1 /*******************************************************************************
2 
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2014 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/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
51 #include <linux/if.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
54 
55 #include "ixgbevf.h"
56 
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
60 
61 #define DRV_VERSION "2.12.1-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64 	"Copyright (c) 2009 - 2012 Intel Corporation.";
65 
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67 	[board_82599_vf] = &ixgbevf_82599_vf_info,
68 	[board_X540_vf]  = &ixgbevf_X540_vf_info,
69 	[board_X550_vf]  = &ixgbevf_X550_vf_info,
70 	[board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
71 };
72 
73 /* ixgbevf_pci_tbl - PCI Device ID Table
74  *
75  * Wildcard entries (PCI_ANY_ID) should come last
76  * Last entry must be all 0s
77  *
78  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
79  *   Class, Class Mask, private data (not used) }
80  */
81 static const struct pci_device_id ixgbevf_pci_tbl[] = {
82 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
83 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
84 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
86 	/* required last entry */
87 	{0, }
88 };
89 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
90 
91 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
92 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
93 MODULE_LICENSE("GPL");
94 MODULE_VERSION(DRV_VERSION);
95 
96 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
97 static int debug = -1;
98 module_param(debug, int, 0);
99 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
100 
101 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
102 {
103 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
104 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
105 	    !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
106 		schedule_work(&adapter->service_task);
107 }
108 
109 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
110 {
111 	BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
112 
113 	/* flush memory to make sure state is correct before next watchdog */
114 	smp_mb__before_atomic();
115 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
116 }
117 
118 /* forward decls */
119 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
120 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
121 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
122 
123 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
124 {
125 	struct ixgbevf_adapter *adapter = hw->back;
126 
127 	if (!hw->hw_addr)
128 		return;
129 	hw->hw_addr = NULL;
130 	dev_err(&adapter->pdev->dev, "Adapter removed\n");
131 	if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
132 		ixgbevf_service_event_schedule(adapter);
133 }
134 
135 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
136 {
137 	u32 value;
138 
139 	/* The following check not only optimizes a bit by not
140 	 * performing a read on the status register when the
141 	 * register just read was a status register read that
142 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
143 	 * potential recursion.
144 	 */
145 	if (reg == IXGBE_VFSTATUS) {
146 		ixgbevf_remove_adapter(hw);
147 		return;
148 	}
149 	value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
150 	if (value == IXGBE_FAILED_READ_REG)
151 		ixgbevf_remove_adapter(hw);
152 }
153 
154 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
155 {
156 	u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
157 	u32 value;
158 
159 	if (IXGBE_REMOVED(reg_addr))
160 		return IXGBE_FAILED_READ_REG;
161 	value = readl(reg_addr + reg);
162 	if (unlikely(value == IXGBE_FAILED_READ_REG))
163 		ixgbevf_check_remove(hw, reg);
164 	return value;
165 }
166 
167 /**
168  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
169  * @adapter: pointer to adapter struct
170  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
171  * @queue: queue to map the corresponding interrupt to
172  * @msix_vector: the vector to map to the corresponding queue
173  */
174 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
175 			     u8 queue, u8 msix_vector)
176 {
177 	u32 ivar, index;
178 	struct ixgbe_hw *hw = &adapter->hw;
179 	if (direction == -1) {
180 		/* other causes */
181 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
182 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
183 		ivar &= ~0xFF;
184 		ivar |= msix_vector;
185 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
186 	} else {
187 		/* tx or rx causes */
188 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
189 		index = ((16 * (queue & 1)) + (8 * direction));
190 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
191 		ivar &= ~(0xFF << index);
192 		ivar |= (msix_vector << index);
193 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
194 	}
195 }
196 
197 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
198 					struct ixgbevf_tx_buffer *tx_buffer)
199 {
200 	if (tx_buffer->skb) {
201 		dev_kfree_skb_any(tx_buffer->skb);
202 		if (dma_unmap_len(tx_buffer, len))
203 			dma_unmap_single(tx_ring->dev,
204 					 dma_unmap_addr(tx_buffer, dma),
205 					 dma_unmap_len(tx_buffer, len),
206 					 DMA_TO_DEVICE);
207 	} else if (dma_unmap_len(tx_buffer, len)) {
208 		dma_unmap_page(tx_ring->dev,
209 			       dma_unmap_addr(tx_buffer, dma),
210 			       dma_unmap_len(tx_buffer, len),
211 			       DMA_TO_DEVICE);
212 	}
213 	tx_buffer->next_to_watch = NULL;
214 	tx_buffer->skb = NULL;
215 	dma_unmap_len_set(tx_buffer, len, 0);
216 	/* tx_buffer must be completely set up in the transmit path */
217 }
218 
219 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
220 {
221 	return ring->stats.packets;
222 }
223 
224 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
225 {
226 	struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
227 	struct ixgbe_hw *hw = &adapter->hw;
228 
229 	u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
230 	u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
231 
232 	if (head != tail)
233 		return (head < tail) ?
234 			tail - head : (tail + ring->count - head);
235 
236 	return 0;
237 }
238 
239 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
240 {
241 	u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
242 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
243 	u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
244 
245 	clear_check_for_tx_hang(tx_ring);
246 
247 	/* Check for a hung queue, but be thorough. This verifies
248 	 * that a transmit has been completed since the previous
249 	 * check AND there is at least one packet pending. The
250 	 * ARMED bit is set to indicate a potential hang.
251 	 */
252 	if ((tx_done_old == tx_done) && tx_pending) {
253 		/* make sure it is true for two checks in a row */
254 		return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
255 					&tx_ring->state);
256 	}
257 	/* reset the countdown */
258 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
259 
260 	/* update completed stats and continue */
261 	tx_ring->tx_stats.tx_done_old = tx_done;
262 
263 	return false;
264 }
265 
266 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
267 {
268 	/* Do the reset outside of interrupt context */
269 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
270 		adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
271 		ixgbevf_service_event_schedule(adapter);
272 	}
273 }
274 
275 /**
276  * ixgbevf_tx_timeout - Respond to a Tx Hang
277  * @netdev: network interface device structure
278  **/
279 static void ixgbevf_tx_timeout(struct net_device *netdev)
280 {
281 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
282 
283 	ixgbevf_tx_timeout_reset(adapter);
284 }
285 
286 /**
287  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
288  * @q_vector: board private structure
289  * @tx_ring: tx ring to clean
290  **/
291 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
292 				 struct ixgbevf_ring *tx_ring)
293 {
294 	struct ixgbevf_adapter *adapter = q_vector->adapter;
295 	struct ixgbevf_tx_buffer *tx_buffer;
296 	union ixgbe_adv_tx_desc *tx_desc;
297 	unsigned int total_bytes = 0, total_packets = 0;
298 	unsigned int budget = tx_ring->count / 2;
299 	unsigned int i = tx_ring->next_to_clean;
300 
301 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
302 		return true;
303 
304 	tx_buffer = &tx_ring->tx_buffer_info[i];
305 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
306 	i -= tx_ring->count;
307 
308 	do {
309 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
310 
311 		/* if next_to_watch is not set then there is no work pending */
312 		if (!eop_desc)
313 			break;
314 
315 		/* prevent any other reads prior to eop_desc */
316 		read_barrier_depends();
317 
318 		/* if DD is not set pending work has not been completed */
319 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
320 			break;
321 
322 		/* clear next_to_watch to prevent false hangs */
323 		tx_buffer->next_to_watch = NULL;
324 
325 		/* update the statistics for this packet */
326 		total_bytes += tx_buffer->bytecount;
327 		total_packets += tx_buffer->gso_segs;
328 
329 		/* free the skb */
330 		dev_kfree_skb_any(tx_buffer->skb);
331 
332 		/* unmap skb header data */
333 		dma_unmap_single(tx_ring->dev,
334 				 dma_unmap_addr(tx_buffer, dma),
335 				 dma_unmap_len(tx_buffer, len),
336 				 DMA_TO_DEVICE);
337 
338 		/* clear tx_buffer data */
339 		tx_buffer->skb = NULL;
340 		dma_unmap_len_set(tx_buffer, len, 0);
341 
342 		/* unmap remaining buffers */
343 		while (tx_desc != eop_desc) {
344 			tx_buffer++;
345 			tx_desc++;
346 			i++;
347 			if (unlikely(!i)) {
348 				i -= tx_ring->count;
349 				tx_buffer = tx_ring->tx_buffer_info;
350 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
351 			}
352 
353 			/* unmap any remaining paged data */
354 			if (dma_unmap_len(tx_buffer, len)) {
355 				dma_unmap_page(tx_ring->dev,
356 					       dma_unmap_addr(tx_buffer, dma),
357 					       dma_unmap_len(tx_buffer, len),
358 					       DMA_TO_DEVICE);
359 				dma_unmap_len_set(tx_buffer, len, 0);
360 			}
361 		}
362 
363 		/* move us one more past the eop_desc for start of next pkt */
364 		tx_buffer++;
365 		tx_desc++;
366 		i++;
367 		if (unlikely(!i)) {
368 			i -= tx_ring->count;
369 			tx_buffer = tx_ring->tx_buffer_info;
370 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
371 		}
372 
373 		/* issue prefetch for next Tx descriptor */
374 		prefetch(tx_desc);
375 
376 		/* update budget accounting */
377 		budget--;
378 	} while (likely(budget));
379 
380 	i += tx_ring->count;
381 	tx_ring->next_to_clean = i;
382 	u64_stats_update_begin(&tx_ring->syncp);
383 	tx_ring->stats.bytes += total_bytes;
384 	tx_ring->stats.packets += total_packets;
385 	u64_stats_update_end(&tx_ring->syncp);
386 	q_vector->tx.total_bytes += total_bytes;
387 	q_vector->tx.total_packets += total_packets;
388 
389 	if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
390 		struct ixgbe_hw *hw = &adapter->hw;
391 		union ixgbe_adv_tx_desc *eop_desc;
392 
393 		eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
394 
395 		pr_err("Detected Tx Unit Hang\n"
396 		       "  Tx Queue             <%d>\n"
397 		       "  TDH, TDT             <%x>, <%x>\n"
398 		       "  next_to_use          <%x>\n"
399 		       "  next_to_clean        <%x>\n"
400 		       "tx_buffer_info[next_to_clean]\n"
401 		       "  next_to_watch        <%p>\n"
402 		       "  eop_desc->wb.status  <%x>\n"
403 		       "  time_stamp           <%lx>\n"
404 		       "  jiffies              <%lx>\n",
405 		       tx_ring->queue_index,
406 		       IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
407 		       IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
408 		       tx_ring->next_to_use, i,
409 		       eop_desc, (eop_desc ? eop_desc->wb.status : 0),
410 		       tx_ring->tx_buffer_info[i].time_stamp, jiffies);
411 
412 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
413 
414 		/* schedule immediate reset if we believe we hung */
415 		ixgbevf_tx_timeout_reset(adapter);
416 
417 		return true;
418 	}
419 
420 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
421 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
422 		     (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
423 		/* Make sure that anybody stopping the queue after this
424 		 * sees the new next_to_clean.
425 		 */
426 		smp_mb();
427 
428 		if (__netif_subqueue_stopped(tx_ring->netdev,
429 					     tx_ring->queue_index) &&
430 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
431 			netif_wake_subqueue(tx_ring->netdev,
432 					    tx_ring->queue_index);
433 			++tx_ring->tx_stats.restart_queue;
434 		}
435 	}
436 
437 	return !!budget;
438 }
439 
440 /**
441  * ixgbevf_rx_skb - Helper function to determine proper Rx method
442  * @q_vector: structure containing interrupt and ring information
443  * @skb: packet to send up
444  **/
445 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
446 			   struct sk_buff *skb)
447 {
448 #ifdef CONFIG_NET_RX_BUSY_POLL
449 	skb_mark_napi_id(skb, &q_vector->napi);
450 
451 	if (ixgbevf_qv_busy_polling(q_vector)) {
452 		netif_receive_skb(skb);
453 		/* exit early if we busy polled */
454 		return;
455 	}
456 #endif /* CONFIG_NET_RX_BUSY_POLL */
457 
458 	napi_gro_receive(&q_vector->napi, skb);
459 }
460 
461 /* ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
462  * @ring: structure containig ring specific data
463  * @rx_desc: current Rx descriptor being processed
464  * @skb: skb currently being received and modified
465  */
466 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
467 				       union ixgbe_adv_rx_desc *rx_desc,
468 				       struct sk_buff *skb)
469 {
470 	skb_checksum_none_assert(skb);
471 
472 	/* Rx csum disabled */
473 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
474 		return;
475 
476 	/* if IP and error */
477 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
478 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
479 		ring->rx_stats.csum_err++;
480 		return;
481 	}
482 
483 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
484 		return;
485 
486 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
487 		ring->rx_stats.csum_err++;
488 		return;
489 	}
490 
491 	/* It must be a TCP or UDP packet with a valid checksum */
492 	skb->ip_summed = CHECKSUM_UNNECESSARY;
493 }
494 
495 /* ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
496  * @rx_ring: rx descriptor ring packet is being transacted on
497  * @rx_desc: pointer to the EOP Rx descriptor
498  * @skb: pointer to current skb being populated
499  *
500  * This function checks the ring, descriptor, and packet information in
501  * order to populate the checksum, VLAN, protocol, and other fields within
502  * the skb.
503  */
504 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
505 				       union ixgbe_adv_rx_desc *rx_desc,
506 				       struct sk_buff *skb)
507 {
508 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
509 
510 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
511 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
512 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
513 
514 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
515 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
516 	}
517 
518 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
519 }
520 
521 /**
522  * ixgbevf_is_non_eop - process handling of non-EOP buffers
523  * @rx_ring: Rx ring being processed
524  * @rx_desc: Rx descriptor for current buffer
525  * @skb: current socket buffer containing buffer in progress
526  *
527  * This function updates next to clean.  If the buffer is an EOP buffer
528  * this function exits returning false, otherwise it will place the
529  * sk_buff in the next buffer to be chained and return true indicating
530  * that this is in fact a non-EOP buffer.
531  **/
532 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
533 			       union ixgbe_adv_rx_desc *rx_desc)
534 {
535 	u32 ntc = rx_ring->next_to_clean + 1;
536 
537 	/* fetch, update, and store next to clean */
538 	ntc = (ntc < rx_ring->count) ? ntc : 0;
539 	rx_ring->next_to_clean = ntc;
540 
541 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
542 
543 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
544 		return false;
545 
546 	return true;
547 }
548 
549 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
550 				      struct ixgbevf_rx_buffer *bi)
551 {
552 	struct page *page = bi->page;
553 	dma_addr_t dma = bi->dma;
554 
555 	/* since we are recycling buffers we should seldom need to alloc */
556 	if (likely(page))
557 		return true;
558 
559 	/* alloc new page for storage */
560 	page = dev_alloc_page();
561 	if (unlikely(!page)) {
562 		rx_ring->rx_stats.alloc_rx_page_failed++;
563 		return false;
564 	}
565 
566 	/* map page for use */
567 	dma = dma_map_page(rx_ring->dev, page, 0,
568 			   PAGE_SIZE, DMA_FROM_DEVICE);
569 
570 	/* if mapping failed free memory back to system since
571 	 * there isn't much point in holding memory we can't use
572 	 */
573 	if (dma_mapping_error(rx_ring->dev, dma)) {
574 		__free_page(page);
575 
576 		rx_ring->rx_stats.alloc_rx_buff_failed++;
577 		return false;
578 	}
579 
580 	bi->dma = dma;
581 	bi->page = page;
582 	bi->page_offset = 0;
583 
584 	return true;
585 }
586 
587 /**
588  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
589  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
590  * @cleaned_count: number of buffers to replace
591  **/
592 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
593 				     u16 cleaned_count)
594 {
595 	union ixgbe_adv_rx_desc *rx_desc;
596 	struct ixgbevf_rx_buffer *bi;
597 	unsigned int i = rx_ring->next_to_use;
598 
599 	/* nothing to do or no valid netdev defined */
600 	if (!cleaned_count || !rx_ring->netdev)
601 		return;
602 
603 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
604 	bi = &rx_ring->rx_buffer_info[i];
605 	i -= rx_ring->count;
606 
607 	do {
608 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
609 			break;
610 
611 		/* Refresh the desc even if pkt_addr didn't change
612 		 * because each write-back erases this info.
