Rx csum disabled */
325 	if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
326 		return;
327 
328 	/* if IP and error */
329 	if ((status_err & IXGBE_RXD_STAT_IPCS) &&
330 	    (status_err & IXGBE_RXDADV_ERR_IPE)) {
331 		adapter->hw_csum_rx_error++;
332 		return;
333 	}
334 
335 	if (!(status_err & IXGBE_RXD_STAT_L4CS))
336 		return;
337 
338 	if (status_err & IXGBE_RXDADV_ERR_TCPE) {
339 		adapter->hw_csum_rx_error++;
340 		return;
341 	}
342 
343 	/* It must be a TCP or UDP packet with a valid checksum */
344 	skb->ip_summed = CHECKSUM_UNNECESSARY;
345 	adapter->hw_csum_rx_good++;
346 }
347 
348 /**
349  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
350  * @adapter: address of board private structure
351  **/
352 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
353 				     struct ixgbevf_ring *rx_ring,
354 				     int cleaned_count)
355 {
356 	struct pci_dev *pdev = adapter->pdev;
357 	union ixgbe_adv_rx_desc *rx_desc;
358 	struct ixgbevf_rx_buffer *bi;
359 	struct sk_buff *skb;
360 	unsigned int i;
361 	unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
362 
363 	i = rx_ring->next_to_use;
364 	bi = &rx_ring->rx_buffer_info[i];
365 
366 	while (cleaned_count--) {
367 		rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
368 
369 		if (!bi->page_dma &&
370 		    (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
371 			if (!bi->page) {
372 				bi->page = alloc_page(GFP_ATOMIC | __GFP_COLD);
373 				if (!bi->page) {
374 					adapter->alloc_rx_page_failed++;
375 					goto no_buffers;
376 				}
377 				bi->page_offset = 0;
378 			} else {
379 				/* use a half page if we're re-using */
380 				bi->page_offset ^= (PAGE_SIZE / 2);
381 			}
382 
383 			bi->page_dma = dma_map_page(&pdev->dev, bi->page,
384 						    bi->page_offset,
385 						    (PAGE_SIZE / 2),
386 						    DMA_FROM_DEVICE);
387 		}
388 
389 		skb = bi->skb;
390 		if (!skb) {
391 			skb = netdev_alloc_skb(adapter->netdev,
392 							       bufsz);
393 
394 			if (!skb) {
395 				adapter->alloc_rx_buff_failed++;
396 				goto no_buffers;
397 			}
398 
399 			/*
400 			 * Make buffer alignment 2 beyond a 16 byte boundary
401 			 * this will result in a 16 byte aligned IP header after
402 			 * the 14 byte MAC header is removed
403 			 */
404 			skb_reserve(skb, NET_IP_ALIGN);
405 
406 			bi->skb = skb;
407 		}
408 		if (!bi->dma) {
409 			bi->dma = dma_map_single(&pdev->dev, skb->data,
410 						 rx_ring->rx_buf_len,
411 						 DMA_FROM_DEVICE);
412 		}
413 		/* Refresh the desc even if buffer_addrs didn't change because
414 		 * each write-back erases this info. */
415 		if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
416 			rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
417 			rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
418 		} else {
419 			rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
420 		}
421 
422 		i++;
423 		if (i == rx_ring->count)
424 			i = 0;
425 		bi = &rx_ring->rx_buffer_info[i];
426 	}
427 
428 no_buffers:
429 	if (rx_ring->next_to_use != i) {
430 		rx_ring->next_to_use = i;
431 		if (i-- == 0)
432 			i = (rx_ring->count - 1);
433 
434 		ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
435 	}
436 }
437 
438 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
439 					     u64 qmask)
440 {
441 	u32 mask;
442 	struct ixgbe_hw *hw = &adapter->hw;
443 
444 	mask = (qmask & 0xFFFFFFFF);
445 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
446 }
447 
448 static inline u16 ixgbevf_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
449 {
450 	return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
451 }
452 
453 static inline u16 ixgbevf_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
454 {
455 	return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
456 }
457 
458 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
459 				 struct ixgbevf_ring *rx_ring,
460 				 int *work_done, int work_to_do)
461 {
462 	struct ixgbevf_adapter *adapter = q_vector->adapter;
463 	struct pci_dev *pdev = adapter->pdev;
464 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
465 	struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
466 	struct sk_buff *skb;
467 	unsigned int i;
468 	u32 len, staterr;
469 	u16 hdr_info;
470 	bool cleaned = false;
471 	int cleaned_count = 0;
472 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
473 
474 	i = rx_ring->next_to_clean;
475 	rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
476 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
477 	rx_buffer_info = &rx_ring->rx_buffer_info[i];
478 
479 	while (staterr & IXGBE_RXD_STAT_DD) {
480 		u32 upper_len = 0;
481 		if (*work_done >= work_to_do)
482 			break;
483 		(*work_done)++;
484 
485 		rmb(); /* read descriptor and rx_buffer_info after status DD */
486 		if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
487 			hdr_info = le16_to_cpu(ixgbevf_get_hdr_info(rx_desc));
488 			len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
489 			       IXGBE_RXDADV_HDRBUFLEN_SHIFT;
490 			if (hdr_info & IXGBE_RXDADV_SPH)
491 				adapter->rx_hdr_split++;
492 			if (len > IXGBEVF_RX_HDR_SIZE)
493 				len = IXGBEVF_RX_HDR_SIZE;
494 			upper_len = le16_to_cpu(rx_desc->wb.upper.length);
495 		} else {
496 			len = le16_to_cpu(rx_desc->wb.upper.length);
497 		}
498 		cleaned = true;
499 		skb = rx_buffer_info->skb;
500 		prefetch(skb->data - NET_IP_ALIGN);
501 		rx_buffer_info->skb = NULL;
502 
503 		if (rx_buffer_info->dma) {
504 			dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
505 					 rx_ring->rx_buf_len,
506 					 DMA_FROM_DEVICE);
507 			rx_buffer_info->dma = 0;
508 			skb_put(skb, len);
509 		}
510 
511 		if (upper_len) {
512 			dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
513 				       PAGE_SIZE / 2, DMA_FROM_DEVICE);
514 			rx_buffer_info->page_dma = 0;
515 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
516 					   rx_buffer_info->page,
517 					   rx_buffer_info->page_offset,
518 					   upper_len);
519 
520 			if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
521 			    (page_count(rx_buffer_info->page) != 1))
522 				rx_buffer_info->page = NULL;
523 			else
524 				get_page(rx_buffer_info->page);
525 
526 			skb->len += upper_len;
527 			skb->data_len += upper_len;
528 			skb->truesize += upper_len;
529 		}
530 
531 		i++;
532 		if (i == rx_ring->count)
533 			i = 0;
534 
535 		next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
536 		prefetch(next_rxd);
537 		cleaned_count++;
538 
539 		next_buffer = &rx_ring->rx_buffer_info[i];
540 
541 		if (!(staterr & IXGBE_RXD_STAT_EOP)) {
542 			if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
543 				rx_buffer_info->skb = next_buffer->skb;
544 				rx_buffer_info->dma = next_buffer->dma;
545 				next_buffer->skb = skb;
546 				next_buffer->dma = 0;
547 			} else {
548 				skb->next = next_buffer->skb;
549 				skb->next->prev = skb;
550 			}
551 			adapter->non_eop_descs++;
552 			goto next_desc;
553 		}
554 
555 		/* ERR_MASK will only have valid bits if EOP set */
556 		if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
557 			dev_kfree_skb_irq(skb);
558 			goto next_desc;
559 		}
560 
561 		ixgbevf_rx_checksum(adapter, staterr, skb);
562 
563 		/* probably a little skewed due to removing CRC */
564 		total_rx_bytes += skb->len;
565 		total_rx_packets++;
566 
567 		/*
568 		 * Work around issue of some types of VM to VM loop back
569 		 * packets not getting split correctly
570 		 */
571 		if (staterr & IXGBE_RXD_STAT_LB) {
572 			u32 header_fixup_len = skb_headlen(skb);
573 			if (header_fixup_len < 14)
574 				skb_push(skb, header_fixup_len);
575 		}
576 		skb->protocol = eth_type_trans(skb, adapter->netdev);
577 
578 		ixgbevf_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
579 
580 next_desc:
581 		rx_desc->wb.upper.status_error = 0;
582 
583 		/* return some buffers to hardware, one at a time is too slow */
584 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
585 			ixgbevf_alloc_rx_buffers(adapter, rx_ring,
586 						 cleaned_count);
587 			cleaned_count = 0;
588 		}
589 
590 		/* use prefetched values */
591 		rx_desc = next_rxd;
592 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
593 
594 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
595 	}
596 
597 	rx_ring->next_to_clean = i;
598 	cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
599 
600 	if (cleaned_count)
601 		ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
602 
603 	u64_stats_update_begin(&rx_ring->syncp);
604 	rx_ring->total_packets += total_rx_packets;
605 	rx_ring->total_bytes += total_rx_bytes;
606 	u64_stats_update_end(&rx_ring->syncp);
607 
608 	return cleaned;
609 }
610 
611 /**
612  * ixgbevf_clean_rxonly - msix (aka one shot) rx clean routine
613  * @napi: napi struct with our devices info in it
614  * @budget: amount of work driver is allowed to do this pass, in packets
615  *
616  * This function is optimized for cleaning one queue only on a single
617  * q_vector!!!
618  **/
619 static int ixgbevf_clean_rxonly(struct napi_struct *napi, int budget)
620 {
621 	struct ixgbevf_q_vector *q_vector =
622 		container_of(napi, struct ixgbevf_q_vector, napi);
623 	struct ixgbevf_adapter *adapter = q_vector->adapter;
624 	struct ixgbevf_ring *rx_ring = NULL;
625 	int work_done = 0;
626 	long r_idx;
627 
628 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
629 	rx_ring = &(adapter->rx_ring[r_idx]);
630 
631 	ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
632 
633 	/* If all Rx work done, exit the polling mode */
634 	if (work_done < budget) {
635 		napi_complete(napi);
636 		if (adapter->itr_setting & 1)
637 			ixgbevf_set_itr_msix(q_vector);
638 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
639 			ixgbevf_irq_enable_queues(adapter, rx_ring->v_idx);
640 	}
641 
642 	return work_done;
643 }
644 
645 /**
646  * ixgbevf_clean_rxonly_many - msix (aka one shot) rx clean routine
647  * @napi: napi struct with our devices info in it
648  * @budget: amount of work driver is allowed to do this pass, in packets
649  *
650  * This function will clean more than one rx queue associated with a
651  * q_vector.
652  **/
653 static int ixgbevf_clean_rxonly_many(struct napi_struct *napi, int budget)
654 {
655 	struct ixgbevf_q_vector *q_vector =
656 		container_of(napi, struct ixgbevf_q_vector, napi);
657 	struct ixgbevf_adapter *adapter = q_vector->adapter;
658 	struct ixgbevf_ring *rx_ring = NULL;
659 	int work_done = 0, i;
660 	long r_idx;
661 	u64 enable_mask = 0;
662 
663 	/* attempt to distribute budget to each queue fairly, but don't allow
664 	 * the budget to go below 1 because we'll exit polling */
665 	budget /= (q_vector->rxr_count ?: 1);
666 	budget = max(budget, 1);
667 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
668 	for (i = 0; i < q_vector->rxr_count; i++) {
669 		rx_ring = &(adapter->rx_ring[r_idx]);
670 		ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
671 		enable_mask |= rx_ring->v_idx;
672 		r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
673 				      r_idx + 1);
674 	}
675 
676 #ifndef HAVE_NETDEV_NAPI_LIST
677 	if (!netif_running(adapter->netdev))
678 		work_done = 0;
679 
680 #endif
681 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
682 	rx_ring = &(adapter->rx_ring[r_idx]);
683 
684 	/* If all Rx work done, exit the polling mode */
685 	if (work_done < budget) {
686 		napi_complete(napi);
687 		if (adapter->itr_setting & 1)
688 			ixgbevf_set_itr_msix(q_vector);
689 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
690 			ixgbevf_irq_enable_queues(adapter, enable_mask);
691 	}
692 
693 	return work_done;
694 }
695 
696 
697 /**
698  * ixgbevf_configure_msix - Configure MSI-X hardware
699  * @adapter: board private structure
700  *
701  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
702  * interrupts.
703  **/
704 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
705 {
706 	struct ixgbevf_q_vector *q_vector;
707 	struct ixgbe_hw *hw = &adapter->hw;
708 	int i, j, q_vectors, v_idx, r_idx;
709 	u32 mask;
710 
711 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
712 
713 	/*
714 	 * Populate the IVAR table and set the ITR values to the
715 	 * corresponding register.
