xref: /linux/drivers/net/ethernet/intel/igbvf/netdev.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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, 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/pci.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/delay.h>
37 #include <linux/netdevice.h>
38 #include <linux/tcp.h>
39 #include <linux/ipv6.h>
40 #include <linux/slab.h>
41 #include <net/checksum.h>
42 #include <net/ip6_checksum.h>
43 #include <linux/mii.h>
44 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/prefetch.h>
47 
48 #include "igbvf.h"
49 
50 #define DRV_VERSION "2.0.2-k"
51 char igbvf_driver_name[] = "igbvf";
52 const char igbvf_driver_version[] = DRV_VERSION;
53 static const char igbvf_driver_string[] =
54 		  "Intel(R) Gigabit Virtual Function Network Driver";
55 static const char igbvf_copyright[] =
56 		  "Copyright (c) 2009 - 2012 Intel Corporation.";
57 
58 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
59 static int debug = -1;
60 module_param(debug, int, 0);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62 
63 static int igbvf_poll(struct napi_struct *napi, int budget);
64 static void igbvf_reset(struct igbvf_adapter *);
65 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
67 
68 static struct igbvf_info igbvf_vf_info = {
69 	.mac                    = e1000_vfadapt,
70 	.flags                  = 0,
71 	.pba                    = 10,
72 	.init_ops               = e1000_init_function_pointers_vf,
73 };
74 
75 static struct igbvf_info igbvf_i350_vf_info = {
76 	.mac			= e1000_vfadapt_i350,
77 	.flags			= 0,
78 	.pba			= 10,
79 	.init_ops		= e1000_init_function_pointers_vf,
80 };
81 
82 static const struct igbvf_info *igbvf_info_tbl[] = {
83 	[board_vf]              = &igbvf_vf_info,
84 	[board_i350_vf]		= &igbvf_i350_vf_info,
85 };
86 
87 /**
88  * igbvf_desc_unused - calculate if we have unused descriptors
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 		if (adapter->rx_ps_hdr_size) {
224 			rx_desc->read.pkt_addr =
225 			     cpu_to_le64(buffer_info->page_dma);
226 			rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
227 		} else {
228 			rx_desc->read.pkt_addr =
229 			     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 		wmb();
252 		writel(i, adapter->hw.hw_addr + rx_ring->tail);
253 	}
254 }
255 
256 /**
257  * igbvf_clean_rx_irq - Send received data up the network stack; legacy
258  * @adapter: board private structure
259  *
260  * the return value indicates whether actual cleaning was done, there
261  * is no guarantee that everything was cleaned
262  **/
263 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
264                                int *work_done, int work_to_do)
265 {
266 	struct igbvf_ring *rx_ring = adapter->rx_ring;
267 	struct net_device *netdev = adapter->netdev;
268 	struct pci_dev *pdev = adapter->pdev;
269 	union e1000_adv_rx_desc *rx_desc, *next_rxd;
270 	struct igbvf_buffer *buffer_info, *next_buffer;
271 	struct sk_buff *skb;
272 	bool cleaned = false;
273 	int cleaned_count = 0;
274 	unsigned int total_bytes = 0, total_packets = 0;
275 	unsigned int i;
276 	u32 length, hlen, staterr;
277 
278 	i = rx_ring->next_to_clean;
279 	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
280 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
281 
282 	while (staterr & E1000_RXD_STAT_DD) {
283 		if (*work_done >= work_to_do)
284 			break;
285 		(*work_done)++;
286 		rmb(); /* read descriptor and rx_buffer_info after status DD */
287 
288 		buffer_info = &rx_ring->buffer_info[i];
289 
290 		/* HW will not DMA in data larger than the given buffer, even
291 		 * if it parses the (NFS, of course) header to be larger.  In
292 		 * that case, it fills the header buffer and spills the rest
293 		 * into the page.
294 		 */
295 		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
296 		  E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
297 		if (hlen > adapter->rx_ps_hdr_size)
298 			hlen = adapter->rx_ps_hdr_size;
299 
300 		length = le16_to_cpu(rx_desc->wb.upper.length);
301 		cleaned = true;
302 		cleaned_count++;
303 
304 		skb = buffer_info->skb;
305 		prefetch(skb->data - NET_IP_ALIGN);
306 		buffer_info->skb = NULL;
307 		if (!adapter->rx_ps_hdr_size) {
308 			dma_unmap_single(&pdev->dev, buffer_info->dma,
309 			                 adapter->rx_buffer_len,
310 					 DMA_FROM_DEVICE);
311 			buffer_info->dma = 0;
312 			skb_put(skb, length);
313 			goto send_up;
314 		}
315 
316 		if (!skb_shinfo(skb)->nr_frags) {
317 			dma_unmap_single(&pdev->dev, buffer_info->dma,
318 			                 adapter->rx_ps_hdr_size,
319 					 DMA_FROM_DEVICE);
320 			skb_put(skb, hlen);
321 		}
322 
323 		if (length) {
324 			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
325 			               PAGE_SIZE / 2,
326 				       DMA_FROM_DEVICE);
327 			buffer_info->page_dma = 0;
328 
329 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
330 			                   buffer_info->page,
331 			                   buffer_info->page_offset,
332 			                   length);
333 
334 			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
335 			    (page_count(buffer_info->page) != 1))
336 				buffer_info->page = NULL;
337 			else
338 				get_page(buffer_info->page);
339 
340 			skb->len += length;
341 			skb->data_len += length;
342 			skb->truesize += PAGE_SIZE / 2;
343 		}
344 send_up:
345 		i++;
346 		if (i == rx_ring->count)
347 			i = 0;
348 		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
349 		prefetch(next_rxd);
350 		next_buffer = &rx_ring->buffer_info[i];
351 
352 		if (!(staterr & E1000_RXD_STAT_EOP)) {
353 			buffer_info->skb = next_buffer->skb;
354 			buffer_info->dma = next_buffer->dma;
355 			next_buffer->skb = skb;
356 			next_buffer->dma = 0;
357 			goto next_desc;
358 		}
359 
360 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
361 			dev_kfree_skb_irq(skb);
362 			goto next_desc;
363 		}
364 
365 		total_bytes += skb->len;
366 		total_packets++;
367 
368 		igbvf_rx_checksum_adv(adapter, staterr, skb);
369 
370 		skb->protocol = eth_type_trans(skb, netdev);
371 
372 		igbvf_receive_skb(adapter, netdev, skb, staterr,
373 		                  rx_desc->wb.upper.vlan);
374 
375 next_desc:
376 		rx_desc->wb.upper.status_error = 0;
377 
378 		/* return some buffers to hardware, one at a time is too slow */
379 		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
380 			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
381 			cleaned_count = 0;
382 		}
383 
384 		/* use prefetched values */
385 		rx_desc = next_rxd;
386 		buffer_info = next_buffer;
387 
388 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
389 	}
390 
391 	rx_ring->next_to_clean = i;
392 	cleaned_count = igbvf_desc_unused(rx_ring);
393 
394 	if (cleaned_count)
395 		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
396 
397 	adapter->total_rx_packets += total_packets;
398 	adapter->total_rx_bytes += total_bytes;
399 	adapter->net_stats.rx_bytes += total_bytes;
400 	adapter->net_stats.rx_packets += total_packets;
401 	return cleaned;
402 }
403 
404 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
405                             struct igbvf_buffer *buffer_info)
406 {
407 	if (buffer_info->dma) {
408 		if (buffer_info->mapped_as_page)
409 			dma_unmap_page(&adapter->pdev->dev,
410 				       buffer_info->dma,
411 				       buffer_info->length,
412 				       DMA_TO_DEVICE);
413 		else
414 			dma_unmap_single(&adapter->pdev->dev,
415 					 buffer_info->dma,
416 					 buffer_info->length,
417 					 DMA_TO_DEVICE);
418 		buffer_info->dma = 0;
419 	}
420 	if (buffer_info->skb) {
421 		dev_kfree_skb_any(buffer_info->skb);
422 		buffer_info->skb = NULL;
423 	}
424 	buffer_info->time_stamp = 0;
425 }
426 
427 /**
428  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
429  * @adapter: board private structure
430  *
431  * Return 0 on success, negative on failure
432  **/
433 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
434                              struct igbvf_ring *tx_ring)
435 {
436 	struct pci_dev *pdev = adapter->pdev;
437 	int size;
438 
439 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
440 	tx_ring->buffer_info = vzalloc(size);
441 	if (!tx_ring->buffer_info)
442 		goto err;
443 
444 	/* round up to nearest 4K */
445 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
446 	tx_ring->size = ALIGN(tx_ring->size, 4096);
447 
448 	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
449 					   &tx_ring->dma, GFP_KERNEL);
450 	if (!tx_ring->desc)
451 		goto err;
452 
453 	tx_ring->adapter = adapter;
454 	tx_ring->next_to_use = 0;
455 	tx_ring->next_to_clean = 0;
456 
457 	return 0;
458 err:
459 	vfree(tx_ring->buffer_info);
460 	dev_err(&adapter->pdev->dev,
461 	        "Unable to allocate memory for the transmit descriptor ring\n");
462 	return -ENOMEM;
463 }
464 
465 /**
466  * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
467  * @adapter: board private structure
468  *
469  * Returns 0 on success, negative on failure
470  **/
471 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
472 			     struct igbvf_ring *rx_ring)
473 {
474 	struct pci_dev *pdev = adapter->pdev;
475 	int size, desc_len;
476 
477 	size = sizeof(struct igbvf_buffer) * rx_ring->count;
478 	rx_ring->buffer_info = vzalloc(size);
479 	if (!rx_ring->buffer_info)
480 		goto err;
481 
482 	desc_len = sizeof(union e1000_adv_rx_desc);
483 
484 	/* Round up to nearest 4K */
485 	rx_ring->size = rx_ring->count * desc_len;
486 	rx_ring->size = ALIGN(rx_ring->size, 4096);
487 
488 	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
489 					   &rx_ring->dma, GFP_KERNEL);
490 	if (!rx_ring->desc)
491 		goto err;
492 
493 	rx_ring->next_to_clean = 0;
494 	rx_ring->next_to_use = 0;
495 
496 	rx_ring->adapter = adapter;
497 
498 	return 0;
499 
500 err:
501 	vfree(rx_ring->buffer_info);
502 	rx_ring->buffer_info = NULL;
503 	dev_err(&adapter->pdev->dev,
504 	        "Unable to allocate memory for the receive descriptor ring\n");
505 	return -ENOMEM;
506 }
507 
508 /**
509  * igbvf_clean_tx_ring - Free Tx Buffers
