xref: /linux/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
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 "2.12.1-k"
60 const char ixgbevf_driver_version[] = DRV_VERSION;
61 static char ixgbevf_copyright[] =
62 	"Copyright (c) 2009 - 2012 Intel Corporation.";
63 
64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
65 	[board_82599_vf] = &ixgbevf_82599_vf_info,
66 	[board_X540_vf]  = &ixgbevf_X540_vf_info,
67 	[board_X550_vf]  = &ixgbevf_X550_vf_info,
68 	[board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
69 };
70 
71 /* ixgbevf_pci_tbl - PCI Device ID Table
72  *
73  * Wildcard entries (PCI_ANY_ID) should come last
74  * Last entry must be all 0s
75  *
76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77  *   Class, Class Mask, private data (not used) }
78  */
79 static const struct pci_device_id ixgbevf_pci_tbl[] = {
80 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
81 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
82 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
83 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
84 	/* required last entry */
85 	{0, }
86 };
87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
88 
89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
90 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
91 MODULE_LICENSE("GPL");
92 MODULE_VERSION(DRV_VERSION);
93 
94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
95 static int debug = -1;
96 module_param(debug, int, 0);
97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
98 
99 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
100 {
101 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
102 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
103 	    !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
104 		schedule_work(&adapter->service_task);
105 }
106 
107 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
108 {
109 	BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
110 
111 	/* flush memory to make sure state is correct before next watchdog */
112 	smp_mb__before_atomic();
113 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
114 }
115 
116 /* forward decls */
117 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
118 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
119 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
120 
121 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
122 {
123 	struct ixgbevf_adapter *adapter = hw->back;
124 
125 	if (!hw->hw_addr)
126 		return;
127 	hw->hw_addr = NULL;
128 	dev_err(&adapter->pdev->dev, "Adapter removed\n");
129 	if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
130 		ixgbevf_service_event_schedule(adapter);
131 }
132 
133 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
134 {
135 	u32 value;
136 
137 	/* The following check not only optimizes a bit by not
138 	 * performing a read on the status register when the
139 	 * register just read was a status register read that
140 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
141 	 * potential recursion.
142 	 */
143 	if (reg == IXGBE_VFSTATUS) {
144 		ixgbevf_remove_adapter(hw);
145 		return;
146 	}
147 	value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
148 	if (value == IXGBE_FAILED_READ_REG)
149 		ixgbevf_remove_adapter(hw);
150 }
151 
152 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
153 {
154 	u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
155 	u32 value;
156 
157 	if (IXGBE_REMOVED(reg_addr))
158 		return IXGBE_FAILED_READ_REG;
159 	value = readl(reg_addr + reg);
160 	if (unlikely(value == IXGBE_FAILED_READ_REG))
161 		ixgbevf_check_remove(hw, reg);
162 	return value;
163 }
164 
165 /**
166  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
167  * @adapter: pointer to adapter struct
168  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
169  * @queue: queue to map the corresponding interrupt to
170  * @msix_vector: the vector to map to the corresponding queue
171  **/
172 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
173 			     u8 queue, u8 msix_vector)
174 {
175 	u32 ivar, index;
176 	struct ixgbe_hw *hw = &adapter->hw;
177 
178 	if (direction == -1) {
179 		/* other causes */
180 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
181 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
182 		ivar &= ~0xFF;
183 		ivar |= msix_vector;
184 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
185 	} else {
186 		/* Tx or Rx causes */
187 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
188 		index = ((16 * (queue & 1)) + (8 * direction));
189 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
190 		ivar &= ~(0xFF << index);
191 		ivar |= (msix_vector << index);
192 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
193 	}
194 }
195 
196 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
197 					struct ixgbevf_tx_buffer *tx_buffer)
198 {
199 	if (tx_buffer->skb) {
200 		dev_kfree_skb_any(tx_buffer->skb);
201 		if (dma_unmap_len(tx_buffer, len))
202 			dma_unmap_single(tx_ring->dev,
203 					 dma_unmap_addr(tx_buffer, dma),
204 					 dma_unmap_len(tx_buffer, len),
205 					 DMA_TO_DEVICE);
206 	} else if (dma_unmap_len(tx_buffer, len)) {
207 		dma_unmap_page(tx_ring->dev,
208 			       dma_unmap_addr(tx_buffer, dma),
209 			       dma_unmap_len(tx_buffer, len),
210 			       DMA_TO_DEVICE);
211 	}
212 	tx_buffer->next_to_watch = NULL;
213 	tx_buffer->skb = NULL;
214 	dma_unmap_len_set(tx_buffer, len, 0);
215 	/* tx_buffer must be completely set up in the transmit path */
216 }
217 
218 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
219 {
220 	return ring->stats.packets;
221 }
222 
223 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
224 {
225 	struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
226 	struct ixgbe_hw *hw = &adapter->hw;
227 
228 	u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
229 	u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
230 
231 	if (head != tail)
232 		return (head < tail) ?
233 			tail - head : (tail + ring->count - head);
234 
235 	return 0;
236 }
237 
238 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
239 {
240 	u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
241 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
242 	u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
243 
244 	clear_check_for_tx_hang(tx_ring);
245 
246 	/* Check for a hung queue, but be thorough. This verifies
247 	 * that a transmit has been completed since the previous
248 	 * check AND there is at least one packet pending. The
249 	 * ARMED bit is set to indicate a potential hang.
250 	 */
251 	if ((tx_done_old == tx_done) && tx_pending) {
252 		/* make sure it is true for two checks in a row */
253 		return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
254 					&tx_ring->state);
255 	}
256 	/* reset the countdown */
257 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
258 
259 	/* update completed stats and continue */
260 	tx_ring->tx_stats.tx_done_old = tx_done;
261 
262 	return false;
263 }
264 
265 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
266 {
267 	/* Do the reset outside of interrupt context */
268 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
269 		adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
270 		ixgbevf_service_event_schedule(adapter);
271 	}
272 }
273 
274 /**
275  * ixgbevf_tx_timeout - Respond to a Tx Hang
276  * @netdev: network interface device structure
277  **/
278 static void ixgbevf_tx_timeout(struct net_device *netdev)
279 {
280 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
281 
282 	ixgbevf_tx_timeout_reset(adapter);
283 }
284 
285 /**
286  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
287  * @q_vector: board private structure
288  * @tx_ring: tx ring to clean
289  **/
290 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
291 				 struct ixgbevf_ring *tx_ring)
292 {
293 	struct ixgbevf_adapter *adapter = q_vector->adapter;
294 	struct ixgbevf_tx_buffer *tx_buffer;
295 	union ixgbe_adv_tx_desc *tx_desc;
296 	unsigned int total_bytes = 0, total_packets = 0;
297 	unsigned int budget = tx_ring->count / 2;
298 	unsigned int i = tx_ring->next_to_clean;
299 
300 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
301 		return true;
302 
303 	tx_buffer = &tx_ring->tx_buffer_info[i];
304 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
305 	i -= tx_ring->count;
306 
307 	do {
308 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
309 
310 		/* if next_to_watch is not set then there is no work pending */
311 		if (!eop_desc)
312 			break;
313 
314 		/* prevent any other reads prior to eop_desc */
315 		read_barrier_depends();
316 
317 		/* if DD is not set pending work has not been completed */
318 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
319 			break;
320 
321 		/* clear next_to_watch to prevent false hangs */
322 		tx_buffer->next_to_watch = NULL;
323 
324 		/* update the statistics for this packet */
325 		total_bytes += tx_buffer->bytecount;
326 		total_packets += tx_buffer->gso_segs;
327 
328 		/* free the skb */
329 		dev_kfree_skb_any(tx_buffer->skb);
330 
331 		/* unmap skb header data */
332 		dma_unmap_single(tx_ring->dev,
333 				 dma_unmap_addr(tx_buffer, dma),
334 				 dma_unmap_len(tx_buffer, len),
335 				 DMA_TO_DEVICE);
336 
337 		/* clear tx_buffer data */
338 		tx_buffer->skb = NULL;
339 		dma_unmap_len_set(tx_buffer, len, 0);
340 
341 		/* unmap remaining buffers */
342 		while (tx_desc != eop_desc) {
343 			tx_buffer++;
344 			tx_desc++;
345 			i++;
346 			if (unlikely(!i)) {
347 				i -= tx_ring->count;
348 				tx_buffer = tx_ring->tx_buffer_info;
349 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
350 			}
351 
352 			/* unmap any remaining paged data */
353 			if (dma_unmap_len(tx_buffer, len)) {
354 				dma_unmap_page(tx_ring->dev,
355 					       dma_unmap_addr(tx_buffer, dma),
356 					       dma_unmap_len(tx_buffer, len),
357 					       DMA_TO_DEVICE);
358 				dma_unmap_len_set(tx_buffer, len, 0);
359 			}
360 		}
361 
362 		/* move us one more past the eop_desc for start of next pkt */
363 		tx_buffer++;
364 		tx_desc++;
365 		i++;
366 		if (unlikely(!i)) {
367 			i -= tx_ring->count;
368 			tx_buffer = tx_ring->tx_buffer_info;
369 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
370 		}
371 
372 		/* issue prefetch for next Tx descriptor */
373 		prefetch(tx_desc);
374 
375 		/* update budget accounting */
376 		budget--;
377 	} while (likely(budget));
378 
379 	i += tx_ring->count;
380 	tx_ring->next_to_clean = i;
381 	u64_stats_update_begin(&tx_ring->syncp);
382 	tx_ring->stats.bytes += total_bytes;
383 	tx_ring->stats.packets += total_packets;
384 	u64_stats_update_end(&tx_ring->syncp);
385 	q_vector->tx.total_bytes += total_bytes;
386 	q_vector->tx.total_packets += total_packets;
387 
388 	if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
389 		struct ixgbe_hw *hw = &adapter->hw;
390 		union ixgbe_adv_tx_desc *eop_desc;
391 
392 		eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
393 
394 		pr_err("Detected Tx Unit Hang\n"
395 		       "  Tx Queue             <%d>\n"
396 		       "  TDH, TDT             <%x>, <%x>\n"
397 		       "  next_to_use          <%x>\n"
398 		       "  next_to_clean        <%x>\n"
399 		       "tx_buffer_info[next_to_clean]\n"
400 		       "  next_to_watch        <%p>\n"
401 		       "  eop_desc->wb.status  <%x>\n"
402 		       "  time_stamp           <%lx>\n"
403 		       "  jiffies              <%lx>\n",
404 		       tx_ring->queue_index,
405 		       IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
406 		       IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
407 		       tx_ring->next_to_use, i,
408 		       eop_desc, (eop_desc ? eop_desc->wb.status : 0),
409 		       tx_ring->tx_buffer_info[i].time_stamp, jiffies);
410 
411 		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
412 
413 		/* schedule immediate reset if we believe we hung */
414 		ixgbevf_tx_timeout_reset(adapter);
415 
416 		return true;
417 	}
418 
419 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
420 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
421 		     (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
422 		/* Make sure that anybody stopping the queue after this
423 		 * sees the new next_to_clean.
424 		 */
425 		smp_mb();
426 
427 		if (__netif_subqueue_stopped(tx_ring->netdev,
428 					     tx_ring->queue_index) &&
429 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
430 			netif_wake_subqueue(tx_ring->netdev,
431 					    tx_ring->queue_index);
432 			++tx_ring->tx_stats.restart_queue;
433 		}
434 	}
435 
436 	return !!budget;
437 }
438 
439 /**
440  * ixgbevf_rx_skb - Helper function to determine proper Rx method
441  * @q_vector: structure containing interrupt and ring information
442  * @skb: packet to send up
443  **/
444 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
445 			   struct sk_buff *skb)
446 {
447 #ifdef CONFIG_NET_RX_BUSY_POLL
448 	skb_mark_napi_id(skb, &q_vector->napi);
449 
450 	if (ixgbevf_qv_busy_polling(q_vector)) {
451 		netif_receive_skb(skb);
452 		/* exit early if we busy polled */
453 		return;
454 	}
455 #endif /* CONFIG_NET_RX_BUSY_POLL */
456 
457 	napi_gro_receive(&q_vector->napi, skb);
458 }
459 
460 #define IXGBE_RSS_L4_TYPES_MASK \
461 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
462 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
463 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
464 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
465 
466 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
467 				   union ixgbe_adv_rx_desc *rx_desc,
468 				   struct sk_buff *skb)
469 {
470 	u16 rss_type;
471 
472 	if (!(ring->netdev->features & NETIF_F_RXHASH))
473 		return;
474 
475 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
476 		   IXGBE_RXDADV_RSSTYPE_MASK;
477 
478 	if (!rss_type)
479 		return;
480 
481 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
482 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
483 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
484 }
485 
486 /**
487  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
488  * @ring: structure containig ring specific data
489  * @rx_desc: current Rx descriptor being processed
490  * @skb: skb currently being received and modified
491  **/
492 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
493 				       union ixgbe_adv_rx_desc *rx_desc,
494 				       struct sk_buff *skb)
495 {
496 	skb_checksum_none_assert(skb);
497 
498 	/* Rx csum disabled */
499 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
500 		return;
501 
502 	/* if IP and error */
503 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
504 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
505 		ring->rx_stats.csum_err++;
506 		return;
507 	}
508 
509 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
510 		return;
511 
512 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
513 		ring->rx_stats.csum_err++;
514 		return;
515 	}
516 
517 	/* It must be a TCP or UDP packet with a valid checksum */
518 	skb->ip_summed = CHECKSUM_UNNECESSARY;
519 }
520 
521 /**
522  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
523  * @rx_ring: rx descriptor ring packet is being transacted on
524  * @rx_desc: pointer to the EOP Rx descriptor
525  * @skb: pointer to current skb being populated
526  *
527  * This function checks the ring, descriptor, and packet information in
528  * order to populate the checksum, VLAN, protocol, and other fields within
529  * the skb.
530  **/
531 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
532 				       union ixgbe_adv_rx_desc *rx_desc,
533 				       struct sk_buff *skb)
534 {
535 	ixgbevf_rx_hash(rx_ring, rx_desc, skb);
536 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
537 
538 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
539 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
540 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
541 
542 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
543 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
544 	}
545 
546 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
547 }
548 
549 /**
550  * ixgbevf_is_non_eop - process handling of non-EOP buffers
551  * @rx_ring: Rx ring being processed
552  * @rx_desc: Rx descriptor for current buffer
553  * @skb: current socket buffer containing buffer in progress
554  *
555  * This function updates next to clean.  If the buffer is an EOP buffer
556  * this function exits returning false, otherwise it will place the
557  * sk_buff in the next buffer to be chained and return true indicating
558  * that this is in fact a non-EOP buffer.