716 	 */
717 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
718 		q_vector = adapter->q_vector[v_idx];
719 		/* XXX for_each_set_bit(...) */
720 		r_idx = find_first_bit(q_vector->rxr_idx,
721 				       adapter->num_rx_queues);
722 
723 		for (i = 0; i < q_vector->rxr_count; i++) {
724 			j = adapter->rx_ring[r_idx].reg_idx;
725 			ixgbevf_set_ivar(adapter, 0, j, v_idx);
726 			r_idx = find_next_bit(q_vector->rxr_idx,
727 					      adapter->num_rx_queues,
728 					      r_idx + 1);
729 		}
730 		r_idx = find_first_bit(q_vector->txr_idx,
731 				       adapter->num_tx_queues);
732 
733 		for (i = 0; i < q_vector->txr_count; i++) {
734 			j = adapter->tx_ring[r_idx].reg_idx;
735 			ixgbevf_set_ivar(adapter, 1, j, v_idx);
736 			r_idx = find_next_bit(q_vector->txr_idx,
737 					      adapter->num_tx_queues,
738 					      r_idx + 1);
739 		}
740 
741 		/* if this is a tx only vector halve the interrupt rate */
742 		if (q_vector->txr_count && !q_vector->rxr_count)
743 			q_vector->eitr = (adapter->eitr_param >> 1);
744 		else if (q_vector->rxr_count)
745 			/* rx only */
746 			q_vector->eitr = adapter->eitr_param;
747 
748 		ixgbevf_write_eitr(adapter, v_idx, q_vector->eitr);
749 	}
750 
751 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
752 
753 	/* set up to autoclear timer, and the vectors */
754 	mask = IXGBE_EIMS_ENABLE_MASK;
755 	mask &= ~IXGBE_EIMS_OTHER;
756 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
757 }
758 
759 enum latency_range {
760 	lowest_latency = 0,
761 	low_latency = 1,
762 	bulk_latency = 2,
763 	latency_invalid = 255
764 };
765 
766 /**
767  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
768  * @adapter: pointer to adapter
769  * @eitr: eitr setting (ints per sec) to give last timeslice
770  * @itr_setting: current throttle rate in ints/second
771  * @packets: the number of packets during this measurement interval
772  * @bytes: the number of bytes during this measurement interval
773  *
774  *      Stores a new ITR value based on packets and byte
775  *      counts during the last interrupt.  The advantage of per interrupt
776  *      computation is faster updates and more accurate ITR for the current
777  *      traffic pattern.  Constants in this function were computed
778  *      based on theoretical maximum wire speed and thresholds were set based
779  *      on testing data as well as attempting to minimize response time
780  *      while increasing bulk throughput.
781  **/
782 static u8 ixgbevf_update_itr(struct ixgbevf_adapter *adapter,
783 			     u32 eitr, u8 itr_setting,
784 			     int packets, int bytes)
785 {
786 	unsigned int retval = itr_setting;
787 	u32 timepassed_us;
788 	u64 bytes_perint;
789 
790 	if (packets == 0)
791 		goto update_itr_done;
792 
793 
794 	/* simple throttlerate management
795 	 *    0-20MB/s lowest (100000 ints/s)
796 	 *   20-100MB/s low   (20000 ints/s)
797 	 *  100-1249MB/s bulk (8000 ints/s)
798 	 */
799 	/* what was last interrupt timeslice? */
800 	timepassed_us = 1000000/eitr;
801 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
802 
803 	switch (itr_setting) {
804 	case lowest_latency:
805 		if (bytes_perint > adapter->eitr_low)
806 			retval = low_latency;
807 		break;
808 	case low_latency:
809 		if (bytes_perint > adapter->eitr_high)
810 			retval = bulk_latency;
811 		else if (bytes_perint <= adapter->eitr_low)
812 			retval = lowest_latency;
813 		break;
814 	case bulk_latency:
815 		if (bytes_perint <= adapter->eitr_high)
816 			retval = low_latency;
817 		break;
818 	}
819 
820 update_itr_done:
821 	return retval;
822 }
823 
824 /**
825  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
826  * @adapter: pointer to adapter struct
827  * @v_idx: vector index into q_vector array
828  * @itr_reg: new value to be written in *register* format, not ints/s
829  *
830  * This function is made to be called by ethtool and by the driver
831  * when it needs to update VTEITR registers at runtime.  Hardware
832  * specific quirks/differences are taken care of here.
833  */
834 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
835 			       u32 itr_reg)
836 {
837 	struct ixgbe_hw *hw = &adapter->hw;
838 
839 	itr_reg = EITR_INTS_PER_SEC_TO_REG(itr_reg);
840 
841 	/*
842 	 * set the WDIS bit to not clear the timer bits and cause an
843 	 * immediate assertion of the interrupt
844 	 */
845 	itr_reg |= IXGBE_EITR_CNT_WDIS;
846 
847 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
848 }
849 
850 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector)
851 {
852 	struct ixgbevf_adapter *adapter = q_vector->adapter;
853 	u32 new_itr;
854 	u8 current_itr, ret_itr;
855 	int i, r_idx, v_idx = q_vector->v_idx;
856 	struct ixgbevf_ring *rx_ring, *tx_ring;
857 
858 	r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
859 	for (i = 0; i < q_vector->txr_count; i++) {
860 		tx_ring = &(adapter->tx_ring[r_idx]);
861 		ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
862 					     q_vector->tx_itr,
863 					     tx_ring->total_packets,
864 					     tx_ring->total_bytes);
865 		/* if the result for this queue would decrease interrupt
866 		 * rate for this vector then use that result */
867 		q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
868 				    q_vector->tx_itr - 1 : ret_itr);
869 		r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
870 				      r_idx + 1);
871 	}
872 
873 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
874 	for (i = 0; i < q_vector->rxr_count; i++) {
875 		rx_ring = &(adapter->rx_ring[r_idx]);
876 		ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
877 					     q_vector->rx_itr,
878 					     rx_ring->total_packets,
879 					     rx_ring->total_bytes);
880 		/* if the result for this queue would decrease interrupt
881 		 * rate for this vector then use that result */
882 		q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
883 				    q_vector->rx_itr - 1 : ret_itr);
884 		r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
885 				      r_idx + 1);
886 	}
887 
888 	current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
889 
890 	switch (current_itr) {
891 	/* counts and packets in update_itr are dependent on these numbers */
892 	case lowest_latency:
893 		new_itr = 100000;
894 		break;
895 	case low_latency:
896 		new_itr = 20000; /* aka hwitr = ~200 */
897 		break;
898 	case bulk_latency:
899 	default:
900 		new_itr = 8000;
901 		break;
902 	}
903 
904 	if (new_itr != q_vector->eitr) {
905 		u32 itr_reg;
906 
907 		/* save the algorithm value here, not the smoothed one */
908 		q_vector->eitr = new_itr;
909 		/* do an exponential smoothing */
910 		new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
911 		itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
912 		ixgbevf_write_eitr(adapter, v_idx, itr_reg);
913 	}
914 }
915 
916 static irqreturn_t ixgbevf_msix_mbx(int irq, void *data)
917 {
918 	struct net_device *netdev = data;
919 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
920 	struct ixgbe_hw *hw = &adapter->hw;
921 	u32 eicr;
922 	u32 msg;
923 	bool got_ack = false;
924 
925 	eicr = IXGBE_READ_REG(hw, IXGBE_VTEICS);
926 	IXGBE_WRITE_REG(hw, IXGBE_VTEICR, eicr);
927 
928 	if (!hw->mbx.ops.check_for_ack(hw))
929 		got_ack = true;
930 
931 	if (!hw->mbx.ops.check_for_msg(hw)) {
932 		hw->mbx.ops.read(hw, &msg, 1);
933 
934 		if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG)
935 			mod_timer(&adapter->watchdog_timer,
936 				  round_jiffies(jiffies + 1));
937 
938 		if (msg & IXGBE_VT_MSGTYPE_NACK)
939 			pr_warn("Last Request of type %2.2x to PF Nacked\n",
940 				msg & 0xFF);
941 		/*
942 		 * Restore the PFSTS bit in case someone is polling for a
943 		 * return message from the PF
944 		 */
945 		hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
946 	}
947 
948 	/*
949 	 * checking for the ack clears the PFACK bit.  Place
950 	 * it back in the v2p_mailbox cache so that anyone
951 	 * polling for an ack will not miss it
952 	 */
953 	if (got_ack)
954 		hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
955 
956 	return IRQ_HANDLED;
957 }
958 
959 static irqreturn_t ixgbevf_msix_clean_tx(int irq, void *data)
960 {
961 	struct ixgbevf_q_vector *q_vector = data;
962 	struct ixgbevf_adapter  *adapter = q_vector->adapter;
963 	struct ixgbevf_ring     *tx_ring;
964 	int i, r_idx;
965 
966 	if (!q_vector->txr_count)
967 		return IRQ_HANDLED;
968 
969 	r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
970 	for (i = 0; i < q_vector->txr_count; i++) {
971 		tx_ring = &(adapter->tx_ring[r_idx]);
972 		tx_ring->total_bytes = 0;
973 		tx_ring->total_packets = 0;
974 		ixgbevf_clean_tx_irq(adapter, tx_ring);
975 		r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
976 				      r_idx + 1);
977 	}
978 
979 	if (adapter->itr_setting & 1)
980 		ixgbevf_set_itr_msix(q_vector);
981 
982 	return IRQ_HANDLED;
983 }
984 
985 /**
986  * ixgbevf_msix_clean_rx - single unshared vector rx clean (all queues)
987  * @irq: unused
988  * @data: pointer to our q_vector struct for this interrupt vector
989  **/
990 static irqreturn_t ixgbevf_msix_clean_rx(int irq, void *data)
991 {
992 	struct ixgbevf_q_vector *q_vector = data;
993 	struct ixgbevf_adapter  *adapter = q_vector->adapter;
994 	struct ixgbe_hw *hw = &adapter->hw;
995 	struct ixgbevf_ring  *rx_ring;
996 	int r_idx;
997 	int i;
998 
999 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1000 	for (i = 0; i < q_vector->rxr_count; i++) {
1001 		rx_ring = &(adapter->rx_ring[r_idx]);
1002 		rx_ring->total_bytes = 0;
1003 		rx_ring->total_packets = 0;
1004 		r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1005 				      r_idx + 1);
1006 	}
1007 
1008 	if (!q_vector->rxr_count)
1009 		return IRQ_HANDLED;
1010 
1011 	r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1012 	rx_ring = &(adapter->rx_ring[r_idx]);
1013 	/* disable interrupts on this vector only */
1014 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, rx_ring->v_idx);
1015 	napi_schedule(&q_vector->napi);
1016 
1017 
1018 	return IRQ_HANDLED;
1019 }
1020 
1021 static irqreturn_t ixgbevf_msix_clean_many(int irq, void *data)
1022 {
1023 	ixgbevf_msix_clean_rx(irq, data);
1024 	ixgbevf_msix_clean_tx(irq, data);
1025 
1026 	return IRQ_HANDLED;
1027 }
1028 
1029 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1030 				     int r_idx)
1031 {
1032 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1033 
1034 	set_bit(r_idx, q_vector->rxr_idx);
1035 	q_vector->rxr_count++;
1036 	a->rx_ring[r_idx].v_idx = 1 << v_idx;
1037 }
1038 
1039 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1040 				     int t_idx)
1041 {
1042 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1043 
1044 	set_bit(t_idx, q_vector->txr_idx);
1045 	q_vector->txr_count++;
1046 	a->tx_ring[t_idx].v_idx = 1 << v_idx;
1047 }
1048 
1049 /**
1050  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1051  * @adapter: board private structure to initialize
1052  *
1053  * This function maps descriptor rings to the queue-specific vectors
1054  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1055  * one vector per ring/queue, but on a constrained vector budget, we
1056  * group the rings as "efficiently" as possible.  You would add new
1057  * mapping configurations in here.
1058  **/
1059 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1060 {
1061 	int q_vectors;
1062 	int v_start = 0;
1063 	int rxr_idx = 0, txr_idx = 0;
1064 	int rxr_remaining = adapter->num_rx_queues;
1065 	int txr_remaining = adapter->num_tx_queues;
1066 	int i, j;
1067 	int rqpv, tqpv;
1068 	int err = 0;
1069 
1070 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1071 
1072 	/*
1073 	 * The ideal configuration...
1074 	 * We have enough vectors to map one per queue.
1075 	 */
1076 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1077 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1078 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1079 
1080 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1081 			map_vector_to_txq(adapter, v_start, txr_idx);
1082 		goto out;
1083 	}
1084 
1085 	/*
1086 	 * If we don't have enough vectors for a 1-to-1
1087 	 * mapping, we'll have to group them so there are
1088 	 * multiple queues per vector.
1089 	 */
1090 	/* Re-adjusting *qpv takes care of the remainder. */
1091 	for (i = v_start; i < q_vectors; i++) {
1092 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1093 		for (j = 0; j < rqpv; j++) {
1094 			map_vector_to_rxq(adapter, i, rxr_idx);
1095 			rxr_idx++;
1096 			rxr_remaining--;
1097 		}
1098 	}
1099 	for (i = v_start; i < q_vectors; i++) {
1100 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1101 		for (j = 0; j < tqpv; j++) {
1102 			map_vector_to_txq(adapter, i, txr_idx);
1103 			txr_idx++;
1104 			txr_remaining--;
1105 		}
1106 	}
1107 
1108 out:
1109 	return err;
1110 }
1111 
1112 /**
1113  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1114  * @adapter: board private structure
1115  *
1116  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1117  * interrupts from the kernel.