510  * @tx_ring: ring to be cleaned
511  **/
512 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
513 {
514 	struct igbvf_adapter *adapter = tx_ring->adapter;
515 	struct igbvf_buffer *buffer_info;
516 	unsigned long size;
517 	unsigned int i;
518 
519 	if (!tx_ring->buffer_info)
520 		return;
521 
522 	/* Free all the Tx ring sk_buffs */
523 	for (i = 0; i < tx_ring->count; i++) {
524 		buffer_info = &tx_ring->buffer_info[i];
525 		igbvf_put_txbuf(adapter, buffer_info);
526 	}
527 
528 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
529 	memset(tx_ring->buffer_info, 0, size);
530 
531 	/* Zero out the descriptor ring */
532 	memset(tx_ring->desc, 0, tx_ring->size);
533 
534 	tx_ring->next_to_use = 0;
535 	tx_ring->next_to_clean = 0;
536 
537 	writel(0, adapter->hw.hw_addr + tx_ring->head);
538 	writel(0, adapter->hw.hw_addr + tx_ring->tail);
539 }
540 
541 /**
542  * igbvf_free_tx_resources - Free Tx Resources per Queue
543  * @tx_ring: ring to free resources from
544  *
545  * Free all transmit software resources
546  **/
547 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
548 {
549 	struct pci_dev *pdev = tx_ring->adapter->pdev;
550 
551 	igbvf_clean_tx_ring(tx_ring);
552 
553 	vfree(tx_ring->buffer_info);
554 	tx_ring->buffer_info = NULL;
555 
556 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
557 			  tx_ring->dma);
558 
559 	tx_ring->desc = NULL;
560 }
561 
562 /**
563  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
564  * @adapter: board private structure
565  **/
566 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
567 {
568 	struct igbvf_adapter *adapter = rx_ring->adapter;
569 	struct igbvf_buffer *buffer_info;
570 	struct pci_dev *pdev = adapter->pdev;
571 	unsigned long size;
572 	unsigned int i;
573 
574 	if (!rx_ring->buffer_info)
575 		return;
576 
577 	/* Free all the Rx ring sk_buffs */
578 	for (i = 0; i < rx_ring->count; i++) {
579 		buffer_info = &rx_ring->buffer_info[i];
580 		if (buffer_info->dma) {
581 			if (adapter->rx_ps_hdr_size){
582 				dma_unmap_single(&pdev->dev, buffer_info->dma,
583 				                 adapter->rx_ps_hdr_size,
584 						 DMA_FROM_DEVICE);
585 			} else {
586 				dma_unmap_single(&pdev->dev, buffer_info->dma,
587 				                 adapter->rx_buffer_len,
588 						 DMA_FROM_DEVICE);
589 			}
590 			buffer_info->dma = 0;
591 		}
592 
593 		if (buffer_info->skb) {
594 			dev_kfree_skb(buffer_info->skb);
595 			buffer_info->skb = NULL;
596 		}
597 
598 		if (buffer_info->page) {
599 			if (buffer_info->page_dma)
600 				dma_unmap_page(&pdev->dev,
601 					       buffer_info->page_dma,
602 				               PAGE_SIZE / 2,
603 					       DMA_FROM_DEVICE);
604 			put_page(buffer_info->page);
605 			buffer_info->page = NULL;
606 			buffer_info->page_dma = 0;
607 			buffer_info->page_offset = 0;
608 		}
609 	}
610 
611 	size = sizeof(struct igbvf_buffer) * rx_ring->count;
612 	memset(rx_ring->buffer_info, 0, size);
613 
614 	/* Zero out the descriptor ring */
615 	memset(rx_ring->desc, 0, rx_ring->size);
616 
617 	rx_ring->next_to_clean = 0;
618 	rx_ring->next_to_use = 0;
619 
620 	writel(0, adapter->hw.hw_addr + rx_ring->head);
621 	writel(0, adapter->hw.hw_addr + rx_ring->tail);
622 }
623 
624 /**
625  * igbvf_free_rx_resources - Free Rx Resources
626  * @rx_ring: ring to clean the resources from
627  *
628  * Free all receive software resources
629  **/
630 
631 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
632 {
633 	struct pci_dev *pdev = rx_ring->adapter->pdev;
634 
635 	igbvf_clean_rx_ring(rx_ring);
636 
637 	vfree(rx_ring->buffer_info);
638 	rx_ring->buffer_info = NULL;
639 
640 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
641 	                  rx_ring->dma);
642 	rx_ring->desc = NULL;
643 }
644 
645 /**
646  * igbvf_update_itr - update the dynamic ITR value based on statistics
647  * @adapter: pointer to adapter
648  * @itr_setting: current adapter->itr
649  * @packets: the number of packets during this measurement interval
650  * @bytes: the number of bytes during this measurement interval
651  *
652  *      Stores a new ITR value based on packets and byte
653  *      counts during the last interrupt.  The advantage of per interrupt
654  *      computation is faster updates and more accurate ITR for the current
655  *      traffic pattern.  Constants in this function were computed
656  *      based on theoretical maximum wire speed and thresholds were set based
657  *      on testing data as well as attempting to minimize response time
658  *      while increasing bulk throughput.
659  **/
660 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
661 					   enum latency_range itr_setting,
662 					   int packets, int bytes)
663 {
664 	enum latency_range retval = itr_setting;
665 
666 	if (packets == 0)
667 		goto update_itr_done;
668 
669 	switch (itr_setting) {
670 	case lowest_latency:
671 		/* handle TSO and jumbo frames */
672 		if (bytes/packets > 8000)
673 			retval = bulk_latency;
674 		else if ((packets < 5) && (bytes > 512))
675 			retval = low_latency;
676 		break;
677 	case low_latency:  /* 50 usec aka 20000 ints/s */
678 		if (bytes > 10000) {
679 			/* this if handles the TSO accounting */
680 			if (bytes/packets > 8000)
681 				retval = bulk_latency;
682 			else if ((packets < 10) || ((bytes/packets) > 1200))
683 				retval = bulk_latency;
684 			else if ((packets > 35))
685 				retval = lowest_latency;
686 		} else if (bytes/packets > 2000) {
687 			retval = bulk_latency;
688 		} else if (packets <= 2 && bytes < 512) {
689 			retval = lowest_latency;
690 		}
691 		break;
692 	case bulk_latency: /* 250 usec aka 4000 ints/s */
693 		if (bytes > 25000) {
694 			if (packets > 35)
695 				retval = low_latency;
696 		} else if (bytes < 6000) {
697 			retval = low_latency;
698 		}
699 		break;
700 	default:
701 		break;
702 	}
703 
704 update_itr_done:
705 	return retval;
706 }
707 
708 static int igbvf_range_to_itr(enum latency_range current_range)
709 {
710 	int new_itr;
711 
712 	switch (current_range) {
713 	/* counts and packets in update_itr are dependent on these numbers */
714 	case lowest_latency:
715 		new_itr = IGBVF_70K_ITR;
716 		break;
717 	case low_latency:
718 		new_itr = IGBVF_20K_ITR;
719 		break;
720 	case bulk_latency:
721 		new_itr = IGBVF_4K_ITR;
722 		break;
723 	default:
724 		new_itr = IGBVF_START_ITR;
725 		break;
726 	}
727 	return new_itr;
728 }
729 
730 static void igbvf_set_itr(struct igbvf_adapter *adapter)
731 {
732 	u32 new_itr;
733 
734 	adapter->tx_ring->itr_range =
735 			igbvf_update_itr(adapter,
736 					 adapter->tx_ring->itr_val,
737 					 adapter->total_tx_packets,
738 					 adapter->total_tx_bytes);
739 
740 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
741 	if (adapter->requested_itr == 3 &&
742 	    adapter->tx_ring->itr_range == lowest_latency)
743 		adapter->tx_ring->itr_range = low_latency;
744 
745 	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
746 
747 
748 	if (new_itr != adapter->tx_ring->itr_val) {
749 		u32 current_itr = adapter->tx_ring->itr_val;
750 		/*
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 		new_itr = new_itr > current_itr ?
776 			     min(current_itr + (new_itr >> 2), new_itr) :
777 			     new_itr;
778 		adapter->rx_ring->itr_val = new_itr;
779 
780 		adapter->rx_ring->set_itr = 1;
781 	}
782 }
783 
784 /**
785  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
786  * @adapter: board private structure
787  *
788  * returns true if ring is completely cleaned
789  **/
790 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
791 {
792 	struct igbvf_adapter *adapter = tx_ring->adapter;
793 	struct net_device *netdev = adapter->netdev;
794 	struct igbvf_buffer *buffer_info;
795 	struct sk_buff *skb;
796 	union e1000_adv_tx_desc *tx_desc, *eop_desc;
797 	unsigned int total_bytes = 0, total_packets = 0;
798 	unsigned int i, count = 0;
799 	bool cleaned = false;
800 
801 	i = tx_ring->next_to_clean;
802 	buffer_info = &tx_ring->buffer_info[i];
803 	eop_desc = buffer_info->next_to_watch;
804 
805 	do {
806 		/* if next_to_watch is not set then there is no work pending */
807 		if (!eop_desc)
808 			break;
809 
810 		/* prevent any other reads prior to eop_desc */
811 		read_barrier_depends();
812 
813 		/* if DD is not set pending work has not been completed */
814 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
815 			break;
816 
817 		/* clear next_to_watch to prevent false hangs */
818 		buffer_info->next_to_watch = NULL;
819 
820 		for (cleaned = false; !cleaned; count++) {
821 			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
822 			cleaned = (tx_desc == eop_desc);
823 			skb = buffer_info->skb;
824 
825 			if (skb) {
826 				unsigned int segs, bytecount;
827 
828 				/* gso_segs is currently only valid for tcp */
829 				segs = skb_shinfo(skb)->gso_segs ?: 1;
830 				/* multiply data chunks by size of headers */
831 				bytecount = ((segs - 1) * skb_headlen(skb)) +
832 				            skb->len;
833 				total_packets += segs;
834 				total_bytes += bytecount;
835 			}
836 
837 			igbvf_put_txbuf(adapter, buffer_info);
838 			tx_desc->wb.status = 0;
839 
840 			i++;
841 			if (i == tx_ring->count)
842 				i = 0;
843 
844 			buffer_info = &tx_ring->buffer_info[i];
845 		}
846 
847 		eop_desc = buffer_info->next_to_watch;
848 	} while (count < tx_ring->count);
849 
850 	tx_ring->next_to_clean = i;
851 
852 	if (unlikely(count &&
853 	             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 reenable the interrupt when we write
906 	 * EICS */
907 	if (!igbvf_clean_tx_irq(tx_ring))
908 		/* Ring was not completely cleaned, so fire another interrupt */
909 		ew32(EICS, tx_ring->eims_value);
910 	else
911 		ew32(EIMS, tx_ring->eims_value);
912 
913 	return IRQ_HANDLED;
914 }
915 
916 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
917 {
918 	struct net_device *netdev = data;
919 	struct igbvf_adapter *adapter = netdev_priv(netdev);
920 
921 	adapter->int_counter0++;
922 
923 	/* Write the ITR value calculated at the end of the
924 	 * previous interrupt.
