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 /**
461  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
462  * @ring: structure containig ring specific data
463  * @rx_desc: current Rx descriptor being processed
464  * @skb: skb currently being received and modified
465  **/
466 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
467 				       union ixgbe_adv_rx_desc *rx_desc,
468 				       struct sk_buff *skb)
469 {
470 	skb_checksum_none_assert(skb);
471 
472 	/* Rx csum disabled */
473 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
474 		return;
475 
476 	/* if IP and error */
477 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
478 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
479 		ring->rx_stats.csum_err++;
480 		return;
481 	}
482 
483 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
484 		return;
485 
486 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
487 		ring->rx_stats.csum_err++;
488 		return;
489 	}
490 
491 	/* It must be a TCP or UDP packet with a valid checksum */
492 	skb->ip_summed = CHECKSUM_UNNECESSARY;
493 }
494 
495 /**
496  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
497  * @rx_ring: rx descriptor ring packet is being transacted on
498  * @rx_desc: pointer to the EOP Rx descriptor
499  * @skb: pointer to current skb being populated
500  *
501  * This function checks the ring, descriptor, and packet information in
502  * order to populate the checksum, VLAN, protocol, and other fields within
503  * the skb.
504  **/
505 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
506 				       union ixgbe_adv_rx_desc *rx_desc,
507 				       struct sk_buff *skb)
508 {
509 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
510 
511 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
512 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
513 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
514 
515 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
516 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
517 	}
518 
519 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
520 }
521 
522 /**
523  * ixgbevf_is_non_eop - process handling of non-EOP buffers
524  * @rx_ring: Rx ring being processed
525  * @rx_desc: Rx descriptor for current buffer
526  * @skb: current socket buffer containing buffer in progress
527  *
528  * This function updates next to clean.  If the buffer is an EOP buffer
529  * this function exits returning false, otherwise it will place the
530  * sk_buff in the next buffer to be chained and return true indicating
531  * that this is in fact a non-EOP buffer.
532  **/
533 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
534 			       union ixgbe_adv_rx_desc *rx_desc)
535 {
536 	u32 ntc = rx_ring->next_to_clean + 1;
537 
538 	/* fetch, update, and store next to clean */
539 	ntc = (ntc < rx_ring->count) ? ntc : 0;
540 	rx_ring->next_to_clean = ntc;
541 
542 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
543 
544 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
545 		return false;
546 
547 	return true;
548 }
549 
550 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
551 				      struct ixgbevf_rx_buffer *bi)
552 {
553 	struct page *page = bi->page;
554 	dma_addr_t dma = bi->dma;
555 
556 	/* since we are recycling buffers we should seldom need to alloc */
557 	if (likely(page))
558 		return true;
559 
560 	/* alloc new page for storage */
561 	page = dev_alloc_page();
562 	if (unlikely(!page)) {
563 		rx_ring->rx_stats.alloc_rx_page_failed++;
564 		return false;
565 	}
566 
567 	/* map page for use */
568 	dma = dma_map_page(rx_ring->dev, page, 0,
569 			   PAGE_SIZE, DMA_FROM_DEVICE);
570 
571 	/* if mapping failed free memory back to system since
572 	 * there isn't much point in holding memory we can't use
573 	 */
574 	if (dma_mapping_error(rx_ring->dev, dma)) {
575 		__free_page(page);
576 
577 		rx_ring->rx_stats.alloc_rx_buff_failed++;
578 		return false;
579 	}
580 
581 	bi->dma = dma;
582 	bi->page = page;
583 	bi->page_offset = 0;
584 
585 	return true;
586 }
587 
588 /**
589  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
590  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
591  * @cleaned_count: number of buffers to replace
592  **/
593 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
594 				     u16 cleaned_count)
595 {
596 	union ixgbe_adv_rx_desc *rx_desc;
597 	struct ixgbevf_rx_buffer *bi;
598 	unsigned int i = rx_ring->next_to_use;
599 
600 	/* nothing to do or no valid netdev defined */
601 	if (!cleaned_count || !rx_ring->netdev)
602 		return;
603 
604 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
605 	bi = &rx_ring->rx_buffer_info[i];
606 	i -= rx_ring->count;
607 
608 	do {
609 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
610 			break;
611 
612 		/* Refresh the desc even if pkt_addr didn't change
613 		 * because each write-back erases this info.
614 		 */
615 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
616 
617 		rx_desc++;
618 		bi++;
619 		i++;
620 		if (unlikely(!i)) {
621 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
622 			bi = rx_ring->rx_buffer_info;
623 			i -= rx_ring->count;
624 		}
625 
626 		/* clear the hdr_addr for the next_to_use descriptor */
627 		rx_desc->read.hdr_addr = 0;
628 
629 		cleaned_count--;
630 	} while (cleaned_count);
631 
632 	i += rx_ring->count;
633 
634 	if (rx_ring->next_to_use != i) {
635 		/* record the next descriptor to use */
636 		rx_ring->next_to_use = i;
637 
638 		/* update next to alloc since we have filled the ring */
639 		rx_ring->next_to_alloc = i;
640 
641 		/* Force memory writes to complete before letting h/w
642 		 * know there are new descriptors to fetch.  (Only
643 		 * applicable for weak-ordered memory model archs,
644 		 * such as IA-64).
645 		 */
646 		wmb();
647 		ixgbevf_write_tail(rx_ring, i);
648 	}
649 }
650 
651 /**
652  * ixgbevf_pull_tail - ixgbevf specific version of skb_pull_tail
653  * @rx_ring: rx descriptor ring packet is being transacted on
654  * @skb: pointer to current skb being adjusted
655  *
656  * This function is an ixgbevf specific version of __pskb_pull_tail.  The
657  * main difference between this version and the original function is that
658  * this function can make several assumptions about the state of things
659  * that allow for significant optimizations versus the standard function.
660  * As a result we can do things like drop a frag and maintain an accurate
661  * truesize for the skb.
662  **/
663 static void ixgbevf_pull_tail(struct ixgbevf_ring *rx_ring,
664 			      struct sk_buff *skb)
665 {
666 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
667 	unsigned char *va;
668 	unsigned int pull_len;
669 
670 	/* it is valid to use page_address instead of kmap since we are
671 	 * working with pages allocated out of the lomem pool per
672 	 * alloc_page(GFP_ATOMIC)
673 	 */
674 	va = skb_frag_address(frag);
675 
676 	/* we need the header to contain the greater of either ETH_HLEN or
677 	 * 60 bytes if the skb->len is less than 60 for skb_pad.
678 	 */
679 	pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
680 
681 	/* align pull length to size of long to optimize memcpy performance */
682 	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
683 
684 	/* update all of the pointers */
685 	skb_frag_size_sub(frag, pull_len);
686 	frag->page_offset += pull_len;
687 	skb->data_len -= pull_len;
688 	skb->tail += pull_len;
689 }
690 
691 /**
692  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
693  * @rx_ring: rx descriptor ring packet is being transacted on
694  * @rx_desc: pointer to the EOP Rx descriptor
695  * @skb: pointer to current skb being fixed
696  *
697  * Check for corrupted packet headers caused by senders on the local L2
698  * embedded NIC switch not setting up their Tx Descriptors right.  These
699  * should be very rare.
700  *
701  * Also address the case where we are pulling data in on pages only
702  * and as such no data is present in the skb header.
703  *
704  * In addition if skb is not at least 60 bytes we need to pad it so that
705  * it is large enough to qualify as a valid Ethernet frame.
706  *
707  * Returns true if an error was encountered and skb was freed.
708  **/
709 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
710 				    union ixgbe_adv_rx_desc *rx_desc,
711 				    struct sk_buff *skb)
712 {
713 	/* verify that the packet does not have any known errors */
714 	if (unlikely(ixgbevf_test_staterr(rx_desc,
715 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
716 		struct net_device *netdev = rx_ring->netdev;
717 
718 		if (!(netdev->features & NETIF_F_RXALL)) {
719 			dev_kfree_skb_any(skb);
720 			return true;
721 		}
722 	}
723 
724 	/* place header in linear portion of buffer */
725 	if (skb_is_nonlinear(skb))
726 		ixgbevf_pull_tail(rx_ring, skb);
727 
728 	/* if eth_skb_pad returns an error the skb was freed */
729 	if (eth_skb_pad(skb))
730 		return true;
731 
732 	return false;
733 }
734 
735 /**
736  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
737  * @rx_ring: rx descriptor ring to store buffers on
738  * @old_buff: donor buffer to have page reused
739  *
740  * Synchronizes page for reuse by the adapter
741  **/
742 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
743 				  struct ixgbevf_rx_buffer *old_buff)
744 {
745 	struct ixgbevf_rx_buffer *new_buff;
746 	u16 nta = rx_ring->next_to_alloc;
747 
748 	new_buff = &rx_ring->rx_buffer_info[nta];
749 
750 	/* update, and store next to alloc */
751 	nta++;
752 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
753 
754 	/* transfer page from old buffer to new buffer */
755 	new_buff->page = old_buff->page;
756 	new_buff->dma = old_buff->dma;
757 	new_buff->page_offset = old_buff->page_offset;
758 
759 	/* sync the buffer for use by the device */
760 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
761 					 new_buff->page_offset,
762 					 IXGBEVF_RX_BUFSZ,
763 					 DMA_FROM_DEVICE);
764 }
765 
766 static inline bool ixgbevf_page_is_reserved(struct page *page)
767 {
768 	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
769 }
770 
771 /**
772  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
773  * @rx_ring: rx descriptor ring to transact packets on
774  * @rx_buffer: buffer containing page to add
775  * @rx_desc: descriptor containing length of buffer written by hardware
776  * @skb: sk_buff to place the data into
777  *
778  * This function will add the data contained in rx_buffer->page to the skb.
779  * This is done either through a direct copy if the data in the buffer is
780  * less than the skb header size, otherwise it will just attach the page as
781  * a frag to the skb.
782  *
783  * The function will then update the page offset if necessary and return
784  * true if the buffer can be reused by the adapter.
785  **/
786 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
787 				struct ixgbevf_rx_buffer *rx_buffer,
788 				union ixgbe_adv_rx_desc *rx_desc,
789 				struct sk_buff *skb)
790 {
791 	struct page *page = rx_buffer->page;
792 	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
793 #if (PAGE_SIZE < 8192)
794 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
795 #else
796 	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
797 #endif
798 
799 	if ((size <= IXGBEVF_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
800 		unsigned char *va = page_address(page) + rx_buffer->page_offset;
801 
802 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
803 
804 		/* page is not reserved, we can reuse buffer as is */
805 		if (likely(!ixgbevf_page_is_reserved(page)))
806 			return true;
807 
808 		/* this page cannot be reused so discard it */
809 		put_page(page);
810 		return false;
811 	}
812 
813 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
814 			rx_buffer->page_offset, size, truesize);
815 
816 	/* avoid re-using remote pages */
817 	if (unlikely(ixgbevf_page_is_reserved(page)))
818 		return false;
819 
820 #if (PAGE_SIZE < 8192)
821 	/* if we are only owner of page we can reuse it */
822 	if (unlikely(page_count(page) != 1))
823 		return false;
824 
825 	/* flip page offset to other buffer */
826 	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
827 
828 #else
829 	/* move offset up to the next cache line */
830 	rx_buffer->page_offset += truesize;
831 
832 	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
833 		return false;
834 
835 #endif
836 	/* Even if we own the page, we are not allowed to use atomic_set()
837 	 * This would break get_page_unless_zero() users.
838 	 */
839 	atomic_inc(&page->_count);
840 
841 	return true;
842 }
843 
844 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
845 					       union ixgbe_adv_rx_desc *rx_desc,
846 					       struct sk_buff *skb)
847 {
848 	struct ixgbevf_rx_buffer *rx_buffer;
849 	struct page *page;
850 
851 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
852 	page = rx_buffer->page;
853 	prefetchw(page);
854 
855 	if (likely(!skb)) {
856 		void *page_addr = page_address(page) +
857 				  rx_buffer->page_offset;
858 
859 		/* prefetch first cache line of first page */
860 		prefetch(page_addr);
861 #if L1_CACHE_BYTES < 128
862 		prefetch(page_addr + L1_CACHE_BYTES);
863 #endif
864 
865 		/* allocate a skb to store the frags */
866 		skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
867 						IXGBEVF_RX_HDR_SIZE);
868 		if (unlikely(!skb)) {
869 			rx_ring->rx_stats.alloc_rx_buff_failed++;
870 			return NULL;
871 		}
872 
873 		/* we will be copying header into skb->data in
874 		 * pskb_may_pull so it is in our interest to prefetch
875 		 * it now to avoid a possible cache miss
876 		 */
877 		prefetchw(skb->data);
878 	}
879 
880 	/* we are reusing so sync this buffer for CPU use */
881 	dma_sync_single_range_for_cpu(rx_ring->dev,
882 				      rx_buffer->dma,
883 				      rx_buffer->page_offset,
884 				      IXGBEVF_RX_BUFSZ,
885 				      DMA_FROM_DEVICE);
886 
887 	/* pull page into skb */
888 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
889 		/* hand second half of page back to the ring */
890 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
891 	} else {
892 		/* we are not reusing the buffer so unmap it */
893 		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
894 			       PAGE_SIZE, DMA_FROM_DEVICE);
895 	}
896 
897 	/* clear contents of buffer_info */
898 	rx_buffer->dma = 0;
899 	rx_buffer->page = NULL;
900 
901 	return skb;
902 }
903 
904 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
905 					     u32 qmask)
906 {
907 	struct ixgbe_hw *hw = &adapter->hw;
908 
909 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
910 }
911 
912 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
913 				struct ixgbevf_ring *rx_ring,
914 				int budget)
915 {
916 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
917 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
918 	struct sk_buff *skb = rx_ring->skb;
919 
920 	while (likely(total_rx_packets < budget)) {
921 		union ixgbe_adv_rx_desc *rx_desc;
922 
923 		/* return some buffers to hardware, one at a time is too slow */
924 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
925 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
926 			cleaned_count = 0;
927 		}
928 
929 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
930 
931 		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
932 			break;
933 
934 		/* This memory barrier is needed to keep us from reading
935 		 * any other fields out of the rx_desc until we know the
936 		 * RXD_STAT_DD bit is set
937 		 */
938 		rmb();
939 
940 		/* retrieve a buffer from the ring */
941 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
942 
943 		/* exit if we failed to retrieve a buffer */
944 		if (!skb)
945 			break;
946 
947 		cleaned_count++;
948 
949 		/* fetch next buffer in frame if non-eop */
950 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
951 			continue;
952 
953 		/* verify the packet layout is correct */
954 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
955 			skb = NULL;
956 			continue;
957 		}
958 
959 		/* probably a little skewed due to removing CRC */
960 		total_rx_bytes += skb->len;
961 
962 		/* Workaround hardware that can't do proper VEPA multicast
963 		 * source pruning.
