xref: /linux/drivers/net/ethernet/intel/igbvf/netdev.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*******************************************************************************
2 
3   Intel(R) 82576 Virtual Function Linux driver
4   Copyright(c) 2009 - 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, see <http://www.gnu.org/licenses/>.
17 
18   The full GNU General Public License is included in this distribution in
19   the file called "COPYING".
20 
21   Contact Information:
22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 
25 *******************************************************************************/
26 
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
28 
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if_vlan.h>
45 #include <linux/prefetch.h>
46 
47 #include "igbvf.h"
48 
49 #define DRV_VERSION "2.0.2-k"
50 char igbvf_driver_name[] = "igbvf";
51 const char igbvf_driver_version[] = DRV_VERSION;
52 static const char igbvf_driver_string[] =
53 		  "Intel(R) Gigabit Virtual Function Network Driver";
54 static const char igbvf_copyright[] =
55 		  "Copyright (c) 2009 - 2012 Intel Corporation.";
56 
57 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
58 static int debug = -1;
59 module_param(debug, int, 0);
60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
61 
62 static int igbvf_poll(struct napi_struct *napi, int budget);
63 static void igbvf_reset(struct igbvf_adapter *);
64 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
65 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
66 
67 static struct igbvf_info igbvf_vf_info = {
68 	.mac		= e1000_vfadapt,
69 	.flags		= 0,
70 	.pba		= 10,
71 	.init_ops	= e1000_init_function_pointers_vf,
72 };
73 
74 static struct igbvf_info igbvf_i350_vf_info = {
75 	.mac		= e1000_vfadapt_i350,
76 	.flags		= 0,
77 	.pba		= 10,
78 	.init_ops	= e1000_init_function_pointers_vf,
79 };
80 
81 static const struct igbvf_info *igbvf_info_tbl[] = {
82 	[board_vf]	= &igbvf_vf_info,
83 	[board_i350_vf]	= &igbvf_i350_vf_info,
84 };
85 
86 /**
87  * igbvf_desc_unused - calculate if we have unused descriptors
88  * @rx_ring: address of receive ring structure
89  **/
90 static int igbvf_desc_unused(struct igbvf_ring *ring)
91 {
92 	if (ring->next_to_clean > ring->next_to_use)
93 		return ring->next_to_clean - ring->next_to_use - 1;
94 
95 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
96 }
97 
98 /**
99  * igbvf_receive_skb - helper function to handle Rx indications
100  * @adapter: board private structure
101  * @status: descriptor status field as written by hardware
102  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
103  * @skb: pointer to sk_buff to be indicated to stack
104  **/
105 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
106 			      struct net_device *netdev,
107 			      struct sk_buff *skb,
108 			      u32 status, u16 vlan)
109 {
110 	u16 vid;
111 
112 	if (status & E1000_RXD_STAT_VP) {
113 		if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
114 		    (status & E1000_RXDEXT_STATERR_LB))
115 			vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
116 		else
117 			vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
118 		if (test_bit(vid, adapter->active_vlans))
119 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
120 	}
121 
122 	napi_gro_receive(&adapter->rx_ring->napi, skb);
123 }
124 
125 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
126 					 u32 status_err, struct sk_buff *skb)
127 {
128 	skb_checksum_none_assert(skb);
129 
130 	/* Ignore Checksum bit is set or checksum is disabled through ethtool */
131 	if ((status_err & E1000_RXD_STAT_IXSM) ||
132 	    (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
133 		return;
134 
135 	/* TCP/UDP checksum error bit is set */
136 	if (status_err &
137 	    (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
138 		/* let the stack verify checksum errors */
139 		adapter->hw_csum_err++;
140 		return;
141 	}
142 
143 	/* It must be a TCP or UDP packet with a valid checksum */
144 	if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
145 		skb->ip_summed = CHECKSUM_UNNECESSARY;
146 
147 	adapter->hw_csum_good++;
148 }
149 
150 /**
151  * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
152  * @rx_ring: address of ring structure to repopulate
153  * @cleaned_count: number of buffers to repopulate
154  **/
155 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
156 				   int cleaned_count)
157 {
158 	struct igbvf_adapter *adapter = rx_ring->adapter;
159 	struct net_device *netdev = adapter->netdev;
160 	struct pci_dev *pdev = adapter->pdev;
161 	union e1000_adv_rx_desc *rx_desc;
162 	struct igbvf_buffer *buffer_info;
163 	struct sk_buff *skb;
164 	unsigned int i;
165 	int bufsz;
166 
167 	i = rx_ring->next_to_use;
168 	buffer_info = &rx_ring->buffer_info[i];
169 
170 	if (adapter->rx_ps_hdr_size)
171 		bufsz = adapter->rx_ps_hdr_size;
172 	else
173 		bufsz = adapter->rx_buffer_len;
174 
175 	while (cleaned_count--) {
176 		rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
177 
178 		if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
179 			if (!buffer_info->page) {
180 				buffer_info->page = alloc_page(GFP_ATOMIC);
181 				if (!buffer_info->page) {
182 					adapter->alloc_rx_buff_failed++;
183 					goto no_buffers;
184 				}
185 				buffer_info->page_offset = 0;
186 			} else {
187 				buffer_info->page_offset ^= PAGE_SIZE / 2;
188 			}
189 			buffer_info->page_dma =
190 				dma_map_page(&pdev->dev, buffer_info->page,
191 					     buffer_info->page_offset,
192 					     PAGE_SIZE / 2,
193 					     DMA_FROM_DEVICE);
194 			if (dma_mapping_error(&pdev->dev,
195 					      buffer_info->page_dma)) {
196 				__free_page(buffer_info->page);
197 				buffer_info->page = NULL;
198 				dev_err(&pdev->dev, "RX DMA map failed\n");
199 				break;
200 			}
201 		}
202 
203 		if (!buffer_info->skb) {
204 			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
205 			if (!skb) {
206 				adapter->alloc_rx_buff_failed++;
207 				goto no_buffers;
208 			}
209 
210 			buffer_info->skb = skb;
211 			buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
212 							  bufsz,
213 							  DMA_FROM_DEVICE);
214 			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
215 				dev_kfree_skb(buffer_info->skb);
216 				buffer_info->skb = NULL;
217 				dev_err(&pdev->dev, "RX DMA map failed\n");
218 				goto no_buffers;
219 			}
220 		}
221 		/* Refresh the desc even if buffer_addrs didn't change because
222 		 * each write-back erases this info.
223 		 */
224 		if (adapter->rx_ps_hdr_size) {
225 			rx_desc->read.pkt_addr =
226 			     cpu_to_le64(buffer_info->page_dma);
227 			rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
228 		} else {
229 			rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
230 			rx_desc->read.hdr_addr = 0;
231 		}
232 
233 		i++;
234 		if (i == rx_ring->count)
235 			i = 0;
236 		buffer_info = &rx_ring->buffer_info[i];
237 	}
238 
239 no_buffers:
240 	if (rx_ring->next_to_use != i) {
241 		rx_ring->next_to_use = i;
242 		if (i == 0)
243 			i = (rx_ring->count - 1);
244 		else
245 			i--;
246 
247 		/* Force memory writes to complete before letting h/w
248 		 * know there are new descriptors to fetch.  (Only
249 		 * applicable for weak-ordered memory model archs,
250 		 * such as IA-64).
251 		*/
252 		wmb();
253 		writel(i, adapter->hw.hw_addr + rx_ring->tail);
254 	}
255 }
256 
257 /**
258  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
259  * @adapter: board private structure
260  *
261  * the return value indicates whether actual cleaning was done, there
262  * is no guarantee that everything was cleaned
263  **/
264 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
265 			       int *work_done, int work_to_do)
266 {
267 	struct igbvf_ring *rx_ring = adapter->rx_ring;
268 	struct net_device *netdev = adapter->netdev;
269 	struct pci_dev *pdev = adapter->pdev;
270 	union e1000_adv_rx_desc *rx_desc, *next_rxd;
271 	struct igbvf_buffer *buffer_info, *next_buffer;
272 	struct sk_buff *skb;
273 	bool cleaned = false;
274 	int cleaned_count = 0;
275 	unsigned int total_bytes = 0, total_packets = 0;
276 	unsigned int i;
277 	u32 length, hlen, staterr;
278 
279 	i = rx_ring->next_to_clean;
280 	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
281 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
282 
283 	while (staterr & E1000_RXD_STAT_DD) {
284 		if (*work_done >= work_to_do)
285 			break;
286 		(*work_done)++;
287 		rmb(); /* read descriptor and rx_buffer_info after status DD */
288 
289 		buffer_info = &rx_ring->buffer_info[i];
290 
291 		/* HW will not DMA in data larger than the given buffer, even
292 		 * if it parses the (NFS, of course) header to be larger.  In
293 		 * that case, it fills the header buffer and spills the rest
294 		 * into the page.