925 	 */
926 	if (adapter->rx_ring->set_itr) {
927 		writel(adapter->rx_ring->itr_val,
928 		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
929 		adapter->rx_ring->set_itr = 0;
930 	}
931 
932 	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
933 		adapter->total_rx_bytes = 0;
934 		adapter->total_rx_packets = 0;
935 		__napi_schedule(&adapter->rx_ring->napi);
936 	}
937 
938 	return IRQ_HANDLED;
939 }
940 
941 #define IGBVF_NO_QUEUE -1
942 
943 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
944                                 int tx_queue, int msix_vector)
945 {
946 	struct e1000_hw *hw = &adapter->hw;
947 	u32 ivar, index;
948 
949 	/* 82576 uses a table-based method for assigning vectors.
950 	   Each queue has a single entry in the table to which we write
951 	   a vector number along with a "valid" bit.  Sadly, the layout
952 	   of the table is somewhat counterintuitive. */
953 	if (rx_queue > IGBVF_NO_QUEUE) {
954 		index = (rx_queue >> 1);
955 		ivar = array_er32(IVAR0, index);
956 		if (rx_queue & 0x1) {
957 			/* vector goes into third byte of register */
958 			ivar = ivar & 0xFF00FFFF;
959 			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
960 		} else {
961 			/* vector goes into low byte of register */
962 			ivar = ivar & 0xFFFFFF00;
963 			ivar |= msix_vector | E1000_IVAR_VALID;
964 		}
965 		adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
966 		array_ew32(IVAR0, index, ivar);
967 	}
968 	if (tx_queue > IGBVF_NO_QUEUE) {
969 		index = (tx_queue >> 1);
970 		ivar = array_er32(IVAR0, index);
971 		if (tx_queue & 0x1) {
972 			/* vector goes into high byte of register */
973 			ivar = ivar & 0x00FFFFFF;
974 			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
975 		} else {
976 			/* vector goes into second byte of register */
977 			ivar = ivar & 0xFFFF00FF;
978 			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
979 		}
980 		adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
981 		array_ew32(IVAR0, index, ivar);
982 	}
983 }
984 
985 /**
986  * igbvf_configure_msix - Configure MSI-X hardware
987  *
988  * igbvf_configure_msix sets up the hardware to properly
989  * generate MSI-X interrupts.
990  **/
991 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
992 {
993 	u32 tmp;
994 	struct e1000_hw *hw = &adapter->hw;
995 	struct igbvf_ring *tx_ring = adapter->tx_ring;
996 	struct igbvf_ring *rx_ring = adapter->rx_ring;
997 	int vector = 0;
998 
999 	adapter->eims_enable_mask = 0;
1000 
1001 	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1002 	adapter->eims_enable_mask |= tx_ring->eims_value;
1003 	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1004 	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1005 	adapter->eims_enable_mask |= rx_ring->eims_value;
1006 	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1007 
1008 	/* set vector for other causes, i.e. link changes */
1009 
1010 	tmp = (vector++ | E1000_IVAR_VALID);
1011 
1012 	ew32(IVAR_MISC, tmp);
1013 
1014 	adapter->eims_enable_mask = (1 << (vector)) - 1;
1015 	adapter->eims_other = 1 << (vector - 1);
1016 	e1e_flush();
1017 }
1018 
1019 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1020 {
1021 	if (adapter->msix_entries) {
1022 		pci_disable_msix(adapter->pdev);
1023 		kfree(adapter->msix_entries);
1024 		adapter->msix_entries = NULL;
1025 	}
1026 }
1027 
1028 /**
1029  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1030  *
1031  * Attempt to configure interrupts using the best available
1032  * capabilities of the hardware and kernel.
1033  **/
1034 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1035 {
1036 	int err = -ENOMEM;
1037 	int i;
1038 
1039 	/* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1040 	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1041 	                                GFP_KERNEL);
1042 	if (adapter->msix_entries) {
1043 		for (i = 0; i < 3; i++)
1044 			adapter->msix_entries[i].entry = i;
1045 
1046 		err = pci_enable_msix_range(adapter->pdev,
1047 		                            adapter->msix_entries, 3, 3);
1048 	}
1049 
1050 	if (err < 0) {
1051 		/* MSI-X failed */
1052 		dev_err(&adapter->pdev->dev,
1053 		        "Failed to initialize MSI-X interrupts.\n");
1054 		igbvf_reset_interrupt_capability(adapter);
1055 	}
1056 }
1057 
1058 /**
1059  * igbvf_request_msix - Initialize MSI-X interrupts
1060  *
1061  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1062  * kernel.
1063  **/
1064 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1065 {
1066 	struct net_device *netdev = adapter->netdev;
1067 	int err = 0, vector = 0;
1068 
1069 	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1070 		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1071 		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1072 	} else {
1073 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1074 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1075 	}
1076 
1077 	err = request_irq(adapter->msix_entries[vector].vector,
1078 	                  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1079 	                  netdev);
1080 	if (err)
1081 		goto out;
1082 
1083 	adapter->tx_ring->itr_register = E1000_EITR(vector);
1084 	adapter->tx_ring->itr_val = adapter->current_itr;
1085 	vector++;
1086 
1087 	err = request_irq(adapter->msix_entries[vector].vector,
1088 	                  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1089 	                  netdev);
1090 	if (err)
1091 		goto out;
1092 
1093 	adapter->rx_ring->itr_register = E1000_EITR(vector);
1094 	adapter->rx_ring->itr_val = adapter->current_itr;
1095 	vector++;
1096 
1097 	err = request_irq(adapter->msix_entries[vector].vector,
1098 	                  igbvf_msix_other, 0, netdev->name, netdev);
1099 	if (err)
1100 		goto out;
1101 
1102 	igbvf_configure_msix(adapter);
1103 	return 0;
1104 out:
1105 	return err;
1106 }
1107 
1108 /**
1109  * igbvf_alloc_queues - Allocate memory for all rings
1110  * @adapter: board private structure to initialize
1111  **/
1112 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1113 {
1114 	struct net_device *netdev = adapter->netdev;
1115 
1116 	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1117 	if (!adapter->tx_ring)
1118 		return -ENOMEM;
1119 
1120 	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1121 	if (!adapter->rx_ring) {
1122 		kfree(adapter->tx_ring);
1123 		return -ENOMEM;
1124 	}
1125 
1126 	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1127 
1128 	return 0;
1129 }
1130 
1131 /**
1132  * igbvf_request_irq - initialize interrupts
1133  *
1134  * Attempts to configure interrupts using the best available
1135  * capabilities of the hardware and kernel.
1136  **/
1137 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1138 {
1139 	int err = -1;
1140 
1141 	/* igbvf supports msi-x only */
1142 	if (adapter->msix_entries)
1143 		err = igbvf_request_msix(adapter);
1144 
1145 	if (!err)
1146 		return err;
1147 
1148 	dev_err(&adapter->pdev->dev,
1149 	        "Unable to allocate interrupt, Error: %d\n", err);
1150 
1151 	return err;
1152 }
1153 
1154 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1155 {
1156 	struct net_device *netdev = adapter->netdev;
1157 	int vector;
1158 
1159 	if (adapter->msix_entries) {
1160 		for (vector = 0; vector < 3; vector++)
1161 			free_irq(adapter->msix_entries[vector].vector, netdev);
1162 	}
1163 }
1164 
1165 /**
1166  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1167  **/
1168 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1169 {
1170 	struct e1000_hw *hw = &adapter->hw;
1171 
1172 	ew32(EIMC, ~0);
1173 
1174 	if (adapter->msix_entries)
1175 		ew32(EIAC, 0);
1176 }
1177 
1178 /**
1179  * igbvf_irq_enable - Enable default interrupt generation settings
1180  **/
1181 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1182 {
1183 	struct e1000_hw *hw = &adapter->hw;
1184 
1185 	ew32(EIAC, adapter->eims_enable_mask);
1186 	ew32(EIAM, adapter->eims_enable_mask);
1187 	ew32(EIMS, adapter->eims_enable_mask);
1188 }
1189 
1190 /**
1191  * igbvf_poll - NAPI Rx polling callback
1192  * @napi: struct associated with this polling callback
1193  * @budget: amount of packets driver is allowed to process this poll
1194  **/
1195 static int igbvf_poll(struct napi_struct *napi, int budget)
1196 {
1197 	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1198 	struct igbvf_adapter *adapter = rx_ring->adapter;
1199 	struct e1000_hw *hw = &adapter->hw;
1200 	int work_done = 0;
1201 
1202 	igbvf_clean_rx_irq(adapter, &work_done, budget);
1203 
1204 	/* If not enough Rx work done, exit the polling mode */
1205 	if (work_done < budget) {
1206 		napi_complete(napi);
1207 
1208 		if (adapter->requested_itr & 3)
1209 			igbvf_set_itr(adapter);
1210 
1211 		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1212 			ew32(EIMS, adapter->rx_ring->eims_value);
1213 	}
1214 
1215 	return work_done;
1216 }
1217 
1218 /**
1219  * igbvf_set_rlpml - set receive large packet maximum length
1220  * @adapter: board private structure
1221  *
1222  * Configure the maximum size of packets that will be received
1223  */
1224 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1225 {
1226 	int max_frame_size;
1227 	struct e1000_hw *hw = &adapter->hw;
1228 
1229 	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1230 	e1000_rlpml_set_vf(hw, max_frame_size);
1231 }
1232 
1233 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1234 				 __be16 proto, u16 vid)
1235 {
1236 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1237 	struct e1000_hw *hw = &adapter->hw;
1238 
1239 	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1240 		dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1241 		return -EINVAL;
1242 	}
1243 	set_bit(vid, adapter->active_vlans);
1244 	return 0;
1245 }
1246 
1247 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1248 				  __be16 proto, u16 vid)
1249 {
1250 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1251 	struct e1000_hw *hw = &adapter->hw;
1252 
1253 	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1254 		dev_err(&adapter->pdev->dev,
1255 		        "Failed to remove vlan id %d\n", vid);
1256 		return -EINVAL;
1257 	}
1258 	clear_bit(vid, adapter->active_vlans);
1259 	return 0;
1260 }
1261 
1262 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1263 {
1264 	u16 vid;
1265 
1266 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1267 		igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1268 }
1269 
1270 /**
1271  * igbvf_configure_tx - Configure Transmit Unit after Reset
1272  * @adapter: board private structure
1273  *
1274  * Configure the Tx unit of the MAC after a reset.