559  **/
560 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
561 			       union ixgbe_adv_rx_desc *rx_desc)
562 {
563 	u32 ntc = rx_ring->next_to_clean + 1;
564 
565 	/* fetch, update, and store next to clean */
566 	ntc = (ntc < rx_ring->count) ? ntc : 0;
567 	rx_ring->next_to_clean = ntc;
568 
569 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
570 
571 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
572 		return false;
573 
574 	return true;
575 }
576 
577 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
578 				      struct ixgbevf_rx_buffer *bi)
579 {
580 	struct page *page = bi->page;
581 	dma_addr_t dma = bi->dma;
582 
583 	/* since we are recycling buffers we should seldom need to alloc */
584 	if (likely(page))
585 		return true;
586 
587 	/* alloc new page for storage */
588 	page = dev_alloc_page();
589 	if (unlikely(!page)) {
590 		rx_ring->rx_stats.alloc_rx_page_failed++;
591 		return false;
592 	}
593 
594 	/* map page for use */
595 	dma = dma_map_page(rx_ring->dev, page, 0,
596 			   PAGE_SIZE, DMA_FROM_DEVICE);
597 
598 	/* if mapping failed free memory back to system since
599 	 * there isn't much point in holding memory we can't use
600 	 */
601 	if (dma_mapping_error(rx_ring->dev, dma)) {
602 		__free_page(page);
603 
604 		rx_ring->rx_stats.alloc_rx_buff_failed++;
605 		return false;
606 	}
607 
608 	bi->dma = dma;
609 	bi->page = page;
610 	bi->page_offset = 0;
611 
612 	return true;
613 }
614 
615 /**
616  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
617  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
618  * @cleaned_count: number of buffers to replace
619  **/
620 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
621 				     u16 cleaned_count)
622 {
623 	union ixgbe_adv_rx_desc *rx_desc;
624 	struct ixgbevf_rx_buffer *bi;
625 	unsigned int i = rx_ring->next_to_use;
626 
627 	/* nothing to do or no valid netdev defined */
628 	if (!cleaned_count || !rx_ring->netdev)
629 		return;
630 
631 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
632 	bi = &rx_ring->rx_buffer_info[i];
633 	i -= rx_ring->count;
634 
635 	do {
636 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
637 			break;
638 
639 		/* Refresh the desc even if pkt_addr didn't change
640 		 * because each write-back erases this info.
641 		 */
642 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
643 
644 		rx_desc++;
645 		bi++;
646 		i++;
647 		if (unlikely(!i)) {
648 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
649 			bi = rx_ring->rx_buffer_info;
650 			i -= rx_ring->count;
651 		}
652 
653 		/* clear the hdr_addr for the next_to_use descriptor */
654 		rx_desc->read.hdr_addr = 0;
655 
656 		cleaned_count--;
657 	} while (cleaned_count);
658 
659 	i += rx_ring->count;
660 
661 	if (rx_ring->next_to_use != i) {
662 		/* record the next descriptor to use */
663 		rx_ring->next_to_use = i;
664 
665 		/* update next to alloc since we have filled the ring */
666 		rx_ring->next_to_alloc = i;
667 
668 		/* Force memory writes to complete before letting h/w
669 		 * know there are new descriptors to fetch.  (Only
670 		 * applicable for weak-ordered memory model archs,
671 		 * such as IA-64).
672 		 */
673 		wmb();
674 		ixgbevf_write_tail(rx_ring, i);
675 	}
676 }
677 
678 /**
679  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
680  * @rx_ring: rx descriptor ring packet is being transacted on
681  * @rx_desc: pointer to the EOP Rx descriptor
682  * @skb: pointer to current skb being fixed
683  *
684  * Check for corrupted packet headers caused by senders on the local L2
685  * embedded NIC switch not setting up their Tx Descriptors right.  These
686  * should be very rare.
687  *
688  * Also address the case where we are pulling data in on pages only
689  * and as such no data is present in the skb header.
690  *
691  * In addition if skb is not at least 60 bytes we need to pad it so that
692  * it is large enough to qualify as a valid Ethernet frame.
693  *
694  * Returns true if an error was encountered and skb was freed.
695  **/
696 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
697 				    union ixgbe_adv_rx_desc *rx_desc,
698 				    struct sk_buff *skb)
699 {
700 	/* verify that the packet does not have any known errors */
701 	if (unlikely(ixgbevf_test_staterr(rx_desc,
702 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
703 		struct net_device *netdev = rx_ring->netdev;
704 
705 		if (!(netdev->features & NETIF_F_RXALL)) {
706 			dev_kfree_skb_any(skb);
707 			return true;
708 		}
709 	}
710 
711 	/* if eth_skb_pad returns an error the skb was freed */
712 	if (eth_skb_pad(skb))
713 		return true;
714 
715 	return false;
716 }
717 
718 /**
719  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
720  * @rx_ring: rx descriptor ring to store buffers on
721  * @old_buff: donor buffer to have page reused
722  *
723  * Synchronizes page for reuse by the adapter
724  **/
725 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
726 				  struct ixgbevf_rx_buffer *old_buff)
727 {
728 	struct ixgbevf_rx_buffer *new_buff;
729 	u16 nta = rx_ring->next_to_alloc;
730 
731 	new_buff = &rx_ring->rx_buffer_info[nta];
732 
733 	/* update, and store next to alloc */
734 	nta++;
735 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
736 
737 	/* transfer page from old buffer to new buffer */
738 	new_buff->page = old_buff->page;
739 	new_buff->dma = old_buff->dma;
740 	new_buff->page_offset = old_buff->page_offset;
741 
742 	/* sync the buffer for use by the device */
743 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
744 					 new_buff->page_offset,
745 					 IXGBEVF_RX_BUFSZ,
746 					 DMA_FROM_DEVICE);
747 }
748 
749 static inline bool ixgbevf_page_is_reserved(struct page *page)
750 {
751 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
752 }
753 
754 /**
755  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
756  * @rx_ring: rx descriptor ring to transact packets on
757  * @rx_buffer: buffer containing page to add
758  * @rx_desc: descriptor containing length of buffer written by hardware
759  * @skb: sk_buff to place the data into
760  *
761  * This function will add the data contained in rx_buffer->page to the skb.
762  * This is done either through a direct copy if the data in the buffer is
763  * less than the skb header size, otherwise it will just attach the page as
764  * a frag to the skb.
765  *
766  * The function will then update the page offset if necessary and return
767  * true if the buffer can be reused by the adapter.
768  **/
769 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
770 				struct ixgbevf_rx_buffer *rx_buffer,
771 				union ixgbe_adv_rx_desc *rx_desc,
772 				struct sk_buff *skb)
773 {
774 	struct page *page = rx_buffer->page;
775 	unsigned char *va = page_address(page) + rx_buffer->page_offset;
776 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
777 #if (PAGE_SIZE < 8192)
778 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
779 #else
780 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
781 #endif
782 	unsigned int pull_len;
783 
784 	if (unlikely(skb_is_nonlinear(skb)))
785 		goto add_tail_frag;
786 
787 	if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
788 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
789 
790 		/* page is not reserved, we can reuse buffer as is */
791 		if (likely(!ixgbevf_page_is_reserved(page)))
792 			return true;
793 
794 		/* this page cannot be reused so discard it */
795 		put_page(page);
796 		return false;
797 	}
798 
799 	/* we need the header to contain the greater of either ETH_HLEN or
800 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
801 	 */
802 	pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
803 
804 	/* align pull length to size of long to optimize memcpy performance */
805 	memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
806 
807 	/* update all of the pointers */
808 	va += pull_len;
809 	size -= pull_len;
810 
811 add_tail_frag:
812 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
813 			(unsigned long)va & ~PAGE_MASK, size, truesize);
814 
815 	/* avoid re-using remote pages */
816 	if (unlikely(ixgbevf_page_is_reserved(page)))
817 		return false;
818 
819 #if (PAGE_SIZE < 8192)
820 	/* if we are only owner of page we can reuse it */
821 	if (unlikely(page_count(page) != 1))
822 		return false;
823 
824 	/* flip page offset to other buffer */
825 	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
826 
827 #else
828 	/* move offset up to the next cache line */
829 	rx_buffer->page_offset += truesize;
830 
831 	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
832 		return false;
833 
834 #endif
835 	/* Even if we own the page, we are not allowed to use atomic_set()
836 	 * This would break get_page_unless_zero() users.
837 	 */
838 	atomic_inc(&page->_count);
839 
840 	return true;
841 }
842 
843 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
844 					       union ixgbe_adv_rx_desc *rx_desc,
845 					       struct sk_buff *skb)
846 {
847 	struct ixgbevf_rx_buffer *rx_buffer;
848 	struct page *page;
849 
850 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
851 	page = rx_buffer->page;
852 	prefetchw(page);
853 
854 	if (likely(!skb)) {
855 		void *page_addr = page_address(page) +
856 				  rx_buffer->page_offset;
857 
858 		/* prefetch first cache line of first page */
859 		prefetch(page_addr);
860 #if L1_CACHE_BYTES < 128
861 		prefetch(page_addr + L1_CACHE_BYTES);
862 #endif
863 
864 		/* allocate a skb to store the frags */
865 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
866 						IXGBEVF_RX_HDR_SIZE);
867 		if (unlikely(!skb)) {
868 			rx_ring->rx_stats.alloc_rx_buff_failed++;
869 			return NULL;
870 		}
871 
872 		/* we will be copying header into skb->data in
873 		 * pskb_may_pull so it is in our interest to prefetch
874 		 * it now to avoid a possible cache miss
875 		 */
876 		prefetchw(skb->data);
877 	}
878 
879 	/* we are reusing so sync this buffer for CPU use */
880 	dma_sync_single_range_for_cpu(rx_ring->dev,
881 				      rx_buffer->dma,
882 				      rx_buffer->page_offset,
883 				      IXGBEVF_RX_BUFSZ,
884 				      DMA_FROM_DEVICE);
885 
886 	/* pull page into skb */
887 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
888 		/* hand second half of page back to the ring */
889 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
890 	} else {
891 		/* we are not reusing the buffer so unmap it */
892 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
893 			       PAGE_SIZE, DMA_FROM_DEVICE);
894 	}
895 
896 	/* clear contents of buffer_info */
897 	rx_buffer->dma = 0;
898 	rx_buffer->page = NULL;
899 
900 	return skb;
901 }
902 
903 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
904 					     u32 qmask)
905 {
906 	struct ixgbe_hw *hw = &adapter->hw;
907 
908 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
909 }
910 
911 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
912 				struct ixgbevf_ring *rx_ring,
913 				int budget)
914 {
915 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
916 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
917 	struct sk_buff *skb = rx_ring->skb;
918 
919 	while (likely(total_rx_packets < budget)) {
920 		union ixgbe_adv_rx_desc *rx_desc;
921 
922 		/* return some buffers to hardware, one at a time is too slow */
923 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
924 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
925 			cleaned_count = 0;
926 		}
927 
928 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
929 
930 		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
931 			break;
932 
933 		/* This memory barrier is needed to keep us from reading
934 		 * any other fields out of the rx_desc until we know the
935 		 * RXD_STAT_DD bit is set
936 		 */
937 		rmb();
938 
939 		/* retrieve a buffer from the ring */
940 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
941 
942 		/* exit if we failed to retrieve a buffer */
943 		if (!skb)
944 			break;
945 
946 		cleaned_count++;
947 
948 		/* fetch next buffer in frame if non-eop */
949 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
950 			continue;
951 
952 		/* verify the packet layout is correct */
953 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
954 			skb = NULL;
955 			continue;
956 		}
957 
958 		/* probably a little skewed due to removing CRC */
959 		total_rx_bytes += skb->len;
960 
961 		/* Workaround hardware that can't do proper VEPA multicast
962 		 * source pruning.
963 		 */
964 		if ((skb->pkt_type == PACKET_BROADCAST ||
965 		     skb->pkt_type == PACKET_MULTICAST) &&
966 		    ether_addr_equal(rx_ring->netdev->dev_addr,
967 				     eth_hdr(skb)->h_source)) {
968 			dev_kfree_skb_irq(skb);
969 			continue;
970 		}
971 
972 		/* populate checksum, VLAN, and protocol */
973 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
974 
975 		ixgbevf_rx_skb(q_vector, skb);
976 
977 		/* reset skb pointer */
978 		skb = NULL;
979 
980 		/* update budget accounting */
981 		total_rx_packets++;
982 	}
983 
984 	/* place incomplete frames back on ring for completion */
985 	rx_ring->skb = skb;
986 
987 	u64_stats_update_begin(&rx_ring->syncp);
988 	rx_ring->stats.packets += total_rx_packets;
989 	rx_ring->stats.bytes += total_rx_bytes;
990 	u64_stats_update_end(&rx_ring->syncp);
991 	q_vector->rx.total_packets += total_rx_packets;
992 	q_vector->rx.total_bytes += total_rx_bytes;
993 
994 	return total_rx_packets;
995 }
996 
997 /**
998  * ixgbevf_poll - NAPI polling calback
999  * @napi: napi struct with our devices info in it
1000  * @budget: amount of work driver is allowed to do this pass, in packets
1001  *
1002  * This function will clean more than one or more rings associated with a
1003  * q_vector.
1004  **/
1005 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1006 {
1007 	struct ixgbevf_q_vector *q_vector =
1008 		container_of(napi, struct ixgbevf_q_vector, napi);
1009 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1010 	struct ixgbevf_ring *ring;
1011 	int per_ring_budget;
1012 	bool clean_complete = true;
1013 
1014 	ixgbevf_for_each_ring(ring, q_vector->tx)
1015 		clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1016 
1017 #ifdef CONFIG_NET_RX_BUSY_POLL
1018 	if (!ixgbevf_qv_lock_napi(q_vector))
1019 		return budget;
1020 #endif
1021 
1022 	/* attempt to distribute budget to each queue fairly, but don't allow
1023 	 * the budget to go below 1 because we'll exit polling
1024 	 */
1025 	if (q_vector->rx.count > 1)
1026 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1027 	else
1028 		per_ring_budget = budget;
1029 
1030 	ixgbevf_for_each_ring(ring, q_vector->rx)
1031 		clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
1032 							per_ring_budget)
1033 				   < per_ring_budget);
1034 
1035 #ifdef CONFIG_NET_RX_BUSY_POLL
1036 	ixgbevf_qv_unlock_napi(q_vector);
1037 #endif
1038 
1039 	/* If all work not completed, return budget and keep polling */
1040 	if (!clean_complete)
1041 		return budget;
1042 	/* all work done, exit the polling mode */
1043 	napi_complete(napi);
1044 	if (adapter->rx_itr_setting & 1)
1045 		ixgbevf_set_itr(q_vector);
1046 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1047 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1048 		ixgbevf_irq_enable_queues(adapter,
1049 					  1 << q_vector->v_idx);
1050 
1051 	return 0;
1052 }
1053 
1054 /**
1055  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1056  * @q_vector: structure containing interrupt and ring information
1057  **/
1058 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1059 {
1060 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1061 	struct ixgbe_hw *hw = &adapter->hw;
1062 	int v_idx = q_vector->v_idx;
1063 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1064 
1065 	/* set the WDIS bit to not clear the timer bits and cause an
1066 	 * immediate assertion of the interrupt
1067 	 */
1068 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1069 
1070 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1071 }
1072 
1073 #ifdef CONFIG_NET_RX_BUSY_POLL
1074 /* must be called with local_bh_disable()d */
1075 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1076 {
1077 	struct ixgbevf_q_vector *q_vector =
1078 			container_of(napi, struct ixgbevf_q_vector, napi);
1079 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1080 	struct ixgbevf_ring  *ring;
1081 	int found = 0;
1082 
1083 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1084 		return LL_FLUSH_FAILED;
1085 
1086 	if (!ixgbevf_qv_lock_poll(q_vector))
1087 		return LL_FLUSH_BUSY;
1088 
1089 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1090 		found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1091 #ifdef BP_EXTENDED_STATS
1092 		if (found)
1093 			ring->stats.cleaned += found;
1094 		else
1095 			ring->stats.misses++;
1096 #endif
1097 		if (found)
1098 			break;
1099 	}
1100 
1101 	ixgbevf_qv_unlock_poll(q_vector);
1102 
1103 	return found;
1104 }
1105 #endif /* CONFIG_NET_RX_BUSY_POLL */
1106 
1107 /**
1108  * ixgbevf_configure_msix - Configure MSI-X hardware
1109  * @adapter: board private structure
1110  *
1111  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1112  * interrupts.