295 		 */
296 		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
297 		       & E1000_RXDADV_HDRBUFLEN_MASK) >>
298 		       E1000_RXDADV_HDRBUFLEN_SHIFT;
299 		if (hlen > adapter->rx_ps_hdr_size)
300 			hlen = adapter->rx_ps_hdr_size;
301 
302 		length = le16_to_cpu(rx_desc->wb.upper.length);
303 		cleaned = true;
304 		cleaned_count++;
305 
306 		skb = buffer_info->skb;
307 		prefetch(skb->data - NET_IP_ALIGN);
308 		buffer_info->skb = NULL;
309 		if (!adapter->rx_ps_hdr_size) {
310 			dma_unmap_single(&pdev->dev, buffer_info->dma,
311 					 adapter->rx_buffer_len,
312 					 DMA_FROM_DEVICE);
313 			buffer_info->dma = 0;
314 			skb_put(skb, length);
315 			goto send_up;
316 		}
317 
318 		if (!skb_shinfo(skb)->nr_frags) {
319 			dma_unmap_single(&pdev->dev, buffer_info->dma,
320 					 adapter->rx_ps_hdr_size,
321 					 DMA_FROM_DEVICE);
322 			buffer_info->dma = 0;
323 			skb_put(skb, hlen);
324 		}
325 
326 		if (length) {
327 			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
328 				       PAGE_SIZE / 2,
329 				       DMA_FROM_DEVICE);
330 			buffer_info->page_dma = 0;
331 
332 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
333 					   buffer_info->page,
334 					   buffer_info->page_offset,
335 					   length);
336 
337 			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
338 			    (page_count(buffer_info->page) != 1))
339 				buffer_info->page = NULL;
340 			else
341 				get_page(buffer_info->page);
342 
343 			skb->len += length;
344 			skb->data_len += length;
345 			skb->truesize += PAGE_SIZE / 2;
346 		}
347 send_up:
348 		i++;
349 		if (i == rx_ring->count)
350 			i = 0;
351 		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
352 		prefetch(next_rxd);
353 		next_buffer = &rx_ring->buffer_info[i];
354 
355 		if (!(staterr & E1000_RXD_STAT_EOP)) {
356 			buffer_info->skb = next_buffer->skb;
357 			buffer_info->dma = next_buffer->dma;
358 			next_buffer->skb = skb;
359 			next_buffer->dma = 0;
360 			goto next_desc;
361 		}
362 
363 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
364 			dev_kfree_skb_irq(skb);
365 			goto next_desc;
366 		}
367 
368 		total_bytes += skb->len;
369 		total_packets++;
370 
371 		igbvf_rx_checksum_adv(adapter, staterr, skb);
372 
373 		skb->protocol = eth_type_trans(skb, netdev);
374 
375 		igbvf_receive_skb(adapter, netdev, skb, staterr,
376 				  rx_desc->wb.upper.vlan);
377 
378 next_desc:
379 		rx_desc->wb.upper.status_error = 0;
380 
381 		/* return some buffers to hardware, one at a time is too slow */
382 		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
383 			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
384 			cleaned_count = 0;
385 		}
386 
387 		/* use prefetched values */
388 		rx_desc = next_rxd;
389 		buffer_info = next_buffer;
390 
391 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
392 	}
393 
394 	rx_ring->next_to_clean = i;
395 	cleaned_count = igbvf_desc_unused(rx_ring);
396 
397 	if (cleaned_count)
398 		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
399 
400 	adapter->total_rx_packets += total_packets;
401 	adapter->total_rx_bytes += total_bytes;
402 	adapter->net_stats.rx_bytes += total_bytes;
403 	adapter->net_stats.rx_packets += total_packets;
404 	return cleaned;
405 }
406 
407 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
408 			    struct igbvf_buffer *buffer_info)
409 {
410 	if (buffer_info->dma) {
411 		if (buffer_info->mapped_as_page)
412 			dma_unmap_page(&adapter->pdev->dev,
413 				       buffer_info->dma,
414 				       buffer_info->length,
415 				       DMA_TO_DEVICE);
416 		else
417 			dma_unmap_single(&adapter->pdev->dev,
418 					 buffer_info->dma,
419 					 buffer_info->length,
420 					 DMA_TO_DEVICE);
421 		buffer_info->dma = 0;
422 	}
423 	if (buffer_info->skb) {
424 		dev_kfree_skb_any(buffer_info->skb);
425 		buffer_info->skb = NULL;
426 	}
427 	buffer_info->time_stamp = 0;
428 }
429 
430 /**
431  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
432  * @adapter: board private structure
433  *
434  * Return 0 on success, negative on failure
435  **/
436 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
437 			     struct igbvf_ring *tx_ring)
438 {
439 	struct pci_dev *pdev = adapter->pdev;
440 	int size;
441 
442 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
443 	tx_ring->buffer_info = vzalloc(size);
444 	if (!tx_ring->buffer_info)
445 		goto err;
446 
447 	/* round up to nearest 4K */
448 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
449 	tx_ring->size = ALIGN(tx_ring->size, 4096);
450 
451 	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
452 					   &tx_ring->dma, GFP_KERNEL);
453 	if (!tx_ring->desc)
454 		goto err;
455 
456 	tx_ring->adapter = adapter;
457 	tx_ring->next_to_use = 0;
458 	tx_ring->next_to_clean = 0;
459 
460 	return 0;
461 err:
462 	vfree(tx_ring->buffer_info);
463 	dev_err(&adapter->pdev->dev,
464 		"Unable to allocate memory for the transmit descriptor ring\n");
465 	return -ENOMEM;
466 }
467 
468 /**
469  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
470  * @adapter: board private structure
471  *
472  * Returns 0 on success, negative on failure
473  **/
474 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
475 			     struct igbvf_ring *rx_ring)
476 {
477 	struct pci_dev *pdev = adapter->pdev;
478 	int size, desc_len;
479 
480 	size = sizeof(struct igbvf_buffer) * rx_ring->count;
481 	rx_ring->buffer_info = vzalloc(size);
482 	if (!rx_ring->buffer_info)
483 		goto err;
484 
485 	desc_len = sizeof(union e1000_adv_rx_desc);
486 
487 	/* Round up to nearest 4K */
488 	rx_ring->size = rx_ring->count * desc_len;
489 	rx_ring->size = ALIGN(rx_ring->size, 4096);
490 
491 	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
492 					   &rx_ring->dma, GFP_KERNEL);
493 	if (!rx_ring->desc)
494 		goto err;
495 
496 	rx_ring->next_to_clean = 0;
497 	rx_ring->next_to_use = 0;
498 
499 	rx_ring->adapter = adapter;
500 
501 	return 0;
502 
503 err:
504 	vfree(rx_ring->buffer_info);
505 	rx_ring->buffer_info = NULL;
506 	dev_err(&adapter->pdev->dev,
507 		"Unable to allocate memory for the receive descriptor ring\n");
508 	return -ENOMEM;
509 }
510 
511 /**
512  * igbvf_clean_tx_ring - Free Tx Buffers
513  * @tx_ring: ring to be cleaned
514  **/
515 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
516 {
517 	struct igbvf_adapter *adapter = tx_ring->adapter;
518 	struct igbvf_buffer *buffer_info;
519 	unsigned long size;
520 	unsigned int i;
521 
522 	if (!tx_ring->buffer_info)
523 		return;
524 
525 	/* Free all the Tx ring sk_buffs */
526 	for (i = 0; i < tx_ring->count; i++) {
527 		buffer_info = &tx_ring->buffer_info[i];
528 		igbvf_put_txbuf(adapter, buffer_info);
529 	}
530 
531 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
532 	memset(tx_ring->buffer_info, 0, size);
533 
534 	/* Zero out the descriptor ring */
535 	memset(tx_ring->desc, 0, tx_ring->size);
536 
537 	tx_ring->next_to_use = 0;
538 	tx_ring->next_to_clean = 0;
539 
540 	writel(0, adapter->hw.hw_addr + tx_ring->head);
541 	writel(0, adapter->hw.hw_addr + tx_ring->tail);
542 }
543 
544 /**
545  * igbvf_free_tx_resources - Free Tx Resources per Queue
546  * @tx_ring: ring to free resources from
547  *
548  * Free all transmit software resources
549  **/
550 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
551 {
552 	struct pci_dev *pdev = tx_ring->adapter->pdev;
553 
554 	igbvf_clean_tx_ring(tx_ring);
555 
556 	vfree(tx_ring->buffer_info);
557 	tx_ring->buffer_info = NULL;
558 
559 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
560 			  tx_ring->dma);
561 
562 	tx_ring->desc = NULL;
563 }
564 
565 /**
566  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
567  * @adapter: board private structure
568  **/
569 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
570 {
571 	struct igbvf_adapter *adapter = rx_ring->adapter;
572 	struct igbvf_buffer *buffer_info;
573 	struct pci_dev *pdev = adapter->pdev;
574 	unsigned long size;
575 	unsigned int i;
576 
577 	if (!rx_ring->buffer_info)
578 		return;
579 
580 	/* Free all the Rx ring sk_buffs */
581 	for (i = 0; i < rx_ring->count; i++) {
582 		buffer_info = &rx_ring->buffer_info[i];
583 		if (buffer_info->dma) {
584 			if (adapter->rx_ps_hdr_size) {
585 				dma_unmap_single(&pdev->dev, buffer_info->dma,
586 						 adapter->rx_ps_hdr_size,
587 						 DMA_FROM_DEVICE);
588 			} else {
589 				dma_unmap_single(&pdev->dev, buffer_info->dma,
590 						 adapter->rx_buffer_len,
591 						 DMA_FROM_DEVICE);
592 			}
593 			buffer_info->dma = 0;
594 		}
595 
596 		if (buffer_info->skb) {
597 			dev_kfree_skb(buffer_info->skb);
598 			buffer_info->skb = NULL;
599 		}
600 
601 		if (buffer_info->page) {
602 			if (buffer_info->page_dma)
603 				dma_unmap_page(&pdev->dev,
604 					       buffer_info->page_dma,
605 					       PAGE_SIZE / 2,
606 					       DMA_FROM_DEVICE);
607 			put_page(buffer_info->page);
608 			buffer_info->page = NULL;
609 			buffer_info->page_dma = 0;
610 			buffer_info->page_offset = 0;
611 		}
612 	}
613 
614 	size = sizeof(struct igbvf_buffer) * rx_ring->count;
615 	memset(rx_ring->buffer_info, 0, size);
616 
617 	/* Zero out the descriptor ring */
618 	memset(rx_ring->desc, 0, rx_ring->size);
619 
620 	rx_ring->next_to_clean = 0;
621 	rx_ring->next_to_use = 0;
622 
623 	writel(0, adapter->hw.hw_addr + rx_ring->head);
624 	writel(0, adapter->hw.hw_addr + rx_ring->tail);
625 }
626 
627 /**
628  * igbvf_free_rx_resources - Free Rx Resources
629  * @rx_ring: ring to clean the resources from
630  *
631  * Free all receive software resources
632  **/
633 
634 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
635 {
636 	struct pci_dev *pdev = rx_ring->adapter->pdev;
637 
638 	igbvf_clean_rx_ring(rx_ring);
639 
640 	vfree(rx_ring->buffer_info);
641 	rx_ring->buffer_info = NULL;
642 
643 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
644 			  rx_ring->dma);
645 	rx_ring->desc = NULL;
646 }
647 
648 /**
649  * igbvf_update_itr - update the dynamic ITR value based on statistics
650  * @adapter: pointer to adapter
651  * @itr_setting: current adapter->itr
652  * @packets: the number of packets during this measurement interval
653  * @bytes: the number of bytes during this measurement interval
654  *
655  * Stores a new ITR value based on packets and byte counts during the last
656  * interrupt.  The advantage of per interrupt computation is faster updates
657  * and more accurate ITR for the current traffic pattern.  Constants in this
658  * function were computed based on theoretical maximum wire speed and thresholds
659  * were set based on testing data as well as attempting to minimize response
660  * time while increasing bulk throughput.
661  **/
662 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
663 					   enum latency_range itr_setting,
664 					   int packets, int bytes)
665 {
666 	enum latency_range retval = itr_setting;
667 
668 	if (packets == 0)
669 		goto update_itr_done;
670 
671 	switch (itr_setting) {
672 	case lowest_latency:
673 		/* handle TSO and jumbo frames */
674 		if (bytes/packets > 8000)
675 			retval = bulk_latency;
676 		else if ((packets < 5) && (bytes > 512))
677 			retval = low_latency;
678 		break;
679 	case low_latency:  /* 50 usec aka 20000 ints/s */
680 		if (bytes > 10000) {
681 			/* this if handles the TSO accounting */
682 			if (bytes/packets > 8000)
683 				retval = bulk_latency;
684 			else if ((packets < 10) || ((bytes/packets) > 1200))
685 				retval = bulk_latency;
686 			else if ((packets > 35))
687 				retval = lowest_latency;
688 		} else if (bytes/packets > 2000) {
689 			retval = bulk_latency;
690 		} else if (packets <= 2 && bytes < 512) {
691 			retval = lowest_latency;
692 		}
693 		break;
694 	case bulk_latency: /* 250 usec aka 4000 ints/s */
695 		if (bytes > 25000) {
696 			if (packets > 35)
697 				retval = low_latency;
698 		} else if (bytes < 6000) {
699 			retval = low_latency;
700 		}
701 		break;
702 	default:
703 		break;
704 	}
705 
706 update_itr_done:
707 	return retval;
708 }
709 
710 static int igbvf_range_to_itr(enum latency_range current_range)
711 {
712 	int new_itr;
713 
714 	switch (current_range) {
715 	/* counts and packets in update_itr are dependent on these numbers */
716 	case lowest_latency:
717 		new_itr = IGBVF_70K_ITR;
718 		break;
719 	case low_latency:
720 		new_itr = IGBVF_20K_ITR;
721 		break;
722 	case bulk_latency:
723 		new_itr = IGBVF_4K_ITR;
724 		break;
725 	default:
726 		new_itr = IGBVF_START_ITR;
727 		break;
728 	}
729 	return new_itr;
730 }
731 
732 static void igbvf_set_itr(struct igbvf_adapter *adapter)
733 {
734 	u32 new_itr;
735 
736 	adapter->tx_ring->itr_range =
737 			igbvf_update_itr(adapter,
738 					 adapter->tx_ring->itr_val,
739 					 adapter->total_tx_packets,
740 					 adapter->total_tx_bytes);
741 
742 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
743 	if (adapter->requested_itr == 3 &&
744 	    adapter->tx_ring->itr_range == lowest_latency)
745 		adapter->tx_ring->itr_range = low_latency;
746 
747 	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
748 
749 	if (new_itr != adapter->tx_ring->itr_val) {
750 		u32 current_itr = adapter->tx_ring->itr_val;
751 		/* this attempts to bias the interrupt rate towards Bulk
752 		 * by adding intermediate steps when interrupt rate is
753 		 * increasing
754 		 */
755 		new_itr = new_itr > current_itr ?