1275  **/
1276 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1277 {
1278 	struct e1000_hw *hw = &adapter->hw;
1279 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1280 	u64 tdba;
1281 	u32 txdctl, dca_txctrl;
1282 
1283 	/* disable transmits */
1284 	txdctl = er32(TXDCTL(0));
1285 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1286 	e1e_flush();
1287 	msleep(10);
1288 
1289 	/* Setup the HW Tx Head and Tail descriptor pointers */
1290 	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1291 	tdba = tx_ring->dma;
1292 	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1293 	ew32(TDBAH(0), (tdba >> 32));
1294 	ew32(TDH(0), 0);
1295 	ew32(TDT(0), 0);
1296 	tx_ring->head = E1000_TDH(0);
1297 	tx_ring->tail = E1000_TDT(0);
1298 
1299 	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1300 	 * MUST be delivered in order or it will completely screw up
1301 	 * our bookeeping.
1302 	 */
1303 	dca_txctrl = er32(DCA_TXCTRL(0));
1304 	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1305 	ew32(DCA_TXCTRL(0), dca_txctrl);
1306 
1307 	/* enable transmits */
1308 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1309 	ew32(TXDCTL(0), txdctl);
1310 
1311 	/* Setup Transmit Descriptor Settings for eop descriptor */
1312 	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1313 
1314 	/* enable Report Status bit */
1315 	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1316 }
1317 
1318 /**
1319  * igbvf_setup_srrctl - configure the receive control registers
1320  * @adapter: Board private structure
1321  **/
1322 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1323 {
1324 	struct e1000_hw *hw = &adapter->hw;
1325 	u32 srrctl = 0;
1326 
1327 	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1328 	            E1000_SRRCTL_BSIZEHDR_MASK |
1329 	            E1000_SRRCTL_BSIZEPKT_MASK);
1330 
1331 	/* Enable queue drop to avoid head of line blocking */
1332 	srrctl |= E1000_SRRCTL_DROP_EN;
1333 
1334 	/* Setup buffer sizes */
1335 	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1336 	          E1000_SRRCTL_BSIZEPKT_SHIFT;
1337 
1338 	if (adapter->rx_buffer_len < 2048) {
1339 		adapter->rx_ps_hdr_size = 0;
1340 		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1341 	} else {
1342 		adapter->rx_ps_hdr_size = 128;
1343 		srrctl |= adapter->rx_ps_hdr_size <<
1344 		          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1345 		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1346 	}
1347 
1348 	ew32(SRRCTL(0), srrctl);
1349 }
1350 
1351 /**
1352  * igbvf_configure_rx - Configure Receive Unit after Reset
1353  * @adapter: board private structure
1354  *
1355  * Configure the Rx unit of the MAC after a reset.
1356  **/
1357 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1358 {
1359 	struct e1000_hw *hw = &adapter->hw;
1360 	struct igbvf_ring *rx_ring = adapter->rx_ring;
1361 	u64 rdba;
1362 	u32 rdlen, rxdctl;
1363 
1364 	/* disable receives */
1365 	rxdctl = er32(RXDCTL(0));
1366 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1367 	e1e_flush();
1368 	msleep(10);
1369 
1370 	rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1371 
1372 	/*
1373 	 * Setup the HW Rx Head and Tail Descriptor Pointers and
1374 	 * the Base and Length of the Rx Descriptor Ring
1375 	 */
1376 	rdba = rx_ring->dma;
1377 	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1378 	ew32(RDBAH(0), (rdba >> 32));
1379 	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1380 	rx_ring->head = E1000_RDH(0);
1381 	rx_ring->tail = E1000_RDT(0);
1382 	ew32(RDH(0), 0);
1383 	ew32(RDT(0), 0);
1384 
1385 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1386 	rxdctl &= 0xFFF00000;
1387 	rxdctl |= IGBVF_RX_PTHRESH;
1388 	rxdctl |= IGBVF_RX_HTHRESH << 8;
1389 	rxdctl |= IGBVF_RX_WTHRESH << 16;
1390 
1391 	igbvf_set_rlpml(adapter);
1392 
1393 	/* enable receives */
1394 	ew32(RXDCTL(0), rxdctl);
1395 }
1396 
1397 /**
1398  * igbvf_set_multi - Multicast and Promiscuous mode set
1399  * @netdev: network interface device structure
1400  *
1401  * The set_multi entry point is called whenever the multicast address
1402  * list or the network interface flags are updated.  This routine is
1403  * responsible for configuring the hardware for proper multicast,
1404  * promiscuous mode, and all-multi behavior.
1405  **/
1406 static void igbvf_set_multi(struct net_device *netdev)
1407 {
1408 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1409 	struct e1000_hw *hw = &adapter->hw;
1410 	struct netdev_hw_addr *ha;
1411 	u8  *mta_list = NULL;
1412 	int i;
1413 
1414 	if (!netdev_mc_empty(netdev)) {
1415 		mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1416 					 GFP_ATOMIC);
1417 		if (!mta_list)
1418 			return;
1419 	}
1420 
1421 	/* prepare a packed array of only addresses. */
1422 	i = 0;
1423 	netdev_for_each_mc_addr(ha, netdev)
1424 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1425 
1426 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1427 	kfree(mta_list);
1428 }
1429 
1430 /**
1431  * igbvf_configure - configure the hardware for Rx and Tx
1432  * @adapter: private board structure
1433  **/
1434 static void igbvf_configure(struct igbvf_adapter *adapter)
1435 {
1436 	igbvf_set_multi(adapter->netdev);
1437 
1438 	igbvf_restore_vlan(adapter);
1439 
1440 	igbvf_configure_tx(adapter);
1441 	igbvf_setup_srrctl(adapter);
1442 	igbvf_configure_rx(adapter);
1443 	igbvf_alloc_rx_buffers(adapter->rx_ring,
1444 	                       igbvf_desc_unused(adapter->rx_ring));
1445 }
1446 
1447 /* igbvf_reset - bring the hardware into a known good state
1448  *
1449  * This function boots the hardware and enables some settings that
1450  * require a configuration cycle of the hardware - those cannot be
1451  * set/changed during runtime. After reset the device needs to be
1452  * properly configured for Rx, Tx etc.
1453  */
1454 static void igbvf_reset(struct igbvf_adapter *adapter)
1455 {
1456 	struct e1000_mac_info *mac = &adapter->hw.mac;
1457 	struct net_device *netdev = adapter->netdev;
1458 	struct e1000_hw *hw = &adapter->hw;
1459 
1460 	/* Allow time for pending master requests to run */
1461 	if (mac->ops.reset_hw(hw))
1462 		dev_err(&adapter->pdev->dev, "PF still resetting\n");
1463 
1464 	mac->ops.init_hw(hw);
1465 
1466 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1467 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1468 		       netdev->addr_len);
1469 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1470 		       netdev->addr_len);
1471 	}
1472 
1473 	adapter->last_reset = jiffies;
1474 }
1475 
1476 int igbvf_up(struct igbvf_adapter *adapter)
1477 {
1478 	struct e1000_hw *hw = &adapter->hw;
1479 
1480 	/* hardware has been reset, we need to reload some things */
1481 	igbvf_configure(adapter);
1482 
1483 	clear_bit(__IGBVF_DOWN, &adapter->state);
1484 
1485 	napi_enable(&adapter->rx_ring->napi);
1486 	if (adapter->msix_entries)
1487 		igbvf_configure_msix(adapter);
1488 
1489 	/* Clear any pending interrupts. */
1490 	er32(EICR);
1491 	igbvf_irq_enable(adapter);
1492 
1493 	/* start the watchdog */
1494 	hw->mac.get_link_status = 1;
1495 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1496 
1497 
1498 	return 0;
1499 }
1500 
1501 void igbvf_down(struct igbvf_adapter *adapter)
1502 {
1503 	struct net_device *netdev = adapter->netdev;
1504 	struct e1000_hw *hw = &adapter->hw;
1505 	u32 rxdctl, txdctl;
1506 
1507 	/*
1508 	 * signal that we're down so the interrupt handler does not
1509 	 * reschedule our watchdog timer
1510 	 */
1511 	set_bit(__IGBVF_DOWN, &adapter->state);
1512 
1513 	/* disable receives in the hardware */
1514 	rxdctl = er32(RXDCTL(0));
1515 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1516 
1517 	netif_stop_queue(netdev);
1518 
1519 	/* disable transmits in the hardware */
1520 	txdctl = er32(TXDCTL(0));
1521 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1522 
1523 	/* flush both disables and wait for them to finish */
1524 	e1e_flush();
1525 	msleep(10);
1526 
1527 	napi_disable(&adapter->rx_ring->napi);
1528 
1529 	igbvf_irq_disable(adapter);
1530 
1531 	del_timer_sync(&adapter->watchdog_timer);
1532 
1533 	netif_carrier_off(netdev);
1534 
1535 	/* record the stats before reset*/
1536 	igbvf_update_stats(adapter);
1537 
1538 	adapter->link_speed = 0;
1539 	adapter->link_duplex = 0;
1540 
1541 	igbvf_reset(adapter);
1542 	igbvf_clean_tx_ring(adapter->tx_ring);
1543 	igbvf_clean_rx_ring(adapter->rx_ring);
1544 }
1545 
1546 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1547 {
1548 	might_sleep();
1549 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1550 		msleep(1);
1551 	igbvf_down(adapter);
1552 	igbvf_up(adapter);
1553 	clear_bit(__IGBVF_RESETTING, &adapter->state);
1554 }
1555 
1556 /**
1557  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1558  * @adapter: board private structure to initialize
1559  *
1560  * igbvf_sw_init initializes the Adapter private data structure.
1561  * Fields are initialized based on PCI device information and
1562  * OS network device settings (MTU size).