1113  **/
1114 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1115 {
1116 	struct ixgbevf_q_vector *q_vector;
1117 	int q_vectors, v_idx;
1118 
1119 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1120 	adapter->eims_enable_mask = 0;
1121 
1122 	/* Populate the IVAR table and set the ITR values to the
1123 	 * corresponding register.
1124 	 */
1125 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1126 		struct ixgbevf_ring *ring;
1127 
1128 		q_vector = adapter->q_vector[v_idx];
1129 
1130 		ixgbevf_for_each_ring(ring, q_vector->rx)
1131 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1132 
1133 		ixgbevf_for_each_ring(ring, q_vector->tx)
1134 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1135 
1136 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1137 			/* Tx only vector */
1138 			if (adapter->tx_itr_setting == 1)
1139 				q_vector->itr = IXGBE_10K_ITR;
1140 			else
1141 				q_vector->itr = adapter->tx_itr_setting;
1142 		} else {
1143 			/* Rx or Rx/Tx vector */
1144 			if (adapter->rx_itr_setting == 1)
1145 				q_vector->itr = IXGBE_20K_ITR;
1146 			else
1147 				q_vector->itr = adapter->rx_itr_setting;
1148 		}
1149 
1150 		/* add q_vector eims value to global eims_enable_mask */
1151 		adapter->eims_enable_mask |= 1 << v_idx;
1152 
1153 		ixgbevf_write_eitr(q_vector);
1154 	}
1155 
1156 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1157 	/* setup eims_other and add value to global eims_enable_mask */
1158 	adapter->eims_other = 1 << v_idx;
1159 	adapter->eims_enable_mask |= adapter->eims_other;
1160 }
1161 
1162 enum latency_range {
1163 	lowest_latency = 0,
1164 	low_latency = 1,
1165 	bulk_latency = 2,
1166 	latency_invalid = 255
1167 };
1168 
1169 /**
1170  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1171  * @q_vector: structure containing interrupt and ring information
1172  * @ring_container: structure containing ring performance data
1173  *
1174  * Stores a new ITR value based on packets and byte
1175  * counts during the last interrupt.  The advantage of per interrupt
1176  * computation is faster updates and more accurate ITR for the current
1177  * traffic pattern.  Constants in this function were computed
1178  * based on theoretical maximum wire speed and thresholds were set based
1179  * on testing data as well as attempting to minimize response time
1180  * while increasing bulk throughput.
1181  **/
1182 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1183 			       struct ixgbevf_ring_container *ring_container)
1184 {
1185 	int bytes = ring_container->total_bytes;
1186 	int packets = ring_container->total_packets;
1187 	u32 timepassed_us;
1188 	u64 bytes_perint;
1189 	u8 itr_setting = ring_container->itr;
1190 
1191 	if (packets == 0)
1192 		return;
1193 
1194 	/* simple throttle rate management
1195 	 *    0-20MB/s lowest (100000 ints/s)
1196 	 *   20-100MB/s low   (20000 ints/s)
1197 	 *  100-1249MB/s bulk (8000 ints/s)
1198 	 */
1199 	/* what was last interrupt timeslice? */
1200 	timepassed_us = q_vector->itr >> 2;
1201 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1202 
1203 	switch (itr_setting) {
1204 	case lowest_latency:
1205 		if (bytes_perint > 10)
1206 			itr_setting = low_latency;
1207 		break;
1208 	case low_latency:
1209 		if (bytes_perint > 20)
1210 			itr_setting = bulk_latency;
1211 		else if (bytes_perint <= 10)
1212 			itr_setting = lowest_latency;
1213 		break;
1214 	case bulk_latency:
1215 		if (bytes_perint <= 20)
1216 			itr_setting = low_latency;
1217 		break;
1218 	}
1219 
1220 	/* clear work counters since we have the values we need */
1221 	ring_container->total_bytes = 0;
1222 	ring_container->total_packets = 0;
1223 
1224 	/* write updated itr to ring container */
1225 	ring_container->itr = itr_setting;
1226 }
1227 
1228 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1229 {
1230 	u32 new_itr = q_vector->itr;
1231 	u8 current_itr;
1232 
1233 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1234 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1235 
1236 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1237 
1238 	switch (current_itr) {
1239 	/* counts and packets in update_itr are dependent on these numbers */
1240 	case lowest_latency:
1241 		new_itr = IXGBE_100K_ITR;
1242 		break;
1243 	case low_latency:
1244 		new_itr = IXGBE_20K_ITR;
1245 		break;
1246 	case bulk_latency:
1247 	default:
1248 		new_itr = IXGBE_8K_ITR;
1249 		break;
1250 	}
1251 
1252 	if (new_itr != q_vector->itr) {
1253 		/* do an exponential smoothing */
1254 		new_itr = (10 * new_itr * q_vector->itr) /
1255 			  ((9 * new_itr) + q_vector->itr);
1256 
1257 		/* save the algorithm value here */
1258 		q_vector->itr = new_itr;
1259 
1260 		ixgbevf_write_eitr(q_vector);
1261 	}
1262 }
1263 
1264 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1265 {
1266 	struct ixgbevf_adapter *adapter = data;
1267 	struct ixgbe_hw *hw = &adapter->hw;
1268 
1269 	hw->mac.get_link_status = 1;
1270 
1271 	ixgbevf_service_event_schedule(adapter);
1272 
1273 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1274 
1275 	return IRQ_HANDLED;
1276 }
1277 
1278 /**
1279  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1280  * @irq: unused
1281  * @data: pointer to our q_vector struct for this interrupt vector
1282  **/
1283 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1284 {
1285 	struct ixgbevf_q_vector *q_vector = data;
1286 
1287 	/* EIAM disabled interrupts (on this vector) for us */
1288 	if (q_vector->rx.ring || q_vector->tx.ring)
1289 		napi_schedule(&q_vector->napi);
1290 
1291 	return IRQ_HANDLED;
1292 }
1293 
1294 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1295 				     int r_idx)
1296 {
1297 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1298 
1299 	a->rx_ring[r_idx]->next = q_vector->rx.ring;
1300 	q_vector->rx.ring = a->rx_ring[r_idx];
1301 	q_vector->rx.count++;
1302 }
1303 
1304 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1305 				     int t_idx)
1306 {
1307 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1308 
1309 	a->tx_ring[t_idx]->next = q_vector->tx.ring;
1310 	q_vector->tx.ring = a->tx_ring[t_idx];
1311 	q_vector->tx.count++;
1312 }
1313 
1314 /**
1315  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1316  * @adapter: board private structure to initialize
1317  *
1318  * This function maps descriptor rings to the queue-specific vectors
1319  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1320  * one vector per ring/queue, but on a constrained vector budget, we
1321  * group the rings as "efficiently" as possible.  You would add new
1322  * mapping configurations in here.
1323  **/
1324 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1325 {
1326 	int q_vectors;
1327 	int v_start = 0;
1328 	int rxr_idx = 0, txr_idx = 0;
1329 	int rxr_remaining = adapter->num_rx_queues;
1330 	int txr_remaining = adapter->num_tx_queues;
1331 	int i, j;
1332 	int rqpv, tqpv;
1333 	int err = 0;
1334 
1335 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1336 
1337 	/* The ideal configuration...
1338 	 * We have enough vectors to map one per queue.
1339 	 */
1340 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1341 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1342 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1343 
1344 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1345 			map_vector_to_txq(adapter, v_start, txr_idx);
1346 		goto out;
1347 	}
1348 
1349 	/* If we don't have enough vectors for a 1-to-1
1350 	 * mapping, we'll have to group them so there are
1351 	 * multiple queues per vector.
1352 	 */
1353 	/* Re-adjusting *qpv takes care of the remainder. */
1354 	for (i = v_start; i < q_vectors; i++) {
1355 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1356 		for (j = 0; j < rqpv; j++) {
1357 			map_vector_to_rxq(adapter, i, rxr_idx);
1358 			rxr_idx++;
1359 			rxr_remaining--;
1360 		}
1361 	}
1362 	for (i = v_start; i < q_vectors; i++) {
1363 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1364 		for (j = 0; j < tqpv; j++) {
1365 			map_vector_to_txq(adapter, i, txr_idx);
1366 			txr_idx++;
1367 			txr_remaining--;
1368 		}
1369 	}
1370 
1371 out:
1372 	return err;
1373 }
1374 
1375 /**
1376  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1377  * @adapter: board private structure
1378  *
1379  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1380  * interrupts from the kernel.
1381  **/
1382 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1383 {
1384 	struct net_device *netdev = adapter->netdev;
1385 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1386 	int vector, err;
1387 	int ri = 0, ti = 0;
1388 
1389 	for (vector = 0; vector < q_vectors; vector++) {
1390 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1391 		struct msix_entry *entry = &adapter->msix_entries[vector];
1392 
1393 		if (q_vector->tx.ring && q_vector->rx.ring) {
1394 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1395 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
1396 			ti++;
1397 		} else if (q_vector->rx.ring) {
1398 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1399 				 "%s-%s-%d", netdev->name, "rx", ri++);
1400 		} else if (q_vector->tx.ring) {
1401 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1402 				 "%s-%s-%d", netdev->name, "tx", ti++);
1403 		} else {
1404 			/* skip this unused q_vector */
1405 			continue;
1406 		}
1407 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1408 				  q_vector->name, q_vector);
1409 		if (err) {
1410 			hw_dbg(&adapter->hw,
1411 			       "request_irq failed for MSIX interrupt Error: %d\n",
1412 			       err);
1413 			goto free_queue_irqs;
1414 		}
1415 	}
1416 
1417 	err = request_irq(adapter->msix_entries[vector].vector,
1418 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1419 	if (err) {
1420 		hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1421 		       err);
1422 		goto free_queue_irqs;
1423 	}
1424 
1425 	return 0;
1426 
1427 free_queue_irqs:
1428 	while (vector) {
1429 		vector--;
1430 		free_irq(adapter->msix_entries[vector].vector,
1431 			 adapter->q_vector[vector]);
1432 	}
1433 	/* This failure is non-recoverable - it indicates the system is
1434 	 * out of MSIX vector resources and the VF driver cannot run
1435 	 * without them.  Set the number of msix vectors to zero
1436 	 * indicating that not enough can be allocated.  The error
1437 	 * will be returned to the user indicating device open failed.
1438 	 * Any further attempts to force the driver to open will also
1439 	 * fail.  The only way to recover is to unload the driver and
1440 	 * reload it again.  If the system has recovered some MSIX
1441 	 * vectors then it may succeed.
1442 	 */
1443 	adapter->num_msix_vectors = 0;
1444 	return err;
1445 }
1446 
1447 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1448 {
1449 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1450 
1451 	for (i = 0; i < q_vectors; i++) {
1452 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1453 
1454 		q_vector->rx.ring = NULL;
1455 		q_vector->tx.ring = NULL;
1456 		q_vector->rx.count = 0;
1457 		q_vector->tx.count = 0;
1458 	}
1459 }
1460 
1461 /**
1462  * ixgbevf_request_irq - initialize interrupts
1463  * @adapter: board private structure
1464  *
1465  * Attempts to configure interrupts using the best available
1466  * capabilities of the hardware and kernel.
1467  **/
1468 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1469 {
1470 	int err = 0;
1471 
1472 	err = ixgbevf_request_msix_irqs(adapter);
1473 
1474 	if (err)
1475 		hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1476 
1477 	return err;
1478 }
1479 
1480 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1481 {
1482 	int i, q_vectors;
1483 
1484 	q_vectors = adapter->num_msix_vectors;
1485 	i = q_vectors - 1;
1486 
1487 	free_irq(adapter->msix_entries[i].vector, adapter);
1488 	i--;
1489 
1490 	for (; i >= 0; i--) {
1491 		/* free only the irqs that were actually requested */
1492 		if (!adapter->q_vector[i]->rx.ring &&
1493 		    !adapter->q_vector[i]->tx.ring)
1494 			continue;
1495 
1496 		free_irq(adapter->msix_entries[i].vector,
1497 			 adapter->q_vector[i]);
1498 	}
1499 
1500 	ixgbevf_reset_q_vectors(adapter);
1501 }
1502 
1503 /**
1504  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1505  * @adapter: board private structure
1506  **/
1507 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1508 {
1509 	struct ixgbe_hw *hw = &adapter->hw;
1510 	int i;
1511 
1512 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1513 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1514 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1515 
1516 	IXGBE_WRITE_FLUSH(hw);
1517 
1518 	for (i = 0; i < adapter->num_msix_vectors; i++)
1519 		synchronize_irq(adapter->msix_entries[i].vector);
1520 }
1521 
1522 /**
1523  * ixgbevf_irq_enable - Enable default interrupt generation settings
1524  * @adapter: board private structure
1525  **/
1526 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1527 {
1528 	struct ixgbe_hw *hw = &adapter->hw;
1529 
1530 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1531 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1532 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1533 }
1534 
1535 /**
1536  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1537  * @adapter: board private structure
1538  * @ring: structure containing ring specific data
1539  *
1540  * Configure the Tx descriptor ring after a reset.
1541  **/
1542 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1543 				      struct ixgbevf_ring *ring)
1544 {
1545 	struct ixgbe_hw *hw = &adapter->hw;
1546 	u64 tdba = ring->dma;
1547 	int wait_loop = 10;
1548 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1549 	u8 reg_idx = ring->reg_idx;
1550 
1551 	/* disable queue to avoid issues while updating state */
1552 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1553 	IXGBE_WRITE_FLUSH(hw);
1554 
1555 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1556 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1557 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1558 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1559 
1560 	/* disable head writeback */
1561 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1562 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1563 
1564 	/* enable relaxed ordering */
1565 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1566 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1567 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1568 
1569 	/* reset head and tail pointers */
1570 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1571 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1572 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1573 
1574 	/* reset ntu and ntc to place SW in sync with hardwdare */
1575 	ring->next_to_clean = 0;
1576 	ring->next_to_use = 0;
1577 
1578 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1579 	 * to or less than the number of on chip descriptors, which is
1580 	 * currently 40.
1581 	 */
1582 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1583 
1584 	/* Setting PTHRESH to 32 both improves performance */
1585 	txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1586 		  32;          /* PTHRESH = 32 */
1587 
1588 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1589 
1590 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1591 
1592 	/* poll to verify queue is enabled */
1593 	do {
1594 		usleep_range(1000, 2000);
1595 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1596 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1597 	if (!wait_loop)
1598 		pr_err("Could not enable Tx Queue %d\n", reg_idx);
1599 }
1600 
1601 /**
1602  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1603  * @adapter: board private structure
1604  *
1605  * Configure the Tx unit of the MAC after a reset.
