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 "2.0.1-k"
51char igbvf_driver_name[] = "igbvf";
52const char igbvf_driver_version[] = DRV_VERSION;
53static const char igbvf_driver_string[] =
54		  "Intel(R) Gigabit Virtual Function Network Driver";
55static 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)
59static int debug = -1;
60module_param(debug, int, 0);
61MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62
63static int igbvf_poll(struct napi_struct *napi, int budget);
64static void igbvf_reset(struct igbvf_adapter *);
65static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
67
68static 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
75static 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
82static 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 **/
90static 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 **/
105static 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	if (status & E1000_RXD_STAT_VP) {
111		u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
112		if (test_bit(vid, adapter->active_vlans))
113			__vlan_hwaccel_put_tag(skb, vid);
114	}
115	netif_receive_skb(skb);
116}
117
118static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
119                                         u32 status_err, struct sk_buff *skb)
120{
121	skb_checksum_none_assert(skb);
122
123	/* Ignore Checksum bit is set or checksum is disabled through ethtool */
124	if ((status_err & E1000_RXD_STAT_IXSM) ||
125	    (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
126		return;
127
128	/* TCP/UDP checksum error bit is set */
129	if (status_err &
130	    (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
131		/* let the stack verify checksum errors */
132		adapter->hw_csum_err++;
133		return;
134	}
135
136	/* It must be a TCP or UDP packet with a valid checksum */
137	if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
138		skb->ip_summed = CHECKSUM_UNNECESSARY;
139
140	adapter->hw_csum_good++;
141}
142
143/**
144 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
145 * @rx_ring: address of ring structure to repopulate
146 * @cleaned_count: number of buffers to repopulate
147 **/
148static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
149                                   int cleaned_count)
150{
151	struct igbvf_adapter *adapter = rx_ring->adapter;
152	struct net_device *netdev = adapter->netdev;
153	struct pci_dev *pdev = adapter->pdev;
154	union e1000_adv_rx_desc *rx_desc;
155	struct igbvf_buffer *buffer_info;
156	struct sk_buff *skb;
157	unsigned int i;
158	int bufsz;
159
160	i = rx_ring->next_to_use;
161	buffer_info = &rx_ring->buffer_info[i];
162
163	if (adapter->rx_ps_hdr_size)
164		bufsz = adapter->rx_ps_hdr_size;
165	else
166		bufsz = adapter->rx_buffer_len;
167
168	while (cleaned_count--) {
169		rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
170
171		if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
172			if (!buffer_info->page) {
173				buffer_info->page = alloc_page(GFP_ATOMIC);
174				if (!buffer_info->page) {
175					adapter->alloc_rx_buff_failed++;
176					goto no_buffers;
177				}
178				buffer_info->page_offset = 0;
179			} else {
180				buffer_info->page_offset ^= PAGE_SIZE / 2;
181			}
182			buffer_info->page_dma =
183				dma_map_page(&pdev->dev, buffer_info->page,
184				             buffer_info->page_offset,
185				             PAGE_SIZE / 2,
186					     DMA_FROM_DEVICE);
187		}
188
189		if (!buffer_info->skb) {
190			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
191			if (!skb) {
192				adapter->alloc_rx_buff_failed++;
193				goto no_buffers;
194			}
195
196			buffer_info->skb = skb;
197			buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
198			                                  bufsz,
199							  DMA_FROM_DEVICE);
200		}
201		/* Refresh the desc even if buffer_addrs didn't change because
202		 * each write-back erases this info. */
203		if (adapter->rx_ps_hdr_size) {
204			rx_desc->read.pkt_addr =
205			     cpu_to_le64(buffer_info->page_dma);
206			rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
207		} else {
208			rx_desc->read.pkt_addr =
209			     cpu_to_le64(buffer_info->dma);
210			rx_desc->read.hdr_addr = 0;
211		}
212
213		i++;
214		if (i == rx_ring->count)
215			i = 0;
216		buffer_info = &rx_ring->buffer_info[i];
217	}
218
219no_buffers:
220	if (rx_ring->next_to_use != i) {
221		rx_ring->next_to_use = i;
222		if (i == 0)
223			i = (rx_ring->count - 1);
224		else
225			i--;
226
227		/* Force memory writes to complete before letting h/w
228		 * know there are new descriptors to fetch.  (Only
229		 * applicable for weak-ordered memory model archs,
230		 * such as IA-64). */
231		wmb();
232		writel(i, adapter->hw.hw_addr + rx_ring->tail);
233	}
234}
235
236/**
237 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
238 * @adapter: board private structure
239 *
240 * the return value indicates whether actual cleaning was done, there
241 * is no guarantee that everything was cleaned
242 **/
243static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
244                               int *work_done, int work_to_do)
245{
246	struct igbvf_ring *rx_ring = adapter->rx_ring;
247	struct net_device *netdev = adapter->netdev;
248	struct pci_dev *pdev = adapter->pdev;
249	union e1000_adv_rx_desc *rx_desc, *next_rxd;
250	struct igbvf_buffer *buffer_info, *next_buffer;
251	struct sk_buff *skb;
252	bool cleaned = false;
253	int cleaned_count = 0;
254	unsigned int total_bytes = 0, total_packets = 0;
255	unsigned int i;
256	u32 length, hlen, staterr;
257
258	i = rx_ring->next_to_clean;
259	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
260	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
261
262	while (staterr & E1000_RXD_STAT_DD) {
263		if (*work_done >= work_to_do)
264			break;
265		(*work_done)++;
266		rmb(); /* read descriptor and rx_buffer_info after status DD */
267
268		buffer_info = &rx_ring->buffer_info[i];
269
270		/* HW will not DMA in data larger than the given buffer, even
271		 * if it parses the (NFS, of course) header to be larger.  In
272		 * that case, it fills the header buffer and spills the rest
273		 * into the page.