613 		 */
614 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
615 
616 		rx_desc++;
617 		bi++;
618 		i++;
619 		if (unlikely(!i)) {
620 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
621 			bi = rx_ring->rx_buffer_info;
622 			i -= rx_ring->count;
623 		}
624 
625 		/* clear the hdr_addr for the next_to_use descriptor */
626 		rx_desc->read.hdr_addr = 0;
627 
628 		cleaned_count--;
629 	} while (cleaned_count);
630 
631 	i += rx_ring->count;
632 
633 	if (rx_ring->next_to_use != i) {
634 		/* record the next descriptor to use */
635 		rx_ring->next_to_use = i;
636 
637 		/* update next to alloc since we have filled the ring */
638 		rx_ring->next_to_alloc = i;
639 
640 		/* Force memory writes to complete before letting h/w
641 		 * know there are new descriptors to fetch.  (Only
642 		 * applicable for weak-ordered memory model archs,
643 		 * such as IA-64).
644 		 */
645 		wmb();
646 		ixgbevf_write_tail(rx_ring, i);
647 	}
648 }
649 
650 /* ixgbevf_pull_tail - ixgbevf specific version of skb_pull_tail
651  * @rx_ring: rx descriptor ring packet is being transacted on
652  * @skb: pointer to current skb being adjusted
653  *
654  * This function is an ixgbevf specific version of __pskb_pull_tail.  The
655  * main difference between this version and the original function is that
656  * this function can make several assumptions about the state of things
657  * that allow for significant optimizations versus the standard function.
658  * As a result we can do things like drop a frag and maintain an accurate
659  * truesize for the skb.
660  */
661 static void ixgbevf_pull_tail(struct ixgbevf_ring *rx_ring,
662 			      struct sk_buff *skb)
663 {
664 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
665 	unsigned char *va;
666 	unsigned int pull_len;
667 
668 	/* it is valid to use page_address instead of kmap since we are
669 	 * working with pages allocated out of the lomem pool per
670 	 * alloc_page(GFP_ATOMIC)
671 	 */
672 	va = skb_frag_address(frag);
673 
674 	/* we need the header to contain the greater of either ETH_HLEN or
675 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
676 	 */
677 	pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
678 
679 	/* align pull length to size of long to optimize memcpy performance */
680 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
681 
682 	/* update all of the pointers */
683 	skb_frag_size_sub(frag, pull_len);
684 	frag->page_offset += pull_len;
685 	skb->data_len -= pull_len;
686 	skb->tail += pull_len;
687 }
688 
689 /* ixgbevf_cleanup_headers - Correct corrupted or empty headers
690  * @rx_ring: rx descriptor ring packet is being transacted on
691  * @rx_desc: pointer to the EOP Rx descriptor
692  * @skb: pointer to current skb being fixed
693  *
694  * Check for corrupted packet headers caused by senders on the local L2
695  * embedded NIC switch not setting up their Tx Descriptors right.  These
696  * should be very rare.
697  *
698  * Also address the case where we are pulling data in on pages only
699  * and as such no data is present in the skb header.
700  *
701  * In addition if skb is not at least 60 bytes we need to pad it so that
702  * it is large enough to qualify as a valid Ethernet frame.
703  *
704  * Returns true if an error was encountered and skb was freed.
705  */
706 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
707 				    union ixgbe_adv_rx_desc *rx_desc,
708 				    struct sk_buff *skb)
709 {
710 	/* verify that the packet does not have any known errors */
711 	if (unlikely(ixgbevf_test_staterr(rx_desc,
712 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
713 		struct net_device *netdev = rx_ring->netdev;
714 
715 		if (!(netdev->features & NETIF_F_RXALL)) {
716 			dev_kfree_skb_any(skb);
717 			return true;
718 		}
719 	}
720 
721 	/* place header in linear portion of buffer */
722 	if (skb_is_nonlinear(skb))
723 		ixgbevf_pull_tail(rx_ring, skb);
724 
725 	/* if eth_skb_pad returns an error the skb was freed */
726 	if (eth_skb_pad(skb))
727 		return true;
728 
729 	return false;
730 }
731 
732 /* ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
733  * @rx_ring: rx descriptor ring to store buffers on
734  * @old_buff: donor buffer to have page reused
735  *
736  * Synchronizes page for reuse by the adapter
737  */
738 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
739 				  struct ixgbevf_rx_buffer *old_buff)
740 {
741 	struct ixgbevf_rx_buffer *new_buff;
742 	u16 nta = rx_ring->next_to_alloc;
743 
744 	new_buff = &rx_ring->rx_buffer_info[nta];
745 
746 	/* update, and store next to alloc */
747 	nta++;
748 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
749 
750 	/* transfer page from old buffer to new buffer */
751 	new_buff->page = old_buff->page;
752 	new_buff->dma = old_buff->dma;
753 	new_buff->page_offset = old_buff->page_offset;
754 
755 	/* sync the buffer for use by the device */
756 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
757 					 new_buff->page_offset,
758 					 IXGBEVF_RX_BUFSZ,
759 					 DMA_FROM_DEVICE);
760 }
761 
762 static inline bool ixgbevf_page_is_reserved(struct page *page)
763 {
764 	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
765 }
766 
767 /* ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
768  * @rx_ring: rx descriptor ring to transact packets on
769  * @rx_buffer: buffer containing page to add
770  * @rx_desc: descriptor containing length of buffer written by hardware
771  * @skb: sk_buff to place the data into
772  *
773  * This function will add the data contained in rx_buffer->page to the skb.
774  * This is done either through a direct copy if the data in the buffer is
775  * less than the skb header size, otherwise it will just attach the page as
776  * a frag to the skb.
777  *
778  * The function will then update the page offset if necessary and return
779  * true if the buffer can be reused by the adapter.
780  */
781 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
782 				struct ixgbevf_rx_buffer *rx_buffer,
783 				union ixgbe_adv_rx_desc *rx_desc,
784 				struct sk_buff *skb)
785 {
786 	struct page *page = rx_buffer->page;
787 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
788 #if (PAGE_SIZE < 8192)
789 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
790 #else
791 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
792 #endif
793 
794 	if ((size <= IXGBEVF_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
795 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
796 
797 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
798 
799 		/* page is not reserved, we can reuse buffer as is */
800 		if (likely(!ixgbevf_page_is_reserved(page)))
801 			return true;
802 
803 		/* this page cannot be reused so discard it */
804 		put_page(page);
805 		return false;
806 	}
807 
808 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
809 			rx_buffer->page_offset, size, truesize);
810 
811 	/* avoid re-using remote pages */
812 	if (unlikely(ixgbevf_page_is_reserved(page)))
813 		return false;
814 
815 #if (PAGE_SIZE < 8192)
816 	/* if we are only owner of page we can reuse it */
817 	if (unlikely(page_count(page) != 1))
818 		return false;
819 
820 	/* flip page offset to other buffer */
821 	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
822 
823 #else
824 	/* move offset up to the next cache line */
825 	rx_buffer->page_offset += truesize;
826 
827 	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
828 		return false;
829 
830 #endif
831 	/* Even if we own the page, we are not allowed to use atomic_set()
832 	 * This would break get_page_unless_zero() users.
833 	 */
834 	atomic_inc(&page->_count);
835 
836 	return true;
837 }
838 
839 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
840 					       union ixgbe_adv_rx_desc *rx_desc,
841 					       struct sk_buff *skb)
842 {
843 	struct ixgbevf_rx_buffer *rx_buffer;
844 	struct page *page;
845 
846 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
847 	page = rx_buffer->page;
848 	prefetchw(page);
849 
850 	if (likely(!skb)) {
851 		void *page_addr = page_address(page) +
852 				  rx_buffer->page_offset;
853 
854 		/* prefetch first cache line of first page */
855 		prefetch(page_addr);
856 #if L1_CACHE_BYTES < 128
857 		prefetch(page_addr + L1_CACHE_BYTES);
858 #endif
859 
860 		/* allocate a skb to store the frags */
861 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
862 						IXGBEVF_RX_HDR_SIZE);
863 		if (unlikely(!skb)) {
864 			rx_ring->rx_stats.alloc_rx_buff_failed++;
865 			return NULL;
866 		}
867 
868 		/* we will be copying header into skb->data in
869 		 * pskb_may_pull so it is in our interest to prefetch
870 		 * it now to avoid a possible cache miss
871 		 */
872 		prefetchw(skb->data);
873 	}
874 
875 	/* we are reusing so sync this buffer for CPU use */
876 	dma_sync_single_range_for_cpu(rx_ring->dev,
877 				      rx_buffer->dma,
878 				      rx_buffer->page_offset,
879 				      IXGBEVF_RX_BUFSZ,
880 				      DMA_FROM_DEVICE);
881 
882 	/* pull page into skb */
883 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
884 		/* hand second half of page back to the ring */
885 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
886 	} else {
887 		/* we are not reusing the buffer so unmap it */
888 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
889 			       PAGE_SIZE, DMA_FROM_DEVICE);
890 	}
891 
892 	/* clear contents of buffer_info */
893 	rx_buffer->dma = 0;
894 	rx_buffer->page = NULL;
895 
896 	return skb;
897 }
898 
899 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
900 					     u32 qmask)
901 {
902 	struct ixgbe_hw *hw = &adapter->hw;
903 
904 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
905 }
906 
907 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
908 				struct ixgbevf_ring *rx_ring,
909 				int budget)
910 {
911 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
912 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
913 	struct sk_buff *skb = rx_ring->skb;
914 
915 	while (likely(total_rx_packets < budget)) {
916 		union ixgbe_adv_rx_desc *rx_desc;
917 
918 		/* return some buffers to hardware, one at a time is too slow */
919 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
920 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
921 			cleaned_count = 0;
922 		}
923 
924 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
925 
926 		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
927 			break;
928 
929 		/* This memory barrier is needed to keep us from reading
930 		 * any other fields out of the rx_desc until we know the
931 		 * RXD_STAT_DD bit is set
932 		 */
933 		rmb();
934 
935 		/* retrieve a buffer from the ring */
936 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
937 
938 		/* exit if we failed to retrieve a buffer */
939 		if (!skb)
940 			break;
941 
942 		cleaned_count++;
943 
944 		/* fetch next buffer in frame if non-eop */
945 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
946 			continue;
947 
948 		/* verify the packet layout is correct */
949 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
950 			skb = NULL;
951 			continue;
952 		}
953 
954 		/* probably a little skewed due to removing CRC */
955 		total_rx_bytes += skb->len;
956 
957 		/* Workaround hardware that can't do proper VEPA multicast
958 		 * source pruning.
959 		 */
960 		if ((skb->pkt_type == PACKET_BROADCAST ||
961 		    skb->pkt_type == PACKET_MULTICAST) &&
962 		    ether_addr_equal(rx_ring->netdev->dev_addr,
963 				     eth_hdr(skb)->h_source)) {
964 			dev_kfree_skb_irq(skb);
965 			continue;
966 		}
967 
968 		/* populate checksum, VLAN, and protocol */
969 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
970 
971 		ixgbevf_rx_skb(q_vector, skb);
972 
973 		/* reset skb pointer */
974 		skb = NULL;
975 
976 		/* update budget accounting */
977 		total_rx_packets++;
978 	}
979 
980 	/* place incomplete frames back on ring for completion */
981 	rx_ring->skb = skb;
982 
983 	u64_stats_update_begin(&rx_ring->syncp);
984 	rx_ring->stats.packets += total_rx_packets;
985 	rx_ring->stats.bytes += total_rx_bytes;
986 	u64_stats_update_end(&rx_ring->syncp);
987 	q_vector->rx.total_packets += total_rx_packets;
988 	q_vector->rx.total_bytes += total_rx_bytes;
989 
990 	return total_rx_packets;
991 }
992 
993 /**
994  * ixgbevf_poll - NAPI polling calback
995  * @napi: napi struct with our devices info in it
996  * @budget: amount of work driver is allowed to do this pass, in packets
997  *
998  * This function will clean more than one or more rings associated with a
999  * q_vector.
1000  **/
1001 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1002 {
1003 	struct ixgbevf_q_vector *q_vector =
1004 		container_of(napi, struct ixgbevf_q_vector, napi);
1005 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1006 	struct ixgbevf_ring *ring;
1007 	int per_ring_budget;
1008 	bool clean_complete = true;
1009 
1010 	ixgbevf_for_each_ring(ring, q_vector->tx)
1011 		clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1012 
1013 #ifdef CONFIG_NET_RX_BUSY_POLL
1014 	if (!ixgbevf_qv_lock_napi(q_vector))
1015 		return budget;
1016 #endif
1017 
1018 	/* attempt to distribute budget to each queue fairly, but don't allow
1019 	 * the budget to go below 1 because we'll exit polling */
1020 	if (q_vector->rx.count > 1)
1021 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1022 	else
1023 		per_ring_budget = budget;
1024 
1025 	ixgbevf_for_each_ring(ring, q_vector->rx)
1026 		clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
1027 							per_ring_budget)
1028 				   < per_ring_budget);
1029 
1030 #ifdef CONFIG_NET_RX_BUSY_POLL
1031 	ixgbevf_qv_unlock_napi(q_vector);
1032 #endif
1033 
1034 	/* If all work not completed, return budget and keep polling */
1035 	if (!clean_complete)
1036 		return budget;
1037 	/* all work done, exit the polling mode */
1038 	napi_complete(napi);
1039 	if (adapter->rx_itr_setting & 1)
1040 		ixgbevf_set_itr(q_vector);
1041 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1042 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1043 		ixgbevf_irq_enable_queues(adapter,
1044 					  1 << q_vector->v_idx);
1045 
1046 	return 0;
1047 }
1048 
1049 /**
1050  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1051  * @q_vector: structure containing interrupt and ring information
1052  */
1053 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1054 {
1055 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1056 	struct ixgbe_hw *hw = &adapter->hw;
1057 	int v_idx = q_vector->v_idx;
1058 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1059 
1060 	/*
1061 	 * set the WDIS bit to not clear the timer bits and cause an
1062 	 * immediate assertion of the interrupt
1063 	 */
1064 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1065 
1066 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1067 }
1068 
1069 #ifdef CONFIG_NET_RX_BUSY_POLL
1070 /* must be called with local_bh_disable()d */
1071 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1072 {
1073 	struct ixgbevf_q_vector *q_vector =
1074 			container_of(napi, struct ixgbevf_q_vector, napi);
1075 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1076 	struct ixgbevf_ring  *ring;
1077 	int found = 0;
1078 
1079 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1080 		return LL_FLUSH_FAILED;
1081 
1082 	if (!ixgbevf_qv_lock_poll(q_vector))
1083 		return LL_FLUSH_BUSY;
1084 
1085 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1086 		found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1087 #ifdef BP_EXTENDED_STATS
1088 		if (found)
1089 			ring->stats.cleaned += found;
1090 		else
1091 			ring->stats.misses++;
1092 #endif
1093 		if (found)
1094 			break;
1095 	}
1096 
1097 	ixgbevf_qv_unlock_poll(q_vector);
1098 
1099 	return found;
1100 }
1101 #endif /* CONFIG_NET_RX_BUSY_POLL */
1102 
1103 /**
1104  * ixgbevf_configure_msix - Configure MSI-X hardware
1105  * @adapter: board private structure
1106  *
1107  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1108  * interrupts.
