1 /*******************************************************************************
2 
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5 
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9 
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14 
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21 
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 
26 *******************************************************************************/
27 
28 
29 /******************************************************************************
30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
32 
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
51 #include <linux/if.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
54 
55 #include "ixgbevf.h"
56 
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
60 
61 #define DRV_VERSION "2.6.0-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64 	"Copyright (c) 2009 - 2012 Intel Corporation.";
65 
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67 	[board_82599_vf] = &ixgbevf_82599_vf_info,
68 	[board_X540_vf]  = &ixgbevf_X540_vf_info,
69 };
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 struct pci_device_id ixgbevf_pci_tbl[] = {
80 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
81 	board_82599_vf},
82 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
83 	board_X540_vf},
84 
85 	/* required last entry */
86 	{0, }
87 };
88 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
89 
90 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
91 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
92 MODULE_LICENSE("GPL");
93 MODULE_VERSION(DRV_VERSION);
94 
95 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
96 static int debug = -1;
97 module_param(debug, int, 0);
98 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
99 
100 /* forward decls */
101 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
102 
103 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
104 					   struct ixgbevf_ring *rx_ring,
105 					   u32 val)
106 {
107 	/*
108 	 * Force memory writes to complete before letting h/w
109 	 * know there are new descriptors to fetch.  (Only
110 	 * applicable for weak-ordered memory model archs,
111 	 * such as IA-64).
112 	 */
113 	wmb();
114 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
115 }
116 
117 /**
118  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
119  * @adapter: pointer to adapter struct
120  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
121  * @queue: queue to map the corresponding interrupt to
122  * @msix_vector: the vector to map to the corresponding queue
123  *
124  */
125 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
126 			     u8 queue, u8 msix_vector)
127 {
128 	u32 ivar, index;
129 	struct ixgbe_hw *hw = &adapter->hw;
130 	if (direction == -1) {
131 		/* other causes */
132 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
133 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
134 		ivar &= ~0xFF;
135 		ivar |= msix_vector;
136 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
137 	} else {
138 		/* tx or rx causes */
139 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
140 		index = ((16 * (queue & 1)) + (8 * direction));
141 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
142 		ivar &= ~(0xFF << index);
143 		ivar |= (msix_vector << index);
144 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
145 	}
146 }
147 
148 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
149 					       struct ixgbevf_tx_buffer
150 					       *tx_buffer_info)
151 {
152 	if (tx_buffer_info->dma) {
153 		if (tx_buffer_info->mapped_as_page)
154 			dma_unmap_page(tx_ring->dev,
155 				       tx_buffer_info->dma,
156 				       tx_buffer_info->length,
157 				       DMA_TO_DEVICE);
158 		else
159 			dma_unmap_single(tx_ring->dev,
160 					 tx_buffer_info->dma,
161 					 tx_buffer_info->length,
162 					 DMA_TO_DEVICE);
163 		tx_buffer_info->dma = 0;
164 	}
165 	if (tx_buffer_info->skb) {
166 		dev_kfree_skb_any(tx_buffer_info->skb);
167 		tx_buffer_info->skb = NULL;
168 	}
169 	tx_buffer_info->time_stamp = 0;
170 	/* tx_buffer_info must be completely set up in the transmit path */
171 }
172 
173 #define IXGBE_MAX_TXD_PWR	14
174 #define IXGBE_MAX_DATA_PER_TXD	(1 << IXGBE_MAX_TXD_PWR)
175 
176 /* Tx Descriptors needed, worst case */
177 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
178 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
179 
180 static void ixgbevf_tx_timeout(struct net_device *netdev);
181 
182 /**
183  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
184  * @q_vector: board private structure
185  * @tx_ring: tx ring to clean
186  **/
187 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
188 				 struct ixgbevf_ring *tx_ring)
189 {
190 	struct ixgbevf_adapter *adapter = q_vector->adapter;
191 	union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
192 	struct ixgbevf_tx_buffer *tx_buffer_info;
193 	unsigned int i, eop, count = 0;
194 	unsigned int total_bytes = 0, total_packets = 0;
195 
196 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
197 		return true;
198 
199 	i = tx_ring->next_to_clean;
200 	eop = tx_ring->tx_buffer_info[i].next_to_watch;
201 	eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
202 
203 	while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
204 	       (count < tx_ring->count)) {
205 		bool cleaned = false;
206 		rmb(); /* read buffer_info after eop_desc */
207 		/* eop could change between read and DD-check */
208 		if (unlikely(eop != tx_ring->tx_buffer_info[i].next_to_watch))
209 			goto cont_loop;
210 		for ( ; !cleaned; count++) {
211 			struct sk_buff *skb;
212 			tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
213 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
214 			cleaned = (i == eop);
215 			skb = tx_buffer_info->skb;
216 
217 			if (cleaned && skb) {
218 				unsigned int segs, bytecount;
219 
220 				/* gso_segs is currently only valid for tcp */
221 				segs = skb_shinfo(skb)->gso_segs ?: 1;
222 				/* multiply data chunks by size of headers */
223 				bytecount = ((segs - 1) * skb_headlen(skb)) +
224 					    skb->len;
225 				total_packets += segs;
226 				total_bytes += bytecount;
227 			}
228 
229 			ixgbevf_unmap_and_free_tx_resource(tx_ring,
230 							   tx_buffer_info);
231 
232 			tx_desc->wb.status = 0;
233 
234 			i++;
235 			if (i == tx_ring->count)
236 				i = 0;
237 		}
238 
239 cont_loop:
240 		eop = tx_ring->tx_buffer_info[i].next_to_watch;
241 		eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
242 	}
243 
244 	tx_ring->next_to_clean = i;
245 
246 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
247 	if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
248 		     (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
249 		/* Make sure that anybody stopping the queue after this
250 		 * sees the new next_to_clean.
251 		 */
252 		smp_mb();
253 		if (__netif_subqueue_stopped(tx_ring->netdev,
254 					     tx_ring->queue_index) &&
255 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
256 			netif_wake_subqueue(tx_ring->netdev,
257 					    tx_ring->queue_index);
258 			++adapter->restart_queue;
259 		}
260 	}
261 
262 	u64_stats_update_begin(&tx_ring->syncp);
263 	tx_ring->total_bytes += total_bytes;
264 	tx_ring->total_packets += total_packets;
265 	u64_stats_update_end(&tx_ring->syncp);
266 
267 	return count < tx_ring->count;
268 }
269 
270 /**
271  * ixgbevf_receive_skb - Send a completed packet up the stack
272  * @q_vector: structure containing interrupt and ring information
273  * @skb: packet to send up
274  * @status: hardware indication of status of receive
275  * @rx_ring: rx descriptor ring (for a specific queue) to setup
276  * @rx_desc: rx descriptor
277  **/
278 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
279 				struct sk_buff *skb, u8 status,
280 				struct ixgbevf_ring *ring,
281 				union ixgbe_adv_rx_desc *rx_desc)
282 {
283 	struct ixgbevf_adapter *adapter = q_vector->adapter;
284 	bool is_vlan = (status & IXGBE_RXD_STAT_VP);
285 	u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
286 
287 	if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
288 		__vlan_hwaccel_put_tag(skb, tag);
289 
290 	napi_gro_receive(&q_vector->napi, skb);
291 }
292 
293 /**
294  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
295  * @adapter: address of board private structure
296  * @status_err: hardware indication of status of receive
297  * @skb: skb currently being received and modified
298  **/
299 static inline void ixgbevf_rx_checksum(struct ixgbevf_adapter *adapter,
300 				       struct ixgbevf_ring *ring,
301 				       u32 status_err, struct sk_buff *skb)
302 {
303 	skb_checksum_none_assert(skb);
304 
305 	/* Rx csum disabled */
306 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
307 		return;
308 
309 	/* if IP and error */
310 	if ((status_err & IXGBE_RXD_STAT_IPCS) &&
311 	    (status_err & IXGBE_RXDADV_ERR_IPE)) {
312 		adapter->hw_csum_rx_error++;
313 		return;
314 	}
315 
316 	if (!(status_err & IXGBE_RXD_STAT_L4CS))
317 		return;
318 
319 	if (status_err & IXGBE_RXDADV_ERR_TCPE) {
320 		adapter->hw_csum_rx_error++;
321 		return;
322 	}
323 
324 	/* It must be a TCP or UDP packet with a valid checksum */
325 	skb->ip_summed = CHECKSUM_UNNECESSARY;
326 	adapter->hw_csum_rx_good++;
327 }
328 
329 /**
330  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
331  * @adapter: address of board private structure
332  **/
333 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
334 				     struct ixgbevf_ring *rx_ring,
335 				     int cleaned_count)
336 {
337 	struct pci_dev *pdev = adapter->pdev;
338 	union ixgbe_adv_rx_desc *rx_desc;
339 	struct ixgbevf_rx_buffer *bi;
340 	struct sk_buff *skb;
341 	unsigned int i = rx_ring->next_to_use;
342 
343 	bi = &rx_ring->rx_buffer_info[i];
344 
345 	while (cleaned_count--) {
346 		rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
347 		skb = bi->skb;
348 		if (!skb) {
349 			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
350 							rx_ring->rx_buf_len);
351 			if (!skb) {
352 				adapter->alloc_rx_buff_failed++;
353 				goto no_buffers;
354 			}
355 			bi->skb = skb;
356 		}
357 		if (!bi->dma) {
358 			bi->dma = dma_map_single(&pdev->dev, skb->data,
359 						 rx_ring->rx_buf_len,
360 						 DMA_FROM_DEVICE);
361 		}
362 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
363 
364 		i++;
365 		if (i == rx_ring->count)
366 			i = 0;
367 		bi = &rx_ring->rx_buffer_info[i];
368 	}
369 
370 no_buffers:
371 	if (rx_ring->next_to_use != i) {
372 		rx_ring->next_to_use = i;
373 
374 		ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
375 	}
376 }
377 
378 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
379 					     u32 qmask)
380 {
381 	struct ixgbe_hw *hw = &adapter->hw;
382 
383 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
384 }
385 
386 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
387 				 struct ixgbevf_ring *rx_ring,
388 				 int budget)
389 {
390 	struct ixgbevf_adapter *adapter = q_vector->adapter;
391 	struct pci_dev *pdev = adapter->pdev;
392 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
393 	struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
394 	struct sk_buff *skb;
395 	unsigned int i;
396 	u32 len, staterr;
397 	int cleaned_count = 0;
398 	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
399 
400 	i = rx_ring->next_to_clean;
401 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
402 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
403 	rx_buffer_info = &rx_ring->rx_buffer_info[i];
404 
405 	while (staterr & IXGBE_RXD_STAT_DD) {
406 		if (!budget)
407 			break;
408 		budget--;
409 
410 		rmb(); /* read descriptor and rx_buffer_info after status DD */
411 		len = le16_to_cpu(rx_desc->wb.upper.length);
412 		skb = rx_buffer_info->skb;
413 		prefetch(skb->data - NET_IP_ALIGN);
414 		rx_buffer_info->skb = NULL;
415 
416 		if (rx_buffer_info->dma) {
417 			dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
418 					 rx_ring->rx_buf_len,
419 					 DMA_FROM_DEVICE);
420 			rx_buffer_info->dma = 0;
421 			skb_put(skb, len);
422 		}
423 
424 		i++;
425 		if (i == rx_ring->count)
426 			i = 0;
427 
428 		next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
429 		prefetch(next_rxd);
430 		cleaned_count++;
431 
432 		next_buffer = &rx_ring->rx_buffer_info[i];
433 
434 		if (!(staterr & IXGBE_RXD_STAT_EOP)) {
435 			skb->next = next_buffer->skb;
436 			skb->next->prev = skb;
437 			adapter->non_eop_descs++;
438 			goto next_desc;
439 		}
440 
441 		/* ERR_MASK will only have valid bits if EOP set */
442 		if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
443 			dev_kfree_skb_irq(skb);
444 			goto next_desc;
445 		}
446 
447 		ixgbevf_rx_checksum(adapter, rx_ring, staterr, skb);
448 
449 		/* probably a little skewed due to removing CRC */
450 		total_rx_bytes += skb->len;
451 		total_rx_packets++;
452 
453 		/*
454 		 * Work around issue of some types of VM to VM loop back
455 		 * packets not getting split correctly
456 		 */
457 		if (staterr & IXGBE_RXD_STAT_LB) {
458 			u32 header_fixup_len = skb_headlen(skb);
459 			if (header_fixup_len < 14)
460 				skb_push(skb, header_fixup_len);
461 		}
462 		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
463 
464 		ixgbevf_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
465 
466 next_desc:
467 		rx_desc->wb.upper.status_error = 0;
468 
469 		/* return some buffers to hardware, one at a time is too slow */
470 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
471 			ixgbevf_alloc_rx_buffers(adapter, rx_ring,
472 						 cleaned_count);
473 			cleaned_count = 0;
474 		}
475 
476 		/* use prefetched values */
477 		rx_desc = next_rxd;
478 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
479 
480 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
481 	}
482 
483 	rx_ring->next_to_clean = i;
484 	cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
485 
486 	if (cleaned_count)
487 		ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
488 
489 	u64_stats_update_begin(&rx_ring->syncp);
490 	rx_ring->total_packets += total_rx_packets;
491 	rx_ring->total_bytes += total_rx_bytes;
492 	u64_stats_update_end(&rx_ring->syncp);
493 
494 	return !!budget;
495 }
496 
497 /**
498  * ixgbevf_poll - NAPI polling calback
499  * @napi: napi struct with our devices info in it
500  * @budget: amount of work driver is allowed to do this pass, in packets
501  *
502  * This function will clean more than one or more rings associated with a
503  * q_vector.