274		 */
275		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
276		  E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
277		if (hlen > adapter->rx_ps_hdr_size)
278			hlen = adapter->rx_ps_hdr_size;
279
280		length = le16_to_cpu(rx_desc->wb.upper.length);
281		cleaned = true;
282		cleaned_count++;
283
284		skb = buffer_info->skb;
285		prefetch(skb->data - NET_IP_ALIGN);
286		buffer_info->skb = NULL;
287		if (!adapter->rx_ps_hdr_size) {
288			dma_unmap_single(&pdev->dev, buffer_info->dma,
289			                 adapter->rx_buffer_len,
290					 DMA_FROM_DEVICE);
291			buffer_info->dma = 0;
292			skb_put(skb, length);
293			goto send_up;
294		}
295
296		if (!skb_shinfo(skb)->nr_frags) {
297			dma_unmap_single(&pdev->dev, buffer_info->dma,
298			                 adapter->rx_ps_hdr_size,
299					 DMA_FROM_DEVICE);
300			skb_put(skb, hlen);
301		}
302
303		if (length) {
304			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
305			               PAGE_SIZE / 2,
306				       DMA_FROM_DEVICE);
307			buffer_info->page_dma = 0;
308
309			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
310			                   buffer_info->page,
311			                   buffer_info->page_offset,
312			                   length);
313
314			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
315			    (page_count(buffer_info->page) != 1))
316				buffer_info->page = NULL;
317			else
318				get_page(buffer_info->page);
319
320			skb->len += length;
321			skb->data_len += length;
322			skb->truesize += PAGE_SIZE / 2;
323		}
324send_up:
325		i++;
326		if (i == rx_ring->count)
327			i = 0;
328		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
329		prefetch(next_rxd);
330		next_buffer = &rx_ring->buffer_info[i];
331
332		if (!(staterr & E1000_RXD_STAT_EOP)) {
333			buffer_info->skb = next_buffer->skb;
334			buffer_info->dma = next_buffer->dma;
335			next_buffer->skb = skb;
336			next_buffer->dma = 0;
337			goto next_desc;
338		}
339
340		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
341			dev_kfree_skb_irq(skb);
342			goto next_desc;
343		}
344
345		total_bytes += skb->len;
346		total_packets++;
347
348		igbvf_rx_checksum_adv(adapter, staterr, skb);
349
350		skb->protocol = eth_type_trans(skb, netdev);
351
352		igbvf_receive_skb(adapter, netdev, skb, staterr,
353		                  rx_desc->wb.upper.vlan);
354
355next_desc:
356		rx_desc->wb.upper.status_error = 0;
357
358		/* return some buffers to hardware, one at a time is too slow */
359		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
360			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
361			cleaned_count = 0;
362		}
363
364		/* use prefetched values */
365		rx_desc = next_rxd;
366		buffer_info = next_buffer;
367
368		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
369	}
370
371	rx_ring->next_to_clean = i;
372	cleaned_count = igbvf_desc_unused(rx_ring);
373
374	if (cleaned_count)
375		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
376
377	adapter->total_rx_packets += total_packets;
378	adapter->total_rx_bytes += total_bytes;
379	adapter->net_stats.rx_bytes += total_bytes;
380	adapter->net_stats.rx_packets += total_packets;
381	return cleaned;
382}
383
384static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
385                            struct igbvf_buffer *buffer_info)
386{
387	if (buffer_info->dma) {
388		if (buffer_info->mapped_as_page)
389			dma_unmap_page(&adapter->pdev->dev,
390				       buffer_info->dma,
391				       buffer_info->length,
392				       DMA_TO_DEVICE);
393		else
394			dma_unmap_single(&adapter->pdev->dev,
395					 buffer_info->dma,
396					 buffer_info->length,
397					 DMA_TO_DEVICE);
398		buffer_info->dma = 0;
399	}
400	if (buffer_info->skb) {
401		dev_kfree_skb_any(buffer_info->skb);
402		buffer_info->skb = NULL;
403	}
404	buffer_info->time_stamp = 0;
405}
406
407/**
408 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
409 * @adapter: board private structure
410 *
411 * Return 0 on success, negative on failure
412 **/
413int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
414                             struct igbvf_ring *tx_ring)
415{
416	struct pci_dev *pdev = adapter->pdev;
417	int size;
418
419	size = sizeof(struct igbvf_buffer) * tx_ring->count;
420	tx_ring->buffer_info = vzalloc(size);
421	if (!tx_ring->buffer_info)
422		goto err;
423
424	/* round up to nearest 4K */
