1 /*******************************************************************************
2 
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2015 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, see <http://www.gnu.org/licenses/>.
17 
18   The full GNU General Public License is included in this distribution in
19   the file called "COPYING".
20 
21   Contact Information:
22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 
25 *******************************************************************************/
26 
27 /******************************************************************************
28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
29 ******************************************************************************/
30 
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32 
33 #include <linux/types.h>
34 #include <linux/bitops.h>
35 #include <linux/module.h>
36 #include <linux/pci.h>
37 #include <linux/netdevice.h>
38 #include <linux/vmalloc.h>
39 #include <linux/string.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/tcp.h>
43 #include <linux/sctp.h>
44 #include <linux/ipv6.h>
45 #include <linux/slab.h>
46 #include <net/checksum.h>
47 #include <net/ip6_checksum.h>
48 #include <linux/ethtool.h>
49 #include <linux/if.h>
50 #include <linux/if_vlan.h>
51 #include <linux/prefetch.h>
52 
53 #include "ixgbevf.h"
54 
55 const char ixgbevf_driver_name[] = "ixgbevf";
56 static const char ixgbevf_driver_string[] =
57 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
58 
59 #define DRV_VERSION "3.2.2-k"
60 const char ixgbevf_driver_version[] = DRV_VERSION;
61 static char ixgbevf_copyright[] =
62 	"Copyright (c) 2009 - 2015 Intel Corporation.";
63 
64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
65 	[board_82599_vf]	= &ixgbevf_82599_vf_info,
66 	[board_82599_vf_hv]	= &ixgbevf_82599_vf_hv_info,
67 	[board_X540_vf]		= &ixgbevf_X540_vf_info,
68 	[board_X540_vf_hv]	= &ixgbevf_X540_vf_hv_info,
69 	[board_X550_vf]		= &ixgbevf_X550_vf_info,
70 	[board_X550_vf_hv]	= &ixgbevf_X550_vf_hv_info,
71 	[board_X550EM_x_vf]	= &ixgbevf_X550EM_x_vf_info,
72 	[board_X550EM_x_vf_hv]	= &ixgbevf_X550EM_x_vf_hv_info,
73 	[board_x550em_a_vf]	= &ixgbevf_x550em_a_vf_info,
74 };
75 
76 /* ixgbevf_pci_tbl - PCI Device ID Table
77  *
78  * Wildcard entries (PCI_ANY_ID) should come last
79  * Last entry must be all 0s
80  *
81  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82  *   Class, Class Mask, private data (not used) }
83  */
84 static const struct pci_device_id ixgbevf_pci_tbl[] = {
85 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
86 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
87 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
88 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
89 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
90 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
91 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
92 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
93 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
94 	/* required last entry */
95 	{0, }
96 };
97 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
98 
99 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
100 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
101 MODULE_LICENSE("GPL");
102 MODULE_VERSION(DRV_VERSION);
103 
104 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
105 static int debug = -1;
106 module_param(debug, int, 0);
107 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
108 
109 static struct workqueue_struct *ixgbevf_wq;
110 
111 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
112 {
113 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
114 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
115 	    !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
116 		queue_work(ixgbevf_wq, &adapter->service_task);
117 }
118 
119 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
120 {
121 	BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
122 
123 	/* flush memory to make sure state is correct before next watchdog */
124 	smp_mb__before_atomic();
125 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
126 }
127 
128 /* forward decls */
129 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
130 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
131 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
132 
133 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
134 {
135 	struct ixgbevf_adapter *adapter = hw->back;
136 
137 	if (!hw->hw_addr)
138 		return;
139 	hw->hw_addr = NULL;
140 	dev_err(&adapter->pdev->dev, "Adapter removed\n");
141 	if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
142 		ixgbevf_service_event_schedule(adapter);
143 }
144 
145 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
146 {
147 	u32 value;
148 
149 	/* The following check not only optimizes a bit by not
150 	 * performing a read on the status register when the
151 	 * register just read was a status register read that
152 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
153 	 * potential recursion.
154 	 */
155 	if (reg == IXGBE_VFSTATUS) {
156 		ixgbevf_remove_adapter(hw);
157 		return;
158 	}
159 	value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
160 	if (value == IXGBE_FAILED_READ_REG)
161 		ixgbevf_remove_adapter(hw);
162 }
163 
164 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
165 {
166 	u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
167 	u32 value;
168 
169 	if (IXGBE_REMOVED(reg_addr))
170 		return IXGBE_FAILED_READ_REG;
171 	value = readl(reg_addr + reg);
172 	if (unlikely(value == IXGBE_FAILED_READ_REG))
173 		ixgbevf_check_remove(hw, reg);
174 	return value;
175 }
176 
177 /**
178  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
179  * @adapter: pointer to adapter struct
180  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
181  * @queue: queue to map the corresponding interrupt to
182  * @msix_vector: the vector to map to the corresponding queue
183  **/
184 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
185 			     u8 queue, u8 msix_vector)
186 {
187 	u32 ivar, index;
188 	struct ixgbe_hw *hw = &adapter->hw;
189 
190 	if (direction == -1) {
191 		/* other causes */
192 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
193 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
194 		ivar &= ~0xFF;
195 		ivar |= msix_vector;
196 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
197 	} else {
198 		/* Tx or Rx causes */
199 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
200 		index = ((16 * (queue & 1)) + (8 * direction));
201 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
202 		ivar &= ~(0xFF << index);
203 		ivar |= (msix_vector << index);
204 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
205 	}
206 }
207 
208 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
209 					struct ixgbevf_tx_buffer *tx_buffer)
210 {
211 	if (tx_buffer->skb) {
212 		dev_kfree_skb_any(tx_buffer->skb);
213 		if (dma_unmap_len(tx_buffer, len))
214 			dma_unmap_single(tx_ring->dev,
215 					 dma_unmap_addr(tx_buffer, dma),
216 					 dma_unmap_len(tx_buffer, len),
217 					 DMA_TO_DEVICE);
218 	} else if (dma_unmap_len(tx_buffer, len)) {
219 		dma_unmap_page(tx_ring->dev,
220 			       dma_unmap_addr(tx_buffer, dma),
221 			       dma_unmap_len(tx_buffer, len),
222 			       DMA_TO_DEVICE);
223 	}
224 	tx_buffer->next_to_watch = NULL;
225 	tx_buffer->skb = NULL;
226 	dma_unmap_len_set(tx_buffer, len, 0);
227 	/* tx_buffer must be completely set up in the transmit path */
228 }
229 
230 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
231 {
232 	return ring->stats.packets;
233 }
234 
235 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
236 {
237 	struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
238 	struct ixgbe_hw *hw = &adapter->hw;
239 
240 	u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
241 	u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
242 
243 	if (head != tail)
244 		return (head < tail) ?
245 			tail - head : (tail + ring->count - head);
246 
247 	return 0;
248 }
249 
250 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
251 {
252 	u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
253 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
254 	u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
255 
256 	clear_check_for_tx_hang(tx_ring);
257 
258 	/* Check for a hung queue, but be thorough. This verifies
259 	 * that a transmit has been completed since the previous
260 	 * check AND there is at least one packet pending. The
261 	 * ARMED bit is set to indicate a potential hang.
262 	 */
263 	if ((tx_done_old == tx_done) && tx_pending) {
264 		/* make sure it is true for two checks in a row */
265 		return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
266 					&tx_ring->state);
267 	}
268 	/* reset the countdown */
269 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
270 
271 	/* update completed stats and continue */
272 	tx_ring->tx_stats.tx_done_old = tx_done;
273 
274 	return false;
275 }
276 
277 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
278 {
279 	/* Do the reset outside of interrupt context */
280 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
281 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
282 		ixgbevf_service_event_schedule(adapter);
283 	}
284 }
285 
286 /**
287  * ixgbevf_tx_timeout - Respond to a Tx Hang
288  * @netdev: network interface device structure
289  **/
290 static void ixgbevf_tx_timeout(struct net_device *netdev)
291 {
292 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
293 
294 	ixgbevf_tx_timeout_reset(adapter);
295 }
296 
297 /**
298  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
299  * @q_vector: board private structure
300  * @tx_ring: tx ring to clean
301  * @napi_budget: Used to determine if we are in netpoll
302  **/
303 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
304 				 struct ixgbevf_ring *tx_ring, int napi_budget)
305 {
306 	struct ixgbevf_adapter *adapter = q_vector->adapter;
307 	struct ixgbevf_tx_buffer *tx_buffer;
308 	union ixgbe_adv_tx_desc *tx_desc;
309 	unsigned int total_bytes = 0, total_packets = 0;
310 	unsigned int budget = tx_ring->count / 2;
311 	unsigned int i = tx_ring->next_to_clean;
312 
313 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
314 		return true;
315 
316 	tx_buffer = &tx_ring->tx_buffer_info[i];
317 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
318 	i -= tx_ring->count;
319 
320 	do {
321 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
322 
323 		/* if next_to_watch is not set then there is no work pending */
324 		if (!eop_desc)
325 			break;
326 
327 		/* prevent any other reads prior to eop_desc */
328 		read_barrier_depends();
329 
330 		/* if DD is not set pending work has not been completed */
331 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
332 			break;
333 
334 		/* clear next_to_watch to prevent false hangs */
335 		tx_buffer->next_to_watch = NULL;
336 
337 		/* update the statistics for this packet */
338 		total_bytes += tx_buffer->bytecount;
339 		total_packets += tx_buffer->gso_segs;
340 
341 		/* free the skb */
342 		napi_consume_skb(tx_buffer->skb, napi_budget);
343 
344 		/* unmap skb header data */
345 		dma_unmap_single(tx_ring->dev,
346 				 dma_unmap_addr(tx_buffer, dma),
347 				 dma_unmap_len(tx_buffer, len),
348 				 DMA_TO_DEVICE);
349 
350 		/* clear tx_buffer data */
351 		tx_buffer->skb = NULL;
352 		dma_unmap_len_set(tx_buffer, len, 0);
353 
354 		/* unmap remaining buffers */
355 		while (tx_desc != eop_desc) {
356 			tx_buffer++;
357 			tx_desc++;
358 			i++;
359 			if (unlikely(!i)) {
360 				i -= tx_ring->count;
361 				tx_buffer = tx_ring->tx_buffer_info;
362 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
363 			}
364 
365 			/* unmap any remaining paged data */
366 			if (dma_unmap_len(tx_buffer, len)) {
367 				dma_unmap_page(tx_ring->dev,
368 					       dma_unmap_addr(tx_buffer, dma),
369 					       dma_unmap_len(tx_buffer, len),
370 					       DMA_TO_DEVICE);
371 				dma_unmap_len_set(tx_buffer, len, 0);
372 			}
373 		}
374 
375 		/* move us one more past the eop_desc for start of next pkt */
376 		tx_buffer++;
377 		tx_desc++;
378 		i++;
379 		if (unlikely(!i)) {
380 			i -= tx_ring->count;
381 			tx_buffer = tx_ring->tx_buffer_info;
382 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
383 		}
384 
385 		/* issue prefetch for next Tx descriptor */
386 		prefetch(tx_desc);
387 
388 		/* update budget accounting */
389 		budget--;
390 	} while (likely(budget));
391 
392 	i += tx_ring->count;
393 	tx_ring->next_to_clean = i;
394 	u64_stats_update_begin(&tx_ring->syncp);
395 	tx_ring->stats.bytes += total_bytes;
396 	tx_ring->stats.packets += total_packets;
397 	u64_stats_update_end(&tx_ring->syncp);
398 	q_vector->tx.total_bytes += total_bytes;
399 	q_vector->tx.total_packets += total_packets;
400 
401 	if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
402 		struct ixgbe_hw *hw = &adapter->hw;
403 		union ixgbe_adv_tx_desc *eop_desc;
404 
405 		eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
406 
407 		pr_err("Detected Tx Unit Hang\n"
408 		       "  Tx Queue             <%d>\n"
409 		       "  TDH, TDT             <%x>, <%x>\n"
410 		       "  next_to_use          <%x>\n"
411 		       "  next_to_clean        <%x>\n"
412 		       "tx_buffer_info[next_to_clean]\n"
413 		       "  next_to_watch        <%p>\n"
414 		       "  eop_desc->wb.status  <%x>\n"
415 		       "  time_stamp           <%lx>\n"
416 		       "  jiffies              <%lx>\n",
417 		       tx_ring->queue_index,
418 		       IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
419 		       IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
420 		       tx_ring->next_to_use, i,
421 		       eop_desc, (eop_desc ? eop_desc->wb.status : 0),
422 		       tx_ring->tx_buffer_info[i].time_stamp, jiffies);
423 
424 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
425 
426 		/* schedule immediate reset if we believe we hung */
427 		ixgbevf_tx_timeout_reset(adapter);
428 
429 		return true;
430 	}
431 
432 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
433 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
434 		     (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
435 		/* Make sure that anybody stopping the queue after this
436 		 * sees the new next_to_clean.
437 		 */
438 		smp_mb();
439 
440 		if (__netif_subqueue_stopped(tx_ring->netdev,
441 					     tx_ring->queue_index) &&
442 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
443 			netif_wake_subqueue(tx_ring->netdev,
444 					    tx_ring->queue_index);
445 			++tx_ring->tx_stats.restart_queue;
446 		}
447 	}
448 
449 	return !!budget;
450 }
451 
452 /**
453  * ixgbevf_rx_skb - Helper function to determine proper Rx method
454  * @q_vector: structure containing interrupt and ring information
455  * @skb: packet to send up
456  **/
457 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
458 			   struct sk_buff *skb)
459 {
460 	napi_gro_receive(&q_vector->napi, skb);
461 }
462 
463 #define IXGBE_RSS_L4_TYPES_MASK \
464 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
465 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
466 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
467 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
468 
469 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
470 				   union ixgbe_adv_rx_desc *rx_desc,
471 				   struct sk_buff *skb)
472 {
473 	u16 rss_type;
474 
475 	if (!(ring->netdev->features & NETIF_F_RXHASH))
476 		return;
477 
478 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
479 		   IXGBE_RXDADV_RSSTYPE_MASK;
480 
481 	if (!rss_type)
482 		return;
483 
484 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
485 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
486 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
487 }
488 
489 /**
490  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
491  * @ring: structure containig ring specific data
492  * @rx_desc: current Rx descriptor being processed
493  * @skb: skb currently being received and modified
494  **/
495 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
496 				       union ixgbe_adv_rx_desc *rx_desc,
497 				       struct sk_buff *skb)
498 {
499 	skb_checksum_none_assert(skb);
500 
501 	/* Rx csum disabled */
502 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
503 		return;
504 
505 	/* if IP and error */
506 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
507 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
508 		ring->rx_stats.csum_err++;
509 		return;
510 	}
511 
512 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
513 		return;
514 
515 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
516 		ring->rx_stats.csum_err++;
517 		return;
518 	}
519 
520 	/* It must be a TCP or UDP packet with a valid checksum */
521 	skb->ip_summed = CHECKSUM_UNNECESSARY;
522 }
523 
524 /**
525  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
526  * @rx_ring: rx descriptor ring packet is being transacted on
527  * @rx_desc: pointer to the EOP Rx descriptor
528  * @skb: pointer to current skb being populated
529  *
530  * This function checks the ring, descriptor, and packet information in
531  * order to populate the checksum, VLAN, protocol, and other fields within
532  * the skb.
533  **/
534 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
535 				       union ixgbe_adv_rx_desc *rx_desc,
536 				       struct sk_buff *skb)
537 {
538 	ixgbevf_rx_hash(rx_ring, rx_desc, skb);
539 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
540 
541 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
542 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
543 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
544 
545 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
546 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
547 	}
548 
549 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
550 }
551 
552 /**
553  * ixgbevf_is_non_eop - process handling of non-EOP buffers
554  * @rx_ring: Rx ring being processed
555  * @rx_desc: Rx descriptor for current buffer
556  * @skb: current socket buffer containing buffer in progress
557  *
558  * This function updates next to clean.  If the buffer is an EOP buffer
559  * this function exits returning false, otherwise it will place the
560  * sk_buff in the next buffer to be chained and return true indicating
561  * that this is in fact a non-EOP buffer.