504  **/
505 static int ixgbevf_poll(struct napi_struct *napi, int budget)
506 {
507 	struct ixgbevf_q_vector *q_vector =
508 		container_of(napi, struct ixgbevf_q_vector, napi);
509 	struct ixgbevf_adapter *adapter = q_vector->adapter;
510 	struct ixgbevf_ring *ring;
511 	int per_ring_budget;
512 	bool clean_complete = true;
513 
514 	ixgbevf_for_each_ring(ring, q_vector->tx)
515 		clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
516 
517 	/* attempt to distribute budget to each queue fairly, but don't allow
518 	 * the budget to go below 1 because we'll exit polling */
519 	if (q_vector->rx.count > 1)
520 		per_ring_budget = max(budget/q_vector->rx.count, 1);
521 	else
522 		per_ring_budget = budget;
523 
524 	ixgbevf_for_each_ring(ring, q_vector->rx)
525 		clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
526 						       per_ring_budget);
527 
528 	/* If all work not completed, return budget and keep polling */
529 	if (!clean_complete)
530 		return budget;
531 	/* all work done, exit the polling mode */
532 	napi_complete(napi);
533 	if (adapter->rx_itr_setting & 1)
534 		ixgbevf_set_itr(q_vector);
535 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
536 		ixgbevf_irq_enable_queues(adapter,
537 					  1 << q_vector->v_idx);
538 
539 	return 0;
540 }
541 
542 /**
543  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
544  * @q_vector: structure containing interrupt and ring information
545  */
546 static void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
547 {
548 	struct ixgbevf_adapter *adapter = q_vector->adapter;
549 	struct ixgbe_hw *hw = &adapter->hw;
550 	int v_idx = q_vector->v_idx;
551 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
552 
553 	/*
554 	 * set the WDIS bit to not clear the timer bits and cause an
555 	 * immediate assertion of the interrupt
556 	 */
557 	itr_reg |= IXGBE_EITR_CNT_WDIS;
558 
559 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
560 }
561 
562 /**
563  * ixgbevf_configure_msix - Configure MSI-X hardware
564  * @adapter: board private structure
565  *
566  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
567  * interrupts.
568  **/
569 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
570 {
571 	struct ixgbevf_q_vector *q_vector;
572 	int q_vectors, v_idx;
573 
574 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
575 	adapter->eims_enable_mask = 0;
576 
577 	/*
578 	 * Populate the IVAR table and set the ITR values to the
579 	 * corresponding register.
580 	 */
581 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
582 		struct ixgbevf_ring *ring;
583 		q_vector = adapter->q_vector[v_idx];
584 
585 		ixgbevf_for_each_ring(ring, q_vector->rx)
586 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
587 
588 		ixgbevf_for_each_ring(ring, q_vector->tx)
589 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
590 
591 		if (q_vector->tx.ring && !q_vector->rx.ring) {
592 			/* tx only vector */
593 			if (adapter->tx_itr_setting == 1)
594 				q_vector->itr = IXGBE_10K_ITR;
595 			else
596 				q_vector->itr = adapter->tx_itr_setting;
597 		} else {
598 			/* rx or rx/tx vector */
599 			if (adapter->rx_itr_setting == 1)
600 				q_vector->itr = IXGBE_20K_ITR;
601 			else
602 				q_vector->itr = adapter->rx_itr_setting;
603 		}
604 
605 		/* add q_vector eims value to global eims_enable_mask */
606 		adapter->eims_enable_mask |= 1 << v_idx;
607 
608 		ixgbevf_write_eitr(q_vector);
609 	}
610 
611 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
612 	/* setup eims_other and add value to global eims_enable_mask */
613 	adapter->eims_other = 1 << v_idx;
614 	adapter->eims_enable_mask |= adapter->eims_other;
615 }
616 
617 enum latency_range {
618 	lowest_latency = 0,
619 	low_latency = 1,
620 	bulk_latency = 2,
621 	latency_invalid = 255
622 };
623 
624 /**
625  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
626  * @q_vector: structure containing interrupt and ring information
627  * @ring_container: structure containing ring performance data
628  *
629  *      Stores a new ITR value based on packets and byte
630  *      counts during the last interrupt.  The advantage of per interrupt
631  *      computation is faster updates and more accurate ITR for the current
632  *      traffic pattern.  Constants in this function were computed
633  *      based on theoretical maximum wire speed and thresholds were set based
634  *      on testing data as well as attempting to minimize response time
635  *      while increasing bulk throughput.
636  **/
637 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
638 			       struct ixgbevf_ring_container *ring_container)
639 {
640 	int bytes = ring_container->total_bytes;
641 	int packets = ring_container->total_packets;
642 	u32 timepassed_us;
643 	u64 bytes_perint;
644 	u8 itr_setting = ring_container->itr;
645 
646 	if (packets == 0)
647 		return;
648 
649 	/* simple throttlerate management
650 	 *    0-20MB/s lowest (100000 ints/s)
651 	 *   20-100MB/s low   (20000 ints/s)
652 	 *  100-1249MB/s bulk (8000 ints/s)
653 	 */
654 	/* what was last interrupt timeslice? */
655 	timepassed_us = q_vector->itr >> 2;
656 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
657 
658 	switch (itr_setting) {
659 	case lowest_latency:
660 		if (bytes_perint > 10)
661 			itr_setting = low_latency;
662 		break;
663 	case low_latency:
664 		if (bytes_perint > 20)
665 			itr_setting = bulk_latency;
666 		else if (bytes_perint <= 10)
667 			itr_setting = lowest_latency;
668 		break;
669 	case bulk_latency:
670 		if (bytes_perint <= 20)
671 			itr_setting = low_latency;
672 		break;
673 	}
674 
675 	/* clear work counters since we have the values we need */
676 	ring_container->total_bytes = 0;
677 	ring_container->total_packets = 0;
678 
679 	/* write updated itr to ring container */
680 	ring_container->itr = itr_setting;
681 }
682 
683 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
684 {
685 	u32 new_itr = q_vector->itr;
686 	u8 current_itr;
687 
688 	ixgbevf_update_itr(q_vector, &q_vector->tx);
689 	ixgbevf_update_itr(q_vector, &q_vector->rx);
690 
691 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
692 
693 	switch (current_itr) {
694 	/* counts and packets in update_itr are dependent on these numbers */
695 	case lowest_latency:
696 		new_itr = IXGBE_100K_ITR;
697 		break;
698 	case low_latency:
699 		new_itr = IXGBE_20K_ITR;
700 		break;
701 	case bulk_latency:
702 	default:
703 		new_itr = IXGBE_8K_ITR;
704 		break;
705 	}
706 
707 	if (new_itr != q_vector->itr) {
708 		/* do an exponential smoothing */
709 		new_itr = (10 * new_itr * q_vector->itr) /
710 			  ((9 * new_itr) + q_vector->itr);
711 
712 		/* save the algorithm value here */
713 		q_vector->itr = new_itr;
714 
715 		ixgbevf_write_eitr(q_vector);
716 	}
717 }
718 
719 static irqreturn_t ixgbevf_msix_mbx(int irq, void *data)
720 {
721 	struct ixgbevf_adapter *adapter = data;
722 	struct ixgbe_hw *hw = &adapter->hw;
723 	u32 msg;
724 	bool got_ack = false;
725 
726 	if (!hw->mbx.ops.check_for_ack(hw))
727 		got_ack = true;
728 
729 	if (!hw->mbx.ops.check_for_msg(hw)) {
730 		hw->mbx.ops.read(hw, &msg, 1);
731 
732 		if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG)
733 			mod_timer(&adapter->watchdog_timer,
734 				  round_jiffies(jiffies + 1));
735 
736 		if (msg & IXGBE_VT_MSGTYPE_NACK)
737 			pr_warn("Last Request of type %2.2x to PF Nacked\n",
738 				msg & 0xFF);
739 		/*
740 		 * Restore the PFSTS bit in case someone is polling for a
741 		 * return message from the PF
742 		 */
743 		hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
744 	}
745 
746 	/*
747 	 * checking for the ack clears the PFACK bit.  Place
748 	 * it back in the v2p_mailbox cache so that anyone
749 	 * polling for an ack will not miss it
750 	 */
751 	if (got_ack)
752 		hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
753 
754 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
755 
756 	return IRQ_HANDLED;
757 }
758 
759 
760 /**
761  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
762  * @irq: unused
763  * @data: pointer to our q_vector struct for this interrupt vector
764  **/
765 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
766 {
767 	struct ixgbevf_q_vector *q_vector = data;
768 
769 	/* EIAM disabled interrupts (on this vector) for us */
770 	if (q_vector->rx.ring || q_vector->tx.ring)
771 		napi_schedule(&q_vector->napi);
772 
773 	return IRQ_HANDLED;
774 }
775 
776 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
777 				     int r_idx)
778 {
779 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
780 
781 	a->rx_ring[r_idx].next = q_vector->rx.ring;
782 	q_vector->rx.ring = &a->rx_ring[r_idx];
783 	q_vector->rx.count++;
784 }
785 
786 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
787 				     int t_idx)
788 {
789 	struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
790 
791 	a->tx_ring[t_idx].next = q_vector->tx.ring;
792 	q_vector->tx.ring = &a->tx_ring[t_idx];
793 	q_vector->tx.count++;
794 }
795 
796 /**
797  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
798  * @adapter: board private structure to initialize
799  *
800  * This function maps descriptor rings to the queue-specific vectors
801  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
802  * one vector per ring/queue, but on a constrained vector budget, we
803  * group the rings as "efficiently" as possible.  You would add new
804  * mapping configurations in here.
805  **/
806 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
807 {
808 	int q_vectors;
809 	int v_start = 0;
810 	int rxr_idx = 0, txr_idx = 0;
811 	int rxr_remaining = adapter->num_rx_queues;
812 	int txr_remaining = adapter->num_tx_queues;
813 	int i, j;
814 	int rqpv, tqpv;
815 	int err = 0;
816 
817 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
818 
819 	/*
820 	 * The ideal configuration...
821 	 * We have enough vectors to map one per queue.
822 	 */
823 	if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
824 		for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
825 			map_vector_to_rxq(adapter, v_start, rxr_idx);
826 
827 		for (; txr_idx < txr_remaining; v_start++, txr_idx++)
828 			map_vector_to_txq(adapter, v_start, txr_idx);
829 		goto out;
830 	}
831 
832 	/*
833 	 * If we don't have enough vectors for a 1-to-1
834 	 * mapping, we'll have to group them so there are
835 	 * multiple queues per vector.
836 	 */
837 	/* Re-adjusting *qpv takes care of the remainder. */
838 	for (i = v_start; i < q_vectors; i++) {
839 		rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
840 		for (j = 0; j < rqpv; j++) {
841 			map_vector_to_rxq(adapter, i, rxr_idx);
842 			rxr_idx++;
843 			rxr_remaining--;
844 		}
845 	}
846 	for (i = v_start; i < q_vectors; i++) {
847 		tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
848 		for (j = 0; j < tqpv; j++) {
849 			map_vector_to_txq(adapter, i, txr_idx);
850 			txr_idx++;
851 			txr_remaining--;
852 		}
853 	}
854 
855 out:
856 	return err;
857 }
858 
859 /**
860  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
861  * @adapter: board private structure
862  *
863  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
864  * interrupts from the kernel.