425	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
426	tx_ring->size = ALIGN(tx_ring->size, 4096);
427
428	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
429					   &tx_ring->dma, GFP_KERNEL);
430
431	if (!tx_ring->desc)
432		goto err;
433
434	tx_ring->adapter = adapter;
435	tx_ring->next_to_use = 0;
436	tx_ring->next_to_clean = 0;
437
438	return 0;
439err:
440	vfree(tx_ring->buffer_info);
441	dev_err(&adapter->pdev->dev,
442	        "Unable to allocate memory for the transmit descriptor ring\n");
443	return -ENOMEM;
444}
445
446/**
447 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
448 * @adapter: board private structure
449 *
450 * Returns 0 on success, negative on failure
451 **/
452int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
453			     struct igbvf_ring *rx_ring)
454{
455	struct pci_dev *pdev = adapter->pdev;
456	int size, desc_len;
457
458	size = sizeof(struct igbvf_buffer) * rx_ring->count;
459	rx_ring->buffer_info = vzalloc(size);
460	if (!rx_ring->buffer_info)
461		goto err;
462
463	desc_len = sizeof(union e1000_adv_rx_desc);
464
465	/* Round up to nearest 4K */
466	rx_ring->size = rx_ring->count * desc_len;
467	rx_ring->size = ALIGN(rx_ring->size, 4096);
468
469	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
470					   &rx_ring->dma, GFP_KERNEL);
471
472	if (!rx_ring->desc)
473		goto err;
474
475	rx_ring->next_to_clean = 0;
476	rx_ring->next_to_use = 0;
477
478	rx_ring->adapter = adapter;
479
480	return 0;
481
482err:
483	vfree(rx_ring->buffer_info);
484	rx_ring->buffer_info = NULL;
485	dev_err(&adapter->pdev->dev,
486	        "Unable to allocate memory for the receive descriptor ring\n");
487	return -ENOMEM;
488}
489
490/**
491 * igbvf_clean_tx_ring - Free Tx Buffers
492 * @tx_ring: ring to be cleaned
493 **/
494static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
495{
496	struct igbvf_adapter *adapter = tx_ring->adapter;
497	struct igbvf_buffer *buffer_info;
498	unsigned long size;
499	unsigned int i;
500
501	if (!tx_ring->buffer_info)
502		return;
503
504	/* Free all the Tx ring sk_buffs */
505	for (i = 0; i < tx_ring->count; i++) {
506		buffer_info = &tx_ring->buffer_info[i];
507		igbvf_put_txbuf(adapter, buffer_info);
508	}
509
510	size = sizeof(struct igbvf_buffer) * tx_ring->count;
511	memset(tx_ring->buffer_info, 0, size);
512
513	/* Zero out the descriptor ring */
514	memset(tx_ring->desc, 0, tx_ring->size);
515
516	tx_ring->next_to_use = 0;
517	tx_ring->next_to_clean = 0;
518
519	writel(0, adapter->hw.hw_addr + tx_ring->head);
520	writel(0, adapter->hw.hw_addr + tx_ring->tail);
521}
522
523/**
524 * igbvf_free_tx_resources - Free Tx Resources per Queue
525 * @tx_ring: ring to free resources from
526 *
527 * Free all transmit software resources
528 **/
529void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
530{
531	struct pci_dev *pdev = tx_ring->adapter->pdev;
532
533	igbvf_clean_tx_ring(tx_ring);
534
535	vfree(tx_ring->buffer_info);
536	tx_ring->buffer_info = NULL;
537
538	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
539			  tx_ring->dma);
540
541	tx_ring->desc = NULL;
542}
543
544/**
545 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
546 * @adapter: board private structure
547 **/
548static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
549{
550	struct igbvf_adapter *adapter = rx_ring->adapter;
551	struct igbvf_buffer *buffer_info;
552	struct pci_dev *pdev = adapter->pdev;
553	unsigned long size;
554	unsigned int i;
555
556	if (!rx_ring->buffer_info)
557		return;
558
559	/* Free all the Rx ring sk_buffs */
560	for (i = 0; i < rx_ring->count; i++) {
561		buffer_info = &rx_ring->buffer_info[i];
562		if (buffer_info->dma) {
563			if (adapter->rx_ps_hdr_size){
564				dma_unmap_single(&pdev->dev, buffer_info->dma,
565				                 adapter->rx_ps_hdr_size,
566						 DMA_FROM_DEVICE);
567			} else {
568				dma_unmap_single(&pdev->dev, buffer_info->dma,
569				                 adapter->rx_buffer_len,
570						 DMA_FROM_DEVICE);
571			}
572			buffer_info->dma = 0;
573		}
574
575		if (buffer_info->skb) {
576			dev_kfree_skb(buffer_info->skb);
577			buffer_info->skb = NULL;
578		}
579
580		if (buffer_info->page) {
581			if (buffer_info->page_dma)
582				dma_unmap_page(&pdev->dev,
583					       buffer_info->page_dma,
584				               PAGE_SIZE / 2,
585					       DMA_FROM_DEVICE);
586			put_page(buffer_info->page);
587			buffer_info->page = NULL;