756 			  min(current_itr + (new_itr >> 2), new_itr) :
757 			  new_itr;
758 		adapter->tx_ring->itr_val = new_itr;
759 
760 		adapter->tx_ring->set_itr = 1;
761 	}
762 
763 	adapter->rx_ring->itr_range =
764 			igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
765 					 adapter->total_rx_packets,
766 					 adapter->total_rx_bytes);
767 	if (adapter->requested_itr == 3 &&
768 	    adapter->rx_ring->itr_range == lowest_latency)
769 		adapter->rx_ring->itr_range = low_latency;
770 
771 	new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
772 
773 	if (new_itr != adapter->rx_ring->itr_val) {
774 		u32 current_itr = adapter->rx_ring->itr_val;
775 
776 		new_itr = new_itr > current_itr ?
777 			  min(current_itr + (new_itr >> 2), new_itr) :
778 			  new_itr;
779 		adapter->rx_ring->itr_val = new_itr;
780 
781 		adapter->rx_ring->set_itr = 1;
782 	}
783 }
784 
785 /**
786  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
787  * @adapter: board private structure
788  *
789  * returns true if ring is completely cleaned
790  **/
791 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
792 {
793 	struct igbvf_adapter *adapter = tx_ring->adapter;
794 	struct net_device *netdev = adapter->netdev;
795 	struct igbvf_buffer *buffer_info;
796 	struct sk_buff *skb;
797 	union e1000_adv_tx_desc *tx_desc, *eop_desc;
798 	unsigned int total_bytes = 0, total_packets = 0;
799 	unsigned int i, count = 0;
800 	bool cleaned = false;
801 
802 	i = tx_ring->next_to_clean;
803 	buffer_info = &tx_ring->buffer_info[i];
804 	eop_desc = buffer_info->next_to_watch;
805 
806 	do {
807 		/* if next_to_watch is not set then there is no work pending */
808 		if (!eop_desc)
809 			break;
810 
811 		/* prevent any other reads prior to eop_desc */
812 		read_barrier_depends();
813 
814 		/* if DD is not set pending work has not been completed */
815 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
816 			break;
817 
818 		/* clear next_to_watch to prevent false hangs */
819 		buffer_info->next_to_watch = NULL;
820 
821 		for (cleaned = false; !cleaned; count++) {
822 			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
823 			cleaned = (tx_desc == eop_desc);
824 			skb = buffer_info->skb;
825 
826 			if (skb) {
827 				unsigned int segs, bytecount;
828 
829 				/* gso_segs is currently only valid for tcp */
830 				segs = skb_shinfo(skb)->gso_segs ?: 1;
831 				/* multiply data chunks by size of headers */
832 				bytecount = ((segs - 1) * skb_headlen(skb)) +
833 					    skb->len;
834 				total_packets += segs;
835 				total_bytes += bytecount;
836 			}
837 
838 			igbvf_put_txbuf(adapter, buffer_info);
839 			tx_desc->wb.status = 0;
840 
841 			i++;
842 			if (i == tx_ring->count)
843 				i = 0;
844 
845 			buffer_info = &tx_ring->buffer_info[i];
846 		}
847 
848 		eop_desc = buffer_info->next_to_watch;
849 	} while (count < tx_ring->count);
850 
851 	tx_ring->next_to_clean = i;
852 
853 	if (unlikely(count && netif_carrier_ok(netdev) &&
854 	    igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
855 		/* Make sure that anybody stopping the queue after this
856 		 * sees the new next_to_clean.
857 		 */
858 		smp_mb();
859 		if (netif_queue_stopped(netdev) &&
860 		    !(test_bit(__IGBVF_DOWN, &adapter->state))) {
861 			netif_wake_queue(netdev);
862 			++adapter->restart_queue;
863 		}
864 	}
865 
866 	adapter->net_stats.tx_bytes += total_bytes;
867 	adapter->net_stats.tx_packets += total_packets;
868 	return count < tx_ring->count;
869 }
870 
871 static irqreturn_t igbvf_msix_other(int irq, void *data)
872 {
873 	struct net_device *netdev = data;
874 	struct igbvf_adapter *adapter = netdev_priv(netdev);
875 	struct e1000_hw *hw = &adapter->hw;
876 
877 	adapter->int_counter1++;
878 
879 	netif_carrier_off(netdev);
880 	hw->mac.get_link_status = 1;
881 	if (!test_bit(__IGBVF_DOWN, &adapter->state))
882 		mod_timer(&adapter->watchdog_timer, jiffies + 1);
883 
884 	ew32(EIMS, adapter->eims_other);
885 
886 	return IRQ_HANDLED;
887 }
888 
889 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
890 {
891 	struct net_device *netdev = data;
892 	struct igbvf_adapter *adapter = netdev_priv(netdev);
893 	struct e1000_hw *hw = &adapter->hw;
894 	struct igbvf_ring *tx_ring = adapter->tx_ring;
895 
896 	if (tx_ring->set_itr) {
897 		writel(tx_ring->itr_val,
898 		       adapter->hw.hw_addr + tx_ring->itr_register);
899 		adapter->tx_ring->set_itr = 0;
900 	}
901 
902 	adapter->total_tx_bytes = 0;
903 	adapter->total_tx_packets = 0;
904 
905 	/* auto mask will automatically re-enable the interrupt when we write
906 	 * EICS
907 	 */
908 	if (!igbvf_clean_tx_irq(tx_ring))
909 		/* Ring was not completely cleaned, so fire another interrupt */
910 		ew32(EICS, tx_ring->eims_value);
911 	else
912 		ew32(EIMS, tx_ring->eims_value);
913 
914 	return IRQ_HANDLED;
915 }
916 
917 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
918 {
919 	struct net_device *netdev = data;
920 	struct igbvf_adapter *adapter = netdev_priv(netdev);
921 
922 	adapter->int_counter0++;
923 
924 	/* Write the ITR value calculated at the end of the
925 	 * previous interrupt.
926 	 */
927 	if (adapter->rx_ring->set_itr) {
928 		writel(adapter->rx_ring->itr_val,
929 		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
930 		adapter->rx_ring->set_itr = 0;
931 	}
932 
933 	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
934 		adapter->total_rx_bytes = 0;
935 		adapter->total_rx_packets = 0;
936 		__napi_schedule(&adapter->rx_ring->napi);
937 	}
938 
939 	return IRQ_HANDLED;
940 }
941 
942 #define IGBVF_NO_QUEUE -1
943 
944 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
945 				int tx_queue, int msix_vector)
946 {
947 	struct e1000_hw *hw = &adapter->hw;
948 	u32 ivar, index;
949 
950 	/* 82576 uses a table-based method for assigning vectors.
951 	 * Each queue has a single entry in the table to which we write
952 	 * a vector number along with a "valid" bit.  Sadly, the layout
953 	 * of the table is somewhat counterintuitive.
954 	 */
955 	if (rx_queue > IGBVF_NO_QUEUE) {
956 		index = (rx_queue >> 1);
957 		ivar = array_er32(IVAR0, index);
958 		if (rx_queue & 0x1) {
959 			/* vector goes into third byte of register */
960 			ivar = ivar & 0xFF00FFFF;
961 			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
962 		} else {
963 			/* vector goes into low byte of register */
964 			ivar = ivar & 0xFFFFFF00;
965 			ivar |= msix_vector | E1000_IVAR_VALID;
966 		}
967 		adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
968 		array_ew32(IVAR0, index, ivar);
969 	}
970 	if (tx_queue > IGBVF_NO_QUEUE) {
971 		index = (tx_queue >> 1);
972 		ivar = array_er32(IVAR0, index);
973 		if (tx_queue & 0x1) {
974 			/* vector goes into high byte of register */
975 			ivar = ivar & 0x00FFFFFF;
976 			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
977 		} else {
978 			/* vector goes into second byte of register */
979 			ivar = ivar & 0xFFFF00FF;
980 			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
981 		}
982 		adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
983 		array_ew32(IVAR0, index, ivar);
984 	}
985 }
986 
987 /**
988  * igbvf_configure_msix - Configure MSI-X hardware
989  * @adapter: board private structure
990  *
991  * igbvf_configure_msix sets up the hardware to properly
992  * generate MSI-X interrupts.
993  **/
994 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
995 {
996 	u32 tmp;
997 	struct e1000_hw *hw = &adapter->hw;
998 	struct igbvf_ring *tx_ring = adapter->tx_ring;
999 	struct igbvf_ring *rx_ring = adapter->rx_ring;
1000 	int vector = 0;
1001 
1002 	adapter->eims_enable_mask = 0;
1003 
1004 	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1005 	adapter->eims_enable_mask |= tx_ring->eims_value;
1006 	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1007 	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1008 	adapter->eims_enable_mask |= rx_ring->eims_value;
1009 	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1010 
1011 	/* set vector for other causes, i.e. link changes */
1012 
1013 	tmp = (vector++ | E1000_IVAR_VALID);
1014 
1015 	ew32(IVAR_MISC, tmp);
1016 
1017 	adapter->eims_enable_mask = (1 << (vector)) - 1;
1018 	adapter->eims_other = 1 << (vector - 1);
1019 	e1e_flush();
1020 }
1021 
1022 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1023 {
1024 	if (adapter->msix_entries) {
1025 		pci_disable_msix(adapter->pdev);
1026 		kfree(adapter->msix_entries);
1027 		adapter->msix_entries = NULL;
1028 	}
1029 }
1030 
1031 /**
1032  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1033  * @adapter: board private structure
1034  *
1035  * Attempt to configure interrupts using the best available
1036  * capabilities of the hardware and kernel.
1037  **/
1038 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1039 {
1040 	int err = -ENOMEM;
1041 	int i;
1042 
1043 	/* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1044 	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1045 					GFP_KERNEL);
1046 	if (adapter->msix_entries) {
1047 		for (i = 0; i < 3; i++)
1048 			adapter->msix_entries[i].entry = i;
1049 
1050 		err = pci_enable_msix_range(adapter->pdev,
1051 					    adapter->msix_entries, 3, 3);
1052 	}
1053 
1054 	if (err < 0) {
1055 		/* MSI-X failed */
1056 		dev_err(&adapter->pdev->dev,
1057 			"Failed to initialize MSI-X interrupts.\n");
1058 		igbvf_reset_interrupt_capability(adapter);
1059 	}
1060 }
1061 
1062 /**
1063  * igbvf_request_msix - Initialize MSI-X interrupts
1064  * @adapter: board private structure
1065  *
1066  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1067  * kernel.