964 		 */
965 		if ((skb->pkt_type == PACKET_BROADCAST ||
966 		     skb->pkt_type == PACKET_MULTICAST) &&
967 		    ether_addr_equal(rx_ring->netdev->dev_addr,
968 				     eth_hdr(skb)->h_source)) {
969 			dev_kfree_skb_irq(skb);
970 			continue;
971 		}
972 
973 		/* populate checksum, VLAN, and protocol */
974 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
975 
976 		ixgbevf_rx_skb(q_vector, skb);
977 
978 		/* reset skb pointer */
979 		skb = NULL;
980 
981 		/* update budget accounting */
982 		total_rx_packets++;
983 	}
984 
985 	/* place incomplete frames back on ring for completion */
986 	rx_ring->skb = skb;
987 
988 	u64_stats_update_begin(&rx_ring->syncp);
989 	rx_ring->stats.packets += total_rx_packets;
990 	rx_ring->stats.bytes += total_rx_bytes;
991 	u64_stats_update_end(&rx_ring->syncp);
992 	q_vector->rx.total_packets += total_rx_packets;
993 	q_vector->rx.total_bytes += total_rx_bytes;
994 
995 	return total_rx_packets;
996 }
997 
998 /**
999  * ixgbevf_poll - NAPI polling calback
1000  * @napi: napi struct with our devices info in it
1001  * @budget: amount of work driver is allowed to do this pass, in packets
1002  *
1003  * This function will clean more than one or more rings associated with a
1004  * q_vector.
1005  **/
1006 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1007 {
1008 	struct ixgbevf_q_vector *q_vector =
1009 		container_of(napi, struct ixgbevf_q_vector, napi);
1010 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1011 	struct ixgbevf_ring *ring;
1012 	int per_ring_budget;
1013 	bool clean_complete = true;
1014 
1015 	ixgbevf_for_each_ring(ring, q_vector->tx)
1016 		clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1017 
1018 #ifdef CONFIG_NET_RX_BUSY_POLL
1019 	if (!ixgbevf_qv_lock_napi(q_vector))
1020 		return budget;
1021 #endif
1022 
1023 	/* attempt to distribute budget to each queue fairly, but don't allow
1024 	 * the budget to go below 1 because we'll exit polling
1025 	 */
1026 	if (q_vector->rx.count > 1)
1027 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1028 	else
1029 		per_ring_budget = budget;
1030 
1031 	ixgbevf_for_each_ring(ring, q_vector->rx)
1032 		clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
1033 							per_ring_budget)
1034 				   < per_ring_budget);
1035 
1036 #ifdef CONFIG_NET_RX_BUSY_POLL
1037 	ixgbevf_qv_unlock_napi(q_vector);
1038 #endif
1039 
1040 	/* If all work not completed, return budget and keep polling */
1041 	if (!clean_complete)
1042 		return budget;
1043 	/* all work done, exit the polling mode */
1044 	napi_complete(napi);
1045 	if (adapter->rx_itr_setting & 1)
1046 		ixgbevf_set_itr(q_vector);
1047 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1048 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1049 		ixgbevf_irq_enable_queues(adapter,
1050 					  1 << q_vector->v_idx);
1051 
1052 	return 0;
1053 }
1054 
1055 /**
1056  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1057  * @q_vector: structure containing interrupt and ring information
1058  **/
1059 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1060 {
1061 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1062 	struct ixgbe_hw *hw = &adapter->hw;
1063 	int v_idx = q_vector->v_idx;
1064 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1065 
1066 	/* set the WDIS bit to not clear the timer bits and cause an
1067 	 * immediate assertion of the interrupt
1068 	 */
1069 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1070 
1071 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1072 }
1073 
1074 #ifdef CONFIG_NET_RX_BUSY_POLL
1075 /* must be called with local_bh_disable()d */
1076 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1077 {
1078 	struct ixgbevf_q_vector *q_vector =
1079 			container_of(napi, struct ixgbevf_q_vector, napi);
1080 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1081 	struct ixgbevf_ring  *ring;
1082 	int found = 0;
1083 
1084 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1085 		return LL_FLUSH_FAILED;
1086 
1087 	if (!ixgbevf_qv_lock_poll(q_vector))
1088 		return LL_FLUSH_BUSY;
1089 
1090 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1091 		found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1092 #ifdef BP_EXTENDED_STATS
1093 		if (found)
1094 			ring->stats.cleaned += found;
1095 		else
1096 			ring->stats.misses++;
1097 #endif
1098 		if (found)
1099 			break;
1100 	}
1101 
1102 	ixgbevf_qv_unlock_poll(q_vector);
1103 
1104 	return found;
1105 }
1106 #endif /* CONFIG_NET_RX_BUSY_POLL */
1107 
1108 /**
1109  * ixgbevf_configure_msix - Configure MSI-X hardware
1110  * @adapter: board private structure
1111  *
1112  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1113  * interrupts.
1114  **/
1115 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1116 {
1117 	struct ixgbevf_q_vector *q_vector;
1118 	int q_vectors, v_idx;
1119 
1120 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1121 	adapter->eims_enable_mask = 0;
1122 
1123 	/* Populate the IVAR table and set the ITR values to the
1124 	 * corresponding register.
1125 	 */
1126 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1127 		struct ixgbevf_ring *ring;
1128 
1129 		q_vector = adapter->q_vector[v_idx];
1130 
1131 		ixgbevf_for_each_ring(ring, q_vector->rx)
1132 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1133 
1134 		ixgbevf_for_each_ring(ring, q_vector->tx)
1135 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1136 
1137 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1138 			/* Tx only vector */
1139 			if (adapter->tx_itr_setting == 1)
1140 				q_vector->itr = IXGBE_10K_ITR;
1141 			else
1142 				q_vector->itr = adapter->tx_itr_setting;
1143 		} else {
1144 			/* Rx or Rx/Tx vector */
1145 			if (adapter->rx_itr_setting == 1)
1146 				q_vector->itr = IXGBE_20K_ITR;
1147 			else
1148 				q_vector->itr = adapter->rx_itr_setting;
1149 		}
1150 
1151 		/* add q_vector eims value to global eims_enable_mask */
1152 		adapter->eims_enable_mask |= 1 << v_idx;
1153 
1154 		ixgbevf_write_eitr(q_vector);
1155 	}
1156 
1157 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1158 	/* setup eims_other and add value to global eims_enable_mask */
1159 	adapter->eims_other = 1 << v_idx;
1160 	adapter->eims_enable_mask |= adapter->eims_other;
1161 }
1162 
1163 enum latency_range {
1164 	lowest_latency = 0,
1165 	low_latency = 1,
1166 	bulk_latency = 2,
1167 	latency_invalid = 255
1168 };
1169 
1170 /**
1171  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1172  * @q_vector: structure containing interrupt and ring information
1173  * @ring_container: structure containing ring performance data
1174  *
1175  * Stores a new ITR value based on packets and byte
1176  * counts during the last interrupt.  The advantage of per interrupt
1177  * computation is faster updates and more accurate ITR for the current
1178  * traffic pattern.  Constants in this function were computed
1179  * based on theoretical maximum wire speed and thresholds were set based
1180  * on testing data as well as attempting to minimize response time
1181  * while increasing bulk throughput.
1182  **/
1183 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1184 			       struct ixgbevf_ring_container *ring_container)
1185 {
1186 	int bytes = ring_container->total_bytes;
1187 	int packets = ring_container->total_packets;
1188 	u32 timepassed_us;
1189 	u64 bytes_perint;
1190 	u8 itr_setting = ring_container->itr;
1191 
1192 	if (packets == 0)
1193 		return;
1194 
1195 	/* simple throttle rate management
1196 	 *    0-20MB/s lowest (100000 ints/s)
1197 	 *   20-100MB/s low   (20000 ints/s)
1198 	 *  100-1249MB/s bulk (8000 ints/s)
1199 	 */
1200 	/* what was last interrupt timeslice? */
1201 	timepassed_us = q_vector->itr >> 2;
1202 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1203 
1204 	switch (itr_setting) {
1205 	case lowest_latency:
1206 		if (bytes_perint > 10)
1207 			itr_setting = low_latency;
1208 		break;
1209 	case low_latency:
1210 		if (bytes_perint > 20)
1211 			itr_setting = bulk_latency;
1212 		else if (bytes_perint <= 10)
1213 			itr_setting = lowest_latency;
1214 		break;
1215 	case bulk_latency:
1216 		if (bytes_perint <= 20)
1217 			itr_setting = low_latency;
1218 		break;
1219 	}
1220 
1221 	/* clear work counters since we have the values we need */
1222 	ring_container->total_bytes = 0;
1223 	ring_container->total_packets = 0;
1224 
1225 	/* write updated itr to ring container */
1226 	ring_container->itr = itr_setting;
1227 }
1228 
1229 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1230 {
1231 	u32 new_itr = q_vector->itr;
1232 	u8 current_itr;
1233 
1234 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1235 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1236 
1237 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1238 
1239 	switch (current_itr) {
1240 	/* counts and packets in update_itr are dependent on these numbers */
1241 	case lowest_latency:
1242 		new_itr = IXGBE_100K_ITR;
1243 		break;
1244 	case low_latency:
1245 		new_itr = IXGBE_20K_ITR;
1246 		break;
1247 	case bulk_latency:
1248 	default:
1249 		new_itr = IXGBE_8K_ITR;
1250 		break;
1251 	}
1252 
1253 	if (new_itr != q_vector->itr) {
1254 		/* do an exponential smoothing */
1255 		new_itr = (10 * new_itr * q_vector->itr) /
1256 			  ((9 * new_itr) + q_vector->itr);
1257 
1258 		/* save the algorithm value here */
1259 		q_vector->itr = new_itr;
1260 
1261 		ixgbevf_write_eitr(q_vector);
1262 	}
1263 }
1264 
1265 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1266 {
1267 	struct ixgbevf_adapter *adapter = data;
1268 	struct ixgbe_hw *hw = &adapter->hw;
1269 
1270 	hw->mac.get_link_status = 1;
1271 
1272 	ixgbevf_service_event_schedule(adapter);
1273 
1274 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1275 
1276 	return IRQ_HANDLED;
1277 }
1278 
1279 /**
1280  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1281  * @irq: unused
1282  * @data: pointer to our q_vector struct for this interrupt vector
1283  **/
1284 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1285 {
1286 	struct ixgbevf_q_vector *q_vector = data;
1287 
1288 	/* EIAM disabled interrupts (on this vector) for us */
1289 	if (q_vector->rx.ring || q_vector->tx.ring)
1290 		napi_schedule(&q_vector->napi);
1291 
1292 	return IRQ_HANDLED;
1293 }
1294 
1295 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1296 				     int r_idx)
1297 {
1298 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1299 
1300 	a->rx_ring[r_idx]->next = q_vector->rx.ring;
1301 	q_vector->rx.ring = a->rx_ring[r_idx];
1302 	q_vector->rx.count++;
1303 }
1304 
1305 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1306 				     int t_idx)
1307 {
1308 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1309 
1310 	a->tx_ring[t_idx]->next = q_vector->tx.ring;
1311 	q_vector->tx.ring = a->tx_ring[t_idx];
1312 	q_vector->tx.count++;
1313 }
1314 
1315 /**
1316  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1317  * @adapter: board private structure to initialize
1318  *
1319  * This function maps descriptor rings to the queue-specific vectors
1320  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1321  * one vector per ring/queue, but on a constrained vector budget, we
1322  * group the rings as "efficiently" as possible.  You would add new
1323  * mapping configurations in here.
1324  **/
1325 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1326 {
1327 	int q_vectors;
1328 	int v_start = 0;
1329 	int rxr_idx = 0, txr_idx = 0;
1330 	int rxr_remaining = adapter->num_rx_queues;
1331 	int txr_remaining = adapter->num_tx_queues;
1332 	int i, j;
1333 	int rqpv, tqpv;
1334 	int err = 0;
1335 
1336 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1337 
1338 	/* The ideal configuration...
1339 	 * We have enough vectors to map one per queue.
1340 	 */
1341 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1342 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1343 			map_vector_to_rxq(adapter, v_start, rxr_idx);
1344 
1345 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1346 			map_vector_to_txq(adapter, v_start, txr_idx);
1347 		goto out;
1348 	}
1349 
1350 	/* If we don't have enough vectors for a 1-to-1
1351 	 * mapping, we'll have to group them so there are
1352 	 * multiple queues per vector.
1353 	 */
1354 	/* Re-adjusting *qpv takes care of the remainder. */
1355 	for (i = v_start; i < q_vectors; i++) {
1356 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1357 		for (j = 0; j < rqpv; j++) {
1358 			map_vector_to_rxq(adapter, i, rxr_idx);
1359 			rxr_idx++;
1360 			rxr_remaining--;
1361 		}
1362 	}
1363 	for (i = v_start; i < q_vectors; i++) {
1364 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1365 		for (j = 0; j < tqpv; j++) {
1366 			map_vector_to_txq(adapter, i, txr_idx);
1367 			txr_idx++;
1368 			txr_remaining--;
1369 		}
1370 	}
1371 
1372 out:
1373 	return err;
1374 }
1375 
1376 /**
1377  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1378  * @adapter: board private structure
1379  *
1380  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1381  * interrupts from the kernel.