1109  **/
1110 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1111 {
1112 	struct ixgbevf_q_vector *q_vector;
1113 	int q_vectors, v_idx;
1114 
1115 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1116 	adapter->eims_enable_mask = 0;
1117 
1118 	/*
1119 	 * Populate the IVAR table and set the ITR values to the
1120 	 * corresponding register.
1121 	 */
1122 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1123 		struct ixgbevf_ring *ring;
1124 		q_vector = adapter->q_vector[v_idx];
1125 
1126 		ixgbevf_for_each_ring(ring, q_vector->rx)
1127 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1128 
1129 		ixgbevf_for_each_ring(ring, q_vector->tx)
1130 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1131 
1132 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1133 			/* tx only vector */
1134 			if (adapter->tx_itr_setting == 1)
1135 				q_vector->itr = IXGBE_10K_ITR;
1136 			else
1137 				q_vector->itr = adapter->tx_itr_setting;
1138 		} else {
1139 			/* rx or rx/tx vector */
1140 			if (adapter->rx_itr_setting == 1)
1141 				q_vector->itr = IXGBE_20K_ITR;
1142 			else
1143 				q_vector->itr = adapter->rx_itr_setting;
1144 		}
1145 
1146 		/* add q_vector eims value to global eims_enable_mask */
1147 		adapter->eims_enable_mask |= 1 << v_idx;
1148 
1149 		ixgbevf_write_eitr(q_vector);
1150 	}
1151 
1152 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1153 	/* setup eims_other and add value to global eims_enable_mask */
1154 	adapter->eims_other = 1 << v_idx;
1155 	adapter->eims_enable_mask |= adapter->eims_other;
1156 }
1157 
1158 enum latency_range {
1159 	lowest_latency = 0,
1160 	low_latency = 1,
1161 	bulk_latency = 2,
1162 	latency_invalid = 255
1163 };
1164 
1165 /**
1166  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1167  * @q_vector: structure containing interrupt and ring information
1168  * @ring_container: structure containing ring performance data
1169  *
1170  *      Stores a new ITR value based on packets and byte
1171  *      counts during the last interrupt.  The advantage of per interrupt
1172  *      computation is faster updates and more accurate ITR for the current
1173  *      traffic pattern.  Constants in this function were computed
1174  *      based on theoretical maximum wire speed and thresholds were set based
1175  *      on testing data as well as attempting to minimize response time
1176  *      while increasing bulk throughput.
1177  **/
1178 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1179 			       struct ixgbevf_ring_container *ring_container)
1180 {
1181 	int bytes = ring_container->total_bytes;
1182 	int packets = ring_container->total_packets;
1183 	u32 timepassed_us;
1184 	u64 bytes_perint;
1185 	u8 itr_setting = ring_container->itr;
1186 
1187 	if (packets == 0)
1188 		return;
1189 
1190 	/* simple throttlerate management
1191 	 *    0-20MB/s lowest (100000 ints/s)
1192 	 *   20-100MB/s low   (20000 ints/s)
1193 	 *  100-1249MB/s bulk (8000 ints/s)
1194 	 */
1195 	/* what was last interrupt timeslice? */
1196 	timepassed_us = q_vector->itr >> 2;
1197 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1198 
1199 	switch (itr_setting) {
1200 	case lowest_latency:
1201 		if (bytes_perint > 10)
1202 			itr_setting = low_latency;
1203 		break;
1204 	case low_latency:
1205 		if (bytes_perint > 20)
1206 			itr_setting = bulk_latency;
1207 		else if (bytes_perint <= 10)
1208 			itr_setting = lowest_latency;
1209 		break;
1210 	case bulk_latency:
1211 		if (bytes_perint <= 20)
1212 			itr_setting = low_latency;
1213 		break;
1214 	}
1215 
1216 	/* clear work counters since we have the values we need */
1217 	ring_container->total_bytes = 0;
1218 	ring_container->total_packets = 0;
1219 
1220 	/* write updated itr to ring container */
1221 	ring_container->itr = itr_setting;
1222 }
1223 
1224 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1225 {
1226 	u32 new_itr = q_vector->itr;
1227 	u8 current_itr;
1228 
1229 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1230 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1231 
1232 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1233 
1234 	switch (current_itr) {
1235 	/* counts and packets in update_itr are dependent on these numbers */
1236 	case lowest_latency:
1237 		new_itr = IXGBE_100K_ITR;
1238 		break;
1239 	case low_latency:
1240 		new_itr = IXGBE_20K_ITR;
1241 		break;
1242 	case bulk_latency:
1243 	default:
1244 		new_itr = IXGBE_8K_ITR;
1245 		break;
1246 	}
1247 
1248 	if (new_itr != q_vector->itr) {
1249 		/* do an exponential smoothing */
1250 		new_itr = (10 * new_itr * q_vector->itr) /
1251 			  ((9 * new_itr) + q_vector->itr);
1252 
1253 		/* save the algorithm value here */
1254 		q_vector->itr = new_itr;
1255 
1256 		ixgbevf_write_eitr(q_vector);
1257 	}
1258 }
1259 
1260 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1261 {
1262 	struct ixgbevf_adapter *adapter = data;
1263 	struct ixgbe_hw *hw = &adapter->hw;
1264 
1265 	hw->mac.get_link_status = 1;
1266 
1267 	ixgbevf_service_event_schedule(adapter);
1268 
1269 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1270 
1271 	return IRQ_HANDLED;
1272 }
1273 
1274 /**
1275  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1276  * @irq: unused
1277  * @data: pointer to our q_vector struct for this interrupt vector
1278  **/
1279 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1280 {
1281 	struct ixgbevf_q_vector *q_vector = data;
1282 
1283 	/* EIAM disabled interrupts (on this vector) for us */
1284 	if (q_vector->rx.ring || q_vector->tx.ring)
1285 		napi_schedule(&q_vector->napi);
1286 
1287 	return IRQ_HANDLED;
1288 }
1289 
1290 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1291 				     int r_idx)
1292 {
1293 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1294 
1295 	a->rx_ring[r_idx]->next = q_vector->rx.ring;
1296 	q_vector->rx.ring = a->rx_ring[r_idx];
1297 	q_vector->rx.count++;
1298 }
1299 
1300 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1301 				     int t_idx)
1302 {
1303 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1304 
1305 	a->tx_ring[t_idx]->next = q_vector->tx.ring;
1306 	q_vector->tx.ring = a->tx_ring[t_idx];
1307 	q_vector->tx.count++;
1308 }
1309 
1310 /**
1311  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1312  * @adapter: board private structure to initialize
1313  *
1314  * This function maps descriptor rings to the queue-specific vectors
1315  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1316  * one vector per ring/queue, but on a constrained vector budget, we
1317  * group the rings as "efficiently" as possible.  You would add new
1318  * mapping configurations in here.
1319  **/
1320 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1321 {
1322 	int q_vectors;
1323 	int v_start = 0;
1324 	int rxr_idx = 0, txr_idx = 0;
1325 	int rxr_remaining = adapter->num_rx_queues;
1326 	int txr_remaining = adapter->num_tx_queues;
1327 	int i, j;
1328 	int rqpv, tqpv;
1329 	int err = 0;
1330 
1331 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1332 
1333 	/*
1334 	 * The ideal configuration...
1335 	 * We have enough vectors to map one per queue.
1336 	 */
1337 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1338 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1339 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1340 
1341 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1342 			map_vector_to_txq(adapter, v_start, txr_idx);
1343 		goto out;
1344 	}
1345 
1346 	/*
1347 	 * If we don't have enough vectors for a 1-to-1
1348 	 * mapping, we'll have to group them so there are
1349 	 * multiple queues per vector.
1350 	 */
1351 	/* Re-adjusting *qpv takes care of the remainder. */
1352 	for (i = v_start; i < q_vectors; i++) {
1353 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1354 		for (j = 0; j < rqpv; j++) {
1355 			map_vector_to_rxq(adapter, i, rxr_idx);
1356 			rxr_idx++;
1357 			rxr_remaining--;
1358 		}
1359 	}
1360 	for (i = v_start; i < q_vectors; i++) {
1361 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1362 		for (j = 0; j < tqpv; j++) {
1363 			map_vector_to_txq(adapter, i, txr_idx);
1364 			txr_idx++;
1365 			txr_remaining--;
1366 		}
1367 	}
1368 
1369 out:
1370 	return err;
1371 }
1372 
1373 /**
1374  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1375  * @adapter: board private structure
1376  *
1377  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1378  * interrupts from the kernel.
1379  **/
1380 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1381 {
1382 	struct net_device *netdev = adapter->netdev;
1383 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1384 	int vector, err;
1385 	int ri = 0, ti = 0;
1386 
1387 	for (vector = 0; vector < q_vectors; vector++) {
1388 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1389 		struct msix_entry *entry = &adapter->msix_entries[vector];
1390 
1391 		if (q_vector->tx.ring && q_vector->rx.ring) {
1392 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1393 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
1394 			ti++;
1395 		} else if (q_vector->rx.ring) {
1396 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1397 				 "%s-%s-%d", netdev->name, "rx", ri++);
1398 		} else if (q_vector->tx.ring) {
1399 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1400 				 "%s-%s-%d", netdev->name, "tx", ti++);
1401 		} else {
1402 			/* skip this unused q_vector */
1403 			continue;
1404 		}
1405 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1406 				  q_vector->name, q_vector);
1407 		if (err) {
1408 			hw_dbg(&adapter->hw,
1409 			       "request_irq failed for MSIX interrupt "
1410 			       "Error: %d\n", err);
1411 			goto free_queue_irqs;
1412 		}
1413 	}
1414 
1415 	err = request_irq(adapter->msix_entries[vector].vector,
1416 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1417 	if (err) {
1418 		hw_dbg(&adapter->hw,
1419 		       "request_irq for msix_other failed: %d\n", err);
1420 		goto free_queue_irqs;
1421 	}
1422 
1423 	return 0;
1424 
1425 free_queue_irqs:
1426 	while (vector) {
1427 		vector--;
1428 		free_irq(adapter->msix_entries[vector].vector,
1429 			 adapter->q_vector[vector]);
1430 	}
1431 	/* This failure is non-recoverable - it indicates the system is
1432 	 * out of MSIX vector resources and the VF driver cannot run
1433 	 * without them.  Set the number of msix vectors to zero
1434 	 * indicating that not enough can be allocated.  The error
1435 	 * will be returned to the user indicating device open failed.
1436 	 * Any further attempts to force the driver to open will also
1437 	 * fail.  The only way to recover is to unload the driver and
1438 	 * reload it again.  If the system has recovered some MSIX
1439 	 * vectors then it may succeed.
1440 	 */
1441 	adapter->num_msix_vectors = 0;
1442 	return err;
1443 }
1444 
1445 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1446 {
1447 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1448 
1449 	for (i = 0; i < q_vectors; i++) {
1450 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1451 		q_vector->rx.ring = NULL;
1452 		q_vector->tx.ring = NULL;
1453 		q_vector->rx.count = 0;
1454 		q_vector->tx.count = 0;
1455 	}
1456 }
1457 
1458 /**
1459  * ixgbevf_request_irq - initialize interrupts
1460  * @adapter: board private structure
1461  *
1462  * Attempts to configure interrupts using the best available
1463  * capabilities of the hardware and kernel.
1464  **/
1465 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1466 {
1467 	int err = 0;
1468 
1469 	err = ixgbevf_request_msix_irqs(adapter);
1470 
1471 	if (err)
1472 		hw_dbg(&adapter->hw,
1473 		       "request_irq failed, Error %d\n", err);
1474 
1475 	return err;
1476 }
1477 
1478 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1479 {
1480 	int i, q_vectors;
1481 
1482 	q_vectors = adapter->num_msix_vectors;
1483 	i = q_vectors - 1;
1484 
1485 	free_irq(adapter->msix_entries[i].vector, adapter);
1486 	i--;
1487 
1488 	for (; i >= 0; i--) {
1489 		/* free only the irqs that were actually requested */
1490 		if (!adapter->q_vector[i]->rx.ring &&
1491 		    !adapter->q_vector[i]->tx.ring)
1492 			continue;
1493 
1494 		free_irq(adapter->msix_entries[i].vector,
1495 			 adapter->q_vector[i]);
1496 	}
1497 
1498 	ixgbevf_reset_q_vectors(adapter);
1499 }
1500 
1501 /**
1502  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1503  * @adapter: board private structure
1504  **/
1505 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1506 {
1507 	struct ixgbe_hw *hw = &adapter->hw;
1508 	int i;
1509 
1510 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1511 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1512 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1513 
1514 	IXGBE_WRITE_FLUSH(hw);
1515 
1516 	for (i = 0; i < adapter->num_msix_vectors; i++)
1517 		synchronize_irq(adapter->msix_entries[i].vector);
1518 }
1519 
1520 /**
1521  * ixgbevf_irq_enable - Enable default interrupt generation settings
1522  * @adapter: board private structure
1523  **/
1524 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1525 {
1526 	struct ixgbe_hw *hw = &adapter->hw;
1527 
1528 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1529 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1530 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1531 }
1532 
1533 /**
1534  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1535  * @adapter: board private structure
1536  * @ring: structure containing ring specific data
1537  *
1538  * Configure the Tx descriptor ring after a reset.
1539  **/
1540 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1541 				      struct ixgbevf_ring *ring)
1542 {
1543 	struct ixgbe_hw *hw = &adapter->hw;
1544 	u64 tdba = ring->dma;
1545 	int wait_loop = 10;
1546 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1547 	u8 reg_idx = ring->reg_idx;
1548 
1549 	/* disable queue to avoid issues while updating state */
1550 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1551 	IXGBE_WRITE_FLUSH(hw);
1552 
1553 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1554 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1555 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1556 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1557 
1558 	/* disable head writeback */
1559 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1560 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1561 
1562 	/* enable relaxed ordering */
1563 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1564 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1565 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1566 
1567 	/* reset head and tail pointers */
1568 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1569 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1570 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1571 
1572 	/* reset ntu and ntc to place SW in sync with hardwdare */
1573 	ring->next_to_clean = 0;
1574 	ring->next_to_use = 0;
1575 
1576 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1577 	 * to or less than the number of on chip descriptors, which is
1578 	 * currently 40.
1579 	 */
1580 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1581 
1582 	/* Setting PTHRESH to 32 both improves performance */
1583 	txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1584 		  32;          /* PTHRESH = 32 */
1585 
1586 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1587 
1588 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1589 
1590 	/* poll to verify queue is enabled */
1591 	do {
1592 		usleep_range(1000, 2000);
1593 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1594 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1595 	if (!wait_loop)
1596 		pr_err("Could not enable Tx Queue %d\n", reg_idx);
1597 }
1598 
1599 /**
1600  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1601  * @adapter: board private structure
1602  *
1603  * Configure the Tx unit of the MAC after a reset.