1068  **/
1069 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1070 {
1071 	struct net_device *netdev = adapter->netdev;
1072 	int err = 0, vector = 0;
1073 
1074 	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1075 		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1076 		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1077 	} else {
1078 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1079 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1080 	}
1081 
1082 	err = request_irq(adapter->msix_entries[vector].vector,
1083 			  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1084 			  netdev);
1085 	if (err)
1086 		goto out;
1087 
1088 	adapter->tx_ring->itr_register = E1000_EITR(vector);
1089 	adapter->tx_ring->itr_val = adapter->current_itr;
1090 	vector++;
1091 
1092 	err = request_irq(adapter->msix_entries[vector].vector,
1093 			  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1094 			  netdev);
1095 	if (err)
1096 		goto out;
1097 
1098 	adapter->rx_ring->itr_register = E1000_EITR(vector);
1099 	adapter->rx_ring->itr_val = adapter->current_itr;
1100 	vector++;
1101 
1102 	err = request_irq(adapter->msix_entries[vector].vector,
1103 			  igbvf_msix_other, 0, netdev->name, netdev);
1104 	if (err)
1105 		goto out;
1106 
1107 	igbvf_configure_msix(adapter);
1108 	return 0;
1109 out:
1110 	return err;
1111 }
1112 
1113 /**
1114  * igbvf_alloc_queues - Allocate memory for all rings
1115  * @adapter: board private structure to initialize
1116  **/
1117 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1118 {
1119 	struct net_device *netdev = adapter->netdev;
1120 
1121 	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1122 	if (!adapter->tx_ring)
1123 		return -ENOMEM;
1124 
1125 	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1126 	if (!adapter->rx_ring) {
1127 		kfree(adapter->tx_ring);
1128 		return -ENOMEM;
1129 	}
1130 
1131 	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1132 
1133 	return 0;
1134 }
1135 
1136 /**
1137  * igbvf_request_irq - initialize interrupts
1138  * @adapter: board private structure
1139  *
1140  * Attempts to configure interrupts using the best available
1141  * capabilities of the hardware and kernel.
1142  **/
1143 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1144 {
1145 	int err = -1;
1146 
1147 	/* igbvf supports msi-x only */
1148 	if (adapter->msix_entries)
1149 		err = igbvf_request_msix(adapter);
1150 
1151 	if (!err)
1152 		return err;
1153 
1154 	dev_err(&adapter->pdev->dev,
1155 		"Unable to allocate interrupt, Error: %d\n", err);
1156 
1157 	return err;
1158 }
1159 
1160 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1161 {
1162 	struct net_device *netdev = adapter->netdev;
1163 	int vector;
1164 
1165 	if (adapter->msix_entries) {
1166 		for (vector = 0; vector < 3; vector++)
1167 			free_irq(adapter->msix_entries[vector].vector, netdev);
1168 	}
1169 }
1170 
1171 /**
1172  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1173  * @adapter: board private structure
1174  **/
1175 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1176 {
1177 	struct e1000_hw *hw = &adapter->hw;
1178 
1179 	ew32(EIMC, ~0);
1180 
1181 	if (adapter->msix_entries)
1182 		ew32(EIAC, 0);
1183 }
1184 
1185 /**
1186  * igbvf_irq_enable - Enable default interrupt generation settings
1187  * @adapter: board private structure
1188  **/
1189 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1190 {
1191 	struct e1000_hw *hw = &adapter->hw;
1192 
1193 	ew32(EIAC, adapter->eims_enable_mask);
1194 	ew32(EIAM, adapter->eims_enable_mask);
1195 	ew32(EIMS, adapter->eims_enable_mask);
1196 }
1197 
1198 /**
1199  * igbvf_poll - NAPI Rx polling callback
1200  * @napi: struct associated with this polling callback
1201  * @budget: amount of packets driver is allowed to process this poll
1202  **/
1203 static int igbvf_poll(struct napi_struct *napi, int budget)
1204 {
1205 	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1206 	struct igbvf_adapter *adapter = rx_ring->adapter;
1207 	struct e1000_hw *hw = &adapter->hw;
1208 	int work_done = 0;
1209 
1210 	igbvf_clean_rx_irq(adapter, &work_done, budget);
1211 
1212 	/* If not enough Rx work done, exit the polling mode */
1213 	if (work_done < budget) {
1214 		napi_complete(napi);
1215 
1216 		if (adapter->requested_itr & 3)
1217 			igbvf_set_itr(adapter);
1218 
1219 		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1220 			ew32(EIMS, adapter->rx_ring->eims_value);
1221 	}
1222 
1223 	return work_done;
1224 }
1225 
1226 /**
1227  * igbvf_set_rlpml - set receive large packet maximum length
1228  * @adapter: board private structure
1229  *
1230  * Configure the maximum size of packets that will be received
1231  */
1232 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1233 {
1234 	int max_frame_size;
1235 	struct e1000_hw *hw = &adapter->hw;
1236 
1237 	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1238 	e1000_rlpml_set_vf(hw, max_frame_size);
1239 }
1240 
1241 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1242 				 __be16 proto, u16 vid)
1243 {
1244 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1245 	struct e1000_hw *hw = &adapter->hw;
1246 
1247 	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1248 		dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1249 		return -EINVAL;
1250 	}
1251 	set_bit(vid, adapter->active_vlans);
1252 	return 0;
1253 }
1254 
1255 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1256 				  __be16 proto, u16 vid)
1257 {
1258 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1259 	struct e1000_hw *hw = &adapter->hw;
1260 
1261 	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1262 		dev_err(&adapter->pdev->dev,
1263 			"Failed to remove vlan id %d\n", vid);
1264 		return -EINVAL;
1265 	}
1266 	clear_bit(vid, adapter->active_vlans);
1267 	return 0;
1268 }
1269 
1270 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1271 {
1272 	u16 vid;
1273 
1274 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1275 		igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1276 }
1277 
1278 /**
1279  * igbvf_configure_tx - Configure Transmit Unit after Reset
1280  * @adapter: board private structure
1281  *
1282  * Configure the Tx unit of the MAC after a reset.
1283  **/
1284 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1285 {
1286 	struct e1000_hw *hw = &adapter->hw;
1287 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1288 	u64 tdba;
1289 	u32 txdctl, dca_txctrl;
1290 
1291 	/* disable transmits */
1292 	txdctl = er32(TXDCTL(0));
1293 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1294 	e1e_flush();
1295 	msleep(10);
1296 
1297 	/* Setup the HW Tx Head and Tail descriptor pointers */
1298 	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1299 	tdba = tx_ring->dma;
1300 	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1301 	ew32(TDBAH(0), (tdba >> 32));
1302 	ew32(TDH(0), 0);
1303 	ew32(TDT(0), 0);
1304 	tx_ring->head = E1000_TDH(0);
1305 	tx_ring->tail = E1000_TDT(0);
1306 
1307 	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1308 	 * MUST be delivered in order or it will completely screw up
1309 	 * our bookkeeping.
1310 	 */
1311 	dca_txctrl = er32(DCA_TXCTRL(0));
1312 	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1313 	ew32(DCA_TXCTRL(0), dca_txctrl);
1314 
1315 	/* enable transmits */
1316 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1317 	ew32(TXDCTL(0), txdctl);
1318 
1319 	/* Setup Transmit Descriptor Settings for eop descriptor */
1320 	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1321 
1322 	/* enable Report Status bit */
1323 	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1324 }
1325 
1326 /**
1327  * igbvf_setup_srrctl - configure the receive control registers
1328  * @adapter: Board private structure
1329  **/
1330 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1331 {
1332 	struct e1000_hw *hw = &adapter->hw;
1333 	u32 srrctl = 0;
1334 
1335 	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1336 		    E1000_SRRCTL_BSIZEHDR_MASK |
1337 		    E1000_SRRCTL_BSIZEPKT_MASK);
1338 
1339 	/* Enable queue drop to avoid head of line blocking */
1340 	srrctl |= E1000_SRRCTL_DROP_EN;
1341 
1342 	/* Setup buffer sizes */
1343 	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1344 		  E1000_SRRCTL_BSIZEPKT_SHIFT;
1345 
1346 	if (adapter->rx_buffer_len < 2048) {
1347 		adapter->rx_ps_hdr_size = 0;
1348 		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1349 	} else {
1350 		adapter->rx_ps_hdr_size = 128;
1351 		srrctl |= adapter->rx_ps_hdr_size <<
1352 			  E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1353 		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1354 	}
1355 
1356 	ew32(SRRCTL(0), srrctl);
1357 }
1358 
1359 /**
1360  * igbvf_configure_rx - Configure Receive Unit after Reset
1361  * @adapter: board private structure
1362  *
1363  * Configure the Rx unit of the MAC after a reset.
1364  **/
1365 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1366 {
1367 	struct e1000_hw *hw = &adapter->hw;
1368 	struct igbvf_ring *rx_ring = adapter->rx_ring;
1369 	u64 rdba;
1370 	u32 rdlen, rxdctl;
1371 
1372 	/* disable receives */
1373 	rxdctl = er32(RXDCTL(0));
1374 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1375 	e1e_flush();
1376 	msleep(10);
1377 
1378 	rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1379 
1380 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1381 	 * the Base and Length of the Rx Descriptor Ring
1382 	 */
1383 	rdba = rx_ring->dma;
1384 	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1385 	ew32(RDBAH(0), (rdba >> 32));
1386 	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1387 	rx_ring->head = E1000_RDH(0);
1388 	rx_ring->tail = E1000_RDT(0);
1389 	ew32(RDH(0), 0);
1390 	ew32(RDT(0), 0);
1391 
1392 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1393 	rxdctl &= 0xFFF00000;
1394 	rxdctl |= IGBVF_RX_PTHRESH;
1395 	rxdctl |= IGBVF_RX_HTHRESH << 8;
1396 	rxdctl |= IGBVF_RX_WTHRESH << 16;
1397 
1398 	igbvf_set_rlpml(adapter);
1399 
1400 	/* enable receives */
1401 	ew32(RXDCTL(0), rxdctl);
1402 }
1403 
1404 /**
1405  * igbvf_set_multi - Multicast and Promiscuous mode set
1406  * @netdev: network interface device structure
1407  *
1408  * The set_multi entry point is called whenever the multicast address
1409  * list or the network interface flags are updated.  This routine is
1410  * responsible for configuring the hardware for proper multicast,
1411  * promiscuous mode, and all-multi behavior.
1412  **/
1413 static void igbvf_set_multi(struct net_device *netdev)
1414 {
1415 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1416 	struct e1000_hw *hw = &adapter->hw;
1417 	struct netdev_hw_addr *ha;
1418 	u8  *mta_list = NULL;
1419 	int i;
1420 
1421 	if (!netdev_mc_empty(netdev)) {
1422 		mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1423 					 GFP_ATOMIC);
1424 		if (!mta_list)
1425 			return;
1426 	}
1427 
1428 	/* prepare a packed array of only addresses. */
1429 	i = 0;
1430 	netdev_for_each_mc_addr(ha, netdev)
1431 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1432 
1433 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1434 	kfree(mta_list);
1435 }
1436 
1437 /**
1438  * igbvf_configure - configure the hardware for Rx and Tx
1439  * @adapter: private board structure
1440  **/
1441 static void igbvf_configure(struct igbvf_adapter *adapter)
1442 {
1443 	igbvf_set_multi(adapter->netdev);
1444 
1445 	igbvf_restore_vlan(adapter);
1446 
1447 	igbvf_configure_tx(adapter);
1448 	igbvf_setup_srrctl(adapter);
1449 	igbvf_configure_rx(adapter);
1450 	igbvf_alloc_rx_buffers(adapter->rx_ring,
1451 			       igbvf_desc_unused(adapter->rx_ring));
1452 }
1453 
1454 /* igbvf_reset - bring the hardware into a known good state
1455  * @adapter: private board structure
1456  *
1457  * This function boots the hardware and enables some settings that
1458  * require a configuration cycle of the hardware - those cannot be
1459  * set/changed during runtime. After reset the device needs to be
1460  * properly configured for Rx, Tx etc.