1606  **/
1607 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1608 {
1609 	u32 i;
1610 
1611 	/* Setup the HW Tx Head and Tail descriptor pointers */
1612 	for (i = 0; i < adapter->num_tx_queues; i++)
1613 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1614 }
1615 
1616 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1617 
1618 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1619 {
1620 	struct ixgbe_hw *hw = &adapter->hw;
1621 	u32 srrctl;
1622 
1623 	srrctl = IXGBE_SRRCTL_DROP_EN;
1624 
1625 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1626 	srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1627 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1628 
1629 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1630 }
1631 
1632 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1633 {
1634 	struct ixgbe_hw *hw = &adapter->hw;
1635 
1636 	/* PSRTYPE must be initialized in 82599 */
1637 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1638 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1639 		      IXGBE_PSRTYPE_L2HDR;
1640 
1641 	if (adapter->num_rx_queues > 1)
1642 		psrtype |= 1 << 29;
1643 
1644 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1645 }
1646 
1647 #define IXGBEVF_MAX_RX_DESC_POLL 10
1648 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1649 				     struct ixgbevf_ring *ring)
1650 {
1651 	struct ixgbe_hw *hw = &adapter->hw;
1652 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1653 	u32 rxdctl;
1654 	u8 reg_idx = ring->reg_idx;
1655 
1656 	if (IXGBE_REMOVED(hw->hw_addr))
1657 		return;
1658 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1659 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1660 
1661 	/* write value back with RXDCTL.ENABLE bit cleared */
1662 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1663 
1664 	/* the hardware may take up to 100us to really disable the Rx queue */
1665 	do {
1666 		udelay(10);
1667 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1668 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1669 
1670 	if (!wait_loop)
1671 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1672 		       reg_idx);
1673 }
1674 
1675 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1676 					 struct ixgbevf_ring *ring)
1677 {
1678 	struct ixgbe_hw *hw = &adapter->hw;
1679 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1680 	u32 rxdctl;
1681 	u8 reg_idx = ring->reg_idx;
1682 
1683 	if (IXGBE_REMOVED(hw->hw_addr))
1684 		return;
1685 	do {
1686 		usleep_range(1000, 2000);
1687 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1688 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1689 
1690 	if (!wait_loop)
1691 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1692 		       reg_idx);
1693 }
1694 
1695 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1696 {
1697 	struct ixgbe_hw *hw = &adapter->hw;
1698 	u32 vfmrqc = 0, vfreta = 0;
1699 	u16 rss_i = adapter->num_rx_queues;
1700 	u8 i, j;
1701 
1702 	/* Fill out hash function seeds */
1703 	netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1704 	for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1705 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1706 
1707 	for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1708 		if (j == rss_i)
1709 			j = 0;
1710 
1711 		adapter->rss_indir_tbl[i] = j;
1712 
1713 		vfreta |= j << (i & 0x3) * 8;
1714 		if ((i & 3) == 3) {
1715 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1716 			vfreta = 0;
1717 		}
1718 	}
1719 
1720 	/* Perform hash on these packet types */
1721 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1722 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1723 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1724 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1725 
1726 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1727 
1728 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1729 }
1730 
1731 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1732 				      struct ixgbevf_ring *ring)
1733 {
1734 	struct ixgbe_hw *hw = &adapter->hw;
1735 	u64 rdba = ring->dma;
1736 	u32 rxdctl;
1737 	u8 reg_idx = ring->reg_idx;
1738 
1739 	/* disable queue to avoid issues while updating state */
1740 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1741 	ixgbevf_disable_rx_queue(adapter, ring);
1742 
1743 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1744 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1745 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1746 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1747 
1748 	/* enable relaxed ordering */
1749 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1750 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1751 
1752 	/* reset head and tail pointers */
1753 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1754 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1755 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1756 
1757 	/* reset ntu and ntc to place SW in sync with hardwdare */
1758 	ring->next_to_clean = 0;
1759 	ring->next_to_use = 0;
1760 	ring->next_to_alloc = 0;
1761 
1762 	ixgbevf_configure_srrctl(adapter, reg_idx);
1763 
1764 	/* allow any size packet since we can handle overflow */
1765 	rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1766 
1767 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1768 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1769 
1770 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1771 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1772 }
1773 
1774 /**
1775  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1776  * @adapter: board private structure
1777  *
1778  * Configure the Rx unit of the MAC after a reset.
1779  **/
1780 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1781 {
1782 	int i;
1783 	struct ixgbe_hw *hw = &adapter->hw;
1784 	struct net_device *netdev = adapter->netdev;
1785 
1786 	ixgbevf_setup_psrtype(adapter);
1787 	if (hw->mac.type >= ixgbe_mac_X550_vf)
1788 		ixgbevf_setup_vfmrqc(adapter);
1789 
1790 	/* notify the PF of our intent to use this size of frame */
1791 	ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1792 
1793 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1794 	 * the Base and Length of the Rx Descriptor Ring
1795 	 */
1796 	for (i = 0; i < adapter->num_rx_queues; i++)
1797 		ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1798 }
1799 
1800 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1801 				   __be16 proto, u16 vid)
1802 {
1803 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1804 	struct ixgbe_hw *hw = &adapter->hw;
1805 	int err;
1806 
1807 	spin_lock_bh(&adapter->mbx_lock);
1808 
1809 	/* add VID to filter table */
1810 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1811 
1812 	spin_unlock_bh(&adapter->mbx_lock);
1813 
1814 	/* translate error return types so error makes sense */
1815 	if (err == IXGBE_ERR_MBX)
1816 		return -EIO;
1817 
1818 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
1819 		return -EACCES;
1820 
1821 	set_bit(vid, adapter->active_vlans);
1822 
1823 	return err;
1824 }
1825 
1826 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1827 				    __be16 proto, u16 vid)
1828 {
1829 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1830 	struct ixgbe_hw *hw = &adapter->hw;
1831 	int err = -EOPNOTSUPP;
1832 
1833 	spin_lock_bh(&adapter->mbx_lock);
1834 
1835 	/* remove VID from filter table */
1836 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1837 
1838 	spin_unlock_bh(&adapter->mbx_lock);
1839 
1840 	clear_bit(vid, adapter->active_vlans);
1841 
1842 	return err;
1843 }
1844 
1845 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1846 {
1847 	u16 vid;
1848 
1849 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1850 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
1851 					htons(ETH_P_8021Q), vid);
1852 }
1853 
1854 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1855 {
1856 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1857 	struct ixgbe_hw *hw = &adapter->hw;
1858 	int count = 0;
1859 
1860 	if ((netdev_uc_count(netdev)) > 10) {
1861 		pr_err("Too many unicast filters - No Space\n");
1862 		return -ENOSPC;
1863 	}
1864 
1865 	if (!netdev_uc_empty(netdev)) {
1866 		struct netdev_hw_addr *ha;
1867 
1868 		netdev_for_each_uc_addr(ha, netdev) {
1869 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1870 			udelay(200);
1871 		}
1872 	} else {
1873 		/* If the list is empty then send message to PF driver to
1874 		 * clear all MAC VLANs on this VF.
1875 		 */
1876 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1877 	}
1878 
1879 	return count;
1880 }
1881 
1882 /**
1883  * ixgbevf_set_rx_mode - Multicast and unicast set
1884  * @netdev: network interface device structure
1885  *
1886  * The set_rx_method entry point is called whenever the multicast address
1887  * list, unicast address list or the network interface flags are updated.
1888  * This routine is responsible for configuring the hardware for proper
1889  * multicast mode and configuring requested unicast filters.
1890  **/
1891 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1892 {
1893 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1894 	struct ixgbe_hw *hw = &adapter->hw;
1895 
1896 	spin_lock_bh(&adapter->mbx_lock);
1897 
1898 	/* reprogram multicast list */
1899 	hw->mac.ops.update_mc_addr_list(hw, netdev);
1900 
1901 	ixgbevf_write_uc_addr_list(netdev);
1902 
1903 	spin_unlock_bh(&adapter->mbx_lock);
1904 }
1905 
1906 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1907 {
1908 	int q_idx;
1909 	struct ixgbevf_q_vector *q_vector;
1910 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1911 
1912 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1913 		q_vector = adapter->q_vector[q_idx];
1914 #ifdef CONFIG_NET_RX_BUSY_POLL
1915 		ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1916 #endif
1917 		napi_enable(&q_vector->napi);
1918 	}
1919 }
1920 
1921 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1922 {
1923 	int q_idx;
1924 	struct ixgbevf_q_vector *q_vector;
1925 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1926 
1927 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1928 		q_vector = adapter->q_vector[q_idx];
1929 		napi_disable(&q_vector->napi);
1930 #ifdef CONFIG_NET_RX_BUSY_POLL
1931 		while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1932 			pr_info("QV %d locked\n", q_idx);
1933 			usleep_range(1000, 20000);
1934 		}
1935 #endif /* CONFIG_NET_RX_BUSY_POLL */
1936 	}
1937 }
1938 
1939 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1940 {
1941 	struct ixgbe_hw *hw = &adapter->hw;
1942 	unsigned int def_q = 0;
1943 	unsigned int num_tcs = 0;
1944 	unsigned int num_rx_queues = adapter->num_rx_queues;
1945 	unsigned int num_tx_queues = adapter->num_tx_queues;
1946 	int err;
1947 
1948 	spin_lock_bh(&adapter->mbx_lock);
1949 
1950 	/* fetch queue configuration from the PF */
1951 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1952 
1953 	spin_unlock_bh(&adapter->mbx_lock);
1954 
1955 	if (err)
1956 		return err;
1957 
1958 	if (num_tcs > 1) {
1959 		/* we need only one Tx queue */
1960 		num_tx_queues = 1;
1961 
1962 		/* update default Tx ring register index */
1963 		adapter->tx_ring[0]->reg_idx = def_q;
1964 
1965 		/* we need as many queues as traffic classes */
1966 		num_rx_queues = num_tcs;
1967 	}
1968 
1969 	/* if we have a bad config abort request queue reset */
1970 	if ((adapter->num_rx_queues != num_rx_queues) ||
1971 	    (adapter->num_tx_queues != num_tx_queues)) {
1972 		/* force mailbox timeout to prevent further messages */
1973 		hw->mbx.timeout = 0;
1974 
1975 		/* wait for watchdog to come around and bail us out */
1976 		adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1977 	}
1978 
1979 	return 0;
1980 }
1981 
1982 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1983 {
1984 	ixgbevf_configure_dcb(adapter);
1985 
1986 	ixgbevf_set_rx_mode(adapter->netdev);
1987 
1988 	ixgbevf_restore_vlan(adapter);
1989 
1990 	ixgbevf_configure_tx(adapter);
1991 	ixgbevf_configure_rx(adapter);
1992 }
1993 
1994 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1995 {
1996 	/* Only save pre-reset stats if there are some */
1997 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1998 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1999 			adapter->stats.base_vfgprc;
2000 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2001 			adapter->stats.base_vfgptc;
2002 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2003 			adapter->stats.base_vfgorc;
2004 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2005 			adapter->stats.base_vfgotc;
2006 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2007 			adapter->stats.base_vfmprc;
2008 	}
2009 }
2010 
2011 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2012 {
2013 	struct ixgbe_hw *hw = &adapter->hw;
2014 
2015 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2016 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2017 	adapter->stats.last_vfgorc |=
2018 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2019 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2020 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2021 	adapter->stats.last_vfgotc |=
2022 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2023 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2024 
2025 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2026 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2027 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2028 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2029 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2030 }
2031 
2032 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2033 {
2034 	struct ixgbe_hw *hw = &adapter->hw;
2035 	int api[] = { ixgbe_mbox_api_12,
2036 		      ixgbe_mbox_api_11,
2037 		      ixgbe_mbox_api_10,
2038 		      ixgbe_mbox_api_unknown };
2039 	int err = 0, idx = 0;
2040 
2041 	spin_lock_bh(&adapter->mbx_lock);
2042 
2043 	while (api[idx] != ixgbe_mbox_api_unknown) {
2044 		err = ixgbevf_negotiate_api_version(hw, api[idx]);
2045 		if (!err)
2046 			break;
2047 		idx++;
2048 	}
2049 
2050 	spin_unlock_bh(&adapter->mbx_lock);
2051 }
2052 
2053 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2054 {
2055 	struct net_device *netdev = adapter->netdev;
2056 	struct ixgbe_hw *hw = &adapter->hw;
2057 
2058 	ixgbevf_configure_msix(adapter);
2059 
2060 	spin_lock_bh(&adapter->mbx_lock);
2061 
2062 	if (is_valid_ether_addr(hw->mac.addr))
2063 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2064 	else
2065 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2066 
2067 	spin_unlock_bh(&adapter->mbx_lock);
2068 
2069 	smp_mb__before_atomic();
2070 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2071 	ixgbevf_napi_enable_all(adapter);
2072 
2073 	/* clear any pending interrupts, may auto mask */
2074 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2075 	ixgbevf_irq_enable(adapter);
2076 
2077 	/* enable transmits */
2078 	netif_tx_start_all_queues(netdev);
2079 
2080 	ixgbevf_save_reset_stats(adapter);
2081 	ixgbevf_init_last_counter_stats(adapter);
2082 
2083 	hw->mac.get_link_status = 1;
2084 	mod_timer(&adapter->service_timer, jiffies);
2085 }
2086 
2087 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2088 {
2089 	ixgbevf_configure(adapter);
2090 
2091 	ixgbevf_up_complete(adapter);
2092 }
2093 
2094 /**
2095  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2096  * @rx_ring: ring to free buffers from
2097  **/
2098 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2099 {
2100 	struct device *dev = rx_ring->dev;
2101 	unsigned long size;
2102 	unsigned int i;
2103 
2104 	/* Free Rx ring sk_buff */
2105 	if (rx_ring->skb) {
2106 		dev_kfree_skb(rx_ring->skb);
2107 		rx_ring->skb = NULL;
2108 	}
2109 
2110 	/* ring already cleared, nothing to do */
2111 	if (!rx_ring->rx_buffer_info)
2112 		return;
2113 
2114 	/* Free all the Rx ring pages */
2115 	for (i = 0; i < rx_ring->count; i++) {
2116 		struct ixgbevf_rx_buffer *rx_buffer;
2117 
2118 		rx_buffer = &rx_ring->rx_buffer_info[i];
2119 		if (rx_buffer->dma)
2120 			dma_unmap_page(dev, rx_buffer->dma,
2121 				       PAGE_SIZE, DMA_FROM_DEVICE);
2122 		rx_buffer->dma = 0;
2123 		if (rx_buffer->page)
2124 			__free_page(rx_buffer->page);
2125 		rx_buffer->page = NULL;
2126 	}
2127 
2128 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2129 	memset(rx_ring->rx_buffer_info, 0, size);
2130 
2131 	/* Zero out the descriptor ring */
2132 	memset(rx_ring->desc, 0, rx_ring->size);
2133 }
2134 
2135 /**
2136  * ixgbevf_clean_tx_ring - Free Tx Buffers
2137  * @tx_ring: ring to be cleaned
2138  **/
2139 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2140 {
2141 	struct ixgbevf_tx_buffer *tx_buffer_info;
2142 	unsigned long size;
2143 	unsigned int i;
2144 
2145 	if (!tx_ring->tx_buffer_info)
2146 		return;
2147 
2148 	/* Free all the Tx ring sk_buffs */
2149 	for (i = 0; i < tx_ring->count; i++) {
2150 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2151 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2152 	}
2153 
2154 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2155 	memset(tx_ring->tx_buffer_info, 0, size);
2156 
2157 	memset(tx_ring->desc, 0, tx_ring->size);
2158 }
2159 
2160 /**
2161  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2162  * @adapter: board private structure
2163  **/
2164 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2165 {
2166 	int i;
2167 
2168 	for (i = 0; i < adapter->num_rx_queues; i++)
2169 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2170 }
2171 
2172 /**
2173  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2174  * @adapter: board private structure
2175  **/
2176 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2177 {
2178 	int i;
2179 
2180 	for (i = 0; i < adapter->num_tx_queues; i++)
2181 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2182 }
2183 
2184 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2185 {
2186 	struct net_device *netdev = adapter->netdev;
2187 	struct ixgbe_hw *hw = &adapter->hw;
2188 	int i;
2189 
2190 	/* signal that we are down to the interrupt handler */
2191 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2192 		return; /* do nothing if already down */
2193 
2194 	/* disable all enabled Rx queues */
2195 	for (i = 0; i < adapter->num_rx_queues; i++)
2196 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2197 
2198 	usleep_range(10000, 20000);
2199 
2200 	netif_tx_stop_all_queues(netdev);
2201 
2202 	/* call carrier off first to avoid false dev_watchdog timeouts */
2203 	netif_carrier_off(netdev);
2204 	netif_tx_disable(netdev);
2205 
2206 	ixgbevf_irq_disable(adapter);
2207 
2208 	ixgbevf_napi_disable_all(adapter);
2209 
2210 	del_timer_sync(&adapter->service_timer);
2211 
2212 	/* disable transmits in the hardware now that interrupts are off */
2213 	for (i = 0; i < adapter->num_tx_queues; i++) {
2214 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2215 
2216 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2217 				IXGBE_TXDCTL_SWFLSH);
2218 	}
2219 
2220 	if (!pci_channel_offline(adapter->pdev))
2221 		ixgbevf_reset(adapter);
2222 
2223 	ixgbevf_clean_all_tx_rings(adapter);
2224 	ixgbevf_clean_all_rx_rings(adapter);
2225 }
2226 
2227 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2228 {
2229 	WARN_ON(in_interrupt());
2230 
2231 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2232 		msleep(1);
2233 
2234 	ixgbevf_down(adapter);
2235 	ixgbevf_up(adapter);
2236 
2237 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2238 }
2239 
2240 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2241 {
2242 	struct ixgbe_hw *hw = &adapter->hw;
2243 	struct net_device *netdev = adapter->netdev;
2244 
2245 	if (hw->mac.ops.reset_hw(hw)) {
2246 		hw_dbg(hw, "PF still resetting\n");
2247 	} else {
2248 		hw->mac.ops.init_hw(hw);
2249 		ixgbevf_negotiate_api(adapter);
2250 	}
2251 
2252 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2253 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2254 		       netdev->addr_len);
2255 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2256 		       netdev->addr_len);
2257 	}
2258 
2259 	adapter->last_reset = jiffies;
2260 }
2261 
2262 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2263 					int vectors)
2264 {
2265 	int vector_threshold;
2266 
2267 	/* We'll want at least 2 (vector_threshold):
2268 	 * 1) TxQ[0] + RxQ[0] handler
2269 	 * 2) Other (Link Status Change, etc.)