1604  **/
1605 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1606 {
1607 	u32 i;
1608 
1609 	/* Setup the HW Tx Head and Tail descriptor pointers */
1610 	for (i = 0; i < adapter->num_tx_queues; i++)
1611 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1612 }
1613 
1614 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1615 
1616 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1617 {
1618 	struct ixgbe_hw *hw = &adapter->hw;
1619 	u32 srrctl;
1620 
1621 	srrctl = IXGBE_SRRCTL_DROP_EN;
1622 
1623 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1624 	srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1625 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1626 
1627 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1628 }
1629 
1630 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1631 {
1632 	struct ixgbe_hw *hw = &adapter->hw;
1633 
1634 	/* PSRTYPE must be initialized in 82599 */
1635 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1636 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1637 		      IXGBE_PSRTYPE_L2HDR;
1638 
1639 	if (adapter->num_rx_queues > 1)
1640 		psrtype |= 1 << 29;
1641 
1642 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1643 }
1644 
1645 #define IXGBEVF_MAX_RX_DESC_POLL 10
1646 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1647 				     struct ixgbevf_ring *ring)
1648 {
1649 	struct ixgbe_hw *hw = &adapter->hw;
1650 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1651 	u32 rxdctl;
1652 	u8 reg_idx = ring->reg_idx;
1653 
1654 	if (IXGBE_REMOVED(hw->hw_addr))
1655 		return;
1656 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1657 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1658 
1659 	/* write value back with RXDCTL.ENABLE bit cleared */
1660 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1661 
1662 	/* the hardware may take up to 100us to really disable the rx queue */
1663 	do {
1664 		udelay(10);
1665 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1666 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1667 
1668 	if (!wait_loop)
1669 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1670 		       reg_idx);
1671 }
1672 
1673 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1674 					 struct ixgbevf_ring *ring)
1675 {
1676 	struct ixgbe_hw *hw = &adapter->hw;
1677 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1678 	u32 rxdctl;
1679 	u8 reg_idx = ring->reg_idx;
1680 
1681 	if (IXGBE_REMOVED(hw->hw_addr))
1682 		return;
1683 	do {
1684 		usleep_range(1000, 2000);
1685 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1686 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1687 
1688 	if (!wait_loop)
1689 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1690 		       reg_idx);
1691 }
1692 
1693 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1694 {
1695 	struct ixgbe_hw *hw = &adapter->hw;
1696 	u32 vfmrqc = 0, vfreta = 0;
1697 	u32 rss_key[10];
1698 	u16 rss_i = adapter->num_rx_queues;
1699 	int i, j;
1700 
1701 	/* Fill out hash function seeds */
1702 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
1703 	for (i = 0; i < 10; i++)
1704 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1705 
1706 	/* Fill out redirection table */
1707 	for (i = 0, j = 0; i < 64; i++, j++) {
1708 		if (j == rss_i)
1709 			j = 0;
1710 		vfreta = (vfreta << 8) | (j * 0x1);
1711 		if ((i & 3) == 3)
1712 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1713 	}
1714 
1715 	/* Perform hash on these packet types */
1716 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1717 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1718 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1719 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1720 
1721 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1722 
1723 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1724 }
1725 
1726 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1727 				      struct ixgbevf_ring *ring)
1728 {
1729 	struct ixgbe_hw *hw = &adapter->hw;
1730 	u64 rdba = ring->dma;
1731 	u32 rxdctl;
1732 	u8 reg_idx = ring->reg_idx;
1733 
1734 	/* disable queue to avoid issues while updating state */
1735 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1736 	ixgbevf_disable_rx_queue(adapter, ring);
1737 
1738 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1739 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1740 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1741 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1742 
1743 	/* enable relaxed ordering */
1744 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1745 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1746 
1747 	/* reset head and tail pointers */
1748 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1749 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1750 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1751 
1752 	/* reset ntu and ntc to place SW in sync with hardwdare */
1753 	ring->next_to_clean = 0;
1754 	ring->next_to_use = 0;
1755 	ring->next_to_alloc = 0;
1756 
1757 	ixgbevf_configure_srrctl(adapter, reg_idx);
1758 
1759 	/* allow any size packet since we can handle overflow */
1760 	rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1761 
1762 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1763 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1764 
1765 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1766 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1767 }
1768 
1769 /**
1770  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1771  * @adapter: board private structure
1772  *
1773  * Configure the Rx unit of the MAC after a reset.
1774  **/
1775 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1776 {
1777 	int i;
1778 	struct ixgbe_hw *hw = &adapter->hw;
1779 	struct net_device *netdev = adapter->netdev;
1780 
1781 	ixgbevf_setup_psrtype(adapter);
1782 	if (hw->mac.type >= ixgbe_mac_X550_vf)
1783 		ixgbevf_setup_vfmrqc(adapter);
1784 
1785 	/* notify the PF of our intent to use this size of frame */
1786 	ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1787 
1788 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1789 	 * the Base and Length of the Rx Descriptor Ring */
1790 	for (i = 0; i < adapter->num_rx_queues; i++)
1791 		ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1792 }
1793 
1794 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1795 				   __be16 proto, u16 vid)
1796 {
1797 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1798 	struct ixgbe_hw *hw = &adapter->hw;
1799 	int err;
1800 
1801 	spin_lock_bh(&adapter->mbx_lock);
1802 
1803 	/* add VID to filter table */
1804 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1805 
1806 	spin_unlock_bh(&adapter->mbx_lock);
1807 
1808 	/* translate error return types so error makes sense */
1809 	if (err == IXGBE_ERR_MBX)
1810 		return -EIO;
1811 
1812 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
1813 		return -EACCES;
1814 
1815 	set_bit(vid, adapter->active_vlans);
1816 
1817 	return err;
1818 }
1819 
1820 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1821 				    __be16 proto, u16 vid)
1822 {
1823 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1824 	struct ixgbe_hw *hw = &adapter->hw;
1825 	int err = -EOPNOTSUPP;
1826 
1827 	spin_lock_bh(&adapter->mbx_lock);
1828 
1829 	/* remove VID from filter table */
1830 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1831 
1832 	spin_unlock_bh(&adapter->mbx_lock);
1833 
1834 	clear_bit(vid, adapter->active_vlans);
1835 
1836 	return err;
1837 }
1838 
1839 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1840 {
1841 	u16 vid;
1842 
1843 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1844 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
1845 					htons(ETH_P_8021Q), vid);
1846 }
1847 
1848 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1849 {
1850 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1851 	struct ixgbe_hw *hw = &adapter->hw;
1852 	int count = 0;
1853 
1854 	if ((netdev_uc_count(netdev)) > 10) {
1855 		pr_err("Too many unicast filters - No Space\n");
1856 		return -ENOSPC;
1857 	}
1858 
1859 	if (!netdev_uc_empty(netdev)) {
1860 		struct netdev_hw_addr *ha;
1861 		netdev_for_each_uc_addr(ha, netdev) {
1862 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1863 			udelay(200);
1864 		}
1865 	} else {
1866 		/*
1867 		 * If the list is empty then send message to PF driver to
1868 		 * clear all macvlans on this VF.
1869 		 */
1870 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1871 	}
1872 
1873 	return count;
1874 }
1875 
1876 /**
1877  * ixgbevf_set_rx_mode - Multicast and unicast set
1878  * @netdev: network interface device structure
1879  *
1880  * The set_rx_method entry point is called whenever the multicast address
1881  * list, unicast address list or the network interface flags are updated.
1882  * This routine is responsible for configuring the hardware for proper
1883  * multicast mode and configuring requested unicast filters.
1884  **/
1885 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1886 {
1887 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1888 	struct ixgbe_hw *hw = &adapter->hw;
1889 
1890 	spin_lock_bh(&adapter->mbx_lock);
1891 
1892 	/* reprogram multicast list */
1893 	hw->mac.ops.update_mc_addr_list(hw, netdev);
1894 
1895 	ixgbevf_write_uc_addr_list(netdev);
1896 
1897 	spin_unlock_bh(&adapter->mbx_lock);
1898 }
1899 
1900 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1901 {
1902 	int q_idx;
1903 	struct ixgbevf_q_vector *q_vector;
1904 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1905 
1906 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1907 		q_vector = adapter->q_vector[q_idx];
1908 #ifdef CONFIG_NET_RX_BUSY_POLL
1909 		ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1910 #endif
1911 		napi_enable(&q_vector->napi);
1912 	}
1913 }
1914 
1915 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1916 {
1917 	int q_idx;
1918 	struct ixgbevf_q_vector *q_vector;
1919 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1920 
1921 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1922 		q_vector = adapter->q_vector[q_idx];
1923 		napi_disable(&q_vector->napi);
1924 #ifdef CONFIG_NET_RX_BUSY_POLL
1925 		while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1926 			pr_info("QV %d locked\n", q_idx);
1927 			usleep_range(1000, 20000);
1928 		}
1929 #endif /* CONFIG_NET_RX_BUSY_POLL */
1930 	}
1931 }
1932 
1933 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1934 {
1935 	struct ixgbe_hw *hw = &adapter->hw;
1936 	unsigned int def_q = 0;
1937 	unsigned int num_tcs = 0;
1938 	unsigned int num_rx_queues = adapter->num_rx_queues;
1939 	unsigned int num_tx_queues = adapter->num_tx_queues;
1940 	int err;
1941 
1942 	spin_lock_bh(&adapter->mbx_lock);
1943 
1944 	/* fetch queue configuration from the PF */
1945 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1946 
1947 	spin_unlock_bh(&adapter->mbx_lock);
1948 
1949 	if (err)
1950 		return err;
1951 
1952 	if (num_tcs > 1) {
1953 		/* we need only one Tx queue */
1954 		num_tx_queues = 1;
1955 
1956 		/* update default Tx ring register index */
1957 		adapter->tx_ring[0]->reg_idx = def_q;
1958 
1959 		/* we need as many queues as traffic classes */
1960 		num_rx_queues = num_tcs;
1961 	}
1962 
1963 	/* if we have a bad config abort request queue reset */
1964 	if ((adapter->num_rx_queues != num_rx_queues) ||
1965 	    (adapter->num_tx_queues != num_tx_queues)) {
1966 		/* force mailbox timeout to prevent further messages */
1967 		hw->mbx.timeout = 0;
1968 
1969 		/* wait for watchdog to come around and bail us out */
1970 		adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1971 	}
1972 
1973 	return 0;
1974 }
1975 
1976 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1977 {
1978 	ixgbevf_configure_dcb(adapter);
1979 
1980 	ixgbevf_set_rx_mode(adapter->netdev);
1981 
1982 	ixgbevf_restore_vlan(adapter);
1983 
1984 	ixgbevf_configure_tx(adapter);
1985 	ixgbevf_configure_rx(adapter);
1986 }
1987 
1988 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1989 {
1990 	/* Only save pre-reset stats if there are some */
1991 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1992 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1993 			adapter->stats.base_vfgprc;
1994 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1995 			adapter->stats.base_vfgptc;
1996 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1997 			adapter->stats.base_vfgorc;
1998 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1999 			adapter->stats.base_vfgotc;
2000 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2001 			adapter->stats.base_vfmprc;
2002 	}
2003 }
2004 
2005 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2006 {
2007 	struct ixgbe_hw *hw = &adapter->hw;
2008 
2009 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2010 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2011 	adapter->stats.last_vfgorc |=
2012 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2013 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2014 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2015 	adapter->stats.last_vfgotc |=
2016 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2017 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2018 
2019 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2020 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2021 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2022 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2023 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2024 }
2025 
2026 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2027 {
2028 	struct ixgbe_hw *hw = &adapter->hw;
2029 	int api[] = { ixgbe_mbox_api_11,
2030 		      ixgbe_mbox_api_10,
2031 		      ixgbe_mbox_api_unknown };
2032 	int err = 0, idx = 0;
2033 
2034 	spin_lock_bh(&adapter->mbx_lock);
2035 
2036 	while (api[idx] != ixgbe_mbox_api_unknown) {
2037 		err = ixgbevf_negotiate_api_version(hw, api[idx]);
2038 		if (!err)
2039 			break;
2040 		idx++;
2041 	}
2042 
2043 	spin_unlock_bh(&adapter->mbx_lock);
2044 }
2045 
2046 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2047 {
2048 	struct net_device *netdev = adapter->netdev;
2049 	struct ixgbe_hw *hw = &adapter->hw;
2050 
2051 	ixgbevf_configure_msix(adapter);
2052 
2053 	spin_lock_bh(&adapter->mbx_lock);
2054 
2055 	if (is_valid_ether_addr(hw->mac.addr))
2056 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2057 	else
2058 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2059 
2060 	spin_unlock_bh(&adapter->mbx_lock);
2061 
2062 	smp_mb__before_atomic();
2063 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2064 	ixgbevf_napi_enable_all(adapter);
2065 
2066 	/* clear any pending interrupts, may auto mask */
2067 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2068 	ixgbevf_irq_enable(adapter);
2069 
2070 	/* enable transmits */
2071 	netif_tx_start_all_queues(netdev);
2072 
2073 	ixgbevf_save_reset_stats(adapter);
2074 	ixgbevf_init_last_counter_stats(adapter);
2075 
2076 	hw->mac.get_link_status = 1;
2077 	mod_timer(&adapter->service_timer, jiffies);
2078 }
2079 
2080 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2081 {
2082 	ixgbevf_configure(adapter);
2083 
2084 	ixgbevf_up_complete(adapter);
2085 }
2086 
2087 /**
2088  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2089  * @rx_ring: ring to free buffers from
2090  **/
2091 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2092 {
2093 	struct device *dev = rx_ring->dev;
2094 	unsigned long size;
2095 	unsigned int i;
2096 
2097 	/* Free Rx ring sk_buff */
2098 	if (rx_ring->skb) {
2099 		dev_kfree_skb(rx_ring->skb);
2100 		rx_ring->skb = NULL;
2101 	}
2102 
2103 	/* ring already cleared, nothing to do */
2104 	if (!rx_ring->rx_buffer_info)
2105 		return;
2106 
2107 	/* Free all the Rx ring pages */
2108 	for (i = 0; i < rx_ring->count; i++) {
2109 		struct ixgbevf_rx_buffer *rx_buffer;
2110 
2111 		rx_buffer = &rx_ring->rx_buffer_info[i];
2112 		if (rx_buffer->dma)
2113 			dma_unmap_page(dev, rx_buffer->dma,
2114 				       PAGE_SIZE, DMA_FROM_DEVICE);
2115 		rx_buffer->dma = 0;
2116 		if (rx_buffer->page)
2117 			__free_page(rx_buffer->page);
2118 		rx_buffer->page = NULL;
2119 	}
2120 
2121 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2122 	memset(rx_ring->rx_buffer_info, 0, size);
2123 
2124 	/* Zero out the descriptor ring */
2125 	memset(rx_ring->desc, 0, rx_ring->size);
2126 }
2127 
2128 /**
2129  * ixgbevf_clean_tx_ring - Free Tx Buffers
2130  * @tx_ring: ring to be cleaned
2131  **/
2132 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2133 {
2134 	struct ixgbevf_tx_buffer *tx_buffer_info;
2135 	unsigned long size;
2136 	unsigned int i;
2137 
2138 	if (!tx_ring->tx_buffer_info)
2139 		return;
2140 
2141 	/* Free all the Tx ring sk_buffs */
2142 	for (i = 0; i < tx_ring->count; i++) {
2143 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2144 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2145 	}
2146 
2147 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2148 	memset(tx_ring->tx_buffer_info, 0, size);
2149 
2150 	memset(tx_ring->desc, 0, tx_ring->size);
2151 }
2152 
2153 /**
2154  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2155  * @adapter: board private structure
2156  **/
2157 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2158 {
2159 	int i;
2160 
2161 	for (i = 0; i < adapter->num_rx_queues; i++)
2162 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2163 }
2164 
2165 /**
2166  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2167  * @adapter: board private structure
2168  **/
2169 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2170 {
2171 	int i;
2172 
2173 	for (i = 0; i < adapter->num_tx_queues; i++)
2174 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2175 }
2176 
2177 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2178 {
2179 	struct net_device *netdev = adapter->netdev;
2180 	struct ixgbe_hw *hw = &adapter->hw;
2181 	int i;
2182 
2183 	/* signal that we are down to the interrupt handler */
2184 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2185 		return; /* do nothing if already down */
2186 
2187 	/* disable all enabled rx queues */
2188 	for (i = 0; i < adapter->num_rx_queues; i++)
2189 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2190 
2191 	usleep_range(10000, 20000);
2192 
2193 	netif_tx_stop_all_queues(netdev);
2194 
2195 	/* call carrier off first to avoid false dev_watchdog timeouts */
2196 	netif_carrier_off(netdev);
2197 	netif_tx_disable(netdev);
2198 
2199 	ixgbevf_irq_disable(adapter);
2200 
2201 	ixgbevf_napi_disable_all(adapter);
2202 
2203 	del_timer_sync(&adapter->service_timer);
2204 
2205 	/* disable transmits in the hardware now that interrupts are off */
2206 	for (i = 0; i < adapter->num_tx_queues; i++) {
2207 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2208 
2209 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2210 				IXGBE_TXDCTL_SWFLSH);
2211 	}
2212 
2213 	if (!pci_channel_offline(adapter->pdev))
2214 		ixgbevf_reset(adapter);
2215 
2216 	ixgbevf_clean_all_tx_rings(adapter);
2217 	ixgbevf_clean_all_rx_rings(adapter);
2218 }
2219 
2220 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2221 {
2222 	WARN_ON(in_interrupt());
2223 
2224 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2225 		msleep(1);
2226 
2227 	ixgbevf_down(adapter);
2228 	ixgbevf_up(adapter);
2229 
2230 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2231 }
2232 
2233 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2234 {
2235 	struct ixgbe_hw *hw = &adapter->hw;
2236 	struct net_device *netdev = adapter->netdev;
2237 
2238 	if (hw->mac.ops.reset_hw(hw)) {
2239 		hw_dbg(hw, "PF still resetting\n");
2240 	} else {
2241 		hw->mac.ops.init_hw(hw);
2242 		ixgbevf_negotiate_api(adapter);
2243 	}
2244 
2245 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2246 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2247 		       netdev->addr_len);
2248 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2249 		       netdev->addr_len);
2250 	}
2251 
2252 	adapter->last_reset = jiffies;
2253 }
2254 
2255 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2256 					int vectors)
2257 {
2258 	int vector_threshold;
2259 
2260 	/* We'll want at least 2 (vector_threshold):
2261 	 * 1) TxQ[0] + RxQ[0] handler
2262 	 * 2) Other (Link Status Change, etc.)