1461  */
1462 static void igbvf_reset(struct igbvf_adapter *adapter)
1463 {
1464 	struct e1000_mac_info *mac = &adapter->hw.mac;
1465 	struct net_device *netdev = adapter->netdev;
1466 	struct e1000_hw *hw = &adapter->hw;
1467 
1468 	/* Allow time for pending master requests to run */
1469 	if (mac->ops.reset_hw(hw))
1470 		dev_err(&adapter->pdev->dev, "PF still resetting\n");
1471 
1472 	mac->ops.init_hw(hw);
1473 
1474 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1475 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1476 		       netdev->addr_len);
1477 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1478 		       netdev->addr_len);
1479 	}
1480 
1481 	adapter->last_reset = jiffies;
1482 }
1483 
1484 int igbvf_up(struct igbvf_adapter *adapter)
1485 {
1486 	struct e1000_hw *hw = &adapter->hw;
1487 
1488 	/* hardware has been reset, we need to reload some things */
1489 	igbvf_configure(adapter);
1490 
1491 	clear_bit(__IGBVF_DOWN, &adapter->state);
1492 
1493 	napi_enable(&adapter->rx_ring->napi);
1494 	if (adapter->msix_entries)
1495 		igbvf_configure_msix(adapter);
1496 
1497 	/* Clear any pending interrupts. */
1498 	er32(EICR);
1499 	igbvf_irq_enable(adapter);
1500 
1501 	/* start the watchdog */
1502 	hw->mac.get_link_status = 1;
1503 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1504 
1505 	return 0;
1506 }
1507 
1508 void igbvf_down(struct igbvf_adapter *adapter)
1509 {
1510 	struct net_device *netdev = adapter->netdev;
1511 	struct e1000_hw *hw = &adapter->hw;
1512 	u32 rxdctl, txdctl;
1513 
1514 	/* signal that we're down so the interrupt handler does not
1515 	 * reschedule our watchdog timer
1516 	 */
1517 	set_bit(__IGBVF_DOWN, &adapter->state);
1518 
1519 	/* disable receives in the hardware */
1520 	rxdctl = er32(RXDCTL(0));
1521 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1522 
1523 	netif_carrier_off(netdev);
1524 	netif_stop_queue(netdev);
1525 
1526 	/* disable transmits in the hardware */
1527 	txdctl = er32(TXDCTL(0));
1528 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1529 
1530 	/* flush both disables and wait for them to finish */
1531 	e1e_flush();
1532 	msleep(10);
1533 
1534 	napi_disable(&adapter->rx_ring->napi);
1535 
1536 	igbvf_irq_disable(adapter);
1537 
1538 	del_timer_sync(&adapter->watchdog_timer);
1539 
1540 	/* record the stats before reset*/
1541 	igbvf_update_stats(adapter);
1542 
1543 	adapter->link_speed = 0;
1544 	adapter->link_duplex = 0;
1545 
1546 	igbvf_reset(adapter);
1547 	igbvf_clean_tx_ring(adapter->tx_ring);
1548 	igbvf_clean_rx_ring(adapter->rx_ring);
1549 }
1550 
1551 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1552 {
1553 	might_sleep();
1554 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1555 		usleep_range(1000, 2000);
1556 	igbvf_down(adapter);
1557 	igbvf_up(adapter);
1558 	clear_bit(__IGBVF_RESETTING, &adapter->state);
1559 }
1560 
1561 /**
1562  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1563  * @adapter: board private structure to initialize
1564  *
1565  * igbvf_sw_init initializes the Adapter private data structure.
1566  * Fields are initialized based on PCI device information and
1567  * OS network device settings (MTU size).
1568  **/
1569 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1570 {
1571 	struct net_device *netdev = adapter->netdev;
1572 	s32 rc;
1573 
1574 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1575 	adapter->rx_ps_hdr_size = 0;
1576 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1577 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1578 
1579 	adapter->tx_int_delay = 8;
1580 	adapter->tx_abs_int_delay = 32;
1581 	adapter->rx_int_delay = 0;
1582 	adapter->rx_abs_int_delay = 8;
1583 	adapter->requested_itr = 3;
1584 	adapter->current_itr = IGBVF_START_ITR;
1585 
1586 	/* Set various function pointers */
1587 	adapter->ei->init_ops(&adapter->hw);
1588 
1589 	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1590 	if (rc)
1591 		return rc;
1592 
1593 	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1594 	if (rc)
1595 		return rc;
1596 
1597 	igbvf_set_interrupt_capability(adapter);
1598 
1599 	if (igbvf_alloc_queues(adapter))
1600 		return -ENOMEM;
1601 
1602 	spin_lock_init(&adapter->tx_queue_lock);
1603 
1604 	/* Explicitly disable IRQ since the NIC can be in any state. */
1605 	igbvf_irq_disable(adapter);
1606 
1607 	spin_lock_init(&adapter->stats_lock);
1608 
1609 	set_bit(__IGBVF_DOWN, &adapter->state);
1610 	return 0;
1611 }
1612 
1613 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1614 {
1615 	struct e1000_hw *hw = &adapter->hw;
1616 
1617 	adapter->stats.last_gprc = er32(VFGPRC);
1618 	adapter->stats.last_gorc = er32(VFGORC);
1619 	adapter->stats.last_gptc = er32(VFGPTC);
1620 	adapter->stats.last_gotc = er32(VFGOTC);
1621 	adapter->stats.last_mprc = er32(VFMPRC);
1622 	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1623 	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1624 	adapter->stats.last_gorlbc = er32(VFGORLBC);
1625 	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1626 
1627 	adapter->stats.base_gprc = er32(VFGPRC);
1628 	adapter->stats.base_gorc = er32(VFGORC);
1629 	adapter->stats.base_gptc = er32(VFGPTC);
1630 	adapter->stats.base_gotc = er32(VFGOTC);
1631 	adapter->stats.base_mprc = er32(VFMPRC);
1632 	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1633 	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1634 	adapter->stats.base_gorlbc = er32(VFGORLBC);
1635 	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1636 }
1637 
1638 /**
1639  * igbvf_open - Called when a network interface is made active
1640  * @netdev: network interface device structure
1641  *
1642  * Returns 0 on success, negative value on failure
1643  *
1644  * The open entry point is called when a network interface is made
1645  * active by the system (IFF_UP).  At this point all resources needed
1646  * for transmit and receive operations are allocated, the interrupt
1647  * handler is registered with the OS, the watchdog timer is started,
1648  * and the stack is notified that the interface is ready.
1649  **/
1650 static int igbvf_open(struct net_device *netdev)
1651 {
1652 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1653 	struct e1000_hw *hw = &adapter->hw;
1654 	int err;
1655 
1656 	/* disallow open during test */
1657 	if (test_bit(__IGBVF_TESTING, &adapter->state))
1658 		return -EBUSY;
1659 
1660 	/* allocate transmit descriptors */
1661 	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1662 	if (err)
1663 		goto err_setup_tx;
1664 
1665 	/* allocate receive descriptors */
1666 	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1667 	if (err)
1668 		goto err_setup_rx;
1669 
1670 	/* before we allocate an interrupt, we must be ready to handle it.
1671 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1672 	 * as soon as we call pci_request_irq, so we have to setup our
1673 	 * clean_rx handler before we do so.
1674 	 */
1675 	igbvf_configure(adapter);
1676 
1677 	err = igbvf_request_irq(adapter);
1678 	if (err)
1679 		goto err_req_irq;
1680 
1681 	/* From here on the code is the same as igbvf_up() */
1682 	clear_bit(__IGBVF_DOWN, &adapter->state);
1683 
1684 	napi_enable(&adapter->rx_ring->napi);
1685 
1686 	/* clear any pending interrupts */
1687 	er32(EICR);
1688 
1689 	igbvf_irq_enable(adapter);
1690 
1691 	/* start the watchdog */
1692 	hw->mac.get_link_status = 1;
1693 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1694 
1695 	return 0;
1696 
1697 err_req_irq:
1698 	igbvf_free_rx_resources(adapter->rx_ring);
1699 err_setup_rx:
1700 	igbvf_free_tx_resources(adapter->tx_ring);
1701 err_setup_tx:
1702 	igbvf_reset(adapter);
1703 
1704 	return err;
1705 }
1706 
1707 /**
1708  * igbvf_close - Disables a network interface
1709  * @netdev: network interface device structure
1710  *
1711  * Returns 0, this is not allowed to fail
1712  *
1713  * The close entry point is called when an interface is de-activated
1714  * by the OS.  The hardware is still under the drivers control, but
1715  * needs to be disabled.  A global MAC reset is issued to stop the
1716  * hardware, and all transmit and receive resources are freed.
1717  **/
1718 static int igbvf_close(struct net_device *netdev)
1719 {
1720 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1721 
1722 	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1723 	igbvf_down(adapter);
1724 
1725 	igbvf_free_irq(adapter);
1726 
1727 	igbvf_free_tx_resources(adapter->tx_ring);
1728 	igbvf_free_rx_resources(adapter->rx_ring);
1729 
1730 	return 0;
1731 }
1732 
1733 /**
1734  * igbvf_set_mac - Change the Ethernet Address of the NIC
1735  * @netdev: network interface device structure
1736  * @p: pointer to an address structure
1737  *
1738  * Returns 0 on success, negative on failure
1739  **/
1740 static int igbvf_set_mac(struct net_device *netdev, void *p)
1741 {
1742 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1743 	struct e1000_hw *hw = &adapter->hw;
1744 	struct sockaddr *addr = p;
1745 
1746 	if (!is_valid_ether_addr(addr->sa_data))
1747 		return -EADDRNOTAVAIL;
1748 
1749 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1750 
1751 	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1752 
1753 	if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1754 		return -EADDRNOTAVAIL;
1755 
1756 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1757 
1758 	return 0;
1759 }
1760 
1761 #define UPDATE_VF_COUNTER(reg, name) \
1762 { \
1763 	u32 current_counter = er32(reg); \
1764 	if (current_counter < adapter->stats.last_##name) \
1765 		adapter->stats.name += 0x100000000LL; \
1766 	adapter->stats.last_##name = current_counter; \
1767 	adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1768 	adapter->stats.name |= current_counter; \
1769 }
1770 
1771 /**
1772  * igbvf_update_stats - Update the board statistics counters
1773  * @adapter: board private structure
1774 **/
1775 void igbvf_update_stats(struct igbvf_adapter *adapter)
1776 {
1777 	struct e1000_hw *hw = &adapter->hw;
1778 	struct pci_dev *pdev = adapter->pdev;
1779 
1780 	/* Prevent stats update while adapter is being reset, link is down
1781 	 * or if the pci connection is down.
1782 	 */
1783 	if (adapter->link_speed == 0)
1784 		return;
1785 
1786 	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1787 		return;
1788 
1789 	if (pci_channel_offline(pdev))
1790 		return;
1791 
1792 	UPDATE_VF_COUNTER(VFGPRC, gprc);
1793 	UPDATE_VF_COUNTER(VFGORC, gorc);
1794 	UPDATE_VF_COUNTER(VFGPTC, gptc);
1795 	UPDATE_VF_COUNTER(VFGOTC, gotc);
1796 	UPDATE_VF_COUNTER(VFMPRC, mprc);
1797 	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1798 	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1799 	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1800 	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1801 
1802 	/* Fill out the OS statistics structure */
1803 	adapter->net_stats.multicast = adapter->stats.mprc;
1804 }
1805 
1806 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1807 {
1808 	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1809 		 adapter->link_speed,
1810 		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1811 }
1812 
1813 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1814 {
1815 	struct e1000_hw *hw = &adapter->hw;
1816 	s32 ret_val = E1000_SUCCESS;
1817 	bool link_active;
1818 
1819 	/* If interface is down, stay link down */
1820 	if (test_bit(__IGBVF_DOWN, &adapter->state))
1821 		return false;
1822 
1823 	ret_val = hw->mac.ops.check_for_link(hw);
1824 	link_active = !hw->mac.get_link_status;
1825 
1826 	/* if check for link returns error we will need to reset */
1827 	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1828 		schedule_work(&adapter->reset_task);
1829 
1830 	return link_active;
1831 }
1832 
1833 /**
1834  * igbvf_watchdog - Timer Call-back
1835  * @data: pointer to adapter cast into an unsigned long
1836  **/
1837 static void igbvf_watchdog(unsigned long data)
1838 {
1839 	struct igbvf_adapter *adapter = (struct igbvf_adapter *)data;
1840 
1841 	/* Do the rest outside of interrupt context */
1842 	schedule_work(&adapter->watchdog_task);
1843 }
1844 
1845 static void igbvf_watchdog_task(struct work_struct *work)
1846 {
1847 	struct igbvf_adapter *adapter = container_of(work,
1848 						     struct igbvf_adapter,
1849 						     watchdog_task);
1850 	struct net_device *netdev = adapter->netdev;
1851 	struct e1000_mac_info *mac = &adapter->hw.mac;
1852 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1853 	struct e1000_hw *hw = &adapter->hw;
1854 	u32 link;
1855 	int tx_pending = 0;
1856 
1857 	link = igbvf_has_link(adapter);
1858 
1859 	if (link) {
1860 		if (!netif_carrier_ok(netdev)) {
1861 			mac->ops.get_link_up_info(&adapter->hw,
1862 						  &adapter->link_speed,
1863 						  &adapter->link_duplex);
1864 			igbvf_print_link_info(adapter);
1865 
1866 			netif_carrier_on(netdev);
1867 			netif_wake_queue(netdev);
1868 		}
1869 	} else {
1870 		if (netif_carrier_ok(netdev)) {
1871 			adapter->link_speed = 0;
1872 			adapter->link_duplex = 0;
1873 			dev_info(&adapter->pdev->dev, "Link is Down\n");
1874 			netif_carrier_off(netdev);
1875 			netif_stop_queue(netdev);
1876 		}
1877 	}
1878 
1879 	if (netif_carrier_ok(netdev)) {
1880 		igbvf_update_stats(adapter);
1881 	} else {
1882 		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1883 			      tx_ring->count);
1884 		if (tx_pending) {
1885 			/* We've lost link, so the controller stops DMA,
1886 			 * but we've got queued Tx work that's never going
1887 			 * to get done, so reset controller to flush Tx.