865  **/
866 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
867 {
868 	struct net_device *netdev = adapter->netdev;
869 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
870 	int vector, err;
871 	int ri = 0, ti = 0;
872 
873 	for (vector = 0; vector < q_vectors; vector++) {
874 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
875 		struct msix_entry *entry = &adapter->msix_entries[vector];
876 
877 		if (q_vector->tx.ring && q_vector->rx.ring) {
878 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
879 				 "%s-%s-%d", netdev->name, "TxRx", ri++);
880 			ti++;
881 		} else if (q_vector->rx.ring) {
882 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
883 				 "%s-%s-%d", netdev->name, "rx", ri++);
884 		} else if (q_vector->tx.ring) {
885 			snprintf(q_vector->name, sizeof(q_vector->name) - 1,
886 				 "%s-%s-%d", netdev->name, "tx", ti++);
887 		} else {
888 			/* skip this unused q_vector */
889 			continue;
890 		}
891 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
892 				  q_vector->name, q_vector);
893 		if (err) {
894 			hw_dbg(&adapter->hw,
895 			       "request_irq failed for MSIX interrupt "
896 			       "Error: %d\n", err);
897 			goto free_queue_irqs;
898 		}
899 	}
900 
901 	err = request_irq(adapter->msix_entries[vector].vector,
902 			  &ixgbevf_msix_mbx, 0, netdev->name, adapter);
903 	if (err) {
904 		hw_dbg(&adapter->hw,
905 		       "request_irq for msix_mbx failed: %d\n", err);
906 		goto free_queue_irqs;
907 	}
908 
909 	return 0;
910 
911 free_queue_irqs:
912 	while (vector) {
913 		vector--;
914 		free_irq(adapter->msix_entries[vector].vector,
915 			 adapter->q_vector[vector]);
916 	}
917 	pci_disable_msix(adapter->pdev);
918 	kfree(adapter->msix_entries);
919 	adapter->msix_entries = NULL;
920 	return err;
921 }
922 
923 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
924 {
925 	int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
926 
927 	for (i = 0; i < q_vectors; i++) {
928 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
929 		q_vector->rx.ring = NULL;
930 		q_vector->tx.ring = NULL;
931 		q_vector->rx.count = 0;
932 		q_vector->tx.count = 0;
933 	}
934 }
935 
936 /**
937  * ixgbevf_request_irq - initialize interrupts
938  * @adapter: board private structure
939  *
940  * Attempts to configure interrupts using the best available
941  * capabilities of the hardware and kernel.
942  **/
943 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
944 {
945 	int err = 0;
946 
947 	err = ixgbevf_request_msix_irqs(adapter);
948 
949 	if (err)
950 		hw_dbg(&adapter->hw,
951 		       "request_irq failed, Error %d\n", err);
952 
953 	return err;
954 }
955 
956 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
957 {
958 	int i, q_vectors;
959 
960 	q_vectors = adapter->num_msix_vectors;
961 	i = q_vectors - 1;
962 
963 	free_irq(adapter->msix_entries[i].vector, adapter);
964 	i--;
965 
966 	for (; i >= 0; i--) {
967 		/* free only the irqs that were actually requested */
968 		if (!adapter->q_vector[i]->rx.ring &&
969 		    !adapter->q_vector[i]->tx.ring)
970 			continue;
971 
972 		free_irq(adapter->msix_entries[i].vector,
973 			 adapter->q_vector[i]);
974 	}
975 
976 	ixgbevf_reset_q_vectors(adapter);
977 }
978 
979 /**
980  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
981  * @adapter: board private structure
982  **/
983 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
984 {
985 	struct ixgbe_hw *hw = &adapter->hw;
986 	int i;
987 
988 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
989 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
990 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
991 
992 	IXGBE_WRITE_FLUSH(hw);
993 
994 	for (i = 0; i < adapter->num_msix_vectors; i++)
995 		synchronize_irq(adapter->msix_entries[i].vector);
996 }
997 
998 /**
999  * ixgbevf_irq_enable - Enable default interrupt generation settings
1000  * @adapter: board private structure
1001  **/
1002 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1003 {
1004 	struct ixgbe_hw *hw = &adapter->hw;
1005 
1006 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1007 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1008 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1009 }
1010 
1011 /**
1012  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1013  * @adapter: board private structure
1014  *
1015  * Configure the Tx unit of the MAC after a reset.
1016  **/
1017 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1018 {
1019 	u64 tdba;
1020 	struct ixgbe_hw *hw = &adapter->hw;
1021 	u32 i, j, tdlen, txctrl;
1022 
1023 	/* Setup the HW Tx Head and Tail descriptor pointers */
1024 	for (i = 0; i < adapter->num_tx_queues; i++) {
1025 		struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1026 		j = ring->reg_idx;
1027 		tdba = ring->dma;
1028 		tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1029 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1030 				(tdba & DMA_BIT_MASK(32)));
1031 		IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1032 		IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1033 		IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1034 		IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1035 		adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1036 		adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1037 		/* Disable Tx Head Writeback RO bit, since this hoses
1038 		 * bookkeeping if things aren't delivered in order.
1039 		 */
1040 		txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1041 		txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1042 		IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1043 	}
1044 }
1045 
1046 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1047 
1048 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1049 {
1050 	struct ixgbevf_ring *rx_ring;
1051 	struct ixgbe_hw *hw = &adapter->hw;
1052 	u32 srrctl;
1053 
1054 	rx_ring = &adapter->rx_ring[index];
1055 
1056 	srrctl = IXGBE_SRRCTL_DROP_EN;
1057 
1058 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1059 
1060 	if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1061 		srrctl |= IXGBEVF_RXBUFFER_2048 >>
1062 			IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1063 	else
1064 		srrctl |= rx_ring->rx_buf_len >>
1065 			IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1066 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1067 }
1068 
1069 /**
1070  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1071  * @adapter: board private structure
1072  *
1073  * Configure the Rx unit of the MAC after a reset.
1074  **/
1075 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1076 {
1077 	u64 rdba;
1078 	struct ixgbe_hw *hw = &adapter->hw;
1079 	struct net_device *netdev = adapter->netdev;
1080 	int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1081 	int i, j;
1082 	u32 rdlen;
1083 	int rx_buf_len;
1084 
1085 	/* PSRTYPE must be initialized in 82599 */
1086 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1087 	if (netdev->mtu <= ETH_DATA_LEN)
1088 		rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1089 	else
1090 		rx_buf_len = ALIGN(max_frame, 1024);
1091 
1092 	rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1093 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1094 	 * the Base and Length of the Rx Descriptor Ring */
1095 	for (i = 0; i < adapter->num_rx_queues; i++) {
1096 		rdba = adapter->rx_ring[i].dma;
1097 		j = adapter->rx_ring[i].reg_idx;
1098 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1099 				(rdba & DMA_BIT_MASK(32)));
1100 		IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1101 		IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1102 		IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1103 		IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1104 		adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1105 		adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1106 		adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1107 
1108 		ixgbevf_configure_srrctl(adapter, j);
1109 	}
1110 }
1111 
1112 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1113 {
1114 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1115 	struct ixgbe_hw *hw = &adapter->hw;
1116 
1117 	spin_lock(&adapter->mbx_lock);
1118 
1119 	/* add VID to filter table */
1120 	if (hw->mac.ops.set_vfta)
1121 		hw->mac.ops.set_vfta(hw, vid, 0, true);
1122 
1123 	spin_unlock(&adapter->mbx_lock);
1124 
1125 	set_bit(vid, adapter->active_vlans);
1126 
1127 	return 0;
1128 }
1129 
1130 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1131 {
1132 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1133 	struct ixgbe_hw *hw = &adapter->hw;
1134 
1135 	spin_lock(&adapter->mbx_lock);
1136 
1137 	/* remove VID from filter table */
1138 	if (hw->mac.ops.set_vfta)
1139 		hw->mac.ops.set_vfta(hw, vid, 0, false);
1140 
1141 	spin_unlock(&adapter->mbx_lock);
1142 
1143 	clear_bit(vid, adapter->active_vlans);
1144 
1145 	return 0;
1146 }
1147 
1148 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1149 {
1150 	u16 vid;
1151 
1152 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1153 		ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1154 }
1155 
1156 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1157 {
1158 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1159 	struct ixgbe_hw *hw = &adapter->hw;
1160 	int count = 0;
1161 
1162 	if ((netdev_uc_count(netdev)) > 10) {
1163 		pr_err("Too many unicast filters - No Space\n");
1164 		return -ENOSPC;
1165 	}
1166 
1167 	if (!netdev_uc_empty(netdev)) {
1168 		struct netdev_hw_addr *ha;
1169 		netdev_for_each_uc_addr(ha, netdev) {
1170 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1171 			udelay(200);
1172 		}
1173 	} else {
1174 		/*
1175 		 * If the list is empty then send message to PF driver to
1176 		 * clear all macvlans on this VF.
1177 		 */
1178 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
1179 	}
1180 
1181 	return count;
1182 }
1183 
1184 /**
1185  * ixgbevf_set_rx_mode - Multicast set
1186  * @netdev: network interface device structure
1187  *
1188  * The set_rx_method entry point is called whenever the multicast address
1189  * list or the network interface flags are updated.  This routine is
1190  * responsible for configuring the hardware for proper multicast mode.
1191  **/
1192 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1193 {
1194 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1195 	struct ixgbe_hw *hw = &adapter->hw;
1196 
1197 	spin_lock(&adapter->mbx_lock);
1198 
1199 	/* reprogram multicast list */
1200 	if (hw->mac.ops.update_mc_addr_list)
1201 		hw->mac.ops.update_mc_addr_list(hw, netdev);
1202 
1203 	ixgbevf_write_uc_addr_list(netdev);
1204 
1205 	spin_unlock(&adapter->mbx_lock);
1206 }
1207 
1208 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1209 {
1210 	int q_idx;
1211 	struct ixgbevf_q_vector *q_vector;
1212 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1213 
1214 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1215 		q_vector = adapter->q_vector[q_idx];
1216 		napi_enable(&q_vector->napi);
1217 	}
1218 }
1219 
1220 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1221 {
1222 	int q_idx;
1223 	struct ixgbevf_q_vector *q_vector;
1224 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1225 
1226 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1227 		q_vector = adapter->q_vector[q_idx];
1228 		napi_disable(&q_vector->napi);
1229 	}
1230 }
1231 
1232 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1233 {
1234 	struct net_device *netdev = adapter->netdev;
1235 	int i;
1236 
1237 	ixgbevf_set_rx_mode(netdev);
1238 
1239 	ixgbevf_restore_vlan(adapter);
1240 
1241 	ixgbevf_configure_tx(adapter);
1242 	ixgbevf_configure_rx(adapter);
1243 	for (i = 0; i < adapter->num_rx_queues; i++) {
1244 		struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1245 		ixgbevf_alloc_rx_buffers(adapter, ring,
1246 					 IXGBE_DESC_UNUSED(ring));
1247 	}
1248 }
1249 
1250 #define IXGBE_MAX_RX_DESC_POLL 10
1251 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1252 						int rxr)
1253 {
1254 	struct ixgbe_hw *hw = &adapter->hw;
1255 	int j = adapter->rx_ring[rxr].reg_idx;
1256 	int k;
1257 
1258 	for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1259 		if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1260 			break;
1261 		else
1262 			msleep(1);
1263 	}
1264 	if (k >= IXGBE_MAX_RX_DESC_POLL) {
1265 		hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1266 		       "not set within the polling period\n", rxr);
1267 	}
1268 
1269 	ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1270 				(adapter->rx_ring[rxr].count - 1));
1271 }
1272 
1273 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1274 {
1275 	/* Only save pre-reset stats if there are some */
1276 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1277 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1278 			adapter->stats.base_vfgprc;
1279 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1280 			adapter->stats.base_vfgptc;
1281 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1282 			adapter->stats.base_vfgorc;
1283 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1284 			adapter->stats.base_vfgotc;
1285 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1286 			adapter->stats.base_vfmprc;
1287 	}
1288 }
1289 
1290 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1291 {
1292 	struct ixgbe_hw *hw = &adapter->hw;
1293 