562  **/
563 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
564 			       union ixgbe_adv_rx_desc *rx_desc)
565 {
566 	u32 ntc = rx_ring->next_to_clean + 1;
567 
568 	/* fetch, update, and store next to clean */
569 	ntc = (ntc < rx_ring->count) ? ntc : 0;
570 	rx_ring->next_to_clean = ntc;
571 
572 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
573 
574 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
575 		return false;
576 
577 	return true;
578 }
579 
580 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
581 				      struct ixgbevf_rx_buffer *bi)
582 {
583 	struct page *page = bi->page;
584 	dma_addr_t dma = bi->dma;
585 
586 	/* since we are recycling buffers we should seldom need to alloc */
587 	if (likely(page))
588 		return true;
589 
590 	/* alloc new page for storage */
591 	page = dev_alloc_page();
592 	if (unlikely(!page)) {
593 		rx_ring->rx_stats.alloc_rx_page_failed++;
594 		return false;
595 	}
596 
597 	/* map page for use */
598 	dma = dma_map_page(rx_ring->dev, page, 0,
599 			   PAGE_SIZE, DMA_FROM_DEVICE);
600 
601 	/* if mapping failed free memory back to system since
602 	 * there isn't much point in holding memory we can't use
603 	 */
604 	if (dma_mapping_error(rx_ring->dev, dma)) {
605 		__free_page(page);
606 
607 		rx_ring->rx_stats.alloc_rx_buff_failed++;
608 		return false;
609 	}
610 
611 	bi->dma = dma;
612 	bi->page = page;
613 	bi->page_offset = 0;
614 
615 	return true;
616 }
617 
618 /**
619  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
620  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
621  * @cleaned_count: number of buffers to replace
622  **/
623 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
624 				     u16 cleaned_count)
625 {
626 	union ixgbe_adv_rx_desc *rx_desc;
627 	struct ixgbevf_rx_buffer *bi;
628 	unsigned int i = rx_ring->next_to_use;
629 
630 	/* nothing to do or no valid netdev defined */
631 	if (!cleaned_count || !rx_ring->netdev)
632 		return;
633 
634 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
635 	bi = &rx_ring->rx_buffer_info[i];
636 	i -= rx_ring->count;
637 
638 	do {
639 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
640 			break;
641 
642 		/* Refresh the desc even if pkt_addr didn't change
643 		 * because each write-back erases this info.
644 		 */
645 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
646 
647 		rx_desc++;
648 		bi++;
649 		i++;
650 		if (unlikely(!i)) {
651 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
652 			bi = rx_ring->rx_buffer_info;
653 			i -= rx_ring->count;
654 		}
655 
656 		/* clear the hdr_addr for the next_to_use descriptor */
657 		rx_desc->read.hdr_addr = 0;
658 
659 		cleaned_count--;
660 	} while (cleaned_count);
661 
662 	i += rx_ring->count;
663 
664 	if (rx_ring->next_to_use != i) {
665 		/* record the next descriptor to use */
666 		rx_ring->next_to_use = i;
667 
668 		/* update next to alloc since we have filled the ring */
669 		rx_ring->next_to_alloc = i;
670 
671 		/* Force memory writes to complete before letting h/w
672 		 * know there are new descriptors to fetch.  (Only
673 		 * applicable for weak-ordered memory model archs,
674 		 * such as IA-64).
675 		 */
676 		wmb();
677 		ixgbevf_write_tail(rx_ring, i);
678 	}
679 }
680 
681 /**
682  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
683  * @rx_ring: rx descriptor ring packet is being transacted on
684  * @rx_desc: pointer to the EOP Rx descriptor
685  * @skb: pointer to current skb being fixed
686  *
687  * Check for corrupted packet headers caused by senders on the local L2
688  * embedded NIC switch not setting up their Tx Descriptors right.  These
689  * should be very rare.
690  *
691  * Also address the case where we are pulling data in on pages only
692  * and as such no data is present in the skb header.
693  *
694  * In addition if skb is not at least 60 bytes we need to pad it so that
695  * it is large enough to qualify as a valid Ethernet frame.
696  *
697  * Returns true if an error was encountered and skb was freed.
698  **/
699 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
700 				    union ixgbe_adv_rx_desc *rx_desc,
701 				    struct sk_buff *skb)
702 {
703 	/* verify that the packet does not have any known errors */
704 	if (unlikely(ixgbevf_test_staterr(rx_desc,
705 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
706 		struct net_device *netdev = rx_ring->netdev;
707 
708 		if (!(netdev->features & NETIF_F_RXALL)) {
709 			dev_kfree_skb_any(skb);
710 			return true;
711 		}
712 	}
713 
714 	/* if eth_skb_pad returns an error the skb was freed */
715 	if (eth_skb_pad(skb))
716 		return true;
717 
718 	return false;
719 }
720 
721 /**
722  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
723  * @rx_ring: rx descriptor ring to store buffers on
724  * @old_buff: donor buffer to have page reused
725  *
726  * Synchronizes page for reuse by the adapter
727  **/
728 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
729 				  struct ixgbevf_rx_buffer *old_buff)
730 {
731 	struct ixgbevf_rx_buffer *new_buff;
732 	u16 nta = rx_ring->next_to_alloc;
733 
734 	new_buff = &rx_ring->rx_buffer_info[nta];
735 
736 	/* update, and store next to alloc */
737 	nta++;
738 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
739 
740 	/* transfer page from old buffer to new buffer */
741 	new_buff->page = old_buff->page;
742 	new_buff->dma = old_buff->dma;
743 	new_buff->page_offset = old_buff->page_offset;
744 
745 	/* sync the buffer for use by the device */
746 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
747 					 new_buff->page_offset,
748 					 IXGBEVF_RX_BUFSZ,
749 					 DMA_FROM_DEVICE);
750 }
751 
752 static inline bool ixgbevf_page_is_reserved(struct page *page)
753 {
754 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
755 }
756 
757 /**
758  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
759  * @rx_ring: rx descriptor ring to transact packets on
760  * @rx_buffer: buffer containing page to add
761  * @rx_desc: descriptor containing length of buffer written by hardware
762  * @skb: sk_buff to place the data into
763  *
764  * This function will add the data contained in rx_buffer->page to the skb.
765  * This is done either through a direct copy if the data in the buffer is
766  * less than the skb header size, otherwise it will just attach the page as
767  * a frag to the skb.
768  *
769  * The function will then update the page offset if necessary and return
770  * true if the buffer can be reused by the adapter.
771  **/
772 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
773 				struct ixgbevf_rx_buffer *rx_buffer,
774 				union ixgbe_adv_rx_desc *rx_desc,
775 				struct sk_buff *skb)
776 {
777 	struct page *page = rx_buffer->page;
778 	unsigned char *va = page_address(page) + rx_buffer->page_offset;
779 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
780 #if (PAGE_SIZE < 8192)
781 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
782 #else
783 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
784 #endif
785 	unsigned int pull_len;
786 
787 	if (unlikely(skb_is_nonlinear(skb)))
788 		goto add_tail_frag;
789 
790 	if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
791 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
792 
793 		/* page is not reserved, we can reuse buffer as is */
794 		if (likely(!ixgbevf_page_is_reserved(page)))
795 			return true;
796 
797 		/* this page cannot be reused so discard it */
798 		put_page(page);
799 		return false;
800 	}
801 
802 	/* we need the header to contain the greater of either ETH_HLEN or
803 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
804 	 */
805 	pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
806 
807 	/* align pull length to size of long to optimize memcpy performance */
808 	memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
809 
810 	/* update all of the pointers */
811 	va += pull_len;
812 	size -= pull_len;
813 
814 add_tail_frag:
815 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
816 			(unsigned long)va & ~PAGE_MASK, size, truesize);
817 
818 	/* avoid re-using remote pages */
819 	if (unlikely(ixgbevf_page_is_reserved(page)))
820 		return false;
821 
822 #if (PAGE_SIZE < 8192)
823 	/* if we are only owner of page we can reuse it */
824 	if (unlikely(page_count(page) != 1))
825 		return false;
826 
827 	/* flip page offset to other buffer */
828 	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
829 
830 #else
831 	/* move offset up to the next cache line */
832 	rx_buffer->page_offset += truesize;
833 
834 	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
835 		return false;
836 
837 #endif
838 	/* Even if we own the page, we are not allowed to use atomic_set()
839 	 * This would break get_page_unless_zero() users.
840 	 */
841 	page_ref_inc(page);
842 
843 	return true;
844 }
845 
846 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
847 					       union ixgbe_adv_rx_desc *rx_desc,
848 					       struct sk_buff *skb)
849 {
850 	struct ixgbevf_rx_buffer *rx_buffer;
851 	struct page *page;
852 
853 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
854 	page = rx_buffer->page;
855 	prefetchw(page);
856 
857 	if (likely(!skb)) {
858 		void *page_addr = page_address(page) +
859 				  rx_buffer->page_offset;
860 
861 		/* prefetch first cache line of first page */
862 		prefetch(page_addr);
863 #if L1_CACHE_BYTES < 128
864 		prefetch(page_addr + L1_CACHE_BYTES);
865 #endif
866 
867 		/* allocate a skb to store the frags */
868 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
869 						IXGBEVF_RX_HDR_SIZE);
870 		if (unlikely(!skb)) {
871 			rx_ring->rx_stats.alloc_rx_buff_failed++;
872 			return NULL;
873 		}
874 
875 		/* we will be copying header into skb->data in
876 		 * pskb_may_pull so it is in our interest to prefetch
877 		 * it now to avoid a possible cache miss
878 		 */
879 		prefetchw(skb->data);
880 	}
881 
882 	/* we are reusing so sync this buffer for CPU use */
883 	dma_sync_single_range_for_cpu(rx_ring->dev,
884 				      rx_buffer->dma,
885 				      rx_buffer->page_offset,
886 				      IXGBEVF_RX_BUFSZ,
887 				      DMA_FROM_DEVICE);
888 
889 	/* pull page into skb */
890 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
891 		/* hand second half of page back to the ring */
892 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
893 	} else {
894 		/* we are not reusing the buffer so unmap it */
895 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
896 			       PAGE_SIZE, DMA_FROM_DEVICE);
897 	}
898 
899 	/* clear contents of buffer_info */
900 	rx_buffer->dma = 0;
901 	rx_buffer->page = NULL;
902 
903 	return skb;
904 }
905 
906 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
907 					     u32 qmask)
908 {
909 	struct ixgbe_hw *hw = &adapter->hw;
910 
911 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
912 }
913 
914 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
915 				struct ixgbevf_ring *rx_ring,
916 				int budget)
917 {
918 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
919 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
920 	struct sk_buff *skb = rx_ring->skb;
921 
922 	while (likely(total_rx_packets < budget)) {
923 		union ixgbe_adv_rx_desc *rx_desc;
924 
925 		/* return some buffers to hardware, one at a time is too slow */
926 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
927 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
928 			cleaned_count = 0;
929 		}
930 
931 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
932 
933 		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
934 			break;
935 
936 		/* This memory barrier is needed to keep us from reading
937 		 * any other fields out of the rx_desc until we know the
938 		 * RXD_STAT_DD bit is set
939 		 */
940 		rmb();
941 
942 		/* retrieve a buffer from the ring */
943 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
944 
945 		/* exit if we failed to retrieve a buffer */
946 		if (!skb)
947 			break;
948 
949 		cleaned_count++;
950 
951 		/* fetch next buffer in frame if non-eop */
952 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
953 			continue;
954 
955 		/* verify the packet layout is correct */
956 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
957 			skb = NULL;
958 			continue;
959 		}
960 
961 		/* probably a little skewed due to removing CRC */
962 		total_rx_bytes += skb->len;
963 
964 		/* Workaround hardware that can't do proper VEPA multicast
965 		 * source pruning.
966 		 */
967 		if ((skb->pkt_type == PACKET_BROADCAST ||
968 		     skb->pkt_type == PACKET_MULTICAST) &&
969 		    ether_addr_equal(rx_ring->netdev->dev_addr,
970 				     eth_hdr(skb)->h_source)) {
971 			dev_kfree_skb_irq(skb);
972 			continue;
973 		}
974 
975 		/* populate checksum, VLAN, and protocol */
976 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
977 
978 		ixgbevf_rx_skb(q_vector, skb);
979 
980 		/* reset skb pointer */
981 		skb = NULL;
982 
983 		/* update budget accounting */
984 		total_rx_packets++;
985 	}
986 
987 	/* place incomplete frames back on ring for completion */
988 	rx_ring->skb = skb;
989 
990 	u64_stats_update_begin(&rx_ring->syncp);
991 	rx_ring->stats.packets += total_rx_packets;
992 	rx_ring->stats.bytes += total_rx_bytes;
993 	u64_stats_update_end(&rx_ring->syncp);
994 	q_vector->rx.total_packets += total_rx_packets;
995 	q_vector->rx.total_bytes += total_rx_bytes;
996 
997 	return total_rx_packets;
998 }
999 
1000 /**
1001  * ixgbevf_poll - NAPI polling calback
1002  * @napi: napi struct with our devices info in it
1003  * @budget: amount of work driver is allowed to do this pass, in packets
1004  *
1005  * This function will clean more than one or more rings associated with a
1006  * q_vector.
1007  **/
1008 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1009 {
1010 	struct ixgbevf_q_vector *q_vector =
1011 		container_of(napi, struct ixgbevf_q_vector, napi);
1012 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1013 	struct ixgbevf_ring *ring;
1014 	int per_ring_budget, work_done = 0;
1015 	bool clean_complete = true;
1016 
1017 	ixgbevf_for_each_ring(ring, q_vector->tx) {
1018 		if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1019 			clean_complete = false;
1020 	}
1021 
1022 	if (budget <= 0)
1023 		return budget;
1024 
1025 	/* attempt to distribute budget to each queue fairly, but don't allow
1026 	 * the budget to go below 1 because we'll exit polling
1027 	 */
1028 	if (q_vector->rx.count > 1)
1029 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1030 	else
1031 		per_ring_budget = budget;
1032 
1033 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1034 		int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1035 						   per_ring_budget);
1036 		work_done += cleaned;
1037 		if (cleaned >= per_ring_budget)
1038 			clean_complete = false;
1039 	}
1040 
1041 	/* If all work not completed, return budget and keep polling */
1042 	if (!clean_complete)
1043 		return budget;
1044 	/* all work done, exit the polling mode */
1045 	napi_complete_done(napi, work_done);
1046 	if (adapter->rx_itr_setting == 1)
1047 		ixgbevf_set_itr(q_vector);
1048 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1049 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1050 		ixgbevf_irq_enable_queues(adapter,
1051 					  BIT(q_vector->v_idx));
1052 
1053 	return 0;
1054 }
1055 
1056 /**
1057  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1058  * @q_vector: structure containing interrupt and ring information
1059  **/
1060 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1061 {
1062 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1063 	struct ixgbe_hw *hw = &adapter->hw;
1064 	int v_idx = q_vector->v_idx;
1065 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1066 
1067 	/* set the WDIS bit to not clear the timer bits and cause an
1068 	 * immediate assertion of the interrupt
1069 	 */
1070 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1071 
1072 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1073 }
1074 
1075 /**
1076  * ixgbevf_configure_msix - Configure MSI-X hardware
1077  * @adapter: board private structure
1078  *
1079  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1080  * interrupts.
1081  **/
1082 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1083 {
1084 	struct ixgbevf_q_vector *q_vector;
1085 	int q_vectors, v_idx;
1086 
1087 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1088 	adapter->eims_enable_mask = 0;
1089 
1090 	/* Populate the IVAR table and set the ITR values to the
1091 	 * corresponding register.