1563  **/
1564 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1565 {
1566 	struct net_device *netdev = adapter->netdev;
1567 	s32 rc;
1568 
1569 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1570 	adapter->rx_ps_hdr_size = 0;
1571 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1572 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1573 
1574 	adapter->tx_int_delay = 8;
1575 	adapter->tx_abs_int_delay = 32;
1576 	adapter->rx_int_delay = 0;
1577 	adapter->rx_abs_int_delay = 8;
1578 	adapter->requested_itr = 3;
1579 	adapter->current_itr = IGBVF_START_ITR;
1580 
1581 	/* Set various function pointers */
1582 	adapter->ei->init_ops(&adapter->hw);
1583 
1584 	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1585 	if (rc)
1586 		return rc;
1587 
1588 	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1589 	if (rc)
1590 		return rc;
1591 
1592 	igbvf_set_interrupt_capability(adapter);
1593 
1594 	if (igbvf_alloc_queues(adapter))
1595 		return -ENOMEM;
1596 
1597 	spin_lock_init(&adapter->tx_queue_lock);
1598 
1599 	/* Explicitly disable IRQ since the NIC can be in any state. */
1600 	igbvf_irq_disable(adapter);
1601 
1602 	spin_lock_init(&adapter->stats_lock);
1603 
1604 	set_bit(__IGBVF_DOWN, &adapter->state);
1605 	return 0;
1606 }
1607 
1608 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1609 {
1610 	struct e1000_hw *hw = &adapter->hw;
1611 
1612 	adapter->stats.last_gprc = er32(VFGPRC);
1613 	adapter->stats.last_gorc = er32(VFGORC);
1614 	adapter->stats.last_gptc = er32(VFGPTC);
1615 	adapter->stats.last_gotc = er32(VFGOTC);
1616 	adapter->stats.last_mprc = er32(VFMPRC);
1617 	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1618 	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1619 	adapter->stats.last_gorlbc = er32(VFGORLBC);
1620 	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1621 
1622 	adapter->stats.base_gprc = er32(VFGPRC);
1623 	adapter->stats.base_gorc = er32(VFGORC);
1624 	adapter->stats.base_gptc = er32(VFGPTC);
1625 	adapter->stats.base_gotc = er32(VFGOTC);
1626 	adapter->stats.base_mprc = er32(VFMPRC);
1627 	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1628 	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1629 	adapter->stats.base_gorlbc = er32(VFGORLBC);
1630 	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1631 }
1632 
1633 /**
1634  * igbvf_open - Called when a network interface is made active
1635  * @netdev: network interface device structure
1636  *
1637  * Returns 0 on success, negative value on failure
1638  *
1639  * The open entry point is called when a network interface is made
1640  * active by the system (IFF_UP).  At this point all resources needed
1641  * for transmit and receive operations are allocated, the interrupt
1642  * handler is registered with the OS, the watchdog timer is started,
1643  * and the stack is notified that the interface is ready.
1644  **/
1645 static int igbvf_open(struct net_device *netdev)
1646 {
1647 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1648 	struct e1000_hw *hw = &adapter->hw;
1649 	int err;
1650 
1651 	/* disallow open during test */
1652 	if (test_bit(__IGBVF_TESTING, &adapter->state))
1653 		return -EBUSY;
1654 
1655 	/* allocate transmit descriptors */
1656 	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1657 	if (err)
1658 		goto err_setup_tx;
1659 
1660 	/* allocate receive descriptors */
1661 	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1662 	if (err)
1663 		goto err_setup_rx;
1664 
1665 	/*
1666 	 * before we allocate an interrupt, we must be ready to handle it.
1667 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1668 	 * as soon as we call pci_request_irq, so we have to setup our
1669 	 * clean_rx handler before we do so.
1670 	 */
1671 	igbvf_configure(adapter);
1672 
1673 	err = igbvf_request_irq(adapter);
1674 	if (err)
1675 		goto err_req_irq;
1676 
1677 	/* From here on the code is the same as igbvf_up() */
1678 	clear_bit(__IGBVF_DOWN, &adapter->state);
1679 
1680 	napi_enable(&adapter->rx_ring->napi);
1681 
1682 	/* clear any pending interrupts */
1683 	er32(EICR);
1684 
1685 	igbvf_irq_enable(adapter);
1686 
1687 	/* start the watchdog */
1688 	hw->mac.get_link_status = 1;
1689 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1690 
1691 	return 0;
1692 
1693 err_req_irq:
1694 	igbvf_free_rx_resources(adapter->rx_ring);
1695 err_setup_rx:
1696 	igbvf_free_tx_resources(adapter->tx_ring);
1697 err_setup_tx:
1698 	igbvf_reset(adapter);
1699 
1700 	return err;
1701 }
1702 
1703 /**
1704  * igbvf_close - Disables a network interface
1705  * @netdev: network interface device structure
1706  *
1707  * Returns 0, this is not allowed to fail
1708  *
1709  * The close entry point is called when an interface is de-activated
1710  * by the OS.  The hardware is still under the drivers control, but
1711  * needs to be disabled.  A global MAC reset is issued to stop the
1712  * hardware, and all transmit and receive resources are freed.
1713  **/
1714 static int igbvf_close(struct net_device *netdev)
1715 {
1716 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1717 
1718 	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1719 	igbvf_down(adapter);
1720 
1721 	igbvf_free_irq(adapter);
1722 
1723 	igbvf_free_tx_resources(adapter->tx_ring);
1724 	igbvf_free_rx_resources(adapter->rx_ring);
1725 
1726 	return 0;
1727 }
1728 /**
1729  * igbvf_set_mac - Change the Ethernet Address of the NIC
1730  * @netdev: network interface device structure
1731  * @p: pointer to an address structure
1732  *
1733  * Returns 0 on success, negative on failure
1734  **/
1735 static int igbvf_set_mac(struct net_device *netdev, void *p)
1736 {
1737 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1738 	struct e1000_hw *hw = &adapter->hw;
1739 	struct sockaddr *addr = p;
1740 
1741 	if (!is_valid_ether_addr(addr->sa_data))
1742 		return -EADDRNOTAVAIL;
1743 
1744 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1745 
1746 	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1747 
1748 	if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1749 		return -EADDRNOTAVAIL;
1750 
1751 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1752 
1753 	return 0;
1754 }
1755 
1756 #define UPDATE_VF_COUNTER(reg, name)                                    \
1757 	{                                                               \
1758 		u32 current_counter = er32(reg);                        \
1759 		if (current_counter < adapter->stats.last_##name)       \
1760 			adapter->stats.name += 0x100000000LL;           \
1761 		adapter->stats.last_##name = current_counter;           \
1762 		adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1763 		adapter->stats.name |= current_counter;                 \
1764 	}
1765 
1766 /**
1767  * igbvf_update_stats - Update the board statistics counters
1768  * @adapter: board private structure
1769 **/
1770 void igbvf_update_stats(struct igbvf_adapter *adapter)
1771 {
1772 	struct e1000_hw *hw = &adapter->hw;
1773 	struct pci_dev *pdev = adapter->pdev;
1774 
1775 	/*
1776 	 * Prevent stats update while adapter is being reset, link is down
1777 	 * or if the pci connection is down.
1778 	 */
1779 	if (adapter->link_speed == 0)
1780 		return;
1781 
1782 	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1783 		return;
1784 
1785 	if (pci_channel_offline(pdev))
1786 		return;
1787 
1788 	UPDATE_VF_COUNTER(VFGPRC, gprc);
1789 	UPDATE_VF_COUNTER(VFGORC, gorc);
1790 	UPDATE_VF_COUNTER(VFGPTC, gptc);
1791 	UPDATE_VF_COUNTER(VFGOTC, gotc);
1792 	UPDATE_VF_COUNTER(VFMPRC, mprc);
1793 	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1794 	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1795 	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1796 	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1797 
1798 	/* Fill out the OS statistics structure */
1799 	adapter->net_stats.multicast = adapter->stats.mprc;
1800 }
1801 
1802 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1803 {
1804 	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1805 		 adapter->link_speed,
1806 		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1807 }
1808 
1809 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1810 {
1811 	struct e1000_hw *hw = &adapter->hw;
1812 	s32 ret_val = E1000_SUCCESS;
1813 	bool link_active;
1814 
1815 	/* If interface is down, stay link down */
1816 	if (test_bit(__IGBVF_DOWN, &adapter->state))
1817 		return false;
1818 
1819 	ret_val = hw->mac.ops.check_for_link(hw);
1820 	link_active = !hw->mac.get_link_status;
1821 
1822 	/* if check for link returns error we will need to reset */
1823 	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1824 		schedule_work(&adapter->reset_task);
1825 
1826 	return link_active;
1827 }
1828 
1829 /**
1830  * igbvf_watchdog - Timer Call-back
1831  * @data: pointer to adapter cast into an unsigned long
1832  **/
1833 static void igbvf_watchdog(unsigned long data)
1834 {
1835 	struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1836 
1837 	/* Do the rest outside of interrupt context */
1838 	schedule_work(&adapter->watchdog_task);
1839 }
1840 
1841 static void igbvf_watchdog_task(struct work_struct *work)
1842 {
1843 	struct igbvf_adapter *adapter = container_of(work,
1844 	                                             struct igbvf_adapter,
1845 	                                             watchdog_task);
1846 	struct net_device *netdev = adapter->netdev;
1847 	struct e1000_mac_info *mac = &adapter->hw.mac;
1848 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1849 	struct e1000_hw *hw = &adapter->hw;
1850 	u32 link;
1851 	int tx_pending = 0;
1852 
1853 	link = igbvf_has_link(adapter);
1854 
1855 	if (link) {
1856 		if (!netif_carrier_ok(netdev)) {
1857 			mac->ops.get_link_up_info(&adapter->hw,
1858 			                          &adapter->link_speed,
1859 			                          &adapter->link_duplex);
1860 			igbvf_print_link_info(adapter);
1861 
1862 			netif_carrier_on(netdev);
1863 			netif_wake_queue(netdev);
1864 		}
1865 	} else {
1866 		if (netif_carrier_ok(netdev)) {
1867 			adapter->link_speed = 0;
1868 			adapter->link_duplex = 0;
1869 			dev_info(&adapter->pdev->dev, "Link is Down\n");
1870 			netif_carrier_off(netdev);
1871 			netif_stop_queue(netdev);
1872 		}
1873 	}
1874 
1875 	if (netif_carrier_ok(netdev)) {
1876 		igbvf_update_stats(adapter);
1877 	} else {
1878 		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1879 		              tx_ring->count);
1880 		if (tx_pending) {
1881 			/*
1882 			 * We've lost link, so the controller stops DMA,
1883 			 * but we've got queued Tx work that's never going
1884 			 * to get done, so reset controller to flush Tx.
1885 			 * (Do the reset outside of interrupt context).