1118  **/
1119 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1120 {
1121 	struct net_device *netdev = adapter->netdev;
1122 	irqreturn_t (*handler)(int, void *);
1123 	int i, vector, q_vectors, err;
1124 	int ri = 0, ti = 0;
1125 
1126 	/* Decrement for Other and TCP Timer vectors */
1127 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1128 
1129 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count)          \
1130 					  ? &ixgbevf_msix_clean_many : \
1131 			  (_v)->rxr_count ? &ixgbevf_msix_clean_rx   : \
1132 			  (_v)->txr_count ? &ixgbevf_msix_clean_tx   : \
1133 			  NULL)
1134 	for (vector = 0; vector < q_vectors; vector++) {
1135 		handler = SET_HANDLER(adapter->q_vector[vector]);
1136 
1137 		if (handler == &ixgbevf_msix_clean_rx) {
1138 			sprintf(adapter->name[vector], "%s-%s-%d",
1139 				netdev->name, "rx", ri++);
1140 		} else if (handler == &ixgbevf_msix_clean_tx) {
1141 			sprintf(adapter->name[vector], "%s-%s-%d",
1142 				netdev->name, "tx", ti++);
1143 		} else if (handler == &ixgbevf_msix_clean_many) {
1144 			sprintf(adapter->name[vector], "%s-%s-%d",
1145 				netdev->name, "TxRx", vector);
1146 		} else {
1147 			/* skip this unused q_vector */
1148 			continue;
1149 		}
1150 		err = request_irq(adapter->msix_entries[vector].vector,
1151 				  handler, 0, adapter->name[vector],
1152 				  adapter->q_vector[vector]);
1153 		if (err) {
1154 			hw_dbg(&adapter->hw,
1155 			       "request_irq failed for MSIX interrupt "
1156 			       "Error: %d\n", err);
1157 			goto free_queue_irqs;
1158 		}
1159 	}
1160 
1161 	sprintf(adapter->name[vector], "%s:mbx", netdev->name);
1162 	err = request_irq(adapter->msix_entries[vector].vector,
1163 			  &ixgbevf_msix_mbx, 0, adapter->name[vector], netdev);
1164 	if (err) {
1165 		hw_dbg(&adapter->hw,
1166 		       "request_irq for msix_mbx failed: %d\n", err);
1167 		goto free_queue_irqs;
1168 	}
1169 
1170 	return 0;
1171 
1172 free_queue_irqs:
1173 	for (i = vector - 1; i >= 0; i--)
1174 		free_irq(adapter->msix_entries[--vector].vector,
1175 			 &(adapter->q_vector[i]));
1176 	pci_disable_msix(adapter->pdev);
1177 	kfree(adapter->msix_entries);
1178 	adapter->msix_entries = NULL;
1179 	return err;
1180 }
1181 
1182 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1183 {
1184 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1185 
1186 	for (i = 0; i < q_vectors; i++) {
1187 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1188 		bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1189 		bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1190 		q_vector->rxr_count = 0;
1191 		q_vector->txr_count = 0;
1192 		q_vector->eitr = adapter->eitr_param;
1193 	}
1194 }
1195 
1196 /**
1197  * ixgbevf_request_irq - initialize interrupts
1198  * @adapter: board private structure
1199  *
1200  * Attempts to configure interrupts using the best available
1201  * capabilities of the hardware and kernel.
1202  **/
1203 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1204 {
1205 	int err = 0;
1206 
1207 	err = ixgbevf_request_msix_irqs(adapter);
1208 
1209 	if (err)
1210 		hw_dbg(&adapter->hw,
1211 		       "request_irq failed, Error %d\n", err);
1212 
1213 	return err;
1214 }
1215 
1216 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1217 {
1218 	struct net_device *netdev = adapter->netdev;
1219 	int i, q_vectors;
1220 
1221 	q_vectors = adapter->num_msix_vectors;
1222 
1223 	i = q_vectors - 1;
1224 
1225 	free_irq(adapter->msix_entries[i].vector, netdev);
1226 	i--;
1227 
1228 	for (; i >= 0; i--) {
1229 		free_irq(adapter->msix_entries[i].vector,
1230 			 adapter->q_vector[i]);
1231 	}
1232 
1233 	ixgbevf_reset_q_vectors(adapter);
1234 }
1235 
1236 /**
1237  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1238  * @adapter: board private structure
1239  **/
1240 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1241 {
1242 	int i;
1243 	struct ixgbe_hw *hw = &adapter->hw;
1244 
1245 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1246 
1247 	IXGBE_WRITE_FLUSH(hw);
1248 
1249 	for (i = 0; i < adapter->num_msix_vectors; i++)
1250 		synchronize_irq(adapter->msix_entries[i].vector);
1251 }
1252 
1253 /**
1254  * ixgbevf_irq_enable - Enable default interrupt generation settings
1255  * @adapter: board private structure
1256  **/
1257 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter,
1258 				      bool queues, bool flush)
1259 {
1260 	struct ixgbe_hw *hw = &adapter->hw;
1261 	u32 mask;
1262 	u64 qmask;
1263 
1264 	mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1265 	qmask = ~0;
1266 
1267 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
1268 
1269 	if (queues)
1270 		ixgbevf_irq_enable_queues(adapter, qmask);
1271 
1272 	if (flush)
1273 		IXGBE_WRITE_FLUSH(hw);
1274 }
1275 
1276 /**
1277  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1278  * @adapter: board private structure
1279  *
1280  * Configure the Tx unit of the MAC after a reset.
1281  **/
1282 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1283 {
1284 	u64 tdba;
1285 	struct ixgbe_hw *hw = &adapter->hw;
1286 	u32 i, j, tdlen, txctrl;
1287 
1288 	/* Setup the HW Tx Head and Tail descriptor pointers */
1289 	for (i = 0; i < adapter->num_tx_queues; i++) {
1290 		struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1291 		j = ring->reg_idx;
1292 		tdba = ring->dma;
1293 		tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1294 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1295 				(tdba & DMA_BIT_MASK(32)));
1296 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1297 		IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1298 		IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1299 		IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1300 		adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1301 		adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1302 		/* Disable Tx Head Writeback RO bit, since this hoses
1303 		 * bookkeeping if things aren't delivered in order.
1304 		 */
1305 		txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1306 		txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1307 		IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1308 	}
1309 }
1310 
1311 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1312 
1313 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1314 {
1315 	struct ixgbevf_ring *rx_ring;
1316 	struct ixgbe_hw *hw = &adapter->hw;
1317 	u32 srrctl;
1318 
1319 	rx_ring = &adapter->rx_ring[index];
1320 
1321 	srrctl = IXGBE_SRRCTL_DROP_EN;
1322 
1323 	if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1324 		u16 bufsz = IXGBEVF_RXBUFFER_2048;
1325 		/* grow the amount we can receive on large page machines */
1326 		if (bufsz < (PAGE_SIZE / 2))
1327 			bufsz = (PAGE_SIZE / 2);
1328 		/* cap the bufsz at our largest descriptor size */
1329 		bufsz = min((u16)IXGBEVF_MAX_RXBUFFER, bufsz);
1330 
1331 		srrctl |= bufsz >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1332 		srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1333 		srrctl |= ((IXGBEVF_RX_HDR_SIZE <<
1334 			   IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1335 			   IXGBE_SRRCTL_BSIZEHDR_MASK);
1336 	} else {
1337 		srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1338 
1339 		if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1340 			srrctl |= IXGBEVF_RXBUFFER_2048 >>
1341 				IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1342 		else
1343 			srrctl |= rx_ring->rx_buf_len >>
1344 				IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1345 	}
1346 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1347 }
1348 
1349 /**
1350  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1351  * @adapter: board private structure
1352  *
1353  * Configure the Rx unit of the MAC after a reset.
1354  **/
1355 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1356 {
1357 	u64 rdba;
1358 	struct ixgbe_hw *hw = &adapter->hw;
1359 	struct net_device *netdev = adapter->netdev;
1360 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1361 	int i, j;
1362 	u32 rdlen;
1363 	int rx_buf_len;
1364 
1365 	/* Decide whether to use packet split mode or not */
1366 	if (netdev->mtu > ETH_DATA_LEN) {
1367 		if (adapter->flags & IXGBE_FLAG_RX_PS_CAPABLE)
1368 			adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1369 		else
1370 			adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
1371 	} else {
1372 		if (adapter->flags & IXGBE_FLAG_RX_1BUF_CAPABLE)
1373 			adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
1374 		else
1375 			adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1376 	}
1377 
1378 	/* Set the RX buffer length according to the mode */
1379 	if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1380 		/* PSRTYPE must be initialized in 82599 */
1381 		u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
1382 			IXGBE_PSRTYPE_UDPHDR |
1383 			IXGBE_PSRTYPE_IPV4HDR |
1384 			IXGBE_PSRTYPE_IPV6HDR |
1385 			IXGBE_PSRTYPE_L2HDR;
1386 		IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1387 		rx_buf_len = IXGBEVF_RX_HDR_SIZE;
1388 	} else {
1389 		IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1390 		if (netdev->mtu <= ETH_DATA_LEN)
1391 			rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1392 		else
1393 			rx_buf_len = ALIGN(max_frame, 1024);
1394 	}
1395 
1396 	rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1397 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1398 	 * the Base and Length of the Rx Descriptor Ring */
1399 	for (i = 0; i < adapter->num_rx_queues; i++) {
1400 		rdba = adapter->rx_ring[i].dma;
1401 		j = adapter->rx_ring[i].reg_idx;
1402 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1403 				(rdba & DMA_BIT_MASK(32)));
1404 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1405 		IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1406 		IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1407 		IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1408 		adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1409 		adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1410 		adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1411 
1412 		ixgbevf_configure_srrctl(adapter, j);
1413 	}
1414 }
1415 
1416 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1417 {
1418 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1419 	struct ixgbe_hw *hw = &adapter->hw;
1420 
1421 	/* add VID to filter table */
1422 	if (hw->mac.ops.set_vfta)
1423 		hw->mac.ops.set_vfta(hw, vid, 0, true);
1424 	set_bit(vid, adapter->active_vlans);
1425 
1426 	return 0;
1427 }
1428 
1429 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1430 {
1431 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1432 	struct ixgbe_hw *hw = &adapter->hw;
1433 
1434 	/* remove VID from filter table */
1435 	if (hw->mac.ops.set_vfta)
1436 		hw->mac.ops.set_vfta(hw, vid, 0, false);
1437 	clear_bit(vid, adapter->active_vlans);
1438 
1439 	return 0;
1440 }
1441 
1442 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1443 {
1444 	u16 vid;
1445 
1446 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1447 		ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1448 }
1449 
1450 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1451 {
1452 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1453 	struct ixgbe_hw *hw = &adapter->hw;
1454 	int count = 0;
1455 
1456 	if ((netdev_uc_count(netdev)) > 10) {
1457 		pr_err("Too many unicast filters - No Space\n");
1458 		return -ENOSPC;
1459 	}
1460 
1461 	if (!netdev_uc_empty(netdev)) {
1462 		struct netdev_hw_addr *ha;
1463 		netdev_for_each_uc_addr(ha, netdev) {
1464 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1465 			udelay(200);
1466 		}
1467 	} else {
1468 		/*
1469 		 * If the list is empty then send message to PF driver to
1470 		 * clear all macvlans on this VF.
1471 		 */
1472 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1473 	}
1474 
1475 	return count;
1476 }
1477 
1478 /**
1479  * ixgbevf_set_rx_mode - Multicast set
1480  * @netdev: network interface device structure
1481  *
1482  * The set_rx_method entry point is called whenever the multicast address
1483  * list or the network interface flags are updated.  This routine is
1484  * responsible for configuring the hardware for proper multicast mode.