1092 	 */
1093 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1094 		struct ixgbevf_ring *ring;
1095 
1096 		q_vector = adapter->q_vector[v_idx];
1097 
1098 		ixgbevf_for_each_ring(ring, q_vector->rx)
1099 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1100 
1101 		ixgbevf_for_each_ring(ring, q_vector->tx)
1102 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1103 
1104 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1105 			/* Tx only vector */
1106 			if (adapter->tx_itr_setting == 1)
1107 				q_vector->itr = IXGBE_12K_ITR;
1108 			else
1109 				q_vector->itr = adapter->tx_itr_setting;
1110 		} else {
1111 			/* Rx or Rx/Tx vector */
1112 			if (adapter->rx_itr_setting == 1)
1113 				q_vector->itr = IXGBE_20K_ITR;
1114 			else
1115 				q_vector->itr = adapter->rx_itr_setting;
1116 		}
1117 
1118 		/* add q_vector eims value to global eims_enable_mask */
1119 		adapter->eims_enable_mask |= BIT(v_idx);
1120 
1121 		ixgbevf_write_eitr(q_vector);
1122 	}
1123 
1124 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1125 	/* setup eims_other and add value to global eims_enable_mask */
1126 	adapter->eims_other = BIT(v_idx);
1127 	adapter->eims_enable_mask |= adapter->eims_other;
1128 }
1129 
1130 enum latency_range {
1131 	lowest_latency = 0,
1132 	low_latency = 1,
1133 	bulk_latency = 2,
1134 	latency_invalid = 255
1135 };
1136 
1137 /**
1138  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1139  * @q_vector: structure containing interrupt and ring information
1140  * @ring_container: structure containing ring performance data
1141  *
1142  * Stores a new ITR value based on packets and byte
1143  * counts during the last interrupt.  The advantage of per interrupt
1144  * computation is faster updates and more accurate ITR for the current
1145  * traffic pattern.  Constants in this function were computed
1146  * based on theoretical maximum wire speed and thresholds were set based
1147  * on testing data as well as attempting to minimize response time
1148  * while increasing bulk throughput.
1149  **/
1150 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1151 			       struct ixgbevf_ring_container *ring_container)
1152 {
1153 	int bytes = ring_container->total_bytes;
1154 	int packets = ring_container->total_packets;
1155 	u32 timepassed_us;
1156 	u64 bytes_perint;
1157 	u8 itr_setting = ring_container->itr;
1158 
1159 	if (packets == 0)
1160 		return;
1161 
1162 	/* simple throttle rate management
1163 	 *    0-20MB/s lowest (100000 ints/s)
1164 	 *   20-100MB/s low   (20000 ints/s)
1165 	 *  100-1249MB/s bulk (12000 ints/s)
1166 	 */
1167 	/* what was last interrupt timeslice? */
1168 	timepassed_us = q_vector->itr >> 2;
1169 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1170 
1171 	switch (itr_setting) {
1172 	case lowest_latency:
1173 		if (bytes_perint > 10)
1174 			itr_setting = low_latency;
1175 		break;
1176 	case low_latency:
1177 		if (bytes_perint > 20)
1178 			itr_setting = bulk_latency;
1179 		else if (bytes_perint <= 10)
1180 			itr_setting = lowest_latency;
1181 		break;
1182 	case bulk_latency:
1183 		if (bytes_perint <= 20)
1184 			itr_setting = low_latency;
1185 		break;
1186 	}
1187 
1188 	/* clear work counters since we have the values we need */
1189 	ring_container->total_bytes = 0;
1190 	ring_container->total_packets = 0;
1191 
1192 	/* write updated itr to ring container */
1193 	ring_container->itr = itr_setting;
1194 }
1195 
1196 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1197 {
1198 	u32 new_itr = q_vector->itr;
1199 	u8 current_itr;
1200 
1201 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1202 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1203 
1204 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1205 
1206 	switch (current_itr) {
1207 	/* counts and packets in update_itr are dependent on these numbers */
1208 	case lowest_latency:
1209 		new_itr = IXGBE_100K_ITR;
1210 		break;
1211 	case low_latency:
1212 		new_itr = IXGBE_20K_ITR;
1213 		break;
1214 	case bulk_latency:
1215 		new_itr = IXGBE_12K_ITR;
1216 		break;
1217 	default:
1218 		break;
1219 	}
1220 
1221 	if (new_itr != q_vector->itr) {
1222 		/* do an exponential smoothing */
1223 		new_itr = (10 * new_itr * q_vector->itr) /
1224 			  ((9 * new_itr) + q_vector->itr);
1225 
1226 		/* save the algorithm value here */
1227 		q_vector->itr = new_itr;
1228 
1229 		ixgbevf_write_eitr(q_vector);
1230 	}
1231 }
1232 
1233 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1234 {
1235 	struct ixgbevf_adapter *adapter = data;
1236 	struct ixgbe_hw *hw = &adapter->hw;
1237 
1238 	hw->mac.get_link_status = 1;
1239 
1240 	ixgbevf_service_event_schedule(adapter);
1241 
1242 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1243 
1244 	return IRQ_HANDLED;
1245 }
1246 
1247 /**
1248  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1249  * @irq: unused
1250  * @data: pointer to our q_vector struct for this interrupt vector
1251  **/
1252 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1253 {
1254 	struct ixgbevf_q_vector *q_vector = data;
1255 
1256 	/* EIAM disabled interrupts (on this vector) for us */
1257 	if (q_vector->rx.ring || q_vector->tx.ring)
1258 		napi_schedule_irqoff(&q_vector->napi);
1259 
1260 	return IRQ_HANDLED;
1261 }
1262 
1263 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1264 				     int r_idx)
1265 {
1266 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1267 
1268 	a->rx_ring[r_idx]->next = q_vector->rx.ring;
1269 	q_vector->rx.ring = a->rx_ring[r_idx];
1270 	q_vector->rx.count++;
1271 }
1272 
1273 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1274 				     int t_idx)
1275 {
1276 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1277 
1278 	a->tx_ring[t_idx]->next = q_vector->tx.ring;
1279 	q_vector->tx.ring = a->tx_ring[t_idx];
1280 	q_vector->tx.count++;
1281 }
1282 
1283 /**
1284  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1285  * @adapter: board private structure to initialize
1286  *
1287  * This function maps descriptor rings to the queue-specific vectors
1288  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1289  * one vector per ring/queue, but on a constrained vector budget, we
1290  * group the rings as "efficiently" as possible.  You would add new
1291  * mapping configurations in here.
1292  **/
1293 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1294 {
1295 	int q_vectors;
1296 	int v_start = 0;
1297 	int rxr_idx = 0, txr_idx = 0;
1298 	int rxr_remaining = adapter->num_rx_queues;
1299 	int txr_remaining = adapter->num_tx_queues;
1300 	int i, j;
1301 	int rqpv, tqpv;
1302 
1303 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1304 
1305 	/* The ideal configuration...
1306 	 * We have enough vectors to map one per queue.
1307 	 */
1308 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1309 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1310 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1311 
1312 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1313 			map_vector_to_txq(adapter, v_start, txr_idx);
1314 		return 0;
1315 	}
1316 
1317 	/* If we don't have enough vectors for a 1-to-1
1318 	 * mapping, we'll have to group them so there are
1319 	 * multiple queues per vector.
1320 	 */
1321 	/* Re-adjusting *qpv takes care of the remainder. */
1322 	for (i = v_start; i < q_vectors; i++) {
1323 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1324 		for (j = 0; j < rqpv; j++) {
1325 			map_vector_to_rxq(adapter, i, rxr_idx);
1326 			rxr_idx++;
1327 			rxr_remaining--;
1328 		}
1329 	}
1330 	for (i = v_start; i < q_vectors; i++) {
1331 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1332 		for (j = 0; j < tqpv; j++) {
1333 			map_vector_to_txq(adapter, i, txr_idx);
1334 			txr_idx++;
1335 			txr_remaining--;
1336 		}
1337 	}
1338 
1339 	return 0;
1340 }
1341 
1342 /**
1343  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1344  * @adapter: board private structure
1345  *
1346  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1347  * interrupts from the kernel.
1348  **/
1349 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1350 {
1351 	struct net_device *netdev = adapter->netdev;
1352 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1353 	int vector, err;
1354 	int ri = 0, ti = 0;
1355 
1356 	for (vector = 0; vector < q_vectors; vector++) {
1357 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1358 		struct msix_entry *entry = &adapter->msix_entries[vector];
1359 
1360 		if (q_vector->tx.ring && q_vector->rx.ring) {
1361 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1362 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
1363 			ti++;
1364 		} else if (q_vector->rx.ring) {
1365 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1366 				 "%s-%s-%d", netdev->name, "rx", ri++);
1367 		} else if (q_vector->tx.ring) {
1368 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1369 				 "%s-%s-%d", netdev->name, "tx", ti++);
1370 		} else {
1371 			/* skip this unused q_vector */
1372 			continue;
1373 		}
1374 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1375 				  q_vector->name, q_vector);
1376 		if (err) {
1377 			hw_dbg(&adapter->hw,
1378 			       "request_irq failed for MSIX interrupt Error: %d\n",
1379 			       err);
1380 			goto free_queue_irqs;
1381 		}
1382 	}
1383 
1384 	err = request_irq(adapter->msix_entries[vector].vector,
1385 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1386 	if (err) {
1387 		hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1388 		       err);
1389 		goto free_queue_irqs;
1390 	}
1391 
1392 	return 0;
1393 
1394 free_queue_irqs:
1395 	while (vector) {
1396 		vector--;
1397 		free_irq(adapter->msix_entries[vector].vector,
1398 			 adapter->q_vector[vector]);
1399 	}
1400 	/* This failure is non-recoverable - it indicates the system is
1401 	 * out of MSIX vector resources and the VF driver cannot run
1402 	 * without them.  Set the number of msix vectors to zero
1403 	 * indicating that not enough can be allocated.  The error
1404 	 * will be returned to the user indicating device open failed.
1405 	 * Any further attempts to force the driver to open will also
1406 	 * fail.  The only way to recover is to unload the driver and
1407 	 * reload it again.  If the system has recovered some MSIX
1408 	 * vectors then it may succeed.
1409 	 */
1410 	adapter->num_msix_vectors = 0;
1411 	return err;
1412 }
1413 
1414 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1415 {
1416 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1417 
1418 	for (i = 0; i < q_vectors; i++) {
1419 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1420 
1421 		q_vector->rx.ring = NULL;
1422 		q_vector->tx.ring = NULL;
1423 		q_vector->rx.count = 0;
1424 		q_vector->tx.count = 0;
1425 	}
1426 }
1427 
1428 /**
1429  * ixgbevf_request_irq - initialize interrupts
1430  * @adapter: board private structure
1431  *
1432  * Attempts to configure interrupts using the best available
1433  * capabilities of the hardware and kernel.
1434  **/
1435 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1436 {
1437 	int err = ixgbevf_request_msix_irqs(adapter);
1438 
1439 	if (err)
1440 		hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1441 
1442 	return err;
1443 }
1444 
1445 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1446 {
1447 	int i, q_vectors;
1448 
1449 	if (!adapter->msix_entries)
1450 		return;
1451 
1452 	q_vectors = adapter->num_msix_vectors;
1453 	i = q_vectors - 1;
1454 
1455 	free_irq(adapter->msix_entries[i].vector, adapter);
1456 	i--;
1457 
1458 	for (; i >= 0; i--) {
1459 		/* free only the irqs that were actually requested */
1460 		if (!adapter->q_vector[i]->rx.ring &&
1461 		    !adapter->q_vector[i]->tx.ring)
1462 			continue;
1463 
1464 		free_irq(adapter->msix_entries[i].vector,
1465 			 adapter->q_vector[i]);
1466 	}
1467 
1468 	ixgbevf_reset_q_vectors(adapter);
1469 }
1470 
1471 /**
1472  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1473  * @adapter: board private structure
1474  **/
1475 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1476 {
1477 	struct ixgbe_hw *hw = &adapter->hw;
1478 	int i;
1479 
1480 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1481 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1482 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1483 
1484 	IXGBE_WRITE_FLUSH(hw);
1485 
1486 	for (i = 0; i < adapter->num_msix_vectors; i++)
1487 		synchronize_irq(adapter->msix_entries[i].vector);
1488 }
1489 
1490 /**
1491  * ixgbevf_irq_enable - Enable default interrupt generation settings
1492  * @adapter: board private structure
1493  **/
1494 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1495 {
1496 	struct ixgbe_hw *hw = &adapter->hw;
1497 
1498 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1499 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1500 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1501 }
1502 
1503 /**
1504  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1505  * @adapter: board private structure
1506  * @ring: structure containing ring specific data
1507  *
1508  * Configure the Tx descriptor ring after a reset.
1509  **/
1510 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1511 				      struct ixgbevf_ring *ring)
1512 {
1513 	struct ixgbe_hw *hw = &adapter->hw;
1514 	u64 tdba = ring->dma;
1515 	int wait_loop = 10;
1516 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1517 	u8 reg_idx = ring->reg_idx;
1518 
1519 	/* disable queue to avoid issues while updating state */
1520 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1521 	IXGBE_WRITE_FLUSH(hw);
1522 
1523 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1524 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1525 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1526 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1527 
1528 	/* disable head writeback */
1529 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1530 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1531 
1532 	/* enable relaxed ordering */
1533 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1534 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1535 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1536 
1537 	/* reset head and tail pointers */
1538 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1539 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1540 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1541 
1542 	/* reset ntu and ntc to place SW in sync with hardwdare */
1543 	ring->next_to_clean = 0;
1544 	ring->next_to_use = 0;
1545 
1546 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1547 	 * to or less than the number of on chip descriptors, which is
1548 	 * currently 40.
1549 	 */
1550 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1551 
1552 	/* Setting PTHRESH to 32 both improves performance */
1553 	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1554 		   32;           /* PTHRESH = 32 */
1555 
1556 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1557 
1558 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1559 
1560 	/* poll to verify queue is enabled */
1561 	do {
1562 		usleep_range(1000, 2000);
1563 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1564 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1565 	if (!wait_loop)
1566 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1567 }
1568 
1569 /**
1570  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1571  * @adapter: board private structure
1572  *
1573  * Configure the Tx unit of the MAC after a reset.