1886 			 */
1887 			adapter->tx_timeout_count++;
1888 			schedule_work(&adapter->reset_task);
1889 		}
1890 	}
1891 
1892 	/* Cause software interrupt to ensure Rx ring is cleaned */
1893 	ew32(EICS, adapter->rx_ring->eims_value);
1894 
1895 	/* Reset the timer */
1896 	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1897 		mod_timer(&adapter->watchdog_timer,
1898 			  round_jiffies(jiffies + (2 * HZ)));
1899 }
1900 
1901 #define IGBVF_TX_FLAGS_CSUM             0x00000001
1902 #define IGBVF_TX_FLAGS_VLAN             0x00000002
1903 #define IGBVF_TX_FLAGS_TSO              0x00000004
1904 #define IGBVF_TX_FLAGS_IPV4             0x00000008
1905 #define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1906 #define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1907 
1908 static int igbvf_tso(struct igbvf_adapter *adapter,
1909                      struct igbvf_ring *tx_ring,
1910                      struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1911 {
1912 	struct e1000_adv_tx_context_desc *context_desc;
1913 	struct igbvf_buffer *buffer_info;
1914 	u32 info = 0, tu_cmd = 0;
1915 	u32 mss_l4len_idx, l4len;
1916 	unsigned int i;
1917 	int err;
1918 
1919 	*hdr_len = 0;
1920 
1921 	err = skb_cow_head(skb, 0);
1922 	if (err < 0) {
1923 		dev_err(&adapter->pdev->dev, "igbvf_tso returning an error\n");
1924 		return err;
1925 	}
1926 
1927 	l4len = tcp_hdrlen(skb);
1928 	*hdr_len += l4len;
1929 
1930 	if (skb->protocol == htons(ETH_P_IP)) {
1931 		struct iphdr *iph = ip_hdr(skb);
1932 		iph->tot_len = 0;
1933 		iph->check = 0;
1934 		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1935 		                                         iph->daddr, 0,
1936 		                                         IPPROTO_TCP,
1937 		                                         0);
1938 	} else if (skb_is_gso_v6(skb)) {
1939 		ipv6_hdr(skb)->payload_len = 0;
1940 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1941 		                                       &ipv6_hdr(skb)->daddr,
1942 		                                       0, IPPROTO_TCP, 0);
1943 	}
1944 
1945 	i = tx_ring->next_to_use;
1946 
1947 	buffer_info = &tx_ring->buffer_info[i];
1948 	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1949 	/* VLAN MACLEN IPLEN */
1950 	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1951 		info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1952 	info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1953 	*hdr_len += skb_network_offset(skb);
1954 	info |= (skb_transport_header(skb) - skb_network_header(skb));
1955 	*hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1956 	context_desc->vlan_macip_lens = cpu_to_le32(info);
1957 
1958 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1959 	tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1960 
1961 	if (skb->protocol == htons(ETH_P_IP))
1962 		tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1963 	tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1964 
1965 	context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1966 
1967 	/* MSS L4LEN IDX */
1968 	mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1969 	mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1970 
1971 	context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1972 	context_desc->seqnum_seed = 0;
1973 
1974 	buffer_info->time_stamp = jiffies;
1975 	buffer_info->dma = 0;
1976 	i++;
1977 	if (i == tx_ring->count)
1978 		i = 0;
1979 
1980 	tx_ring->next_to_use = i;
1981 
1982 	return true;
1983 }
1984 
1985 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1986                                  struct igbvf_ring *tx_ring,
1987                                  struct sk_buff *skb, u32 tx_flags)
1988 {
1989 	struct e1000_adv_tx_context_desc *context_desc;
1990 	unsigned int i;
1991 	struct igbvf_buffer *buffer_info;
1992 	u32 info = 0, tu_cmd = 0;
1993 
1994 	if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1995 	    (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1996 		i = tx_ring->next_to_use;
1997 		buffer_info = &tx_ring->buffer_info[i];
1998 		context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1999 
2000 		if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2001 			info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2002 
2003 		info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2004 		if (skb->ip_summed == CHECKSUM_PARTIAL)
2005 			info |= (skb_transport_header(skb) -
2006 			         skb_network_header(skb));
2007 
2008 
2009 		context_desc->vlan_macip_lens = cpu_to_le32(info);
2010 
2011 		tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2012 
2013 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
2014 			switch (skb->protocol) {
2015 			case htons(ETH_P_IP):
2016 				tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2017 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2018 					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2019 				break;
2020 			case htons(ETH_P_IPV6):
2021 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2022 					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2023 				break;
2024 			default:
2025 				break;
2026 			}
2027 		}
2028 
2029 		context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2030 		context_desc->seqnum_seed = 0;
2031 		context_desc->mss_l4len_idx = 0;
2032 
2033 		buffer_info->time_stamp = jiffies;
2034 		buffer_info->dma = 0;
2035 		i++;
2036 		if (i == tx_ring->count)
2037 			i = 0;
2038 		tx_ring->next_to_use = i;
2039 
2040 		return true;
2041 	}
2042 
2043 	return false;
2044 }
2045 
2046 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2047 {
2048 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2049 
2050 	/* there is enough descriptors then we don't need to worry  */
2051 	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2052 		return 0;
2053 
2054 	netif_stop_queue(netdev);
2055 
2056 	smp_mb();
2057 
2058 	/* We need to check again just in case room has been made available */
2059 	if (igbvf_desc_unused(adapter->tx_ring) < size)
2060 		return -EBUSY;
2061 
2062 	netif_wake_queue(netdev);
2063 
2064 	++adapter->restart_queue;
2065 	return 0;
2066 }
2067 
2068 #define IGBVF_MAX_TXD_PWR       16
2069 #define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2070 
2071 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2072                                    struct igbvf_ring *tx_ring,
2073 				   struct sk_buff *skb)
2074 {
2075 	struct igbvf_buffer *buffer_info;
2076 	struct pci_dev *pdev = adapter->pdev;
2077 	unsigned int len = skb_headlen(skb);
2078 	unsigned int count = 0, i;
2079 	unsigned int f;
2080 
2081 	i = tx_ring->next_to_use;
2082 
2083 	buffer_info = &tx_ring->buffer_info[i];
2084 	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2085 	buffer_info->length = len;
2086 	/* set time_stamp *before* dma to help avoid a possible race */
2087 	buffer_info->time_stamp = jiffies;
2088 	buffer_info->mapped_as_page = false;
2089 	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2090 					  DMA_TO_DEVICE);
2091 	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2092 		goto dma_error;
2093 
2094 
2095 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2096 		const struct skb_frag_struct *frag;
2097 
2098 		count++;
2099 		i++;
2100 		if (i == tx_ring->count)
2101 			i = 0;
2102 
2103 		frag = &skb_shinfo(skb)->frags[f];
2104 		len = skb_frag_size(frag);
2105 
2106 		buffer_info = &tx_ring->buffer_info[i];
2107 		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2108 		buffer_info->length = len;
2109 		buffer_info->time_stamp = jiffies;
2110 		buffer_info->mapped_as_page = true;
2111 		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2112 						DMA_TO_DEVICE);
2113 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2114 			goto dma_error;
2115 	}
2116 
2117 	tx_ring->buffer_info[i].skb = skb;
2118 
2119 	return ++count;
2120 
2121 dma_error:
2122 	dev_err(&pdev->dev, "TX DMA map failed\n");
2123 
2124 	/* clear timestamp and dma mappings for failed buffer_info mapping */
2125 	buffer_info->dma = 0;
2126 	buffer_info->time_stamp = 0;
2127 	buffer_info->length = 0;
2128 	buffer_info->mapped_as_page = false;
2129 	if (count)
2130 		count--;
2131 
2132 	/* clear timestamp and dma mappings for remaining portion of packet */
2133 	while (count--) {
2134 		if (i==0)
2135 			i += tx_ring->count;
2136 		i--;
2137 		buffer_info = &tx_ring->buffer_info[i];
2138 		igbvf_put_txbuf(adapter, buffer_info);
2139 	}
2140 
2141 	return 0;
2142 }
2143 
2144 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2145                                       struct igbvf_ring *tx_ring,
2146 				      int tx_flags, int count,
2147 				      unsigned int first, u32 paylen,
2148                                       u8 hdr_len)
2149 {
2150 	union e1000_adv_tx_desc *tx_desc = NULL;
2151 	struct igbvf_buffer *buffer_info;
2152 	u32 olinfo_status = 0, cmd_type_len;
2153 	unsigned int i;
2154 
2155 	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2156 	                E1000_ADVTXD_DCMD_DEXT);
2157 
2158 	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2159 		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2160 
2161 	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2162 		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2163 
2164 		/* insert tcp checksum */
2165 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2166 
2167 		/* insert ip checksum */
2168 		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2169 			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2170 
2171 	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2172 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2173 	}
2174 
2175 	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2176 
2177 	i = tx_ring->next_to_use;
2178 	while (count--) {
2179 		buffer_info = &tx_ring->buffer_info[i];
2180 		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2181 		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2182 		tx_desc->read.cmd_type_len =
2183 		         cpu_to_le32(cmd_type_len | buffer_info->length);
2184 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2185 		i++;
2186 		if (i == tx_ring->count)
2187 			i = 0;
2188 	}
2189 
2190 	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2191 	/* Force memory writes to complete before letting h/w
2192 	 * know there are new descriptors to fetch.  (Only
2193 	 * applicable for weak-ordered memory model archs,
2194 	 * such as IA-64). */
2195 	wmb();
2196 
2197 	tx_ring->buffer_info[first].next_to_watch = tx_desc;
2198 	tx_ring->next_to_use = i;
2199 	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2200 	/* we need this if more than one processor can write to our tail
2201 	 * at a time, it syncronizes IO on IA64/Altix systems */
2202 	mmiowb();
2203 }
2204 
2205 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2206 					     struct net_device *netdev,
2207 					     struct igbvf_ring *tx_ring)
2208 {
2209 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2210 	unsigned int first, tx_flags = 0;
2211 	u8 hdr_len = 0;
2212 	int count = 0;
2213 	int tso = 0;
2214 
2215 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2216 		dev_kfree_skb_any(skb);
2217 		return NETDEV_TX_OK;
2218 	}
2219 
2220 	if (skb->len <= 0) {
2221 		dev_kfree_skb_any(skb);
2222 		return NETDEV_TX_OK;
2223 	}
2224 
2225 	/*
2226 	 * need: count + 4 desc gap to keep tail from touching
2227          *       + 2 desc gap to keep tail from touching head,
2228          *       + 1 desc for skb->data,
2229          *       + 1 desc for context descriptor,
2230 	 * head, otherwise try next time
2231 	 */
2232 	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2233 		/* this is a hard error */
2234 		return NETDEV_TX_BUSY;
2235 	}
2236 
2237 	if (vlan_tx_tag_present(skb)) {
2238 		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2239 		tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2240 	}
2241 
2242 	if (skb->protocol == htons(ETH_P_IP))
2243 		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2244 
2245 	first = tx_ring->next_to_use;
2246 
2247 	tso = skb_is_gso(skb) ?