1485  **/
1486 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1487 {
1488 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1489 	struct ixgbe_hw *hw = &adapter->hw;
1490 
1491 	/* reprogram multicast list */
1492 	if (hw->mac.ops.update_mc_addr_list)
1493 		hw->mac.ops.update_mc_addr_list(hw, netdev);
1494 
1495 	ixgbevf_write_uc_addr_list(netdev);
1496 }
1497 
1498 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1499 {
1500 	int q_idx;
1501 	struct ixgbevf_q_vector *q_vector;
1502 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1503 
1504 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1505 		struct napi_struct *napi;
1506 		q_vector = adapter->q_vector[q_idx];
1507 		if (!q_vector->rxr_count)
1508 			continue;
1509 		napi = &q_vector->napi;
1510 		if (q_vector->rxr_count > 1)
1511 			napi->poll = &ixgbevf_clean_rxonly_many;
1512 
1513 		napi_enable(napi);
1514 	}
1515 }
1516 
1517 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1518 {
1519 	int q_idx;
1520 	struct ixgbevf_q_vector *q_vector;
1521 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1522 
1523 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1524 		q_vector = adapter->q_vector[q_idx];
1525 		if (!q_vector->rxr_count)
1526 			continue;
1527 		napi_disable(&q_vector->napi);
1528 	}
1529 }
1530 
1531 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1532 {
1533 	struct net_device *netdev = adapter->netdev;
1534 	int i;
1535 
1536 	ixgbevf_set_rx_mode(netdev);
1537 
1538 	ixgbevf_restore_vlan(adapter);
1539 
1540 	ixgbevf_configure_tx(adapter);
1541 	ixgbevf_configure_rx(adapter);
1542 	for (i = 0; i < adapter->num_rx_queues; i++) {
1543 		struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1544 		ixgbevf_alloc_rx_buffers(adapter, ring, ring->count);
1545 		ring->next_to_use = ring->count - 1;
1546 		writel(ring->next_to_use, adapter->hw.hw_addr + ring->tail);
1547 	}
1548 }
1549 
1550 #define IXGBE_MAX_RX_DESC_POLL 10
1551 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1552 						int rxr)
1553 {
1554 	struct ixgbe_hw *hw = &adapter->hw;
1555 	int j = adapter->rx_ring[rxr].reg_idx;
1556 	int k;
1557 
1558 	for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1559 		if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1560 			break;
1561 		else
1562 			msleep(1);
1563 	}
1564 	if (k >= IXGBE_MAX_RX_DESC_POLL) {
1565 		hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1566 		       "not set within the polling period\n", rxr);
1567 	}
1568 
1569 	ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1570 				(adapter->rx_ring[rxr].count - 1));
1571 }
1572 
1573 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1574 {
1575 	/* Only save pre-reset stats if there are some */
1576 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1577 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1578 			adapter->stats.base_vfgprc;
1579 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1580 			adapter->stats.base_vfgptc;
1581 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1582 			adapter->stats.base_vfgorc;
1583 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1584 			adapter->stats.base_vfgotc;
1585 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1586 			adapter->stats.base_vfmprc;
1587 	}
1588 }
1589 
1590 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1591 {
1592 	struct ixgbe_hw *hw = &adapter->hw;
1593 
1594 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1595 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1596 	adapter->stats.last_vfgorc |=
1597 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1598 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1599 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1600 	adapter->stats.last_vfgotc |=
1601 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1602 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1603 
1604 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1605 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1606 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1607 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1608 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1609 }
1610 
1611 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1612 {
1613 	struct net_device *netdev = adapter->netdev;
1614 	struct ixgbe_hw *hw = &adapter->hw;
1615 	int i, j = 0;
1616 	int num_rx_rings = adapter->num_rx_queues;
1617 	u32 txdctl, rxdctl;
1618 	u32 msg[2];
1619 
1620 	for (i = 0; i < adapter->num_tx_queues; i++) {
1621 		j = adapter->tx_ring[i].reg_idx;
1622 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1623 		/* enable WTHRESH=8 descriptors, to encourage burst writeback */
1624 		txdctl |= (8 << 16);
1625 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1626 	}
1627 
1628 	for (i = 0; i < adapter->num_tx_queues; i++) {
1629 		j = adapter->tx_ring[i].reg_idx;
1630 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1631 		txdctl |= IXGBE_TXDCTL_ENABLE;
1632 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1633 	}
1634 
1635 	for (i = 0; i < num_rx_rings; i++) {
1636 		j = adapter->rx_ring[i].reg_idx;
1637 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1638 		rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1639 		if (hw->mac.type == ixgbe_mac_X540_vf) {
1640 			rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1641 			rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1642 				   IXGBE_RXDCTL_RLPML_EN);
1643 		}
1644 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1645 		ixgbevf_rx_desc_queue_enable(adapter, i);
1646 	}
1647 
1648 	ixgbevf_configure_msix(adapter);
1649 
1650 	if (hw->mac.ops.set_rar) {
1651 		if (is_valid_ether_addr(hw->mac.addr))
1652 			hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1653 		else
1654 			hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1655 	}
1656 
1657 	msg[0] = IXGBE_VF_SET_LPE;
1658 	msg[1] = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1659 	hw->mbx.ops.write_posted(hw, msg, 2);
1660 
1661 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
1662 	ixgbevf_napi_enable_all(adapter);
1663 
1664 	/* enable transmits */
1665 	netif_tx_start_all_queues(netdev);
1666 
1667 	ixgbevf_save_reset_stats(adapter);
1668 	ixgbevf_init_last_counter_stats(adapter);
1669 
1670 	/* bring the link up in the watchdog, this could race with our first
1671 	 * link up interrupt but shouldn't be a problem */
1672 	adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1673 	adapter->link_check_timeout = jiffies;
1674 	mod_timer(&adapter->watchdog_timer, jiffies);
1675 }
1676 
1677 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1678 {
1679 	struct ixgbe_hw *hw = &adapter->hw;
1680 
1681 	ixgbevf_configure(adapter);
1682 
1683 	ixgbevf_up_complete(adapter);
1684 
1685 	/* clear any pending interrupts, may auto mask */
1686 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
1687 
1688 	ixgbevf_irq_enable(adapter, true, true);
1689 }
1690 
1691 /**
1692  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1693  * @adapter: board private structure
1694  * @rx_ring: ring to free buffers from
1695  **/
1696 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1697 				  struct ixgbevf_ring *rx_ring)
1698 {
1699 	struct pci_dev *pdev = adapter->pdev;
1700 	unsigned long size;
1701 	unsigned int i;
1702 
1703 	if (!rx_ring->rx_buffer_info)
1704 		return;
1705 
1706 	/* Free all the Rx ring sk_buffs */
1707 	for (i = 0; i < rx_ring->count; i++) {
1708 		struct ixgbevf_rx_buffer *rx_buffer_info;
1709 
1710 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
1711 		if (rx_buffer_info->dma) {
1712 			dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1713 					 rx_ring->rx_buf_len,
1714 					 DMA_FROM_DEVICE);
1715 			rx_buffer_info->dma = 0;
1716 		}
1717 		if (rx_buffer_info->skb) {
1718 			struct sk_buff *skb = rx_buffer_info->skb;
1719 			rx_buffer_info->skb = NULL;
1720 			do {
1721 				struct sk_buff *this = skb;
1722 				skb = skb->prev;
1723 				dev_kfree_skb(this);
1724 			} while (skb);
1725 		}
1726 		if (!rx_buffer_info->page)
1727 			continue;
1728 		dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1729 			       PAGE_SIZE / 2, DMA_FROM_DEVICE);
1730 		rx_buffer_info->page_dma = 0;
1731 		put_page(rx_buffer_info->page);
1732 		rx_buffer_info->page = NULL;
1733 		rx_buffer_info->page_offset = 0;
1734 	}
1735 
1736 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1737 	memset(rx_ring->rx_buffer_info, 0, size);
1738 
1739 	/* Zero out the descriptor ring */
1740 	memset(rx_ring->desc, 0, rx_ring->size);
1741 
1742 	rx_ring->next_to_clean = 0;
1743 	rx_ring->next_to_use = 0;
1744 
1745 	if (rx_ring->head)
1746 		writel(0, adapter->hw.hw_addr + rx_ring->head);
1747 	if (rx_ring->tail)
1748 		writel(0, adapter->hw.hw_addr + rx_ring->tail);
1749 }
1750 
1751 /**
1752  * ixgbevf_clean_tx_ring - Free Tx Buffers
1753  * @adapter: board private structure
1754  * @tx_ring: ring to be cleaned
1755  **/
1756 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1757 				  struct ixgbevf_ring *tx_ring)
1758 {
1759 	struct ixgbevf_tx_buffer *tx_buffer_info;
1760 	unsigned long size;
1761 	unsigned int i;
1762 
1763 	if (!tx_ring->tx_buffer_info)
1764 		return;
1765 
1766 	/* Free all the Tx ring sk_buffs */
1767 
1768 	for (i = 0; i < tx_ring->count; i++) {
1769 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
1770 		ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
1771 	}
1772 
1773 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1774 	memset(tx_ring->tx_buffer_info, 0, size);
1775 
1776 	memset(tx_ring->desc, 0, tx_ring->size);
1777 
1778 	tx_ring->next_to_use = 0;
1779 	tx_ring->next_to_clean = 0;
1780 
1781 	if (tx_ring->head)
1782 		writel(0, adapter->hw.hw_addr + tx_ring->head);
1783 	if (tx_ring->tail)
1784 		writel(0, adapter->hw.hw_addr + tx_ring->tail);
1785 }
1786 
1787 /**
1788  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1789  * @adapter: board private structure
1790  **/
1791 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1792 {
1793 	int i;
1794 
1795 	for (i = 0; i < adapter->num_rx_queues; i++)
1796 		ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1797 }
1798 
1799 /**
1800  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1801  * @adapter: board private structure
1802  **/
1803 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1804 {
1805 	int i;
1806 
1807 	for (i = 0; i < adapter->num_tx_queues; i++)
1808 		ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1809 }
1810 
1811 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1812 {
1813 	struct net_device *netdev = adapter->netdev;
1814 	struct ixgbe_hw *hw = &adapter->hw;
1815 	u32 txdctl;
1816 	int i, j;
1817 
1818 	/* signal that we are down to the interrupt handler */
1819 	set_bit(__IXGBEVF_DOWN, &adapter->state);
1820 	/* disable receives */
1821 
1822 	netif_tx_disable(netdev);
1823 
1824 	msleep(10);
1825 
1826 	netif_tx_stop_all_queues(netdev);
1827 
1828 	ixgbevf_irq_disable(adapter);
1829 
1830 	ixgbevf_napi_disable_all(adapter);
1831 
1832 	del_timer_sync(&adapter->watchdog_timer);
1833 	/* can't call flush scheduled work here because it can deadlock
1834 	 * if linkwatch_event tries to acquire the rtnl_lock which we are
1835 	 * holding */
1836 	while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1837 		msleep(1);
1838 
1839 	/* disable transmits in the hardware now that interrupts are off */
1840 	for (i = 0; i < adapter->num_tx_queues; i++) {
1841 		j = adapter->tx_ring[i].reg_idx;
1842 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1843 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1844 				(txdctl & ~IXGBE_TXDCTL_ENABLE));
1845 	}
1846 
1847 	netif_carrier_off(netdev);
1848 
1849 	if (!pci_channel_offline(adapter->pdev))
1850 		ixgbevf_reset(adapter);
1851 
1852 	ixgbevf_clean_all_tx_rings(adapter);
1853 	ixgbevf_clean_all_rx_rings(adapter);
1854 }
1855 
1856 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1857 {
1858 	struct ixgbe_hw *hw = &adapter->hw;
1859 
1860 	WARN_ON(in_interrupt());
1861 
1862 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1863 		msleep(1);
1864 
1865 	/*
1866 	 * Check if PF is up before re-init.  If not then skip until
1867 	 * later when the PF is up and ready to service requests from
1868 	 * the VF via mailbox.  If the VF is up and running then the
1869 	 * watchdog task will continue to schedule reset tasks until
1870 	 * the PF is up and running.
1871 	 */
1872 	if (!hw->mac.ops.reset_hw(hw)) {
1873 		ixgbevf_down(adapter);
1874 		ixgbevf_up(adapter);
1875 	}
1876 
1877 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1878 }
1879 
1880 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1881 {
1882 	struct ixgbe_hw *hw = &adapter->hw;
1883 	struct net_device *netdev = adapter->netdev;
1884 
1885 	if (hw->mac.ops.reset_hw(hw))
1886 		hw_dbg(hw, "PF still resetting\n");
1887 	else
1888 		hw->mac.ops.init_hw(hw);
1889 
1890 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1891 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1892 		       netdev->addr_len);
1893 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1894 		       netdev->addr_len);
1895 	}
1896 }
1897 
1898 static void ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1899 					 int vectors)
1900 {
1901 	int err, vector_threshold;
1902 
1903 	/* We'll want at least 3 (vector_threshold):
1904 	 * 1) TxQ[0] Cleanup
1905 	 * 2) RxQ[0] Cleanup
1906 	 * 3) Other (Link Status Change, etc.)
1907 	 */
1908 	vector_threshold = MIN_MSIX_COUNT;
1909 
1910 	/* The more we get, the more we will assign to Tx/Rx Cleanup
1911 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1912 	 * Right now, we simply care about how many we'll get; we'll
1913 	 * set them up later while requesting irq's.
1914 	 */
1915 	while (vectors >= vector_threshold) {
1916 		err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1917 				      vectors);
1918 		if (!err) /* Success in acquiring all requested vectors. */
1919 			break;
1920 		else if (err < 0)
1921 			vectors = 0; /* Nasty failure, quit now */
1922 		else /* err == number of vectors we should try again with */
1923 			vectors = err;
1924 	}
1925 
1926 	if (vectors < vector_threshold) {
1927 		/* Can't allocate enough MSI-X interrupts?  Oh well.
1928 		 * This just means we'll go with either a single MSI
1929 		 * vector or fall back to legacy interrupts.
1930 		 */
1931 		hw_dbg(&adapter->hw,
1932 		       "Unable to allocate MSI-X interrupts\n");
1933 		kfree(adapter->msix_entries);
1934 		adapter->msix_entries = NULL;
1935 	} else {
1936 		/*
1937 		 * Adjust for only the vectors we'll use, which is minimum
1938 		 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1939 		 * vectors we were allocated.
1940 		 */
1941 		adapter->num_msix_vectors = vectors;
1942 	}
1943 }
1944 
1945 /*
1946  * ixgbevf_set_num_queues: Allocate queues for device, feature dependent
1947  * @adapter: board private structure to initialize
1948  *
1949  * This is the top level queue allocation routine.  The order here is very
1950  * important, starting with the "most" number of features turned on at once,
1951  * and ending with the smallest set of features.  This way large combinations
1952  * can be allocated if they're turned on, and smaller combinations are the
1953  * fallthrough conditions.