1574  **/
1575 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1576 {
1577 	u32 i;
1578 
1579 	/* Setup the HW Tx Head and Tail descriptor pointers */
1580 	for (i = 0; i < adapter->num_tx_queues; i++)
1581 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1582 }
1583 
1584 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1585 
1586 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1587 {
1588 	struct ixgbe_hw *hw = &adapter->hw;
1589 	u32 srrctl;
1590 
1591 	srrctl = IXGBE_SRRCTL_DROP_EN;
1592 
1593 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1594 	srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1595 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1596 
1597 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1598 }
1599 
1600 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1601 {
1602 	struct ixgbe_hw *hw = &adapter->hw;
1603 
1604 	/* PSRTYPE must be initialized in 82599 */
1605 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1606 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1607 		      IXGBE_PSRTYPE_L2HDR;
1608 
1609 	if (adapter->num_rx_queues > 1)
1610 		psrtype |= BIT(29);
1611 
1612 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1613 }
1614 
1615 #define IXGBEVF_MAX_RX_DESC_POLL 10
1616 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1617 				     struct ixgbevf_ring *ring)
1618 {
1619 	struct ixgbe_hw *hw = &adapter->hw;
1620 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1621 	u32 rxdctl;
1622 	u8 reg_idx = ring->reg_idx;
1623 
1624 	if (IXGBE_REMOVED(hw->hw_addr))
1625 		return;
1626 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1627 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1628 
1629 	/* write value back with RXDCTL.ENABLE bit cleared */
1630 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1631 
1632 	/* the hardware may take up to 100us to really disable the Rx queue */
1633 	do {
1634 		udelay(10);
1635 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1636 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1637 
1638 	if (!wait_loop)
1639 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1640 		       reg_idx);
1641 }
1642 
1643 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1644 					 struct ixgbevf_ring *ring)
1645 {
1646 	struct ixgbe_hw *hw = &adapter->hw;
1647 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1648 	u32 rxdctl;
1649 	u8 reg_idx = ring->reg_idx;
1650 
1651 	if (IXGBE_REMOVED(hw->hw_addr))
1652 		return;
1653 	do {
1654 		usleep_range(1000, 2000);
1655 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1656 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1657 
1658 	if (!wait_loop)
1659 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1660 		       reg_idx);
1661 }
1662 
1663 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1664 {
1665 	struct ixgbe_hw *hw = &adapter->hw;
1666 	u32 vfmrqc = 0, vfreta = 0;
1667 	u16 rss_i = adapter->num_rx_queues;
1668 	u8 i, j;
1669 
1670 	/* Fill out hash function seeds */
1671 	netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1672 	for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1673 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1674 
1675 	for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1676 		if (j == rss_i)
1677 			j = 0;
1678 
1679 		adapter->rss_indir_tbl[i] = j;
1680 
1681 		vfreta |= j << (i & 0x3) * 8;
1682 		if ((i & 3) == 3) {
1683 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1684 			vfreta = 0;
1685 		}
1686 	}
1687 
1688 	/* Perform hash on these packet types */
1689 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1690 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1691 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1692 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1693 
1694 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1695 
1696 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1697 }
1698 
1699 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1700 				      struct ixgbevf_ring *ring)
1701 {
1702 	struct ixgbe_hw *hw = &adapter->hw;
1703 	u64 rdba = ring->dma;
1704 	u32 rxdctl;
1705 	u8 reg_idx = ring->reg_idx;
1706 
1707 	/* disable queue to avoid issues while updating state */
1708 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1709 	ixgbevf_disable_rx_queue(adapter, ring);
1710 
1711 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1712 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1713 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1714 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1715 
1716 #ifndef CONFIG_SPARC
1717 	/* enable relaxed ordering */
1718 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1719 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1720 #else
1721 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1722 			IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1723 			IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1724 #endif
1725 
1726 	/* reset head and tail pointers */
1727 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1728 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1729 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1730 
1731 	/* reset ntu and ntc to place SW in sync with hardwdare */
1732 	ring->next_to_clean = 0;
1733 	ring->next_to_use = 0;
1734 	ring->next_to_alloc = 0;
1735 
1736 	ixgbevf_configure_srrctl(adapter, reg_idx);
1737 
1738 	/* allow any size packet since we can handle overflow */
1739 	rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1740 
1741 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1742 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1743 
1744 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1745 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1746 }
1747 
1748 /**
1749  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1750  * @adapter: board private structure
1751  *
1752  * Configure the Rx unit of the MAC after a reset.
1753  **/
1754 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1755 {
1756 	struct ixgbe_hw *hw = &adapter->hw;
1757 	struct net_device *netdev = adapter->netdev;
1758 	int i, ret;
1759 
1760 	ixgbevf_setup_psrtype(adapter);
1761 	if (hw->mac.type >= ixgbe_mac_X550_vf)
1762 		ixgbevf_setup_vfmrqc(adapter);
1763 
1764 	spin_lock_bh(&adapter->mbx_lock);
1765 	/* notify the PF of our intent to use this size of frame */
1766 	ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1767 	spin_unlock_bh(&adapter->mbx_lock);
1768 	if (ret)
1769 		dev_err(&adapter->pdev->dev,
1770 			"Failed to set MTU at %d\n", netdev->mtu);
1771 
1772 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1773 	 * the Base and Length of the Rx Descriptor Ring
1774 	 */
1775 	for (i = 0; i < adapter->num_rx_queues; i++)
1776 		ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1777 }
1778 
1779 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1780 				   __be16 proto, u16 vid)
1781 {
1782 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1783 	struct ixgbe_hw *hw = &adapter->hw;
1784 	int err;
1785 
1786 	spin_lock_bh(&adapter->mbx_lock);
1787 
1788 	/* add VID to filter table */
1789 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1790 
1791 	spin_unlock_bh(&adapter->mbx_lock);
1792 
1793 	/* translate error return types so error makes sense */
1794 	if (err == IXGBE_ERR_MBX)
1795 		return -EIO;
1796 
1797 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
1798 		return -EACCES;
1799 
1800 	set_bit(vid, adapter->active_vlans);
1801 
1802 	return err;
1803 }
1804 
1805 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1806 				    __be16 proto, u16 vid)
1807 {
1808 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1809 	struct ixgbe_hw *hw = &adapter->hw;
1810 	int err;
1811 
1812 	spin_lock_bh(&adapter->mbx_lock);
1813 
1814 	/* remove VID from filter table */
1815 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1816 
1817 	spin_unlock_bh(&adapter->mbx_lock);
1818 
1819 	clear_bit(vid, adapter->active_vlans);
1820 
1821 	return err;
1822 }
1823 
1824 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1825 {
1826 	u16 vid;
1827 
1828 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1829 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
1830 					htons(ETH_P_8021Q), vid);
1831 }
1832 
1833 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1834 {
1835 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1836 	struct ixgbe_hw *hw = &adapter->hw;
1837 	int count = 0;
1838 
1839 	if ((netdev_uc_count(netdev)) > 10) {
1840 		pr_err("Too many unicast filters - No Space\n");
1841 		return -ENOSPC;
1842 	}
1843 
1844 	if (!netdev_uc_empty(netdev)) {
1845 		struct netdev_hw_addr *ha;
1846 
1847 		netdev_for_each_uc_addr(ha, netdev) {
1848 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1849 			udelay(200);
1850 		}
1851 	} else {
1852 		/* If the list is empty then send message to PF driver to
1853 		 * clear all MAC VLANs on this VF.
1854 		 */
1855 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1856 	}
1857 
1858 	return count;
1859 }
1860 
1861 /**
1862  * ixgbevf_set_rx_mode - Multicast and unicast set
1863  * @netdev: network interface device structure
1864  *
1865  * The set_rx_method entry point is called whenever the multicast address
1866  * list, unicast address list or the network interface flags are updated.
1867  * This routine is responsible for configuring the hardware for proper
1868  * multicast mode and configuring requested unicast filters.
1869  **/
1870 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1871 {
1872 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1873 	struct ixgbe_hw *hw = &adapter->hw;
1874 	unsigned int flags = netdev->flags;
1875 	int xcast_mode;
1876 
1877 	xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1878 		     (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1879 		     IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1880 
1881 	/* request the most inclusive mode we need */
1882 	if (flags & IFF_PROMISC)
1883 		xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
1884 	else if (flags & IFF_ALLMULTI)
1885 		xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
1886 	else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
1887 		xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
1888 	else
1889 		xcast_mode = IXGBEVF_XCAST_MODE_NONE;
1890 
1891 	spin_lock_bh(&adapter->mbx_lock);
1892 
1893 	hw->mac.ops.update_xcast_mode(hw, xcast_mode);
1894 
1895 	/* reprogram multicast list */
1896 	hw->mac.ops.update_mc_addr_list(hw, netdev);
1897 
1898 	ixgbevf_write_uc_addr_list(netdev);
1899 
1900 	spin_unlock_bh(&adapter->mbx_lock);
1901 }
1902 
1903 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1904 {
1905 	int q_idx;
1906 	struct ixgbevf_q_vector *q_vector;
1907 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1908 
1909 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1910 		q_vector = adapter->q_vector[q_idx];
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 	}
1925 }
1926 
1927 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1928 {
1929 	struct ixgbe_hw *hw = &adapter->hw;
1930 	unsigned int def_q = 0;
1931 	unsigned int num_tcs = 0;
1932 	unsigned int num_rx_queues = adapter->num_rx_queues;
1933 	unsigned int num_tx_queues = adapter->num_tx_queues;
1934 	int err;
1935 
1936 	spin_lock_bh(&adapter->mbx_lock);
1937 
1938 	/* fetch queue configuration from the PF */
1939 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1940 
1941 	spin_unlock_bh(&adapter->mbx_lock);
1942 
1943 	if (err)
1944 		return err;
1945 
1946 	if (num_tcs > 1) {
1947 		/* we need only one Tx queue */
1948 		num_tx_queues = 1;
1949 
1950 		/* update default Tx ring register index */
1951 		adapter->tx_ring[0]->reg_idx = def_q;
1952 
1953 		/* we need as many queues as traffic classes */
1954 		num_rx_queues = num_tcs;
1955 	}
1956 
1957 	/* if we have a bad config abort request queue reset */
1958 	if ((adapter->num_rx_queues != num_rx_queues) ||
1959 	    (adapter->num_tx_queues != num_tx_queues)) {
1960 		/* force mailbox timeout to prevent further messages */
1961 		hw->mbx.timeout = 0;
1962 
1963 		/* wait for watchdog to come around and bail us out */
1964 		set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
1965 	}
1966 
1967 	return 0;
1968 }
1969 
1970 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1971 {
1972 	ixgbevf_configure_dcb(adapter);
1973 
1974 	ixgbevf_set_rx_mode(adapter->netdev);
1975 
1976 	ixgbevf_restore_vlan(adapter);
1977 
1978 	ixgbevf_configure_tx(adapter);
1979 	ixgbevf_configure_rx(adapter);
1980 }
1981 
1982 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1983 {
1984 	/* Only save pre-reset stats if there are some */
1985 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1986 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1987 			adapter->stats.base_vfgprc;
1988 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1989 			adapter->stats.base_vfgptc;
1990 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1991 			adapter->stats.base_vfgorc;
1992 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1993 			adapter->stats.base_vfgotc;
1994 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1995 			adapter->stats.base_vfmprc;
1996 	}
1997 }
1998 
1999 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2000 {
2001 	struct ixgbe_hw *hw = &adapter->hw;
2002 
2003 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2004 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2005 	adapter->stats.last_vfgorc |=
2006 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2007 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2008 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2009 	adapter->stats.last_vfgotc |=
2010 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2011 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2012 
2013 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2014 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2015 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2016 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2017 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2018 }
2019 
2020 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2021 {
2022 	struct ixgbe_hw *hw = &adapter->hw;
2023 	int api[] = { ixgbe_mbox_api_13,
2024 		      ixgbe_mbox_api_12,
2025 		      ixgbe_mbox_api_11,
2026 		      ixgbe_mbox_api_10,
2027 		      ixgbe_mbox_api_unknown };
2028 	int err, idx = 0;
2029 
2030 	spin_lock_bh(&adapter->mbx_lock);
2031 
2032 	while (api[idx] != ixgbe_mbox_api_unknown) {
2033 		err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2034 		if (!err)
2035 			break;
2036 		idx++;
2037 	}
2038 
2039 	spin_unlock_bh(&adapter->mbx_lock);
2040 }
2041 
2042 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2043 {
2044 	struct net_device *netdev = adapter->netdev;
2045 	struct ixgbe_hw *hw = &adapter->hw;
2046 
2047 	ixgbevf_configure_msix(adapter);
2048 
2049 	spin_lock_bh(&adapter->mbx_lock);
2050 
2051 	if (is_valid_ether_addr(hw->mac.addr))
2052 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2053 	else
2054 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2055 
2056 	spin_unlock_bh(&adapter->mbx_lock);
2057 
2058 	smp_mb__before_atomic();
2059 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2060 	ixgbevf_napi_enable_all(adapter);
2061 
2062 	/* clear any pending interrupts, may auto mask */
2063 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2064 	ixgbevf_irq_enable(adapter);
2065 
2066 	/* enable transmits */
2067 	netif_tx_start_all_queues(netdev);
2068 
2069 	ixgbevf_save_reset_stats(adapter);
2070 	ixgbevf_init_last_counter_stats(adapter);
2071 
2072 	hw->mac.get_link_status = 1;
2073 	mod_timer(&adapter->service_timer, jiffies);
2074 }
2075 
2076 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2077 {
2078 	ixgbevf_configure(adapter);
2079 
2080 	ixgbevf_up_complete(adapter);
2081 }
2082 
2083 /**
2084  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2085  * @rx_ring: ring to free buffers from
2086  **/
2087 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2088 {
2089 	struct device *dev = rx_ring->dev;
2090 	unsigned long size;
2091 	unsigned int i;
2092 
2093 	/* Free Rx ring sk_buff */
2094 	if (rx_ring->skb) {
2095 		dev_kfree_skb(rx_ring->skb);
2096 		rx_ring->skb = NULL;
2097 	}
2098 
2099 	/* ring already cleared, nothing to do */
2100 	if (!rx_ring->rx_buffer_info)
2101 		return;
2102 
2103 	/* Free all the Rx ring pages */
2104 	for (i = 0; i < rx_ring->count; i++) {
2105 		struct ixgbevf_rx_buffer *rx_buffer;
2106 
2107 		rx_buffer = &rx_ring->rx_buffer_info[i];
2108 		if (rx_buffer->dma)
2109 			dma_unmap_page(dev, rx_buffer->dma,
2110 				       PAGE_SIZE, DMA_FROM_DEVICE);
2111 		rx_buffer->dma = 0;
2112 		if (rx_buffer->page)
2113 			__free_page(rx_buffer->page);
2114 		rx_buffer->page = NULL;
2115 	}
2116 
2117 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2118 	memset(rx_ring->rx_buffer_info, 0, size);
2119 
2120 	/* Zero out the descriptor ring */
2121 	memset(rx_ring->desc, 0, rx_ring->size);
2122 }
2123 
2124 /**
2125  * ixgbevf_clean_tx_ring - Free Tx Buffers
2126  * @tx_ring: ring to be cleaned
2127  **/
2128 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2129 {
2130 	struct ixgbevf_tx_buffer *tx_buffer_info;
2131 	unsigned long size;
2132 	unsigned int i;
2133 
2134 	if (!tx_ring->tx_buffer_info)
2135 		return;
2136 
2137 	/* Free all the Tx ring sk_buffs */
2138 	for (i = 0; i < tx_ring->count; i++) {
2139 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2140 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2141 	}
2142 
2143 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2144 	memset(tx_ring->tx_buffer_info, 0, size);
2145 
2146 	memset(tx_ring->desc, 0, tx_ring->size);
2147 }
2148 
2149 /**
2150  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2151  * @adapter: board private structure
2152  **/
2153 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2154 {
2155 	int i;
2156 
2157 	for (i = 0; i < adapter->num_rx_queues; i++)
2158 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2159 }
2160 
2161 /**
2162  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2163  * @adapter: board private structure
2164  **/
2165 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2166 {
2167 	int i;
2168 
2169 	for (i = 0; i < adapter->num_tx_queues; i++)
2170 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2171 }
2172 
2173 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2174 {
2175 	struct net_device *netdev = adapter->netdev;
2176 	struct ixgbe_hw *hw = &adapter->hw;
2177 	int i;
2178 
2179 	/* signal that we are down to the interrupt handler */
2180 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2181 		return; /* do nothing if already down */
2182 
2183 	/* disable all enabled Rx queues */
2184 	for (i = 0; i < adapter->num_rx_queues; i++)
2185 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2186 
2187 	usleep_range(10000, 20000);
2188 
2189 	netif_tx_stop_all_queues(netdev);
2190 
2191 	/* call carrier off first to avoid false dev_watchdog timeouts */
2192 	netif_carrier_off(netdev);
2193 	netif_tx_disable(netdev);
2194 
2195 	ixgbevf_irq_disable(adapter);
2196 
2197 	ixgbevf_napi_disable_all(adapter);
2198 
2199 	del_timer_sync(&adapter->service_timer);
2200 
2201 	/* disable transmits in the hardware now that interrupts are off */
2202 	for (i = 0; i < adapter->num_tx_queues; i++) {
2203 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2204 
2205 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2206 				IXGBE_TXDCTL_SWFLSH);
2207 	}
2208 
2209 	if (!pci_channel_offline(adapter->pdev))
2210 		ixgbevf_reset(adapter);
2211 
2212 	ixgbevf_clean_all_tx_rings(adapter);
2213 	ixgbevf_clean_all_rx_rings(adapter);
2214 }
2215 
2216 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2217 {
2218 	WARN_ON(in_interrupt());
2219 
2220 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2221 		msleep(1);
2222 
2223 	ixgbevf_down(adapter);
2224 	ixgbevf_up(adapter);
2225 
2226 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2227 }
2228 
2229 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2230 {
2231 	struct ixgbe_hw *hw = &adapter->hw;
2232 	struct net_device *netdev = adapter->netdev;
2233 
2234 	if (hw->mac.ops.reset_hw(hw)) {
2235 		hw_dbg(hw, "PF still resetting\n");
2236 	} else {
2237 		hw->mac.ops.init_hw(hw);
2238 		ixgbevf_negotiate_api(adapter);
2239 	}
2240 
2241 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2242 		ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2243 		ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2244 	}
2245 
2246 	adapter->last_reset = jiffies;
2247 }
2248 
2249 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2250 					int vectors)
2251 {
2252 	int vector_threshold;
2253 
2254 	/* We'll want at least 2 (vector_threshold):
2255 	 * 1) TxQ[0] + RxQ[0] handler
2256 	 * 2) Other (Link Status Change, etc.)