2263 	 */
2264 	vector_threshold = MIN_MSIX_COUNT;
2265 
2266 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2267 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2268 	 * Right now, we simply care about how many we'll get; we'll
2269 	 * set them up later while requesting irq's.
2270 	 */
2271 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2272 					vector_threshold, vectors);
2273 
2274 	if (vectors < 0) {
2275 		dev_err(&adapter->pdev->dev,
2276 			"Unable to allocate MSI-X interrupts\n");
2277 		kfree(adapter->msix_entries);
2278 		adapter->msix_entries = NULL;
2279 		return vectors;
2280 	}
2281 
2282 	/* Adjust for only the vectors we'll use, which is minimum
2283 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2284 	 * vectors we were allocated.
2285 	 */
2286 	adapter->num_msix_vectors = vectors;
2287 
2288 	return 0;
2289 }
2290 
2291 /**
2292  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2293  * @adapter: board private structure to initialize
2294  *
2295  * This is the top level queue allocation routine.  The order here is very
2296  * important, starting with the "most" number of features turned on at once,
2297  * and ending with the smallest set of features.  This way large combinations
2298  * can be allocated if they're turned on, and smaller combinations are the
2299  * fallthrough conditions.
2300  *
2301  **/
2302 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2303 {
2304 	struct ixgbe_hw *hw = &adapter->hw;
2305 	unsigned int def_q = 0;
2306 	unsigned int num_tcs = 0;
2307 	int err;
2308 
2309 	/* Start with base case */
2310 	adapter->num_rx_queues = 1;
2311 	adapter->num_tx_queues = 1;
2312 
2313 	spin_lock_bh(&adapter->mbx_lock);
2314 
2315 	/* fetch queue configuration from the PF */
2316 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2317 
2318 	spin_unlock_bh(&adapter->mbx_lock);
2319 
2320 	if (err)
2321 		return;
2322 
2323 	/* we need as many queues as traffic classes */
2324 	if (num_tcs > 1) {
2325 		adapter->num_rx_queues = num_tcs;
2326 	} else {
2327 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2328 
2329 		switch (hw->api_version) {
2330 		case ixgbe_mbox_api_11:
2331 			adapter->num_rx_queues = rss;
2332 			adapter->num_tx_queues = rss;
2333 		default:
2334 			break;
2335 		}
2336 	}
2337 }
2338 
2339 /**
2340  * ixgbevf_alloc_queues - Allocate memory for all rings
2341  * @adapter: board private structure to initialize
2342  *
2343  * We allocate one ring per queue at run-time since we don't know the
2344  * number of queues at compile-time.  The polling_netdev array is
2345  * intended for Multiqueue, but should work fine with a single queue.
2346  **/
2347 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2348 {
2349 	struct ixgbevf_ring *ring;
2350 	int rx = 0, tx = 0;
2351 
2352 	for (; tx < adapter->num_tx_queues; tx++) {
2353 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2354 		if (!ring)
2355 			goto err_allocation;
2356 
2357 		ring->dev = &adapter->pdev->dev;
2358 		ring->netdev = adapter->netdev;
2359 		ring->count = adapter->tx_ring_count;
2360 		ring->queue_index = tx;
2361 		ring->reg_idx = tx;
2362 
2363 		adapter->tx_ring[tx] = ring;
2364 	}
2365 
2366 	for (; rx < adapter->num_rx_queues; rx++) {
2367 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2368 		if (!ring)
2369 			goto err_allocation;
2370 
2371 		ring->dev = &adapter->pdev->dev;
2372 		ring->netdev = adapter->netdev;
2373 
2374 		ring->count = adapter->rx_ring_count;
2375 		ring->queue_index = rx;
2376 		ring->reg_idx = rx;
2377 
2378 		adapter->rx_ring[rx] = ring;
2379 	}
2380 
2381 	return 0;
2382 
2383 err_allocation:
2384 	while (tx) {
2385 		kfree(adapter->tx_ring[--tx]);
2386 		adapter->tx_ring[tx] = NULL;
2387 	}
2388 
2389 	while (rx) {
2390 		kfree(adapter->rx_ring[--rx]);
2391 		adapter->rx_ring[rx] = NULL;
2392 	}
2393 	return -ENOMEM;
2394 }
2395 
2396 /**
2397  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2398  * @adapter: board private structure to initialize
2399  *
2400  * Attempt to configure the interrupts using the best available
2401  * capabilities of the hardware and the kernel.
2402  **/
2403 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2404 {
2405 	struct net_device *netdev = adapter->netdev;
2406 	int err = 0;
2407 	int vector, v_budget;
2408 
2409 	/*
2410 	 * It's easy to be greedy for MSI-X vectors, but it really
2411 	 * doesn't do us much good if we have a lot more vectors
2412 	 * than CPU's.  So let's be conservative and only ask for
2413 	 * (roughly) the same number of vectors as there are CPU's.
2414 	 * The default is to use pairs of vectors.
2415 	 */
2416 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2417 	v_budget = min_t(int, v_budget, num_online_cpus());
2418 	v_budget += NON_Q_VECTORS;
2419 
2420 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2421 	 * mean we disable MSI-X capabilities of the adapter. */
2422 	adapter->msix_entries = kcalloc(v_budget,
2423 					sizeof(struct msix_entry), GFP_KERNEL);
2424 	if (!adapter->msix_entries) {
2425 		err = -ENOMEM;
2426 		goto out;
2427 	}
2428 
2429 	for (vector = 0; vector < v_budget; vector++)
2430 		adapter->msix_entries[vector].entry = vector;
2431 
2432 	err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2433 	if (err)
2434 		goto out;
2435 
2436 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2437 	if (err)
2438 		goto out;
2439 
2440 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2441 
2442 out:
2443 	return err;
2444 }
2445 
2446 /**
2447  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2448  * @adapter: board private structure to initialize
2449  *
2450  * We allocate one q_vector per queue interrupt.  If allocation fails we
2451  * return -ENOMEM.
2452  **/
2453 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2454 {
2455 	int q_idx, num_q_vectors;
2456 	struct ixgbevf_q_vector *q_vector;
2457 
2458 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2459 
2460 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2461 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2462 		if (!q_vector)
2463 			goto err_out;
2464 		q_vector->adapter = adapter;
2465 		q_vector->v_idx = q_idx;
2466 		netif_napi_add(adapter->netdev, &q_vector->napi,
2467 			       ixgbevf_poll, 64);
2468 #ifdef CONFIG_NET_RX_BUSY_POLL
2469 		napi_hash_add(&q_vector->napi);
2470 #endif
2471 		adapter->q_vector[q_idx] = q_vector;
2472 	}
2473 
2474 	return 0;
2475 
2476 err_out:
2477 	while (q_idx) {
2478 		q_idx--;
2479 		q_vector = adapter->q_vector[q_idx];
2480 #ifdef CONFIG_NET_RX_BUSY_POLL
2481 		napi_hash_del(&q_vector->napi);
2482 #endif
2483 		netif_napi_del(&q_vector->napi);
2484 		kfree(q_vector);
2485 		adapter->q_vector[q_idx] = NULL;
2486 	}
2487 	return -ENOMEM;
2488 }
2489 
2490 /**
2491  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2492  * @adapter: board private structure to initialize
2493  *
2494  * This function frees the memory allocated to the q_vectors.  In addition if
2495  * NAPI is enabled it will delete any references to the NAPI struct prior
2496  * to freeing the q_vector.
2497  **/
2498 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2499 {
2500 	int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2501 
2502 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2503 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2504 
2505 		adapter->q_vector[q_idx] = NULL;
2506 #ifdef CONFIG_NET_RX_BUSY_POLL
2507 		napi_hash_del(&q_vector->napi);
2508 #endif
2509 		netif_napi_del(&q_vector->napi);
2510 		kfree(q_vector);
2511 	}
2512 }
2513 
2514 /**
2515  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2516  * @adapter: board private structure
2517  *
2518  **/
2519 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2520 {
2521 	pci_disable_msix(adapter->pdev);
2522 	kfree(adapter->msix_entries);
2523 	adapter->msix_entries = NULL;
2524 }
2525 
2526 /**
2527  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2528  * @adapter: board private structure to initialize
2529  *
2530  **/
2531 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2532 {
2533 	int err;
2534 
2535 	/* Number of supported queues */
2536 	ixgbevf_set_num_queues(adapter);
2537 
2538 	err = ixgbevf_set_interrupt_capability(adapter);
2539 	if (err) {
2540 		hw_dbg(&adapter->hw,
2541 		       "Unable to setup interrupt capabilities\n");
2542 		goto err_set_interrupt;
2543 	}
2544 
2545 	err = ixgbevf_alloc_q_vectors(adapter);
2546 	if (err) {
2547 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2548 		       "vectors\n");
2549 		goto err_alloc_q_vectors;
2550 	}
2551 
2552 	err = ixgbevf_alloc_queues(adapter);
2553 	if (err) {
2554 		pr_err("Unable to allocate memory for queues\n");
2555 		goto err_alloc_queues;
2556 	}
2557 
2558 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2559 	       "Tx Queue count = %u\n",
2560 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2561 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2562 
2563 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2564 
2565 	return 0;
2566 err_alloc_queues:
2567 	ixgbevf_free_q_vectors(adapter);
2568 err_alloc_q_vectors:
2569 	ixgbevf_reset_interrupt_capability(adapter);
2570 err_set_interrupt:
2571 	return err;
2572 }
2573 
2574 /**
2575  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2576  * @adapter: board private structure to clear interrupt scheme on
2577  *
2578  * We go through and clear interrupt specific resources and reset the structure
2579  * to pre-load conditions
2580  **/
2581 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2582 {
2583 	int i;
2584 
2585 	for (i = 0; i < adapter->num_tx_queues; i++) {
2586 		kfree(adapter->tx_ring[i]);
2587 		adapter->tx_ring[i] = NULL;
2588 	}
2589 	for (i = 0; i < adapter->num_rx_queues; i++) {
2590 		kfree(adapter->rx_ring[i]);
2591 		adapter->rx_ring[i] = NULL;
2592 	}
2593 
2594 	adapter->num_tx_queues = 0;
2595 	adapter->num_rx_queues = 0;
2596 
2597 	ixgbevf_free_q_vectors(adapter);
2598 	ixgbevf_reset_interrupt_capability(adapter);
2599 }
2600 
2601 /**
2602  * ixgbevf_sw_init - Initialize general software structures
2603  * (struct ixgbevf_adapter)
2604  * @adapter: board private structure to initialize
2605  *
2606  * ixgbevf_sw_init initializes the Adapter private data structure.
2607  * Fields are initialized based on PCI device information and
2608  * OS network device settings (MTU size).
2609  **/
2610 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2611 {
2612 	struct ixgbe_hw *hw = &adapter->hw;
2613 	struct pci_dev *pdev = adapter->pdev;
2614 	struct net_device *netdev = adapter->netdev;
2615 	int err;
2616 
2617 	/* PCI config space info */
2618 
2619 	hw->vendor_id = pdev->vendor;
2620 	hw->device_id = pdev->device;
2621 	hw->revision_id = pdev->revision;
2622 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2623 	hw->subsystem_device_id = pdev->subsystem_device;
2624 
2625 	hw->mbx.ops.init_params(hw);
2626 
2627 	/* assume legacy case in which PF would only give VF 2 queues */
2628 	hw->mac.max_tx_queues = 2;
2629 	hw->mac.max_rx_queues = 2;
2630 
2631 	/* lock to protect mailbox accesses */
2632 	spin_lock_init(&adapter->mbx_lock);
2633 
2634 	err = hw->mac.ops.reset_hw(hw);
2635 	if (err) {
2636 		dev_info(&pdev->dev,
2637 			 "PF still in reset state.  Is the PF interface up?\n");
2638 	} else {
2639 		err = hw->mac.ops.init_hw(hw);
2640 		if (err) {
2641 			pr_err("init_shared_code failed: %d\n", err);
2642 			goto out;
2643 		}
2644 		ixgbevf_negotiate_api(adapter);
2645 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2646 		if (err)
2647 			dev_info(&pdev->dev, "Error reading MAC address\n");
2648 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2649 			dev_info(&pdev->dev,
2650 				 "MAC address not assigned by administrator.\n");
2651 		memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2652 	}
2653 
2654 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2655 		dev_info(&pdev->dev, "Assigning random MAC address\n");
2656 		eth_hw_addr_random(netdev);
2657 		memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2658 	}
2659 
2660 	/* Enable dynamic interrupt throttling rates */
2661 	adapter->rx_itr_setting = 1;
2662 	adapter->tx_itr_setting = 1;
2663 
2664 	/* set default ring sizes */
2665 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2666 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2667 
2668 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2669 	return 0;
2670 
2671 out:
2672 	return err;
2673 }
2674 
2675 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2676 	{							\
2677 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2678 		if (current_counter < last_counter)		\
2679 			counter += 0x100000000LL;		\
2680 		last_counter = current_counter;			\
2681 		counter &= 0xFFFFFFFF00000000LL;		\
2682 		counter |= current_counter;			\
2683 	}
2684 
2685 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2686 	{								 \
2687 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2688 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2689 		u64 current_counter = (current_counter_msb << 32) |      \
2690 			current_counter_lsb;                             \
2691 		if (current_counter < last_counter)			 \
2692 			counter += 0x1000000000LL;			 \
2693 		last_counter = current_counter;				 \
2694 		counter &= 0xFFFFFFF000000000LL;			 \
2695 		counter |= current_counter;				 \
2696 	}
2697 /**
2698  * ixgbevf_update_stats - Update the board statistics counters.