1954  *
1955  **/
1956 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1957 {
1958 	/* Start with base case */
1959 	adapter->num_rx_queues = 1;
1960 	adapter->num_tx_queues = 1;
1961 	adapter->num_rx_pools = adapter->num_rx_queues;
1962 	adapter->num_rx_queues_per_pool = 1;
1963 }
1964 
1965 /**
1966  * ixgbevf_alloc_queues - Allocate memory for all rings
1967  * @adapter: board private structure to initialize
1968  *
1969  * We allocate one ring per queue at run-time since we don't know the
1970  * number of queues at compile-time.  The polling_netdev array is
1971  * intended for Multiqueue, but should work fine with a single queue.
1972  **/
1973 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1974 {
1975 	int i;
1976 
1977 	adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1978 				   sizeof(struct ixgbevf_ring), GFP_KERNEL);
1979 	if (!adapter->tx_ring)
1980 		goto err_tx_ring_allocation;
1981 
1982 	adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1983 				   sizeof(struct ixgbevf_ring), GFP_KERNEL);
1984 	if (!adapter->rx_ring)
1985 		goto err_rx_ring_allocation;
1986 
1987 	for (i = 0; i < adapter->num_tx_queues; i++) {
1988 		adapter->tx_ring[i].count = adapter->tx_ring_count;
1989 		adapter->tx_ring[i].queue_index = i;
1990 		adapter->tx_ring[i].reg_idx = i;
1991 	}
1992 
1993 	for (i = 0; i < adapter->num_rx_queues; i++) {
1994 		adapter->rx_ring[i].count = adapter->rx_ring_count;
1995 		adapter->rx_ring[i].queue_index = i;
1996 		adapter->rx_ring[i].reg_idx = i;
1997 	}
1998 
1999 	return 0;
2000 
2001 err_rx_ring_allocation:
2002 	kfree(adapter->tx_ring);
2003 err_tx_ring_allocation:
2004 	return -ENOMEM;
2005 }
2006 
2007 /**
2008  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2009  * @adapter: board private structure to initialize
2010  *
2011  * Attempt to configure the interrupts using the best available
2012  * capabilities of the hardware and the kernel.
2013  **/
2014 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2015 {
2016 	int err = 0;
2017 	int vector, v_budget;
2018 
2019 	/*
2020 	 * It's easy to be greedy for MSI-X vectors, but it really
2021 	 * doesn't do us much good if we have a lot more vectors
2022 	 * than CPU's.  So let's be conservative and only ask for
2023 	 * (roughly) twice the number of vectors as there are CPU's.
2024 	 */
2025 	v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2026 		       (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2027 
2028 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2029 	 * mean we disable MSI-X capabilities of the adapter. */
2030 	adapter->msix_entries = kcalloc(v_budget,
2031 					sizeof(struct msix_entry), GFP_KERNEL);
2032 	if (!adapter->msix_entries) {
2033 		err = -ENOMEM;
2034 		goto out;
2035 	}
2036 
2037 	for (vector = 0; vector < v_budget; vector++)
2038 		adapter->msix_entries[vector].entry = vector;
2039 
2040 	ixgbevf_acquire_msix_vectors(adapter, v_budget);
2041 
2042 out:
2043 	return err;
2044 }
2045 
2046 /**
2047  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2048  * @adapter: board private structure to initialize
2049  *
2050  * We allocate one q_vector per queue interrupt.  If allocation fails we
2051  * return -ENOMEM.
2052  **/
2053 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2054 {
2055 	int q_idx, num_q_vectors;
2056 	struct ixgbevf_q_vector *q_vector;
2057 	int napi_vectors;
2058 	int (*poll)(struct napi_struct *, int);
2059 
2060 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2061 	napi_vectors = adapter->num_rx_queues;
2062 	poll = &ixgbevf_clean_rxonly;
2063 
2064 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2065 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2066 		if (!q_vector)
2067 			goto err_out;
2068 		q_vector->adapter = adapter;
2069 		q_vector->v_idx = q_idx;
2070 		q_vector->eitr = adapter->eitr_param;
2071 		if (q_idx < napi_vectors)
2072 			netif_napi_add(adapter->netdev, &q_vector->napi,
2073 				       (*poll), 64);
2074 		adapter->q_vector[q_idx] = q_vector;
2075 	}
2076 
2077 	return 0;
2078 
2079 err_out:
2080 	while (q_idx) {
2081 		q_idx--;
2082 		q_vector = adapter->q_vector[q_idx];
2083 		netif_napi_del(&q_vector->napi);
2084 		kfree(q_vector);
2085 		adapter->q_vector[q_idx] = NULL;
2086 	}
2087 	return -ENOMEM;
2088 }
2089 
2090 /**
2091  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2092  * @adapter: board private structure to initialize
2093  *
2094  * This function frees the memory allocated to the q_vectors.  In addition if
2095  * NAPI is enabled it will delete any references to the NAPI struct prior
2096  * to freeing the q_vector.
2097  **/
2098 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2099 {
2100 	int q_idx, num_q_vectors;
2101 	int napi_vectors;
2102 
2103 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2104 	napi_vectors = adapter->num_rx_queues;
2105 
2106 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2107 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2108 
2109 		adapter->q_vector[q_idx] = NULL;
2110 		if (q_idx < napi_vectors)
2111 			netif_napi_del(&q_vector->napi);
2112 		kfree(q_vector);
2113 	}
2114 }
2115 
2116 /**
2117  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2118  * @adapter: board private structure
2119  *
2120  **/
2121 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2122 {
2123 	pci_disable_msix(adapter->pdev);
2124 	kfree(adapter->msix_entries);
2125 	adapter->msix_entries = NULL;
2126 }
2127 
2128 /**
2129  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2130  * @adapter: board private structure to initialize
2131  *
2132  **/
2133 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2134 {
2135 	int err;
2136 
2137 	/* Number of supported queues */
2138 	ixgbevf_set_num_queues(adapter);
2139 
2140 	err = ixgbevf_set_interrupt_capability(adapter);
2141 	if (err) {
2142 		hw_dbg(&adapter->hw,
2143 		       "Unable to setup interrupt capabilities\n");
2144 		goto err_set_interrupt;
2145 	}
2146 
2147 	err = ixgbevf_alloc_q_vectors(adapter);
2148 	if (err) {
2149 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2150 		       "vectors\n");
2151 		goto err_alloc_q_vectors;
2152 	}
2153 
2154 	err = ixgbevf_alloc_queues(adapter);
2155 	if (err) {
2156 		pr_err("Unable to allocate memory for queues\n");
2157 		goto err_alloc_queues;
2158 	}
2159 
2160 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2161 	       "Tx Queue count = %u\n",
2162 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2163 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2164 
2165 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2166 
2167 	return 0;
2168 err_alloc_queues:
2169 	ixgbevf_free_q_vectors(adapter);
2170 err_alloc_q_vectors:
2171 	ixgbevf_reset_interrupt_capability(adapter);
2172 err_set_interrupt:
2173 	return err;
2174 }
2175 
2176 /**
2177  * ixgbevf_sw_init - Initialize general software structures
2178  * (struct ixgbevf_adapter)
2179  * @adapter: board private structure to initialize
2180  *
2181  * ixgbevf_sw_init initializes the Adapter private data structure.
2182  * Fields are initialized based on PCI device information and
2183  * OS network device settings (MTU size).
2184  **/
2185 static int __devinit ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2186 {
2187 	struct ixgbe_hw *hw = &adapter->hw;
2188 	struct pci_dev *pdev = adapter->pdev;
2189 	int err;
2190 
2191 	/* PCI config space info */
2192 
2193 	hw->vendor_id = pdev->vendor;
2194 	hw->device_id = pdev->device;
2195 	hw->revision_id = pdev->revision;
2196 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2197 	hw->subsystem_device_id = pdev->subsystem_device;
2198 
2199 	hw->mbx.ops.init_params(hw);
2200 	hw->mac.max_tx_queues = MAX_TX_QUEUES;
2201 	hw->mac.max_rx_queues = MAX_RX_QUEUES;
2202 	err = hw->mac.ops.reset_hw(hw);
2203 	if (err) {
2204 		dev_info(&pdev->dev,
2205 		         "PF still in reset state, assigning new address\n");
2206 		eth_hw_addr_random(adapter->netdev);
2207 		memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
2208 			adapter->netdev->addr_len);
2209 	} else {
2210 		err = hw->mac.ops.init_hw(hw);
2211 		if (err) {
2212 			pr_err("init_shared_code failed: %d\n", err);
2213 			goto out;
2214 		}
2215 		memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
2216 			adapter->netdev->addr_len);
2217 	}
2218 
2219 	/* Enable dynamic interrupt throttling rates */
2220 	adapter->eitr_param = 20000;
2221 	adapter->itr_setting = 1;
2222 
2223 	/* set defaults for eitr in MegaBytes */
2224 	adapter->eitr_low = 10;
2225 	adapter->eitr_high = 20;
2226 
2227 	/* set default ring sizes */
2228 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2229 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2230 
2231 	/* enable rx csum by default */
2232 	adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2233 
2234 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2235 	return 0;
2236 
2237 out:
2238 	return err;
2239 }
2240 
2241 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2242 	{							\
2243 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2244 		if (current_counter < last_counter)		\
2245 			counter += 0x100000000LL;		\
2246 		last_counter = current_counter;			\
2247 		counter &= 0xFFFFFFFF00000000LL;		\
2248 		counter |= current_counter;			\
2249 	}
2250 
2251 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2252 	{								 \
2253 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2254 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2255 		u64 current_counter = (current_counter_msb << 32) |      \
2256 			current_counter_lsb;                             \
2257 		if (current_counter < last_counter)			 \
2258 			counter += 0x1000000000LL;			 \
2259 		last_counter = current_counter;				 \
2260 		counter &= 0xFFFFFFF000000000LL;			 \
2261 		counter |= current_counter;				 \
2262 	}
2263 /**
2264  * ixgbevf_update_stats - Update the board statistics counters.