2257 	 */
2258 	vector_threshold = MIN_MSIX_COUNT;
2259 
2260 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2261 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2262 	 * Right now, we simply care about how many we'll get; we'll
2263 	 * set them up later while requesting irq's.
2264 	 */
2265 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2266 					vector_threshold, vectors);
2267 
2268 	if (vectors < 0) {
2269 		dev_err(&adapter->pdev->dev,
2270 			"Unable to allocate MSI-X interrupts\n");
2271 		kfree(adapter->msix_entries);
2272 		adapter->msix_entries = NULL;
2273 		return vectors;
2274 	}
2275 
2276 	/* Adjust for only the vectors we'll use, which is minimum
2277 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2278 	 * vectors we were allocated.
2279 	 */
2280 	adapter->num_msix_vectors = vectors;
2281 
2282 	return 0;
2283 }
2284 
2285 /**
2286  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2287  * @adapter: board private structure to initialize
2288  *
2289  * This is the top level queue allocation routine.  The order here is very
2290  * important, starting with the "most" number of features turned on at once,
2291  * and ending with the smallest set of features.  This way large combinations
2292  * can be allocated if they're turned on, and smaller combinations are the
2293  * fallthrough conditions.
2294  *
2295  **/
2296 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2297 {
2298 	struct ixgbe_hw *hw = &adapter->hw;
2299 	unsigned int def_q = 0;
2300 	unsigned int num_tcs = 0;
2301 	int err;
2302 
2303 	/* Start with base case */
2304 	adapter->num_rx_queues = 1;
2305 	adapter->num_tx_queues = 1;
2306 
2307 	spin_lock_bh(&adapter->mbx_lock);
2308 
2309 	/* fetch queue configuration from the PF */
2310 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2311 
2312 	spin_unlock_bh(&adapter->mbx_lock);
2313 
2314 	if (err)
2315 		return;
2316 
2317 	/* we need as many queues as traffic classes */
2318 	if (num_tcs > 1) {
2319 		adapter->num_rx_queues = num_tcs;
2320 	} else {
2321 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2322 
2323 		switch (hw->api_version) {
2324 		case ixgbe_mbox_api_11:
2325 		case ixgbe_mbox_api_12:
2326 		case ixgbe_mbox_api_13:
2327 			adapter->num_rx_queues = rss;
2328 			adapter->num_tx_queues = rss;
2329 		default:
2330 			break;
2331 		}
2332 	}
2333 }
2334 
2335 /**
2336  * ixgbevf_alloc_queues - Allocate memory for all rings
2337  * @adapter: board private structure to initialize
2338  *
2339  * We allocate one ring per queue at run-time since we don't know the
2340  * number of queues at compile-time.  The polling_netdev array is
2341  * intended for Multiqueue, but should work fine with a single queue.
2342  **/
2343 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2344 {
2345 	struct ixgbevf_ring *ring;
2346 	int rx = 0, tx = 0;
2347 
2348 	for (; tx < adapter->num_tx_queues; tx++) {
2349 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2350 		if (!ring)
2351 			goto err_allocation;
2352 
2353 		ring->dev = &adapter->pdev->dev;
2354 		ring->netdev = adapter->netdev;
2355 		ring->count = adapter->tx_ring_count;
2356 		ring->queue_index = tx;
2357 		ring->reg_idx = tx;
2358 
2359 		adapter->tx_ring[tx] = ring;
2360 	}
2361 
2362 	for (; rx < adapter->num_rx_queues; rx++) {
2363 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2364 		if (!ring)
2365 			goto err_allocation;
2366 
2367 		ring->dev = &adapter->pdev->dev;
2368 		ring->netdev = adapter->netdev;
2369 
2370 		ring->count = adapter->rx_ring_count;
2371 		ring->queue_index = rx;
2372 		ring->reg_idx = rx;
2373 
2374 		adapter->rx_ring[rx] = ring;
2375 	}
2376 
2377 	return 0;
2378 
2379 err_allocation:
2380 	while (tx) {
2381 		kfree(adapter->tx_ring[--tx]);
2382 		adapter->tx_ring[tx] = NULL;
2383 	}
2384 
2385 	while (rx) {
2386 		kfree(adapter->rx_ring[--rx]);
2387 		adapter->rx_ring[rx] = NULL;
2388 	}
2389 	return -ENOMEM;
2390 }
2391 
2392 /**
2393  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2394  * @adapter: board private structure to initialize
2395  *
2396  * Attempt to configure the interrupts using the best available
2397  * capabilities of the hardware and the kernel.
2398  **/
2399 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2400 {
2401 	struct net_device *netdev = adapter->netdev;
2402 	int err;
2403 	int vector, v_budget;
2404 
2405 	/* It's easy to be greedy for MSI-X vectors, but it really
2406 	 * doesn't do us much good if we have a lot more vectors
2407 	 * than CPU's.  So let's be conservative and only ask for
2408 	 * (roughly) the same number of vectors as there are CPU's.
2409 	 * The default is to use pairs of vectors.
2410 	 */
2411 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2412 	v_budget = min_t(int, v_budget, num_online_cpus());
2413 	v_budget += NON_Q_VECTORS;
2414 
2415 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2416 	 * mean we disable MSI-X capabilities of the adapter.
2417 	 */
2418 	adapter->msix_entries = kcalloc(v_budget,
2419 					sizeof(struct msix_entry), GFP_KERNEL);
2420 	if (!adapter->msix_entries)
2421 		return -ENOMEM;
2422 
2423 	for (vector = 0; vector < v_budget; vector++)
2424 		adapter->msix_entries[vector].entry = vector;
2425 
2426 	err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2427 	if (err)
2428 		return err;
2429 
2430 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2431 	if (err)
2432 		return err;
2433 
2434 	return netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2435 }
2436 
2437 /**
2438  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2439  * @adapter: board private structure to initialize
2440  *
2441  * We allocate one q_vector per queue interrupt.  If allocation fails we
2442  * return -ENOMEM.
2443  **/
2444 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2445 {
2446 	int q_idx, num_q_vectors;
2447 	struct ixgbevf_q_vector *q_vector;
2448 
2449 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2450 
2451 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2452 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2453 		if (!q_vector)
2454 			goto err_out;
2455 		q_vector->adapter = adapter;
2456 		q_vector->v_idx = q_idx;
2457 		netif_napi_add(adapter->netdev, &q_vector->napi,
2458 			       ixgbevf_poll, 64);
2459 		adapter->q_vector[q_idx] = q_vector;
2460 	}
2461 
2462 	return 0;
2463 
2464 err_out:
2465 	while (q_idx) {
2466 		q_idx--;
2467 		q_vector = adapter->q_vector[q_idx];
2468 #ifdef CONFIG_NET_RX_BUSY_POLL
2469 		napi_hash_del(&q_vector->napi);
2470 #endif
2471 		netif_napi_del(&q_vector->napi);
2472 		kfree(q_vector);
2473 		adapter->q_vector[q_idx] = NULL;
2474 	}
2475 	return -ENOMEM;
2476 }
2477 
2478 /**
2479  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2480  * @adapter: board private structure to initialize
2481  *
2482  * This function frees the memory allocated to the q_vectors.  In addition if
2483  * NAPI is enabled it will delete any references to the NAPI struct prior
2484  * to freeing the q_vector.
2485  **/
2486 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2487 {
2488 	int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2489 
2490 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2491 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2492 
2493 		adapter->q_vector[q_idx] = NULL;
2494 #ifdef CONFIG_NET_RX_BUSY_POLL
2495 		napi_hash_del(&q_vector->napi);
2496 #endif
2497 		netif_napi_del(&q_vector->napi);
2498 		kfree(q_vector);
2499 	}
2500 }
2501 
2502 /**
2503  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2504  * @adapter: board private structure
2505  *
2506  **/
2507 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2508 {
2509 	if (!adapter->msix_entries)
2510 		return;
2511 
2512 	pci_disable_msix(adapter->pdev);
2513 	kfree(adapter->msix_entries);
2514 	adapter->msix_entries = NULL;
2515 }
2516 
2517 /**
2518  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2519  * @adapter: board private structure to initialize
2520  *
2521  **/
2522 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2523 {
2524 	int err;
2525 
2526 	/* Number of supported queues */
2527 	ixgbevf_set_num_queues(adapter);
2528 
2529 	err = ixgbevf_set_interrupt_capability(adapter);
2530 	if (err) {
2531 		hw_dbg(&adapter->hw,
2532 		       "Unable to setup interrupt capabilities\n");
2533 		goto err_set_interrupt;
2534 	}
2535 
2536 	err = ixgbevf_alloc_q_vectors(adapter);
2537 	if (err) {
2538 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2539 		goto err_alloc_q_vectors;
2540 	}
2541 
2542 	err = ixgbevf_alloc_queues(adapter);
2543 	if (err) {
2544 		pr_err("Unable to allocate memory for queues\n");
2545 		goto err_alloc_queues;
2546 	}
2547 
2548 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2549 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2550 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2551 
2552 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2553 
2554 	return 0;
2555 err_alloc_queues:
2556 	ixgbevf_free_q_vectors(adapter);
2557 err_alloc_q_vectors:
2558 	ixgbevf_reset_interrupt_capability(adapter);
2559 err_set_interrupt:
2560 	return err;
2561 }
2562 
2563 /**
2564  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2565  * @adapter: board private structure to clear interrupt scheme on
2566  *
2567  * We go through and clear interrupt specific resources and reset the structure
2568  * to pre-load conditions
2569  **/
2570 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2571 {
2572 	int i;
2573 
2574 	for (i = 0; i < adapter->num_tx_queues; i++) {
2575 		kfree(adapter->tx_ring[i]);
2576 		adapter->tx_ring[i] = NULL;
2577 	}
2578 	for (i = 0; i < adapter->num_rx_queues; i++) {
2579 		kfree(adapter->rx_ring[i]);
2580 		adapter->rx_ring[i] = NULL;
2581 	}
2582 
2583 	adapter->num_tx_queues = 0;
2584 	adapter->num_rx_queues = 0;
2585 
2586 	ixgbevf_free_q_vectors(adapter);
2587 	ixgbevf_reset_interrupt_capability(adapter);
2588 }
2589 
2590 /**
2591  * ixgbevf_sw_init - Initialize general software structures
2592  * @adapter: board private structure to initialize
2593  *
2594  * ixgbevf_sw_init initializes the Adapter private data structure.
2595  * Fields are initialized based on PCI device information and
2596  * OS network device settings (MTU size).
2597  **/
2598 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2599 {
2600 	struct ixgbe_hw *hw = &adapter->hw;
2601 	struct pci_dev *pdev = adapter->pdev;
2602 	struct net_device *netdev = adapter->netdev;
2603 	int err;
2604 
2605 	/* PCI config space info */
2606 	hw->vendor_id = pdev->vendor;
2607 	hw->device_id = pdev->device;
2608 	hw->revision_id = pdev->revision;
2609 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2610 	hw->subsystem_device_id = pdev->subsystem_device;
2611 
2612 	hw->mbx.ops.init_params(hw);
2613 
2614 	/* assume legacy case in which PF would only give VF 2 queues */
2615 	hw->mac.max_tx_queues = 2;
2616 	hw->mac.max_rx_queues = 2;
2617 
2618 	/* lock to protect mailbox accesses */
2619 	spin_lock_init(&adapter->mbx_lock);
2620 
2621 	err = hw->mac.ops.reset_hw(hw);
2622 	if (err) {
2623 		dev_info(&pdev->dev,
2624 			 "PF still in reset state.  Is the PF interface up?\n");
2625 	} else {
2626 		err = hw->mac.ops.init_hw(hw);
2627 		if (err) {
2628 			pr_err("init_shared_code failed: %d\n", err);
2629 			goto out;
2630 		}
2631 		ixgbevf_negotiate_api(adapter);
2632 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2633 		if (err)
2634 			dev_info(&pdev->dev, "Error reading MAC address\n");
2635 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2636 			dev_info(&pdev->dev,
2637 				 "MAC address not assigned by administrator.\n");
2638 		ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2639 	}
2640 
2641 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2642 		dev_info(&pdev->dev, "Assigning random MAC address\n");
2643 		eth_hw_addr_random(netdev);
2644 		ether_addr_copy(hw->mac.addr, netdev->dev_addr);
2645 		ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
2646 	}
2647 
2648 	/* Enable dynamic interrupt throttling rates */
2649 	adapter->rx_itr_setting = 1;
2650 	adapter->tx_itr_setting = 1;
2651 
2652 	/* set default ring sizes */
2653 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2654 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2655 
2656 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2657 	return 0;
2658 
2659 out:
2660 	return err;
2661 }
2662 
2663 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2664 	{							\
2665 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2666 		if (current_counter < last_counter)		\
2667 			counter += 0x100000000LL;		\
2668 		last_counter = current_counter;			\
2669 		counter &= 0xFFFFFFFF00000000LL;		\
2670 		counter |= current_counter;			\
2671 	}
2672 
2673 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2674 	{								 \
2675 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2676 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2677 		u64 current_counter = (current_counter_msb << 32) |	 \
2678 			current_counter_lsb;				 \
2679 		if (current_counter < last_counter)			 \
2680 			counter += 0x1000000000LL;			 \
2681 		last_counter = current_counter;				 \
2682 		counter &= 0xFFFFFFF000000000LL;			 \
2683 		counter |= current_counter;				 \
2684 	}
2685 /**
2686  * ixgbevf_update_stats - Update the board statistics counters.
2687  * @adapter: board private structure
2688  **/
2689 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2690 {
2691 	struct ixgbe_hw *hw = &adapter->hw;
2692 	int i;
2693 
2694 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2695 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2696 		return;
2697 
2698 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2699 				adapter->stats.vfgprc);
2700 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2701 				adapter->stats.vfgptc);
2702 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2703 				adapter->stats.last_vfgorc,
2704 				adapter->stats.vfgorc);
2705 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2706 				adapter->stats.last_vfgotc,
2707 				adapter->stats.vfgotc);
2708 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2709 				adapter->stats.vfmprc);
2710 
2711 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2712 		adapter->hw_csum_rx_error +=
2713 			adapter->rx_ring[i]->hw_csum_rx_error;
2714 		adapter->rx_ring[i]->hw_csum_rx_error = 0;
2715 	}
2716 }
2717 
2718 /**
2719  * ixgbevf_service_timer - Timer Call-back
2720  * @data: pointer to adapter cast into an unsigned long
2721  **/
2722 static void ixgbevf_service_timer(unsigned long data)
2723 {
2724 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2725 
2726 	/* Reset the timer */
2727 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2728 
2729 	ixgbevf_service_event_schedule(adapter);
2730 }
2731 
2732 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2733 {
2734 	if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
2735 		return;
2736 
2737 	/* If we're already down or resetting, just bail */
2738 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2739 	    test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
2740 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2741 		return;
2742 
2743 	adapter->tx_timeout_count++;
2744 
2745 	rtnl_lock();
2746 	ixgbevf_reinit_locked(adapter);
2747 	rtnl_unlock();
2748 }
2749 
2750 /**
2751  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2752  * @adapter: pointer to the device adapter structure
2753  *
2754  * This function serves two purposes.  First it strobes the interrupt lines
2755  * in order to make certain interrupts are occurring.  Secondly it sets the
2756  * bits needed to check for TX hangs.  As a result we should immediately
2757  * determine if a hang has occurred.