1294 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1295 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1296 	adapter->stats.last_vfgorc |=
1297 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1298 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1299 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1300 	adapter->stats.last_vfgotc |=
1301 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1302 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1303 
1304 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1305 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1306 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1307 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1308 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1309 }
1310 
1311 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1312 {
1313 	struct net_device *netdev = adapter->netdev;
1314 	struct ixgbe_hw *hw = &adapter->hw;
1315 	int i, j = 0;
1316 	int num_rx_rings = adapter->num_rx_queues;
1317 	u32 txdctl, rxdctl;
1318 	u32 msg[2];
1319 
1320 	for (i = 0; i < adapter->num_tx_queues; i++) {
1321 		j = adapter->tx_ring[i].reg_idx;
1322 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1323 		/* enable WTHRESH=8 descriptors, to encourage burst writeback */
1324 		txdctl |= (8 << 16);
1325 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1326 	}
1327 
1328 	for (i = 0; i < adapter->num_tx_queues; i++) {
1329 		j = adapter->tx_ring[i].reg_idx;
1330 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1331 		txdctl |= IXGBE_TXDCTL_ENABLE;
1332 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1333 	}
1334 
1335 	for (i = 0; i < num_rx_rings; i++) {
1336 		j = adapter->rx_ring[i].reg_idx;
1337 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1338 		rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1339 		if (hw->mac.type == ixgbe_mac_X540_vf) {
1340 			rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1341 			rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1342 				   IXGBE_RXDCTL_RLPML_EN);
1343 		}
1344 		IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1345 		ixgbevf_rx_desc_queue_enable(adapter, i);
1346 	}
1347 
1348 	ixgbevf_configure_msix(adapter);
1349 
1350 	spin_lock(&adapter->mbx_lock);
1351 
1352 	if (hw->mac.ops.set_rar) {
1353 		if (is_valid_ether_addr(hw->mac.addr))
1354 			hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1355 		else
1356 			hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1357 	}
1358 
1359 	msg[0] = IXGBE_VF_SET_LPE;
1360 	msg[1] = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1361 	hw->mbx.ops.write_posted(hw, msg, 2);
1362 
1363 	spin_unlock(&adapter->mbx_lock);
1364 
1365 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
1366 	ixgbevf_napi_enable_all(adapter);
1367 
1368 	/* enable transmits */
1369 	netif_tx_start_all_queues(netdev);
1370 
1371 	ixgbevf_save_reset_stats(adapter);
1372 	ixgbevf_init_last_counter_stats(adapter);
1373 
1374 	mod_timer(&adapter->watchdog_timer, jiffies);
1375 }
1376 
1377 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1378 {
1379 	struct ixgbe_hw *hw = &adapter->hw;
1380 
1381 	ixgbevf_configure(adapter);
1382 
1383 	ixgbevf_up_complete(adapter);
1384 
1385 	/* clear any pending interrupts, may auto mask */
1386 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
1387 
1388 	ixgbevf_irq_enable(adapter);
1389 }
1390 
1391 /**
1392  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1393  * @adapter: board private structure
1394  * @rx_ring: ring to free buffers from
1395  **/
1396 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1397 				  struct ixgbevf_ring *rx_ring)
1398 {
1399 	struct pci_dev *pdev = adapter->pdev;
1400 	unsigned long size;
1401 	unsigned int i;
1402 
1403 	if (!rx_ring->rx_buffer_info)
1404 		return;
1405 
1406 	/* Free all the Rx ring sk_buffs */
1407 	for (i = 0; i < rx_ring->count; i++) {
1408 		struct ixgbevf_rx_buffer *rx_buffer_info;
1409 
1410 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
1411 		if (rx_buffer_info->dma) {
1412 			dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1413 					 rx_ring->rx_buf_len,
1414 					 DMA_FROM_DEVICE);
1415 			rx_buffer_info->dma = 0;
1416 		}
1417 		if (rx_buffer_info->skb) {
1418 			struct sk_buff *skb = rx_buffer_info->skb;
1419 			rx_buffer_info->skb = NULL;
1420 			do {
1421 				struct sk_buff *this = skb;
1422 				skb = skb->prev;
1423 				dev_kfree_skb(this);
1424 			} while (skb);
1425 		}
1426 	}
1427 
1428 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1429 	memset(rx_ring->rx_buffer_info, 0, size);
1430 
1431 	/* Zero out the descriptor ring */
1432 	memset(rx_ring->desc, 0, rx_ring->size);
1433 
1434 	rx_ring->next_to_clean = 0;
1435 	rx_ring->next_to_use = 0;
1436 
1437 	if (rx_ring->head)
1438 		writel(0, adapter->hw.hw_addr + rx_ring->head);
1439 	if (rx_ring->tail)
1440 		writel(0, adapter->hw.hw_addr + rx_ring->tail);
1441 }
1442 
1443 /**
1444  * ixgbevf_clean_tx_ring - Free Tx Buffers
1445  * @adapter: board private structure
1446  * @tx_ring: ring to be cleaned
1447  **/
1448 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1449 				  struct ixgbevf_ring *tx_ring)
1450 {
1451 	struct ixgbevf_tx_buffer *tx_buffer_info;
1452 	unsigned long size;
1453 	unsigned int i;
1454 
1455 	if (!tx_ring->tx_buffer_info)
1456 		return;
1457 
1458 	/* Free all the Tx ring sk_buffs */
1459 
1460 	for (i = 0; i < tx_ring->count; i++) {
1461 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
1462 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1463 	}
1464 
1465 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1466 	memset(tx_ring->tx_buffer_info, 0, size);
1467 
1468 	memset(tx_ring->desc, 0, tx_ring->size);
1469 
1470 	tx_ring->next_to_use = 0;
1471 	tx_ring->next_to_clean = 0;
1472 
1473 	if (tx_ring->head)
1474 		writel(0, adapter->hw.hw_addr + tx_ring->head);
1475 	if (tx_ring->tail)
1476 		writel(0, adapter->hw.hw_addr + tx_ring->tail);
1477 }
1478 
1479 /**
1480  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1481  * @adapter: board private structure
1482  **/
1483 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1484 {
1485 	int i;
1486 
1487 	for (i = 0; i < adapter->num_rx_queues; i++)
1488 		ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1489 }
1490 
1491 /**
1492  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1493  * @adapter: board private structure
1494  **/
1495 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1496 {
1497 	int i;
1498 
1499 	for (i = 0; i < adapter->num_tx_queues; i++)
1500 		ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1501 }
1502 
1503 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1504 {
1505 	struct net_device *netdev = adapter->netdev;
1506 	struct ixgbe_hw *hw = &adapter->hw;
1507 	u32 txdctl;
1508 	int i, j;
1509 
1510 	/* signal that we are down to the interrupt handler */
1511 	set_bit(__IXGBEVF_DOWN, &adapter->state);
1512 	/* disable receives */
1513 
1514 	netif_tx_disable(netdev);
1515 
1516 	msleep(10);
1517 
1518 	netif_tx_stop_all_queues(netdev);
1519 
1520 	ixgbevf_irq_disable(adapter);
1521 
1522 	ixgbevf_napi_disable_all(adapter);
1523 
1524 	del_timer_sync(&adapter->watchdog_timer);
1525 	/* can't call flush scheduled work here because it can deadlock
1526 	 * if linkwatch_event tries to acquire the rtnl_lock which we are
1527 	 * holding */
1528 	while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1529 		msleep(1);
1530 
1531 	/* disable transmits in the hardware now that interrupts are off */
1532 	for (i = 0; i < adapter->num_tx_queues; i++) {
1533 		j = adapter->tx_ring[i].reg_idx;
1534 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1535 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1536 				(txdctl & ~IXGBE_TXDCTL_ENABLE));
1537 	}
1538 
1539 	netif_carrier_off(netdev);
1540 
1541 	if (!pci_channel_offline(adapter->pdev))
1542 		ixgbevf_reset(adapter);
1543 
1544 	ixgbevf_clean_all_tx_rings(adapter);
1545 	ixgbevf_clean_all_rx_rings(adapter);
1546 }
1547 
1548 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1549 {
1550 	struct ixgbe_hw *hw = &adapter->hw;
1551 
1552 	WARN_ON(in_interrupt());
1553 
1554 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1555 		msleep(1);
1556 
1557 	/*
1558 	 * Check if PF is up before re-init.  If not then skip until
1559 	 * later when the PF is up and ready to service requests from
1560 	 * the VF via mailbox.  If the VF is up and running then the
1561 	 * watchdog task will continue to schedule reset tasks until
1562 	 * the PF is up and running.
1563 	 */
1564 	if (!hw->mac.ops.reset_hw(hw)) {
1565 		ixgbevf_down(adapter);
1566 		ixgbevf_up(adapter);
1567 	}
1568 
1569 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1570 }
1571 
1572 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1573 {
1574 	struct ixgbe_hw *hw = &adapter->hw;
1575 	struct net_device *netdev = adapter->netdev;
1576 
1577 	spin_lock(&adapter->mbx_lock);
1578 
1579 	if (hw->mac.ops.reset_hw(hw))
1580 		hw_dbg(hw, "PF still resetting\n");
1581 	else
1582 		hw->mac.ops.init_hw(hw);
1583 
1584 	spin_unlock(&adapter->mbx_lock);
1585 
1586 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1587 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1588 		       netdev->addr_len);
1589 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1590 		       netdev->addr_len);
1591 	}
1592 }
1593 
1594 static void ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1595 					 int vectors)
1596 {
1597 	int err, vector_threshold;
1598 
1599 	/* We'll want at least 2 (vector_threshold):
1600 	 * 1) TxQ[0] + RxQ[0] handler
1601 	 * 2) Other (Link Status Change, etc.)
1602 	 */
1603 	vector_threshold = MIN_MSIX_COUNT;
1604 
1605 	/* The more we get, the more we will assign to Tx/Rx Cleanup
1606 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1607 	 * Right now, we simply care about how many we'll get; we'll
1608 	 * set them up later while requesting irq's.
1609 	 */
1610 	while (vectors >= vector_threshold) {
1611 		err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1612 				      vectors);
1613 		if (!err) /* Success in acquiring all requested vectors. */
1614 			break;
1615 		else if (err < 0)
1616 			vectors = 0; /* Nasty failure, quit now */
1617 		else /* err == number of vectors we should try again with */
1618 			vectors = err;
1619 	}
1620 
1621 	if (vectors < vector_threshold) {
1622 		/* Can't allocate enough MSI-X interrupts?  Oh well.
1623 		 * This just means we'll go with either a single MSI
1624 		 * vector or fall back to legacy interrupts.
1625 		 */
1626 		hw_dbg(&adapter->hw,
1627 		       "Unable to allocate MSI-X interrupts\n");
1628 		kfree(adapter->msix_entries);
1629 		adapter->msix_entries = NULL;
1630 	} else {
1631 		/*
1632 		 * Adjust for only the vectors we'll use, which is minimum
1633 		 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1634 		 * vectors we were allocated.
1635 		 */
1636 		adapter->num_msix_vectors = vectors;
1637 	}
1638 }
1639 
1640 /**
1641  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1642  * @adapter: board private structure to initialize
1643  *
1644  * This is the top level queue allocation routine.  The order here is very
1645  * important, starting with the "most" number of features turned on at once,
1646  * and ending with the smallest set of features.  This way large combinations
1647  * can be allocated if they're turned on, and smaller combinations are the
1648  * fallthrough conditions.
1649  *
1650  **/
1651 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1652 {
1653 	/* Start with base case */
1654 	adapter->num_rx_queues = 1;
1655 	adapter->num_tx_queues = 1;
1656 }
1657 
1658 /**
1659  * ixgbevf_alloc_queues - Allocate memory for all rings
1660  * @adapter: board private structure to initialize
1661  *
1662  * We allocate one ring per queue at run-time since we don't know the
1663  * number of queues at compile-time.  The polling_netdev array is
1664  * intended for Multiqueue, but should work fine with a single queue.
1665  **/
1666 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1667 {
1668 	int i;
1669 
1670 	adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1671 				   sizeof(struct ixgbevf_ring), GFP_KERNEL);
1672 	if (!adapter->tx_ring)
1673 		goto err_tx_ring_allocation;
1674 
1675 	adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1676 				   sizeof(struct ixgbevf_ring), GFP_KERNEL);
1677 	if (!adapter->rx_ring)
1678 		goto err_rx_ring_allocation;
1679 
1680 	for (i = 0; i < adapter->num_tx_queues; i++) {
1681 		adapter->tx_ring[i].count = adapter->tx_ring_count;
1682 		adapter->tx_ring[i].queue_index = i;
1683 		adapter->tx_ring[i].reg_idx = i;
1684 		adapter->tx_ring[i].dev = &adapter->pdev->dev;
1685 		adapter->tx_ring[i].netdev = adapter->netdev;
1686 	}
1687 
1688 	for (i = 0; i < adapter->num_rx_queues; i++) {
1689 		adapter->rx_ring[i].count = adapter->rx_ring_count;
1690 		adapter->rx_ring[i].queue_index = i;
1691 		adapter->rx_ring[i].reg_idx = i;
1692 		adapter->rx_ring[i].dev = &adapter->pdev->dev;
1693 		adapter->rx_ring[i].netdev = adapter->netdev;
1694 	}
1695 
1696 	return 0;
1697 
1698 err_rx_ring_allocation:
1699 	kfree(adapter->tx_ring);
1700 err_tx_ring_allocation:
1701 	return -ENOMEM;
1702 }
1703 
1704 /**
1705  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1706  * @adapter: board private structure to initialize
1707  *
1708  * Attempt to configure the interrupts using the best available
1709  * capabilities of the hardware and the kernel.