588			buffer_info->page_dma = 0;
589			buffer_info->page_offset = 0;
590		}
591	}
592
593	size = sizeof(struct igbvf_buffer) * rx_ring->count;
594	memset(rx_ring->buffer_info, 0, size);
595
596	/* Zero out the descriptor ring */
597	memset(rx_ring->desc, 0, rx_ring->size);
598
599	rx_ring->next_to_clean = 0;
600	rx_ring->next_to_use = 0;
601
602	writel(0, adapter->hw.hw_addr + rx_ring->head);
603	writel(0, adapter->hw.hw_addr + rx_ring->tail);
604}
605
606/**
607 * igbvf_free_rx_resources - Free Rx Resources
608 * @rx_ring: ring to clean the resources from
609 *
610 * Free all receive software resources
611 **/
612
613void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
614{
615	struct pci_dev *pdev = rx_ring->adapter->pdev;
616
617	igbvf_clean_rx_ring(rx_ring);
618
619	vfree(rx_ring->buffer_info);
620	rx_ring->buffer_info = NULL;
621
622	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
623	                  rx_ring->dma);
624	rx_ring->desc = NULL;
625}
626
627/**
628 * igbvf_update_itr - update the dynamic ITR value based on statistics
629 * @adapter: pointer to adapter
630 * @itr_setting: current adapter->itr
631 * @packets: the number of packets during this measurement interval
632 * @bytes: the number of bytes during this measurement interval
633 *
634 *      Stores a new ITR value based on packets and byte
635 *      counts during the last interrupt.  The advantage of per interrupt
636 *      computation is faster updates and more accurate ITR for the current
637 *      traffic pattern.  Constants in this function were computed
638 *      based on theoretical maximum wire speed and thresholds were set based
639 *      on testing data as well as attempting to minimize response time
640 *      while increasing bulk throughput.
641 **/
642static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
643					   enum latency_range itr_setting,
644					   int packets, int bytes)
645{
646	enum latency_range retval = itr_setting;
647
648	if (packets == 0)
649		goto update_itr_done;
650
651	switch (itr_setting) {
652	case lowest_latency:
653		/* handle TSO and jumbo frames */
654		if (bytes/packets > 8000)
655			retval = bulk_latency;
656		else if ((packets < 5) && (bytes > 512))
657			retval = low_latency;
658		break;
659	case low_latency:  /* 50 usec aka 20000 ints/s */
660		if (bytes > 10000) {
661			/* this if handles the TSO accounting */
662			if (bytes/packets > 8000)
663				retval = bulk_latency;
664			else if ((packets < 10) || ((bytes/packets) > 1200))
665				retval = bulk_latency;
666			else if ((packets > 35))
667				retval = lowest_latency;
668		} else if (bytes/packets > 2000) {
669			retval = bulk_latency;
670		} else if (packets <= 2 && bytes < 512) {
671			retval = lowest_latency;
672		}
673		break;
674	case bulk_latency: /* 250 usec aka 4000 ints/s */
675		if (bytes > 25000) {
676			if (packets > 35)
677				retval = low_latency;
678		} else if (bytes < 6000) {
679			retval = low_latency;
680		}
681		break;
682	default:
683		break;
684	}
685
686update_itr_done:
687	return retval;
688}
689
690static int igbvf_range_to_itr(enum latency_range current_range)
691{
692	int new_itr;
693
694	switch (current_range) {
695	/* counts and packets in update_itr are dependent on these numbers */
696	case lowest_latency:
697		new_itr = IGBVF_70K_ITR;
698		break;
699	case low_latency:
700		new_itr = IGBVF_20K_ITR;
701		break;
702	case bulk_latency:
703		new_itr = IGBVF_4K_ITR;
704		break;
705	default:
706		new_itr = IGBVF_START_ITR;
707		break;
708	}
709	return new_itr;
710}
711
712static void igbvf_set_itr(struct igbvf_adapter *adapter)
713{
714	u32 new_itr;
715
716	adapter->tx_ring->itr_range =
717			igbvf_update_itr(adapter,
718					 adapter->tx_ring->itr_val,
719					 adapter->total_tx_packets,
720					 adapter->total_tx_bytes);
721
722	/* conservative mode (itr 3) eliminates the lowest_latency setting */
723	if (adapter->requested_itr == 3 &&
724	    adapter->tx_ring->itr_range == lowest_latency)
725		adapter->tx_ring->itr_range = low_latency;
726
727	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
728
729
730	if (new_itr != adapter->tx_ring->itr_val) {
731		u32 current_itr = adapter->tx_ring->itr_val;
732		/*
733		 * this attempts to bias the interrupt rate towards Bulk
734		 * by adding intermediate steps when interrupt rate is
735		 * increasing
736		 */
737		new_itr = new_itr > current_itr ?