2270 	 */
2271 	vector_threshold = MIN_MSIX_COUNT;
2272 
2273 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2274 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2275 	 * Right now, we simply care about how many we'll get; we'll
2276 	 * set them up later while requesting irq's.
2277 	 */
2278 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2279 					vector_threshold, vectors);
2280 
2281 	if (vectors < 0) {
2282 		dev_err(&adapter->pdev->dev,
2283 			"Unable to allocate MSI-X interrupts\n");
2284 		kfree(adapter->msix_entries);
2285 		adapter->msix_entries = NULL;
2286 		return vectors;
2287 	}
2288 
2289 	/* Adjust for only the vectors we'll use, which is minimum
2290 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2291 	 * vectors we were allocated.
2292 	 */
2293 	adapter->num_msix_vectors = vectors;
2294 
2295 	return 0;
2296 }
2297 
2298 /**
2299  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2300  * @adapter: board private structure to initialize
2301  *
2302  * This is the top level queue allocation routine.  The order here is very
2303  * important, starting with the "most" number of features turned on at once,
2304  * and ending with the smallest set of features.  This way large combinations
2305  * can be allocated if they're turned on, and smaller combinations are the
2306  * fallthrough conditions.
2307  *
2308  **/
2309 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2310 {
2311 	struct ixgbe_hw *hw = &adapter->hw;
2312 	unsigned int def_q = 0;
2313 	unsigned int num_tcs = 0;
2314 	int err;
2315 
2316 	/* Start with base case */
2317 	adapter->num_rx_queues = 1;
2318 	adapter->num_tx_queues = 1;
2319 
2320 	spin_lock_bh(&adapter->mbx_lock);
2321 
2322 	/* fetch queue configuration from the PF */
2323 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2324 
2325 	spin_unlock_bh(&adapter->mbx_lock);
2326 
2327 	if (err)
2328 		return;
2329 
2330 	/* we need as many queues as traffic classes */
2331 	if (num_tcs > 1) {
2332 		adapter->num_rx_queues = num_tcs;
2333 	} else {
2334 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2335 
2336 		switch (hw->api_version) {
2337 		case ixgbe_mbox_api_11:
2338 		case ixgbe_mbox_api_12:
2339 			adapter->num_rx_queues = rss;
2340 			adapter->num_tx_queues = rss;
2341 		default:
2342 			break;
2343 		}
2344 	}
2345 }
2346 
2347 /**
2348  * ixgbevf_alloc_queues - Allocate memory for all rings
2349  * @adapter: board private structure to initialize
2350  *
2351  * We allocate one ring per queue at run-time since we don't know the
2352  * number of queues at compile-time.  The polling_netdev array is
2353  * intended for Multiqueue, but should work fine with a single queue.
2354  **/
2355 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2356 {
2357 	struct ixgbevf_ring *ring;
2358 	int rx = 0, tx = 0;
2359 
2360 	for (; tx < adapter->num_tx_queues; tx++) {
2361 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2362 		if (!ring)
2363 			goto err_allocation;
2364 
2365 		ring->dev = &adapter->pdev->dev;
2366 		ring->netdev = adapter->netdev;
2367 		ring->count = adapter->tx_ring_count;
2368 		ring->queue_index = tx;
2369 		ring->reg_idx = tx;
2370 
2371 		adapter->tx_ring[tx] = ring;
2372 	}
2373 
2374 	for (; rx < adapter->num_rx_queues; rx++) {
2375 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2376 		if (!ring)
2377 			goto err_allocation;
2378 
2379 		ring->dev = &adapter->pdev->dev;
2380 		ring->netdev = adapter->netdev;
2381 
2382 		ring->count = adapter->rx_ring_count;
2383 		ring->queue_index = rx;
2384 		ring->reg_idx = rx;
2385 
2386 		adapter->rx_ring[rx] = ring;
2387 	}
2388 
2389 	return 0;
2390 
2391 err_allocation:
2392 	while (tx) {
2393 		kfree(adapter->tx_ring[--tx]);
2394 		adapter->tx_ring[tx] = NULL;
2395 	}
2396 
2397 	while (rx) {
2398 		kfree(adapter->rx_ring[--rx]);
2399 		adapter->rx_ring[rx] = NULL;
2400 	}
2401 	return -ENOMEM;
2402 }
2403 
2404 /**
2405  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2406  * @adapter: board private structure to initialize
2407  *
2408  * Attempt to configure the interrupts using the best available
2409  * capabilities of the hardware and the kernel.
2410  **/
2411 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2412 {
2413 	struct net_device *netdev = adapter->netdev;
2414 	int err = 0;
2415 	int vector, v_budget;
2416 
2417 	/* It's easy to be greedy for MSI-X vectors, but it really
2418 	 * doesn't do us much good if we have a lot more vectors
2419 	 * than CPU's.  So let's be conservative and only ask for
2420 	 * (roughly) the same number of vectors as there are CPU's.
2421 	 * The default is to use pairs of vectors.
2422 	 */
2423 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2424 	v_budget = min_t(int, v_budget, num_online_cpus());
2425 	v_budget += NON_Q_VECTORS;
2426 
2427 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2428 	 * mean we disable MSI-X capabilities of the adapter.
2429 	 */
2430 	adapter->msix_entries = kcalloc(v_budget,
2431 					sizeof(struct msix_entry), GFP_KERNEL);
2432 	if (!adapter->msix_entries) {
2433 		err = -ENOMEM;
2434 		goto out;
2435 	}
2436 
2437 	for (vector = 0; vector < v_budget; vector++)
2438 		adapter->msix_entries[vector].entry = vector;
2439 
2440 	err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2441 	if (err)
2442 		goto out;
2443 
2444 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2445 	if (err)
2446 		goto out;
2447 
2448 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2449 
2450 out:
2451 	return err;
2452 }
2453 
2454 /**
2455  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2456  * @adapter: board private structure to initialize
2457  *
2458  * We allocate one q_vector per queue interrupt.  If allocation fails we
2459  * return -ENOMEM.
2460  **/
2461 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2462 {
2463 	int q_idx, num_q_vectors;
2464 	struct ixgbevf_q_vector *q_vector;
2465 
2466 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2467 
2468 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2469 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2470 		if (!q_vector)
2471 			goto err_out;
2472 		q_vector->adapter = adapter;
2473 		q_vector->v_idx = q_idx;
2474 		netif_napi_add(adapter->netdev, &q_vector->napi,
2475 			       ixgbevf_poll, 64);
2476 #ifdef CONFIG_NET_RX_BUSY_POLL
2477 		napi_hash_add(&q_vector->napi);
2478 #endif
2479 		adapter->q_vector[q_idx] = q_vector;
2480 	}
2481 
2482 	return 0;
2483 
2484 err_out:
2485 	while (q_idx) {
2486 		q_idx--;
2487 		q_vector = adapter->q_vector[q_idx];
2488 #ifdef CONFIG_NET_RX_BUSY_POLL
2489 		napi_hash_del(&q_vector->napi);
2490 #endif
2491 		netif_napi_del(&q_vector->napi);
2492 		kfree(q_vector);
2493 		adapter->q_vector[q_idx] = NULL;
2494 	}
2495 	return -ENOMEM;
2496 }
2497 
2498 /**
2499  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2500  * @adapter: board private structure to initialize
2501  *
2502  * This function frees the memory allocated to the q_vectors.  In addition if
2503  * NAPI is enabled it will delete any references to the NAPI struct prior
2504  * to freeing the q_vector.
2505  **/
2506 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2507 {
2508 	int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2509 
2510 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2511 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2512 
2513 		adapter->q_vector[q_idx] = NULL;
2514 #ifdef CONFIG_NET_RX_BUSY_POLL
2515 		napi_hash_del(&q_vector->napi);
2516 #endif
2517 		netif_napi_del(&q_vector->napi);
2518 		kfree(q_vector);
2519 	}
2520 }
2521 
2522 /**
2523  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2524  * @adapter: board private structure
2525  *
2526  **/
2527 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2528 {
2529 	pci_disable_msix(adapter->pdev);
2530 	kfree(adapter->msix_entries);
2531 	adapter->msix_entries = NULL;
2532 }
2533 
2534 /**
2535  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2536  * @adapter: board private structure to initialize
2537  *
2538  **/
2539 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2540 {
2541 	int err;
2542 
2543 	/* Number of supported queues */
2544 	ixgbevf_set_num_queues(adapter);
2545 
2546 	err = ixgbevf_set_interrupt_capability(adapter);
2547 	if (err) {
2548 		hw_dbg(&adapter->hw,
2549 		       "Unable to setup interrupt capabilities\n");
2550 		goto err_set_interrupt;
2551 	}
2552 
2553 	err = ixgbevf_alloc_q_vectors(adapter);
2554 	if (err) {
2555 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2556 		goto err_alloc_q_vectors;
2557 	}
2558 
2559 	err = ixgbevf_alloc_queues(adapter);
2560 	if (err) {
2561 		pr_err("Unable to allocate memory for queues\n");
2562 		goto err_alloc_queues;
2563 	}
2564 
2565 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2566 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2567 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2568 
2569 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2570 
2571 	return 0;
2572 err_alloc_queues:
2573 	ixgbevf_free_q_vectors(adapter);
2574 err_alloc_q_vectors:
2575 	ixgbevf_reset_interrupt_capability(adapter);
2576 err_set_interrupt:
2577 	return err;
2578 }
2579 
2580 /**
2581  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2582  * @adapter: board private structure to clear interrupt scheme on
2583  *
2584  * We go through and clear interrupt specific resources and reset the structure
2585  * to pre-load conditions
2586  **/
2587 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2588 {
2589 	int i;
2590 
2591 	for (i = 0; i < adapter->num_tx_queues; i++) {
2592 		kfree(adapter->tx_ring[i]);
2593 		adapter->tx_ring[i] = NULL;
2594 	}
2595 	for (i = 0; i < adapter->num_rx_queues; i++) {
2596 		kfree(adapter->rx_ring[i]);
2597 		adapter->rx_ring[i] = NULL;
2598 	}
2599 
2600 	adapter->num_tx_queues = 0;
2601 	adapter->num_rx_queues = 0;
2602 
2603 	ixgbevf_free_q_vectors(adapter);
2604 	ixgbevf_reset_interrupt_capability(adapter);
2605 }
2606 
2607 /**
2608  * ixgbevf_sw_init - Initialize general software structures
2609  * @adapter: board private structure to initialize
2610  *
2611  * ixgbevf_sw_init initializes the Adapter private data structure.
2612  * Fields are initialized based on PCI device information and
2613  * OS network device settings (MTU size).
2614  **/
2615 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2616 {
2617 	struct ixgbe_hw *hw = &adapter->hw;
2618 	struct pci_dev *pdev = adapter->pdev;
2619 	struct net_device *netdev = adapter->netdev;
2620 	int err;
2621 
2622 	/* PCI config space info */
2623 	hw->vendor_id = pdev->vendor;
2624 	hw->device_id = pdev->device;
2625 	hw->revision_id = pdev->revision;
2626 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2627 	hw->subsystem_device_id = pdev->subsystem_device;
2628 
2629 	hw->mbx.ops.init_params(hw);
2630 
2631 	/* assume legacy case in which PF would only give VF 2 queues */
2632 	hw->mac.max_tx_queues = 2;
2633 	hw->mac.max_rx_queues = 2;
2634 
2635 	/* lock to protect mailbox accesses */
2636 	spin_lock_init(&adapter->mbx_lock);
2637 
2638 	err = hw->mac.ops.reset_hw(hw);
2639 	if (err) {
2640 		dev_info(&pdev->dev,
2641 			 "PF still in reset state.  Is the PF interface up?\n");
2642 	} else {
2643 		err = hw->mac.ops.init_hw(hw);
2644 		if (err) {
2645 			pr_err("init_shared_code failed: %d\n", err);
2646 			goto out;
2647 		}
2648 		ixgbevf_negotiate_api(adapter);
2649 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2650 		if (err)
2651 			dev_info(&pdev->dev, "Error reading MAC address\n");
2652 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2653 			dev_info(&pdev->dev,
2654 				 "MAC address not assigned by administrator.\n");
2655 		memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2656 	}
2657 
2658 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2659 		dev_info(&pdev->dev, "Assigning random MAC address\n");
2660 		eth_hw_addr_random(netdev);
2661 		memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2662 	}
2663 
2664 	/* Enable dynamic interrupt throttling rates */
2665 	adapter->rx_itr_setting = 1;
2666 	adapter->tx_itr_setting = 1;
2667 
2668 	/* set default ring sizes */
2669 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2670 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2671 
2672 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2673 	return 0;
2674 
2675 out:
2676 	return err;
2677 }
2678 
2679 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2680 	{							\
2681 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2682 		if (current_counter < last_counter)		\
2683 			counter += 0x100000000LL;		\
2684 		last_counter = current_counter;			\
2685 		counter &= 0xFFFFFFFF00000000LL;		\
2686 		counter |= current_counter;			\
2687 	}
2688 
2689 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2690 	{								 \
2691 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2692 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2693 		u64 current_counter = (current_counter_msb << 32) |	 \
2694 			current_counter_lsb;				 \
2695 		if (current_counter < last_counter)			 \
2696 			counter += 0x1000000000LL;			 \
2697 		last_counter = current_counter;				 \
2698 		counter &= 0xFFFFFFF000000000LL;			 \
2699 		counter |= current_counter;				 \
2700 	}
2701 /**
2702  * ixgbevf_update_stats - Update the board statistics counters.