1888 			 * (Do the reset outside of interrupt context).
1889 			 */
1890 			adapter->tx_timeout_count++;
1891 			schedule_work(&adapter->reset_task);
1892 		}
1893 	}
1894 
1895 	/* Cause software interrupt to ensure Rx ring is cleaned */
1896 	ew32(EICS, adapter->rx_ring->eims_value);
1897 
1898 	/* Reset the timer */
1899 	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1900 		mod_timer(&adapter->watchdog_timer,
1901 			  round_jiffies(jiffies + (2 * HZ)));
1902 }
1903 
1904 #define IGBVF_TX_FLAGS_CSUM		0x00000001
1905 #define IGBVF_TX_FLAGS_VLAN		0x00000002
1906 #define IGBVF_TX_FLAGS_TSO		0x00000004
1907 #define IGBVF_TX_FLAGS_IPV4		0x00000008
1908 #define IGBVF_TX_FLAGS_VLAN_MASK	0xffff0000
1909 #define IGBVF_TX_FLAGS_VLAN_SHIFT	16
1910 
1911 static int igbvf_tso(struct igbvf_adapter *adapter,
1912 		     struct igbvf_ring *tx_ring,
1913 		     struct sk_buff *skb, u32 tx_flags, u8 *hdr_len,
1914 		     __be16 protocol)
1915 {
1916 	struct e1000_adv_tx_context_desc *context_desc;
1917 	struct igbvf_buffer *buffer_info;
1918 	u32 info = 0, tu_cmd = 0;
1919 	u32 mss_l4len_idx, l4len;
1920 	unsigned int i;
1921 	int err;
1922 
1923 	*hdr_len = 0;
1924 
1925 	err = skb_cow_head(skb, 0);
1926 	if (err < 0) {
1927 		dev_err(&adapter->pdev->dev, "igbvf_tso returning an error\n");
1928 		return err;
1929 	}
1930 
1931 	l4len = tcp_hdrlen(skb);
1932 	*hdr_len += l4len;
1933 
1934 	if (protocol == htons(ETH_P_IP)) {
1935 		struct iphdr *iph = ip_hdr(skb);
1936 
1937 		iph->tot_len = 0;
1938 		iph->check = 0;
1939 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1940 							 iph->daddr, 0,
1941 							 IPPROTO_TCP,
1942 							 0);
1943 	} else if (skb_is_gso_v6(skb)) {
1944 		ipv6_hdr(skb)->payload_len = 0;
1945 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1946 						       &ipv6_hdr(skb)->daddr,
1947 						       0, IPPROTO_TCP, 0);
1948 	}
1949 
1950 	i = tx_ring->next_to_use;
1951 
1952 	buffer_info = &tx_ring->buffer_info[i];
1953 	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1954 	/* VLAN MACLEN IPLEN */
1955 	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1956 		info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1957 	info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1958 	*hdr_len += skb_network_offset(skb);
1959 	info |= (skb_transport_header(skb) - skb_network_header(skb));
1960 	*hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1961 	context_desc->vlan_macip_lens = cpu_to_le32(info);
1962 
1963 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1964 	tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1965 
1966 	if (protocol == htons(ETH_P_IP))
1967 		tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1968 	tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1969 
1970 	context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1971 
1972 	/* MSS L4LEN IDX */
1973 	mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1974 	mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1975 
1976 	context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1977 	context_desc->seqnum_seed = 0;
1978 
1979 	buffer_info->time_stamp = jiffies;
1980 	buffer_info->dma = 0;
1981 	i++;
1982 	if (i == tx_ring->count)
1983 		i = 0;
1984 
1985 	tx_ring->next_to_use = i;
1986 
1987 	return true;
1988 }
1989 
1990 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1991 				 struct igbvf_ring *tx_ring,
1992 				 struct sk_buff *skb, u32 tx_flags,
1993 				 __be16 protocol)
1994 {
1995 	struct e1000_adv_tx_context_desc *context_desc;
1996 	unsigned int i;
1997 	struct igbvf_buffer *buffer_info;
1998 	u32 info = 0, tu_cmd = 0;
1999 
2000 	if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2001 	    (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2002 		i = tx_ring->next_to_use;
2003 		buffer_info = &tx_ring->buffer_info[i];
2004 		context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2005 
2006 		if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2007 			info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2008 
2009 		info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2010 		if (skb->ip_summed == CHECKSUM_PARTIAL)
2011 			info |= (skb_transport_header(skb) -
2012 				 skb_network_header(skb));
2013 
2014 		context_desc->vlan_macip_lens = cpu_to_le32(info);
2015 
2016 		tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2017 
2018 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2019 			switch (protocol) {
2020 			case htons(ETH_P_IP):
2021 				tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2022 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2023 					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2024 				break;
2025 			case htons(ETH_P_IPV6):
2026 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2027 					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2028 				break;
2029 			default:
2030 				break;
2031 			}
2032 		}
2033 
2034 		context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2035 		context_desc->seqnum_seed = 0;
2036 		context_desc->mss_l4len_idx = 0;
2037 
2038 		buffer_info->time_stamp = jiffies;
2039 		buffer_info->dma = 0;
2040 		i++;
2041 		if (i == tx_ring->count)
2042 			i = 0;
2043 		tx_ring->next_to_use = i;
2044 
2045 		return true;
2046 	}
2047 
2048 	return false;
2049 }
2050 
2051 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2052 {
2053 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2054 
2055 	/* there is enough descriptors then we don't need to worry  */
2056 	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2057 		return 0;
2058 
2059 	netif_stop_queue(netdev);
2060 
2061 	/* Herbert's original patch had:
2062 	 *  smp_mb__after_netif_stop_queue();
2063 	 * but since that doesn't exist yet, just open code it.
2064 	 */
2065 	smp_mb();
2066 
2067 	/* We need to check again just in case room has been made available */
2068 	if (igbvf_desc_unused(adapter->tx_ring) < size)
2069 		return -EBUSY;
2070 
2071 	netif_wake_queue(netdev);
2072 
2073 	++adapter->restart_queue;
2074 	return 0;
2075 }
2076 
2077 #define IGBVF_MAX_TXD_PWR	16
2078 #define IGBVF_MAX_DATA_PER_TXD	(1 << IGBVF_MAX_TXD_PWR)
2079 
2080 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2081 				   struct igbvf_ring *tx_ring,
2082 				   struct sk_buff *skb)
2083 {
2084 	struct igbvf_buffer *buffer_info;
2085 	struct pci_dev *pdev = adapter->pdev;
2086 	unsigned int len = skb_headlen(skb);
2087 	unsigned int count = 0, i;
2088 	unsigned int f;
2089 
2090 	i = tx_ring->next_to_use;
2091 
2092 	buffer_info = &tx_ring->buffer_info[i];
2093 	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2094 	buffer_info->length = len;
2095 	/* set time_stamp *before* dma to help avoid a possible race */
2096 	buffer_info->time_stamp = jiffies;
2097 	buffer_info->mapped_as_page = false;
2098 	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2099 					  DMA_TO_DEVICE);
2100 	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2101 		goto dma_error;
2102 
2103 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2104 		const struct skb_frag_struct *frag;
2105 
2106 		count++;
2107 		i++;
2108 		if (i == tx_ring->count)
2109 			i = 0;
2110 
2111 		frag = &skb_shinfo(skb)->frags[f];
2112 		len = skb_frag_size(frag);
2113 
2114 		buffer_info = &tx_ring->buffer_info[i];
2115 		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2116 		buffer_info->length = len;
2117 		buffer_info->time_stamp = jiffies;
2118 		buffer_info->mapped_as_page = true;
2119 		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2120 						    DMA_TO_DEVICE);
2121 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2122 			goto dma_error;
2123 	}
2124 
2125 	tx_ring->buffer_info[i].skb = skb;
2126 
2127 	return ++count;
2128 
2129 dma_error:
2130 	dev_err(&pdev->dev, "TX DMA map failed\n");
2131 
2132 	/* clear timestamp and dma mappings for failed buffer_info mapping */
2133 	buffer_info->dma = 0;
2134 	buffer_info->time_stamp = 0;
2135 	buffer_info->length = 0;
2136 	buffer_info->mapped_as_page = false;
2137 	if (count)
2138 		count--;
2139 
2140 	/* clear timestamp and dma mappings for remaining portion of packet */
2141 	while (count--) {
2142 		if (i == 0)
2143 			i += tx_ring->count;
2144 		i--;
2145 		buffer_info = &tx_ring->buffer_info[i];
2146 		igbvf_put_txbuf(adapter, buffer_info);
2147 	}
2148 
2149 	return 0;
2150 }
2151 
2152 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2153 				      struct igbvf_ring *tx_ring,
2154 				      int tx_flags, int count,
2155 				      unsigned int first, u32 paylen,
2156 				      u8 hdr_len)
2157 {
2158 	union e1000_adv_tx_desc *tx_desc = NULL;
2159 	struct igbvf_buffer *buffer_info;
2160 	u32 olinfo_status = 0, cmd_type_len;
2161 	unsigned int i;
2162 
2163 	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2164 			E1000_ADVTXD_DCMD_DEXT);
2165 
2166 	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2167 		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2168 
2169 	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2170 		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2171 
2172 		/* insert tcp checksum */
2173 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2174 
2175 		/* insert ip checksum */
2176 		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2177 			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2178 
2179 	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2180 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2181 	}
2182 
2183 	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2184 
2185 	i = tx_ring->next_to_use;
2186 	while (count--) {
2187 		buffer_info = &tx_ring->buffer_info[i];
2188 		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2189 		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2190 		tx_desc->read.cmd_type_len =
2191 			 cpu_to_le32(cmd_type_len | buffer_info->length);
2192 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2193 		i++;
2194 		if (i == tx_ring->count)
2195 			i = 0;
2196 	}
2197 
2198 	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2199 	/* Force memory writes to complete before letting h/w
2200 	 * know there are new descriptors to fetch.  (Only
2201 	 * applicable for weak-ordered memory model archs,
2202 	 * such as IA-64).