1710  **/
1711 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1712 {
1713 	int err = 0;
1714 	int vector, v_budget;
1715 
1716 	/*
1717 	 * It's easy to be greedy for MSI-X vectors, but it really
1718 	 * doesn't do us much good if we have a lot more vectors
1719 	 * than CPU's.  So let's be conservative and only ask for
1720 	 * (roughly) the same number of vectors as there are CPU's.
1721 	 * The default is to use pairs of vectors.
1722 	 */
1723 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1724 	v_budget = min_t(int, v_budget, num_online_cpus());
1725 	v_budget += NON_Q_VECTORS;
1726 
1727 	/* A failure in MSI-X entry allocation isn't fatal, but it does
1728 	 * mean we disable MSI-X capabilities of the adapter. */
1729 	adapter->msix_entries = kcalloc(v_budget,
1730 					sizeof(struct msix_entry), GFP_KERNEL);
1731 	if (!adapter->msix_entries) {
1732 		err = -ENOMEM;
1733 		goto out;
1734 	}
1735 
1736 	for (vector = 0; vector < v_budget; vector++)
1737 		adapter->msix_entries[vector].entry = vector;
1738 
1739 	ixgbevf_acquire_msix_vectors(adapter, v_budget);
1740 
1741 out:
1742 	return err;
1743 }
1744 
1745 /**
1746  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1747  * @adapter: board private structure to initialize
1748  *
1749  * We allocate one q_vector per queue interrupt.  If allocation fails we
1750  * return -ENOMEM.
1751  **/
1752 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1753 {
1754 	int q_idx, num_q_vectors;
1755 	struct ixgbevf_q_vector *q_vector;
1756 
1757 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1758 
1759 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1760 		q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1761 		if (!q_vector)
1762 			goto err_out;
1763 		q_vector->adapter = adapter;
1764 		q_vector->v_idx = q_idx;
1765 		netif_napi_add(adapter->netdev, &q_vector->napi,
1766 			       ixgbevf_poll, 64);
1767 		adapter->q_vector[q_idx] = q_vector;
1768 	}
1769 
1770 	return 0;
1771 
1772 err_out:
1773 	while (q_idx) {
1774 		q_idx--;
1775 		q_vector = adapter->q_vector[q_idx];
1776 		netif_napi_del(&q_vector->napi);
1777 		kfree(q_vector);
1778 		adapter->q_vector[q_idx] = NULL;
1779 	}
1780 	return -ENOMEM;
1781 }
1782 
1783 /**
1784  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1785  * @adapter: board private structure to initialize
1786  *
1787  * This function frees the memory allocated to the q_vectors.  In addition if
1788  * NAPI is enabled it will delete any references to the NAPI struct prior
1789  * to freeing the q_vector.
1790  **/
1791 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1792 {
1793 	int q_idx, num_q_vectors;
1794 	int napi_vectors;
1795 
1796 	num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1797 	napi_vectors = adapter->num_rx_queues;
1798 
1799 	for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1800 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1801 
1802 		adapter->q_vector[q_idx] = NULL;
1803 		if (q_idx < napi_vectors)
1804 			netif_napi_del(&q_vector->napi);
1805 		kfree(q_vector);
1806 	}
1807 }
1808 
1809 /**
1810  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1811  * @adapter: board private structure
1812  *
1813  **/
1814 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1815 {
1816 	pci_disable_msix(adapter->pdev);
1817 	kfree(adapter->msix_entries);
1818 	adapter->msix_entries = NULL;
1819 }
1820 
1821 /**
1822  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1823  * @adapter: board private structure to initialize
1824  *
1825  **/
1826 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1827 {
1828 	int err;
1829 
1830 	/* Number of supported queues */
1831 	ixgbevf_set_num_queues(adapter);
1832 
1833 	err = ixgbevf_set_interrupt_capability(adapter);
1834 	if (err) {
1835 		hw_dbg(&adapter->hw,
1836 		       "Unable to setup interrupt capabilities\n");
1837 		goto err_set_interrupt;
1838 	}
1839 
1840 	err = ixgbevf_alloc_q_vectors(adapter);
1841 	if (err) {
1842 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
1843 		       "vectors\n");
1844 		goto err_alloc_q_vectors;
1845 	}
1846 
1847 	err = ixgbevf_alloc_queues(adapter);
1848 	if (err) {
1849 		pr_err("Unable to allocate memory for queues\n");
1850 		goto err_alloc_queues;
1851 	}
1852 
1853 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
1854 	       "Tx Queue count = %u\n",
1855 	       (adapter->num_rx_queues > 1) ? "Enabled" :
1856 	       "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
1857 
1858 	set_bit(__IXGBEVF_DOWN, &adapter->state);
1859 
1860 	return 0;
1861 err_alloc_queues:
1862 	ixgbevf_free_q_vectors(adapter);
1863 err_alloc_q_vectors:
1864 	ixgbevf_reset_interrupt_capability(adapter);
1865 err_set_interrupt:
1866 	return err;
1867 }
1868 
1869 /**
1870  * ixgbevf_sw_init - Initialize general software structures
1871  * (struct ixgbevf_adapter)
1872  * @adapter: board private structure to initialize
1873  *
1874  * ixgbevf_sw_init initializes the Adapter private data structure.
1875  * Fields are initialized based on PCI device information and
1876  * OS network device settings (MTU size).
1877  **/
1878 static int __devinit ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
1879 {
1880 	struct ixgbe_hw *hw = &adapter->hw;
1881 	struct pci_dev *pdev = adapter->pdev;
1882 	int err;
1883 
1884 	/* PCI config space info */
1885 
1886 	hw->vendor_id = pdev->vendor;
1887 	hw->device_id = pdev->device;
1888 	hw->revision_id = pdev->revision;
1889 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
1890 	hw->subsystem_device_id = pdev->subsystem_device;
1891 
1892 	hw->mbx.ops.init_params(hw);
1893 	hw->mac.max_tx_queues = MAX_TX_QUEUES;
1894 	hw->mac.max_rx_queues = MAX_RX_QUEUES;
1895 	err = hw->mac.ops.reset_hw(hw);
1896 	if (err) {
1897 		dev_info(&pdev->dev,
1898 		         "PF still in reset state, assigning new address\n");
1899 		eth_hw_addr_random(adapter->netdev);
1900 		memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
1901 			adapter->netdev->addr_len);
1902 	} else {
1903 		err = hw->mac.ops.init_hw(hw);
1904 		if (err) {
1905 			pr_err("init_shared_code failed: %d\n", err);
1906 			goto out;
1907 		}
1908 		memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
1909 			adapter->netdev->addr_len);
1910 	}
1911 
1912 	/* lock to protect mailbox accesses */
1913 	spin_lock_init(&adapter->mbx_lock);
1914 
1915 	/* Enable dynamic interrupt throttling rates */
1916 	adapter->rx_itr_setting = 1;
1917 	adapter->tx_itr_setting = 1;
1918 
1919 	/* set default ring sizes */
1920 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
1921 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
1922 
1923 	set_bit(__IXGBEVF_DOWN, &adapter->state);
1924 	return 0;
1925 
1926 out:
1927 	return err;
1928 }
1929 
1930 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
1931 	{							\
1932 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
1933 		if (current_counter < last_counter)		\
1934 			counter += 0x100000000LL;		\
1935 		last_counter = current_counter;			\
1936 		counter &= 0xFFFFFFFF00000000LL;		\
1937 		counter |= current_counter;			\
1938 	}
1939 
1940 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
1941 	{								 \
1942 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
1943 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
1944 		u64 current_counter = (current_counter_msb << 32) |      \
1945 			current_counter_lsb;                             \
1946 		if (current_counter < last_counter)			 \
1947 			counter += 0x1000000000LL;			 \
1948 		last_counter = current_counter;				 \
1949 		counter &= 0xFFFFFFF000000000LL;			 \
1950 		counter |= current_counter;				 \
1951 	}
1952 /**
1953  * ixgbevf_update_stats - Update the board statistics counters.
1954  * @adapter: board private structure
1955  **/
1956 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
1957 {
1958 	struct ixgbe_hw *hw = &adapter->hw;
1959 
1960 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
1961 				adapter->stats.vfgprc);
1962 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
1963 				adapter->stats.vfgptc);
1964 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
1965 				adapter->stats.last_vfgorc,
1966 				adapter->stats.vfgorc);
1967 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
1968 				adapter->stats.last_vfgotc,
1969 				adapter->stats.vfgotc);
1970 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
1971 				adapter->stats.vfmprc);
1972 }
1973 
1974 /**
1975  * ixgbevf_watchdog - Timer Call-back
1976  * @data: pointer to adapter cast into an unsigned long
1977  **/
1978 static void ixgbevf_watchdog(unsigned long data)
1979 {
1980 	struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
1981 	struct ixgbe_hw *hw = &adapter->hw;
1982 	u32 eics = 0;
1983 	int i;
1984 
1985 	/*
1986 	 * Do the watchdog outside of interrupt context due to the lovely
1987 	 * delays that some of the newer hardware requires
1988 	 */
1989 
1990 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1991 		goto watchdog_short_circuit;
1992 
1993 	/* get one bit for every active tx/rx interrupt vector */
1994 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
1995 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
1996 		if (qv->rx.ring || qv->tx.ring)
1997 			eics |= 1 << i;
1998 	}
1999 
2000 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2001 
2002 watchdog_short_circuit:
2003 	schedule_work(&adapter->watchdog_task);
2004 }
2005 
2006 /**
2007  * ixgbevf_tx_timeout - Respond to a Tx Hang
2008  * @netdev: network interface device structure
2009  **/
2010 static void ixgbevf_tx_timeout(struct net_device *netdev)
2011 {
2012 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2013 
2014 	/* Do the reset outside of interrupt context */
2015 	schedule_work(&adapter->reset_task);
2016 }
2017 
2018 static void ixgbevf_reset_task(struct work_struct *work)
2019 {
2020 	struct ixgbevf_adapter *adapter;
2021 	adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2022 
2023 	/* If we're already down or resetting, just bail */
2024 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2025 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
2026 		return;
2027 
2028 	adapter->tx_timeout_count++;
2029 
2030 	ixgbevf_reinit_locked(adapter);
2031 }
2032 
2033 /**
2034  * ixgbevf_watchdog_task - worker thread to bring link up
2035  * @work: pointer to work_struct containing our data
2036  **/
2037 static void ixgbevf_watchdog_task(struct work_struct *work)
2038 {
2039 	struct ixgbevf_adapter *adapter = container_of(work,
2040 						       struct ixgbevf_adapter,
2041 						       watchdog_task);
2042 	struct net_device *netdev = adapter->netdev;
2043 	struct ixgbe_hw *hw = &adapter->hw;
2044 	u32 link_speed = adapter->link_speed;
2045 	bool link_up = adapter->link_up;
2046 
2047 	adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2048 
2049 	/*
2050 	 * Always check the link on the watchdog because we have
2051 	 * no LSC interrupt
2052 	 */
2053 	if (hw->mac.ops.check_link) {
2054 		s32 need_reset;
2055 
2056 		spin_lock(&adapter->mbx_lock);
2057 
2058 		need_reset = hw->mac.ops.check_link(hw, &link_speed,
2059 						    &link_up, false);
2060 
2061 		spin_unlock(&adapter->mbx_lock);
2062 
2063 		if (need_reset) {
2064 			adapter->link_up = link_up;
2065 			adapter->link_speed = link_speed;
2066 			netif_carrier_off(netdev);
2067 			netif_tx_stop_all_queues(netdev);
2068 			schedule_work(&adapter->reset_task);
2069 			goto pf_has_reset;
2070 		}
2071 	} else {
2072 		/* always assume link is up, if no check link
2073 		 * function */
2074 		link_speed = IXGBE_LINK_SPEED_10GB_FULL;
2075 		link_up = true;
2076 	}
2077 	adapter->link_up = link_up;
2078 	adapter->link_speed = link_speed;
2079 
2080 	if (link_up) {
2081 		if (!netif_carrier_ok(netdev)) {
2082 			hw_dbg(&adapter->hw, "NIC Link is Up, %u Gbps\n",
2083 			       (link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2084 			       10 : 1);
2085 			netif_carrier_on(netdev);
2086 			netif_tx_wake_all_queues(netdev);
2087 		}
2088 	} else {
2089 		adapter->link_up = false;
2090 		adapter->link_speed = 0;
2091 		if (netif_carrier_ok(netdev)) {
2092 			hw_dbg(&adapter->hw, "NIC Link is Down\n");
2093 			netif_carrier_off(netdev);
2094 			netif_tx_stop_all_queues(netdev);
2095 		}
2096 	}
2097 
2098 	ixgbevf_update_stats(adapter);
2099 
2100 pf_has_reset:
2101 	/* Reset the timer */
2102 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2103 		mod_timer(&adapter->watchdog_timer,
2104 			  round_jiffies(jiffies + (2 * HZ)));
2105 
2106 	adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2107 }
2108 
2109 /**
2110  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2111  * @adapter: board private structure
2112  * @tx_ring: Tx descriptor ring for a specific queue
2113  *
2114  * Free all transmit software resources
2115  **/
2116 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2117 			       struct ixgbevf_ring *tx_ring)
2118 {
2119 	struct pci_dev *pdev = adapter->pdev;
2120 
2121 	ixgbevf_clean_tx_ring(adapter, tx_ring);
2122 
2123 	vfree(tx_ring->tx_buffer_info);
2124 	tx_ring->tx_buffer_info = NULL;
2125 
2126 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2127 			  tx_ring->dma);
2128 
2129 	tx_ring->desc = NULL;
2130 }
2131 
2132 /**
2133  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2134  * @adapter: board private structure
2135  *
2136  * Free all transmit software resources
2137  **/
2138 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2139 {
2140 	int i;
2141 
2142 	for (i = 0; i < adapter->num_tx_queues; i++)
2143 		if (adapter->tx_ring[i].desc)
2144 			ixgbevf_free_tx_resources(adapter,
2145 						  &adapter->tx_ring[i]);
2146 
2147 }
2148 
2149 /**
2150  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2151  * @adapter: board private structure
2152  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2153  *
2154  * Return 0 on success, negative on failure
2155  **/
2156 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2157 			       struct ixgbevf_ring *tx_ring)
2158 {
2159 	struct pci_dev *pdev = adapter->pdev;
2160 	int size;
2161 
2162 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2163 	tx_ring->tx_buffer_info = vzalloc(size);
2164 	if (!tx_ring->tx_buffer_info)
2165 		goto err;
2166 
2167 	/* round up to nearest 4K */
2168 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2169 	tx_ring->size = ALIGN(tx_ring->size, 4096);
2170 
2171 	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2172 					   &tx_ring->dma, GFP_KERNEL);
2173 	if (!tx_ring->desc)
2174 		goto err;
2175 
2176 	tx_ring->next_to_use = 0;
2177 	tx_ring->next_to_clean = 0;
2178 	return 0;
2179 
2180 err:
2181 	vfree(tx_ring->tx_buffer_info);
2182 	tx_ring->tx_buffer_info = NULL;
2183 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2184 	       "descriptor ring\n");
2185 	return -ENOMEM;
2186 }
2187 
2188 /**
2189  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2190  * @adapter: board private structure
2191  *
2192  * If this function returns with an error, then it's possible one or
2193  * more of the rings is populated (while the rest are not).  It is the
2194  * callers duty to clean those orphaned rings.