2703  * @adapter: board private structure
2704  **/
2705 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2706 {
2707 	struct ixgbe_hw *hw = &adapter->hw;
2708 	int i;
2709 
2710 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2711 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2712 		return;
2713 
2714 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2715 				adapter->stats.vfgprc);
2716 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2717 				adapter->stats.vfgptc);
2718 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2719 				adapter->stats.last_vfgorc,
2720 				adapter->stats.vfgorc);
2721 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2722 				adapter->stats.last_vfgotc,
2723 				adapter->stats.vfgotc);
2724 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2725 				adapter->stats.vfmprc);
2726 
2727 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2728 		adapter->hw_csum_rx_error +=
2729 			adapter->rx_ring[i]->hw_csum_rx_error;
2730 		adapter->rx_ring[i]->hw_csum_rx_error = 0;
2731 	}
2732 }
2733 
2734 /**
2735  * ixgbevf_service_timer - Timer Call-back
2736  * @data: pointer to adapter cast into an unsigned long
2737  **/
2738 static void ixgbevf_service_timer(unsigned long data)
2739 {
2740 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2741 
2742 	/* Reset the timer */
2743 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2744 
2745 	ixgbevf_service_event_schedule(adapter);
2746 }
2747 
2748 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2749 {
2750 	if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2751 		return;
2752 
2753 	adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2754 
2755 	/* If we're already down or resetting, just bail */
2756 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2757 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2758 		return;
2759 
2760 	adapter->tx_timeout_count++;
2761 
2762 	ixgbevf_reinit_locked(adapter);
2763 }
2764 
2765 /**
2766  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2767  * @adapter: pointer to the device adapter structure
2768  *
2769  * This function serves two purposes.  First it strobes the interrupt lines
2770  * in order to make certain interrupts are occurring.  Secondly it sets the
2771  * bits needed to check for TX hangs.  As a result we should immediately
2772  * determine if a hang has occurred.
2773  **/
2774 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2775 {
2776 	struct ixgbe_hw *hw = &adapter->hw;
2777 	u32 eics = 0;
2778 	int i;
2779 
2780 	/* If we're down or resetting, just bail */
2781 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2782 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2783 		return;
2784 
2785 	/* Force detection of hung controller */
2786 	if (netif_carrier_ok(adapter->netdev)) {
2787 		for (i = 0; i < adapter->num_tx_queues; i++)
2788 			set_check_for_tx_hang(adapter->tx_ring[i]);
2789 	}
2790 
2791 	/* get one bit for every active Tx/Rx interrupt vector */
2792 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2793 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2794 
2795 		if (qv->rx.ring || qv->tx.ring)
2796 			eics |= 1 << i;
2797 	}
2798 
2799 	/* Cause software interrupt to ensure rings are cleaned */
2800 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2801 }
2802 
2803 /**
2804  * ixgbevf_watchdog_update_link - update the link status
2805  * @adapter: pointer to the device adapter structure
2806  **/
2807 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2808 {
2809 	struct ixgbe_hw *hw = &adapter->hw;
2810 	u32 link_speed = adapter->link_speed;
2811 	bool link_up = adapter->link_up;
2812 	s32 err;
2813 
2814 	spin_lock_bh(&adapter->mbx_lock);
2815 
2816 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2817 
2818 	spin_unlock_bh(&adapter->mbx_lock);
2819 
2820 	/* if check for link returns error we will need to reset */
2821 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2822 		adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2823 		link_up = false;
2824 	}
2825 
2826 	adapter->link_up = link_up;
2827 	adapter->link_speed = link_speed;
2828 }
2829 
2830 /**
2831  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2832  *				 print link up message
2833  * @adapter: pointer to the device adapter structure
2834  **/
2835 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2836 {
2837 	struct net_device *netdev = adapter->netdev;
2838 
2839 	/* only continue if link was previously down */
2840 	if (netif_carrier_ok(netdev))
2841 		return;
2842 
2843 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2844 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2845 		 "10 Gbps" :
2846 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2847 		 "1 Gbps" :
2848 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2849 		 "100 Mbps" :
2850 		 "unknown speed");
2851 
2852 	netif_carrier_on(netdev);
2853 }
2854 
2855 /**
2856  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2857  *				   print link down message
2858  * @adapter: pointer to the adapter structure
2859  **/
2860 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2861 {
2862 	struct net_device *netdev = adapter->netdev;
2863 
2864 	adapter->link_speed = 0;
2865 
2866 	/* only continue if link was up previously */
2867 	if (!netif_carrier_ok(netdev))
2868 		return;
2869 
2870 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2871 
2872 	netif_carrier_off(netdev);
2873 }
2874 
2875 /**
2876  * ixgbevf_watchdog_subtask - worker thread to bring link up
2877  * @work: pointer to work_struct containing our data
2878  **/
2879 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2880 {
2881 	/* if interface is down do nothing */
2882 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2883 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2884 		return;
2885 
2886 	ixgbevf_watchdog_update_link(adapter);
2887 
2888 	if (adapter->link_up)
2889 		ixgbevf_watchdog_link_is_up(adapter);
2890 	else
2891 		ixgbevf_watchdog_link_is_down(adapter);
2892 
2893 	ixgbevf_update_stats(adapter);
2894 }
2895 
2896 /**
2897  * ixgbevf_service_task - manages and runs subtasks
2898  * @work: pointer to work_struct containing our data
2899  **/
2900 static void ixgbevf_service_task(struct work_struct *work)
2901 {
2902 	struct ixgbevf_adapter *adapter = container_of(work,
2903 						       struct ixgbevf_adapter,
2904 						       service_task);
2905 	struct ixgbe_hw *hw = &adapter->hw;
2906 
2907 	if (IXGBE_REMOVED(hw->hw_addr)) {
2908 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2909 			rtnl_lock();
2910 			ixgbevf_down(adapter);
2911 			rtnl_unlock();
2912 		}
2913 		return;
2914 	}
2915 
2916 	ixgbevf_queue_reset_subtask(adapter);
2917 	ixgbevf_reset_subtask(adapter);
2918 	ixgbevf_watchdog_subtask(adapter);
2919 	ixgbevf_check_hang_subtask(adapter);
2920 
2921 	ixgbevf_service_event_complete(adapter);
2922 }
2923 
2924 /**
2925  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2926  * @tx_ring: Tx descriptor ring for a specific queue
2927  *
2928  * Free all transmit software resources
2929  **/
2930 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2931 {
2932 	ixgbevf_clean_tx_ring(tx_ring);
2933 
2934 	vfree(tx_ring->tx_buffer_info);
2935 	tx_ring->tx_buffer_info = NULL;
2936 
2937 	/* if not set, then don't free */
2938 	if (!tx_ring->desc)
2939 		return;
2940 
2941 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2942 			  tx_ring->dma);
2943 
2944 	tx_ring->desc = NULL;
2945 }
2946 
2947 /**
2948  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2949  * @adapter: board private structure
2950  *
2951  * Free all transmit software resources
2952  **/
2953 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2954 {
2955 	int i;
2956 
2957 	for (i = 0; i < adapter->num_tx_queues; i++)
2958 		if (adapter->tx_ring[i]->desc)
2959 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2960 }
2961 
2962 /**
2963  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2964  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2965  *
2966  * Return 0 on success, negative on failure
2967  **/
2968 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2969 {
2970 	int size;
2971 
2972 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2973 	tx_ring->tx_buffer_info = vzalloc(size);
2974 	if (!tx_ring->tx_buffer_info)
2975 		goto err;
2976 
2977 	/* round up to nearest 4K */
2978 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2979 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2980 
2981 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2982 					   &tx_ring->dma, GFP_KERNEL);
2983 	if (!tx_ring->desc)
2984 		goto err;
2985 
2986 	return 0;
2987 
2988 err:
2989 	vfree(tx_ring->tx_buffer_info);
2990 	tx_ring->tx_buffer_info = NULL;
2991 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
2992 	return -ENOMEM;
2993 }
2994 
2995 /**
2996  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2997  * @adapter: board private structure
2998  *
2999  * If this function returns with an error, then it's possible one or
3000  * more of the rings is populated (while the rest are not).  It is the
3001  * callers duty to clean those orphaned rings.
3002  *
3003  * Return 0 on success, negative on failure
3004  **/
3005 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3006 {
3007 	int i, err = 0;
3008 
3009 	for (i = 0; i < adapter->num_tx_queues; i++) {
3010 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3011 		if (!err)
3012 			continue;
3013 		hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3014 		break;
3015 	}
3016 
3017 	return err;
3018 }
3019 
3020 /**
3021  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3022  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3023  *
3024  * Returns 0 on success, negative on failure
3025  **/
3026 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3027 {
3028 	int size;
3029 
3030 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3031 	rx_ring->rx_buffer_info = vzalloc(size);
3032 	if (!rx_ring->rx_buffer_info)
3033 		goto err;
3034 
3035 	/* Round up to nearest 4K */
3036 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3037 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3038 
3039 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3040 					   &rx_ring->dma, GFP_KERNEL);
3041 
3042 	if (!rx_ring->desc)
3043 		goto err;
3044 
3045 	return 0;
3046 err:
3047 	vfree(rx_ring->rx_buffer_info);
3048 	rx_ring->rx_buffer_info = NULL;
3049 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3050 	return -ENOMEM;
3051 }
3052 
3053 /**
3054  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3055  * @adapter: board private structure
3056  *
3057  * If this function returns with an error, then it's possible one or
3058  * more of the rings is populated (while the rest are not).  It is the
3059  * callers duty to clean those orphaned rings.
3060  *
3061  * Return 0 on success, negative on failure
3062  **/
3063 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3064 {
3065 	int i, err = 0;
3066 
3067 	for (i = 0; i < adapter->num_rx_queues; i++) {
3068 		err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3069 		if (!err)
3070 			continue;
3071 		hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3072 		break;
3073 	}
3074 	return err;
3075 }
3076 
3077 /**
3078  * ixgbevf_free_rx_resources - Free Rx Resources
3079  * @rx_ring: ring to clean the resources from
3080  *
3081  * Free all receive software resources
3082  **/
3083 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3084 {
3085 	ixgbevf_clean_rx_ring(rx_ring);
3086 
3087 	vfree(rx_ring->rx_buffer_info);
3088 	rx_ring->rx_buffer_info = NULL;
3089 
3090 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3091 			  rx_ring->dma);
3092 
3093 	rx_ring->desc = NULL;
3094 }
3095 
3096 /**
3097  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3098  * @adapter: board private structure
3099  *
3100  * Free all receive software resources
3101  **/
3102 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3103 {
3104 	int i;
3105 
3106 	for (i = 0; i < adapter->num_rx_queues; i++)
3107 		if (adapter->rx_ring[i]->desc)
3108 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3109 }
3110 
3111 /**
3112  * ixgbevf_open - Called when a network interface is made active
3113  * @netdev: network interface device structure
3114  *
3115  * Returns 0 on success, negative value on failure
3116  *
3117  * The open entry point is called when a network interface is made
3118  * active by the system (IFF_UP).  At this point all resources needed
3119  * for transmit and receive operations are allocated, the interrupt
3120  * handler is registered with the OS, the watchdog timer is started,
3121  * and the stack is notified that the interface is ready.
3122  **/
3123 static int ixgbevf_open(struct net_device *netdev)
3124 {
3125 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3126 	struct ixgbe_hw *hw = &adapter->hw;
3127 	int err;
3128 
3129 	/* A previous failure to open the device because of a lack of
3130 	 * available MSIX vector resources may have reset the number
3131 	 * of msix vectors variable to zero.  The only way to recover
3132 	 * is to unload/reload the driver and hope that the system has
3133 	 * been able to recover some MSIX vector resources.
3134 	 */
3135 	if (!adapter->num_msix_vectors)
3136 		return -ENOMEM;
3137 
3138 	if (hw->adapter_stopped) {
3139 		ixgbevf_reset(adapter);
3140 		/* if adapter is still stopped then PF isn't up and
3141 		 * the VF can't start.
3142 		 */
3143 		if (hw->adapter_stopped) {
3144 			err = IXGBE_ERR_MBX;
3145 			pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3146 			goto err_setup_reset;
3147 		}
3148 	}
3149 
3150 	/* disallow open during test */
3151 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3152 		return -EBUSY;
3153 
3154 	netif_carrier_off(netdev);
3155 
3156 	/* allocate transmit descriptors */
3157 	err = ixgbevf_setup_all_tx_resources(adapter);
3158 	if (err)
3159 		goto err_setup_tx;
3160 
3161 	/* allocate receive descriptors */
3162 	err = ixgbevf_setup_all_rx_resources(adapter);
3163 	if (err)
3164 		goto err_setup_rx;
3165 
3166 	ixgbevf_configure(adapter);
3167 
3168 	/* Map the Tx/Rx rings to the vectors we were allotted.
3169 	 * if request_irq will be called in this function map_rings
3170 	 * must be called *before* up_complete
3171 	 */
3172 	ixgbevf_map_rings_to_vectors(adapter);
3173 
3174 	err = ixgbevf_request_irq(adapter);
3175 	if (err)
3176 		goto err_req_irq;
3177 
3178 	ixgbevf_up_complete(adapter);
3179 
3180 	return 0;
3181 
3182 err_req_irq:
3183 	ixgbevf_down(adapter);
3184 err_setup_rx:
3185 	ixgbevf_free_all_rx_resources(adapter);
3186 err_setup_tx:
3187 	ixgbevf_free_all_tx_resources(adapter);
3188 	ixgbevf_reset(adapter);
3189 
3190 err_setup_reset:
3191 
3192 	return err;
3193 }
3194 
3195 /**
3196  * ixgbevf_close - Disables a network interface
3197  * @netdev: network interface device structure
3198  *
3199  * Returns 0, this is not allowed to fail
3200  *
3201  * The close entry point is called when an interface is de-activated
3202  * by the OS.  The hardware is still under the drivers control, but
3203  * needs to be disabled.  A global MAC reset is issued to stop the
3204  * hardware, and all transmit and receive resources are freed.
3205  **/
3206 static int ixgbevf_close(struct net_device *netdev)
3207 {
3208 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3209 
3210 	ixgbevf_down(adapter);
3211 	ixgbevf_free_irq(adapter);
3212 
3213 	ixgbevf_free_all_tx_resources(adapter);
3214 	ixgbevf_free_all_rx_resources(adapter);
3215 
3216 	return 0;
3217 }
3218 
3219 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3220 {
3221 	struct net_device *dev = adapter->netdev;
3222 
3223 	if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3224 		return;
3225 
3226 	adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3227 
3228 	/* if interface is down do nothing */
3229 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3230 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3231 		return;
3232 
3233 	/* Hardware has to reinitialize queues and interrupts to
3234 	 * match packet buffer alignment. Unfortunately, the
3235 	 * hardware is not flexible enough to do this dynamically.