2699  * @adapter: board private structure
2700  **/
2701 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2702 {
2703 	struct ixgbe_hw *hw = &adapter->hw;
2704 	int i;
2705 
2706 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2707 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2708 		return;
2709 
2710 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2711 				adapter->stats.vfgprc);
2712 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2713 				adapter->stats.vfgptc);
2714 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2715 				adapter->stats.last_vfgorc,
2716 				adapter->stats.vfgorc);
2717 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2718 				adapter->stats.last_vfgotc,
2719 				adapter->stats.vfgotc);
2720 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2721 				adapter->stats.vfmprc);
2722 
2723 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2724 		adapter->hw_csum_rx_error +=
2725 			adapter->rx_ring[i]->hw_csum_rx_error;
2726 		adapter->rx_ring[i]->hw_csum_rx_error = 0;
2727 	}
2728 }
2729 
2730 /**
2731  * ixgbevf_service_timer - Timer Call-back
2732  * @data: pointer to adapter cast into an unsigned long
2733  **/
2734 static void ixgbevf_service_timer(unsigned long data)
2735 {
2736 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2737 
2738 	/* Reset the timer */
2739 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2740 
2741 	ixgbevf_service_event_schedule(adapter);
2742 }
2743 
2744 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2745 {
2746 	if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2747 		return;
2748 
2749 	adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2750 
2751 	/* If we're already down or resetting, just bail */
2752 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2753 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2754 		return;
2755 
2756 	adapter->tx_timeout_count++;
2757 
2758 	ixgbevf_reinit_locked(adapter);
2759 }
2760 
2761 /* ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2762  * @adapter - pointer to the device adapter structure
2763  *
2764  * This function serves two purposes.  First it strobes the interrupt lines
2765  * in order to make certain interrupts are occurring.  Secondly it sets the
2766  * bits needed to check for TX hangs.  As a result we should immediately
2767  * determine if a hang has occurred.
2768  */
2769 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2770 {
2771 	struct ixgbe_hw *hw = &adapter->hw;
2772 	u32 eics = 0;
2773 	int i;
2774 
2775 	/* If we're down or resetting, just bail */
2776 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2777 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2778 		return;
2779 
2780 	/* Force detection of hung controller */
2781 	if (netif_carrier_ok(adapter->netdev)) {
2782 		for (i = 0; i < adapter->num_tx_queues; i++)
2783 			set_check_for_tx_hang(adapter->tx_ring[i]);
2784 	}
2785 
2786 	/* get one bit for every active tx/rx interrupt vector */
2787 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2788 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2789 
2790 		if (qv->rx.ring || qv->tx.ring)
2791 			eics |= 1 << i;
2792 	}
2793 
2794 	/* Cause software interrupt to ensure rings are cleaned */
2795 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2796 }
2797 
2798 /**
2799  * ixgbevf_watchdog_update_link - update the link status
2800  * @adapter - pointer to the device adapter structure
2801  **/
2802 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2803 {
2804 	struct ixgbe_hw *hw = &adapter->hw;
2805 	u32 link_speed = adapter->link_speed;
2806 	bool link_up = adapter->link_up;
2807 	s32 err;
2808 
2809 	spin_lock_bh(&adapter->mbx_lock);
2810 
2811 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2812 
2813 	spin_unlock_bh(&adapter->mbx_lock);
2814 
2815 	/* if check for link returns error we will need to reset */
2816 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2817 		adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2818 		link_up = false;
2819 	}
2820 
2821 	adapter->link_up = link_up;
2822 	adapter->link_speed = link_speed;
2823 }
2824 
2825 /**
2826  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2827  *				 print link up message
2828  * @adapter - pointer to the device adapter structure
2829  **/
2830 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2831 {
2832 	struct net_device *netdev = adapter->netdev;
2833 
2834 	/* only continue if link was previously down */
2835 	if (netif_carrier_ok(netdev))
2836 		return;
2837 
2838 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2839 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2840 		 "10 Gbps" :
2841 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2842 		 "1 Gbps" :
2843 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2844 		 "100 Mbps" :
2845 		 "unknown speed");
2846 
2847 	netif_carrier_on(netdev);
2848 }
2849 
2850 /**
2851  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2852  *				   print link down message
2853  * @adapter - pointer to the adapter structure
2854  **/
2855 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2856 {
2857 	struct net_device *netdev = adapter->netdev;
2858 
2859 	adapter->link_speed = 0;
2860 
2861 	/* only continue if link was up previously */
2862 	if (!netif_carrier_ok(netdev))
2863 		return;
2864 
2865 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2866 
2867 	netif_carrier_off(netdev);
2868 }
2869 
2870 /**
2871  * ixgbevf_watchdog_subtask - worker thread to bring link up
2872  * @work: pointer to work_struct containing our data
2873  **/
2874 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2875 {
2876 	/* if interface is down do nothing */
2877 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2878 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2879 		return;
2880 
2881 	ixgbevf_watchdog_update_link(adapter);
2882 
2883 	if (adapter->link_up)
2884 		ixgbevf_watchdog_link_is_up(adapter);
2885 	else
2886 		ixgbevf_watchdog_link_is_down(adapter);
2887 
2888 	ixgbevf_update_stats(adapter);
2889 }
2890 
2891 /**
2892  * ixgbevf_service_task - manages and runs subtasks
2893  * @work: pointer to work_struct containing our data
2894  **/
2895 static void ixgbevf_service_task(struct work_struct *work)
2896 {
2897 	struct ixgbevf_adapter *adapter = container_of(work,
2898 						       struct ixgbevf_adapter,
2899 						       service_task);
2900 	struct ixgbe_hw *hw = &adapter->hw;
2901 
2902 	if (IXGBE_REMOVED(hw->hw_addr)) {
2903 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2904 			rtnl_lock();
2905 			ixgbevf_down(adapter);
2906 			rtnl_unlock();
2907 		}
2908 		return;
2909 	}
2910 
2911 	ixgbevf_queue_reset_subtask(adapter);
2912 	ixgbevf_reset_subtask(adapter);
2913 	ixgbevf_watchdog_subtask(adapter);
2914 	ixgbevf_check_hang_subtask(adapter);
2915 
2916 	ixgbevf_service_event_complete(adapter);
2917 }
2918 
2919 /**
2920  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2921  * @tx_ring: Tx descriptor ring for a specific queue
2922  *
2923  * Free all transmit software resources
2924  **/
2925 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2926 {
2927 	ixgbevf_clean_tx_ring(tx_ring);
2928 
2929 	vfree(tx_ring->tx_buffer_info);
2930 	tx_ring->tx_buffer_info = NULL;
2931 
2932 	/* if not set, then don't free */
2933 	if (!tx_ring->desc)
2934 		return;
2935 
2936 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2937 			  tx_ring->dma);
2938 
2939 	tx_ring->desc = NULL;
2940 }
2941 
2942 /**
2943  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2944  * @adapter: board private structure
2945  *
2946  * Free all transmit software resources
2947  **/
2948 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2949 {
2950 	int i;
2951 
2952 	for (i = 0; i < adapter->num_tx_queues; i++)
2953 		if (adapter->tx_ring[i]->desc)
2954 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2955 }
2956 
2957 /**
2958  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2959  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2960  *
2961  * Return 0 on success, negative on failure
2962  **/
2963 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2964 {
2965 	int size;
2966 
2967 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2968 	tx_ring->tx_buffer_info = vzalloc(size);
2969 	if (!tx_ring->tx_buffer_info)
2970 		goto err;
2971 
2972 	/* round up to nearest 4K */
2973 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2974 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2975 
2976 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2977 					   &tx_ring->dma, GFP_KERNEL);
2978 	if (!tx_ring->desc)
2979 		goto err;
2980 
2981 	return 0;
2982 
2983 err:
2984 	vfree(tx_ring->tx_buffer_info);
2985 	tx_ring->tx_buffer_info = NULL;
2986 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2987 	       "descriptor ring\n");
2988 	return -ENOMEM;
2989 }
2990 
2991 /**
2992  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2993  * @adapter: board private structure
2994  *
2995  * If this function returns with an error, then it's possible one or
2996  * more of the rings is populated (while the rest are not).  It is the
2997  * callers duty to clean those orphaned rings.
2998  *
2999  * Return 0 on success, negative on failure
3000  **/
3001 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3002 {
3003 	int i, err = 0;
3004 
3005 	for (i = 0; i < adapter->num_tx_queues; i++) {
3006 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3007 		if (!err)
3008 			continue;
3009 		hw_dbg(&adapter->hw,
3010 		       "Allocation for Tx Queue %u failed\n", i);
3011 		break;
3012 	}
3013 
3014 	return err;
3015 }
3016 
3017 /**
3018  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3019  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
3020  *
3021  * Returns 0 on success, negative on failure
3022  **/
3023 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3024 {
3025 	int size;
3026 
3027 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3028 	rx_ring->rx_buffer_info = vzalloc(size);
3029 	if (!rx_ring->rx_buffer_info)
3030 		goto err;
3031 
3032 	/* Round up to nearest 4K */
3033 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3034 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3035 
3036 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3037 					   &rx_ring->dma, GFP_KERNEL);
3038 
3039 	if (!rx_ring->desc)
3040 		goto err;
3041 
3042 	return 0;
3043 err:
3044 	vfree(rx_ring->rx_buffer_info);
3045 	rx_ring->rx_buffer_info = NULL;
3046 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3047 	return -ENOMEM;
3048 }
3049 
3050 /**
3051  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3052  * @adapter: board private structure
3053  *
3054  * If this function returns with an error, then it's possible one or
3055  * more of the rings is populated (while the rest are not).  It is the
3056  * callers duty to clean those orphaned rings.
3057  *
3058  * Return 0 on success, negative on failure
3059  **/
3060 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3061 {
3062 	int i, err = 0;
3063 
3064 	for (i = 0; i < adapter->num_rx_queues; i++) {
3065 		err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3066 		if (!err)
3067 			continue;
3068 		hw_dbg(&adapter->hw,
3069 		       "Allocation for Rx Queue %u failed\n", i);
3070 		break;
3071 	}
3072 	return err;
3073 }
3074 
3075 /**
3076  * ixgbevf_free_rx_resources - Free Rx Resources
3077  * @rx_ring: ring to clean the resources from
3078  *
3079  * Free all receive software resources
3080  **/
3081 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3082 {
3083 	ixgbevf_clean_rx_ring(rx_ring);
3084 
3085 	vfree(rx_ring->rx_buffer_info);
3086 	rx_ring->rx_buffer_info = NULL;
3087 
3088 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3089 			  rx_ring->dma);
3090 
3091 	rx_ring->desc = NULL;
3092 }
3093 
3094 /**
3095  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3096  * @adapter: board private structure
3097  *
3098  * Free all receive software resources
3099  **/
3100 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3101 {
3102 	int i;
3103 
3104 	for (i = 0; i < adapter->num_rx_queues; i++)
3105 		if (adapter->rx_ring[i]->desc)
3106 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3107 }
3108 
3109 /**
3110  * ixgbevf_open - Called when a network interface is made active
3111  * @netdev: network interface device structure
3112  *
3113  * Returns 0 on success, negative value on failure
3114  *
3115  * The open entry point is called when a network interface is made
3116  * active by the system (IFF_UP).  At this point all resources needed
3117  * for transmit and receive operations are allocated, the interrupt
3118  * handler is registered with the OS, the watchdog timer is started,
3119  * and the stack is notified that the interface is ready.
3120  **/
3121 static int ixgbevf_open(struct net_device *netdev)
3122 {
3123 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3124 	struct ixgbe_hw *hw = &adapter->hw;
3125 	int err;
3126 
3127 	/* A previous failure to open the device because of a lack of
3128 	 * available MSIX vector resources may have reset the number
3129 	 * of msix vectors variable to zero.  The only way to recover
3130 	 * is to unload/reload the driver and hope that the system has
3131 	 * been able to recover some MSIX vector resources.
3132 	 */
3133 	if (!adapter->num_msix_vectors)
3134 		return -ENOMEM;
3135 
3136 	if (hw->adapter_stopped) {
3137 		ixgbevf_reset(adapter);
3138 		/* if adapter is still stopped then PF isn't up and
3139 		 * the vf can't start. */
3140 		if (hw->adapter_stopped) {
3141 			err = IXGBE_ERR_MBX;
3142 			pr_err("Unable to start - perhaps the PF Driver isn't "
3143 			       "up yet\n");
3144 			goto err_setup_reset;
3145 		}
3146 	}
3147 
3148 	/* disallow open during test */
3149 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3150 		return -EBUSY;
3151 
3152 	netif_carrier_off(netdev);
3153 
3154 	/* allocate transmit descriptors */
3155 	err = ixgbevf_setup_all_tx_resources(adapter);
3156 	if (err)
3157 		goto err_setup_tx;
3158 
3159 	/* allocate receive descriptors */
3160 	err = ixgbevf_setup_all_rx_resources(adapter);
3161 	if (err)
3162 		goto err_setup_rx;
3163 
3164 	ixgbevf_configure(adapter);
3165 
3166 	/*
3167 	 * Map the Tx/Rx rings to the vectors we were allotted.
3168 	 * if request_irq will be called in this function map_rings
3169 	 * must be called *before* up_complete
3170 	 */
3171 	ixgbevf_map_rings_to_vectors(adapter);
3172 
3173 	err = ixgbevf_request_irq(adapter);
3174 	if (err)
3175 		goto err_req_irq;
3176 
3177 	ixgbevf_up_complete(adapter);
3178 
3179 	return 0;
3180 
3181 err_req_irq:
3182 	ixgbevf_down(adapter);
3183 err_setup_rx:
3184 	ixgbevf_free_all_rx_resources(adapter);
3185 err_setup_tx:
3186 	ixgbevf_free_all_tx_resources(adapter);
3187 	ixgbevf_reset(adapter);
3188 
3189 err_setup_reset:
3190 
3191 	return err;
3192 }
3193 
3194 /**
3195  * ixgbevf_close - Disables a network interface
3196  * @netdev: network interface device structure
3197  *
3198  * Returns 0, this is not allowed to fail
3199  *
3200  * The close entry point is called when an interface is de-activated
3201  * by the OS.  The hardware is still under the drivers control, but
3202  * needs to be disabled.  A global MAC reset is issued to stop the
3203  * hardware, and all transmit and receive resources are freed.
3204  **/
3205 static int ixgbevf_close(struct net_device *netdev)
3206 {
3207 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3208 
3209 	ixgbevf_down(adapter);
3210 	ixgbevf_free_irq(adapter);
3211 
3212 	ixgbevf_free_all_tx_resources(adapter);
3213 	ixgbevf_free_all_rx_resources(adapter);
3214 
3215 	return 0;
3216 }
3217 
3218 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3219 {
3220 	struct net_device *dev = adapter->netdev;
3221 
3222 	if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3223 		return;
3224 
3225 	adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3226 
3227 	/* if interface is down do nothing */
3228 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3229 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3230 		return;
3231 
3232 	/* Hardware has to reinitialize queues and interrupts to
3233 	 * match packet buffer alignment. Unfortunately, the
3234 	 * hardware is not flexible enough to do this dynamically.