1382  **/
1383 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1384 {
1385 	struct net_device *netdev = adapter->netdev;
1386 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1387 	int vector, err;
1388 	int ri = 0, ti = 0;
1389 
1390 	for (vector = 0; vector < q_vectors; vector++) {
1391 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1392 		struct msix_entry *entry = &adapter->msix_entries[vector];
1393 
1394 		if (q_vector->tx.ring && q_vector->rx.ring) {
1395 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1396 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
1397 			ti++;
1398 		} else if (q_vector->rx.ring) {
1399 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1400 				 "%s-%s-%d", netdev->name, "rx", ri++);
1401 		} else if (q_vector->tx.ring) {
1402 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1403 				 "%s-%s-%d", netdev->name, "tx", ti++);
1404 		} else {
1405 			/* skip this unused q_vector */
1406 			continue;
1407 		}
1408 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1409 				  q_vector->name, q_vector);
1410 		if (err) {
1411 			hw_dbg(&adapter->hw,
1412 			       "request_irq failed for MSIX interrupt Error: %d\n",
1413 			       err);
1414 			goto free_queue_irqs;
1415 		}
1416 	}
1417 
1418 	err = request_irq(adapter->msix_entries[vector].vector,
1419 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1420 	if (err) {
1421 		hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1422 		       err);
1423 		goto free_queue_irqs;
1424 	}
1425 
1426 	return 0;
1427 
1428 free_queue_irqs:
1429 	while (vector) {
1430 		vector--;
1431 		free_irq(adapter->msix_entries[vector].vector,
1432 			 adapter->q_vector[vector]);
1433 	}
1434 	/* This failure is non-recoverable - it indicates the system is
1435 	 * out of MSIX vector resources and the VF driver cannot run
1436 	 * without them.  Set the number of msix vectors to zero
1437 	 * indicating that not enough can be allocated.  The error
1438 	 * will be returned to the user indicating device open failed.
1439 	 * Any further attempts to force the driver to open will also
1440 	 * fail.  The only way to recover is to unload the driver and
1441 	 * reload it again.  If the system has recovered some MSIX
1442 	 * vectors then it may succeed.
1443 	 */
1444 	adapter->num_msix_vectors = 0;
1445 	return err;
1446 }
1447 
1448 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1449 {
1450 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1451 
1452 	for (i = 0; i < q_vectors; i++) {
1453 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1454 
1455 		q_vector->rx.ring = NULL;
1456 		q_vector->tx.ring = NULL;
1457 		q_vector->rx.count = 0;
1458 		q_vector->tx.count = 0;
1459 	}
1460 }
1461 
1462 /**
1463  * ixgbevf_request_irq - initialize interrupts
1464  * @adapter: board private structure
1465  *
1466  * Attempts to configure interrupts using the best available
1467  * capabilities of the hardware and kernel.
1468  **/
1469 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1470 {
1471 	int err = 0;
1472 
1473 	err = ixgbevf_request_msix_irqs(adapter);
1474 
1475 	if (err)
1476 		hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1477 
1478 	return err;
1479 }
1480 
1481 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1482 {
1483 	int i, q_vectors;
1484 
1485 	q_vectors = adapter->num_msix_vectors;
1486 	i = q_vectors - 1;
1487 
1488 	free_irq(adapter->msix_entries[i].vector, adapter);
1489 	i--;
1490 
1491 	for (; i >= 0; i--) {
1492 		/* free only the irqs that were actually requested */
1493 		if (!adapter->q_vector[i]->rx.ring &&
1494 		    !adapter->q_vector[i]->tx.ring)
1495 			continue;
1496 
1497 		free_irq(adapter->msix_entries[i].vector,
1498 			 adapter->q_vector[i]);
1499 	}
1500 
1501 	ixgbevf_reset_q_vectors(adapter);
1502 }
1503 
1504 /**
1505  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1506  * @adapter: board private structure
1507  **/
1508 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1509 {
1510 	struct ixgbe_hw *hw = &adapter->hw;
1511 	int i;
1512 
1513 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1514 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1515 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1516 
1517 	IXGBE_WRITE_FLUSH(hw);
1518 
1519 	for (i = 0; i < adapter->num_msix_vectors; i++)
1520 		synchronize_irq(adapter->msix_entries[i].vector);
1521 }
1522 
1523 /**
1524  * ixgbevf_irq_enable - Enable default interrupt generation settings
1525  * @adapter: board private structure
1526  **/
1527 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1528 {
1529 	struct ixgbe_hw *hw = &adapter->hw;
1530 
1531 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1532 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1533 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1534 }
1535 
1536 /**
1537  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1538  * @adapter: board private structure
1539  * @ring: structure containing ring specific data
1540  *
1541  * Configure the Tx descriptor ring after a reset.
1542  **/
1543 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1544 				      struct ixgbevf_ring *ring)
1545 {
1546 	struct ixgbe_hw *hw = &adapter->hw;
1547 	u64 tdba = ring->dma;
1548 	int wait_loop = 10;
1549 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1550 	u8 reg_idx = ring->reg_idx;
1551 
1552 	/* disable queue to avoid issues while updating state */
1553 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1554 	IXGBE_WRITE_FLUSH(hw);
1555 
1556 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1557 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1558 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1559 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1560 
1561 	/* disable head writeback */
1562 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1563 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1564 
1565 	/* enable relaxed ordering */
1566 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1567 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1568 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1569 
1570 	/* reset head and tail pointers */
1571 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1572 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1573 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1574 
1575 	/* reset ntu and ntc to place SW in sync with hardwdare */
1576 	ring->next_to_clean = 0;
1577 	ring->next_to_use = 0;
1578 
1579 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1580 	 * to or less than the number of on chip descriptors, which is
1581 	 * currently 40.
1582 	 */
1583 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1584 
1585 	/* Setting PTHRESH to 32 both improves performance */
1586 	txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1587 		  32;          /* PTHRESH = 32 */
1588 
1589 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1590 
1591 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1592 
1593 	/* poll to verify queue is enabled */
1594 	do {
1595 		usleep_range(1000, 2000);
1596 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1597 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1598 	if (!wait_loop)
1599 		pr_err("Could not enable Tx Queue %d\n", reg_idx);
1600 }
1601 
1602 /**
1603  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1604  * @adapter: board private structure
1605  *
1606  * Configure the Tx unit of the MAC after a reset.
1607  **/
1608 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1609 {
1610 	u32 i;
1611 
1612 	/* Setup the HW Tx Head and Tail descriptor pointers */
1613 	for (i = 0; i < adapter->num_tx_queues; i++)
1614 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1615 }
1616 
1617 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1618 
1619 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1620 {
1621 	struct ixgbe_hw *hw = &adapter->hw;
1622 	u32 srrctl;
1623 
1624 	srrctl = IXGBE_SRRCTL_DROP_EN;
1625 
1626 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1627 	srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1628 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1629 
1630 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1631 }
1632 
1633 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1634 {
1635 	struct ixgbe_hw *hw = &adapter->hw;
1636 
1637 	/* PSRTYPE must be initialized in 82599 */
1638 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1639 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1640 		      IXGBE_PSRTYPE_L2HDR;
1641 
1642 	if (adapter->num_rx_queues > 1)
1643 		psrtype |= 1 << 29;
1644 
1645 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1646 }
1647 
1648 #define IXGBEVF_MAX_RX_DESC_POLL 10
1649 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1650 				     struct ixgbevf_ring *ring)
1651 {
1652 	struct ixgbe_hw *hw = &adapter->hw;
1653 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1654 	u32 rxdctl;
1655 	u8 reg_idx = ring->reg_idx;
1656 
1657 	if (IXGBE_REMOVED(hw->hw_addr))
1658 		return;
1659 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1660 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1661 
1662 	/* write value back with RXDCTL.ENABLE bit cleared */
1663 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1664 
1665 	/* the hardware may take up to 100us to really disable the Rx queue */
1666 	do {
1667 		udelay(10);
1668 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1669 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1670 
1671 	if (!wait_loop)
1672 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1673 		       reg_idx);
1674 }
1675 
1676 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1677 					 struct ixgbevf_ring *ring)
1678 {
1679 	struct ixgbe_hw *hw = &adapter->hw;
1680 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1681 	u32 rxdctl;
1682 	u8 reg_idx = ring->reg_idx;
1683 
1684 	if (IXGBE_REMOVED(hw->hw_addr))
1685 		return;
1686 	do {
1687 		usleep_range(1000, 2000);
1688 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1689 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1690 
1691 	if (!wait_loop)
1692 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1693 		       reg_idx);
1694 }
1695 
1696 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1697 {
1698 	struct ixgbe_hw *hw = &adapter->hw;
1699 	u32 vfmrqc = 0, vfreta = 0;
1700 	u32 rss_key[10];
1701 	u16 rss_i = adapter->num_rx_queues;
1702 	int i, j;
1703 
1704 	/* Fill out hash function seeds */
1705 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
1706 	for (i = 0; i < 10; i++)
1707 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), rss_key[i]);
1708 
1709 	/* Fill out redirection table */
1710 	for (i = 0, j = 0; i < 64; i++, j++) {
1711 		if (j == rss_i)
1712 			j = 0;
1713 		vfreta = (vfreta << 8) | (j * 0x1);
1714 		if ((i & 3) == 3)
1715 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1716 	}
1717 
1718 	/* Perform hash on these packet types */
1719 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1720 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1721 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1722 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1723 
1724 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1725 
1726 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1727 }
1728 
1729 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1730 				      struct ixgbevf_ring *ring)
1731 {
1732 	struct ixgbe_hw *hw = &adapter->hw;
1733 	u64 rdba = ring->dma;
1734 	u32 rxdctl;
1735 	u8 reg_idx = ring->reg_idx;
1736 
1737 	/* disable queue to avoid issues while updating state */
1738 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1739 	ixgbevf_disable_rx_queue(adapter, ring);
1740 
1741 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1742 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1743 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1744 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1745 
1746 	/* enable relaxed ordering */
1747 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1748 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1749 
1750 	/* reset head and tail pointers */
1751 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1752 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1753 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1754 
1755 	/* reset ntu and ntc to place SW in sync with hardwdare */
1756 	ring->next_to_clean = 0;
1757 	ring->next_to_use = 0;
1758 	ring->next_to_alloc = 0;
1759 
1760 	ixgbevf_configure_srrctl(adapter, reg_idx);
1761 
1762 	/* allow any size packet since we can handle overflow */
1763 	rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1764 
1765 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1766 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1767 
1768 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1769 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1770 }
1771 
1772 /**
1773  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1774  * @adapter: board private structure
1775  *
1776  * Configure the Rx unit of the MAC after a reset.
1777  **/
1778 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1779 {
1780 	int i;
1781 	struct ixgbe_hw *hw = &adapter->hw;
1782 	struct net_device *netdev = adapter->netdev;
1783 
1784 	ixgbevf_setup_psrtype(adapter);
1785 	if (hw->mac.type >= ixgbe_mac_X550_vf)
1786 		ixgbevf_setup_vfmrqc(adapter);
1787 
1788 	/* notify the PF of our intent to use this size of frame */
1789 	ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1790 
1791 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1792 	 * the Base and Length of the Rx Descriptor Ring
1793 	 */
1794 	for (i = 0; i < adapter->num_rx_queues; i++)
1795 		ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1796 }
1797 
1798 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1799 				   __be16 proto, u16 vid)
1800 {
1801 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1802 	struct ixgbe_hw *hw = &adapter->hw;
1803 	int err;
1804 
1805 	spin_lock_bh(&adapter->mbx_lock);
1806 
1807 	/* add VID to filter table */
1808 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1809 
1810 	spin_unlock_bh(&adapter->mbx_lock);
1811 
1812 	/* translate error return types so error makes sense */
1813 	if (err == IXGBE_ERR_MBX)
1814 		return -EIO;
1815 
1816 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
1817 		return -EACCES;
1818 
1819 	set_bit(vid, adapter->active_vlans);
1820 
1821 	return err;
1822 }
1823 
1824 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1825 				    __be16 proto, u16 vid)
1826 {
1827 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1828 	struct ixgbe_hw *hw = &adapter->hw;
1829 	int err = -EOPNOTSUPP;
1830 
1831 	spin_lock_bh(&adapter->mbx_lock);
1832 
1833 	/* remove VID from filter table */
1834 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1835 
1836 	spin_unlock_bh(&adapter->mbx_lock);
1837 
1838 	clear_bit(vid, adapter->active_vlans);
1839 
1840 	return err;
1841 }
1842 
1843 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1844 {
1845 	u16 vid;
1846 
1847 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1848 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
1849 					htons(ETH_P_8021Q), vid);
1850 }
1851 
1852 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1853 {
1854 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1855 	struct ixgbe_hw *hw = &adapter->hw;
1856 	int count = 0;
1857 
1858 	if ((netdev_uc_count(netdev)) > 10) {
1859 		pr_err("Too many unicast filters - No Space\n");
1860 		return -ENOSPC;
1861 	}
1862 
1863 	if (!netdev_uc_empty(netdev)) {
1864 		struct netdev_hw_addr *ha;
1865 
1866 		netdev_for_each_uc_addr(ha, netdev) {
1867 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1868 			udelay(200);
1869 		}
1870 	} else {
1871 		/* If the list is empty then send message to PF driver to
1872 		 * clear all MAC VLANs on this VF.
1873 		 */
1874 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1875 	}
1876 
1877 	return count;
1878 }
1879 
1880 /**
1881  * ixgbevf_set_rx_mode - Multicast and unicast set
1882  * @netdev: network interface device structure
1883  *
1884  * The set_rx_method entry point is called whenever the multicast address
1885  * list, unicast address list or the network interface flags are updated.
1886  * This routine is responsible for configuring the hardware for proper
1887  * multicast mode and configuring requested unicast filters.