2265  * @adapter: board private structure
2266  **/
2267 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2268 {
2269 	struct ixgbe_hw *hw = &adapter->hw;
2270 
2271 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2272 				adapter->stats.vfgprc);
2273 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2274 				adapter->stats.vfgptc);
2275 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2276 				adapter->stats.last_vfgorc,
2277 				adapter->stats.vfgorc);
2278 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2279 				adapter->stats.last_vfgotc,
2280 				adapter->stats.vfgotc);
2281 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2282 				adapter->stats.vfmprc);
2283 }
2284 
2285 /**
2286  * ixgbevf_watchdog - Timer Call-back
2287  * @data: pointer to adapter cast into an unsigned long
2288  **/
2289 static void ixgbevf_watchdog(unsigned long data)
2290 {
2291 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2292 	struct ixgbe_hw *hw = &adapter->hw;
2293 	u64 eics = 0;
2294 	int i;
2295 
2296 	/*
2297 	 * Do the watchdog outside of interrupt context due to the lovely
2298 	 * delays that some of the newer hardware requires
2299 	 */
2300 
2301 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2302 		goto watchdog_short_circuit;
2303 
2304 	/* get one bit for every active tx/rx interrupt vector */
2305 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2306 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2307 		if (qv->rxr_count || qv->txr_count)
2308 			eics |= (1 << i);
2309 	}
2310 
2311 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, (u32)eics);
2312 
2313 watchdog_short_circuit:
2314 	schedule_work(&adapter->watchdog_task);
2315 }
2316 
2317 /**
2318  * ixgbevf_tx_timeout - Respond to a Tx Hang
2319  * @netdev: network interface device structure
2320  **/
2321 static void ixgbevf_tx_timeout(struct net_device *netdev)
2322 {
2323 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2324 
2325 	/* Do the reset outside of interrupt context */
2326 	schedule_work(&adapter->reset_task);
2327 }
2328 
2329 static void ixgbevf_reset_task(struct work_struct *work)
2330 {
2331 	struct ixgbevf_adapter *adapter;
2332 	adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2333 
2334 	/* If we're already down or resetting, just bail */
2335 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2336 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2337 		return;
2338 
2339 	adapter->tx_timeout_count++;
2340 
2341 	ixgbevf_reinit_locked(adapter);
2342 }
2343 
2344 /**
2345  * ixgbevf_watchdog_task - worker thread to bring link up
2346  * @work: pointer to work_struct containing our data
2347  **/
2348 static void ixgbevf_watchdog_task(struct work_struct *work)
2349 {
2350 	struct ixgbevf_adapter *adapter = container_of(work,
2351 						       struct ixgbevf_adapter,
2352 						       watchdog_task);
2353 	struct net_device *netdev = adapter->netdev;
2354 	struct ixgbe_hw *hw = &adapter->hw;
2355 	u32 link_speed = adapter->link_speed;
2356 	bool link_up = adapter->link_up;
2357 
2358 	adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2359 
2360 	/*
2361 	 * Always check the link on the watchdog because we have
2362 	 * no LSC interrupt
2363 	 */
2364 	if (hw->mac.ops.check_link) {
2365 		if ((hw->mac.ops.check_link(hw, &link_speed,
2366 					    &link_up, false)) != 0) {
2367 			adapter->link_up = link_up;
2368 			adapter->link_speed = link_speed;
2369 			netif_carrier_off(netdev);
2370 			netif_tx_stop_all_queues(netdev);
2371 			schedule_work(&adapter->reset_task);
2372 			goto pf_has_reset;
2373 		}
2374 	} else {
2375 		/* always assume link is up, if no check link
2376 		 * function */
2377 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
2378 		link_up = true;
2379 	}
2380 	adapter->link_up = link_up;
2381 	adapter->link_speed = link_speed;
2382 
2383 	if (link_up) {
2384 		if (!netif_carrier_ok(netdev)) {
2385 			hw_dbg(&adapter->hw, "NIC Link is Up, %u Gbps\n",
2386 			       (link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2387 			       10 : 1);
2388 			netif_carrier_on(netdev);
2389 			netif_tx_wake_all_queues(netdev);
2390 		}
2391 	} else {
2392 		adapter->link_up = false;
2393 		adapter->link_speed = 0;
2394 		if (netif_carrier_ok(netdev)) {
2395 			hw_dbg(&adapter->hw, "NIC Link is Down\n");
2396 			netif_carrier_off(netdev);
2397 			netif_tx_stop_all_queues(netdev);
2398 		}
2399 	}
2400 
2401 	ixgbevf_update_stats(adapter);
2402 
2403 pf_has_reset:
2404 	/* Reset the timer */
2405 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2406 		mod_timer(&adapter->watchdog_timer,
2407 			  round_jiffies(jiffies + (2 * HZ)));
2408 
2409 	adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2410 }
2411 
2412 /**
2413  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2414  * @adapter: board private structure
2415  * @tx_ring: Tx descriptor ring for a specific queue
2416  *
2417  * Free all transmit software resources
2418  **/
2419 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2420 			       struct ixgbevf_ring *tx_ring)
2421 {
2422 	struct pci_dev *pdev = adapter->pdev;
2423 
2424 	ixgbevf_clean_tx_ring(adapter, tx_ring);
2425 
2426 	vfree(tx_ring->tx_buffer_info);
2427 	tx_ring->tx_buffer_info = NULL;
2428 
2429 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2430 			  tx_ring->dma);
2431 
2432 	tx_ring->desc = NULL;
2433 }
2434 
2435 /**
2436  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2437  * @adapter: board private structure
2438  *
2439  * Free all transmit software resources
2440  **/
2441 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2442 {
2443 	int i;
2444 
2445 	for (i = 0; i < adapter->num_tx_queues; i++)
2446 		if (adapter->tx_ring[i].desc)
2447 			ixgbevf_free_tx_resources(adapter,
2448 						  &adapter->tx_ring[i]);
2449 
2450 }
2451 
2452 /**
2453  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2454  * @adapter: board private structure
2455  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2456  *
2457  * Return 0 on success, negative on failure
2458  **/
2459 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2460 			       struct ixgbevf_ring *tx_ring)
2461 {
2462 	struct pci_dev *pdev = adapter->pdev;
2463 	int size;
2464 
2465 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2466 	tx_ring->tx_buffer_info = vzalloc(size);
2467 	if (!tx_ring->tx_buffer_info)
2468 		goto err;
2469 
2470 	/* round up to nearest 4K */
2471 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2472 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2473 
2474 	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2475 					   &tx_ring->dma, GFP_KERNEL);
2476 	if (!tx_ring->desc)
2477 		goto err;
2478 
2479 	tx_ring->next_to_use = 0;
2480 	tx_ring->next_to_clean = 0;
2481 	tx_ring->work_limit = tx_ring->count;
2482 	return 0;
2483 
2484 err:
2485 	vfree(tx_ring->tx_buffer_info);
2486 	tx_ring->tx_buffer_info = NULL;
2487 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2488 	       "descriptor ring\n");
2489 	return -ENOMEM;
2490 }
2491 
2492 /**
2493  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2494  * @adapter: board private structure
2495  *
2496  * If this function returns with an error, then it's possible one or
2497  * more of the rings is populated (while the rest are not).  It is the
2498  * callers duty to clean those orphaned rings.
2499  *
2500  * Return 0 on success, negative on failure
2501  **/
2502 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2503 {
2504 	int i, err = 0;
2505 
2506 	for (i = 0; i < adapter->num_tx_queues; i++) {
2507 		err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2508 		if (!err)
2509 			continue;
2510 		hw_dbg(&adapter->hw,
2511 		       "Allocation for Tx Queue %u failed\n", i);
2512 		break;
2513 	}
2514 
2515 	return err;
2516 }
2517 
2518 /**
2519  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2520  * @adapter: board private structure
2521  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2522  *
2523  * Returns 0 on success, negative on failure
2524  **/
2525 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2526 			       struct ixgbevf_ring *rx_ring)
2527 {
2528 	struct pci_dev *pdev = adapter->pdev;
2529 	int size;
2530 
2531 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2532 	rx_ring->rx_buffer_info = vzalloc(size);
2533 	if (!rx_ring->rx_buffer_info)
2534 		goto alloc_failed;
2535 
2536 	/* Round up to nearest 4K */
2537 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2538 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2539 
2540 	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2541 					   &rx_ring->dma, GFP_KERNEL);
2542 
2543 	if (!rx_ring->desc) {
2544 		hw_dbg(&adapter->hw,
2545 		       "Unable to allocate memory for "
2546 		       "the receive descriptor ring\n");
2547 		vfree(rx_ring->rx_buffer_info);
2548 		rx_ring->rx_buffer_info = NULL;
2549 		goto alloc_failed;
2550 	}
2551 
2552 	rx_ring->next_to_clean = 0;
2553 	rx_ring->next_to_use = 0;
2554 
2555 	return 0;
2556 alloc_failed:
2557 	return -ENOMEM;
2558 }
2559 
2560 /**
2561  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2562  * @adapter: board private structure
2563  *
2564  * If this function returns with an error, then it's possible one or
2565  * more of the rings is populated (while the rest are not).  It is the
2566  * callers duty to clean those orphaned rings.
2567  *
2568  * Return 0 on success, negative on failure
2569  **/
2570 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2571 {
2572 	int i, err = 0;
2573 
2574 	for (i = 0; i < adapter->num_rx_queues; i++) {
2575 		err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2576 		if (!err)
2577 			continue;
2578 		hw_dbg(&adapter->hw,
2579 		       "Allocation for Rx Queue %u failed\n", i);
2580 		break;
2581 	}
2582 	return err;
2583 }
2584 
2585 /**
2586  * ixgbevf_free_rx_resources - Free Rx Resources
2587  * @adapter: board private structure
2588  * @rx_ring: ring to clean the resources from
2589  *
2590  * Free all receive software resources
2591  **/
2592 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2593 			       struct ixgbevf_ring *rx_ring)
2594 {
2595 	struct pci_dev *pdev = adapter->pdev;
2596 
2597 	ixgbevf_clean_rx_ring(adapter, rx_ring);
2598 
2599 	vfree(rx_ring->rx_buffer_info);
2600 	rx_ring->rx_buffer_info = NULL;
2601 
2602 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2603 			  rx_ring->dma);
2604 
2605 	rx_ring->desc = NULL;
2606 }
2607 
2608 /**
2609  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2610  * @adapter: board private structure
2611  *
2612  * Free all receive software resources
2613  **/
2614 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2615 {
2616 	int i;
2617 
2618 	for (i = 0; i < adapter->num_rx_queues; i++)
2619 		if (adapter->rx_ring[i].desc)
2620 			ixgbevf_free_rx_resources(adapter,
2621 						  &adapter->rx_ring[i]);
2622 }
2623 
2624 /**
2625  * ixgbevf_open - Called when a network interface is made active
2626  * @netdev: network interface device structure
2627  *
2628  * Returns 0 on success, negative value on failure
2629  *
2630  * The open entry point is called when a network interface is made
2631  * active by the system (IFF_UP).  At this point all resources needed
2632  * for transmit and receive operations are allocated, the interrupt
2633  * handler is registered with the OS, the watchdog timer is started,
2634  * and the stack is notified that the interface is ready.
2635  **/
2636 static int ixgbevf_open(struct net_device *netdev)
2637 {
2638 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2639 	struct ixgbe_hw *hw = &adapter->hw;
2640 	int err;
2641 
2642 	/* disallow open during test */
2643 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2644 		return -EBUSY;
2645 
2646 	if (hw->adapter_stopped) {
2647 		ixgbevf_reset(adapter);
2648 		/* if adapter is still stopped then PF isn't up and
2649 		 * the vf can't start. */
2650 		if (hw->adapter_stopped) {
2651 			err = IXGBE_ERR_MBX;
2652 			pr_err("Unable to start - perhaps the PF Driver isn't "
2653 			       "up yet\n");
2654 			goto err_setup_reset;
2655 		}
2656 	}
2657 
2658 	/* allocate transmit descriptors */
2659 	err = ixgbevf_setup_all_tx_resources(adapter);
2660 	if (err)
2661 		goto err_setup_tx;
2662 
2663 	/* allocate receive descriptors */
2664 	err = ixgbevf_setup_all_rx_resources(adapter);
2665 	if (err)
2666 		goto err_setup_rx;
2667 
2668 	ixgbevf_configure(adapter);
2669 
2670 	/*
2671 	 * Map the Tx/Rx rings to the vectors we were allotted.
2672 	 * if request_irq will be called in this function map_rings
2673 	 * must be called *before* up_complete
2674 	 */
2675 	ixgbevf_map_rings_to_vectors(adapter);
2676 
2677 	ixgbevf_up_complete(adapter);
2678 
2679 	/* clear any pending interrupts, may auto mask */
2680 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2681 	err = ixgbevf_request_irq(adapter);
2682 	if (err)
2683 		goto err_req_irq;
2684 
2685 	ixgbevf_irq_enable(adapter, true, true);
2686 
2687 	return 0;
2688 
2689 err_req_irq:
2690 	ixgbevf_down(adapter);
2691 	ixgbevf_free_irq(adapter);
2692 err_setup_rx:
2693 	ixgbevf_free_all_rx_resources(adapter);
2694 err_setup_tx:
2695 	ixgbevf_free_all_tx_resources(adapter);
2696 	ixgbevf_reset(adapter);
2697 
2698 err_setup_reset:
2699 
2700 	return err;
2701 }
2702 
2703 /**
2704  * ixgbevf_close - Disables a network interface
2705  * @netdev: network interface device structure
2706  *
2707  * Returns 0, this is not allowed to fail
2708  *
2709  * The close entry point is called when an interface is de-activated
2710  * by the OS.  The hardware is still under the drivers control, but
2711  * needs to be disabled.  A global MAC reset is issued to stop the
2712  * hardware, and all transmit and receive resources are freed.