3235 	 */
3236 	if (netif_running(dev))
3237 		ixgbevf_close(dev);
3238 
3239 	ixgbevf_clear_interrupt_scheme(adapter);
3240 	ixgbevf_init_interrupt_scheme(adapter);
3241 
3242 	if (netif_running(dev))
3243 		ixgbevf_open(dev);
3244 }
3245 
3246 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3247 				u32 vlan_macip_lens, u32 type_tucmd,
3248 				u32 mss_l4len_idx)
3249 {
3250 	struct ixgbe_adv_tx_context_desc *context_desc;
3251 	u16 i = tx_ring->next_to_use;
3252 
3253 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3254 
3255 	i++;
3256 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3257 
3258 	/* set bits to identify this as an advanced context descriptor */
3259 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3260 
3261 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3262 	context_desc->seqnum_seed	= 0;
3263 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3264 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3265 }
3266 
3267 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3268 		       struct ixgbevf_tx_buffer *first,
3269 		       u8 *hdr_len)
3270 {
3271 	struct sk_buff *skb = first->skb;
3272 	u32 vlan_macip_lens, type_tucmd;
3273 	u32 mss_l4len_idx, l4len;
3274 	int err;
3275 
3276 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3277 		return 0;
3278 
3279 	if (!skb_is_gso(skb))
3280 		return 0;
3281 
3282 	err = skb_cow_head(skb, 0);
3283 	if (err < 0)
3284 		return err;
3285 
3286 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3287 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3288 
3289 	if (first->protocol == htons(ETH_P_IP)) {
3290 		struct iphdr *iph = ip_hdr(skb);
3291 		iph->tot_len = 0;
3292 		iph->check = 0;
3293 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3294 							 iph->daddr, 0,
3295 							 IPPROTO_TCP,
3296 							 0);
3297 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3298 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3299 				   IXGBE_TX_FLAGS_CSUM |
3300 				   IXGBE_TX_FLAGS_IPV4;
3301 	} else if (skb_is_gso_v6(skb)) {
3302 		ipv6_hdr(skb)->payload_len = 0;
3303 		tcp_hdr(skb)->check =
3304 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3305 				     &ipv6_hdr(skb)->daddr,
3306 				     0, IPPROTO_TCP, 0);
3307 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3308 				   IXGBE_TX_FLAGS_CSUM;
3309 	}
3310 
3311 	/* compute header lengths */
3312 	l4len = tcp_hdrlen(skb);
3313 	*hdr_len += l4len;
3314 	*hdr_len = skb_transport_offset(skb) + l4len;
3315 
3316 	/* update gso size and bytecount with header size */
3317 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3318 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3319 
3320 	/* mss_l4len_id: use 1 as index for TSO */
3321 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3322 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3323 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3324 
3325 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3326 	vlan_macip_lens = skb_network_header_len(skb);
3327 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3328 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3329 
3330 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3331 			    type_tucmd, mss_l4len_idx);
3332 
3333 	return 1;
3334 }
3335 
3336 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3337 			    struct ixgbevf_tx_buffer *first)
3338 {
3339 	struct sk_buff *skb = first->skb;
3340 	u32 vlan_macip_lens = 0;
3341 	u32 mss_l4len_idx = 0;
3342 	u32 type_tucmd = 0;
3343 
3344 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3345 		u8 l4_hdr = 0;
3346 		switch (first->protocol) {
3347 		case htons(ETH_P_IP):
3348 			vlan_macip_lens |= skb_network_header_len(skb);
3349 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3350 			l4_hdr = ip_hdr(skb)->protocol;
3351 			break;
3352 		case htons(ETH_P_IPV6):
3353 			vlan_macip_lens |= skb_network_header_len(skb);
3354 			l4_hdr = ipv6_hdr(skb)->nexthdr;
3355 			break;
3356 		default:
3357 			if (unlikely(net_ratelimit())) {
3358 				dev_warn(tx_ring->dev,
3359 				 "partial checksum but proto=%x!\n",
3360 				 first->protocol);
3361 			}
3362 			break;
3363 		}
3364 
3365 		switch (l4_hdr) {
3366 		case IPPROTO_TCP:
3367 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3368 			mss_l4len_idx = tcp_hdrlen(skb) <<
3369 					IXGBE_ADVTXD_L4LEN_SHIFT;
3370 			break;
3371 		case IPPROTO_SCTP:
3372 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3373 			mss_l4len_idx = sizeof(struct sctphdr) <<
3374 					IXGBE_ADVTXD_L4LEN_SHIFT;
3375 			break;
3376 		case IPPROTO_UDP:
3377 			mss_l4len_idx = sizeof(struct udphdr) <<
3378 					IXGBE_ADVTXD_L4LEN_SHIFT;
3379 			break;
3380 		default:
3381 			if (unlikely(net_ratelimit())) {
3382 				dev_warn(tx_ring->dev,
3383 				 "partial checksum but l4 proto=%x!\n",
3384 				 l4_hdr);
3385 			}
3386 			break;
3387 		}
3388 
3389 		/* update TX checksum flag */
3390 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3391 	}
3392 
3393 	/* vlan_macip_lens: MACLEN, VLAN tag */
3394 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3395 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3396 
3397 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3398 			    type_tucmd, mss_l4len_idx);
3399 }
3400 
3401 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3402 {
3403 	/* set type for advanced descriptor with frame checksum insertion */
3404 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3405 				      IXGBE_ADVTXD_DCMD_IFCS |
3406 				      IXGBE_ADVTXD_DCMD_DEXT);
3407 
3408 	/* set HW vlan bit if vlan is present */
3409 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3410 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3411 
3412 	/* set segmentation enable bits for TSO/FSO */
3413 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3414 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3415 
3416 	return cmd_type;
3417 }
3418 
3419 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3420 				     u32 tx_flags, unsigned int paylen)
3421 {
3422 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3423 
3424 	/* enable L4 checksum for TSO and TX checksum offload */
3425 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3426 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3427 
3428 	/* enble IPv4 checksum for TSO */
3429 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3430 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3431 
3432 	/* use index 1 context for TSO/FSO/FCOE */
3433 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3434 		olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3435 
3436 	/* Check Context must be set if Tx switch is enabled, which it
3437 	 * always is for case where virtual functions are running
3438 	 */
3439 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3440 
3441 	tx_desc->read.olinfo_status = olinfo_status;
3442 }
3443 
3444 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3445 			   struct ixgbevf_tx_buffer *first,
3446 			   const u8 hdr_len)
3447 {
3448 	dma_addr_t dma;
3449 	struct sk_buff *skb = first->skb;
3450 	struct ixgbevf_tx_buffer *tx_buffer;
3451 	union ixgbe_adv_tx_desc *tx_desc;
3452 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3453 	unsigned int data_len = skb->data_len;
3454 	unsigned int size = skb_headlen(skb);
3455 	unsigned int paylen = skb->len - hdr_len;
3456 	u32 tx_flags = first->tx_flags;
3457 	__le32 cmd_type;
3458 	u16 i = tx_ring->next_to_use;
3459 
3460 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3461 
3462 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3463 	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3464 
3465 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3466 	if (dma_mapping_error(tx_ring->dev, dma))
3467 		goto dma_error;
3468 
3469 	/* record length, and DMA address */
3470 	dma_unmap_len_set(first, len, size);
3471 	dma_unmap_addr_set(first, dma, dma);
3472 
3473 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
3474 
3475 	for (;;) {
3476 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3477 			tx_desc->read.cmd_type_len =
3478 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3479 
3480 			i++;
3481 			tx_desc++;
3482 			if (i == tx_ring->count) {
3483 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3484 				i = 0;
3485 			}
3486 
3487 			dma += IXGBE_MAX_DATA_PER_TXD;
3488 			size -= IXGBE_MAX_DATA_PER_TXD;
3489 
3490 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3491 			tx_desc->read.olinfo_status = 0;
3492 		}
3493 
3494 		if (likely(!data_len))
3495 			break;
3496 
3497 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3498 
3499 		i++;
3500 		tx_desc++;
3501 		if (i == tx_ring->count) {
3502 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3503 			i = 0;
3504 		}
3505 
3506 		size = skb_frag_size(frag);
3507 		data_len -= size;
3508 
3509 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3510 				       DMA_TO_DEVICE);
3511 		if (dma_mapping_error(tx_ring->dev, dma))
3512 			goto dma_error;
3513 
3514 		tx_buffer = &tx_ring->tx_buffer_info[i];
3515 		dma_unmap_len_set(tx_buffer, len, size);
3516 		dma_unmap_addr_set(tx_buffer, dma, dma);
3517 
3518 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3519 		tx_desc->read.olinfo_status = 0;
3520 
3521 		frag++;
3522 	}
3523 
3524 	/* write last descriptor with RS and EOP bits */
3525 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3526 	tx_desc->read.cmd_type_len = cmd_type;
3527 
3528 	/* set the timestamp */
3529 	first->time_stamp = jiffies;
3530 
3531 	/* Force memory writes to complete before letting h/w know there
3532 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
3533 	 * memory model archs, such as IA-64).
3534 	 *
3535 	 * We also need this memory barrier (wmb) to make certain all of the
3536 	 * status bits have been updated before next_to_watch is written.
3537 	 */
3538 	wmb();
3539 
3540 	/* set next_to_watch value indicating a packet is present */
3541 	first->next_to_watch = tx_desc;
3542 
3543 	i++;
3544 	if (i == tx_ring->count)
3545 		i = 0;
3546 
3547 	tx_ring->next_to_use = i;
3548 
3549 	/* notify HW of packet */
3550 	ixgbevf_write_tail(tx_ring, i);
3551 
3552 	return;
3553 dma_error:
3554 	dev_err(tx_ring->dev, "TX DMA map failed\n");
3555 
3556 	/* clear dma mappings for failed tx_buffer_info map */
3557 	for (;;) {
3558 		tx_buffer = &tx_ring->tx_buffer_info[i];
3559 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3560 		if (tx_buffer == first)
3561 			break;
3562 		if (i == 0)
3563 			i = tx_ring->count;
3564 		i--;
3565 	}
3566 
3567 	tx_ring->next_to_use = i;
3568 }
3569 
3570 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3571 {
3572 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3573 	/* Herbert's original patch had:
3574 	 *  smp_mb__after_netif_stop_queue();
3575 	 * but since that doesn't exist yet, just open code it. */
3576 	smp_mb();
3577 
3578 	/* We need to check again in a case another CPU has just
3579 	 * made room available. */
3580 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
3581 		return -EBUSY;
3582 
3583 	/* A reprieve! - use start_queue because it doesn't call schedule */
3584 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3585 	++tx_ring->tx_stats.restart_queue;
3586 
3587 	return 0;
3588 }
3589 
3590 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3591 {
3592 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3593 		return 0;
3594 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
3595 }
3596 
3597 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3598 {
3599 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3600 	struct ixgbevf_tx_buffer *first;
3601 	struct ixgbevf_ring *tx_ring;
3602 	int tso;
3603 	u32 tx_flags = 0;
3604 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
3605 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3606 	unsigned short f;
3607 #endif
3608 	u8 hdr_len = 0;
3609 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3610 
3611 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3612 		dev_kfree_skb(skb);
3613 		return NETDEV_TX_OK;
3614 	}
3615 
3616 	tx_ring = adapter->tx_ring[skb->queue_mapping];
3617 
3618 	/*
3619 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3620 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3621 	 *       + 2 desc gap to keep tail from touching head,
3622 	 *       + 1 desc for context descriptor,
3623 	 * otherwise try next time
3624 	 */
3625 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3626 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3627 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3628 #else
3629 	count += skb_shinfo(skb)->nr_frags;
3630 #endif
3631 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3632 		tx_ring->tx_stats.tx_busy++;
3633 		return NETDEV_TX_BUSY;
3634 	}
3635 
3636 	/* record the location of the first descriptor for this packet */
3637 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3638 	first->skb = skb;
3639 	first->bytecount = skb->len;
3640 	first->gso_segs = 1;
3641 
3642 	if (skb_vlan_tag_present(skb)) {
3643 		tx_flags |= skb_vlan_tag_get(skb);
3644 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3645 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3646 	}
3647 
3648 	/* record initial flags and protocol */
3649 	first->tx_flags = tx_flags;
3650 	first->protocol = vlan_get_protocol(skb);
3651 
3652 	tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3653 	if (tso < 0)
3654 		goto out_drop;
3655 	else if (!tso)
3656 		ixgbevf_tx_csum(tx_ring, first);
3657 
3658 	ixgbevf_tx_map(tx_ring, first, hdr_len);
3659 
3660 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3661 
3662 	return NETDEV_TX_OK;
3663 
3664 out_drop:
3665 	dev_kfree_skb_any(first->skb);
3666 	first->skb = NULL;
3667 
3668 	return NETDEV_TX_OK;
3669 }
3670 
3671 /**
3672  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3673  * @netdev: network interface device structure
3674  * @p: pointer to an address structure
3675  *
3676  * Returns 0 on success, negative on failure
3677  **/
3678 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3679 {
3680 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3681 	struct ixgbe_hw *hw = &adapter->hw;
3682 	struct sockaddr *addr = p;
3683 
3684 	if (!is_valid_ether_addr(addr->sa_data))
3685 		return -EADDRNOTAVAIL;
3686 
3687 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3688 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3689 
3690 	spin_lock_bh(&adapter->mbx_lock);
3691 
3692 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3693 
3694 	spin_unlock_bh(&adapter->mbx_lock);
3695 
3696 	return 0;
3697 }
3698 
3699 /**
3700  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3701  * @netdev: network interface device structure
3702  * @new_mtu: new value for maximum frame size
3703  *
3704  * Returns 0 on success, negative on failure
3705  **/
3706 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3707 {
3708 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3709 	struct ixgbe_hw *hw = &adapter->hw;
3710 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3711 	int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3712 
3713 	switch (adapter->hw.api_version) {
3714 	case ixgbe_mbox_api_11:
3715 		max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3716 		break;
3717 	default:
3718 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3719 			max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3720 		break;
3721 	}
3722 
3723 	/* MTU < 68 is an error and causes problems on some kernels */
3724 	if ((new_mtu < 68) || (max_frame > max_possible_frame))
3725 		return -EINVAL;
3726 
3727 	hw_dbg(hw, "changing MTU from %d to %d\n",
3728 	       netdev->mtu, new_mtu);
3729 	/* must set new MTU before calling down or up */
3730 	netdev->mtu = new_mtu;
3731 
3732 	/* notify the PF of our intent to use this size of frame */
3733 	ixgbevf_rlpml_set_vf(hw, max_frame);
3734 
3735 	return 0;
3736 }
3737 
3738 #ifdef CONFIG_NET_POLL_CONTROLLER
3739 /* Polling 'interrupt' - used by things like netconsole to send skbs
3740  * without having to re-enable interrupts. It's not called while
3741  * the interrupt routine is executing.
3742  */
3743 static void ixgbevf_netpoll(struct net_device *netdev)
3744 {
3745 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3746 	int i;
3747 
3748 	/* if interface is down do nothing */
3749 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3750 		return;
3751 	for (i = 0; i < adapter->num_rx_queues; i++)
3752 		ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3753 }
3754 #endif /* CONFIG_NET_POLL_CONTROLLER */
3755 
3756 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3757 {
3758 	struct net_device *netdev = pci_get_drvdata(pdev);
3759 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3760 #ifdef CONFIG_PM
3761 	int retval = 0;
3762 #endif
3763 
3764 	netif_device_detach(netdev);
3765 
3766 	if (netif_running(netdev)) {
3767 		rtnl_lock();
3768 		ixgbevf_down(adapter);
3769 		ixgbevf_free_irq(adapter);
3770 		ixgbevf_free_all_tx_resources(adapter);
3771 		ixgbevf_free_all_rx_resources(adapter);
3772 		rtnl_unlock();
3773 	}
3774 
3775 	ixgbevf_clear_interrupt_scheme(adapter);
3776 
3777 #ifdef CONFIG_PM
3778 	retval = pci_save_state(pdev);
3779 	if (retval)
3780 		return retval;
3781 
3782 #endif
3783 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3784 		pci_disable_device(pdev);
3785 
3786 	return 0;
3787 }
3788 
3789 #ifdef CONFIG_PM
3790 static int ixgbevf_resume(struct pci_dev *pdev)
3791 {
3792 	struct net_device *netdev = pci_get_drvdata(pdev);
3793 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3794 	u32 err;
3795 
3796 	pci_restore_state(pdev);
3797 	/*
3798 	 * pci_restore_state clears dev->state_saved so call
3799 	 * pci_save_state to restore it.