2758  **/
2759 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2760 {
2761 	struct ixgbe_hw *hw = &adapter->hw;
2762 	u32 eics = 0;
2763 	int i;
2764 
2765 	/* If we're down or resetting, just bail */
2766 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2767 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2768 		return;
2769 
2770 	/* Force detection of hung controller */
2771 	if (netif_carrier_ok(adapter->netdev)) {
2772 		for (i = 0; i < adapter->num_tx_queues; i++)
2773 			set_check_for_tx_hang(adapter->tx_ring[i]);
2774 	}
2775 
2776 	/* get one bit for every active Tx/Rx interrupt vector */
2777 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2778 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2779 
2780 		if (qv->rx.ring || qv->tx.ring)
2781 			eics |= BIT(i);
2782 	}
2783 
2784 	/* Cause software interrupt to ensure rings are cleaned */
2785 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2786 }
2787 
2788 /**
2789  * ixgbevf_watchdog_update_link - update the link status
2790  * @adapter: pointer to the device adapter structure
2791  **/
2792 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2793 {
2794 	struct ixgbe_hw *hw = &adapter->hw;
2795 	u32 link_speed = adapter->link_speed;
2796 	bool link_up = adapter->link_up;
2797 	s32 err;
2798 
2799 	spin_lock_bh(&adapter->mbx_lock);
2800 
2801 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2802 
2803 	spin_unlock_bh(&adapter->mbx_lock);
2804 
2805 	/* if check for link returns error we will need to reset */
2806 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2807 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
2808 		link_up = false;
2809 	}
2810 
2811 	adapter->link_up = link_up;
2812 	adapter->link_speed = link_speed;
2813 }
2814 
2815 /**
2816  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2817  *				 print link up message
2818  * @adapter: pointer to the device adapter structure
2819  **/
2820 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2821 {
2822 	struct net_device *netdev = adapter->netdev;
2823 
2824 	/* only continue if link was previously down */
2825 	if (netif_carrier_ok(netdev))
2826 		return;
2827 
2828 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2829 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2830 		 "10 Gbps" :
2831 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2832 		 "1 Gbps" :
2833 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2834 		 "100 Mbps" :
2835 		 "unknown speed");
2836 
2837 	netif_carrier_on(netdev);
2838 }
2839 
2840 /**
2841  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2842  *				   print link down message
2843  * @adapter: pointer to the adapter structure
2844  **/
2845 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2846 {
2847 	struct net_device *netdev = adapter->netdev;
2848 
2849 	adapter->link_speed = 0;
2850 
2851 	/* only continue if link was up previously */
2852 	if (!netif_carrier_ok(netdev))
2853 		return;
2854 
2855 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2856 
2857 	netif_carrier_off(netdev);
2858 }
2859 
2860 /**
2861  * ixgbevf_watchdog_subtask - worker thread to bring link up
2862  * @work: pointer to work_struct containing our data
2863  **/
2864 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2865 {
2866 	/* if interface is down do nothing */
2867 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2868 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2869 		return;
2870 
2871 	ixgbevf_watchdog_update_link(adapter);
2872 
2873 	if (adapter->link_up)
2874 		ixgbevf_watchdog_link_is_up(adapter);
2875 	else
2876 		ixgbevf_watchdog_link_is_down(adapter);
2877 
2878 	ixgbevf_update_stats(adapter);
2879 }
2880 
2881 /**
2882  * ixgbevf_service_task - manages and runs subtasks
2883  * @work: pointer to work_struct containing our data
2884  **/
2885 static void ixgbevf_service_task(struct work_struct *work)
2886 {
2887 	struct ixgbevf_adapter *adapter = container_of(work,
2888 						       struct ixgbevf_adapter,
2889 						       service_task);
2890 	struct ixgbe_hw *hw = &adapter->hw;
2891 
2892 	if (IXGBE_REMOVED(hw->hw_addr)) {
2893 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2894 			rtnl_lock();
2895 			ixgbevf_down(adapter);
2896 			rtnl_unlock();
2897 		}
2898 		return;
2899 	}
2900 
2901 	ixgbevf_queue_reset_subtask(adapter);
2902 	ixgbevf_reset_subtask(adapter);
2903 	ixgbevf_watchdog_subtask(adapter);
2904 	ixgbevf_check_hang_subtask(adapter);
2905 
2906 	ixgbevf_service_event_complete(adapter);
2907 }
2908 
2909 /**
2910  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2911  * @tx_ring: Tx descriptor ring for a specific queue
2912  *
2913  * Free all transmit software resources
2914  **/
2915 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2916 {
2917 	ixgbevf_clean_tx_ring(tx_ring);
2918 
2919 	vfree(tx_ring->tx_buffer_info);
2920 	tx_ring->tx_buffer_info = NULL;
2921 
2922 	/* if not set, then don't free */
2923 	if (!tx_ring->desc)
2924 		return;
2925 
2926 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2927 			  tx_ring->dma);
2928 
2929 	tx_ring->desc = NULL;
2930 }
2931 
2932 /**
2933  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2934  * @adapter: board private structure
2935  *
2936  * Free all transmit software resources
2937  **/
2938 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2939 {
2940 	int i;
2941 
2942 	for (i = 0; i < adapter->num_tx_queues; i++)
2943 		if (adapter->tx_ring[i]->desc)
2944 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2945 }
2946 
2947 /**
2948  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2949  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2950  *
2951  * Return 0 on success, negative on failure
2952  **/
2953 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2954 {
2955 	struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2956 	int size;
2957 
2958 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2959 	tx_ring->tx_buffer_info = vzalloc(size);
2960 	if (!tx_ring->tx_buffer_info)
2961 		goto err;
2962 
2963 	/* round up to nearest 4K */
2964 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2965 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2966 
2967 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2968 					   &tx_ring->dma, GFP_KERNEL);
2969 	if (!tx_ring->desc)
2970 		goto err;
2971 
2972 	return 0;
2973 
2974 err:
2975 	vfree(tx_ring->tx_buffer_info);
2976 	tx_ring->tx_buffer_info = NULL;
2977 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
2978 	return -ENOMEM;
2979 }
2980 
2981 /**
2982  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2983  * @adapter: board private structure
2984  *
2985  * If this function returns with an error, then it's possible one or
2986  * more of the rings is populated (while the rest are not).  It is the
2987  * callers duty to clean those orphaned rings.
2988  *
2989  * Return 0 on success, negative on failure
2990  **/
2991 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2992 {
2993 	int i, err = 0;
2994 
2995 	for (i = 0; i < adapter->num_tx_queues; i++) {
2996 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
2997 		if (!err)
2998 			continue;
2999 		hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3000 		break;
3001 	}
3002 
3003 	return err;
3004 }
3005 
3006 /**
3007  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3008  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3009  *
3010  * Returns 0 on success, negative on failure
3011  **/
3012 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3013 {
3014 	int size;
3015 
3016 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3017 	rx_ring->rx_buffer_info = vzalloc(size);
3018 	if (!rx_ring->rx_buffer_info)
3019 		goto err;
3020 
3021 	/* Round up to nearest 4K */
3022 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3023 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3024 
3025 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3026 					   &rx_ring->dma, GFP_KERNEL);
3027 
3028 	if (!rx_ring->desc)
3029 		goto err;
3030 
3031 	return 0;
3032 err:
3033 	vfree(rx_ring->rx_buffer_info);
3034 	rx_ring->rx_buffer_info = NULL;
3035 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3036 	return -ENOMEM;
3037 }
3038 
3039 /**
3040  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3041  * @adapter: board private structure
3042  *
3043  * If this function returns with an error, then it's possible one or
3044  * more of the rings is populated (while the rest are not).  It is the
3045  * callers duty to clean those orphaned rings.
3046  *
3047  * Return 0 on success, negative on failure
3048  **/
3049 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3050 {
3051 	int i, err = 0;
3052 
3053 	for (i = 0; i < adapter->num_rx_queues; i++) {
3054 		err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3055 		if (!err)
3056 			continue;
3057 		hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3058 		break;
3059 	}
3060 	return err;
3061 }
3062 
3063 /**
3064  * ixgbevf_free_rx_resources - Free Rx Resources
3065  * @rx_ring: ring to clean the resources from
3066  *
3067  * Free all receive software resources
3068  **/
3069 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3070 {
3071 	ixgbevf_clean_rx_ring(rx_ring);
3072 
3073 	vfree(rx_ring->rx_buffer_info);
3074 	rx_ring->rx_buffer_info = NULL;
3075 
3076 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3077 			  rx_ring->dma);
3078 
3079 	rx_ring->desc = NULL;
3080 }
3081 
3082 /**
3083  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3084  * @adapter: board private structure
3085  *
3086  * Free all receive software resources
3087  **/
3088 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3089 {
3090 	int i;
3091 
3092 	for (i = 0; i < adapter->num_rx_queues; i++)
3093 		if (adapter->rx_ring[i]->desc)
3094 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3095 }
3096 
3097 /**
3098  * ixgbevf_open - Called when a network interface is made active
3099  * @netdev: network interface device structure
3100  *
3101  * Returns 0 on success, negative value on failure
3102  *
3103  * The open entry point is called when a network interface is made
3104  * active by the system (IFF_UP).  At this point all resources needed
3105  * for transmit and receive operations are allocated, the interrupt
3106  * handler is registered with the OS, the watchdog timer is started,
3107  * and the stack is notified that the interface is ready.
3108  **/
3109 int ixgbevf_open(struct net_device *netdev)
3110 {
3111 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3112 	struct ixgbe_hw *hw = &adapter->hw;
3113 	int err;
3114 
3115 	/* A previous failure to open the device because of a lack of
3116 	 * available MSIX vector resources may have reset the number
3117 	 * of msix vectors variable to zero.  The only way to recover
3118 	 * is to unload/reload the driver and hope that the system has
3119 	 * been able to recover some MSIX vector resources.
3120 	 */
3121 	if (!adapter->num_msix_vectors)
3122 		return -ENOMEM;
3123 
3124 	if (hw->adapter_stopped) {
3125 		ixgbevf_reset(adapter);
3126 		/* if adapter is still stopped then PF isn't up and
3127 		 * the VF can't start.
3128 		 */
3129 		if (hw->adapter_stopped) {
3130 			err = IXGBE_ERR_MBX;
3131 			pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3132 			goto err_setup_reset;
3133 		}
3134 	}
3135 
3136 	/* disallow open during test */
3137 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3138 		return -EBUSY;
3139 
3140 	netif_carrier_off(netdev);
3141 
3142 	/* allocate transmit descriptors */
3143 	err = ixgbevf_setup_all_tx_resources(adapter);
3144 	if (err)
3145 		goto err_setup_tx;
3146 
3147 	/* allocate receive descriptors */
3148 	err = ixgbevf_setup_all_rx_resources(adapter);
3149 	if (err)
3150 		goto err_setup_rx;
3151 
3152 	ixgbevf_configure(adapter);
3153 
3154 	/* Map the Tx/Rx rings to the vectors we were allotted.
3155 	 * if request_irq will be called in this function map_rings
3156 	 * must be called *before* up_complete
3157 	 */
3158 	ixgbevf_map_rings_to_vectors(adapter);
3159 
3160 	err = ixgbevf_request_irq(adapter);
3161 	if (err)
3162 		goto err_req_irq;
3163 
3164 	ixgbevf_up_complete(adapter);
3165 
3166 	return 0;
3167 
3168 err_req_irq:
3169 	ixgbevf_down(adapter);
3170 err_setup_rx:
3171 	ixgbevf_free_all_rx_resources(adapter);
3172 err_setup_tx:
3173 	ixgbevf_free_all_tx_resources(adapter);
3174 	ixgbevf_reset(adapter);
3175 
3176 err_setup_reset:
3177 
3178 	return err;
3179 }
3180 
3181 /**
3182  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3183  * @adapter: the private adapter struct
3184  *
3185  * This function should contain the necessary work common to both suspending
3186  * and closing of the device.
3187  */
3188 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3189 {
3190 	ixgbevf_down(adapter);
3191 	ixgbevf_free_irq(adapter);
3192 	ixgbevf_free_all_tx_resources(adapter);
3193 	ixgbevf_free_all_rx_resources(adapter);
3194 }
3195 
3196 /**
3197  * ixgbevf_close - Disables a network interface
3198  * @netdev: network interface device structure
3199  *
3200  * Returns 0, this is not allowed to fail
3201  *
3202  * The close entry point is called when an interface is de-activated
3203  * by the OS.  The hardware is still under the drivers control, but
3204  * needs to be disabled.  A global MAC reset is issued to stop the
3205  * hardware, and all transmit and receive resources are freed.
3206  **/
3207 int ixgbevf_close(struct net_device *netdev)
3208 {
3209 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3210 
3211 	if (netif_device_present(netdev))
3212 		ixgbevf_close_suspend(adapter);
3213 
3214 	return 0;
3215 }
3216 
3217 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3218 {
3219 	struct net_device *dev = adapter->netdev;
3220 
3221 	if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3222 				&adapter->state))
3223 		return;
3224 
3225 	/* if interface is down do nothing */
3226 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3227 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3228 		return;
3229 
3230 	/* Hardware has to reinitialize queues and interrupts to
3231 	 * match packet buffer alignment. Unfortunately, the
3232 	 * hardware is not flexible enough to do this dynamically.