1888  **/
1889 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1890 {
1891 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1892 	struct ixgbe_hw *hw = &adapter->hw;
1893 
1894 	spin_lock_bh(&adapter->mbx_lock);
1895 
1896 	/* reprogram multicast list */
1897 	hw->mac.ops.update_mc_addr_list(hw, netdev);
1898 
1899 	ixgbevf_write_uc_addr_list(netdev);
1900 
1901 	spin_unlock_bh(&adapter->mbx_lock);
1902 }
1903 
1904 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1905 {
1906 	int q_idx;
1907 	struct ixgbevf_q_vector *q_vector;
1908 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1909 
1910 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1911 		q_vector = adapter->q_vector[q_idx];
1912 #ifdef CONFIG_NET_RX_BUSY_POLL
1913 		ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1914 #endif
1915 		napi_enable(&q_vector->napi);
1916 	}
1917 }
1918 
1919 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1920 {
1921 	int q_idx;
1922 	struct ixgbevf_q_vector *q_vector;
1923 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1924 
1925 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1926 		q_vector = adapter->q_vector[q_idx];
1927 		napi_disable(&q_vector->napi);
1928 #ifdef CONFIG_NET_RX_BUSY_POLL
1929 		while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1930 			pr_info("QV %d locked\n", q_idx);
1931 			usleep_range(1000, 20000);
1932 		}
1933 #endif /* CONFIG_NET_RX_BUSY_POLL */
1934 	}
1935 }
1936 
1937 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1938 {
1939 	struct ixgbe_hw *hw = &adapter->hw;
1940 	unsigned int def_q = 0;
1941 	unsigned int num_tcs = 0;
1942 	unsigned int num_rx_queues = adapter->num_rx_queues;
1943 	unsigned int num_tx_queues = adapter->num_tx_queues;
1944 	int err;
1945 
1946 	spin_lock_bh(&adapter->mbx_lock);
1947 
1948 	/* fetch queue configuration from the PF */
1949 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1950 
1951 	spin_unlock_bh(&adapter->mbx_lock);
1952 
1953 	if (err)
1954 		return err;
1955 
1956 	if (num_tcs > 1) {
1957 		/* we need only one Tx queue */
1958 		num_tx_queues = 1;
1959 
1960 		/* update default Tx ring register index */
1961 		adapter->tx_ring[0]->reg_idx = def_q;
1962 
1963 		/* we need as many queues as traffic classes */
1964 		num_rx_queues = num_tcs;
1965 	}
1966 
1967 	/* if we have a bad config abort request queue reset */
1968 	if ((adapter->num_rx_queues != num_rx_queues) ||
1969 	    (adapter->num_tx_queues != num_tx_queues)) {
1970 		/* force mailbox timeout to prevent further messages */
1971 		hw->mbx.timeout = 0;
1972 
1973 		/* wait for watchdog to come around and bail us out */
1974 		adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1975 	}
1976 
1977 	return 0;
1978 }
1979 
1980 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1981 {
1982 	ixgbevf_configure_dcb(adapter);
1983 
1984 	ixgbevf_set_rx_mode(adapter->netdev);
1985 
1986 	ixgbevf_restore_vlan(adapter);
1987 
1988 	ixgbevf_configure_tx(adapter);
1989 	ixgbevf_configure_rx(adapter);
1990 }
1991 
1992 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1993 {
1994 	/* Only save pre-reset stats if there are some */
1995 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1996 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1997 			adapter->stats.base_vfgprc;
1998 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1999 			adapter->stats.base_vfgptc;
2000 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2001 			adapter->stats.base_vfgorc;
2002 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2003 			adapter->stats.base_vfgotc;
2004 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2005 			adapter->stats.base_vfmprc;
2006 	}
2007 }
2008 
2009 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2010 {
2011 	struct ixgbe_hw *hw = &adapter->hw;
2012 
2013 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2014 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2015 	adapter->stats.last_vfgorc |=
2016 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2017 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2018 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2019 	adapter->stats.last_vfgotc |=
2020 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2021 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2022 
2023 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2024 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2025 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2026 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2027 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2028 }
2029 
2030 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2031 {
2032 	struct ixgbe_hw *hw = &adapter->hw;
2033 	int api[] = { ixgbe_mbox_api_12,
2034 		      ixgbe_mbox_api_11,
2035 		      ixgbe_mbox_api_10,
2036 		      ixgbe_mbox_api_unknown };
2037 	int err = 0, idx = 0;
2038 
2039 	spin_lock_bh(&adapter->mbx_lock);
2040 
2041 	while (api[idx] != ixgbe_mbox_api_unknown) {
2042 		err = ixgbevf_negotiate_api_version(hw, api[idx]);
2043 		if (!err)
2044 			break;
2045 		idx++;
2046 	}
2047 
2048 	spin_unlock_bh(&adapter->mbx_lock);
2049 }
2050 
2051 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2052 {
2053 	struct net_device *netdev = adapter->netdev;
2054 	struct ixgbe_hw *hw = &adapter->hw;
2055 
2056 	ixgbevf_configure_msix(adapter);
2057 
2058 	spin_lock_bh(&adapter->mbx_lock);
2059 
2060 	if (is_valid_ether_addr(hw->mac.addr))
2061 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2062 	else
2063 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2064 
2065 	spin_unlock_bh(&adapter->mbx_lock);
2066 
2067 	smp_mb__before_atomic();
2068 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2069 	ixgbevf_napi_enable_all(adapter);
2070 
2071 	/* clear any pending interrupts, may auto mask */
2072 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2073 	ixgbevf_irq_enable(adapter);
2074 
2075 	/* enable transmits */
2076 	netif_tx_start_all_queues(netdev);
2077 
2078 	ixgbevf_save_reset_stats(adapter);
2079 	ixgbevf_init_last_counter_stats(adapter);
2080 
2081 	hw->mac.get_link_status = 1;
2082 	mod_timer(&adapter->service_timer, jiffies);
2083 }
2084 
2085 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2086 {
2087 	ixgbevf_configure(adapter);
2088 
2089 	ixgbevf_up_complete(adapter);
2090 }
2091 
2092 /**
2093  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2094  * @rx_ring: ring to free buffers from
2095  **/
2096 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2097 {
2098 	struct device *dev = rx_ring->dev;
2099 	unsigned long size;
2100 	unsigned int i;
2101 
2102 	/* Free Rx ring sk_buff */
2103 	if (rx_ring->skb) {
2104 		dev_kfree_skb(rx_ring->skb);
2105 		rx_ring->skb = NULL;
2106 	}
2107 
2108 	/* ring already cleared, nothing to do */
2109 	if (!rx_ring->rx_buffer_info)
2110 		return;
2111 
2112 	/* Free all the Rx ring pages */
2113 	for (i = 0; i < rx_ring->count; i++) {
2114 		struct ixgbevf_rx_buffer *rx_buffer;
2115 
2116 		rx_buffer = &rx_ring->rx_buffer_info[i];
2117 		if (rx_buffer->dma)
2118 			dma_unmap_page(dev, rx_buffer->dma,
2119 				       PAGE_SIZE, DMA_FROM_DEVICE);
2120 		rx_buffer->dma = 0;
2121 		if (rx_buffer->page)
2122 			__free_page(rx_buffer->page);
2123 		rx_buffer->page = NULL;
2124 	}
2125 
2126 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2127 	memset(rx_ring->rx_buffer_info, 0, size);
2128 
2129 	/* Zero out the descriptor ring */
2130 	memset(rx_ring->desc, 0, rx_ring->size);
2131 }
2132 
2133 /**
2134  * ixgbevf_clean_tx_ring - Free Tx Buffers
2135  * @tx_ring: ring to be cleaned
2136  **/
2137 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2138 {
2139 	struct ixgbevf_tx_buffer *tx_buffer_info;
2140 	unsigned long size;
2141 	unsigned int i;
2142 
2143 	if (!tx_ring->tx_buffer_info)
2144 		return;
2145 
2146 	/* Free all the Tx ring sk_buffs */
2147 	for (i = 0; i < tx_ring->count; i++) {
2148 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2149 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2150 	}
2151 
2152 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2153 	memset(tx_ring->tx_buffer_info, 0, size);
2154 
2155 	memset(tx_ring->desc, 0, tx_ring->size);
2156 }
2157 
2158 /**
2159  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2160  * @adapter: board private structure
2161  **/
2162 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2163 {
2164 	int i;
2165 
2166 	for (i = 0; i < adapter->num_rx_queues; i++)
2167 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2168 }
2169 
2170 /**
2171  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2172  * @adapter: board private structure
2173  **/
2174 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2175 {
2176 	int i;
2177 
2178 	for (i = 0; i < adapter->num_tx_queues; i++)
2179 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2180 }
2181 
2182 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2183 {
2184 	struct net_device *netdev = adapter->netdev;
2185 	struct ixgbe_hw *hw = &adapter->hw;
2186 	int i;
2187 
2188 	/* signal that we are down to the interrupt handler */
2189 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2190 		return; /* do nothing if already down */
2191 
2192 	/* disable all enabled Rx queues */
2193 	for (i = 0; i < adapter->num_rx_queues; i++)
2194 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2195 
2196 	usleep_range(10000, 20000);
2197 
2198 	netif_tx_stop_all_queues(netdev);
2199 
2200 	/* call carrier off first to avoid false dev_watchdog timeouts */
2201 	netif_carrier_off(netdev);
2202 	netif_tx_disable(netdev);
2203 
2204 	ixgbevf_irq_disable(adapter);
2205 
2206 	ixgbevf_napi_disable_all(adapter);
2207 
2208 	del_timer_sync(&adapter->service_timer);
2209 
2210 	/* disable transmits in the hardware now that interrupts are off */
2211 	for (i = 0; i < adapter->num_tx_queues; i++) {
2212 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2213 
2214 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2215 				IXGBE_TXDCTL_SWFLSH);
2216 	}
2217 
2218 	if (!pci_channel_offline(adapter->pdev))
2219 		ixgbevf_reset(adapter);
2220 
2221 	ixgbevf_clean_all_tx_rings(adapter);
2222 	ixgbevf_clean_all_rx_rings(adapter);
2223 }
2224 
2225 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2226 {
2227 	WARN_ON(in_interrupt());
2228 
2229 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2230 		msleep(1);
2231 
2232 	ixgbevf_down(adapter);
2233 	ixgbevf_up(adapter);
2234 
2235 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2236 }
2237 
2238 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2239 {
2240 	struct ixgbe_hw *hw = &adapter->hw;
2241 	struct net_device *netdev = adapter->netdev;
2242 
2243 	if (hw->mac.ops.reset_hw(hw)) {
2244 		hw_dbg(hw, "PF still resetting\n");
2245 	} else {
2246 		hw->mac.ops.init_hw(hw);
2247 		ixgbevf_negotiate_api(adapter);
2248 	}
2249 
2250 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2251 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2252 		       netdev->addr_len);
2253 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2254 		       netdev->addr_len);
2255 	}
2256 
2257 	adapter->last_reset = jiffies;
2258 }
2259 
2260 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2261 					int vectors)
2262 {
2263 	int vector_threshold;
2264 
2265 	/* We'll want at least 2 (vector_threshold):
2266 	 * 1) TxQ[0] + RxQ[0] handler
2267 	 * 2) Other (Link Status Change, etc.)
2268 	 */
2269 	vector_threshold = MIN_MSIX_COUNT;
2270 
2271 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2272 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2273 	 * Right now, we simply care about how many we'll get; we'll
2274 	 * set them up later while requesting irq's.
2275 	 */
2276 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2277 					vector_threshold, vectors);
2278 
2279 	if (vectors < 0) {
2280 		dev_err(&adapter->pdev->dev,
2281 			"Unable to allocate MSI-X interrupts\n");
2282 		kfree(adapter->msix_entries);
2283 		adapter->msix_entries = NULL;
2284 		return vectors;
2285 	}
2286 
2287 	/* Adjust for only the vectors we'll use, which is minimum
2288 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2289 	 * vectors we were allocated.
2290 	 */
2291 	adapter->num_msix_vectors = vectors;
2292 
2293 	return 0;
2294 }
2295 
2296 /**
2297  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2298  * @adapter: board private structure to initialize
2299  *
2300  * This is the top level queue allocation routine.  The order here is very
2301  * important, starting with the "most" number of features turned on at once,
2302  * and ending with the smallest set of features.  This way large combinations
2303  * can be allocated if they're turned on, and smaller combinations are the
2304  * fallthrough conditions.
2305  *
2306  **/
2307 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2308 {
2309 	struct ixgbe_hw *hw = &adapter->hw;
2310 	unsigned int def_q = 0;
2311 	unsigned int num_tcs = 0;
2312 	int err;
2313 
2314 	/* Start with base case */
2315 	adapter->num_rx_queues = 1;
2316 	adapter->num_tx_queues = 1;
2317 
2318 	spin_lock_bh(&adapter->mbx_lock);
2319 
2320 	/* fetch queue configuration from the PF */
2321 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2322 
2323 	spin_unlock_bh(&adapter->mbx_lock);
2324 
2325 	if (err)
2326 		return;
2327 
2328 	/* we need as many queues as traffic classes */
2329 	if (num_tcs > 1) {
2330 		adapter->num_rx_queues = num_tcs;
2331 	} else {
2332 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2333 
2334 		switch (hw->api_version) {
2335 		case ixgbe_mbox_api_11:
2336 		case ixgbe_mbox_api_12:
2337 			adapter->num_rx_queues = rss;
2338 			adapter->num_tx_queues = rss;
2339 		default:
2340 			break;
2341 		}
2342 	}
2343 }
2344 
2345 /**
2346  * ixgbevf_alloc_queues - Allocate memory for all rings
2347  * @adapter: board private structure to initialize
2348  *
2349  * We allocate one ring per queue at run-time since we don't know the
2350  * number of queues at compile-time.  The polling_netdev array is
2351  * intended for Multiqueue, but should work fine with a single queue.
2352  **/
2353 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2354 {
2355 	struct ixgbevf_ring *ring;
2356 	int rx = 0, tx = 0;
2357 
2358 	for (; tx < adapter->num_tx_queues; tx++) {
2359 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2360 		if (!ring)
2361 			goto err_allocation;
2362 
2363 		ring->dev = &adapter->pdev->dev;
2364 		ring->netdev = adapter->netdev;
2365 		ring->count = adapter->tx_ring_count;
2366 		ring->queue_index = tx;
2367 		ring->reg_idx = tx;
2368 
2369 		adapter->tx_ring[tx] = ring;
2370 	}
2371 
2372 	for (; rx < adapter->num_rx_queues; rx++) {
2373 		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2374 		if (!ring)
2375 			goto err_allocation;
2376 
2377 		ring->dev = &adapter->pdev->dev;
2378 		ring->netdev = adapter->netdev;
2379 
2380 		ring->count = adapter->rx_ring_count;
2381 		ring->queue_index = rx;
2382 		ring->reg_idx = rx;
2383 
2384 		adapter->rx_ring[rx] = ring;
2385 	}
2386 
2387 	return 0;
2388 
2389 err_allocation:
2390 	while (tx) {
2391 		kfree(adapter->tx_ring[--tx]);
2392 		adapter->tx_ring[tx] = NULL;
2393 	}
2394 
2395 	while (rx) {
2396 		kfree(adapter->rx_ring[--rx]);
2397 		adapter->rx_ring[rx] = NULL;
2398 	}
2399 	return -ENOMEM;
2400 }
2401 
2402 /**
2403  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2404  * @adapter: board private structure to initialize
2405  *
2406  * Attempt to configure the interrupts using the best available
2407  * capabilities of the hardware and the kernel.