2713  **/
2714 static int ixgbevf_close(struct net_device *netdev)
2715 {
2716 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2717 
2718 	ixgbevf_down(adapter);
2719 	ixgbevf_free_irq(adapter);
2720 
2721 	ixgbevf_free_all_tx_resources(adapter);
2722 	ixgbevf_free_all_rx_resources(adapter);
2723 
2724 	return 0;
2725 }
2726 
2727 static int ixgbevf_tso(struct ixgbevf_adapter *adapter,
2728 		       struct ixgbevf_ring *tx_ring,
2729 		       struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2730 {
2731 	struct ixgbe_adv_tx_context_desc *context_desc;
2732 	unsigned int i;
2733 	int err;
2734 	struct ixgbevf_tx_buffer *tx_buffer_info;
2735 	u32 vlan_macip_lens = 0, type_tucmd_mlhl;
2736 	u32 mss_l4len_idx, l4len;
2737 
2738 	if (skb_is_gso(skb)) {
2739 		if (skb_header_cloned(skb)) {
2740 			err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2741 			if (err)
2742 				return err;
2743 		}
2744 		l4len = tcp_hdrlen(skb);
2745 		*hdr_len += l4len;
2746 
2747 		if (skb->protocol == htons(ETH_P_IP)) {
2748 			struct iphdr *iph = ip_hdr(skb);
2749 			iph->tot_len = 0;
2750 			iph->check = 0;
2751 			tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2752 								 iph->daddr, 0,
2753 								 IPPROTO_TCP,
2754 								 0);
2755 			adapter->hw_tso_ctxt++;
2756 		} else if (skb_is_gso_v6(skb)) {
2757 			ipv6_hdr(skb)->payload_len = 0;
2758 			tcp_hdr(skb)->check =
2759 			    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2760 					     &ipv6_hdr(skb)->daddr,
2761 					     0, IPPROTO_TCP, 0);
2762 			adapter->hw_tso6_ctxt++;
2763 		}
2764 
2765 		i = tx_ring->next_to_use;
2766 
2767 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2768 		context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2769 
2770 		/* VLAN MACLEN IPLEN */
2771 		if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2772 			vlan_macip_lens |=
2773 				(tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
2774 		vlan_macip_lens |= ((skb_network_offset(skb)) <<
2775 				    IXGBE_ADVTXD_MACLEN_SHIFT);
2776 		*hdr_len += skb_network_offset(skb);
2777 		vlan_macip_lens |=
2778 			(skb_transport_header(skb) - skb_network_header(skb));
2779 		*hdr_len +=
2780 			(skb_transport_header(skb) - skb_network_header(skb));
2781 		context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2782 		context_desc->seqnum_seed = 0;
2783 
2784 		/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2785 		type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
2786 				    IXGBE_ADVTXD_DTYP_CTXT);
2787 
2788 		if (skb->protocol == htons(ETH_P_IP))
2789 			type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2790 		type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2791 		context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2792 
2793 		/* MSS L4LEN IDX */
2794 		mss_l4len_idx =
2795 			(skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
2796 		mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
2797 		/* use index 1 for TSO */
2798 		mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2799 		context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2800 
2801 		tx_buffer_info->time_stamp = jiffies;
2802 		tx_buffer_info->next_to_watch = i;
2803 
2804 		i++;
2805 		if (i == tx_ring->count)
2806 			i = 0;
2807 		tx_ring->next_to_use = i;
2808 
2809 		return true;
2810 	}
2811 
2812 	return false;
2813 }
2814 
2815 static bool ixgbevf_tx_csum(struct ixgbevf_adapter *adapter,
2816 			    struct ixgbevf_ring *tx_ring,
2817 			    struct sk_buff *skb, u32 tx_flags)
2818 {
2819 	struct ixgbe_adv_tx_context_desc *context_desc;
2820 	unsigned int i;
2821 	struct ixgbevf_tx_buffer *tx_buffer_info;
2822 	u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
2823 
2824 	if (skb->ip_summed == CHECKSUM_PARTIAL ||
2825 	    (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
2826 		i = tx_ring->next_to_use;
2827 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2828 		context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2829 
2830 		if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2831 			vlan_macip_lens |= (tx_flags &
2832 					    IXGBE_TX_FLAGS_VLAN_MASK);
2833 		vlan_macip_lens |= (skb_network_offset(skb) <<
2834 				    IXGBE_ADVTXD_MACLEN_SHIFT);
2835 		if (skb->ip_summed == CHECKSUM_PARTIAL)
2836 			vlan_macip_lens |= (skb_transport_header(skb) -
2837 					    skb_network_header(skb));
2838 
2839 		context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2840 		context_desc->seqnum_seed = 0;
2841 
2842 		type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
2843 				    IXGBE_ADVTXD_DTYP_CTXT);
2844 
2845 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2846 			switch (skb->protocol) {
2847 			case __constant_htons(ETH_P_IP):
2848 				type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2849 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2850 					type_tucmd_mlhl |=
2851 					    IXGBE_ADVTXD_TUCMD_L4T_TCP;
2852 				break;
2853 			case __constant_htons(ETH_P_IPV6):
2854 				/* XXX what about other V6 headers?? */
2855 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2856 					type_tucmd_mlhl |=
2857 						IXGBE_ADVTXD_TUCMD_L4T_TCP;
2858 				break;
2859 			default:
2860 				if (unlikely(net_ratelimit())) {
2861 					pr_warn("partial checksum but "
2862 						"proto=%x!\n", skb->protocol);
2863 				}
2864 				break;
2865 			}
2866 		}
2867 
2868 		context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2869 		/* use index zero for tx checksum offload */
2870 		context_desc->mss_l4len_idx = 0;
2871 
2872 		tx_buffer_info->time_stamp = jiffies;
2873 		tx_buffer_info->next_to_watch = i;
2874 
2875 		adapter->hw_csum_tx_good++;
2876 		i++;
2877 		if (i == tx_ring->count)
2878 			i = 0;
2879 		tx_ring->next_to_use = i;
2880 
2881 		return true;
2882 	}
2883 
2884 	return false;
2885 }
2886 
2887 static int ixgbevf_tx_map(struct ixgbevf_adapter *adapter,
2888 			  struct ixgbevf_ring *tx_ring,
2889 			  struct sk_buff *skb, u32 tx_flags,
2890 			  unsigned int first)
2891 {
2892 	struct pci_dev *pdev = adapter->pdev;
2893 	struct ixgbevf_tx_buffer *tx_buffer_info;
2894 	unsigned int len;
2895 	unsigned int total = skb->len;
2896 	unsigned int offset = 0, size;
2897 	int count = 0;
2898 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2899 	unsigned int f;
2900 	int i;
2901 
2902 	i = tx_ring->next_to_use;
2903 
2904 	len = min(skb_headlen(skb), total);
2905 	while (len) {
2906 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2907 		size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2908 
2909 		tx_buffer_info->length = size;
2910 		tx_buffer_info->mapped_as_page = false;
2911 		tx_buffer_info->dma = dma_map_single(&adapter->pdev->dev,
2912 						     skb->data + offset,
2913 						     size, DMA_TO_DEVICE);
2914 		if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2915 			goto dma_error;
2916 		tx_buffer_info->time_stamp = jiffies;
2917 		tx_buffer_info->next_to_watch = i;
2918 
2919 		len -= size;
2920 		total -= size;
2921 		offset += size;
2922 		count++;
2923 		i++;
2924 		if (i == tx_ring->count)
2925 			i = 0;
2926 	}
2927 
2928 	for (f = 0; f < nr_frags; f++) {
2929 		const struct skb_frag_struct *frag;
2930 
2931 		frag = &skb_shinfo(skb)->frags[f];
2932 		len = min((unsigned int)skb_frag_size(frag), total);
2933 		offset = 0;
2934 
2935 		while (len) {
2936 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
2937 			size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2938 
2939 			tx_buffer_info->length = size;
2940 			tx_buffer_info->dma =
2941 				skb_frag_dma_map(&adapter->pdev->dev, frag,
2942 						 offset, size, DMA_TO_DEVICE);
2943 			tx_buffer_info->mapped_as_page = true;
2944 			if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2945 				goto dma_error;
2946 			tx_buffer_info->time_stamp = jiffies;
2947 			tx_buffer_info->next_to_watch = i;
2948 
2949 			len -= size;
2950 			total -= size;
2951 			offset += size;
2952 			count++;
2953 			i++;
2954 			if (i == tx_ring->count)
2955 				i = 0;
2956 		}
2957 		if (total == 0)
2958 			break;
2959 	}
2960 
2961 	if (i == 0)
2962 		i = tx_ring->count - 1;
2963 	else
2964 		i = i - 1;
2965 	tx_ring->tx_buffer_info[i].skb = skb;
2966 	tx_ring->tx_buffer_info[first].next_to_watch = i;
2967 
2968 	return count;
2969 
2970 dma_error:
2971 	dev_err(&pdev->dev, "TX DMA map failed\n");
2972 
2973 	/* clear timestamp and dma mappings for failed tx_buffer_info map */
2974 	tx_buffer_info->dma = 0;
2975 	tx_buffer_info->time_stamp = 0;
2976 	tx_buffer_info->next_to_watch = 0;
2977 	count--;
2978 
2979 	/* clear timestamp and dma mappings for remaining portion of packet */
2980 	while (count >= 0) {
2981 		count--;
2982 		i--;
2983 		if (i < 0)
2984 			i += tx_ring->count;
2985 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2986 		ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2987 	}
2988 
2989 	return count;
2990 }
2991 
2992 static void ixgbevf_tx_queue(struct ixgbevf_adapter *adapter,
2993 			     struct ixgbevf_ring *tx_ring, int tx_flags,
2994 			     int count, u32 paylen, u8 hdr_len)
2995 {
2996 	union ixgbe_adv_tx_desc *tx_desc = NULL;
2997 	struct ixgbevf_tx_buffer *tx_buffer_info;
2998 	u32 olinfo_status = 0, cmd_type_len = 0;
2999 	unsigned int i;
3000 
3001 	u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3002 
3003 	cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3004 
3005 	cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3006 
3007 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3008 		cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3009 
3010 	if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3011 		cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3012 
3013 		olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3014 			IXGBE_ADVTXD_POPTS_SHIFT;
3015 
3016 		/* use index 1 context for tso */
3017 		olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3018 		if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3019 			olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3020 				IXGBE_ADVTXD_POPTS_SHIFT;
3021 
3022 	} else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3023 		olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3024 			IXGBE_ADVTXD_POPTS_SHIFT;
3025 
3026 	olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3027 
3028 	i = tx_ring->next_to_use;
3029 	while (count--) {
3030 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
3031 		tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3032 		tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3033 		tx_desc->read.cmd_type_len =
3034 			cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3035 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3036 		i++;
3037 		if (i == tx_ring->count)
3038 			i = 0;
3039 	}
3040 
3041 	tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3042 
3043 	/*
3044 	 * Force memory writes to complete before letting h/w
3045 	 * know there are new descriptors to fetch.  (Only
3046 	 * applicable for weak-ordered memory model archs,
3047 	 * such as IA-64).
3048 	 */
3049 	wmb();
3050 
3051 	tx_ring->next_to_use = i;
3052 	writel(i, adapter->hw.hw_addr + tx_ring->tail);
3053 }
3054 
3055 static int __ixgbevf_maybe_stop_tx(struct net_device *netdev,
3056 				   struct ixgbevf_ring *tx_ring, int size)
3057 {
3058 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3059 
3060 	netif_stop_subqueue(netdev, tx_ring->queue_index);
3061 	/* Herbert's original patch had:
3062 	 *  smp_mb__after_netif_stop_queue();
3063 	 * but since that doesn't exist yet, just open code it. */
3064 	smp_mb();
3065 
3066 	/* We need to check again in a case another CPU has just
3067 	 * made room available. */
3068 	if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3069 		return -EBUSY;
3070 
3071 	/* A reprieve! - use start_queue because it doesn't call schedule */
3072 	netif_start_subqueue(netdev, tx_ring->queue_index);
3073 	++adapter->restart_queue;
3074 	return 0;
3075 }
3076 
3077 static int ixgbevf_maybe_stop_tx(struct net_device *netdev,
3078 				 struct ixgbevf_ring *tx_ring, int size)
3079 {
3080 	if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3081 		return 0;
3082 	return __ixgbevf_maybe_stop_tx(netdev, tx_ring, size);
3083 }
3084 
3085 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3086 {
3087 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3088 	struct ixgbevf_ring *tx_ring;
3089 	unsigned int first;
3090 	unsigned int tx_flags = 0;
3091 	u8 hdr_len = 0;
3092 	int r_idx = 0, tso;
3093 	int count = 0;
3094 
3095 	unsigned int f;
3096 
3097 	tx_ring = &adapter->tx_ring[r_idx];
3098 
3099 	if (vlan_tx_tag_present(skb)) {
3100 		tx_flags |= vlan_tx_tag_get(skb);
3101 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3102 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3103 	}
3104 
3105 	/* four things can cause us to need a context descriptor */
3106 	if (skb_is_gso(skb) ||
3107 	    (skb->ip_summed == CHECKSUM_PARTIAL) ||
3108 	    (tx_flags & IXGBE_TX_FLAGS_VLAN))
3109 		count++;
3110 
3111 	count += TXD_USE_COUNT(skb_headlen(skb));
3112 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3113 		count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]));
3114 
3115 	if (ixgbevf_maybe_stop_tx(netdev, tx_ring, count)) {
3116 		adapter->tx_busy++;
3117 		return NETDEV_TX_BUSY;
3118 	}
3119 
3120 	first = tx_ring->next_to_use;
3121 
3122 	if (skb->protocol == htons(ETH_P_IP))
3123 		tx_flags |= IXGBE_TX_FLAGS_IPV4;
3124 	tso = ixgbevf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3125 	if (tso < 0) {
3126 		dev_kfree_skb_any(skb);
3127 		return NETDEV_TX_OK;
3128 	}
3129 
3130 	if (tso)