3233 	 */
3234 	rtnl_lock();
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 	rtnl_unlock();
3246 }
3247 
3248 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3249 				u32 vlan_macip_lens, u32 type_tucmd,
3250 				u32 mss_l4len_idx)
3251 {
3252 	struct ixgbe_adv_tx_context_desc *context_desc;
3253 	u16 i = tx_ring->next_to_use;
3254 
3255 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3256 
3257 	i++;
3258 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3259 
3260 	/* set bits to identify this as an advanced context descriptor */
3261 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3262 
3263 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3264 	context_desc->seqnum_seed	= 0;
3265 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3266 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3267 }
3268 
3269 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3270 		       struct ixgbevf_tx_buffer *first,
3271 		       u8 *hdr_len)
3272 {
3273 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3274 	struct sk_buff *skb = first->skb;
3275 	union {
3276 		struct iphdr *v4;
3277 		struct ipv6hdr *v6;
3278 		unsigned char *hdr;
3279 	} ip;
3280 	union {
3281 		struct tcphdr *tcp;
3282 		unsigned char *hdr;
3283 	} l4;
3284 	u32 paylen, l4_offset;
3285 	int err;
3286 
3287 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3288 		return 0;
3289 
3290 	if (!skb_is_gso(skb))
3291 		return 0;
3292 
3293 	err = skb_cow_head(skb, 0);
3294 	if (err < 0)
3295 		return err;
3296 
3297 	ip.hdr = skb_network_header(skb);
3298 	l4.hdr = skb_checksum_start(skb);
3299 
3300 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3301 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3302 
3303 	/* initialize outer IP header fields */
3304 	if (ip.v4->version == 4) {
3305 		unsigned char *csum_start = skb_checksum_start(skb);
3306 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3307 
3308 		/* IP header will have to cancel out any data that
3309 		 * is not a part of the outer IP header
3310 		 */
3311 		ip.v4->check = csum_fold(csum_partial(trans_start,
3312 						      csum_start - trans_start,
3313 						      0));
3314 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3315 
3316 		ip.v4->tot_len = 0;
3317 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3318 				   IXGBE_TX_FLAGS_CSUM |
3319 				   IXGBE_TX_FLAGS_IPV4;
3320 	} else {
3321 		ip.v6->payload_len = 0;
3322 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3323 				   IXGBE_TX_FLAGS_CSUM;
3324 	}
3325 
3326 	/* determine offset of inner transport header */
3327 	l4_offset = l4.hdr - skb->data;
3328 
3329 	/* compute length of segmentation header */
3330 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
3331 
3332 	/* remove payload length from inner checksum */
3333 	paylen = skb->len - l4_offset;
3334 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3335 
3336 	/* update gso size and bytecount with header size */
3337 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3338 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3339 
3340 	/* mss_l4len_id: use 1 as index for TSO */
3341 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3342 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3343 	mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3344 
3345 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3346 	vlan_macip_lens = l4.hdr - ip.hdr;
3347 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3348 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3349 
3350 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3351 			    type_tucmd, mss_l4len_idx);
3352 
3353 	return 1;
3354 }
3355 
3356 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3357 {
3358 	unsigned int offset = 0;
3359 
3360 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3361 
3362 	return offset == skb_checksum_start_offset(skb);
3363 }
3364 
3365 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3366 			    struct ixgbevf_tx_buffer *first)
3367 {
3368 	struct sk_buff *skb = first->skb;
3369 	u32 vlan_macip_lens = 0;
3370 	u32 type_tucmd = 0;
3371 
3372 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3373 		goto no_csum;
3374 
3375 	switch (skb->csum_offset) {
3376 	case offsetof(struct tcphdr, check):
3377 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3378 		/* fall through */
3379 	case offsetof(struct udphdr, check):
3380 		break;
3381 	case offsetof(struct sctphdr, checksum):
3382 		/* validate that this is actually an SCTP request */
3383 		if (((first->protocol == htons(ETH_P_IP)) &&
3384 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3385 		    ((first->protocol == htons(ETH_P_IPV6)) &&
3386 		     ixgbevf_ipv6_csum_is_sctp(skb))) {
3387 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3388 			break;
3389 		}
3390 		/* fall through */
3391 	default:
3392 		skb_checksum_help(skb);
3393 		goto no_csum;
3394 	}
3395 	/* update TX checksum flag */
3396 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3397 	vlan_macip_lens = skb_checksum_start_offset(skb) -
3398 			  skb_network_offset(skb);
3399 no_csum:
3400 	/* vlan_macip_lens: MACLEN, VLAN tag */
3401 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3402 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3403 
3404 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3405 }
3406 
3407 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3408 {
3409 	/* set type for advanced descriptor with frame checksum insertion */
3410 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3411 				      IXGBE_ADVTXD_DCMD_IFCS |
3412 				      IXGBE_ADVTXD_DCMD_DEXT);
3413 
3414 	/* set HW VLAN bit if VLAN is present */
3415 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3416 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3417 
3418 	/* set segmentation enable bits for TSO/FSO */
3419 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3420 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3421 
3422 	return cmd_type;
3423 }
3424 
3425 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3426 				     u32 tx_flags, unsigned int paylen)
3427 {
3428 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3429 
3430 	/* enable L4 checksum for TSO and TX checksum offload */
3431 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3432 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3433 
3434 	/* enble IPv4 checksum for TSO */
3435 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3436 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3437 
3438 	/* use index 1 context for TSO/FSO/FCOE */
3439 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3440 		olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3441 
3442 	/* Check Context must be set if Tx switch is enabled, which it
3443 	 * always is for case where virtual functions are running
3444 	 */
3445 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3446 
3447 	tx_desc->read.olinfo_status = olinfo_status;
3448 }
3449 
3450 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3451 			   struct ixgbevf_tx_buffer *first,
3452 			   const u8 hdr_len)
3453 {
3454 	dma_addr_t dma;
3455 	struct sk_buff *skb = first->skb;
3456 	struct ixgbevf_tx_buffer *tx_buffer;
3457 	union ixgbe_adv_tx_desc *tx_desc;
3458 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3459 	unsigned int data_len = skb->data_len;
3460 	unsigned int size = skb_headlen(skb);
3461 	unsigned int paylen = skb->len - hdr_len;
3462 	u32 tx_flags = first->tx_flags;
3463 	__le32 cmd_type;
3464 	u16 i = tx_ring->next_to_use;
3465 
3466 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3467 
3468 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3469 	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3470 
3471 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3472 	if (dma_mapping_error(tx_ring->dev, dma))
3473 		goto dma_error;
3474 
3475 	/* record length, and DMA address */
3476 	dma_unmap_len_set(first, len, size);
3477 	dma_unmap_addr_set(first, dma, dma);
3478 
3479 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
3480 
3481 	for (;;) {
3482 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3483 			tx_desc->read.cmd_type_len =
3484 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3485 
3486 			i++;
3487 			tx_desc++;
3488 			if (i == tx_ring->count) {
3489 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3490 				i = 0;
3491 			}
3492 
3493 			dma += IXGBE_MAX_DATA_PER_TXD;
3494 			size -= IXGBE_MAX_DATA_PER_TXD;
3495 
3496 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3497 			tx_desc->read.olinfo_status = 0;
3498 		}
3499 
3500 		if (likely(!data_len))
3501 			break;
3502 
3503 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3504 
3505 		i++;
3506 		tx_desc++;
3507 		if (i == tx_ring->count) {
3508 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3509 			i = 0;
3510 		}
3511 
3512 		size = skb_frag_size(frag);
3513 		data_len -= size;
3514 
3515 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3516 				       DMA_TO_DEVICE);
3517 		if (dma_mapping_error(tx_ring->dev, dma))
3518 			goto dma_error;
3519 
3520 		tx_buffer = &tx_ring->tx_buffer_info[i];
3521 		dma_unmap_len_set(tx_buffer, len, size);
3522 		dma_unmap_addr_set(tx_buffer, dma, dma);
3523 
3524 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3525 		tx_desc->read.olinfo_status = 0;
3526 
3527 		frag++;
3528 	}
3529 
3530 	/* write last descriptor with RS and EOP bits */
3531 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3532 	tx_desc->read.cmd_type_len = cmd_type;
3533 
3534 	/* set the timestamp */
3535 	first->time_stamp = jiffies;
3536 
3537 	/* Force memory writes to complete before letting h/w know there
3538 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
3539 	 * memory model archs, such as IA-64).
3540 	 *
3541 	 * We also need this memory barrier (wmb) to make certain all of the
3542 	 * status bits have been updated before next_to_watch is written.
3543 	 */
3544 	wmb();
3545 
3546 	/* set next_to_watch value indicating a packet is present */
3547 	first->next_to_watch = tx_desc;
3548 
3549 	i++;
3550 	if (i == tx_ring->count)
3551 		i = 0;
3552 
3553 	tx_ring->next_to_use = i;
3554 
3555 	/* notify HW of packet */
3556 	ixgbevf_write_tail(tx_ring, i);
3557 
3558 	return;
3559 dma_error:
3560 	dev_err(tx_ring->dev, "TX DMA map failed\n");
3561 
3562 	/* clear dma mappings for failed tx_buffer_info map */
3563 	for (;;) {
3564 		tx_buffer = &tx_ring->tx_buffer_info[i];
3565 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3566 		if (tx_buffer == first)
3567 			break;
3568 		if (i == 0)
3569 			i = tx_ring->count;
3570 		i--;
3571 	}
3572 
3573 	tx_ring->next_to_use = i;
3574 }
3575 
3576 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3577 {
3578 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3579 	/* Herbert's original patch had:
3580 	 *  smp_mb__after_netif_stop_queue();
3581 	 * but since that doesn't exist yet, just open code it.
3582 	 */
3583 	smp_mb();
3584 
3585 	/* We need to check again in a case another CPU has just
3586 	 * made room available.
3587 	 */
3588 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
3589 		return -EBUSY;
3590 
3591 	/* A reprieve! - use start_queue because it doesn't call schedule */
3592 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3593 	++tx_ring->tx_stats.restart_queue;
3594 
3595 	return 0;
3596 }
3597 
3598 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3599 {
3600 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3601 		return 0;
3602 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
3603 }
3604 
3605 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3606 {
3607 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3608 	struct ixgbevf_tx_buffer *first;
3609 	struct ixgbevf_ring *tx_ring;
3610 	int tso;
3611 	u32 tx_flags = 0;
3612 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
3613 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3614 	unsigned short f;
3615 #endif
3616 	u8 hdr_len = 0;
3617 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3618 
3619 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3620 		dev_kfree_skb_any(skb);
3621 		return NETDEV_TX_OK;
3622 	}
3623 
3624 	tx_ring = adapter->tx_ring[skb->queue_mapping];
3625 
3626 	/* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3627 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3628 	 *       + 2 desc gap to keep tail from touching head,
3629 	 *       + 1 desc for context descriptor,
3630 	 * otherwise try next time
3631 	 */
3632 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3633 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3634 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3635 #else
3636 	count += skb_shinfo(skb)->nr_frags;
3637 #endif
3638 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3639 		tx_ring->tx_stats.tx_busy++;
3640 		return NETDEV_TX_BUSY;
3641 	}
3642 
3643 	/* record the location of the first descriptor for this packet */
3644 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3645 	first->skb = skb;
3646 	first->bytecount = skb->len;
3647 	first->gso_segs = 1;
3648 
3649 	if (skb_vlan_tag_present(skb)) {
3650 		tx_flags |= skb_vlan_tag_get(skb);
3651 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3652 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3653 	}
3654 
3655 	/* record initial flags and protocol */
3656 	first->tx_flags = tx_flags;
3657 	first->protocol = vlan_get_protocol(skb);
3658 
3659 	tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3660 	if (tso < 0)
3661 		goto out_drop;
3662 	else if (!tso)
3663 		ixgbevf_tx_csum(tx_ring, first);
3664 
3665 	ixgbevf_tx_map(tx_ring, first, hdr_len);
3666 
3667 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3668 
3669 	return NETDEV_TX_OK;
3670 
3671 out_drop:
3672 	dev_kfree_skb_any(first->skb);
3673 	first->skb = NULL;
3674 
3675 	return NETDEV_TX_OK;
3676 }
3677 
3678 /**
3679  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3680  * @netdev: network interface device structure
3681  * @p: pointer to an address structure
3682  *
3683  * Returns 0 on success, negative on failure
3684  **/
3685 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3686 {
3687 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3688 	struct ixgbe_hw *hw = &adapter->hw;
3689 	struct sockaddr *addr = p;
3690 	int err;
3691 
3692 	if (!is_valid_ether_addr(addr->sa_data))
3693 		return -EADDRNOTAVAIL;
3694 
3695 	spin_lock_bh(&adapter->mbx_lock);
3696 
3697 	err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
3698 
3699 	spin_unlock_bh(&adapter->mbx_lock);
3700 
3701 	if (err)
3702 		return -EPERM;
3703 
3704 	ether_addr_copy(hw->mac.addr, addr->sa_data);
3705 	ether_addr_copy(netdev->dev_addr, addr->sa_data);
3706 
3707 	return 0;
3708 }
3709 
3710 /**
3711  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3712  * @netdev: network interface device structure
3713  * @new_mtu: new value for maximum frame size
3714  *
3715  * Returns 0 on success, negative on failure
3716  **/
3717 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3718 {
3719 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3720 	struct ixgbe_hw *hw = &adapter->hw;
3721 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3722 	int ret;
3723 
3724 	spin_lock_bh(&adapter->mbx_lock);
3725 	/* notify the PF of our intent to use this size of frame */
3726 	ret = hw->mac.ops.set_rlpml(hw, max_frame);
3727 	spin_unlock_bh(&adapter->mbx_lock);
3728 	if (ret)
3729 		return -EINVAL;
3730 
3731 	hw_dbg(hw, "changing MTU from %d to %d\n",
3732 	       netdev->mtu, new_mtu);
3733 
3734 	/* must set new MTU before calling down or up */
3735 	netdev->mtu = new_mtu;
3736 
3737 	return 0;
3738 }
3739 
3740 #ifdef CONFIG_NET_POLL_CONTROLLER
3741 /* Polling 'interrupt' - used by things like netconsole to send skbs
3742  * without having to re-enable interrupts. It's not called while
3743  * the interrupt routine is executing.
3744  */
3745 static void ixgbevf_netpoll(struct net_device *netdev)
3746 {
3747 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3748 	int i;
3749 
3750 	/* if interface is down do nothing */
3751 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3752 		return;
3753 	for (i = 0; i < adapter->num_rx_queues; i++)
3754 		ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3755 }
3756 #endif /* CONFIG_NET_POLL_CONTROLLER */
3757 
3758 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3759 {
3760 	struct net_device *netdev = pci_get_drvdata(pdev);
3761 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3762 #ifdef CONFIG_PM
3763 	int retval = 0;
3764 #endif
3765 
3766 	rtnl_lock();
3767 	netif_device_detach(netdev);
3768 
3769 	if (netif_running(netdev))
3770 		ixgbevf_close_suspend(adapter);
3771 
3772 	ixgbevf_clear_interrupt_scheme(adapter);
3773 	rtnl_unlock();
3774 
3775 #ifdef CONFIG_PM
3776 	retval = pci_save_state(pdev);
3777 	if (retval)
3778 		return retval;
3779 
3780 #endif
3781 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3782 		pci_disable_device(pdev);
3783 
3784 	return 0;
3785 }
3786 
3787 #ifdef CONFIG_PM
3788 static int ixgbevf_resume(struct pci_dev *pdev)
3789 {
3790 	struct net_device *netdev = pci_get_drvdata(pdev);
3791 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3792 	u32 err;
3793 
3794 	pci_restore_state(pdev);
3795 	/* pci_restore_state clears dev->state_saved so call
3796 	 * pci_save_state to restore it.