3236 	 */
3237 	if (netif_running(dev))
3238 		ixgbevf_close(dev);
3239 
3240 	ixgbevf_clear_interrupt_scheme(adapter);
3241 	ixgbevf_init_interrupt_scheme(adapter);
3242 
3243 	if (netif_running(dev))
3244 		ixgbevf_open(dev);
3245 }
3246 
3247 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3248 				u32 vlan_macip_lens, u32 type_tucmd,
3249 				u32 mss_l4len_idx)
3250 {
3251 	struct ixgbe_adv_tx_context_desc *context_desc;
3252 	u16 i = tx_ring->next_to_use;
3253 
3254 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3255 
3256 	i++;
3257 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3258 
3259 	/* set bits to identify this as an advanced context descriptor */
3260 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3261 
3262 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3263 	context_desc->seqnum_seed	= 0;
3264 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3265 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3266 }
3267 
3268 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3269 		       struct ixgbevf_tx_buffer *first,
3270 		       u8 *hdr_len)
3271 {
3272 	struct sk_buff *skb = first->skb;
3273 	u32 vlan_macip_lens, type_tucmd;
3274 	u32 mss_l4len_idx, l4len;
3275 	int err;
3276 
3277 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3278 		return 0;
3279 
3280 	if (!skb_is_gso(skb))
3281 		return 0;
3282 
3283 	err = skb_cow_head(skb, 0);
3284 	if (err < 0)
3285 		return err;
3286 
3287 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3288 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3289 
3290 	if (first->protocol == htons(ETH_P_IP)) {
3291 		struct iphdr *iph = ip_hdr(skb);
3292 
3293 		iph->tot_len = 0;
3294 		iph->check = 0;
3295 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3296 							 iph->daddr, 0,
3297 							 IPPROTO_TCP,
3298 							 0);
3299 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3300 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3301 				   IXGBE_TX_FLAGS_CSUM |
3302 				   IXGBE_TX_FLAGS_IPV4;
3303 	} else if (skb_is_gso_v6(skb)) {
3304 		ipv6_hdr(skb)->payload_len = 0;
3305 		tcp_hdr(skb)->check =
3306 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3307 				     &ipv6_hdr(skb)->daddr,
3308 				     0, IPPROTO_TCP, 0);
3309 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3310 				   IXGBE_TX_FLAGS_CSUM;
3311 	}
3312 
3313 	/* compute header lengths */
3314 	l4len = tcp_hdrlen(skb);
3315 	*hdr_len += l4len;
3316 	*hdr_len = skb_transport_offset(skb) + l4len;
3317 
3318 	/* update GSO size and bytecount with header size */
3319 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3320 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3321 
3322 	/* mss_l4len_id: use 1 as index for TSO */
3323 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3324 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3325 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3326 
3327 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3328 	vlan_macip_lens = skb_network_header_len(skb);
3329 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3330 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3331 
3332 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3333 			    type_tucmd, mss_l4len_idx);
3334 
3335 	return 1;
3336 }
3337 
3338 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3339 			    struct ixgbevf_tx_buffer *first)
3340 {
3341 	struct sk_buff *skb = first->skb;
3342 	u32 vlan_macip_lens = 0;
3343 	u32 mss_l4len_idx = 0;
3344 	u32 type_tucmd = 0;
3345 
3346 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3347 		u8 l4_hdr = 0;
3348 
3349 		switch (first->protocol) {
3350 		case htons(ETH_P_IP):
3351 			vlan_macip_lens |= skb_network_header_len(skb);
3352 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3353 			l4_hdr = ip_hdr(skb)->protocol;
3354 			break;
3355 		case htons(ETH_P_IPV6):
3356 			vlan_macip_lens |= skb_network_header_len(skb);
3357 			l4_hdr = ipv6_hdr(skb)->nexthdr;
3358 			break;
3359 		default:
3360 			if (unlikely(net_ratelimit())) {
3361 				dev_warn(tx_ring->dev,
3362 					 "partial checksum but proto=%x!\n",
3363 					 first->protocol);
3364 			}
3365 			break;
3366 		}
3367 
3368 		switch (l4_hdr) {
3369 		case IPPROTO_TCP:
3370 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3371 			mss_l4len_idx = tcp_hdrlen(skb) <<
3372 					IXGBE_ADVTXD_L4LEN_SHIFT;
3373 			break;
3374 		case IPPROTO_SCTP:
3375 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3376 			mss_l4len_idx = sizeof(struct sctphdr) <<
3377 					IXGBE_ADVTXD_L4LEN_SHIFT;
3378 			break;
3379 		case IPPROTO_UDP:
3380 			mss_l4len_idx = sizeof(struct udphdr) <<
3381 					IXGBE_ADVTXD_L4LEN_SHIFT;
3382 			break;
3383 		default:
3384 			if (unlikely(net_ratelimit())) {
3385 				dev_warn(tx_ring->dev,
3386 					 "partial checksum but l4 proto=%x!\n",
3387 					 l4_hdr);
3388 			}
3389 			break;
3390 		}
3391 
3392 		/* update TX checksum flag */
3393 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3394 	}
3395 
3396 	/* vlan_macip_lens: MACLEN, VLAN tag */
3397 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3398 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3399 
3400 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3401 			    type_tucmd, mss_l4len_idx);
3402 }
3403 
3404 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3405 {
3406 	/* set type for advanced descriptor with frame checksum insertion */
3407 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3408 				      IXGBE_ADVTXD_DCMD_IFCS |
3409 				      IXGBE_ADVTXD_DCMD_DEXT);
3410 
3411 	/* set HW VLAN bit if VLAN is present */
3412 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3413 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3414 
3415 	/* set segmentation enable bits for TSO/FSO */
3416 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3417 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3418 
3419 	return cmd_type;
3420 }
3421 
3422 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3423 				     u32 tx_flags, unsigned int paylen)
3424 {
3425 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3426 
3427 	/* enable L4 checksum for TSO and TX checksum offload */
3428 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3429 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3430 
3431 	/* enble IPv4 checksum for TSO */
3432 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3433 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3434 
3435 	/* use index 1 context for TSO/FSO/FCOE */
3436 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3437 		olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3438 
3439 	/* Check Context must be set if Tx switch is enabled, which it
3440 	 * always is for case where virtual functions are running
3441 	 */
3442 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3443 
3444 	tx_desc->read.olinfo_status = olinfo_status;
3445 }
3446 
3447 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3448 			   struct ixgbevf_tx_buffer *first,
3449 			   const u8 hdr_len)
3450 {
3451 	dma_addr_t dma;
3452 	struct sk_buff *skb = first->skb;
3453 	struct ixgbevf_tx_buffer *tx_buffer;
3454 	union ixgbe_adv_tx_desc *tx_desc;
3455 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3456 	unsigned int data_len = skb->data_len;
3457 	unsigned int size = skb_headlen(skb);
3458 	unsigned int paylen = skb->len - hdr_len;
3459 	u32 tx_flags = first->tx_flags;
3460 	__le32 cmd_type;
3461 	u16 i = tx_ring->next_to_use;
3462 
3463 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3464 
3465 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3466 	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3467 
3468 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3469 	if (dma_mapping_error(tx_ring->dev, dma))
3470 		goto dma_error;
3471 
3472 	/* record length, and DMA address */
3473 	dma_unmap_len_set(first, len, size);
3474 	dma_unmap_addr_set(first, dma, dma);
3475 
3476 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
3477 
3478 	for (;;) {
3479 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3480 			tx_desc->read.cmd_type_len =
3481 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3482 
3483 			i++;
3484 			tx_desc++;
3485 			if (i == tx_ring->count) {
3486 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3487 				i = 0;
3488 			}
3489 
3490 			dma += IXGBE_MAX_DATA_PER_TXD;
3491 			size -= IXGBE_MAX_DATA_PER_TXD;
3492 
3493 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3494 			tx_desc->read.olinfo_status = 0;
3495 		}
3496 
3497 		if (likely(!data_len))
3498 			break;
3499 
3500 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3501 
3502 		i++;
3503 		tx_desc++;
3504 		if (i == tx_ring->count) {
3505 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3506 			i = 0;
3507 		}
3508 
3509 		size = skb_frag_size(frag);
3510 		data_len -= size;
3511 
3512 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3513 				       DMA_TO_DEVICE);
3514 		if (dma_mapping_error(tx_ring->dev, dma))
3515 			goto dma_error;
3516 
3517 		tx_buffer = &tx_ring->tx_buffer_info[i];
3518 		dma_unmap_len_set(tx_buffer, len, size);
3519 		dma_unmap_addr_set(tx_buffer, dma, dma);
3520 
3521 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3522 		tx_desc->read.olinfo_status = 0;
3523 
3524 		frag++;
3525 	}
3526 
3527 	/* write last descriptor with RS and EOP bits */
3528 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3529 	tx_desc->read.cmd_type_len = cmd_type;
3530 
3531 	/* set the timestamp */
3532 	first->time_stamp = jiffies;
3533 
3534 	/* Force memory writes to complete before letting h/w know there
3535 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
3536 	 * memory model archs, such as IA-64).
3537 	 *
3538 	 * We also need this memory barrier (wmb) to make certain all of the
3539 	 * status bits have been updated before next_to_watch is written.
3540 	 */
3541 	wmb();
3542 
3543 	/* set next_to_watch value indicating a packet is present */
3544 	first->next_to_watch = tx_desc;
3545 
3546 	i++;
3547 	if (i == tx_ring->count)
3548 		i = 0;
3549 
3550 	tx_ring->next_to_use = i;
3551 
3552 	/* notify HW of packet */
3553 	ixgbevf_write_tail(tx_ring, i);
3554 
3555 	return;
3556 dma_error:
3557 	dev_err(tx_ring->dev, "TX DMA map failed\n");
3558 
3559 	/* clear dma mappings for failed tx_buffer_info map */
3560 	for (;;) {
3561 		tx_buffer = &tx_ring->tx_buffer_info[i];
3562 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3563 		if (tx_buffer == first)
3564 			break;
3565 		if (i == 0)
3566 			i = tx_ring->count;
3567 		i--;
3568 	}
3569 
3570 	tx_ring->next_to_use = i;
3571 }
3572 
3573 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3574 {
3575 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3576 	/* Herbert's original patch had:
3577 	 *  smp_mb__after_netif_stop_queue();
3578 	 * but since that doesn't exist yet, just open code it.
3579 	 */
3580 	smp_mb();
3581 
3582 	/* We need to check again in a case another CPU has just
3583 	 * made room available.
3584 	 */
3585 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
3586 		return -EBUSY;
3587 
3588 	/* A reprieve! - use start_queue because it doesn't call schedule */
3589 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3590 	++tx_ring->tx_stats.restart_queue;
3591 
3592 	return 0;
3593 }
3594 
3595 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3596 {
3597 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3598 		return 0;
3599 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
3600 }
3601 
3602 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3603 {
3604 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3605 	struct ixgbevf_tx_buffer *first;
3606 	struct ixgbevf_ring *tx_ring;
3607 	int tso;
3608 	u32 tx_flags = 0;
3609 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
3610 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3611 	unsigned short f;
3612 #endif
3613 	u8 hdr_len = 0;
3614 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3615 
3616 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3617 		dev_kfree_skb_any(skb);
3618 		return NETDEV_TX_OK;
3619 	}
3620 
3621 	tx_ring = adapter->tx_ring[skb->queue_mapping];
3622 
3623 	/* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3624 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3625 	 *       + 2 desc gap to keep tail from touching head,
3626 	 *       + 1 desc for context descriptor,
3627 	 * otherwise try next time
3628 	 */
3629 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3630 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3631 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3632 #else
3633 	count += skb_shinfo(skb)->nr_frags;
3634 #endif
3635 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3636 		tx_ring->tx_stats.tx_busy++;
3637 		return NETDEV_TX_BUSY;
3638 	}
3639 
3640 	/* record the location of the first descriptor for this packet */
3641 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3642 	first->skb = skb;
3643 	first->bytecount = skb->len;
3644 	first->gso_segs = 1;
3645 
3646 	if (skb_vlan_tag_present(skb)) {
3647 		tx_flags |= skb_vlan_tag_get(skb);
3648 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3649 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3650 	}
3651 
3652 	/* record initial flags and protocol */
3653 	first->tx_flags = tx_flags;
3654 	first->protocol = vlan_get_protocol(skb);
3655 
3656 	tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3657 	if (tso < 0)
3658 		goto out_drop;
3659 	else if (!tso)
3660 		ixgbevf_tx_csum(tx_ring, first);
3661 
3662 	ixgbevf_tx_map(tx_ring, first, hdr_len);
3663 
3664 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3665 
3666 	return NETDEV_TX_OK;
3667 
3668 out_drop:
3669 	dev_kfree_skb_any(first->skb);
3670 	first->skb = NULL;
3671 
3672 	return NETDEV_TX_OK;
3673 }
3674 
3675 /**
3676  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3677  * @netdev: network interface device structure
3678  * @p: pointer to an address structure
3679  *
3680  * Returns 0 on success, negative on failure
3681  **/
3682 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3683 {
3684 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3685 	struct ixgbe_hw *hw = &adapter->hw;
3686 	struct sockaddr *addr = p;
3687 
3688 	if (!is_valid_ether_addr(addr->sa_data))
3689 		return -EADDRNOTAVAIL;
3690 
3691 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3692 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3693 
3694 	spin_lock_bh(&adapter->mbx_lock);
3695 
3696 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3697 
3698 	spin_unlock_bh(&adapter->mbx_lock);
3699 
3700 	return 0;
3701 }
3702 
3703 /**
3704  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3705  * @netdev: network interface device structure
3706  * @new_mtu: new value for maximum frame size
3707  *
3708  * Returns 0 on success, negative on failure
3709  **/
3710 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3711 {
3712 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3713 	struct ixgbe_hw *hw = &adapter->hw;
3714 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3715 	int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3716 
3717 	switch (adapter->hw.api_version) {
3718 	case ixgbe_mbox_api_11:
3719 	case ixgbe_mbox_api_12:
3720 		max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3721 		break;
3722 	default:
3723 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3724 			max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3725 		break;
3726 	}
3727 
3728 	/* MTU < 68 is an error and causes problems on some kernels */
3729 	if ((new_mtu < 68) || (max_frame > max_possible_frame))
3730 		return -EINVAL;
3731 
3732 	hw_dbg(hw, "changing MTU from %d to %d\n",
3733 	       netdev->mtu, new_mtu);
3734 	/* must set new MTU before calling down or up */
3735 	netdev->mtu = new_mtu;
3736 
3737 	/* notify the PF of our intent to use this size of frame */
3738 	ixgbevf_rlpml_set_vf(hw, max_frame);
3739 
3740 	return 0;
3741 }
3742 
3743 #ifdef CONFIG_NET_POLL_CONTROLLER
3744 /* Polling 'interrupt' - used by things like netconsole to send skbs
3745  * without having to re-enable interrupts. It's not called while
3746  * the interrupt routine is executing.
3747  */
3748 static void ixgbevf_netpoll(struct net_device *netdev)
3749 {
3750 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3751 	int i;
3752 
3753 	/* if interface is down do nothing */
3754 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3755 		return;
3756 	for (i = 0; i < adapter->num_rx_queues; i++)
3757 		ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3758 }
3759 #endif /* CONFIG_NET_POLL_CONTROLLER */
3760 
3761 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3762 {
3763 	struct net_device *netdev = pci_get_drvdata(pdev);
3764 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3765 #ifdef CONFIG_PM
3766 	int retval = 0;
3767 #endif
3768 
3769 	netif_device_detach(netdev);
3770 
3771 	if (netif_running(netdev)) {
3772 		rtnl_lock();
3773 		ixgbevf_down(adapter);
3774 		ixgbevf_free_irq(adapter);
3775 		ixgbevf_free_all_tx_resources(adapter);
3776 		ixgbevf_free_all_rx_resources(adapter);
3777 		rtnl_unlock();
3778 	}
3779 
3780 	ixgbevf_clear_interrupt_scheme(adapter);
3781 
3782 #ifdef CONFIG_PM
3783 	retval = pci_save_state(pdev);
3784 	if (retval)
3785 		return retval;
3786 
3787 #endif
3788 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3789 		pci_disable_device(pdev);
3790 
3791 	return 0;
3792 }
3793 
3794 #ifdef CONFIG_PM
3795 static int ixgbevf_resume(struct pci_dev *pdev)
3796 {
3797 	struct net_device *netdev = pci_get_drvdata(pdev);
3798 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3799 	u32 err;
3800 
3801 	pci_restore_state(pdev);
3802 	/* pci_restore_state clears dev->state_saved so call
3803 	 * pci_save_state to restore it.