2408  **/
2409 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2410 {
2411 	struct net_device *netdev = adapter->netdev;
2412 	int err = 0;
2413 	int vector, v_budget;
2414 
2415 	/* It's easy to be greedy for MSI-X vectors, but it really
2416 	 * doesn't do us much good if we have a lot more vectors
2417 	 * than CPU's.  So let's be conservative and only ask for
2418 	 * (roughly) the same number of vectors as there are CPU's.
2419 	 * The default is to use pairs of vectors.
2420 	 */
2421 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2422 	v_budget = min_t(int, v_budget, num_online_cpus());
2423 	v_budget += NON_Q_VECTORS;
2424 
2425 	/* A failure in MSI-X entry allocation isn't fatal, but it does
2426 	 * mean we disable MSI-X capabilities of the adapter.
2427 	 */
2428 	adapter->msix_entries = kcalloc(v_budget,
2429 					sizeof(struct msix_entry), GFP_KERNEL);
2430 	if (!adapter->msix_entries) {
2431 		err = -ENOMEM;
2432 		goto out;
2433 	}
2434 
2435 	for (vector = 0; vector < v_budget; vector++)
2436 		adapter->msix_entries[vector].entry = vector;
2437 
2438 	err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2439 	if (err)
2440 		goto out;
2441 
2442 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2443 	if (err)
2444 		goto out;
2445 
2446 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2447 
2448 out:
2449 	return err;
2450 }
2451 
2452 /**
2453  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2454  * @adapter: board private structure to initialize
2455  *
2456  * We allocate one q_vector per queue interrupt.  If allocation fails we
2457  * return -ENOMEM.
2458  **/
2459 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2460 {
2461 	int q_idx, num_q_vectors;
2462 	struct ixgbevf_q_vector *q_vector;
2463 
2464 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2465 
2466 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2467 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2468 		if (!q_vector)
2469 			goto err_out;
2470 		q_vector->adapter = adapter;
2471 		q_vector->v_idx = q_idx;
2472 		netif_napi_add(adapter->netdev, &q_vector->napi,
2473 			       ixgbevf_poll, 64);
2474 #ifdef CONFIG_NET_RX_BUSY_POLL
2475 		napi_hash_add(&q_vector->napi);
2476 #endif
2477 		adapter->q_vector[q_idx] = q_vector;
2478 	}
2479 
2480 	return 0;
2481 
2482 err_out:
2483 	while (q_idx) {
2484 		q_idx--;
2485 		q_vector = adapter->q_vector[q_idx];
2486 #ifdef CONFIG_NET_RX_BUSY_POLL
2487 		napi_hash_del(&q_vector->napi);
2488 #endif
2489 		netif_napi_del(&q_vector->napi);
2490 		kfree(q_vector);
2491 		adapter->q_vector[q_idx] = NULL;
2492 	}
2493 	return -ENOMEM;
2494 }
2495 
2496 /**
2497  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2498  * @adapter: board private structure to initialize
2499  *
2500  * This function frees the memory allocated to the q_vectors.  In addition if
2501  * NAPI is enabled it will delete any references to the NAPI struct prior
2502  * to freeing the q_vector.
2503  **/
2504 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2505 {
2506 	int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2507 
2508 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2509 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2510 
2511 		adapter->q_vector[q_idx] = NULL;
2512 #ifdef CONFIG_NET_RX_BUSY_POLL
2513 		napi_hash_del(&q_vector->napi);
2514 #endif
2515 		netif_napi_del(&q_vector->napi);
2516 		kfree(q_vector);
2517 	}
2518 }
2519 
2520 /**
2521  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2522  * @adapter: board private structure
2523  *
2524  **/
2525 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2526 {
2527 	pci_disable_msix(adapter->pdev);
2528 	kfree(adapter->msix_entries);
2529 	adapter->msix_entries = NULL;
2530 }
2531 
2532 /**
2533  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2534  * @adapter: board private structure to initialize
2535  *
2536  **/
2537 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2538 {
2539 	int err;
2540 
2541 	/* Number of supported queues */
2542 	ixgbevf_set_num_queues(adapter);
2543 
2544 	err = ixgbevf_set_interrupt_capability(adapter);
2545 	if (err) {
2546 		hw_dbg(&adapter->hw,
2547 		       "Unable to setup interrupt capabilities\n");
2548 		goto err_set_interrupt;
2549 	}
2550 
2551 	err = ixgbevf_alloc_q_vectors(adapter);
2552 	if (err) {
2553 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2554 		goto err_alloc_q_vectors;
2555 	}
2556 
2557 	err = ixgbevf_alloc_queues(adapter);
2558 	if (err) {
2559 		pr_err("Unable to allocate memory for queues\n");
2560 		goto err_alloc_queues;
2561 	}
2562 
2563 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2564 	       (adapter->num_rx_queues > 1) ? "Enabled" :
2565 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2566 
2567 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2568 
2569 	return 0;
2570 err_alloc_queues:
2571 	ixgbevf_free_q_vectors(adapter);
2572 err_alloc_q_vectors:
2573 	ixgbevf_reset_interrupt_capability(adapter);
2574 err_set_interrupt:
2575 	return err;
2576 }
2577 
2578 /**
2579  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2580  * @adapter: board private structure to clear interrupt scheme on
2581  *
2582  * We go through and clear interrupt specific resources and reset the structure
2583  * to pre-load conditions
2584  **/
2585 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2586 {
2587 	int i;
2588 
2589 	for (i = 0; i < adapter->num_tx_queues; i++) {
2590 		kfree(adapter->tx_ring[i]);
2591 		adapter->tx_ring[i] = NULL;
2592 	}
2593 	for (i = 0; i < adapter->num_rx_queues; i++) {
2594 		kfree(adapter->rx_ring[i]);
2595 		adapter->rx_ring[i] = NULL;
2596 	}
2597 
2598 	adapter->num_tx_queues = 0;
2599 	adapter->num_rx_queues = 0;
2600 
2601 	ixgbevf_free_q_vectors(adapter);
2602 	ixgbevf_reset_interrupt_capability(adapter);
2603 }
2604 
2605 /**
2606  * ixgbevf_sw_init - Initialize general software structures
2607  * @adapter: board private structure to initialize
2608  *
2609  * ixgbevf_sw_init initializes the Adapter private data structure.
2610  * Fields are initialized based on PCI device information and
2611  * OS network device settings (MTU size).
2612  **/
2613 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2614 {
2615 	struct ixgbe_hw *hw = &adapter->hw;
2616 	struct pci_dev *pdev = adapter->pdev;
2617 	struct net_device *netdev = adapter->netdev;
2618 	int err;
2619 
2620 	/* PCI config space info */
2621 	hw->vendor_id = pdev->vendor;
2622 	hw->device_id = pdev->device;
2623 	hw->revision_id = pdev->revision;
2624 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2625 	hw->subsystem_device_id = pdev->subsystem_device;
2626 
2627 	hw->mbx.ops.init_params(hw);
2628 
2629 	/* assume legacy case in which PF would only give VF 2 queues */
2630 	hw->mac.max_tx_queues = 2;
2631 	hw->mac.max_rx_queues = 2;
2632 
2633 	/* lock to protect mailbox accesses */
2634 	spin_lock_init(&adapter->mbx_lock);
2635 
2636 	err = hw->mac.ops.reset_hw(hw);
2637 	if (err) {
2638 		dev_info(&pdev->dev,
2639 			 "PF still in reset state.  Is the PF interface up?\n");
2640 	} else {
2641 		err = hw->mac.ops.init_hw(hw);
2642 		if (err) {
2643 			pr_err("init_shared_code failed: %d\n", err);
2644 			goto out;
2645 		}
2646 		ixgbevf_negotiate_api(adapter);
2647 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2648 		if (err)
2649 			dev_info(&pdev->dev, "Error reading MAC address\n");
2650 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2651 			dev_info(&pdev->dev,
2652 				 "MAC address not assigned by administrator.\n");
2653 		memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2654 	}
2655 
2656 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2657 		dev_info(&pdev->dev, "Assigning random MAC address\n");
2658 		eth_hw_addr_random(netdev);
2659 		memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2660 	}
2661 
2662 	/* Enable dynamic interrupt throttling rates */
2663 	adapter->rx_itr_setting = 1;
2664 	adapter->tx_itr_setting = 1;
2665 
2666 	/* set default ring sizes */
2667 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2668 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2669 
2670 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2671 	return 0;
2672 
2673 out:
2674 	return err;
2675 }
2676 
2677 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
2678 	{							\
2679 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
2680 		if (current_counter < last_counter)		\
2681 			counter += 0x100000000LL;		\
2682 		last_counter = current_counter;			\
2683 		counter &= 0xFFFFFFFF00000000LL;		\
2684 		counter |= current_counter;			\
2685 	}
2686 
2687 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2688 	{								 \
2689 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
2690 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
2691 		u64 current_counter = (current_counter_msb << 32) |	 \
2692 			current_counter_lsb;				 \
2693 		if (current_counter < last_counter)			 \
2694 			counter += 0x1000000000LL;			 \
2695 		last_counter = current_counter;				 \
2696 		counter &= 0xFFFFFFF000000000LL;			 \
2697 		counter |= current_counter;				 \
2698 	}
2699 /**
2700  * ixgbevf_update_stats - Update the board statistics counters.
2701  * @adapter: board private structure
2702  **/
2703 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2704 {
2705 	struct ixgbe_hw *hw = &adapter->hw;
2706 	int i;
2707 
2708 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2709 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2710 		return;
2711 
2712 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2713 				adapter->stats.vfgprc);
2714 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2715 				adapter->stats.vfgptc);
2716 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2717 				adapter->stats.last_vfgorc,
2718 				adapter->stats.vfgorc);
2719 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2720 				adapter->stats.last_vfgotc,
2721 				adapter->stats.vfgotc);
2722 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2723 				adapter->stats.vfmprc);
2724 
2725 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2726 		adapter->hw_csum_rx_error +=
2727 			adapter->rx_ring[i]->hw_csum_rx_error;
2728 		adapter->rx_ring[i]->hw_csum_rx_error = 0;
2729 	}
2730 }
2731 
2732 /**
2733  * ixgbevf_service_timer - Timer Call-back
2734  * @data: pointer to adapter cast into an unsigned long
2735  **/
2736 static void ixgbevf_service_timer(unsigned long data)
2737 {
2738 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2739 
2740 	/* Reset the timer */
2741 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2742 
2743 	ixgbevf_service_event_schedule(adapter);
2744 }
2745 
2746 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2747 {
2748 	if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2749 		return;
2750 
2751 	adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2752 
2753 	/* If we're already down or resetting, just bail */
2754 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2755 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2756 		return;
2757 
2758 	adapter->tx_timeout_count++;
2759 
2760 	ixgbevf_reinit_locked(adapter);
2761 }
2762 
2763 /**
2764  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2765  * @adapter: pointer to the device adapter structure
2766  *
2767  * This function serves two purposes.  First it strobes the interrupt lines
2768  * in order to make certain interrupts are occurring.  Secondly it sets the
2769  * bits needed to check for TX hangs.  As a result we should immediately
2770  * determine if a hang has occurred.