738			     min(current_itr + (new_itr >> 2), new_itr) :
739			     new_itr;
740		adapter->tx_ring->itr_val = new_itr;
741
742		adapter->tx_ring->set_itr = 1;
743	}
744
745	adapter->rx_ring->itr_range =
746			igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
747					 adapter->total_rx_packets,
748					 adapter->total_rx_bytes);
749	if (adapter->requested_itr == 3 &&
750	    adapter->rx_ring->itr_range == lowest_latency)
751		adapter->rx_ring->itr_range = low_latency;
752
753	new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
754
755	if (new_itr != adapter->rx_ring->itr_val) {
756		u32 current_itr = adapter->rx_ring->itr_val;
757		new_itr = new_itr > current_itr ?
758			     min(current_itr + (new_itr >> 2), new_itr) :
759			     new_itr;
760		adapter->rx_ring->itr_val = new_itr;
761
762		adapter->rx_ring->set_itr = 1;
763	}
764}
765
766/**
767 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
768 * @adapter: board private structure
769 * returns true if ring is completely cleaned
770 **/
771static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
772{
773	struct igbvf_adapter *adapter = tx_ring->adapter;
774	struct net_device *netdev = adapter->netdev;
775	struct igbvf_buffer *buffer_info;
776	struct sk_buff *skb;
777	union e1000_adv_tx_desc *tx_desc, *eop_desc;
778	unsigned int total_bytes = 0, total_packets = 0;
779	unsigned int i, eop, count = 0;
780	bool cleaned = false;
781
782	i = tx_ring->next_to_clean;
783	eop = tx_ring->buffer_info[i].next_to_watch;
784	eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
785
786	while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
787	       (count < tx_ring->count)) {
788		rmb();	/* read buffer_info after eop_desc status */
789		for (cleaned = false; !cleaned; count++) {
790			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
791			buffer_info = &tx_ring->buffer_info[i];
792			cleaned = (i == eop);
793			skb = buffer_info->skb;
794
795			if (skb) {
796				unsigned int segs, bytecount;
797
798				/* gso_segs is currently only valid for tcp */
799				segs = skb_shinfo(skb)->gso_segs ?: 1;
800				/* multiply data chunks by size of headers */
801				bytecount = ((segs - 1) * skb_headlen(skb)) +
802				            skb->len;
803				total_packets += segs;
804				total_bytes += bytecount;
805			}
806
807			igbvf_put_txbuf(adapter, buffer_info);
808			tx_desc->wb.status = 0;
809
810			i++;
811			if (i == tx_ring->count)
812				i = 0;
813		}
814		eop = tx_ring->buffer_info[i].next_to_watch;
815		eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
816	}
817
818	tx_ring->next_to_clean = i;
819
820	if (unlikely(count &&
821	             netif_carrier_ok(netdev) &&
822	             igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
823		/* Make sure that anybody stopping the queue after this
824		 * sees the new next_to_clean.
825		 */
826		smp_mb();
827		if (netif_queue_stopped(netdev) &&
828		    !(test_bit(__IGBVF_DOWN, &adapter->state))) {
829			netif_wake_queue(netdev);
830			++adapter->restart_queue;
831		}
832	}
833
834	adapter->net_stats.tx_bytes += total_bytes;
835	adapter->net_stats.tx_packets += total_packets;
836	return count < tx_ring->count;
837}
838
839static irqreturn_t igbvf_msix_other(int irq, void *data)
840{
841	struct net_device *netdev = data;
842	struct igbvf_adapter *adapter = netdev_priv(netdev);
843	struct e1000_hw *hw = &adapter->hw;
844
845	adapter->int_counter1++;
846
847	netif_carrier_off(netdev);
848	hw->mac.get_link_status = 1;
849	if (!test_bit(__IGBVF_DOWN, &adapter->state))
850		mod_timer(&adapter->watchdog_timer, jiffies + 1);
851
852	ew32(EIMS, adapter->eims_other);
853
854	return IRQ_HANDLED;
855}
856
857static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
858{
859	struct net_device *netdev = data;
860	struct igbvf_adapter *adapter = netdev_priv(netdev);
861	struct e1000_hw *hw = &adapter->hw;
862	struct igbvf_ring *tx_ring = adapter->tx_ring;
863
864	if (tx_ring->set_itr) {
865		writel(tx_ring->itr_val,
866		       adapter->hw.hw_addr + tx_ring->itr_register);
867		adapter->tx_ring->set_itr = 0;
868	}
869
870	adapter->total_tx_bytes = 0;
871	adapter->total_tx_packets = 0;
872
873	/* auto mask will automatically reenable the interrupt when we write
874	 * EICS */
875	if (!igbvf_clean_tx_irq(tx_ring))
876		/* Ring was not completely cleaned, so fire another interrupt */
877		ew32(EICS, tx_ring->eims_value);
878	else
879		ew32(EIMS, tx_ring->eims_value);
880
881	return IRQ_HANDLED;
882}
883
884static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
885{
886	struct net_device *netdev = data;
887	struct igbvf_adapter *adapter = netdev_priv(netdev);
888
889	adapter->int_counter0++;
890
891	/* Write the ITR value calculated at the end of the
892	 * previous interrupt.