3797 	 */
3798 	pci_save_state(pdev);
3799 
3800 	err = pci_enable_device_mem(pdev);
3801 	if (err) {
3802 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3803 		return err;
3804 	}
3805 
3806 	adapter->hw.hw_addr = adapter->io_addr;
3807 	smp_mb__before_atomic();
3808 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3809 	pci_set_master(pdev);
3810 
3811 	ixgbevf_reset(adapter);
3812 
3813 	rtnl_lock();
3814 	err = ixgbevf_init_interrupt_scheme(adapter);
3815 	rtnl_unlock();
3816 	if (err) {
3817 		dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3818 		return err;
3819 	}
3820 
3821 	if (netif_running(netdev)) {
3822 		err = ixgbevf_open(netdev);
3823 		if (err)
3824 			return err;
3825 	}
3826 
3827 	netif_device_attach(netdev);
3828 
3829 	return err;
3830 }
3831 
3832 #endif /* CONFIG_PM */
3833 static void ixgbevf_shutdown(struct pci_dev *pdev)
3834 {
3835 	ixgbevf_suspend(pdev, PMSG_SUSPEND);
3836 }
3837 
3838 static void ixgbevf_get_stats(struct net_device *netdev,
3839 			      struct rtnl_link_stats64 *stats)
3840 {
3841 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3842 	unsigned int start;
3843 	u64 bytes, packets;
3844 	const struct ixgbevf_ring *ring;
3845 	int i;
3846 
3847 	ixgbevf_update_stats(adapter);
3848 
3849 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3850 
3851 	for (i = 0; i < adapter->num_rx_queues; i++) {
3852 		ring = adapter->rx_ring[i];
3853 		do {
3854 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3855 			bytes = ring->stats.bytes;
3856 			packets = ring->stats.packets;
3857 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3858 		stats->rx_bytes += bytes;
3859 		stats->rx_packets += packets;
3860 	}
3861 
3862 	for (i = 0; i < adapter->num_tx_queues; i++) {
3863 		ring = adapter->tx_ring[i];
3864 		do {
3865 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3866 			bytes = ring->stats.bytes;
3867 			packets = ring->stats.packets;
3868 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3869 		stats->tx_bytes += bytes;
3870 		stats->tx_packets += packets;
3871 	}
3872 }
3873 
3874 #define IXGBEVF_MAX_MAC_HDR_LEN		127
3875 #define IXGBEVF_MAX_NETWORK_HDR_LEN	511
3876 
3877 static netdev_features_t
3878 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
3879 		       netdev_features_t features)
3880 {
3881 	unsigned int network_hdr_len, mac_hdr_len;
3882 
3883 	/* Make certain the headers can be described by a context descriptor */
3884 	mac_hdr_len = skb_network_header(skb) - skb->data;
3885 	if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
3886 		return features & ~(NETIF_F_HW_CSUM |
3887 				    NETIF_F_SCTP_CRC |
3888 				    NETIF_F_HW_VLAN_CTAG_TX |
3889 				    NETIF_F_TSO |
3890 				    NETIF_F_TSO6);
3891 
3892 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3893 	if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
3894 		return features & ~(NETIF_F_HW_CSUM |
3895 				    NETIF_F_SCTP_CRC |
3896 				    NETIF_F_TSO |
3897 				    NETIF_F_TSO6);
3898 
3899 	/* We can only support IPV4 TSO in tunnels if we can mangle the
3900 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
3901 	 */
3902 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3903 		features &= ~NETIF_F_TSO;
3904 
3905 	return features;
3906 }
3907 
3908 static const struct net_device_ops ixgbevf_netdev_ops = {
3909 	.ndo_open		= ixgbevf_open,
3910 	.ndo_stop		= ixgbevf_close,
3911 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3912 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3913 	.ndo_get_stats64	= ixgbevf_get_stats,
3914 	.ndo_validate_addr	= eth_validate_addr,
3915 	.ndo_set_mac_address	= ixgbevf_set_mac,
3916 	.ndo_change_mtu		= ixgbevf_change_mtu,
3917 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3918 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3919 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3920 #ifdef CONFIG_NET_POLL_CONTROLLER
3921 	.ndo_poll_controller	= ixgbevf_netpoll,
3922 #endif
3923 	.ndo_features_check	= ixgbevf_features_check,
3924 };
3925 
3926 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3927 {
3928 	dev->netdev_ops = &ixgbevf_netdev_ops;
3929 	ixgbevf_set_ethtool_ops(dev);
3930 	dev->watchdog_timeo = 5 * HZ;
3931 }
3932 
3933 /**
3934  * ixgbevf_probe - Device Initialization Routine
3935  * @pdev: PCI device information struct
3936  * @ent: entry in ixgbevf_pci_tbl
3937  *
3938  * Returns 0 on success, negative on failure
3939  *
3940  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3941  * The OS initialization, configuring of the adapter private structure,
3942  * and a hardware reset occur.
3943  **/
3944 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3945 {
3946 	struct net_device *netdev;
3947 	struct ixgbevf_adapter *adapter = NULL;
3948 	struct ixgbe_hw *hw = NULL;
3949 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3950 	int err, pci_using_dac;
3951 	bool disable_dev = false;
3952 
3953 	err = pci_enable_device(pdev);
3954 	if (err)
3955 		return err;
3956 
3957 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3958 		pci_using_dac = 1;
3959 	} else {
3960 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3961 		if (err) {
3962 			dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3963 			goto err_dma;
3964 		}
3965 		pci_using_dac = 0;
3966 	}
3967 
3968 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3969 	if (err) {
3970 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3971 		goto err_pci_reg;
3972 	}
3973 
3974 	pci_set_master(pdev);
3975 
3976 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3977 				   MAX_TX_QUEUES);
3978 	if (!netdev) {
3979 		err = -ENOMEM;
3980 		goto err_alloc_etherdev;
3981 	}
3982 
3983 	SET_NETDEV_DEV(netdev, &pdev->dev);
3984 
3985 	adapter = netdev_priv(netdev);
3986 
3987 	adapter->netdev = netdev;
3988 	adapter->pdev = pdev;
3989 	hw = &adapter->hw;
3990 	hw->back = adapter;
3991 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3992 
3993 	/* call save state here in standalone driver because it relies on
3994 	 * adapter struct to exist, and needs to call netdev_priv
3995 	 */
3996 	pci_save_state(pdev);
3997 
3998 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3999 			      pci_resource_len(pdev, 0));
4000 	adapter->io_addr = hw->hw_addr;
4001 	if (!hw->hw_addr) {
4002 		err = -EIO;
4003 		goto err_ioremap;
4004 	}
4005 
4006 	ixgbevf_assign_netdev_ops(netdev);
4007 
4008 	/* Setup HW API */
4009 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4010 	hw->mac.type  = ii->mac;
4011 
4012 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4013 	       sizeof(struct ixgbe_mbx_operations));
4014 
4015 	/* setup the private structure */
4016 	err = ixgbevf_sw_init(adapter);
4017 	if (err)
4018 		goto err_sw_init;
4019 
4020 	/* The HW MAC address was set and/or determined in sw_init */
4021 	if (!is_valid_ether_addr(netdev->dev_addr)) {
4022 		pr_err("invalid MAC address\n");
4023 		err = -EIO;
4024 		goto err_sw_init;
4025 	}
4026 
4027 	netdev->hw_features = NETIF_F_SG |
4028 			      NETIF_F_TSO |
4029 			      NETIF_F_TSO6 |
4030 			      NETIF_F_RXCSUM |
4031 			      NETIF_F_HW_CSUM |
4032 			      NETIF_F_SCTP_CRC;
4033 
4034 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4035 				      NETIF_F_GSO_GRE_CSUM | \
4036 				      NETIF_F_GSO_IPXIP4 | \
4037 				      NETIF_F_GSO_IPXIP6 | \
4038 				      NETIF_F_GSO_UDP_TUNNEL | \
4039 				      NETIF_F_GSO_UDP_TUNNEL_CSUM)
4040 
4041 	netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4042 	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4043 			       IXGBEVF_GSO_PARTIAL_FEATURES;
4044 
4045 	netdev->features = netdev->hw_features;
4046 
4047 	if (pci_using_dac)
4048 		netdev->features |= NETIF_F_HIGHDMA;
4049 
4050 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4051 	netdev->mpls_features |= NETIF_F_HW_CSUM;
4052 	netdev->hw_enc_features |= netdev->vlan_features;
4053 
4054 	/* set this bit last since it cannot be part of vlan_features */
4055 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4056 			    NETIF_F_HW_VLAN_CTAG_RX |
4057 			    NETIF_F_HW_VLAN_CTAG_TX;
4058 
4059 	netdev->priv_flags |= IFF_UNICAST_FLT;
4060 
4061 	/* MTU range: 68 - 1504 or 9710 */
4062 	netdev->min_mtu = ETH_MIN_MTU;
4063 	switch (adapter->hw.api_version) {
4064 	case ixgbe_mbox_api_11:
4065 	case ixgbe_mbox_api_12:
4066 	case ixgbe_mbox_api_13:
4067 		netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4068 				  (ETH_HLEN + ETH_FCS_LEN);
4069 		break;
4070 	default:
4071 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4072 			netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4073 					  (ETH_HLEN + ETH_FCS_LEN);
4074 		else
4075 			netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4076 		break;
4077 	}
4078 
4079 	if (IXGBE_REMOVED(hw->hw_addr)) {
4080 		err = -EIO;
4081 		goto err_sw_init;
4082 	}
4083 
4084 	setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4085 		    (unsigned long)adapter);
4086 
4087 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4088 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4089 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4090 
4091 	err = ixgbevf_init_interrupt_scheme(adapter);
4092 	if (err)
4093 		goto err_sw_init;
4094 
4095 	strcpy(netdev->name, "eth%d");
4096 
4097 	err = register_netdev(netdev);
4098 	if (err)
4099 		goto err_register;
4100 
4101 	pci_set_drvdata(pdev, netdev);
4102 	netif_carrier_off(netdev);
4103 
4104 	ixgbevf_init_last_counter_stats(adapter);
4105 
4106 	/* print the VF info */
4107 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4108 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4109 
4110 	switch (hw->mac.type) {
4111 	case ixgbe_mac_X550_vf:
4112 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4113 		break;
4114 	case ixgbe_mac_X540_vf:
4115 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4116 		break;
4117 	case ixgbe_mac_82599_vf:
4118 	default:
4119 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4120 		break;
4121 	}
4122 
4123 	return 0;
4124 
4125 err_register:
4126 	ixgbevf_clear_interrupt_scheme(adapter);
4127 err_sw_init:
4128 	ixgbevf_reset_interrupt_capability(adapter);
4129 	iounmap(adapter->io_addr);
4130 err_ioremap:
4131 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4132 	free_netdev(netdev);
4133 err_alloc_etherdev:
4134 	pci_release_regions(pdev);
4135 err_pci_reg:
4136 err_dma:
4137 	if (!adapter || disable_dev)
4138 		pci_disable_device(pdev);
4139 	return err;
4140 }
4141 
4142 /**
4143  * ixgbevf_remove - Device Removal Routine
4144  * @pdev: PCI device information struct
4145  *
4146  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4147  * that it should release a PCI device.  The could be caused by a
4148  * Hot-Plug event, or because the driver is going to be removed from
4149  * memory.
4150  **/
4151 static void ixgbevf_remove(struct pci_dev *pdev)
4152 {
4153 	struct net_device *netdev = pci_get_drvdata(pdev);
4154 	struct ixgbevf_adapter *adapter;
4155 	bool disable_dev;
4156 
4157 	if (!netdev)
4158 		return;
4159 
4160 	adapter = netdev_priv(netdev);
4161 
4162 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4163 	cancel_work_sync(&adapter->service_task);
4164 
4165 	if (netdev->reg_state == NETREG_REGISTERED)
4166 		unregister_netdev(netdev);
4167 
4168 	ixgbevf_clear_interrupt_scheme(adapter);
4169 	ixgbevf_reset_interrupt_capability(adapter);
4170 
4171 	iounmap(adapter->io_addr);
4172 	pci_release_regions(pdev);
4173 
4174 	hw_dbg(&adapter->hw, "Remove complete\n");
4175 
4176 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4177 	free_netdev(netdev);
4178 
4179 	if (disable_dev)
4180 		pci_disable_device(pdev);
4181 }
4182 
4183 /**
4184  * ixgbevf_io_error_detected - called when PCI error is detected
4185  * @pdev: Pointer to PCI device
4186  * @state: The current pci connection state
4187  *
4188  * This function is called after a PCI bus error affecting
4189  * this device has been detected.
4190  **/
4191 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4192 						  pci_channel_state_t state)
4193 {
4194 	struct net_device *netdev = pci_get_drvdata(pdev);
4195 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4196 
4197 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4198 		return PCI_ERS_RESULT_DISCONNECT;
4199 
4200 	rtnl_lock();
4201 	netif_device_detach(netdev);
4202 
4203 	if (state == pci_channel_io_perm_failure) {
4204 		rtnl_unlock();
4205 		return PCI_ERS_RESULT_DISCONNECT;
4206 	}
4207 
4208 	if (netif_running(netdev))
4209 		ixgbevf_close_suspend(adapter);
4210 
4211 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4212 		pci_disable_device(pdev);
4213 	rtnl_unlock();
4214 
4215 	/* Request a slot slot reset. */
4216 	return PCI_ERS_RESULT_NEED_RESET;
4217 }
4218 
4219 /**
4220  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4221  * @pdev: Pointer to PCI device
4222  *
4223  * Restart the card from scratch, as if from a cold-boot. Implementation
4224  * resembles the first-half of the ixgbevf_resume routine.
4225  **/
4226 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4227 {
4228 	struct net_device *netdev = pci_get_drvdata(pdev);
4229 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4230 
4231 	if (pci_enable_device_mem(pdev)) {
4232 		dev_err(&pdev->dev,
4233 			"Cannot re-enable PCI device after reset.\n");
4234 		return PCI_ERS_RESULT_DISCONNECT;
4235 	}
4236 
4237 	adapter->hw.hw_addr = adapter->io_addr;
4238 	smp_mb__before_atomic();
4239 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4240 	pci_set_master(pdev);
4241 
4242 	ixgbevf_reset(adapter);
4243 
4244 	return PCI_ERS_RESULT_RECOVERED;
4245 }
4246 
4247 /**
4248  * ixgbevf_io_resume - called when traffic can start flowing again.
4249  * @pdev: Pointer to PCI device
4250  *
4251  * This callback is called when the error recovery driver tells us that
4252  * its OK to resume normal operation. Implementation resembles the
4253  * second-half of the ixgbevf_resume routine.
4254  **/
4255 static void ixgbevf_io_resume(struct pci_dev *pdev)
4256 {
4257 	struct net_device *netdev = pci_get_drvdata(pdev);
4258 
4259 	rtnl_lock();
4260 	if (netif_running(netdev))
4261 		ixgbevf_open(netdev);
4262 
4263 	netif_device_attach(netdev);
4264 	rtnl_unlock();
4265 }
4266 
4267 /* PCI Error Recovery (ERS) */
4268 static const struct pci_error_handlers ixgbevf_err_handler = {
4269 	.error_detected = ixgbevf_io_error_detected,
4270 	.slot_reset = ixgbevf_io_slot_reset,
4271 	.resume = ixgbevf_io_resume,
4272 };
4273 
4274 static struct pci_driver ixgbevf_driver = {
4275 	.name		= ixgbevf_driver_name,
4276 	.id_table	= ixgbevf_pci_tbl,
4277 	.probe		= ixgbevf_probe,
4278 	.remove		= ixgbevf_remove,
4279 #ifdef CONFIG_PM
4280 	/* Power Management Hooks */
4281 	.suspend	= ixgbevf_suspend,
4282 	.resume		= ixgbevf_resume,
4283 #endif
4284 	.shutdown	= ixgbevf_shutdown,
4285 	.err_handler	= &ixgbevf_err_handler
4286 };
4287 
4288 /**
4289  * ixgbevf_init_module - Driver Registration Routine
4290  *
4291  * ixgbevf_init_module is the first routine called when the driver is
4292  * loaded. All it does is register with the PCI subsystem.
4293  **/
4294 static int __init ixgbevf_init_module(void)
4295 {
4296 	pr_info("%s - version %s\n", ixgbevf_driver_string,
4297 		ixgbevf_driver_version);
4298 
4299 	pr_info("%s\n", ixgbevf_copyright);
4300 	ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4301 	if (!ixgbevf_wq) {
4302 		pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4303 		return -ENOMEM;
4304 	}
4305 
4306 	return pci_register_driver(&ixgbevf_driver);
4307 }
4308 
4309 module_init(ixgbevf_init_module);
4310 
4311 /**
4312  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4313  *
4314  * ixgbevf_exit_module is called just before the driver is removed
4315  * from memory.
4316  **/
4317 static void __exit ixgbevf_exit_module(void)
4318 {
4319 	pci_unregister_driver(&ixgbevf_driver);
4320 	if (ixgbevf_wq) {
4321 		destroy_workqueue(ixgbevf_wq);
4322 		ixgbevf_wq = NULL;
4323 	}
4324 }
4325 
4326 #ifdef DEBUG
4327 /**
4328  * ixgbevf_get_hw_dev_name - return device name string
4329  * used by hardware layer to print debugging information
4330  **/
4331 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4332 {
4333 	struct ixgbevf_adapter *adapter = hw->back;
4334 
4335 	return adapter->netdev->name;
4336 }
4337 
4338 #endif
4339 module_exit(ixgbevf_exit_module);
4340 
4341 /* ixgbevf_main.c */
4342