2203 	 */
2204 	wmb();
2205 
2206 	tx_ring->buffer_info[first].next_to_watch = tx_desc;
2207 	tx_ring->next_to_use = i;
2208 	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2209 	/* we need this if more than one processor can write to our tail
2210 	 * at a time, it synchronizes IO on IA64/Altix systems
2211 	 */
2212 	mmiowb();
2213 }
2214 
2215 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2216 					     struct net_device *netdev,
2217 					     struct igbvf_ring *tx_ring)
2218 {
2219 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2220 	unsigned int first, tx_flags = 0;
2221 	u8 hdr_len = 0;
2222 	int count = 0;
2223 	int tso = 0;
2224 	__be16 protocol = vlan_get_protocol(skb);
2225 
2226 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2227 		dev_kfree_skb_any(skb);
2228 		return NETDEV_TX_OK;
2229 	}
2230 
2231 	if (skb->len <= 0) {
2232 		dev_kfree_skb_any(skb);
2233 		return NETDEV_TX_OK;
2234 	}
2235 
2236 	/* need: count + 4 desc gap to keep tail from touching
2237 	 *       + 2 desc gap to keep tail from touching head,
2238 	 *       + 1 desc for skb->data,
2239 	 *       + 1 desc for context descriptor,
2240 	 * head, otherwise try next time
2241 	 */
2242 	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2243 		/* this is a hard error */
2244 		return NETDEV_TX_BUSY;
2245 	}
2246 
2247 	if (skb_vlan_tag_present(skb)) {
2248 		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2249 		tx_flags |= (skb_vlan_tag_get(skb) <<
2250 			     IGBVF_TX_FLAGS_VLAN_SHIFT);
2251 	}
2252 
2253 	if (protocol == htons(ETH_P_IP))
2254 		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2255 
2256 	first = tx_ring->next_to_use;
2257 
2258 	tso = skb_is_gso(skb) ?
2259 		igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len, protocol) : 0;
2260 	if (unlikely(tso < 0)) {
2261 		dev_kfree_skb_any(skb);
2262 		return NETDEV_TX_OK;
2263 	}
2264 
2265 	if (tso)
2266 		tx_flags |= IGBVF_TX_FLAGS_TSO;
2267 	else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags, protocol) &&
2268 		 (skb->ip_summed == CHECKSUM_PARTIAL))
2269 		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2270 
2271 	/* count reflects descriptors mapped, if 0 then mapping error
2272 	 * has occurred and we need to rewind the descriptor queue
2273 	 */
2274 	count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2275 
2276 	if (count) {
2277 		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2278 				   first, skb->len, hdr_len);
2279 		/* Make sure there is space in the ring for the next send. */
2280 		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2281 	} else {
2282 		dev_kfree_skb_any(skb);
2283 		tx_ring->buffer_info[first].time_stamp = 0;
2284 		tx_ring->next_to_use = first;
2285 	}
2286 
2287 	return NETDEV_TX_OK;
2288 }
2289 
2290 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2291 				    struct net_device *netdev)
2292 {
2293 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2294 	struct igbvf_ring *tx_ring;
2295 
2296 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2297 		dev_kfree_skb_any(skb);
2298 		return NETDEV_TX_OK;
2299 	}
2300 
2301 	tx_ring = &adapter->tx_ring[0];
2302 
2303 	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2304 }
2305 
2306 /**
2307  * igbvf_tx_timeout - Respond to a Tx Hang
2308  * @netdev: network interface device structure
2309  **/
2310 static void igbvf_tx_timeout(struct net_device *netdev)
2311 {
2312 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2313 
2314 	/* Do the reset outside of interrupt context */
2315 	adapter->tx_timeout_count++;
2316 	schedule_work(&adapter->reset_task);
2317 }
2318 
2319 static void igbvf_reset_task(struct work_struct *work)
2320 {
2321 	struct igbvf_adapter *adapter;
2322 
2323 	adapter = container_of(work, struct igbvf_adapter, reset_task);
2324 
2325 	igbvf_reinit_locked(adapter);
2326 }
2327 
2328 /**
2329  * igbvf_get_stats - Get System Network Statistics
2330  * @netdev: network interface device structure
2331  *
2332  * Returns the address of the device statistics structure.
2333  * The statistics are actually updated from the timer callback.
2334  **/
2335 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2336 {
2337 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2338 
2339 	/* only return the current stats */
2340 	return &adapter->net_stats;
2341 }
2342 
2343 /**
2344  * igbvf_change_mtu - Change the Maximum Transfer Unit
2345  * @netdev: network interface device structure
2346  * @new_mtu: new value for maximum frame size
2347  *
2348  * Returns 0 on success, negative on failure
2349  **/
2350 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2351 {
2352 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2353 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2354 
2355 	if (new_mtu < 68 || new_mtu > INT_MAX - ETH_HLEN - ETH_FCS_LEN ||
2356 	    max_frame > MAX_JUMBO_FRAME_SIZE)
2357 		return -EINVAL;
2358 
2359 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2360 	if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2361 		dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2362 		return -EINVAL;
2363 	}
2364 
2365 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2366 		usleep_range(1000, 2000);
2367 	/* igbvf_down has a dependency on max_frame_size */
2368 	adapter->max_frame_size = max_frame;
2369 	if (netif_running(netdev))
2370 		igbvf_down(adapter);
2371 
2372 	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2373 	 * means we reserve 2 more, this pushes us to allocate from the next
2374 	 * larger slab size.
2375 	 * i.e. RXBUFFER_2048 --> size-4096 slab
2376 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2377 	 * fragmented skbs
2378 	 */
2379 
2380 	if (max_frame <= 1024)
2381 		adapter->rx_buffer_len = 1024;
2382 	else if (max_frame <= 2048)
2383 		adapter->rx_buffer_len = 2048;
2384 	else
2385 #if (PAGE_SIZE / 2) > 16384
2386 		adapter->rx_buffer_len = 16384;
2387 #else
2388 		adapter->rx_buffer_len = PAGE_SIZE / 2;
2389 #endif
2390 
2391 	/* adjust allocation if LPE protects us, and we aren't using SBP */
2392 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2393 	    (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2394 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2395 					 ETH_FCS_LEN;
2396 
2397 	dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2398 		 netdev->mtu, new_mtu);
2399 	netdev->mtu = new_mtu;
2400 
2401 	if (netif_running(netdev))
2402 		igbvf_up(adapter);
2403 	else
2404 		igbvf_reset(adapter);
2405 
2406 	clear_bit(__IGBVF_RESETTING, &adapter->state);
2407 
2408 	return 0;
2409 }
2410 
2411 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2412 {
2413 	switch (cmd) {
2414 	default:
2415 		return -EOPNOTSUPP;
2416 	}
2417 }
2418 
2419 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2420 {
2421 	struct net_device *netdev = pci_get_drvdata(pdev);
2422 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2423 #ifdef CONFIG_PM
2424 	int retval = 0;
2425 #endif
2426 
2427 	netif_device_detach(netdev);
2428 
2429 	if (netif_running(netdev)) {
2430 		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2431 		igbvf_down(adapter);
2432 		igbvf_free_irq(adapter);
2433 	}
2434 
2435 #ifdef CONFIG_PM
2436 	retval = pci_save_state(pdev);
2437 	if (retval)
2438 		return retval;
2439 #endif
2440 
2441 	pci_disable_device(pdev);
2442 
2443 	return 0;
2444 }
2445 
2446 #ifdef CONFIG_PM
2447 static int igbvf_resume(struct pci_dev *pdev)
2448 {
2449 	struct net_device *netdev = pci_get_drvdata(pdev);
2450 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2451 	u32 err;
2452 
2453 	pci_restore_state(pdev);
2454 	err = pci_enable_device_mem(pdev);
2455 	if (err) {
2456 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2457 		return err;
2458 	}
2459 
2460 	pci_set_master(pdev);
2461 
2462 	if (netif_running(netdev)) {
2463 		err = igbvf_request_irq(adapter);
2464 		if (err)
2465 			return err;
2466 	}
2467 
2468 	igbvf_reset(adapter);
2469 
2470 	if (netif_running(netdev))
2471 		igbvf_up(adapter);
2472 
2473 	netif_device_attach(netdev);
2474 
2475 	return 0;
2476 }
2477 #endif
2478 
2479 static void igbvf_shutdown(struct pci_dev *pdev)
2480 {
2481 	igbvf_suspend(pdev, PMSG_SUSPEND);
2482 }
2483 
2484 #ifdef CONFIG_NET_POLL_CONTROLLER
2485 /* Polling 'interrupt' - used by things like netconsole to send skbs
2486  * without having to re-enable interrupts. It's not called while
2487  * the interrupt routine is executing.
2488  */
2489 static void igbvf_netpoll(struct net_device *netdev)
2490 {
2491 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2492 
2493 	disable_irq(adapter->pdev->irq);
2494 
2495 	igbvf_clean_tx_irq(adapter->tx_ring);
2496 
2497 	enable_irq(adapter->pdev->irq);
2498 }
2499 #endif
2500 
2501 /**
2502  * igbvf_io_error_detected - called when PCI error is detected
2503  * @pdev: Pointer to PCI device
2504  * @state: The current pci connection state
2505  *
2506  * This function is called after a PCI bus error affecting
2507  * this device has been detected.
2508  */
2509 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2510 						pci_channel_state_t state)
2511 {
2512 	struct net_device *netdev = pci_get_drvdata(pdev);
2513 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2514 
2515 	netif_device_detach(netdev);
2516 
2517 	if (state == pci_channel_io_perm_failure)
2518 		return PCI_ERS_RESULT_DISCONNECT;
2519 
2520 	if (netif_running(netdev))
2521 		igbvf_down(adapter);
2522 	pci_disable_device(pdev);
2523 
2524 	/* Request a slot slot reset. */
2525 	return PCI_ERS_RESULT_NEED_RESET;
2526 }
2527 
2528 /**
2529  * igbvf_io_slot_reset - called after the pci bus has been reset.
2530  * @pdev: Pointer to PCI device
2531  *
2532  * Restart the card from scratch, as if from a cold-boot. Implementation
2533  * resembles the first-half of the igbvf_resume routine.
2534  */
2535 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2536 {
2537 	struct net_device *netdev = pci_get_drvdata(pdev);
2538 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2539 
2540 	if (pci_enable_device_mem(pdev)) {
2541 		dev_err(&pdev->dev,
2542 			"Cannot re-enable PCI device after reset.\n");
2543 		return PCI_ERS_RESULT_DISCONNECT;
2544 	}
2545 	pci_set_master(pdev);
2546 
2547 	igbvf_reset(adapter);
2548 
2549 	return PCI_ERS_RESULT_RECOVERED;
2550 }
2551 
2552 /**
2553  * igbvf_io_resume - called when traffic can start flowing again.
2554  * @pdev: Pointer to PCI device
2555  *
2556  * This callback is called when the error recovery driver tells us that
2557  * its OK to resume normal operation. Implementation resembles the
2558  * second-half of the igbvf_resume routine.