893	 */
894	if (adapter->rx_ring->set_itr) {
895		writel(adapter->rx_ring->itr_val,
896		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
897		adapter->rx_ring->set_itr = 0;
898	}
899
900	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
901		adapter->total_rx_bytes = 0;
902		adapter->total_rx_packets = 0;
903		__napi_schedule(&adapter->rx_ring->napi);
904	}
905
906	return IRQ_HANDLED;
907}
908
909#define IGBVF_NO_QUEUE -1
910
911static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
912                                int tx_queue, int msix_vector)
913{
914	struct e1000_hw *hw = &adapter->hw;
915	u32 ivar, index;
916
917	/* 82576 uses a table-based method for assigning vectors.
918	   Each queue has a single entry in the table to which we write
919	   a vector number along with a "valid" bit.  Sadly, the layout
920	   of the table is somewhat counterintuitive. */
921	if (rx_queue > IGBVF_NO_QUEUE) {
922		index = (rx_queue >> 1);
923		ivar = array_er32(IVAR0, index);
924		if (rx_queue & 0x1) {
925			/* vector goes into third byte of register */
926			ivar = ivar & 0xFF00FFFF;
927			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
928		} else {
929			/* vector goes into low byte of register */
930			ivar = ivar & 0xFFFFFF00;
931			ivar |= msix_vector | E1000_IVAR_VALID;
932		}
933		adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
934		array_ew32(IVAR0, index, ivar);
935	}
936	if (tx_queue > IGBVF_NO_QUEUE) {
937		index = (tx_queue >> 1);
938		ivar = array_er32(IVAR0, index);
939		if (tx_queue & 0x1) {
940			/* vector goes into high byte of register */
941			ivar = ivar & 0x00FFFFFF;
942			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
943		} else {
944			/* vector goes into second byte of register */
945			ivar = ivar & 0xFFFF00FF;
946			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
947		}
948		adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
949		array_ew32(IVAR0, index, ivar);
950	}
951}
952
953/**
954 * igbvf_configure_msix - Configure MSI-X hardware
955 *
956 * igbvf_configure_msix sets up the hardware to properly
957 * generate MSI-X interrupts.
958 **/
959static void igbvf_configure_msix(struct igbvf_adapter *adapter)
960{
961	u32 tmp;
962	struct e1000_hw *hw = &adapter->hw;
963	struct igbvf_ring *tx_ring = adapter->tx_ring;
964	struct igbvf_ring *rx_ring = adapter->rx_ring;
965	int vector = 0;
966
967	adapter->eims_enable_mask = 0;
968
969	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
970	adapter->eims_enable_mask |= tx_ring->eims_value;
971	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
972	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
973	adapter->eims_enable_mask |= rx_ring->eims_value;
974	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
975
976	/* set vector for other causes, i.e. link changes */
977
978	tmp = (vector++ | E1000_IVAR_VALID);
979
980	ew32(IVAR_MISC, tmp);
981
982	adapter->eims_enable_mask = (1 << (vector)) - 1;
983	adapter->eims_other = 1 << (vector - 1);
984	e1e_flush();
985}
986
987static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
988{
989	if (adapter->msix_entries) {
990		pci_disable_msix(adapter->pdev);
991		kfree(adapter->msix_entries);
992		adapter->msix_entries = NULL;
993	}
994}
995
996/**
997 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
998 *
999 * Attempt to configure interrupts using the best available
1000 * capabilities of the hardware and kernel.
1001 **/
1002static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1003{
1004	int err = -ENOMEM;
1005	int i;
1006
1007	/* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1008	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1009	                                GFP_KERNEL);
1010	if (adapter->msix_entries) {
1011		for (i = 0; i < 3; i++)
1012			adapter->msix_entries[i].entry = i;
1013
1014		err = pci_enable_msix(adapter->pdev,
1015		                      adapter->msix_entries, 3);
1016	}
1017
1018	if (err) {
1019		/* MSI-X failed */
1020		dev_err(&adapter->pdev->dev,
1021		        "Failed to initialize MSI-X interrupts.\n");
1022		igbvf_reset_interrupt_capability(adapter);
1023	}
1024}
1025
1026/**
1027 * igbvf_request_msix - Initialize MSI-X interrupts
1028 *
1029 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1030 * kernel.