2195  *
2196  * Return 0 on success, negative on failure
2197  **/
2198 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2199 {
2200 	int i, err = 0;
2201 
2202 	for (i = 0; i < adapter->num_tx_queues; i++) {
2203 		err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2204 		if (!err)
2205 			continue;
2206 		hw_dbg(&adapter->hw,
2207 		       "Allocation for Tx Queue %u failed\n", i);
2208 		break;
2209 	}
2210 
2211 	return err;
2212 }
2213 
2214 /**
2215  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2216  * @adapter: board private structure
2217  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2218  *
2219  * Returns 0 on success, negative on failure
2220  **/
2221 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2222 			       struct ixgbevf_ring *rx_ring)
2223 {
2224 	struct pci_dev *pdev = adapter->pdev;
2225 	int size;
2226 
2227 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2228 	rx_ring->rx_buffer_info = vzalloc(size);
2229 	if (!rx_ring->rx_buffer_info)
2230 		goto alloc_failed;
2231 
2232 	/* Round up to nearest 4K */
2233 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2234 	rx_ring->size = ALIGN(rx_ring->size, 4096);
2235 
2236 	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2237 					   &rx_ring->dma, GFP_KERNEL);
2238 
2239 	if (!rx_ring->desc) {
2240 		hw_dbg(&adapter->hw,
2241 		       "Unable to allocate memory for "
2242 		       "the receive descriptor ring\n");
2243 		vfree(rx_ring->rx_buffer_info);
2244 		rx_ring->rx_buffer_info = NULL;
2245 		goto alloc_failed;
2246 	}
2247 
2248 	rx_ring->next_to_clean = 0;
2249 	rx_ring->next_to_use = 0;
2250 
2251 	return 0;
2252 alloc_failed:
2253 	return -ENOMEM;
2254 }
2255 
2256 /**
2257  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2258  * @adapter: board private structure
2259  *
2260  * If this function returns with an error, then it's possible one or
2261  * more of the rings is populated (while the rest are not).  It is the
2262  * callers duty to clean those orphaned rings.
2263  *
2264  * Return 0 on success, negative on failure
2265  **/
2266 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2267 {
2268 	int i, err = 0;
2269 
2270 	for (i = 0; i < adapter->num_rx_queues; i++) {
2271 		err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2272 		if (!err)
2273 			continue;
2274 		hw_dbg(&adapter->hw,
2275 		       "Allocation for Rx Queue %u failed\n", i);
2276 		break;
2277 	}
2278 	return err;
2279 }
2280 
2281 /**
2282  * ixgbevf_free_rx_resources - Free Rx Resources
2283  * @adapter: board private structure
2284  * @rx_ring: ring to clean the resources from
2285  *
2286  * Free all receive software resources
2287  **/
2288 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2289 			       struct ixgbevf_ring *rx_ring)
2290 {
2291 	struct pci_dev *pdev = adapter->pdev;
2292 
2293 	ixgbevf_clean_rx_ring(adapter, rx_ring);
2294 
2295 	vfree(rx_ring->rx_buffer_info);
2296 	rx_ring->rx_buffer_info = NULL;
2297 
2298 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2299 			  rx_ring->dma);
2300 
2301 	rx_ring->desc = NULL;
2302 }
2303 
2304 /**
2305  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2306  * @adapter: board private structure
2307  *
2308  * Free all receive software resources
2309  **/
2310 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2311 {
2312 	int i;
2313 
2314 	for (i = 0; i < adapter->num_rx_queues; i++)
2315 		if (adapter->rx_ring[i].desc)
2316 			ixgbevf_free_rx_resources(adapter,
2317 						  &adapter->rx_ring[i]);
2318 }
2319 
2320 /**
2321  * ixgbevf_open - Called when a network interface is made active
2322  * @netdev: network interface device structure
2323  *
2324  * Returns 0 on success, negative value on failure
2325  *
2326  * The open entry point is called when a network interface is made
2327  * active by the system (IFF_UP).  At this point all resources needed
2328  * for transmit and receive operations are allocated, the interrupt
2329  * handler is registered with the OS, the watchdog timer is started,
2330  * and the stack is notified that the interface is ready.
2331  **/
2332 static int ixgbevf_open(struct net_device *netdev)
2333 {
2334 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2335 	struct ixgbe_hw *hw = &adapter->hw;
2336 	int err;
2337 
2338 	/* disallow open during test */
2339 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2340 		return -EBUSY;
2341 
2342 	if (hw->adapter_stopped) {
2343 		ixgbevf_reset(adapter);
2344 		/* if adapter is still stopped then PF isn't up and
2345 		 * the vf can't start. */
2346 		if (hw->adapter_stopped) {
2347 			err = IXGBE_ERR_MBX;
2348 			pr_err("Unable to start - perhaps the PF Driver isn't "
2349 			       "up yet\n");
2350 			goto err_setup_reset;
2351 		}
2352 	}
2353 
2354 	/* allocate transmit descriptors */
2355 	err = ixgbevf_setup_all_tx_resources(adapter);
2356 	if (err)
2357 		goto err_setup_tx;
2358 
2359 	/* allocate receive descriptors */
2360 	err = ixgbevf_setup_all_rx_resources(adapter);
2361 	if (err)
2362 		goto err_setup_rx;
2363 
2364 	ixgbevf_configure(adapter);
2365 
2366 	/*
2367 	 * Map the Tx/Rx rings to the vectors we were allotted.
2368 	 * if request_irq will be called in this function map_rings
2369 	 * must be called *before* up_complete
2370 	 */
2371 	ixgbevf_map_rings_to_vectors(adapter);
2372 
2373 	ixgbevf_up_complete(adapter);
2374 
2375 	/* clear any pending interrupts, may auto mask */
2376 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2377 	err = ixgbevf_request_irq(adapter);
2378 	if (err)
2379 		goto err_req_irq;
2380 
2381 	ixgbevf_irq_enable(adapter);
2382 
2383 	return 0;
2384 
2385 err_req_irq:
2386 	ixgbevf_down(adapter);
2387 	ixgbevf_free_irq(adapter);
2388 err_setup_rx:
2389 	ixgbevf_free_all_rx_resources(adapter);
2390 err_setup_tx:
2391 	ixgbevf_free_all_tx_resources(adapter);
2392 	ixgbevf_reset(adapter);
2393 
2394 err_setup_reset:
2395 
2396 	return err;
2397 }
2398 
2399 /**
2400  * ixgbevf_close - Disables a network interface
2401  * @netdev: network interface device structure
2402  *
2403  * Returns 0, this is not allowed to fail
2404  *
2405  * The close entry point is called when an interface is de-activated
2406  * by the OS.  The hardware is still under the drivers control, but
2407  * needs to be disabled.  A global MAC reset is issued to stop the
2408  * hardware, and all transmit and receive resources are freed.