2771  **/
2772 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2773 {
2774 	struct ixgbe_hw *hw = &adapter->hw;
2775 	u32 eics = 0;
2776 	int i;
2777 
2778 	/* If we're down or resetting, just bail */
2779 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2780 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2781 		return;
2782 
2783 	/* Force detection of hung controller */
2784 	if (netif_carrier_ok(adapter->netdev)) {
2785 		for (i = 0; i < adapter->num_tx_queues; i++)
2786 			set_check_for_tx_hang(adapter->tx_ring[i]);
2787 	}
2788 
2789 	/* get one bit for every active Tx/Rx interrupt vector */
2790 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2791 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2792 
2793 		if (qv->rx.ring || qv->tx.ring)
2794 			eics |= 1 << i;
2795 	}
2796 
2797 	/* Cause software interrupt to ensure rings are cleaned */
2798 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2799 }
2800 
2801 /**
2802  * ixgbevf_watchdog_update_link - update the link status
2803  * @adapter: pointer to the device adapter structure
2804  **/
2805 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2806 {
2807 	struct ixgbe_hw *hw = &adapter->hw;
2808 	u32 link_speed = adapter->link_speed;
2809 	bool link_up = adapter->link_up;
2810 	s32 err;
2811 
2812 	spin_lock_bh(&adapter->mbx_lock);
2813 
2814 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2815 
2816 	spin_unlock_bh(&adapter->mbx_lock);
2817 
2818 	/* if check for link returns error we will need to reset */
2819 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2820 		adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2821 		link_up = false;
2822 	}
2823 
2824 	adapter->link_up = link_up;
2825 	adapter->link_speed = link_speed;
2826 }
2827 
2828 /**
2829  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2830  *				 print link up message
2831  * @adapter: pointer to the device adapter structure
2832  **/
2833 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2834 {
2835 	struct net_device *netdev = adapter->netdev;
2836 
2837 	/* only continue if link was previously down */
2838 	if (netif_carrier_ok(netdev))
2839 		return;
2840 
2841 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2842 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2843 		 "10 Gbps" :
2844 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2845 		 "1 Gbps" :
2846 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2847 		 "100 Mbps" :
2848 		 "unknown speed");
2849 
2850 	netif_carrier_on(netdev);
2851 }
2852 
2853 /**
2854  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2855  *				   print link down message
2856  * @adapter: pointer to the adapter structure
2857  **/
2858 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2859 {
2860 	struct net_device *netdev = adapter->netdev;
2861 
2862 	adapter->link_speed = 0;
2863 
2864 	/* only continue if link was up previously */
2865 	if (!netif_carrier_ok(netdev))
2866 		return;
2867 
2868 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2869 
2870 	netif_carrier_off(netdev);
2871 }
2872 
2873 /**
2874  * ixgbevf_watchdog_subtask - worker thread to bring link up
2875  * @work: pointer to work_struct containing our data
2876  **/
2877 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2878 {
2879 	/* if interface is down do nothing */
2880 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2881 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2882 		return;
2883 
2884 	ixgbevf_watchdog_update_link(adapter);
2885 
2886 	if (adapter->link_up)
2887 		ixgbevf_watchdog_link_is_up(adapter);
2888 	else
2889 		ixgbevf_watchdog_link_is_down(adapter);
2890 
2891 	ixgbevf_update_stats(adapter);
2892 }
2893 
2894 /**
2895  * ixgbevf_service_task - manages and runs subtasks
2896  * @work: pointer to work_struct containing our data
2897  **/
2898 static void ixgbevf_service_task(struct work_struct *work)
2899 {
2900 	struct ixgbevf_adapter *adapter = container_of(work,
2901 						       struct ixgbevf_adapter,
2902 						       service_task);
2903 	struct ixgbe_hw *hw = &adapter->hw;
2904 
2905 	if (IXGBE_REMOVED(hw->hw_addr)) {
2906 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2907 			rtnl_lock();
2908 			ixgbevf_down(adapter);
2909 			rtnl_unlock();
2910 		}
2911 		return;
2912 	}
2913 
2914 	ixgbevf_queue_reset_subtask(adapter);
2915 	ixgbevf_reset_subtask(adapter);
2916 	ixgbevf_watchdog_subtask(adapter);
2917 	ixgbevf_check_hang_subtask(adapter);
2918 
2919 	ixgbevf_service_event_complete(adapter);
2920 }
2921 
2922 /**
2923  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2924  * @tx_ring: Tx descriptor ring for a specific queue
2925  *
2926  * Free all transmit software resources
2927  **/
2928 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2929 {
2930 	ixgbevf_clean_tx_ring(tx_ring);
2931 
2932 	vfree(tx_ring->tx_buffer_info);
2933 	tx_ring->tx_buffer_info = NULL;
2934 
2935 	/* if not set, then don't free */
2936 	if (!tx_ring->desc)
2937 		return;
2938 
2939 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2940 			  tx_ring->dma);
2941 
2942 	tx_ring->desc = NULL;
2943 }
2944 
2945 /**
2946  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2947  * @adapter: board private structure
2948  *
2949  * Free all transmit software resources
2950  **/
2951 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2952 {
2953 	int i;
2954 
2955 	for (i = 0; i < adapter->num_tx_queues; i++)
2956 		if (adapter->tx_ring[i]->desc)
2957 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2958 }
2959 
2960 /**
2961  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2962  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2963  *
2964  * Return 0 on success, negative on failure
2965  **/
2966 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2967 {
2968 	int size;
2969 
2970 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2971 	tx_ring->tx_buffer_info = vzalloc(size);
2972 	if (!tx_ring->tx_buffer_info)
2973 		goto err;
2974 
2975 	/* round up to nearest 4K */
2976 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2977 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2978 
2979 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2980 					   &tx_ring->dma, GFP_KERNEL);
2981 	if (!tx_ring->desc)
2982 		goto err;
2983 
2984 	return 0;
2985 
2986 err:
2987 	vfree(tx_ring->tx_buffer_info);
2988 	tx_ring->tx_buffer_info = NULL;
2989 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
2990 	return -ENOMEM;
2991 }
2992 
2993 /**
2994  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2995  * @adapter: board private structure
2996  *
2997  * If this function returns with an error, then it's possible one or
2998  * more of the rings is populated (while the rest are not).  It is the
2999  * callers duty to clean those orphaned rings.
3000  *
3001  * Return 0 on success, negative on failure
3002  **/
3003 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3004 {
3005 	int i, err = 0;
3006 
3007 	for (i = 0; i < adapter->num_tx_queues; i++) {
3008 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3009 		if (!err)
3010 			continue;
3011 		hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3012 		break;
3013 	}
3014 
3015 	return err;
3016 }
3017 
3018 /**
3019  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3020  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3021  *
3022  * Returns 0 on success, negative on failure
3023  **/
3024 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3025 {
3026 	int size;
3027 
3028 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3029 	rx_ring->rx_buffer_info = vzalloc(size);
3030 	if (!rx_ring->rx_buffer_info)
3031 		goto err;
3032 
3033 	/* Round up to nearest 4K */
3034 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3035 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3036 
3037 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3038 					   &rx_ring->dma, GFP_KERNEL);
3039 
3040 	if (!rx_ring->desc)
3041 		goto err;
3042 
3043 	return 0;
3044 err:
3045 	vfree(rx_ring->rx_buffer_info);
3046 	rx_ring->rx_buffer_info = NULL;
3047 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3048 	return -ENOMEM;
3049 }
3050 
3051 /**
3052  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3053  * @adapter: board private structure
3054  *
3055  * If this function returns with an error, then it's possible one or
3056  * more of the rings is populated (while the rest are not).  It is the
3057  * callers duty to clean those orphaned rings.
3058  *
3059  * Return 0 on success, negative on failure
3060  **/
3061 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3062 {
3063 	int i, err = 0;
3064 
3065 	for (i = 0; i < adapter->num_rx_queues; i++) {
3066 		err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3067 		if (!err)
3068 			continue;
3069 		hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3070 		break;
3071 	}
3072 	return err;
3073 }
3074 
3075 /**
3076  * ixgbevf_free_rx_resources - Free Rx Resources
3077  * @rx_ring: ring to clean the resources from
3078  *
3079  * Free all receive software resources
3080  **/
3081 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3082 {
3083 	ixgbevf_clean_rx_ring(rx_ring);
3084 
3085 	vfree(rx_ring->rx_buffer_info);
3086 	rx_ring->rx_buffer_info = NULL;
3087 
3088 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3089 			  rx_ring->dma);
3090 
3091 	rx_ring->desc = NULL;
3092 }
3093 
3094 /**
3095  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3096  * @adapter: board private structure
3097  *
3098  * Free all receive software resources
3099  **/
3100 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3101 {
3102 	int i;
3103 
3104 	for (i = 0; i < adapter->num_rx_queues; i++)
3105 		if (adapter->rx_ring[i]->desc)
3106 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3107 }
3108 
3109 /**
3110  * ixgbevf_open - Called when a network interface is made active
3111  * @netdev: network interface device structure
3112  *
3113  * Returns 0 on success, negative value on failure
3114  *
3115  * The open entry point is called when a network interface is made
3116  * active by the system (IFF_UP).  At this point all resources needed
3117  * for transmit and receive operations are allocated, the interrupt
3118  * handler is registered with the OS, the watchdog timer is started,
3119  * and the stack is notified that the interface is ready.
3120  **/
3121 static int ixgbevf_open(struct net_device *netdev)
3122 {
3123 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3124 	struct ixgbe_hw *hw = &adapter->hw;
3125 	int err;
3126 
3127 	/* A previous failure to open the device because of a lack of
3128 	 * available MSIX vector resources may have reset the number
3129 	 * of msix vectors variable to zero.  The only way to recover
3130 	 * is to unload/reload the driver and hope that the system has
3131 	 * been able to recover some MSIX vector resources.
3132 	 */
3133 	if (!adapter->num_msix_vectors)
3134 		return -ENOMEM;
3135 
3136 	if (hw->adapter_stopped) {
3137 		ixgbevf_reset(adapter);
3138 		/* if adapter is still stopped then PF isn't up and
3139 		 * the VF can't start.
3140 		 */
3141 		if (hw->adapter_stopped) {
3142 			err = IXGBE_ERR_MBX;
3143 			pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3144 			goto err_setup_reset;
3145 		}
3146 	}
3147 
3148 	/* disallow open during test */
3149 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3150 		return -EBUSY;
3151 
3152 	netif_carrier_off(netdev);
3153 
3154 	/* allocate transmit descriptors */
3155 	err = ixgbevf_setup_all_tx_resources(adapter);
3156 	if (err)
3157 		goto err_setup_tx;
3158 
3159 	/* allocate receive descriptors */
3160 	err = ixgbevf_setup_all_rx_resources(adapter);
3161 	if (err)
3162 		goto err_setup_rx;
3163 
3164 	ixgbevf_configure(adapter);
3165 
3166 	/* Map the Tx/Rx rings to the vectors we were allotted.
3167 	 * if request_irq will be called in this function map_rings
3168 	 * must be called *before* up_complete
3169 	 */
3170 	ixgbevf_map_rings_to_vectors(adapter);
3171 
3172 	err = ixgbevf_request_irq(adapter);
3173 	if (err)
3174 		goto err_req_irq;
3175 
3176 	ixgbevf_up_complete(adapter);
3177 
3178 	return 0;
3179 
3180 err_req_irq:
3181 	ixgbevf_down(adapter);
3182 err_setup_rx:
3183 	ixgbevf_free_all_rx_resources(adapter);
3184 err_setup_tx:
3185 	ixgbevf_free_all_tx_resources(adapter);
3186 	ixgbevf_reset(adapter);
3187 
3188 err_setup_reset:
3189 
3190 	return err;
3191 }
3192 
3193 /**
3194  * ixgbevf_close - Disables a network interface
3195  * @netdev: network interface device structure
3196  *
3197  * Returns 0, this is not allowed to fail
3198  *
3199  * The close entry point is called when an interface is de-activated
3200  * by the OS.  The hardware is still under the drivers control, but
3201  * needs to be disabled.  A global MAC reset is issued to stop the
3202  * hardware, and all transmit and receive resources are freed.
3203  **/
3204 static int ixgbevf_close(struct net_device *netdev)
3205 {
3206 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3207 
3208 	ixgbevf_down(adapter);
3209 	ixgbevf_free_irq(adapter);
3210 
3211 	ixgbevf_free_all_tx_resources(adapter);
3212 	ixgbevf_free_all_rx_resources(adapter);
3213 
3214 	return 0;
3215 }
3216 
3217 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3218 {
3219 	struct net_device *dev = adapter->netdev;
3220 
3221 	if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3222 		return;
3223 
3224 	adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3225 
3226 	/* if interface is down do nothing */
3227 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3228 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3229 		return;
3230 
3231 	/* Hardware has to reinitialize queues and interrupts to
3232 	 * match packet buffer alignment. Unfortunately, the
3233 	 * hardware is not flexible enough to do this dynamically.