2248 		igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2249 	if (unlikely(tso < 0)) {
2250 		dev_kfree_skb_any(skb);
2251 		return NETDEV_TX_OK;
2252 	}
2253 
2254 	if (tso)
2255 		tx_flags |= IGBVF_TX_FLAGS_TSO;
2256 	else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2257 	         (skb->ip_summed == CHECKSUM_PARTIAL))
2258 		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2259 
2260 	/*
2261 	 * count reflects descriptors mapped, if 0 then mapping error
2262 	 * has occurred and we need to rewind the descriptor queue
2263 	 */
2264 	count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2265 
2266 	if (count) {
2267 		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2268 				   first, skb->len, hdr_len);
2269 		/* Make sure there is space in the ring for the next send. */
2270 		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2271 	} else {
2272 		dev_kfree_skb_any(skb);
2273 		tx_ring->buffer_info[first].time_stamp = 0;
2274 		tx_ring->next_to_use = first;
2275 	}
2276 
2277 	return NETDEV_TX_OK;
2278 }
2279 
2280 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2281 				    struct net_device *netdev)
2282 {
2283 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2284 	struct igbvf_ring *tx_ring;
2285 
2286 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2287 		dev_kfree_skb_any(skb);
2288 		return NETDEV_TX_OK;
2289 	}
2290 
2291 	tx_ring = &adapter->tx_ring[0];
2292 
2293 	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2294 }
2295 
2296 /**
2297  * igbvf_tx_timeout - Respond to a Tx Hang
2298  * @netdev: network interface device structure
2299  **/
2300 static void igbvf_tx_timeout(struct net_device *netdev)
2301 {
2302 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2303 
2304 	/* Do the reset outside of interrupt context */
2305 	adapter->tx_timeout_count++;
2306 	schedule_work(&adapter->reset_task);
2307 }
2308 
2309 static void igbvf_reset_task(struct work_struct *work)
2310 {
2311 	struct igbvf_adapter *adapter;
2312 	adapter = container_of(work, struct igbvf_adapter, reset_task);
2313 
2314 	igbvf_reinit_locked(adapter);
2315 }
2316 
2317 /**
2318  * igbvf_get_stats - Get System Network Statistics
2319  * @netdev: network interface device structure
2320  *
2321  * Returns the address of the device statistics structure.
2322  * The statistics are actually updated from the timer callback.
2323  **/
2324 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2325 {
2326 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2327 
2328 	/* only return the current stats */
2329 	return &adapter->net_stats;
2330 }
2331 
2332 /**
2333  * igbvf_change_mtu - Change the Maximum Transfer Unit
2334  * @netdev: network interface device structure
2335  * @new_mtu: new value for maximum frame size
2336  *
2337  * Returns 0 on success, negative on failure
2338  **/
2339 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2340 {
2341 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2342 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2343 
2344 	if (new_mtu < 68 || new_mtu > INT_MAX - ETH_HLEN - ETH_FCS_LEN ||
2345 	    max_frame > MAX_JUMBO_FRAME_SIZE)
2346 		return -EINVAL;
2347 
2348 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2349 	if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2350 		dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2351 		return -EINVAL;
2352 	}
2353 
2354 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2355 		msleep(1);
2356 	/* igbvf_down has a dependency on max_frame_size */
2357 	adapter->max_frame_size = max_frame;
2358 	if (netif_running(netdev))
2359 		igbvf_down(adapter);
2360 
2361 	/*
2362 	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2363 	 * means we reserve 2 more, this pushes us to allocate from the next
2364 	 * larger slab size.
2365 	 * i.e. RXBUFFER_2048 --> size-4096 slab
2366 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2367 	 * fragmented skbs
2368 	 */
2369 
2370 	if (max_frame <= 1024)
2371 		adapter->rx_buffer_len = 1024;
2372 	else if (max_frame <= 2048)
2373 		adapter->rx_buffer_len = 2048;
2374 	else
2375 #if (PAGE_SIZE / 2) > 16384
2376 		adapter->rx_buffer_len = 16384;
2377 #else
2378 		adapter->rx_buffer_len = PAGE_SIZE / 2;
2379 #endif
2380 
2381 
2382 	/* adjust allocation if LPE protects us, and we aren't using SBP */
2383 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2384 	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2385 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2386 		                         ETH_FCS_LEN;
2387 
2388 	dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2389 	         netdev->mtu, new_mtu);
2390 	netdev->mtu = new_mtu;
2391 
2392 	if (netif_running(netdev))
2393 		igbvf_up(adapter);
2394 	else
2395 		igbvf_reset(adapter);
2396 
2397 	clear_bit(__IGBVF_RESETTING, &adapter->state);
2398 
2399 	return 0;
2400 }
2401 
2402 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2403 {
2404 	switch (cmd) {
2405 	default:
2406 		return -EOPNOTSUPP;
2407 	}
2408 }
2409 
2410 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2411 {
2412 	struct net_device *netdev = pci_get_drvdata(pdev);
2413 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2414 #ifdef CONFIG_PM
2415 	int retval = 0;
2416 #endif
2417 
2418 	netif_device_detach(netdev);
2419 
2420 	if (netif_running(netdev)) {
2421 		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2422 		igbvf_down(adapter);
2423 		igbvf_free_irq(adapter);
2424 	}
2425 
2426 #ifdef CONFIG_PM
2427 	retval = pci_save_state(pdev);
2428 	if (retval)
2429 		return retval;
2430 #endif
2431 
2432 	pci_disable_device(pdev);
2433 
2434 	return 0;
2435 }
2436 
2437 #ifdef CONFIG_PM
2438 static int igbvf_resume(struct pci_dev *pdev)
2439 {
2440 	struct net_device *netdev = pci_get_drvdata(pdev);
2441 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2442 	u32 err;
2443 
2444 	pci_restore_state(pdev);
2445 	err = pci_enable_device_mem(pdev);
2446 	if (err) {
2447 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2448 		return err;
2449 	}
2450 
2451 	pci_set_master(pdev);
2452 
2453 	if (netif_running(netdev)) {
2454 		err = igbvf_request_irq(adapter);
2455 		if (err)
2456 			return err;
2457 	}
2458 
2459 	igbvf_reset(adapter);
2460 
2461 	if (netif_running(netdev))
2462 		igbvf_up(adapter);
2463 
2464 	netif_device_attach(netdev);
2465 
2466 	return 0;
2467 }
2468 #endif
2469 
2470 static void igbvf_shutdown(struct pci_dev *pdev)
2471 {
2472 	igbvf_suspend(pdev, PMSG_SUSPEND);
2473 }
2474 
2475 #ifdef CONFIG_NET_POLL_CONTROLLER
2476 /*
2477  * Polling 'interrupt' - used by things like netconsole to send skbs
2478  * without having to re-enable interrupts. It's not called while
2479  * the interrupt routine is executing.
2480  */
2481 static void igbvf_netpoll(struct net_device *netdev)
2482 {
2483 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2484 
2485 	disable_irq(adapter->pdev->irq);
2486 
2487 	igbvf_clean_tx_irq(adapter->tx_ring);
2488 
2489 	enable_irq(adapter->pdev->irq);
2490 }
2491 #endif
2492 
2493 /**
2494  * igbvf_io_error_detected - called when PCI error is detected
2495  * @pdev: Pointer to PCI device
2496  * @state: The current pci connection state
2497  *
2498  * This function is called after a PCI bus error affecting
2499  * this device has been detected.
2500  */
2501 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2502                                                 pci_channel_state_t state)
2503 {
2504 	struct net_device *netdev = pci_get_drvdata(pdev);
2505 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2506 
2507 	netif_device_detach(netdev);
2508 
2509 	if (state == pci_channel_io_perm_failure)
2510 		return PCI_ERS_RESULT_DISCONNECT;
2511 
2512 	if (netif_running(netdev))
2513 		igbvf_down(adapter);
2514 	pci_disable_device(pdev);
2515 
2516 	/* Request a slot slot reset. */
2517 	return PCI_ERS_RESULT_NEED_RESET;
2518 }
2519 
2520 /**
2521  * igbvf_io_slot_reset - called after the pci bus has been reset.
2522  * @pdev: Pointer to PCI device
2523  *
2524  * Restart the card from scratch, as if from a cold-boot. Implementation
2525  * resembles the first-half of the igbvf_resume routine.
2526  */
2527 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2528 {
2529 	struct net_device *netdev = pci_get_drvdata(pdev);
2530 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2531 
2532 	if (pci_enable_device_mem(pdev)) {
2533 		dev_err(&pdev->dev,
2534 			"Cannot re-enable PCI device after reset.\n");
2535 		return PCI_ERS_RESULT_DISCONNECT;
2536 	}
2537 	pci_set_master(pdev);
2538 
2539 	igbvf_reset(adapter);
2540 
2541 	return PCI_ERS_RESULT_RECOVERED;
2542 }
2543 
2544 /**
2545  * igbvf_io_resume - called when traffic can start flowing again.
2546  * @pdev: Pointer to PCI device
2547  *
2548  * This callback is called when the error recovery driver tells us that
2549  * its OK to resume normal operation. Implementation resembles the
2550  * second-half of the igbvf_resume routine.
2551  */
2552 static void igbvf_io_resume(struct pci_dev *pdev)
2553 {
2554 	struct net_device *netdev = pci_get_drvdata(pdev);
2555 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2556 
2557 	if (netif_running(netdev)) {
2558 		if (igbvf_up(adapter)) {
2559 			dev_err(&pdev->dev,
2560 				"can't bring device back up after reset\n");
2561 			return;
2562 		}
2563 	}
2564 
2565 	netif_device_attach(netdev);
2566 }
2567 
2568 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2569 {
2570 	struct e1000_hw *hw = &adapter->hw;
2571 	struct net_device *netdev = adapter->netdev;
2572 	struct pci_dev *pdev = adapter->pdev;
2573 
2574 	if (hw->mac.type == e1000_vfadapt_i350)
2575 		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2576 	else
2577 		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2578 	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2579 }
2580 
2581 static int igbvf_set_features(struct net_device *netdev,
2582 	netdev_features_t features)
2583 {
2584 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2585 
2586 	if (features & NETIF_F_RXCSUM)
2587 		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2588 	else
2589 		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2590 
2591 	return 0;
2592 }
2593 
2594 static const struct net_device_ops igbvf_netdev_ops = {
2595 	.ndo_open                       = igbvf_open,
2596 	.ndo_stop                       = igbvf_close,
2597 	.ndo_start_xmit                 = igbvf_xmit_frame,
2598 	.ndo_get_stats                  = igbvf_get_stats,
2599 	.ndo_set_rx_mode		= igbvf_set_multi,
2600 	.ndo_set_mac_address            = igbvf_set_mac,
2601 	.ndo_change_mtu                 = igbvf_change_mtu,
2602 	.ndo_do_ioctl                   = igbvf_ioctl,
2603 	.ndo_tx_timeout                 = igbvf_tx_timeout,
2604 	.ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2605 	.ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2606 #ifdef CONFIG_NET_POLL_CONTROLLER
2607 	.ndo_poll_controller            = igbvf_netpoll,
2608 #endif
2609 	.ndo_set_features               = igbvf_set_features,
2610 };
2611 
2612 /**
2613  * igbvf_probe - Device Initialization Routine
2614  * @pdev: PCI device information struct
2615  * @ent: entry in igbvf_pci_tbl
2616  *
2617  * Returns 0 on success, negative on failure
2618  *
2619  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2620  * The OS initialization, configuring of the adapter private structure,
2621  * and a hardware reset occur.