1031 **/
1032static int igbvf_request_msix(struct igbvf_adapter *adapter)
1033{
1034	struct net_device *netdev = adapter->netdev;
1035	int err = 0, vector = 0;
1036
1037	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1038		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1039		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1040	} else {
1041		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1042		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1043	}
1044
1045	err = request_irq(adapter->msix_entries[vector].vector,
1046	                  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1047	                  netdev);
1048	if (err)
1049		goto out;
1050
1051	adapter->tx_ring->itr_register = E1000_EITR(vector);
1052	adapter->tx_ring->itr_val = adapter->current_itr;
1053	vector++;
1054
1055	err = request_irq(adapter->msix_entries[vector].vector,
1056	                  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1057	                  netdev);
1058	if (err)
1059		goto out;
1060
1061	adapter->rx_ring->itr_register = E1000_EITR(vector);
1062	adapter->rx_ring->itr_val = adapter->current_itr;
1063	vector++;
1064
1065	err = request_irq(adapter->msix_entries[vector].vector,
1066	                  igbvf_msix_other, 0, netdev->name, netdev);
1067	if (err)
1068		goto out;
1069
1070	igbvf_configure_msix(adapter);
1071	return 0;
1072out:
1073	return err;
1074}
1075
1076/**
1077 * igbvf_alloc_queues - Allocate memory for all rings
1078 * @adapter: board private structure to initialize
1079 **/
1080static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1081{
1082	struct net_device *netdev = adapter->netdev;
1083
1084	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1085	if (!adapter->tx_ring)
1086		return -ENOMEM;
1087
1088	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1089	if (!adapter->rx_ring) {
1090		kfree(adapter->tx_ring);
1091		return -ENOMEM;
1092	}
1093
1094	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1095
1096	return 0;
1097}
1098
1099/**
1100 * igbvf_request_irq - initialize interrupts
1101 *
1102 * Attempts to configure interrupts using the best available
1103 * capabilities of the hardware and kernel.
1104 **/
1105static int igbvf_request_irq(struct igbvf_adapter *adapter)
1106{
1107	int err = -1;
1108
1109	/* igbvf supports msi-x only */
1110	if (adapter->msix_entries)
1111		err = igbvf_request_msix(adapter);
1112
1113	if (!err)
1114		return err;
1115
1116	dev_err(&adapter->pdev->dev,
1117	        "Unable to allocate interrupt, Error: %d\n", err);
1118
1119	return err;
1120}
1121
1122static void igbvf_free_irq(struct igbvf_adapter *adapter)
1123{
1124	struct net_device *netdev = adapter->netdev;
1125	int vector;
1126
1127	if (adapter->msix_entries) {
1128		for (vector = 0; vector < 3; vector++)
1129			free_irq(adapter->msix_entries[vector].vector, netdev);
1130	}
1131}
1132
1133/**
1134 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1135 **/
1136static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1137{
1138	struct e1000_hw *hw = &adapter->hw;
1139
1140	ew32(EIMC, ~0);
1141
1142	if (adapter->msix_entries)
1143		ew32(EIAC, 0);
1144}
1145
1146/**
1147 * igbvf_irq_enable - Enable default interrupt generation settings
1148 **/
1149static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1150{
1151	struct e1000_hw *hw = &adapter->hw;
1152
1153	ew32(EIAC, adapter->eims_enable_mask);
1154	ew32(EIAM, adapter->eims_enable_mask);
1155	ew32(EIMS, adapter->eims_enable_mask);
1156}
1157
1158/**
1159 * igbvf_poll - NAPI Rx polling callback
1160 * @napi: struct associated with this polling callback
1161 * @budget: amount of packets driver is allowed to process this poll
1162 **/
1163static int igbvf_poll(struct napi_struct *napi, int budget)
1164{
1165	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1166	struct igbvf_adapter *adapter = rx_ring->adapter;
1167	struct e1000_hw *hw = &adapter->hw;
1168	int work_done = 0;
1169
1170	igbvf_clean_rx_irq(adapter, &work_done, budget);
1171
1172	/* If not enough Rx work done, exit the polling mode */
1173	if (work_done < budget) {
1174		napi_complete(napi);
1175
1176		if (adapter->requested_itr & 3)
1177			igbvf_set_itr(adapter);
1178
1179		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1180			ew32(EIMS, adapter->rx_ring->eims_value);
1181	}
1182
1183	return work_done;
1184}
1185
1186/**
1187 * igbvf_set_rlpml - set receive large packet maximum length
1188 * @adapter: board private structure
1189 *
1190 * Configure the maximum size of packets that will be received
1191 */
1192static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1193{
1194	int max_frame_size;
1195	struct e1000_hw *hw = &adapter->hw;
1196
1197	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1198	e1000_rlpml_set_vf(hw, max_frame_size);
1199}
1200
1201static int igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1202{
1203	struct igbvf_adapter *adapter = netdev_priv(netdev);
1204	struct e1000_hw *hw = &adapter->hw;
1205
1206	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1207		dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1208		return -EINVAL;
1209	}
1210	set_bit(vid, adapter->active_vlans);
1211	return 0;
1212}
1213
1214static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1215{
1216	struct igbvf_adapter *adapter = netdev_priv(netdev);
1217	struct e1000_hw *hw = &adapter->hw;
1218
1219	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1220		dev_err(&adapter->pdev->dev,
1221		        "Failed to remove vlan id %d\n", vid);
1222		return -EINVAL;
1223	}
1224	clear_bit(vid, adapter->active_vlans);
1225	return 0;
1226}
1227
1228static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1229{
1230	u16 vid;
1231
1232	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1233		igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1234}
1235
1236/**
1237 * igbvf_configure_tx - Configure Transmit Unit after Reset
1238 * @adapter: board private structure
1239 *
1240 * Configure the Tx unit of the MAC after a reset.