3234 	 */
3235 	if (netif_running(dev))
3236 		ixgbevf_close(dev);
3237 
3238 	ixgbevf_clear_interrupt_scheme(adapter);
3239 	ixgbevf_init_interrupt_scheme(adapter);
3240 
3241 	if (netif_running(dev))
3242 		ixgbevf_open(dev);
3243 }
3244 
3245 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3246 				u32 vlan_macip_lens, u32 type_tucmd,
3247 				u32 mss_l4len_idx)
3248 {
3249 	struct ixgbe_adv_tx_context_desc *context_desc;
3250 	u16 i = tx_ring->next_to_use;
3251 
3252 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3253 
3254 	i++;
3255 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3256 
3257 	/* set bits to identify this as an advanced context descriptor */
3258 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3259 
3260 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3261 	context_desc->seqnum_seed	= 0;
3262 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3263 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3264 }
3265 
3266 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3267 		       struct ixgbevf_tx_buffer *first,
3268 		       u8 *hdr_len)
3269 {
3270 	struct sk_buff *skb = first->skb;
3271 	u32 vlan_macip_lens, type_tucmd;
3272 	u32 mss_l4len_idx, l4len;
3273 	int err;
3274 
3275 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3276 		return 0;
3277 
3278 	if (!skb_is_gso(skb))
3279 		return 0;
3280 
3281 	err = skb_cow_head(skb, 0);
3282 	if (err < 0)
3283 		return err;
3284 
3285 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3286 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3287 
3288 	if (first->protocol == htons(ETH_P_IP)) {
3289 		struct iphdr *iph = ip_hdr(skb);
3290 
3291 		iph->tot_len = 0;
3292 		iph->check = 0;
3293 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3294 							 iph->daddr, 0,
3295 							 IPPROTO_TCP,
3296 							 0);
3297 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3298 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3299 				   IXGBE_TX_FLAGS_CSUM |
3300 				   IXGBE_TX_FLAGS_IPV4;
3301 	} else if (skb_is_gso_v6(skb)) {
3302 		ipv6_hdr(skb)->payload_len = 0;
3303 		tcp_hdr(skb)->check =
3304 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3305 				     &ipv6_hdr(skb)->daddr,
3306 				     0, IPPROTO_TCP, 0);
3307 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3308 				   IXGBE_TX_FLAGS_CSUM;
3309 	}
3310 
3311 	/* compute header lengths */
3312 	l4len = tcp_hdrlen(skb);
3313 	*hdr_len += l4len;
3314 	*hdr_len = skb_transport_offset(skb) + l4len;
3315 
3316 	/* update GSO size and bytecount with header size */
3317 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3318 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3319 
3320 	/* mss_l4len_id: use 1 as index for TSO */
3321 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3322 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3323 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3324 
3325 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3326 	vlan_macip_lens = skb_network_header_len(skb);
3327 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3328 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3329 
3330 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3331 			    type_tucmd, mss_l4len_idx);
3332 
3333 	return 1;
3334 }
3335 
3336 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3337 			    struct ixgbevf_tx_buffer *first)
3338 {
3339 	struct sk_buff *skb = first->skb;
3340 	u32 vlan_macip_lens = 0;
3341 	u32 mss_l4len_idx = 0;
3342 	u32 type_tucmd = 0;
3343 
3344 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3345 		u8 l4_hdr = 0;
3346 
3347 		switch (first->protocol) {
3348 		case htons(ETH_P_IP):
3349 			vlan_macip_lens |= skb_network_header_len(skb);
3350 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3351 			l4_hdr = ip_hdr(skb)->protocol;
3352 			break;
3353 		case htons(ETH_P_IPV6):
3354 			vlan_macip_lens |= skb_network_header_len(skb);
3355 			l4_hdr = ipv6_hdr(skb)->nexthdr;
3356 			break;
3357 		default:
3358 			if (unlikely(net_ratelimit())) {
3359 				dev_warn(tx_ring->dev,
3360 					 "partial checksum but proto=%x!\n",
3361 					 first->protocol);
3362 			}
3363 			break;
3364 		}
3365 
3366 		switch (l4_hdr) {
3367 		case IPPROTO_TCP:
3368 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3369 			mss_l4len_idx = tcp_hdrlen(skb) <<
3370 					IXGBE_ADVTXD_L4LEN_SHIFT;
3371 			break;
3372 		case IPPROTO_SCTP:
3373 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3374 			mss_l4len_idx = sizeof(struct sctphdr) <<
3375 					IXGBE_ADVTXD_L4LEN_SHIFT;
3376 			break;
3377 		case IPPROTO_UDP:
3378 			mss_l4len_idx = sizeof(struct udphdr) <<
3379 					IXGBE_ADVTXD_L4LEN_SHIFT;
3380 			break;
3381 		default:
3382 			if (unlikely(net_ratelimit())) {
3383 				dev_warn(tx_ring->dev,
3384 					 "partial checksum but l4 proto=%x!\n",
3385 					 l4_hdr);
3386 			}
3387 			break;
3388 		}
3389 
3390 		/* update TX checksum flag */
3391 		first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3392 	}
3393 
3394 	/* vlan_macip_lens: MACLEN, VLAN tag */
3395 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3396 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3397 
3398 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3399 			    type_tucmd, mss_l4len_idx);
3400 }
3401 
3402 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3403 {
3404 	/* set type for advanced descriptor with frame checksum insertion */
3405 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3406 				      IXGBE_ADVTXD_DCMD_IFCS |
3407 				      IXGBE_ADVTXD_DCMD_DEXT);
3408 
3409 	/* set HW VLAN bit if VLAN is present */
3410 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3411 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3412 
3413 	/* set segmentation enable bits for TSO/FSO */
3414 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3415 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3416 
3417 	return cmd_type;
3418 }
3419 
3420 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3421 				     u32 tx_flags, unsigned int paylen)
3422 {
3423 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3424 
3425 	/* enable L4 checksum for TSO and TX checksum offload */
3426 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3427 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3428 
3429 	/* enble IPv4 checksum for TSO */
3430 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3431 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3432 
3433 	/* use index 1 context for TSO/FSO/FCOE */
3434 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3435 		olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3436 
3437 	/* Check Context must be set if Tx switch is enabled, which it
3438 	 * always is for case where virtual functions are running
3439 	 */
3440 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3441 
3442 	tx_desc->read.olinfo_status = olinfo_status;
3443 }
3444 
3445 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3446 			   struct ixgbevf_tx_buffer *first,
3447 			   const u8 hdr_len)
3448 {
3449 	dma_addr_t dma;
3450 	struct sk_buff *skb = first->skb;
3451 	struct ixgbevf_tx_buffer *tx_buffer;
3452 	union ixgbe_adv_tx_desc *tx_desc;
3453 	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3454 	unsigned int data_len = skb->data_len;
3455 	unsigned int size = skb_headlen(skb);
3456 	unsigned int paylen = skb->len - hdr_len;
3457 	u32 tx_flags = first->tx_flags;
3458 	__le32 cmd_type;
3459 	u16 i = tx_ring->next_to_use;
3460 
3461 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3462 
3463 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3464 	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3465 
3466 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3467 	if (dma_mapping_error(tx_ring->dev, dma))
3468 		goto dma_error;
3469 
3470 	/* record length, and DMA address */
3471 	dma_unmap_len_set(first, len, size);
3472 	dma_unmap_addr_set(first, dma, dma);
3473 
3474 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
3475 
3476 	for (;;) {
3477 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3478 			tx_desc->read.cmd_type_len =
3479 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3480 
3481 			i++;
3482 			tx_desc++;
3483 			if (i == tx_ring->count) {
3484 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3485 				i = 0;
3486 			}
3487 
3488 			dma += IXGBE_MAX_DATA_PER_TXD;
3489 			size -= IXGBE_MAX_DATA_PER_TXD;
3490 
3491 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3492 			tx_desc->read.olinfo_status = 0;
3493 		}
3494 
3495 		if (likely(!data_len))
3496 			break;
3497 
3498 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3499 
3500 		i++;
3501 		tx_desc++;
3502 		if (i == tx_ring->count) {
3503 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3504 			i = 0;
3505 		}
3506 
3507 		size = skb_frag_size(frag);
3508 		data_len -= size;
3509 
3510 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3511 				       DMA_TO_DEVICE);
3512 		if (dma_mapping_error(tx_ring->dev, dma))
3513 			goto dma_error;
3514 
3515 		tx_buffer = &tx_ring->tx_buffer_info[i];
3516 		dma_unmap_len_set(tx_buffer, len, size);
3517 		dma_unmap_addr_set(tx_buffer, dma, dma);
3518 
3519 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3520 		tx_desc->read.olinfo_status = 0;
3521 
3522 		frag++;
3523 	}
3524 
3525 	/* write last descriptor with RS and EOP bits */
3526 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3527 	tx_desc->read.cmd_type_len = cmd_type;
3528 
3529 	/* set the timestamp */
3530 	first->time_stamp = jiffies;
3531 
3532 	/* Force memory writes to complete before letting h/w know there
3533 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
3534 	 * memory model archs, such as IA-64).
3535 	 *
3536 	 * We also need this memory barrier (wmb) to make certain all of the
3537 	 * status bits have been updated before next_to_watch is written.
3538 	 */
3539 	wmb();
3540 
3541 	/* set next_to_watch value indicating a packet is present */
3542 	first->next_to_watch = tx_desc;
3543 
3544 	i++;
3545 	if (i == tx_ring->count)
3546 		i = 0;
3547 
3548 	tx_ring->next_to_use = i;
3549 
3550 	/* notify HW of packet */
3551 	ixgbevf_write_tail(tx_ring, i);
3552 
3553 	return;
3554 dma_error:
3555 	dev_err(tx_ring->dev, "TX DMA map failed\n");
3556 
3557 	/* clear dma mappings for failed tx_buffer_info map */
3558 	for (;;) {
3559 		tx_buffer = &tx_ring->tx_buffer_info[i];
3560 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3561 		if (tx_buffer == first)
3562 			break;
3563 		if (i == 0)
3564 			i = tx_ring->count;
3565 		i--;
3566 	}
3567 
3568 	tx_ring->next_to_use = i;
3569 }
3570 
3571 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3572 {
3573 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3574 	/* Herbert's original patch had:
3575 	 *  smp_mb__after_netif_stop_queue();
3576 	 * but since that doesn't exist yet, just open code it.
3577 	 */
3578 	smp_mb();
3579 
3580 	/* We need to check again in a case another CPU has just
3581 	 * made room available.
3582 	 */
3583 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
3584 		return -EBUSY;
3585 
3586 	/* A reprieve! - use start_queue because it doesn't call schedule */
3587 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3588 	++tx_ring->tx_stats.restart_queue;
3589 
3590 	return 0;
3591 }
3592 
3593 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3594 {
3595 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3596 		return 0;
3597 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
3598 }
3599 
3600 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3601 {
3602 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3603 	struct ixgbevf_tx_buffer *first;
3604 	struct ixgbevf_ring *tx_ring;
3605 	int tso;
3606 	u32 tx_flags = 0;
3607 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
3608 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3609 	unsigned short f;
3610 #endif
3611 	u8 hdr_len = 0;
3612 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3613 
3614 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3615 		dev_kfree_skb_any(skb);
3616 		return NETDEV_TX_OK;
3617 	}
3618 
3619 	tx_ring = adapter->tx_ring[skb->queue_mapping];
3620 
3621 	/* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3622 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3623 	 *       + 2 desc gap to keep tail from touching head,
3624 	 *       + 1 desc for context descriptor,
3625 	 * otherwise try next time
3626 	 */
3627 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3628 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3629 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3630 #else
3631 	count += skb_shinfo(skb)->nr_frags;
3632 #endif
3633 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3634 		tx_ring->tx_stats.tx_busy++;
3635 		return NETDEV_TX_BUSY;
3636 	}
3637 
3638 	/* record the location of the first descriptor for this packet */
3639 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3640 	first->skb = skb;
3641 	first->bytecount = skb->len;
3642 	first->gso_segs = 1;
3643 
3644 	if (skb_vlan_tag_present(skb)) {
3645 		tx_flags |= skb_vlan_tag_get(skb);
3646 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3647 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
3648 	}
3649 
3650 	/* record initial flags and protocol */
3651 	first->tx_flags = tx_flags;
3652 	first->protocol = vlan_get_protocol(skb);
3653 
3654 	tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3655 	if (tso < 0)
3656 		goto out_drop;
3657 	else if (!tso)
3658 		ixgbevf_tx_csum(tx_ring, first);
3659 
3660 	ixgbevf_tx_map(tx_ring, first, hdr_len);
3661 
3662 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3663 
3664 	return NETDEV_TX_OK;
3665 
3666 out_drop:
3667 	dev_kfree_skb_any(first->skb);
3668 	first->skb = NULL;
3669 
3670 	return NETDEV_TX_OK;
3671 }
3672 
3673 /**
3674  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3675  * @netdev: network interface device structure
3676  * @p: pointer to an address structure
3677  *
3678  * Returns 0 on success, negative on failure
3679  **/
3680 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3681 {
3682 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3683 	struct ixgbe_hw *hw = &adapter->hw;
3684 	struct sockaddr *addr = p;
3685 
3686 	if (!is_valid_ether_addr(addr->sa_data))
3687 		return -EADDRNOTAVAIL;
3688 
3689 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3690 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3691 
3692 	spin_lock_bh(&adapter->mbx_lock);
3693 
3694 	hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3695 
3696 	spin_unlock_bh(&adapter->mbx_lock);
3697 
3698 	return 0;
3699 }
3700 
3701 /**
3702  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3703  * @netdev: network interface device structure
3704  * @new_mtu: new value for maximum frame size
3705  *
3706  * Returns 0 on success, negative on failure
3707  **/
3708 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3709 {
3710 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3711 	struct ixgbe_hw *hw = &adapter->hw;
3712 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3713 	int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3714 
3715 	switch (adapter->hw.api_version) {
3716 	case ixgbe_mbox_api_11:
3717 	case ixgbe_mbox_api_12:
3718 		max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3719 		break;
3720 	default:
3721 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3722 			max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3723 		break;
3724 	}
3725 
3726 	/* MTU < 68 is an error and causes problems on some kernels */
3727 	if ((new_mtu < 68) || (max_frame > max_possible_frame))
3728 		return -EINVAL;
3729 
3730 	hw_dbg(hw, "changing MTU from %d to %d\n",
3731 	       netdev->mtu, new_mtu);
3732 	/* must set new MTU before calling down or up */
3733 	netdev->mtu = new_mtu;
3734 
3735 	/* notify the PF of our intent to use this size of frame */
3736 	ixgbevf_rlpml_set_vf(hw, max_frame);
3737 
3738 	return 0;
3739 }
3740 
3741 #ifdef CONFIG_NET_POLL_CONTROLLER
3742 /* Polling 'interrupt' - used by things like netconsole to send skbs
3743  * without having to re-enable interrupts. It's not called while
3744  * the interrupt routine is executing.
3745  */
3746 static void ixgbevf_netpoll(struct net_device *netdev)
3747 {
3748 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3749 	int i;
3750 
3751 	/* if interface is down do nothing */
3752 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3753 		return;
3754 	for (i = 0; i < adapter->num_rx_queues; i++)
3755 		ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3756 }
3757 #endif /* CONFIG_NET_POLL_CONTROLLER */
3758 
3759 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3760 {
3761 	struct net_device *netdev = pci_get_drvdata(pdev);
3762 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3763 #ifdef CONFIG_PM
3764 	int retval = 0;
3765 #endif
3766 
3767 	netif_device_detach(netdev);
3768 
3769 	if (netif_running(netdev)) {
3770 		rtnl_lock();
3771 		ixgbevf_down(adapter);
3772 		ixgbevf_free_irq(adapter);
3773 		ixgbevf_free_all_tx_resources(adapter);
3774 		ixgbevf_free_all_rx_resources(adapter);
3775 		rtnl_unlock();
3776 	}
3777 
3778 	ixgbevf_clear_interrupt_scheme(adapter);
3779 
3780 #ifdef CONFIG_PM
3781 	retval = pci_save_state(pdev);
3782 	if (retval)
3783 		return retval;
3784 
3785 #endif
3786 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3787 		pci_disable_device(pdev);
3788 
3789 	return 0;
3790 }
3791 
3792 #ifdef CONFIG_PM
3793 static int ixgbevf_resume(struct pci_dev *pdev)
3794 {
3795 	struct net_device *netdev = pci_get_drvdata(pdev);
3796 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3797 	u32 err;
3798 
3799 	pci_restore_state(pdev);
3800 	/* pci_restore_state clears dev->state_saved so call
3801 	 * pci_save_state to restore it.