3131 		tx_flags |= IXGBE_TX_FLAGS_TSO;
3132 	else if (ixgbevf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3133 		 (skb->ip_summed == CHECKSUM_PARTIAL))
3134 		tx_flags |= IXGBE_TX_FLAGS_CSUM;
3135 
3136 	ixgbevf_tx_queue(adapter, tx_ring, tx_flags,
3137 			 ixgbevf_tx_map(adapter, tx_ring, skb, tx_flags, first),
3138 			 skb->len, hdr_len);
3139 
3140 	ixgbevf_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3141 
3142 	return NETDEV_TX_OK;
3143 }
3144 
3145 /**
3146  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3147  * @netdev: network interface device structure
3148  * @p: pointer to an address structure
3149  *
3150  * Returns 0 on success, negative on failure
3151  **/
3152 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3153 {
3154 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3155 	struct ixgbe_hw *hw = &adapter->hw;
3156 	struct sockaddr *addr = p;
3157 
3158 	if (!is_valid_ether_addr(addr->sa_data))
3159 		return -EADDRNOTAVAIL;
3160 
3161 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3162 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3163 
3164 	if (hw->mac.ops.set_rar)
3165 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3166 
3167 	return 0;
3168 }
3169 
3170 /**
3171  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3172  * @netdev: network interface device structure
3173  * @new_mtu: new value for maximum frame size
3174  *
3175  * Returns 0 on success, negative on failure
3176  **/
3177 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3178 {
3179 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3180 	struct ixgbe_hw *hw = &adapter->hw;
3181 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3182 	int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3183 	u32 msg[2];
3184 
3185 	if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3186 		max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3187 
3188 	/* MTU < 68 is an error and causes problems on some kernels */
3189 	if ((new_mtu < 68) || (max_frame > max_possible_frame))
3190 		return -EINVAL;
3191 
3192 	hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3193 	       netdev->mtu, new_mtu);
3194 	/* must set new MTU before calling down or up */
3195 	netdev->mtu = new_mtu;
3196 
3197 	if (!netif_running(netdev)) {
3198 		msg[0] = IXGBE_VF_SET_LPE;
3199 		msg[1] = max_frame;
3200 		hw->mbx.ops.write_posted(hw, msg, 2);
3201 	}
3202 
3203 	if (netif_running(netdev))
3204 		ixgbevf_reinit_locked(adapter);
3205 
3206 	return 0;
3207 }
3208 
3209 static void ixgbevf_shutdown(struct pci_dev *pdev)
3210 {
3211 	struct net_device *netdev = pci_get_drvdata(pdev);
3212 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3213 
3214 	netif_device_detach(netdev);
3215 
3216 	if (netif_running(netdev)) {
3217 		ixgbevf_down(adapter);
3218 		ixgbevf_free_irq(adapter);
3219 		ixgbevf_free_all_tx_resources(adapter);
3220 		ixgbevf_free_all_rx_resources(adapter);
3221 	}
3222 
3223 #ifdef CONFIG_PM
3224 	pci_save_state(pdev);
3225 #endif
3226 
3227 	pci_disable_device(pdev);
3228 }
3229 
3230 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3231 						struct rtnl_link_stats64 *stats)
3232 {
3233 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3234 	unsigned int start;
3235 	u64 bytes, packets;
3236 	const struct ixgbevf_ring *ring;
3237 	int i;
3238 
3239 	ixgbevf_update_stats(adapter);
3240 
3241 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3242 
3243 	for (i = 0; i < adapter->num_rx_queues; i++) {
3244 		ring = &adapter->rx_ring[i];
3245 		do {
3246 			start = u64_stats_fetch_begin_bh(&ring->syncp);
3247 			bytes = ring->total_bytes;
3248 			packets = ring->total_packets;
3249 		} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3250 		stats->rx_bytes += bytes;
3251 		stats->rx_packets += packets;
3252 	}
3253 
3254 	for (i = 0; i < adapter->num_tx_queues; i++) {
3255 		ring = &adapter->tx_ring[i];
3256 		do {
3257 			start = u64_stats_fetch_begin_bh(&ring->syncp);
3258 			bytes = ring->total_bytes;
3259 			packets = ring->total_packets;
3260 		} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3261 		stats->tx_bytes += bytes;
3262 		stats->tx_packets += packets;
3263 	}
3264 
3265 	return stats;
3266 }
3267 
3268 static int ixgbevf_set_features(struct net_device *netdev,
3269 	netdev_features_t features)
3270 {
3271 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3272 
3273 	if (features & NETIF_F_RXCSUM)
3274 		adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
3275 	else
3276 		adapter->flags &= ~IXGBE_FLAG_RX_CSUM_ENABLED;
3277 
3278 	return 0;
3279 }
3280 
3281 static const struct net_device_ops ixgbe_netdev_ops = {
3282 	.ndo_open		= ixgbevf_open,
3283 	.ndo_stop		= ixgbevf_close,
3284 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3285 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3286 	.ndo_get_stats64	= ixgbevf_get_stats,
3287 	.ndo_validate_addr	= eth_validate_addr,
3288 	.ndo_set_mac_address	= ixgbevf_set_mac,
3289 	.ndo_change_mtu		= ixgbevf_change_mtu,
3290 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3291 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3292 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3293 	.ndo_set_features	= ixgbevf_set_features,
3294 };
3295 
3296 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3297 {
3298 	dev->netdev_ops = &ixgbe_netdev_ops;
3299 	ixgbevf_set_ethtool_ops(dev);
3300 	dev->watchdog_timeo = 5 * HZ;
3301 }
3302 
3303 /**
3304  * ixgbevf_probe - Device Initialization Routine
3305  * @pdev: PCI device information struct
3306  * @ent: entry in ixgbevf_pci_tbl
3307  *
3308  * Returns 0 on success, negative on failure
3309  *
3310  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3311  * The OS initialization, configuring of the adapter private structure,
3312  * and a hardware reset occur.
3313  **/
3314 static int __devinit ixgbevf_probe(struct pci_dev *pdev,
3315 				   const struct pci_device_id *ent)
3316 {
3317 	struct net_device *netdev;
3318 	struct ixgbevf_adapter *adapter = NULL;
3319 	struct ixgbe_hw *hw = NULL;
3320 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3321 	static int cards_found;
3322 	int err, pci_using_dac;
3323 
3324 	err = pci_enable_device(pdev);
3325 	if (err)
3326 		return err;
3327 
3328 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3329 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3330 		pci_using_dac = 1;
3331 	} else {
3332 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3333 		if (err) {
3334 			err = dma_set_coherent_mask(&pdev->dev,
3335 						    DMA_BIT_MASK(32));
3336 			if (err) {
3337 				dev_err(&pdev->dev, "No usable DMA "
3338 					"configuration, aborting\n");
3339 				goto err_dma;
3340 			}
3341 		}
3342 		pci_using_dac = 0;
3343 	}
3344 
3345 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3346 	if (err) {
3347 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3348 		goto err_pci_reg;
3349 	}
3350 
3351 	pci_set_master(pdev);
3352 
3353 #ifdef HAVE_TX_MQ
3354 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3355 				   MAX_TX_QUEUES);
3356 #else
3357 	netdev = alloc_etherdev(sizeof(struct ixgbevf_adapter));
3358 #endif
3359 	if (!netdev) {
3360 		err = -ENOMEM;
3361 		goto err_alloc_etherdev;
3362 	}
3363 
3364 	SET_NETDEV_DEV(netdev, &pdev->dev);
3365 
3366 	pci_set_drvdata(pdev, netdev);
3367 	adapter = netdev_priv(netdev);
3368 
3369 	adapter->netdev = netdev;
3370 	adapter->pdev = pdev;
3371 	hw = &adapter->hw;
3372 	hw->back = adapter;
3373 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3374 
3375 	/*
3376 	 * call save state here in standalone driver because it relies on
3377 	 * adapter struct to exist, and needs to call netdev_priv
3378 	 */
3379 	pci_save_state(pdev);
3380 
3381 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3382 			      pci_resource_len(pdev, 0));
3383 	if (!hw->hw_addr) {
3384 		err = -EIO;
3385 		goto err_ioremap;
3386 	}
3387 
3388 	ixgbevf_assign_netdev_ops(netdev);
3389 
3390 	adapter->bd_number = cards_found;
3391 
3392 	/* Setup hw api */
3393 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3394 	hw->mac.type  = ii->mac;
3395 
3396 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3397 	       sizeof(struct ixgbe_mbx_operations));
3398 
3399 	adapter->flags &= ~IXGBE_FLAG_RX_PS_CAPABLE;
3400 	adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
3401 	adapter->flags |= IXGBE_FLAG_RX_1BUF_CAPABLE;
3402 
3403 	/* setup the private structure */
3404 	err = ixgbevf_sw_init(adapter);
3405 	if (err)
3406 		goto err_sw_init;
3407 
3408 	/* The HW MAC address was set and/or determined in sw_init */
3409 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3410 
3411 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3412 		pr_err("invalid MAC address\n");
3413 		err = -EIO;
3414 		goto err_sw_init;
3415 	}
3416 
3417 	netdev->hw_features = NETIF_F_SG |
3418 			   NETIF_F_IP_CSUM |
3419 			   NETIF_F_IPV6_CSUM |
3420 			   NETIF_F_TSO |
3421 			   NETIF_F_TSO6 |
3422 			   NETIF_F_RXCSUM;
3423 
3424 	netdev->features = netdev->hw_features |
3425 			   NETIF_F_HW_VLAN_TX |
3426 			   NETIF_F_HW_VLAN_RX |
3427 			   NETIF_F_HW_VLAN_FILTER;
3428 
3429 	netdev->vlan_features |= NETIF_F_TSO;
3430 	netdev->vlan_features |= NETIF_F_TSO6;
3431 	netdev->vlan_features |= NETIF_F_IP_CSUM;
3432 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3433 	netdev->vlan_features |= NETIF_F_SG;
3434 
3435 	if (pci_using_dac)
3436 		netdev->features |= NETIF_F_HIGHDMA;
3437 
3438 	netdev->priv_flags |= IFF_UNICAST_FLT;
3439 
3440 	init_timer(&adapter->watchdog_timer);
3441 	adapter->watchdog_timer.function = ixgbevf_watchdog;
3442 	adapter->watchdog_timer.data = (unsigned long)adapter;
3443 
3444 	INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3445 	INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3446 
3447 	err = ixgbevf_init_interrupt_scheme(adapter);
3448 	if (err)
3449 		goto err_sw_init;
3450 
3451 	/* pick up the PCI bus settings for reporting later */
3452 	if (hw->mac.ops.get_bus_info)
3453 		hw->mac.ops.get_bus_info(hw);
3454 
3455 	strcpy(netdev->name, "eth%d");
3456 
3457 	err = register_netdev(netdev);
3458 	if (err)
3459 		goto err_register;
3460 
3461 	adapter->netdev_registered = true;
3462 
3463 	netif_carrier_off(netdev);
3464 
3465 	ixgbevf_init_last_counter_stats(adapter);
3466 
3467 	/* print the MAC address */
3468 	hw_dbg(hw, "%pM\n", netdev->dev_addr);
3469 
3470 	hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3471 
3472 	hw_dbg(hw, "LRO is disabled\n");
3473 
3474 	hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3475 	cards_found++;
3476 	return 0;
3477 
3478 err_register:
3479 err_sw_init:
3480 	ixgbevf_reset_interrupt_capability(adapter);
3481 	iounmap(hw->hw_addr);
3482 err_ioremap:
3483 	free_netdev(netdev);
3484 err_alloc_etherdev:
3485 	pci_release_regions(pdev);
3486 err_pci_reg:
3487 err_dma:
3488 	pci_disable_device(pdev);
3489 	return err;
3490 }
3491 
3492 /**
3493  * ixgbevf_remove - Device Removal Routine
3494  * @pdev: PCI device information struct
3495  *
3496  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3497  * that it should release a PCI device.  The could be caused by a
3498  * Hot-Plug event, or because the driver is going to be removed from
3499  * memory.
3500  **/
3501 static void __devexit ixgbevf_remove(struct pci_dev *pdev)
3502 {
3503 	struct net_device *netdev = pci_get_drvdata(pdev);
3504 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3505 
3506 	set_bit(__IXGBEVF_DOWN, &adapter->state);
3507 
3508 	del_timer_sync(&adapter->watchdog_timer);
3509 
3510 	cancel_work_sync(&adapter->reset_task);
3511 	cancel_work_sync(&adapter->watchdog_task);
3512 
3513 	if (adapter->netdev_registered) {
3514 		unregister_netdev(netdev);
3515 		adapter->netdev_registered = false;
3516 	}
3517 
3518 	ixgbevf_reset_interrupt_capability(adapter);
3519 
3520 	iounmap(adapter->hw.hw_addr);
3521 	pci_release_regions(pdev);
3522 
3523 	hw_dbg(&adapter->hw, "Remove complete\n");
3524 
3525 	kfree(adapter->tx_ring);
3526 	kfree(adapter->rx_ring);
3527 
3528 	free_netdev(netdev);
3529 
3530 	pci_disable_device(pdev);
3531 }
3532 
3533 static struct pci_driver ixgbevf_driver = {
3534 	.name     = ixgbevf_driver_name,
3535 	.id_table = ixgbevf_pci_tbl,
3536 	.probe    = ixgbevf_probe,
3537 	.remove   = __devexit_p(ixgbevf_remove),
3538 	.shutdown = ixgbevf_shutdown,
3539 };
3540 
3541 /**
3542  * ixgbevf_init_module - Driver Registration Routine
3543  *
3544  * ixgbevf_init_module is the first routine called when the driver is
3545  * loaded. All it does is register with the PCI subsystem.
3546  **/
3547 static int __init ixgbevf_init_module(void)
3548 {
3549 	int ret;
3550 	pr_info("%s - version %s\n", ixgbevf_driver_string,
3551 		ixgbevf_driver_version);
3552 
3553 	pr_info("%s\n", ixgbevf_copyright);
3554 
3555 	ret = pci_register_driver(&ixgbevf_driver);
3556 	return ret;
3557 }
3558 
3559 module_init(ixgbevf_init_module);
3560 
3561 /**
3562  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3563  *
3564  * ixgbevf_exit_module is called just before the driver is removed
3565  * from memory.
3566  **/
3567 static void __exit ixgbevf_exit_module(void)
3568 {
3569 	pci_unregister_driver(&ixgbevf_driver);
3570 }
3571 
3572 #ifdef DEBUG
3573 /**
3574  * ixgbevf_get_hw_dev_name - return device name string
3575  * used by hardware layer to print debugging information
3576  **/
3577 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3578 {
3579 	struct ixgbevf_adapter *adapter = hw->back;
3580 	return adapter->netdev->name;
3581 }
3582 
3583 #endif
3584 module_exit(ixgbevf_exit_module);
3585 
3586 /* ixgbevf_main.c */
3587