2409  **/
2410 static int ixgbevf_close(struct net_device *netdev)
2411 {
2412 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2413 
2414 	ixgbevf_down(adapter);
2415 	ixgbevf_free_irq(adapter);
2416 
2417 	ixgbevf_free_all_tx_resources(adapter);
2418 	ixgbevf_free_all_rx_resources(adapter);
2419 
2420 	return 0;
2421 }
2422 
2423 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2424 				u32 vlan_macip_lens, u32 type_tucmd,
2425 				u32 mss_l4len_idx)
2426 {
2427 	struct ixgbe_adv_tx_context_desc *context_desc;
2428 	u16 i = tx_ring->next_to_use;
2429 
2430 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2431 
2432 	i++;
2433 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2434 
2435 	/* set bits to identify this as an advanced context descriptor */
2436 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2437 
2438 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
2439 	context_desc->seqnum_seed	= 0;
2440 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
2441 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
2442 }
2443 
2444 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2445 		       struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2446 {
2447 	u32 vlan_macip_lens, type_tucmd;
2448 	u32 mss_l4len_idx, l4len;
2449 
2450 	if (!skb_is_gso(skb))
2451 		return 0;
2452 
2453 	if (skb_header_cloned(skb)) {
2454 		int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2455 		if (err)
2456 			return err;
2457 	}
2458 
2459 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2460 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2461 
2462 	if (skb->protocol == htons(ETH_P_IP)) {
2463 		struct iphdr *iph = ip_hdr(skb);
2464 		iph->tot_len = 0;
2465 		iph->check = 0;
2466 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2467 							 iph->daddr, 0,
2468 							 IPPROTO_TCP,
2469 							 0);
2470 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2471 	} else if (skb_is_gso_v6(skb)) {
2472 		ipv6_hdr(skb)->payload_len = 0;
2473 		tcp_hdr(skb)->check =
2474 		    ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2475 				     &ipv6_hdr(skb)->daddr,
2476 				     0, IPPROTO_TCP, 0);
2477 	}
2478 
2479 	/* compute header lengths */
2480 	l4len = tcp_hdrlen(skb);
2481 	*hdr_len += l4len;
2482 	*hdr_len = skb_transport_offset(skb) + l4len;
2483 
2484 	/* mss_l4len_id: use 1 as index for TSO */
2485 	mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2486 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2487 	mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2488 
2489 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2490 	vlan_macip_lens = skb_network_header_len(skb);
2491 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2492 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2493 
2494 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2495 			    type_tucmd, mss_l4len_idx);
2496 
2497 	return 1;
2498 }
2499 
2500 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2501 			    struct sk_buff *skb, u32 tx_flags)
2502 {
2503 
2504 
2505 
2506 	u32 vlan_macip_lens = 0;
2507 	u32 mss_l4len_idx = 0;
2508 	u32 type_tucmd = 0;
2509 
2510 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2511 		u8 l4_hdr = 0;
2512 		switch (skb->protocol) {
2513 		case __constant_htons(ETH_P_IP):
2514 			vlan_macip_lens |= skb_network_header_len(skb);
2515 			type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2516 			l4_hdr = ip_hdr(skb)->protocol;
2517 			break;
2518 		case __constant_htons(ETH_P_IPV6):
2519 			vlan_macip_lens |= skb_network_header_len(skb);
2520 			l4_hdr = ipv6_hdr(skb)->nexthdr;
2521 			break;
2522 		default:
2523 			if (unlikely(net_ratelimit())) {
2524 				dev_warn(tx_ring->dev,
2525 				 "partial checksum but proto=%x!\n",
2526 				 skb->protocol);
2527 			}
2528 			break;
2529 		}
2530 
2531 		switch (l4_hdr) {
2532 		case IPPROTO_TCP:
2533 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2534 			mss_l4len_idx = tcp_hdrlen(skb) <<
2535 					IXGBE_ADVTXD_L4LEN_SHIFT;
2536 			break;
2537 		case IPPROTO_SCTP:
2538 			type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2539 			mss_l4len_idx = sizeof(struct sctphdr) <<
2540 					IXGBE_ADVTXD_L4LEN_SHIFT;
2541 			break;
2542 		case IPPROTO_UDP:
2543 			mss_l4len_idx = sizeof(struct udphdr) <<
2544 					IXGBE_ADVTXD_L4LEN_SHIFT;
2545 			break;
2546 		default:
2547 			if (unlikely(net_ratelimit())) {
2548 				dev_warn(tx_ring->dev,
2549 				 "partial checksum but l4 proto=%x!\n",
2550 				 l4_hdr);
2551 			}
2552 			break;
2553 		}
2554 	}
2555 
2556 	/* vlan_macip_lens: MACLEN, VLAN tag */
2557 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2558 	vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2559 
2560 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2561 			    type_tucmd, mss_l4len_idx);
2562 
2563 	return (skb->ip_summed == CHECKSUM_PARTIAL);
2564 }
2565 
2566 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2567 			  struct sk_buff *skb, u32 tx_flags,
2568 			  unsigned int first)
2569 {
2570 	struct ixgbevf_tx_buffer *tx_buffer_info;
2571 	unsigned int len;
2572 	unsigned int total = skb->len;
2573 	unsigned int offset = 0, size;
2574 	int count = 0;
2575 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2576 	unsigned int f;
2577 	int i;
2578 
2579 	i = tx_ring->next_to_use;
2580 
2581 	len = min(skb_headlen(skb), total);
2582 	while (len) {
2583 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2584 		size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2585 
2586 		tx_buffer_info->length = size;
2587 		tx_buffer_info->mapped_as_page = false;
2588 		tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2589 						     skb->data + offset,
2590 						     size, DMA_TO_DEVICE);
2591 		if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2592 			goto dma_error;
2593 		tx_buffer_info->next_to_watch = i;
2594 
2595 		len -= size;
2596 		total -= size;
2597 		offset += size;
2598 		count++;
2599 		i++;
2600 		if (i == tx_ring->count)
2601 			i = 0;
2602 	}
2603 
2604 	for (f = 0; f < nr_frags; f++) {
2605 		const struct skb_frag_struct *frag;
2606 
2607 		frag = &skb_shinfo(skb)->frags[f];
2608 		len = min((unsigned int)skb_frag_size(frag), total);
2609 		offset = 0;
2610 
2611 		while (len) {
2612 			tx_buffer_info = &tx_ring->tx_buffer_info[i];
2613 			size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2614 
2615 			tx_buffer_info->length = size;
2616 			tx_buffer_info->dma =
2617 				skb_frag_dma_map(tx_ring->dev, frag,
2618 						 offset, size, DMA_TO_DEVICE);
2619 			tx_buffer_info->mapped_as_page = true;
2620 			if (dma_mapping_error(tx_ring->dev,
2621 					      tx_buffer_info->dma))
2622 				goto dma_error;
2623 			tx_buffer_info->next_to_watch = i;
2624 
2625 			len -= size;
2626 			total -= size;
2627 			offset += size;
2628 			count++;
2629 			i++;
2630 			if (i == tx_ring->count)
2631 				i = 0;
2632 		}
2633 		if (total == 0)
2634 			break;
2635 	}
2636 
2637 	if (i == 0)
2638 		i = tx_ring->count - 1;
2639 	else
2640 		i = i - 1;
2641 	tx_ring->tx_buffer_info[i].skb = skb;
2642 	tx_ring->tx_buffer_info[first].next_to_watch = i;
2643 	tx_ring->tx_buffer_info[first].time_stamp = jiffies;
2644 
2645 	return count;
2646 
2647 dma_error:
2648 	dev_err(tx_ring->dev, "TX DMA map failed\n");
2649 
2650 	/* clear timestamp and dma mappings for failed tx_buffer_info map */
2651 	tx_buffer_info->dma = 0;
2652 	tx_buffer_info->next_to_watch = 0;
2653 	count--;
2654 
2655 	/* clear timestamp and dma mappings for remaining portion of packet */
2656 	while (count >= 0) {
2657 		count--;
2658 		i--;
2659 		if (i < 0)
2660 			i += tx_ring->count;
2661 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2662 		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2663 	}
2664 
2665 	return count;
2666 }
2667 
2668 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2669 			     int count, u32 paylen, u8 hdr_len)
2670 {
2671 	union ixgbe_adv_tx_desc *tx_desc = NULL;
2672 	struct ixgbevf_tx_buffer *tx_buffer_info;
2673 	u32 olinfo_status = 0, cmd_type_len = 0;
2674 	unsigned int i;
2675 
2676 	u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2677 
2678 	cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2679 
2680 	cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2681 
2682 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2683 		cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2684 
2685 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2686 		olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2687 
2688 	if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2689 		cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2690 
2691 		/* use index 1 context for tso */
2692 		olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2693 		if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2694 			olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2695 
2696 	}
2697 
2698 	/*
2699 	 * Check Context must be set if Tx switch is enabled, which it
2700 	 * always is for case where virtual functions are running
2701 	 */
2702 	olinfo_status |= IXGBE_ADVTXD_CC;
2703 
2704 	olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2705 
2706 	i = tx_ring->next_to_use;
2707 	while (count--) {
2708 		tx_buffer_info = &tx_ring->tx_buffer_info[i];
2709 		tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2710 		tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2711 		tx_desc->read.cmd_type_len =
2712 			cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2713 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2714 		i++;
2715 		if (i == tx_ring->count)
2716 			i = 0;
2717 	}
2718 
2719 	tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2720 
2721 	tx_ring->next_to_use = i;
2722 }
2723 
2724 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2725 {
2726 	struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2727 
2728 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2729 	/* Herbert's original patch had:
2730 	 *  smp_mb__after_netif_stop_queue();
2731 	 * but since that doesn't exist yet, just open code it. */
2732 	smp_mb();
2733 
2734 	/* We need to check again in a case another CPU has just
2735 	 * made room available. */
2736 	if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
2737 		return -EBUSY;
2738 
2739 	/* A reprieve! - use start_queue because it doesn't call schedule */
2740 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2741 	++adapter->restart_queue;
2742 	return 0;
2743 }
2744 
2745 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2746 {
2747 	if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
2748 		return 0;
2749 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
2750 }
2751 
2752 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2753 {
2754 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2755 	struct ixgbevf_ring *tx_ring;
2756 	unsigned int first;
2757 	unsigned int tx_flags = 0;
2758 	u8 hdr_len = 0;
2759 	int r_idx = 0, tso;
2760 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
2761 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
2762 	unsigned short f;
2763 #endif
2764 
2765 	tx_ring = &adapter->tx_ring[r_idx];
2766 
2767 	/*
2768 	 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
2769 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
2770 	 *       + 2 desc gap to keep tail from touching head,
2771 	 *       + 1 desc for context descriptor,
2772 	 * otherwise try next time
2773 	 */
2774 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
2775 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
2776 		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
2777 #else
2778 	count += skb_shinfo(skb)->nr_frags;
2779 #endif
2780 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
2781 		adapter->tx_busy++;
2782 		return NETDEV_TX_BUSY;
2783 	}
2784 
2785 	if (vlan_tx_tag_present(skb)) {
2786 		tx_flags |= vlan_tx_tag_get(skb);
2787 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
2788 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
2789 	}
2790 
2791 	first = tx_ring->next_to_use;
2792 
2793 	if (skb->protocol == htons(ETH_P_IP))
2794 		tx_flags |= IXGBE_TX_FLAGS_IPV4;
2795 	tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
2796 	if (tso < 0) {
2797 		dev_kfree_skb_any(skb);
2798 		return NETDEV_TX_OK;
2799 	}
2800 
2801 	if (tso)
2802 		tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
2803 	else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
2804 		tx_flags |= IXGBE_TX_FLAGS_CSUM;
2805 
2806 	ixgbevf_tx_queue(tx_ring, tx_flags,
2807 			 ixgbevf_tx_map(tx_ring, skb, tx_flags, first),
2808 			 skb->len, hdr_len);
2809 	/*
2810 	 * Force memory writes to complete before letting h/w
2811 	 * know there are new descriptors to fetch.  (Only
2812 	 * applicable for weak-ordered memory model archs,
2813 	 * such as IA-64).
2814 	 */
2815 	wmb();
2816 
2817 	writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
2818 
2819 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
2820 
2821 	return NETDEV_TX_OK;
2822 }
2823 
2824 /**
2825  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
2826  * @netdev: network interface device structure
2827  * @p: pointer to an address structure
2828  *
2829  * Returns 0 on success, negative on failure
2830  **/
2831 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
2832 {
2833 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2834 	struct ixgbe_hw *hw = &adapter->hw;
2835 	struct sockaddr *addr = p;
2836 
2837 	if (!is_valid_ether_addr(addr->sa_data))
2838 		return -EADDRNOTAVAIL;
2839 
2840 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2841 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2842 
2843 	spin_lock(&adapter->mbx_lock);
2844 
2845 	if (hw->mac.ops.set_rar)
2846 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2847 
2848 	spin_unlock(&adapter->mbx_lock);
2849 
2850 	return 0;
2851 }
2852 
2853 /**
2854  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
2855  * @netdev: network interface device structure
2856  * @new_mtu: new value for maximum frame size
2857  *
2858  * Returns 0 on success, negative on failure
2859  **/
2860 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
2861 {
2862 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2863 	struct ixgbe_hw *hw = &adapter->hw;
2864 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2865 	int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
2866 	u32 msg[2];
2867 
2868 	if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
2869 		max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
2870 
2871 	/* MTU < 68 is an error and causes problems on some kernels */
2872 	if ((new_mtu < 68) || (max_frame > max_possible_frame))
2873 		return -EINVAL;
2874 
2875 	hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
2876 	       netdev->mtu, new_mtu);
2877 	/* must set new MTU before calling down or up */
2878 	netdev->mtu = new_mtu;
2879 
2880 	if (!netif_running(netdev)) {
2881 		msg[0] = IXGBE_VF_SET_LPE;
2882 		msg[1] = max_frame;
2883 		hw->mbx.ops.write_posted(hw, msg, 2);
2884 	}
2885 
2886 	if (netif_running(netdev))
2887 		ixgbevf_reinit_locked(adapter);
2888 
2889 	return 0;
2890 }
2891 
2892 static void ixgbevf_shutdown(struct pci_dev *pdev)
2893 {
2894 	struct net_device *netdev = pci_get_drvdata(pdev);
2895 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2896 
2897 	netif_device_detach(netdev);
2898 
2899 	if (netif_running(netdev)) {
2900 		ixgbevf_down(adapter);
2901 		ixgbevf_free_irq(adapter);
2902 		ixgbevf_free_all_tx_resources(adapter);
2903 		ixgbevf_free_all_rx_resources(adapter);
2904 	}
2905 
2906 	pci_save_state(pdev);
2907 
2908 	pci_disable_device(pdev);
2909 }
2910 
2911 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
2912 						struct rtnl_link_stats64 *stats)
2913 {
2914 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2915 	unsigned int start;
2916 	u64 bytes, packets;
2917 	const struct ixgbevf_ring *ring;
2918 	int i;
2919 
2920 	ixgbevf_update_stats(adapter);
2921 
2922 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
2923 
2924 	for (i = 0; i < adapter->num_rx_queues; i++) {
2925 		ring = &adapter->rx_ring[i];
2926 		do {
2927 			start = u64_stats_fetch_begin_bh(&ring->syncp);
2928 			bytes = ring->total_bytes;
2929 			packets = ring->total_packets;
2930 		} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
2931 		stats->rx_bytes += bytes;
2932 		stats->rx_packets += packets;
2933 	}
2934 
2935 	for (i = 0; i < adapter->num_tx_queues; i++) {
2936 		ring = &adapter->tx_ring[i];
2937 		do {
2938 			start = u64_stats_fetch_begin_bh(&ring->syncp);
2939 			bytes = ring->total_bytes;
2940 			packets = ring->total_packets;
2941 		} while (u64_stats_fetch_retry_bh(&ring->syncp, start));
2942 		stats->tx_bytes += bytes;
2943 		stats->tx_packets += packets;
2944 	}
2945 
2946 	return stats;
2947 }
2948 
2949 static const struct net_device_ops ixgbe_netdev_ops = {
2950 	.ndo_open		= ixgbevf_open,
2951 	.ndo_stop		= ixgbevf_close,
2952 	.ndo_start_xmit		= ixgbevf_xmit_frame,
2953 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
2954 	.ndo_get_stats64	= ixgbevf_get_stats,
2955 	.ndo_validate_addr	= eth_validate_addr,
2956 	.ndo_set_mac_address	= ixgbevf_set_mac,
2957 	.ndo_change_mtu		= ixgbevf_change_mtu,
2958 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
2959 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
2960 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
2961 };
2962 
2963 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
2964 {
2965 	dev->netdev_ops = &ixgbe_netdev_ops;
2966 	ixgbevf_set_ethtool_ops(dev);
2967 	dev->watchdog_timeo = 5 * HZ;
2968 }
2969 
2970 /**
2971  * ixgbevf_probe - Device Initialization Routine
2972  * @pdev: PCI device information struct
2973  * @ent: entry in ixgbevf_pci_tbl
2974  *
2975  * Returns 0 on success, negative on failure
2976  *
2977  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
2978  * The OS initialization, configuring of the adapter private structure,
2979  * and a hardware reset occur.