2622  **/
2623 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2624 {
2625 	struct net_device *netdev;
2626 	struct igbvf_adapter *adapter;
2627 	struct e1000_hw *hw;
2628 	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2629 
2630 	static int cards_found;
2631 	int err, pci_using_dac;
2632 
2633 	err = pci_enable_device_mem(pdev);
2634 	if (err)
2635 		return err;
2636 
2637 	pci_using_dac = 0;
2638 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2639 	if (!err) {
2640 		pci_using_dac = 1;
2641 	} else {
2642 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2643 		if (err) {
2644 			dev_err(&pdev->dev, "No usable DMA "
2645 			        "configuration, aborting\n");
2646 			goto err_dma;
2647 		}
2648 	}
2649 
2650 	err = pci_request_regions(pdev, igbvf_driver_name);
2651 	if (err)
2652 		goto err_pci_reg;
2653 
2654 	pci_set_master(pdev);
2655 
2656 	err = -ENOMEM;
2657 	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2658 	if (!netdev)
2659 		goto err_alloc_etherdev;
2660 
2661 	SET_NETDEV_DEV(netdev, &pdev->dev);
2662 
2663 	pci_set_drvdata(pdev, netdev);
2664 	adapter = netdev_priv(netdev);
2665 	hw = &adapter->hw;
2666 	adapter->netdev = netdev;
2667 	adapter->pdev = pdev;
2668 	adapter->ei = ei;
2669 	adapter->pba = ei->pba;
2670 	adapter->flags = ei->flags;
2671 	adapter->hw.back = adapter;
2672 	adapter->hw.mac.type = ei->mac;
2673 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2674 
2675 	/* PCI config space info */
2676 
2677 	hw->vendor_id = pdev->vendor;
2678 	hw->device_id = pdev->device;
2679 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2680 	hw->subsystem_device_id = pdev->subsystem_device;
2681 	hw->revision_id = pdev->revision;
2682 
2683 	err = -EIO;
2684 	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2685 	                              pci_resource_len(pdev, 0));
2686 
2687 	if (!adapter->hw.hw_addr)
2688 		goto err_ioremap;
2689 
2690 	if (ei->get_variants) {
2691 		err = ei->get_variants(adapter);
2692 		if (err)
2693 			goto err_get_variants;
2694 	}
2695 
2696 	/* setup adapter struct */
2697 	err = igbvf_sw_init(adapter);
2698 	if (err)
2699 		goto err_sw_init;
2700 
2701 	/* construct the net_device struct */
2702 	netdev->netdev_ops = &igbvf_netdev_ops;
2703 
2704 	igbvf_set_ethtool_ops(netdev);
2705 	netdev->watchdog_timeo = 5 * HZ;
2706 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2707 
2708 	adapter->bd_number = cards_found++;
2709 
2710 	netdev->hw_features = NETIF_F_SG |
2711 	                   NETIF_F_IP_CSUM |
2712 			   NETIF_F_IPV6_CSUM |
2713 			   NETIF_F_TSO |
2714 			   NETIF_F_TSO6 |
2715 			   NETIF_F_RXCSUM;
2716 
2717 	netdev->features = netdev->hw_features |
2718 	                   NETIF_F_HW_VLAN_CTAG_TX |
2719 	                   NETIF_F_HW_VLAN_CTAG_RX |
2720 	                   NETIF_F_HW_VLAN_CTAG_FILTER;
2721 
2722 	if (pci_using_dac)
2723 		netdev->features |= NETIF_F_HIGHDMA;
2724 
2725 	netdev->vlan_features |= NETIF_F_TSO;
2726 	netdev->vlan_features |= NETIF_F_TSO6;
2727 	netdev->vlan_features |= NETIF_F_IP_CSUM;
2728 	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2729 	netdev->vlan_features |= NETIF_F_SG;
2730 
2731 	/*reset the controller to put the device in a known good state */
2732 	err = hw->mac.ops.reset_hw(hw);
2733 	if (err) {
2734 		dev_info(&pdev->dev,
2735 			 "PF still in reset state. Is the PF interface up?\n");
2736 	} else {
2737 		err = hw->mac.ops.read_mac_addr(hw);
2738 		if (err)
2739 			dev_info(&pdev->dev, "Error reading MAC address.\n");
2740 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2741 			dev_info(&pdev->dev, "MAC address not assigned by administrator.\n");
2742 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2743 		       netdev->addr_len);
2744 	}
2745 
2746 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2747 		dev_info(&pdev->dev, "Assigning random MAC address.\n");
2748 		eth_hw_addr_random(netdev);
2749 		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2750 			netdev->addr_len);
2751 	}
2752 
2753 	setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2754 	            (unsigned long) adapter);
2755 
2756 	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2757 	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2758 
2759 	/* ring size defaults */
2760 	adapter->rx_ring->count = 1024;
2761 	adapter->tx_ring->count = 1024;
2762 
2763 	/* reset the hardware with the new settings */
2764 	igbvf_reset(adapter);
2765 
2766 	/* set hardware-specific flags */
2767 	if (adapter->hw.mac.type == e1000_vfadapt_i350)
2768 		adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2769 
2770 	strcpy(netdev->name, "eth%d");
2771 	err = register_netdev(netdev);
2772 	if (err)
2773 		goto err_hw_init;
2774 
2775 	/* tell the stack to leave us alone until igbvf_open() is called */
2776 	netif_carrier_off(netdev);
2777 	netif_stop_queue(netdev);
2778 
2779 	igbvf_print_device_info(adapter);
2780 
2781 	igbvf_initialize_last_counter_stats(adapter);
2782 
2783 	return 0;
2784 
2785 err_hw_init:
2786 	kfree(adapter->tx_ring);
2787 	kfree(adapter->rx_ring);
2788 err_sw_init:
2789 	igbvf_reset_interrupt_capability(adapter);
2790 err_get_variants:
2791 	iounmap(adapter->hw.hw_addr);
2792 err_ioremap:
2793 	free_netdev(netdev);
2794 err_alloc_etherdev:
2795 	pci_release_regions(pdev);
2796 err_pci_reg:
2797 err_dma:
2798 	pci_disable_device(pdev);
2799 	return err;
2800 }
2801 
2802 /**
2803  * igbvf_remove - Device Removal Routine
2804  * @pdev: PCI device information struct
2805  *
2806  * igbvf_remove is called by the PCI subsystem to alert the driver
2807  * that it should release a PCI device.  The could be caused by a
2808  * Hot-Plug event, or because the driver is going to be removed from
2809  * memory.
2810  **/
2811 static void igbvf_remove(struct pci_dev *pdev)
2812 {
2813 	struct net_device *netdev = pci_get_drvdata(pdev);
2814 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2815 	struct e1000_hw *hw = &adapter->hw;
2816 
2817 	/*
2818 	 * The watchdog timer may be rescheduled, so explicitly
2819 	 * disable it from being rescheduled.
2820 	 */
2821 	set_bit(__IGBVF_DOWN, &adapter->state);
2822 	del_timer_sync(&adapter->watchdog_timer);
2823 
2824 	cancel_work_sync(&adapter->reset_task);
2825 	cancel_work_sync(&adapter->watchdog_task);
2826 
2827 	unregister_netdev(netdev);
2828 
2829 	igbvf_reset_interrupt_capability(adapter);
2830 
2831 	/*
2832 	 * it is important to delete the napi struct prior to freeing the
2833 	 * rx ring so that you do not end up with null pointer refs
2834 	 */
2835 	netif_napi_del(&adapter->rx_ring->napi);
2836 	kfree(adapter->tx_ring);
2837 	kfree(adapter->rx_ring);
2838 
2839 	iounmap(hw->hw_addr);
2840 	if (hw->flash_address)
2841 		iounmap(hw->flash_address);
2842 	pci_release_regions(pdev);
2843 
2844 	free_netdev(netdev);
2845 
2846 	pci_disable_device(pdev);
2847 }
2848 
2849 /* PCI Error Recovery (ERS) */
2850 static const struct pci_error_handlers igbvf_err_handler = {
2851 	.error_detected = igbvf_io_error_detected,
2852 	.slot_reset = igbvf_io_slot_reset,
2853 	.resume = igbvf_io_resume,
2854 };
2855 
2856 static const struct pci_device_id igbvf_pci_tbl[] = {
2857 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2858 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2859 	{ } /* terminate list */
2860 };
2861 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2862 
2863 /* PCI Device API Driver */
2864 static struct pci_driver igbvf_driver = {
2865 	.name     = igbvf_driver_name,
2866 	.id_table = igbvf_pci_tbl,
2867 	.probe    = igbvf_probe,
2868 	.remove   = igbvf_remove,
2869 #ifdef CONFIG_PM
2870 	/* Power Management Hooks */
2871 	.suspend  = igbvf_suspend,
2872 	.resume   = igbvf_resume,
2873 #endif
2874 	.shutdown = igbvf_shutdown,
2875 	.err_handler = &igbvf_err_handler
2876 };
2877 
2878 /**
2879  * igbvf_init_module - Driver Registration Routine
2880  *
2881  * igbvf_init_module is the first routine called when the driver is
2882  * loaded. All it does is register with the PCI subsystem.
2883  **/
2884 static int __init igbvf_init_module(void)
2885 {
2886 	int ret;
2887 	pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2888 	pr_info("%s\n", igbvf_copyright);
2889 
2890 	ret = pci_register_driver(&igbvf_driver);
2891 
2892 	return ret;
2893 }
2894 module_init(igbvf_init_module);
2895 
2896 /**
2897  * igbvf_exit_module - Driver Exit Cleanup Routine
2898  *
2899  * igbvf_exit_module is called just before the driver is removed
2900  * from memory.
2901  **/
2902 static void __exit igbvf_exit_module(void)
2903 {
2904 	pci_unregister_driver(&igbvf_driver);
2905 }
2906 module_exit(igbvf_exit_module);
2907 
2908 
2909 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2910 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2911 MODULE_LICENSE("GPL");
2912 MODULE_VERSION(DRV_VERSION);
2913 
2914 /* netdev.c */
2915