3802 	 */
3803 	pci_save_state(pdev);
3804 
3805 	err = pci_enable_device_mem(pdev);
3806 	if (err) {
3807 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3808 		return err;
3809 	}
3810 	smp_mb__before_atomic();
3811 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3812 	pci_set_master(pdev);
3813 
3814 	ixgbevf_reset(adapter);
3815 
3816 	rtnl_lock();
3817 	err = ixgbevf_init_interrupt_scheme(adapter);
3818 	rtnl_unlock();
3819 	if (err) {
3820 		dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3821 		return err;
3822 	}
3823 
3824 	if (netif_running(netdev)) {
3825 		err = ixgbevf_open(netdev);
3826 		if (err)
3827 			return err;
3828 	}
3829 
3830 	netif_device_attach(netdev);
3831 
3832 	return err;
3833 }
3834 
3835 #endif /* CONFIG_PM */
3836 static void ixgbevf_shutdown(struct pci_dev *pdev)
3837 {
3838 	ixgbevf_suspend(pdev, PMSG_SUSPEND);
3839 }
3840 
3841 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3842 						struct rtnl_link_stats64 *stats)
3843 {
3844 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3845 	unsigned int start;
3846 	u64 bytes, packets;
3847 	const struct ixgbevf_ring *ring;
3848 	int i;
3849 
3850 	ixgbevf_update_stats(adapter);
3851 
3852 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3853 
3854 	for (i = 0; i < adapter->num_rx_queues; i++) {
3855 		ring = adapter->rx_ring[i];
3856 		do {
3857 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3858 			bytes = ring->stats.bytes;
3859 			packets = ring->stats.packets;
3860 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3861 		stats->rx_bytes += bytes;
3862 		stats->rx_packets += packets;
3863 	}
3864 
3865 	for (i = 0; i < adapter->num_tx_queues; i++) {
3866 		ring = adapter->tx_ring[i];
3867 		do {
3868 			start = u64_stats_fetch_begin_irq(&ring->syncp);
3869 			bytes = ring->stats.bytes;
3870 			packets = ring->stats.packets;
3871 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3872 		stats->tx_bytes += bytes;
3873 		stats->tx_packets += packets;
3874 	}
3875 
3876 	return stats;
3877 }
3878 
3879 static const struct net_device_ops ixgbevf_netdev_ops = {
3880 	.ndo_open		= ixgbevf_open,
3881 	.ndo_stop		= ixgbevf_close,
3882 	.ndo_start_xmit		= ixgbevf_xmit_frame,
3883 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
3884 	.ndo_get_stats64	= ixgbevf_get_stats,
3885 	.ndo_validate_addr	= eth_validate_addr,
3886 	.ndo_set_mac_address	= ixgbevf_set_mac,
3887 	.ndo_change_mtu		= ixgbevf_change_mtu,
3888 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
3889 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
3890 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
3891 #ifdef CONFIG_NET_RX_BUSY_POLL
3892 	.ndo_busy_poll		= ixgbevf_busy_poll_recv,
3893 #endif
3894 #ifdef CONFIG_NET_POLL_CONTROLLER
3895 	.ndo_poll_controller	= ixgbevf_netpoll,
3896 #endif
3897 };
3898 
3899 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3900 {
3901 	dev->netdev_ops = &ixgbevf_netdev_ops;
3902 	ixgbevf_set_ethtool_ops(dev);
3903 	dev->watchdog_timeo = 5 * HZ;
3904 }
3905 
3906 /**
3907  * ixgbevf_probe - Device Initialization Routine
3908  * @pdev: PCI device information struct
3909  * @ent: entry in ixgbevf_pci_tbl
3910  *
3911  * Returns 0 on success, negative on failure
3912  *
3913  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3914  * The OS initialization, configuring of the adapter private structure,
3915  * and a hardware reset occur.
3916  **/
3917 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3918 {
3919 	struct net_device *netdev;
3920 	struct ixgbevf_adapter *adapter = NULL;
3921 	struct ixgbe_hw *hw = NULL;
3922 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3923 	int err, pci_using_dac;
3924 	bool disable_dev = false;
3925 
3926 	err = pci_enable_device(pdev);
3927 	if (err)
3928 		return err;
3929 
3930 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3931 		pci_using_dac = 1;
3932 	} else {
3933 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3934 		if (err) {
3935 			dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3936 			goto err_dma;
3937 		}
3938 		pci_using_dac = 0;
3939 	}
3940 
3941 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3942 	if (err) {
3943 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3944 		goto err_pci_reg;
3945 	}
3946 
3947 	pci_set_master(pdev);
3948 
3949 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3950 				   MAX_TX_QUEUES);
3951 	if (!netdev) {
3952 		err = -ENOMEM;
3953 		goto err_alloc_etherdev;
3954 	}
3955 
3956 	SET_NETDEV_DEV(netdev, &pdev->dev);
3957 
3958 	adapter = netdev_priv(netdev);
3959 
3960 	adapter->netdev = netdev;
3961 	adapter->pdev = pdev;
3962 	hw = &adapter->hw;
3963 	hw->back = adapter;
3964 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3965 
3966 	/* call save state here in standalone driver because it relies on
3967 	 * adapter struct to exist, and needs to call netdev_priv
3968 	 */
3969 	pci_save_state(pdev);
3970 
3971 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3972 			      pci_resource_len(pdev, 0));
3973 	adapter->io_addr = hw->hw_addr;
3974 	if (!hw->hw_addr) {
3975 		err = -EIO;
3976 		goto err_ioremap;
3977 	}
3978 
3979 	ixgbevf_assign_netdev_ops(netdev);
3980 
3981 	/* Setup HW API */
3982 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3983 	hw->mac.type  = ii->mac;
3984 
3985 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3986 	       sizeof(struct ixgbe_mbx_operations));
3987 
3988 	/* setup the private structure */
3989 	err = ixgbevf_sw_init(adapter);
3990 	if (err)
3991 		goto err_sw_init;
3992 
3993 	/* The HW MAC address was set and/or determined in sw_init */
3994 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3995 		pr_err("invalid MAC address\n");
3996 		err = -EIO;
3997 		goto err_sw_init;
3998 	}
3999 
4000 	netdev->hw_features = NETIF_F_SG |
4001 			      NETIF_F_IP_CSUM |
4002 			      NETIF_F_IPV6_CSUM |
4003 			      NETIF_F_TSO |
4004 			      NETIF_F_TSO6 |
4005 			      NETIF_F_RXCSUM;
4006 
4007 	netdev->features = netdev->hw_features |
4008 			   NETIF_F_HW_VLAN_CTAG_TX |
4009 			   NETIF_F_HW_VLAN_CTAG_RX |
4010 			   NETIF_F_HW_VLAN_CTAG_FILTER;
4011 
4012 	netdev->vlan_features |= NETIF_F_TSO |
4013 				 NETIF_F_TSO6 |
4014 				 NETIF_F_IP_CSUM |
4015 				 NETIF_F_IPV6_CSUM |
4016 				 NETIF_F_SG;
4017 
4018 	if (pci_using_dac)
4019 		netdev->features |= NETIF_F_HIGHDMA;
4020 
4021 	netdev->priv_flags |= IFF_UNICAST_FLT;
4022 
4023 	if (IXGBE_REMOVED(hw->hw_addr)) {
4024 		err = -EIO;
4025 		goto err_sw_init;
4026 	}
4027 
4028 	setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4029 		    (unsigned long)adapter);
4030 
4031 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4032 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4033 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4034 
4035 	err = ixgbevf_init_interrupt_scheme(adapter);
4036 	if (err)
4037 		goto err_sw_init;
4038 
4039 	strcpy(netdev->name, "eth%d");
4040 
4041 	err = register_netdev(netdev);
4042 	if (err)
4043 		goto err_register;
4044 
4045 	pci_set_drvdata(pdev, netdev);
4046 	netif_carrier_off(netdev);
4047 
4048 	ixgbevf_init_last_counter_stats(adapter);
4049 
4050 	/* print the VF info */
4051 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4052 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4053 
4054 	switch (hw->mac.type) {
4055 	case ixgbe_mac_X550_vf:
4056 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4057 		break;
4058 	case ixgbe_mac_X540_vf:
4059 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4060 		break;
4061 	case ixgbe_mac_82599_vf:
4062 	default:
4063 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4064 		break;
4065 	}
4066 
4067 	return 0;
4068 
4069 err_register:
4070 	ixgbevf_clear_interrupt_scheme(adapter);
4071 err_sw_init:
4072 	ixgbevf_reset_interrupt_capability(adapter);
4073 	iounmap(adapter->io_addr);
4074 err_ioremap:
4075 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4076 	free_netdev(netdev);
4077 err_alloc_etherdev:
4078 	pci_release_regions(pdev);
4079 err_pci_reg:
4080 err_dma:
4081 	if (!adapter || disable_dev)
4082 		pci_disable_device(pdev);
4083 	return err;
4084 }
4085 
4086 /**
4087  * ixgbevf_remove - Device Removal Routine
4088  * @pdev: PCI device information struct
4089  *
4090  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4091  * that it should release a PCI device.  The could be caused by a
4092  * Hot-Plug event, or because the driver is going to be removed from
4093  * memory.
4094  **/
4095 static void ixgbevf_remove(struct pci_dev *pdev)
4096 {
4097 	struct net_device *netdev = pci_get_drvdata(pdev);
4098 	struct ixgbevf_adapter *adapter;
4099 	bool disable_dev;
4100 
4101 	if (!netdev)
4102 		return;
4103 
4104 	adapter = netdev_priv(netdev);
4105 
4106 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4107 	cancel_work_sync(&adapter->service_task);
4108 
4109 	if (netdev->reg_state == NETREG_REGISTERED)
4110 		unregister_netdev(netdev);
4111 
4112 	ixgbevf_clear_interrupt_scheme(adapter);
4113 	ixgbevf_reset_interrupt_capability(adapter);
4114 
4115 	iounmap(adapter->io_addr);
4116 	pci_release_regions(pdev);
4117 
4118 	hw_dbg(&adapter->hw, "Remove complete\n");
4119 
4120 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4121 	free_netdev(netdev);
4122 
4123 	if (disable_dev)
4124 		pci_disable_device(pdev);
4125 }
4126 
4127 /**
4128  * ixgbevf_io_error_detected - called when PCI error is detected
4129  * @pdev: Pointer to PCI device
4130  * @state: The current pci connection state
4131  *
4132  * This function is called after a PCI bus error affecting
4133  * this device has been detected.
4134  **/
4135 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4136 						  pci_channel_state_t state)
4137 {
4138 	struct net_device *netdev = pci_get_drvdata(pdev);
4139 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4140 
4141 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4142 		return PCI_ERS_RESULT_DISCONNECT;
4143 
4144 	rtnl_lock();
4145 	netif_device_detach(netdev);
4146 
4147 	if (state == pci_channel_io_perm_failure) {
4148 		rtnl_unlock();
4149 		return PCI_ERS_RESULT_DISCONNECT;
4150 	}
4151 
4152 	if (netif_running(netdev))
4153 		ixgbevf_down(adapter);
4154 
4155 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4156 		pci_disable_device(pdev);
4157 	rtnl_unlock();
4158 
4159 	/* Request a slot slot reset. */
4160 	return PCI_ERS_RESULT_NEED_RESET;
4161 }
4162 
4163 /**
4164  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4165  * @pdev: Pointer to PCI device
4166  *
4167  * Restart the card from scratch, as if from a cold-boot. Implementation
4168  * resembles the first-half of the ixgbevf_resume routine.
4169  **/
4170 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4171 {
4172 	struct net_device *netdev = pci_get_drvdata(pdev);
4173 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4174 
4175 	if (pci_enable_device_mem(pdev)) {
4176 		dev_err(&pdev->dev,
4177 			"Cannot re-enable PCI device after reset.\n");
4178 		return PCI_ERS_RESULT_DISCONNECT;
4179 	}
4180 
4181 	smp_mb__before_atomic();
4182 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4183 	pci_set_master(pdev);
4184 
4185 	ixgbevf_reset(adapter);
4186 
4187 	return PCI_ERS_RESULT_RECOVERED;
4188 }
4189 
4190 /**
4191  * ixgbevf_io_resume - called when traffic can start flowing again.
4192  * @pdev: Pointer to PCI device
4193  *
4194  * This callback is called when the error recovery driver tells us that
4195  * its OK to resume normal operation. Implementation resembles the
4196  * second-half of the ixgbevf_resume routine.
4197  **/
4198 static void ixgbevf_io_resume(struct pci_dev *pdev)
4199 {
4200 	struct net_device *netdev = pci_get_drvdata(pdev);
4201 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4202 
4203 	if (netif_running(netdev))
4204 		ixgbevf_up(adapter);
4205 
4206 	netif_device_attach(netdev);
4207 }
4208 
4209 /* PCI Error Recovery (ERS) */
4210 static const struct pci_error_handlers ixgbevf_err_handler = {
4211 	.error_detected = ixgbevf_io_error_detected,
4212 	.slot_reset = ixgbevf_io_slot_reset,
4213 	.resume = ixgbevf_io_resume,
4214 };
4215 
4216 static struct pci_driver ixgbevf_driver = {
4217 	.name		= ixgbevf_driver_name,
4218 	.id_table	= ixgbevf_pci_tbl,
4219 	.probe		= ixgbevf_probe,
4220 	.remove		= ixgbevf_remove,
4221 #ifdef CONFIG_PM
4222 	/* Power Management Hooks */
4223 	.suspend	= ixgbevf_suspend,
4224 	.resume		= ixgbevf_resume,
4225 #endif
4226 	.shutdown	= ixgbevf_shutdown,
4227 	.err_handler	= &ixgbevf_err_handler
4228 };
4229 
4230 /**
4231  * ixgbevf_init_module - Driver Registration Routine
4232  *
4233  * ixgbevf_init_module is the first routine called when the driver is
4234  * loaded. All it does is register with the PCI subsystem.
4235  **/
4236 static int __init ixgbevf_init_module(void)
4237 {
4238 	int ret;
4239 
4240 	pr_info("%s - version %s\n", ixgbevf_driver_string,
4241 		ixgbevf_driver_version);
4242 
4243 	pr_info("%s\n", ixgbevf_copyright);
4244 
4245 	ret = pci_register_driver(&ixgbevf_driver);
4246 	return ret;
4247 }
4248 
4249 module_init(ixgbevf_init_module);
4250 
4251 /**
4252  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4253  *
4254  * ixgbevf_exit_module is called just before the driver is removed
4255  * from memory.
4256  **/
4257 static void __exit ixgbevf_exit_module(void)
4258 {
4259 	pci_unregister_driver(&ixgbevf_driver);
4260 }
4261 
4262 #ifdef DEBUG
4263 /**
4264  * ixgbevf_get_hw_dev_name - return device name string
4265  * used by hardware layer to print debugging information
4266  **/
4267 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4268 {
4269 	struct ixgbevf_adapter *adapter = hw->back;
4270 
4271 	return adapter->netdev->name;
4272 }
4273 
4274 #endif
4275 module_exit(ixgbevf_exit_module);
4276 
4277 /* ixgbevf_main.c */
4278