2980  **/
2981 static int __devinit ixgbevf_probe(struct pci_dev *pdev,
2982 				   const struct pci_device_id *ent)
2983 {
2984 	struct net_device *netdev;
2985 	struct ixgbevf_adapter *adapter = NULL;
2986 	struct ixgbe_hw *hw = NULL;
2987 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
2988 	static int cards_found;
2989 	int err, pci_using_dac;
2990 
2991 	err = pci_enable_device(pdev);
2992 	if (err)
2993 		return err;
2994 
2995 	if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
2996 	    !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
2997 		pci_using_dac = 1;
2998 	} else {
2999 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3000 		if (err) {
3001 			err = dma_set_coherent_mask(&pdev->dev,
3002 						    DMA_BIT_MASK(32));
3003 			if (err) {
3004 				dev_err(&pdev->dev, "No usable DMA "
3005 					"configuration, aborting\n");
3006 				goto err_dma;
3007 			}
3008 		}
3009 		pci_using_dac = 0;
3010 	}
3011 
3012 	err = pci_request_regions(pdev, ixgbevf_driver_name);
3013 	if (err) {
3014 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3015 		goto err_pci_reg;
3016 	}
3017 
3018 	pci_set_master(pdev);
3019 
3020 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3021 				   MAX_TX_QUEUES);
3022 	if (!netdev) {
3023 		err = -ENOMEM;
3024 		goto err_alloc_etherdev;
3025 	}
3026 
3027 	SET_NETDEV_DEV(netdev, &pdev->dev);
3028 
3029 	pci_set_drvdata(pdev, netdev);
3030 	adapter = netdev_priv(netdev);
3031 
3032 	adapter->netdev = netdev;
3033 	adapter->pdev = pdev;
3034 	hw = &adapter->hw;
3035 	hw->back = adapter;
3036 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3037 
3038 	/*
3039 	 * call save state here in standalone driver because it relies on
3040 	 * adapter struct to exist, and needs to call netdev_priv
3041 	 */
3042 	pci_save_state(pdev);
3043 
3044 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3045 			      pci_resource_len(pdev, 0));
3046 	if (!hw->hw_addr) {
3047 		err = -EIO;
3048 		goto err_ioremap;
3049 	}
3050 
3051 	ixgbevf_assign_netdev_ops(netdev);
3052 
3053 	adapter->bd_number = cards_found;
3054 
3055 	/* Setup hw api */
3056 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3057 	hw->mac.type  = ii->mac;
3058 
3059 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3060 	       sizeof(struct ixgbe_mbx_operations));
3061 
3062 	/* setup the private structure */
3063 	err = ixgbevf_sw_init(adapter);
3064 	if (err)
3065 		goto err_sw_init;
3066 
3067 	/* The HW MAC address was set and/or determined in sw_init */
3068 	memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3069 
3070 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3071 		pr_err("invalid MAC address\n");
3072 		err = -EIO;
3073 		goto err_sw_init;
3074 	}
3075 
3076 	netdev->hw_features = NETIF_F_SG |
3077 			   NETIF_F_IP_CSUM |
3078 			   NETIF_F_IPV6_CSUM |
3079 			   NETIF_F_TSO |
3080 			   NETIF_F_TSO6 |
3081 			   NETIF_F_RXCSUM;
3082 
3083 	netdev->features = netdev->hw_features |
3084 			   NETIF_F_HW_VLAN_TX |
3085 			   NETIF_F_HW_VLAN_RX |
3086 			   NETIF_F_HW_VLAN_FILTER;
3087 
3088 	netdev->vlan_features |= NETIF_F_TSO;
3089 	netdev->vlan_features |= NETIF_F_TSO6;
3090 	netdev->vlan_features |= NETIF_F_IP_CSUM;
3091 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3092 	netdev->vlan_features |= NETIF_F_SG;
3093 
3094 	if (pci_using_dac)
3095 		netdev->features |= NETIF_F_HIGHDMA;
3096 
3097 	netdev->priv_flags |= IFF_UNICAST_FLT;
3098 
3099 	init_timer(&adapter->watchdog_timer);
3100 	adapter->watchdog_timer.function = ixgbevf_watchdog;
3101 	adapter->watchdog_timer.data = (unsigned long)adapter;
3102 
3103 	INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3104 	INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3105 
3106 	err = ixgbevf_init_interrupt_scheme(adapter);
3107 	if (err)
3108 		goto err_sw_init;
3109 
3110 	/* pick up the PCI bus settings for reporting later */
3111 	if (hw->mac.ops.get_bus_info)
3112 		hw->mac.ops.get_bus_info(hw);
3113 
3114 	strcpy(netdev->name, "eth%d");
3115 
3116 	err = register_netdev(netdev);
3117 	if (err)
3118 		goto err_register;
3119 
3120 	netif_carrier_off(netdev);
3121 
3122 	ixgbevf_init_last_counter_stats(adapter);
3123 
3124 	/* print the MAC address */
3125 	hw_dbg(hw, "%pM\n", netdev->dev_addr);
3126 
3127 	hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3128 
3129 	hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3130 	cards_found++;
3131 	return 0;
3132 
3133 err_register:
3134 err_sw_init:
3135 	ixgbevf_reset_interrupt_capability(adapter);
3136 	iounmap(hw->hw_addr);
3137 err_ioremap:
3138 	free_netdev(netdev);
3139 err_alloc_etherdev:
3140 	pci_release_regions(pdev);
3141 err_pci_reg:
3142 err_dma:
3143 	pci_disable_device(pdev);
3144 	return err;
3145 }
3146 
3147 /**
3148  * ixgbevf_remove - Device Removal Routine
3149  * @pdev: PCI device information struct
3150  *
3151  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3152  * that it should release a PCI device.  The could be caused by a
3153  * Hot-Plug event, or because the driver is going to be removed from
3154  * memory.
3155  **/
3156 static void __devexit ixgbevf_remove(struct pci_dev *pdev)
3157 {
3158 	struct net_device *netdev = pci_get_drvdata(pdev);
3159 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3160 
3161 	set_bit(__IXGBEVF_DOWN, &adapter->state);
3162 
3163 	del_timer_sync(&adapter->watchdog_timer);
3164 
3165 	cancel_work_sync(&adapter->reset_task);
3166 	cancel_work_sync(&adapter->watchdog_task);
3167 
3168 	if (netdev->reg_state == NETREG_REGISTERED)
3169 		unregister_netdev(netdev);
3170 
3171 	ixgbevf_reset_interrupt_capability(adapter);
3172 
3173 	iounmap(adapter->hw.hw_addr);
3174 	pci_release_regions(pdev);
3175 
3176 	hw_dbg(&adapter->hw, "Remove complete\n");
3177 
3178 	kfree(adapter->tx_ring);
3179 	kfree(adapter->rx_ring);
3180 
3181 	free_netdev(netdev);
3182 
3183 	pci_disable_device(pdev);
3184 }
3185 
3186 /**
3187  * ixgbevf_io_error_detected - called when PCI error is detected
3188  * @pdev: Pointer to PCI device
3189  * @state: The current pci connection state
3190  *
3191  * This function is called after a PCI bus error affecting
3192  * this device has been detected.
3193  */
3194 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3195 						  pci_channel_state_t state)
3196 {
3197 	struct net_device *netdev = pci_get_drvdata(pdev);
3198 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3199 
3200 	netif_device_detach(netdev);
3201 
3202 	if (state == pci_channel_io_perm_failure)
3203 		return PCI_ERS_RESULT_DISCONNECT;
3204 
3205 	if (netif_running(netdev))
3206 		ixgbevf_down(adapter);
3207 
3208 	pci_disable_device(pdev);
3209 
3210 	/* Request a slot slot reset. */
3211 	return PCI_ERS_RESULT_NEED_RESET;
3212 }
3213 
3214 /**
3215  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3216  * @pdev: Pointer to PCI device
3217  *
3218  * Restart the card from scratch, as if from a cold-boot. Implementation
3219  * resembles the first-half of the ixgbevf_resume routine.
3220  */
3221 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3222 {
3223 	struct net_device *netdev = pci_get_drvdata(pdev);
3224 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3225 
3226 	if (pci_enable_device_mem(pdev)) {
3227 		dev_err(&pdev->dev,
3228 			"Cannot re-enable PCI device after reset.\n");
3229 		return PCI_ERS_RESULT_DISCONNECT;
3230 	}
3231 
3232 	pci_set_master(pdev);
3233 
3234 	ixgbevf_reset(adapter);
3235 
3236 	return PCI_ERS_RESULT_RECOVERED;
3237 }
3238 
3239 /**
3240  * ixgbevf_io_resume - called when traffic can start flowing again.
3241  * @pdev: Pointer to PCI device
3242  *
3243  * This callback is called when the error recovery driver tells us that
3244  * its OK to resume normal operation. Implementation resembles the
3245  * second-half of the ixgbevf_resume routine.
3246  */
3247 static void ixgbevf_io_resume(struct pci_dev *pdev)
3248 {
3249 	struct net_device *netdev = pci_get_drvdata(pdev);
3250 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3251 
3252 	if (netif_running(netdev))
3253 		ixgbevf_up(adapter);
3254 
3255 	netif_device_attach(netdev);
3256 }
3257 
3258 /* PCI Error Recovery (ERS) */
3259 static struct pci_error_handlers ixgbevf_err_handler = {
3260 	.error_detected = ixgbevf_io_error_detected,
3261 	.slot_reset = ixgbevf_io_slot_reset,
3262 	.resume = ixgbevf_io_resume,
3263 };
3264 
3265 static struct pci_driver ixgbevf_driver = {
3266 	.name     = ixgbevf_driver_name,
3267 	.id_table = ixgbevf_pci_tbl,
3268 	.probe    = ixgbevf_probe,
3269 	.remove   = __devexit_p(ixgbevf_remove),
3270 	.shutdown = ixgbevf_shutdown,
3271 	.err_handler = &ixgbevf_err_handler
3272 };
3273 
3274 /**
3275  * ixgbevf_init_module - Driver Registration Routine
3276  *
3277  * ixgbevf_init_module is the first routine called when the driver is
3278  * loaded. All it does is register with the PCI subsystem.
3279  **/
3280 static int __init ixgbevf_init_module(void)
3281 {
3282 	int ret;
3283 	pr_info("%s - version %s\n", ixgbevf_driver_string,
3284 		ixgbevf_driver_version);
3285 
3286 	pr_info("%s\n", ixgbevf_copyright);
3287 
3288 	ret = pci_register_driver(&ixgbevf_driver);
3289 	return ret;
3290 }
3291 
3292 module_init(ixgbevf_init_module);
3293 
3294 /**
3295  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3296  *
3297  * ixgbevf_exit_module is called just before the driver is removed
3298  * from memory.
3299  **/
3300 static void __exit ixgbevf_exit_module(void)
3301 {
3302 	pci_unregister_driver(&ixgbevf_driver);
3303 }
3304 
3305 #ifdef DEBUG
3306 /**
3307  * ixgbevf_get_hw_dev_name - return device name string
3308  * used by hardware layer to print debugging information
3309  **/
3310 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3311 {
3312 	struct ixgbevf_adapter *adapter = hw->back;
3313 	return adapter->netdev->name;
3314 }
3315 
3316 #endif
3317 module_exit(ixgbevf_exit_module);
3318 
3319 /* ixgbevf_main.c */
3320