3800 	 */
3801 	pci_save_state(pdev);
3802 
3803 	err = pci_enable_device_mem(pdev);
3804 	if (err) {
3805 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3806 		return err;
3807 	}
3808 	smp_mb__before_atomic();
3809 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3810 	pci_set_master(pdev);
3811 
3812 	ixgbevf_reset(adapter);
3813 
3814 	rtnl_lock();
3815 	err = ixgbevf_init_interrupt_scheme(adapter);
3816 	rtnl_unlock();
3817 	if (err) {
3818 		dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3819 		return err;
3820 	}
3821 
3822 	if (netif_running(netdev)) {
3823 		err = ixgbevf_open(netdev);
3824 		if (err)
3825 			return err;
3826 	}
3827 
3828 	netif_device_attach(netdev);
3829 
3830 	return err;
3831 }
3832 
3833 #endif /* CONFIG_PM */
3834 static void ixgbevf_shutdown(struct pci_dev *pdev)
3835 {
3836 	ixgbevf_suspend(pdev, PMSG_SUSPEND);
3837 }
3838 
3839 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3840 						struct rtnl_link_stats64 *stats)
3841 {
3842 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3843 	unsigned int start;
3844 	u64 bytes, packets;
3845 	const struct ixgbevf_ring *ring;
3846 	int i;
3847 
3848 	ixgbevf_update_stats(adapter);
3849 
3850 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3851 
3852 	for (i = 0; i < adapter->num_rx_queues; i++) {
3853 		ring = adapter->rx_ring[i];
3854 		do {
3855 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3856 			bytes = ring->stats.bytes;
3857 			packets = ring->stats.packets;
3858 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3859 		stats->rx_bytes += bytes;
3860 		stats->rx_packets += packets;
3861 	}
3862 
3863 	for (i = 0; i < adapter->num_tx_queues; i++) {
3864 		ring = adapter->tx_ring[i];
3865 		do {
3866 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3867 			bytes = ring->stats.bytes;
3868 			packets = ring->stats.packets;
3869 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3870 		stats->tx_bytes += bytes;
3871 		stats->tx_packets += packets;
3872 	}
3873 
3874 	return stats;
3875 }
3876 
3877 static const struct net_device_ops ixgbevf_netdev_ops = {
3878 	.ndo_open		= ixgbevf_open,
3879 	.ndo_stop		= ixgbevf_close,
3880 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3881 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3882 	.ndo_get_stats64	= ixgbevf_get_stats,
3883 	.ndo_validate_addr	= eth_validate_addr,
3884 	.ndo_set_mac_address	= ixgbevf_set_mac,
3885 	.ndo_change_mtu		= ixgbevf_change_mtu,
3886 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3887 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3888 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3889 #ifdef CONFIG_NET_RX_BUSY_POLL
3890 	.ndo_busy_poll		= ixgbevf_busy_poll_recv,
3891 #endif
3892 #ifdef CONFIG_NET_POLL_CONTROLLER
3893 	.ndo_poll_controller	= ixgbevf_netpoll,
3894 #endif
3895 };
3896 
3897 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3898 {
3899 	dev->netdev_ops = &ixgbevf_netdev_ops;
3900 	ixgbevf_set_ethtool_ops(dev);
3901 	dev->watchdog_timeo = 5 * HZ;
3902 }
3903 
3904 /**
3905  * ixgbevf_probe - Device Initialization Routine
3906  * @pdev: PCI device information struct
3907  * @ent: entry in ixgbevf_pci_tbl
3908  *
3909  * Returns 0 on success, negative on failure
3910  *
3911  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3912  * The OS initialization, configuring of the adapter private structure,
3913  * and a hardware reset occur.
3914  **/
3915 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3916 {
3917 	struct net_device *netdev;
3918 	struct ixgbevf_adapter *adapter = NULL;
3919 	struct ixgbe_hw *hw = NULL;
3920 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3921 	int err, pci_using_dac;
3922 	bool disable_dev = false;
3923 
3924 	err = pci_enable_device(pdev);
3925 	if (err)
3926 		return err;
3927 
3928 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3929 		pci_using_dac = 1;
3930 	} else {
3931 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3932 		if (err) {
3933 			dev_err(&pdev->dev, "No usable DMA "
3934 				"configuration, aborting\n");
3935 			goto err_dma;
3936 		}
3937 		pci_using_dac = 0;
3938 	}
3939 
3940 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3941 	if (err) {
3942 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3943 		goto err_pci_reg;
3944 	}
3945 
3946 	pci_set_master(pdev);
3947 
3948 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3949 				   MAX_TX_QUEUES);
3950 	if (!netdev) {
3951 		err = -ENOMEM;
3952 		goto err_alloc_etherdev;
3953 	}
3954 
3955 	SET_NETDEV_DEV(netdev, &pdev->dev);
3956 
3957 	adapter = netdev_priv(netdev);
3958 
3959 	adapter->netdev = netdev;
3960 	adapter->pdev = pdev;
3961 	hw = &adapter->hw;
3962 	hw->back = adapter;
3963 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3964 
3965 	/*
3966 	 * call save state here in standalone driver because it relies on
3967 	 * adapter struct to exist, and needs to call netdev_priv
3968 	 */
3969 	pci_save_state(pdev);
3970 
3971 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3972 			      pci_resource_len(pdev, 0));
3973 	adapter->io_addr = hw->hw_addr;
3974 	if (!hw->hw_addr) {
3975 		err = -EIO;
3976 		goto err_ioremap;
3977 	}
3978 
3979 	ixgbevf_assign_netdev_ops(netdev);
3980 
3981 	/* Setup hw api */
3982 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3983 	hw->mac.type  = ii->mac;
3984 
3985 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3986 	       sizeof(struct ixgbe_mbx_operations));
3987 
3988 	/* setup the private structure */
3989 	err = ixgbevf_sw_init(adapter);
3990 	if (err)
3991 		goto err_sw_init;
3992 
3993 	/* The HW MAC address was set and/or determined in sw_init */
3994 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3995 		pr_err("invalid MAC address\n");
3996 		err = -EIO;
3997 		goto err_sw_init;
3998 	}
3999 
4000 	netdev->hw_features = NETIF_F_SG |
4001 			   NETIF_F_IP_CSUM |
4002 			   NETIF_F_IPV6_CSUM |
4003 			   NETIF_F_TSO |
4004 			   NETIF_F_TSO6 |
4005 			   NETIF_F_RXCSUM;
4006 
4007 	netdev->features = netdev->hw_features |
4008 			   NETIF_F_HW_VLAN_CTAG_TX |
4009 			   NETIF_F_HW_VLAN_CTAG_RX |
4010 			   NETIF_F_HW_VLAN_CTAG_FILTER;
4011 
4012 	netdev->vlan_features |= NETIF_F_TSO |
4013 				 NETIF_F_TSO6 |
4014 				 NETIF_F_IP_CSUM |
4015 				 NETIF_F_IPV6_CSUM |
4016 				 NETIF_F_SG;
4017 
4018 	if (pci_using_dac)
4019 		netdev->features |= NETIF_F_HIGHDMA;
4020 
4021 	netdev->priv_flags |= IFF_UNICAST_FLT;
4022 
4023 	if (IXGBE_REMOVED(hw->hw_addr)) {
4024 		err = -EIO;
4025 		goto err_sw_init;
4026 	}
4027 
4028 	setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4029 		    (unsigned long)adapter);
4030 
4031 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4032 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4033 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4034 
4035 	err = ixgbevf_init_interrupt_scheme(adapter);
4036 	if (err)
4037 		goto err_sw_init;
4038 
4039 	strcpy(netdev->name, "eth%d");
4040 
4041 	err = register_netdev(netdev);
4042 	if (err)
4043 		goto err_register;
4044 
4045 	pci_set_drvdata(pdev, netdev);
4046 	netif_carrier_off(netdev);
4047 
4048 	ixgbevf_init_last_counter_stats(adapter);
4049 
4050 	/* print the VF info */
4051 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4052 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4053 
4054 	switch (hw->mac.type) {
4055 	case ixgbe_mac_X550_vf:
4056 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4057 		break;
4058 	case ixgbe_mac_X540_vf:
4059 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4060 		break;
4061 	case ixgbe_mac_82599_vf:
4062 	default:
4063 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4064 		break;
4065 	}
4066 
4067 	return 0;
4068 
4069 err_register:
4070 	ixgbevf_clear_interrupt_scheme(adapter);
4071 err_sw_init:
4072 	ixgbevf_reset_interrupt_capability(adapter);
4073 	iounmap(adapter->io_addr);
4074 err_ioremap:
4075 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4076 	free_netdev(netdev);
4077 err_alloc_etherdev:
4078 	pci_release_regions(pdev);
4079 err_pci_reg:
4080 err_dma:
4081 	if (!adapter || disable_dev)
4082 		pci_disable_device(pdev);
4083 	return err;
4084 }
4085 
4086 /**
4087  * ixgbevf_remove - Device Removal Routine
4088  * @pdev: PCI device information struct
4089  *
4090  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4091  * that it should release a PCI device.  The could be caused by a
4092  * Hot-Plug event, or because the driver is going to be removed from
4093  * memory.
4094  **/
4095 static void ixgbevf_remove(struct pci_dev *pdev)
4096 {
4097 	struct net_device *netdev = pci_get_drvdata(pdev);
4098 	struct ixgbevf_adapter *adapter;
4099 	bool disable_dev;
4100 
4101 	if (!netdev)
4102 		return;
4103 
4104 	adapter = netdev_priv(netdev);
4105 
4106 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4107 	cancel_work_sync(&adapter->service_task);
4108 
4109 	if (netdev->reg_state == NETREG_REGISTERED)
4110 		unregister_netdev(netdev);
4111 
4112 	ixgbevf_clear_interrupt_scheme(adapter);
4113 	ixgbevf_reset_interrupt_capability(adapter);
4114 
4115 	iounmap(adapter->io_addr);
4116 	pci_release_regions(pdev);
4117 
4118 	hw_dbg(&adapter->hw, "Remove complete\n");
4119 
4120 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4121 	free_netdev(netdev);
4122 
4123 	if (disable_dev)
4124 		pci_disable_device(pdev);
4125 }
4126 
4127 /**
4128  * ixgbevf_io_error_detected - called when PCI error is detected
4129  * @pdev: Pointer to PCI device
4130  * @state: The current pci connection state
4131  *
4132  * This function is called after a PCI bus error affecting
4133  * this device has been detected.
4134  */
4135 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4136 						  pci_channel_state_t state)
4137 {
4138 	struct net_device *netdev = pci_get_drvdata(pdev);
4139 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4140 
4141 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4142 		return PCI_ERS_RESULT_DISCONNECT;
4143 
4144 	rtnl_lock();
4145 	netif_device_detach(netdev);
4146 
4147 	if (state == pci_channel_io_perm_failure) {
4148 		rtnl_unlock();
4149 		return PCI_ERS_RESULT_DISCONNECT;
4150 	}
4151 
4152 	if (netif_running(netdev))
4153 		ixgbevf_down(adapter);
4154 
4155 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4156 		pci_disable_device(pdev);
4157 	rtnl_unlock();
4158 
4159 	/* Request a slot slot reset. */
4160 	return PCI_ERS_RESULT_NEED_RESET;
4161 }
4162 
4163 /**
4164  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4165  * @pdev: Pointer to PCI device
4166  *
4167  * Restart the card from scratch, as if from a cold-boot. Implementation
4168  * resembles the first-half of the ixgbevf_resume routine.
4169  */
4170 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4171 {
4172 	struct net_device *netdev = pci_get_drvdata(pdev);
4173 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4174 
4175 	if (pci_enable_device_mem(pdev)) {
4176 		dev_err(&pdev->dev,
4177 			"Cannot re-enable PCI device after reset.\n");
4178 		return PCI_ERS_RESULT_DISCONNECT;
4179 	}
4180 
4181 	smp_mb__before_atomic();
4182 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4183 	pci_set_master(pdev);
4184 
4185 	ixgbevf_reset(adapter);
4186 
4187 	return PCI_ERS_RESULT_RECOVERED;
4188 }
4189 
4190 /**
4191  * ixgbevf_io_resume - called when traffic can start flowing again.
4192  * @pdev: Pointer to PCI device
4193  *
4194  * This callback is called when the error recovery driver tells us that
4195  * its OK to resume normal operation. Implementation resembles the
4196  * second-half of the ixgbevf_resume routine.
4197  */
4198 static void ixgbevf_io_resume(struct pci_dev *pdev)
4199 {
4200 	struct net_device *netdev = pci_get_drvdata(pdev);
4201 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4202 
4203 	if (netif_running(netdev))
4204 		ixgbevf_up(adapter);
4205 
4206 	netif_device_attach(netdev);
4207 }
4208 
4209 /* PCI Error Recovery (ERS) */
4210 static const struct pci_error_handlers ixgbevf_err_handler = {
4211 	.error_detected = ixgbevf_io_error_detected,
4212 	.slot_reset = ixgbevf_io_slot_reset,
4213 	.resume = ixgbevf_io_resume,
4214 };
4215 
4216 static struct pci_driver ixgbevf_driver = {
4217 	.name     = ixgbevf_driver_name,
4218 	.id_table = ixgbevf_pci_tbl,
4219 	.probe    = ixgbevf_probe,
4220 	.remove   = ixgbevf_remove,
4221 #ifdef CONFIG_PM
4222 	/* Power Management Hooks */
4223 	.suspend  = ixgbevf_suspend,
4224 	.resume   = ixgbevf_resume,
4225 #endif
4226 	.shutdown = ixgbevf_shutdown,
4227 	.err_handler = &ixgbevf_err_handler
4228 };
4229 
4230 /**
4231  * ixgbevf_init_module - Driver Registration Routine
4232  *
4233  * ixgbevf_init_module is the first routine called when the driver is
4234  * loaded. All it does is register with the PCI subsystem.
4235  **/
4236 static int __init ixgbevf_init_module(void)
4237 {
4238 	int ret;
4239 	pr_info("%s - version %s\n", ixgbevf_driver_string,
4240 		ixgbevf_driver_version);
4241 
4242 	pr_info("%s\n", ixgbevf_copyright);
4243 
4244 	ret = pci_register_driver(&ixgbevf_driver);
4245 	return ret;
4246 }
4247 
4248 module_init(ixgbevf_init_module);
4249 
4250 /**
4251  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4252  *
4253  * ixgbevf_exit_module is called just before the driver is removed
4254  * from memory.
4255  **/
4256 static void __exit ixgbevf_exit_module(void)
4257 {
4258 	pci_unregister_driver(&ixgbevf_driver);
4259 }
4260 
4261 #ifdef DEBUG
4262 /**
4263  * ixgbevf_get_hw_dev_name - return device name string
4264  * used by hardware layer to print debugging information
4265  **/
4266 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4267 {
4268 	struct ixgbevf_adapter *adapter = hw->back;
4269 	return adapter->netdev->name;
4270 }
4271 
4272 #endif
4273 module_exit(ixgbevf_exit_module);
4274 
4275 /* ixgbevf_main.c */
4276