2559  */
2560 static void igbvf_io_resume(struct pci_dev *pdev)
2561 {
2562 	struct net_device *netdev = pci_get_drvdata(pdev);
2563 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2564 
2565 	if (netif_running(netdev)) {
2566 		if (igbvf_up(adapter)) {
2567 			dev_err(&pdev->dev,
2568 				"can't bring device back up after reset\n");
2569 			return;
2570 		}
2571 	}
2572 
2573 	netif_device_attach(netdev);
2574 }
2575 
2576 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2577 {
2578 	struct e1000_hw *hw = &adapter->hw;
2579 	struct net_device *netdev = adapter->netdev;
2580 	struct pci_dev *pdev = adapter->pdev;
2581 
2582 	if (hw->mac.type == e1000_vfadapt_i350)
2583 		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2584 	else
2585 		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2586 	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2587 }
2588 
2589 static int igbvf_set_features(struct net_device *netdev,
2590 			      netdev_features_t features)
2591 {
2592 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2593 
2594 	if (features & NETIF_F_RXCSUM)
2595 		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2596 	else
2597 		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2598 
2599 	return 0;
2600 }
2601 
2602 static const struct net_device_ops igbvf_netdev_ops = {
2603 	.ndo_open		= igbvf_open,
2604 	.ndo_stop		= igbvf_close,
2605 	.ndo_start_xmit		= igbvf_xmit_frame,
2606 	.ndo_get_stats		= igbvf_get_stats,
2607 	.ndo_set_rx_mode	= igbvf_set_multi,
2608 	.ndo_set_mac_address	= igbvf_set_mac,
2609 	.ndo_change_mtu		= igbvf_change_mtu,
2610 	.ndo_do_ioctl		= igbvf_ioctl,
2611 	.ndo_tx_timeout		= igbvf_tx_timeout,
2612 	.ndo_vlan_rx_add_vid	= igbvf_vlan_rx_add_vid,
2613 	.ndo_vlan_rx_kill_vid	= igbvf_vlan_rx_kill_vid,
2614 #ifdef CONFIG_NET_POLL_CONTROLLER
2615 	.ndo_poll_controller	= igbvf_netpoll,
2616 #endif
2617 	.ndo_set_features	= igbvf_set_features,
2618 };
2619 
2620 /**
2621  * igbvf_probe - Device Initialization Routine
2622  * @pdev: PCI device information struct
2623  * @ent: entry in igbvf_pci_tbl
2624  *
2625  * Returns 0 on success, negative on failure
2626  *
2627  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2628  * The OS initialization, configuring of the adapter private structure,
2629  * and a hardware reset occur.
2630  **/
2631 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2632 {
2633 	struct net_device *netdev;
2634 	struct igbvf_adapter *adapter;
2635 	struct e1000_hw *hw;
2636 	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2637 
2638 	static int cards_found;
2639 	int err, pci_using_dac;
2640 
2641 	err = pci_enable_device_mem(pdev);
2642 	if (err)
2643 		return err;
2644 
2645 	pci_using_dac = 0;
2646 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2647 	if (!err) {
2648 		pci_using_dac = 1;
2649 	} else {
2650 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2651 		if (err) {
2652 			dev_err(&pdev->dev,
2653 				"No usable DMA configuration, aborting\n");
2654 			goto err_dma;
2655 		}
2656 	}
2657 
2658 	err = pci_request_regions(pdev, igbvf_driver_name);
2659 	if (err)
2660 		goto err_pci_reg;
2661 
2662 	pci_set_master(pdev);
2663 
2664 	err = -ENOMEM;
2665 	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2666 	if (!netdev)
2667 		goto err_alloc_etherdev;
2668 
2669 	SET_NETDEV_DEV(netdev, &pdev->dev);
2670 
2671 	pci_set_drvdata(pdev, netdev);
2672 	adapter = netdev_priv(netdev);
2673 	hw = &adapter->hw;
2674 	adapter->netdev = netdev;
2675 	adapter->pdev = pdev;
2676 	adapter->ei = ei;
2677 	adapter->pba = ei->pba;
2678 	adapter->flags = ei->flags;
2679 	adapter->hw.back = adapter;
2680 	adapter->hw.mac.type = ei->mac;
2681 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2682 
2683 	/* PCI config space info */
2684 
2685 	hw->vendor_id = pdev->vendor;
2686 	hw->device_id = pdev->device;
2687 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2688 	hw->subsystem_device_id = pdev->subsystem_device;
2689 	hw->revision_id = pdev->revision;
2690 
2691 	err = -EIO;
2692 	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2693 				      pci_resource_len(pdev, 0));
2694 
2695 	if (!adapter->hw.hw_addr)
2696 		goto err_ioremap;
2697 
2698 	if (ei->get_variants) {
2699 		err = ei->get_variants(adapter);
2700 		if (err)
2701 			goto err_get_variants;
2702 	}
2703 
2704 	/* setup adapter struct */
2705 	err = igbvf_sw_init(adapter);
2706 	if (err)
2707 		goto err_sw_init;
2708 
2709 	/* construct the net_device struct */
2710 	netdev->netdev_ops = &igbvf_netdev_ops;
2711 
2712 	igbvf_set_ethtool_ops(netdev);
2713 	netdev->watchdog_timeo = 5 * HZ;
2714 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2715 
2716 	adapter->bd_number = cards_found++;
2717 
2718 	netdev->hw_features = NETIF_F_SG |
2719 			   NETIF_F_IP_CSUM |
2720 			   NETIF_F_IPV6_CSUM |
2721 			   NETIF_F_TSO |
2722 			   NETIF_F_TSO6 |
2723 			   NETIF_F_RXCSUM;
2724 
2725 	netdev->features = netdev->hw_features |
2726 			   NETIF_F_HW_VLAN_CTAG_TX |
2727 			   NETIF_F_HW_VLAN_CTAG_RX |
2728 			   NETIF_F_HW_VLAN_CTAG_FILTER;
2729 
2730 	if (pci_using_dac)
2731 		netdev->features |= NETIF_F_HIGHDMA;
2732 
2733 	netdev->vlan_features |= NETIF_F_TSO;
2734 	netdev->vlan_features |= NETIF_F_TSO6;
2735 	netdev->vlan_features |= NETIF_F_IP_CSUM;
2736 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2737 	netdev->vlan_features |= NETIF_F_SG;
2738 
2739 	/*reset the controller to put the device in a known good state */
2740 	err = hw->mac.ops.reset_hw(hw);
2741 	if (err) {
2742 		dev_info(&pdev->dev,
2743 			 "PF still in reset state. Is the PF interface up?\n");
2744 	} else {
2745 		err = hw->mac.ops.read_mac_addr(hw);
2746 		if (err)
2747 			dev_info(&pdev->dev, "Error reading MAC address.\n");
2748 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2749 			dev_info(&pdev->dev,
2750 				 "MAC address not assigned by administrator.\n");
2751 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2752 		       netdev->addr_len);
2753 	}
2754 
2755 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2756 		dev_info(&pdev->dev, "Assigning random MAC address.\n");
2757 		eth_hw_addr_random(netdev);
2758 		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2759 		       netdev->addr_len);
2760 	}
2761 
2762 	setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2763 		    (unsigned long)adapter);
2764 
2765 	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2766 	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2767 
2768 	/* ring size defaults */
2769 	adapter->rx_ring->count = 1024;
2770 	adapter->tx_ring->count = 1024;
2771 
2772 	/* reset the hardware with the new settings */
2773 	igbvf_reset(adapter);
2774 
2775 	/* set hardware-specific flags */
2776 	if (adapter->hw.mac.type == e1000_vfadapt_i350)
2777 		adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2778 
2779 	strcpy(netdev->name, "eth%d");
2780 	err = register_netdev(netdev);
2781 	if (err)
2782 		goto err_hw_init;
2783 
2784 	/* tell the stack to leave us alone until igbvf_open() is called */
2785 	netif_carrier_off(netdev);
2786 	netif_stop_queue(netdev);
2787 
2788 	igbvf_print_device_info(adapter);
2789 
2790 	igbvf_initialize_last_counter_stats(adapter);
2791 
2792 	return 0;
2793 
2794 err_hw_init:
2795 	kfree(adapter->tx_ring);
2796 	kfree(adapter->rx_ring);
2797 err_sw_init:
2798 	igbvf_reset_interrupt_capability(adapter);
2799 err_get_variants:
2800 	iounmap(adapter->hw.hw_addr);
2801 err_ioremap:
2802 	free_netdev(netdev);
2803 err_alloc_etherdev:
2804 	pci_release_regions(pdev);
2805 err_pci_reg:
2806 err_dma:
2807 	pci_disable_device(pdev);
2808 	return err;
2809 }
2810 
2811 /**
2812  * igbvf_remove - Device Removal Routine
2813  * @pdev: PCI device information struct
2814  *
2815  * igbvf_remove is called by the PCI subsystem to alert the driver
2816  * that it should release a PCI device.  The could be caused by a
2817  * Hot-Plug event, or because the driver is going to be removed from
2818  * memory.
2819  **/
2820 static void igbvf_remove(struct pci_dev *pdev)
2821 {
2822 	struct net_device *netdev = pci_get_drvdata(pdev);
2823 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2824 	struct e1000_hw *hw = &adapter->hw;
2825 
2826 	/* The watchdog timer may be rescheduled, so explicitly
2827 	 * disable it from being rescheduled.
2828 	 */
2829 	set_bit(__IGBVF_DOWN, &adapter->state);
2830 	del_timer_sync(&adapter->watchdog_timer);
2831 
2832 	cancel_work_sync(&adapter->reset_task);
2833 	cancel_work_sync(&adapter->watchdog_task);
2834 
2835 	unregister_netdev(netdev);
2836 
2837 	igbvf_reset_interrupt_capability(adapter);
2838 
2839 	/* it is important to delete the NAPI struct prior to freeing the
2840 	 * Rx ring so that you do not end up with null pointer refs
2841 	 */
2842 	netif_napi_del(&adapter->rx_ring->napi);
2843 	kfree(adapter->tx_ring);
2844 	kfree(adapter->rx_ring);
2845 
2846 	iounmap(hw->hw_addr);
2847 	if (hw->flash_address)
2848 		iounmap(hw->flash_address);
2849 	pci_release_regions(pdev);
2850 
2851 	free_netdev(netdev);
2852 
2853 	pci_disable_device(pdev);
2854 }
2855 
2856 /* PCI Error Recovery (ERS) */
2857 static const struct pci_error_handlers igbvf_err_handler = {
2858 	.error_detected = igbvf_io_error_detected,
2859 	.slot_reset = igbvf_io_slot_reset,
2860 	.resume = igbvf_io_resume,
2861 };
2862 
2863 static const struct pci_device_id igbvf_pci_tbl[] = {
2864 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2865 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2866 	{ } /* terminate list */
2867 };
2868 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2869 
2870 /* PCI Device API Driver */
2871 static struct pci_driver igbvf_driver = {
2872 	.name		= igbvf_driver_name,
2873 	.id_table	= igbvf_pci_tbl,
2874 	.probe		= igbvf_probe,
2875 	.remove		= igbvf_remove,
2876 #ifdef CONFIG_PM
2877 	/* Power Management Hooks */
2878 	.suspend	= igbvf_suspend,
2879 	.resume		= igbvf_resume,
2880 #endif
2881 	.shutdown	= igbvf_shutdown,
2882 	.err_handler	= &igbvf_err_handler
2883 };
2884 
2885 /**
2886  * igbvf_init_module - Driver Registration Routine
2887  *
2888  * igbvf_init_module is the first routine called when the driver is
2889  * loaded. All it does is register with the PCI subsystem.
2890  **/
2891 static int __init igbvf_init_module(void)
2892 {
2893 	int ret;
2894 
2895 	pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2896 	pr_info("%s\n", igbvf_copyright);
2897 
2898 	ret = pci_register_driver(&igbvf_driver);
2899 
2900 	return ret;
2901 }
2902 module_init(igbvf_init_module);
2903 
2904 /**
2905  * igbvf_exit_module - Driver Exit Cleanup Routine
2906  *
2907  * igbvf_exit_module is called just before the driver is removed
2908  * from memory.
2909  **/
2910 static void __exit igbvf_exit_module(void)
2911 {
2912 	pci_unregister_driver(&igbvf_driver);
2913 }
2914 module_exit(igbvf_exit_module);
2915 
2916 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2917 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2918 MODULE_LICENSE("GPL");
2919 MODULE_VERSION(DRV_VERSION);
2920 
2921 /* netdev.c */
2922