3804 	 */
3805 	pci_save_state(pdev);
3806 
3807 	err = pci_enable_device_mem(pdev);
3808 	if (err) {
3809 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3810 		return err;
3811 	}
3812 	smp_mb__before_atomic();
3813 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3814 	pci_set_master(pdev);
3815 
3816 	ixgbevf_reset(adapter);
3817 
3818 	rtnl_lock();
3819 	err = ixgbevf_init_interrupt_scheme(adapter);
3820 	rtnl_unlock();
3821 	if (err) {
3822 		dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3823 		return err;
3824 	}
3825 
3826 	if (netif_running(netdev)) {
3827 		err = ixgbevf_open(netdev);
3828 		if (err)
3829 			return err;
3830 	}
3831 
3832 	netif_device_attach(netdev);
3833 
3834 	return err;
3835 }
3836 
3837 #endif /* CONFIG_PM */
3838 static void ixgbevf_shutdown(struct pci_dev *pdev)
3839 {
3840 	ixgbevf_suspend(pdev, PMSG_SUSPEND);
3841 }
3842 
3843 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3844 						struct rtnl_link_stats64 *stats)
3845 {
3846 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3847 	unsigned int start;
3848 	u64 bytes, packets;
3849 	const struct ixgbevf_ring *ring;
3850 	int i;
3851 
3852 	ixgbevf_update_stats(adapter);
3853 
3854 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3855 
3856 	for (i = 0; i < adapter->num_rx_queues; i++) {
3857 		ring = adapter->rx_ring[i];
3858 		do {
3859 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3860 			bytes = ring->stats.bytes;
3861 			packets = ring->stats.packets;
3862 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3863 		stats->rx_bytes += bytes;
3864 		stats->rx_packets += packets;
3865 	}
3866 
3867 	for (i = 0; i < adapter->num_tx_queues; i++) {
3868 		ring = adapter->tx_ring[i];
3869 		do {
3870 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3871 			bytes = ring->stats.bytes;
3872 			packets = ring->stats.packets;
3873 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3874 		stats->tx_bytes += bytes;
3875 		stats->tx_packets += packets;
3876 	}
3877 
3878 	return stats;
3879 }
3880 
3881 static const struct net_device_ops ixgbevf_netdev_ops = {
3882 	.ndo_open		= ixgbevf_open,
3883 	.ndo_stop		= ixgbevf_close,
3884 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3885 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3886 	.ndo_get_stats64	= ixgbevf_get_stats,
3887 	.ndo_validate_addr	= eth_validate_addr,
3888 	.ndo_set_mac_address	= ixgbevf_set_mac,
3889 	.ndo_change_mtu		= ixgbevf_change_mtu,
3890 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3891 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3892 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3893 #ifdef CONFIG_NET_RX_BUSY_POLL
3894 	.ndo_busy_poll		= ixgbevf_busy_poll_recv,
3895 #endif
3896 #ifdef CONFIG_NET_POLL_CONTROLLER
3897 	.ndo_poll_controller	= ixgbevf_netpoll,
3898 #endif
3899 };
3900 
3901 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3902 {
3903 	dev->netdev_ops = &ixgbevf_netdev_ops;
3904 	ixgbevf_set_ethtool_ops(dev);
3905 	dev->watchdog_timeo = 5 * HZ;
3906 }
3907 
3908 /**
3909  * ixgbevf_probe - Device Initialization Routine
3910  * @pdev: PCI device information struct
3911  * @ent: entry in ixgbevf_pci_tbl
3912  *
3913  * Returns 0 on success, negative on failure
3914  *
3915  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3916  * The OS initialization, configuring of the adapter private structure,
3917  * and a hardware reset occur.
3918  **/
3919 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3920 {
3921 	struct net_device *netdev;
3922 	struct ixgbevf_adapter *adapter = NULL;
3923 	struct ixgbe_hw *hw = NULL;
3924 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3925 	int err, pci_using_dac;
3926 	bool disable_dev = false;
3927 
3928 	err = pci_enable_device(pdev);
3929 	if (err)
3930 		return err;
3931 
3932 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3933 		pci_using_dac = 1;
3934 	} else {
3935 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3936 		if (err) {
3937 			dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3938 			goto err_dma;
3939 		}
3940 		pci_using_dac = 0;
3941 	}
3942 
3943 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3944 	if (err) {
3945 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3946 		goto err_pci_reg;
3947 	}
3948 
3949 	pci_set_master(pdev);
3950 
3951 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3952 				   MAX_TX_QUEUES);
3953 	if (!netdev) {
3954 		err = -ENOMEM;
3955 		goto err_alloc_etherdev;
3956 	}
3957 
3958 	SET_NETDEV_DEV(netdev, &pdev->dev);
3959 
3960 	adapter = netdev_priv(netdev);
3961 
3962 	adapter->netdev = netdev;
3963 	adapter->pdev = pdev;
3964 	hw = &adapter->hw;
3965 	hw->back = adapter;
3966 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3967 
3968 	/* call save state here in standalone driver because it relies on
3969 	 * adapter struct to exist, and needs to call netdev_priv
3970 	 */
3971 	pci_save_state(pdev);
3972 
3973 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3974 			      pci_resource_len(pdev, 0));
3975 	adapter->io_addr = hw->hw_addr;
3976 	if (!hw->hw_addr) {
3977 		err = -EIO;
3978 		goto err_ioremap;
3979 	}
3980 
3981 	ixgbevf_assign_netdev_ops(netdev);
3982 
3983 	/* Setup HW API */
3984 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3985 	hw->mac.type  = ii->mac;
3986 
3987 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3988 	       sizeof(struct ixgbe_mbx_operations));
3989 
3990 	/* setup the private structure */
3991 	err = ixgbevf_sw_init(adapter);
3992 	if (err)
3993 		goto err_sw_init;
3994 
3995 	/* The HW MAC address was set and/or determined in sw_init */
3996 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3997 		pr_err("invalid MAC address\n");
3998 		err = -EIO;
3999 		goto err_sw_init;
4000 	}
4001 
4002 	netdev->hw_features = NETIF_F_SG |
4003 			      NETIF_F_IP_CSUM |
4004 			      NETIF_F_IPV6_CSUM |
4005 			      NETIF_F_TSO |
4006 			      NETIF_F_TSO6 |
4007 			      NETIF_F_RXCSUM;
4008 
4009 	netdev->features = netdev->hw_features |
4010 			   NETIF_F_HW_VLAN_CTAG_TX |
4011 			   NETIF_F_HW_VLAN_CTAG_RX |
4012 			   NETIF_F_HW_VLAN_CTAG_FILTER;
4013 
4014 	netdev->vlan_features |= NETIF_F_TSO |
4015 				 NETIF_F_TSO6 |
4016 				 NETIF_F_IP_CSUM |
4017 				 NETIF_F_IPV6_CSUM |
4018 				 NETIF_F_SG;
4019 
4020 	if (pci_using_dac)
4021 		netdev->features |= NETIF_F_HIGHDMA;
4022 
4023 	netdev->priv_flags |= IFF_UNICAST_FLT;
4024 
4025 	if (IXGBE_REMOVED(hw->hw_addr)) {
4026 		err = -EIO;
4027 		goto err_sw_init;
4028 	}
4029 
4030 	setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4031 		    (unsigned long)adapter);
4032 
4033 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4034 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4035 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4036 
4037 	err = ixgbevf_init_interrupt_scheme(adapter);
4038 	if (err)
4039 		goto err_sw_init;
4040 
4041 	strcpy(netdev->name, "eth%d");
4042 
4043 	err = register_netdev(netdev);
4044 	if (err)
4045 		goto err_register;
4046 
4047 	pci_set_drvdata(pdev, netdev);
4048 	netif_carrier_off(netdev);
4049 
4050 	ixgbevf_init_last_counter_stats(adapter);
4051 
4052 	/* print the VF info */
4053 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4054 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4055 
4056 	switch (hw->mac.type) {
4057 	case ixgbe_mac_X550_vf:
4058 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4059 		break;
4060 	case ixgbe_mac_X540_vf:
4061 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4062 		break;
4063 	case ixgbe_mac_82599_vf:
4064 	default:
4065 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4066 		break;
4067 	}
4068 
4069 	return 0;
4070 
4071 err_register:
4072 	ixgbevf_clear_interrupt_scheme(adapter);
4073 err_sw_init:
4074 	ixgbevf_reset_interrupt_capability(adapter);
4075 	iounmap(adapter->io_addr);
4076 err_ioremap:
4077 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4078 	free_netdev(netdev);
4079 err_alloc_etherdev:
4080 	pci_release_regions(pdev);
4081 err_pci_reg:
4082 err_dma:
4083 	if (!adapter || disable_dev)
4084 		pci_disable_device(pdev);
4085 	return err;
4086 }
4087 
4088 /**
4089  * ixgbevf_remove - Device Removal Routine
4090  * @pdev: PCI device information struct
4091  *
4092  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4093  * that it should release a PCI device.  The could be caused by a
4094  * Hot-Plug event, or because the driver is going to be removed from
4095  * memory.
4096  **/
4097 static void ixgbevf_remove(struct pci_dev *pdev)
4098 {
4099 	struct net_device *netdev = pci_get_drvdata(pdev);
4100 	struct ixgbevf_adapter *adapter;
4101 	bool disable_dev;
4102 
4103 	if (!netdev)
4104 		return;
4105 
4106 	adapter = netdev_priv(netdev);
4107 
4108 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4109 	cancel_work_sync(&adapter->service_task);
4110 
4111 	if (netdev->reg_state == NETREG_REGISTERED)
4112 		unregister_netdev(netdev);
4113 
4114 	ixgbevf_clear_interrupt_scheme(adapter);
4115 	ixgbevf_reset_interrupt_capability(adapter);
4116 
4117 	iounmap(adapter->io_addr);
4118 	pci_release_regions(pdev);
4119 
4120 	hw_dbg(&adapter->hw, "Remove complete\n");
4121 
4122 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4123 	free_netdev(netdev);
4124 
4125 	if (disable_dev)
4126 		pci_disable_device(pdev);
4127 }
4128 
4129 /**
4130  * ixgbevf_io_error_detected - called when PCI error is detected
4131  * @pdev: Pointer to PCI device
4132  * @state: The current pci connection state
4133  *
4134  * This function is called after a PCI bus error affecting
4135  * this device has been detected.
4136  **/
4137 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4138 						  pci_channel_state_t state)
4139 {
4140 	struct net_device *netdev = pci_get_drvdata(pdev);
4141 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4142 
4143 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4144 		return PCI_ERS_RESULT_DISCONNECT;
4145 
4146 	rtnl_lock();
4147 	netif_device_detach(netdev);
4148 
4149 	if (state == pci_channel_io_perm_failure) {
4150 		rtnl_unlock();
4151 		return PCI_ERS_RESULT_DISCONNECT;
4152 	}
4153 
4154 	if (netif_running(netdev))
4155 		ixgbevf_down(adapter);
4156 
4157 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4158 		pci_disable_device(pdev);
4159 	rtnl_unlock();
4160 
4161 	/* Request a slot slot reset. */
4162 	return PCI_ERS_RESULT_NEED_RESET;
4163 }
4164 
4165 /**
4166  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4167  * @pdev: Pointer to PCI device
4168  *
4169  * Restart the card from scratch, as if from a cold-boot. Implementation
4170  * resembles the first-half of the ixgbevf_resume routine.
4171  **/
4172 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4173 {
4174 	struct net_device *netdev = pci_get_drvdata(pdev);
4175 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4176 
4177 	if (pci_enable_device_mem(pdev)) {
4178 		dev_err(&pdev->dev,
4179 			"Cannot re-enable PCI device after reset.\n");
4180 		return PCI_ERS_RESULT_DISCONNECT;
4181 	}
4182 
4183 	smp_mb__before_atomic();
4184 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4185 	pci_set_master(pdev);
4186 
4187 	ixgbevf_reset(adapter);
4188 
4189 	return PCI_ERS_RESULT_RECOVERED;
4190 }
4191 
4192 /**
4193  * ixgbevf_io_resume - called when traffic can start flowing again.
4194  * @pdev: Pointer to PCI device
4195  *
4196  * This callback is called when the error recovery driver tells us that
4197  * its OK to resume normal operation. Implementation resembles the
4198  * second-half of the ixgbevf_resume routine.
4199  **/
4200 static void ixgbevf_io_resume(struct pci_dev *pdev)
4201 {
4202 	struct net_device *netdev = pci_get_drvdata(pdev);
4203 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4204 
4205 	if (netif_running(netdev))
4206 		ixgbevf_up(adapter);
4207 
4208 	netif_device_attach(netdev);
4209 }
4210 
4211 /* PCI Error Recovery (ERS) */
4212 static const struct pci_error_handlers ixgbevf_err_handler = {
4213 	.error_detected = ixgbevf_io_error_detected,
4214 	.slot_reset = ixgbevf_io_slot_reset,
4215 	.resume = ixgbevf_io_resume,
4216 };
4217 
4218 static struct pci_driver ixgbevf_driver = {
4219 	.name		= ixgbevf_driver_name,
4220 	.id_table	= ixgbevf_pci_tbl,
4221 	.probe		= ixgbevf_probe,
4222 	.remove		= ixgbevf_remove,
4223 #ifdef CONFIG_PM
4224 	/* Power Management Hooks */
4225 	.suspend	= ixgbevf_suspend,
4226 	.resume		= ixgbevf_resume,
4227 #endif
4228 	.shutdown	= ixgbevf_shutdown,
4229 	.err_handler	= &ixgbevf_err_handler
4230 };
4231 
4232 /**
4233  * ixgbevf_init_module - Driver Registration Routine
4234  *
4235  * ixgbevf_init_module is the first routine called when the driver is
4236  * loaded. All it does is register with the PCI subsystem.
4237  **/
4238 static int __init ixgbevf_init_module(void)
4239 {
4240 	int ret;
4241 
4242 	pr_info("%s - version %s\n", ixgbevf_driver_string,
4243 		ixgbevf_driver_version);
4244 
4245 	pr_info("%s\n", ixgbevf_copyright);
4246 
4247 	ret = pci_register_driver(&ixgbevf_driver);
4248 	return ret;
4249 }
4250 
4251 module_init(ixgbevf_init_module);
4252 
4253 /**
4254  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4255  *
4256  * ixgbevf_exit_module is called just before the driver is removed
4257  * from memory.
4258  **/
4259 static void __exit ixgbevf_exit_module(void)
4260 {
4261 	pci_unregister_driver(&ixgbevf_driver);
4262 }
4263 
4264 #ifdef DEBUG
4265 /**
4266  * ixgbevf_get_hw_dev_name - return device name string
4267  * used by hardware layer to print debugging information
4268  **/
4269 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4270 {
4271 	struct ixgbevf_adapter *adapter = hw->back;
4272 
4273 	return adapter->netdev->name;
4274 }
4275 
4276 #endif
4277 module_exit(ixgbevf_exit_module);
4278 
4279 /* ixgbevf_main.c */
4280