1241 **/
1242static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1243{
1244	struct e1000_hw *hw = &adapter->hw;
1245	struct igbvf_ring *tx_ring = adapter->tx_ring;
1246	u64 tdba;
1247	u32 txdctl, dca_txctrl;
1248
1249	/* disable transmits */
1250	txdctl = er32(TXDCTL(0));
1251	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1252	e1e_flush();
1253	msleep(10);
1254
1255	/* Setup the HW Tx Head and Tail descriptor pointers */
1256	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1257	tdba = tx_ring->dma;
1258	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1259	ew32(TDBAH(0), (tdba >> 32));
1260	ew32(TDH(0), 0);
1261	ew32(TDT(0), 0);
1262	tx_ring->head = E1000_TDH(0);
1263	tx_ring->tail = E1000_TDT(0);
1264
1265	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1266	 * MUST be delivered in order or it will completely screw up
1267	 * our bookeeping.
1268	 */
1269	dca_txctrl = er32(DCA_TXCTRL(0));
1270	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1271	ew32(DCA_TXCTRL(0), dca_txctrl);
1272
1273	/* enable transmits */
1274	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1275	ew32(TXDCTL(0), txdctl);
1276
1277	/* Setup Transmit Descriptor Settings for eop descriptor */
1278	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1279
1280	/* enable Report Status bit */
1281	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1282}
1283
1284/**
1285 * igbvf_setup_srrctl - configure the receive control registers
1286 * @adapter: Board private structure
1287 **/
1288static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1289{
1290	struct e1000_hw *hw = &adapter->hw;
1291	u32 srrctl = 0;
1292
1293	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1294	            E1000_SRRCTL_BSIZEHDR_MASK |
1295	            E1000_SRRCTL_BSIZEPKT_MASK);
1296
1297	/* Enable queue drop to avoid head of line blocking */
1298	srrctl |= E1000_SRRCTL_DROP_EN;
1299
1300	/* Setup buffer sizes */
1301	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1302	          E1000_SRRCTL_BSIZEPKT_SHIFT;
1303
1304	if (adapter->rx_buffer_len < 2048) {
1305		adapter->rx_ps_hdr_size = 0;
1306		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1307	} else {
1308		adapter->rx_ps_hdr_size = 128;
1309		srrctl |= adapter->rx_ps_hdr_size <<
1310		          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1311		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1312	}
1313
1314	ew32(SRRCTL(0), srrctl);
1315}
1316
1317/**
1318 * igbvf_configure_rx - Configure Receive Unit after Reset
1319 * @adapter: board private structure
1320 *
1321 * Configure the Rx unit of the MAC after a reset.
1322 **/
1323static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1324{
1325	struct e1000_hw *hw = &adapter->hw;
1326	struct igbvf_ring *rx_ring = adapter->rx_ring;
1327	u64 rdba;
1328	u32 rdlen, rxdctl;
1329
1330	/* disable receives */
1331	rxdctl = er32(RXDCTL(0));
1332	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1333	e1e_flush();
1334	msleep(10);
1335
1336	rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1337
1338	/*
1339	 * Setup the HW Rx Head and Tail Descriptor Pointers and
1340	 * the Base and Length of the Rx Descriptor Ring
1341	 */
1342	rdba = rx_ring->dma;
1343	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1344	ew32(RDBAH(0), (rdba >> 32));
1345	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1346	rx_ring->head = E1000_RDH(0);
1347	rx_ring->tail = E1000_RDT(0);
1348	ew32(RDH(0), 0);
1349	ew32(RDT(0), 0);
1350
1351	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1352	rxdctl &= 0xFFF00000;
1353	rxdctl |= IGBVF_RX_PTHRESH;
1354	rxdctl |= IGBVF_RX_HTHRESH << 8;
1355	rxdctl |= IGBVF_RX_WTHRESH << 16;
1356
1357	igbvf_set_rlpml(adapter);
1358
1359	/* enable receives */
1360	ew32(RXDCTL(0), rxdctl);
1361}
1362
1363/**
1364 * igbvf_set_multi - Multicast and Promiscuous mode set
1365 * @netdev: network interface device structure
1366 *
1367 * The set_multi entry point is called whenever the multicast address
1368 * list or the network interface flags are updated.  This routine is
1369 * responsible for configuring the hardware for proper multicast,
1370 * promiscuous mode, and all-multi behavior.
1371 **/
1372static void igbvf_set_multi(struct net_device *netdev)
1373{
1374	struct igbvf_adapter *adapter = netdev_priv(netdev);
1375	struct e1000_hw *hw = &adapter->hw;
1376	struct netdev_hw_addr *ha;
1377	u8  *mta_list = NULL;
1378	int i;
1379
1380	if (!netdev_mc_empty(netdev)) {
1381		mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1382		if (!mta_list) {
1383			dev_err(&adapter->pdev->dev,
1384			        "failed to allocate multicast filter list\n");
1385			return;
1386		}
1387	}
1388
1389	/* prepare a packed array of only addresses. */
1390	i = 0;
1391	netdev_for_each_mc_addr(ha, netdev)
1392		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1393
1394	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1395	kfree(mta_list);
1396}
1397
1398/**
1399 * igbvf_configure - configure the hardware for Rx and Tx
1400 * @adapter: private board structure
1401 **/
1402static void igbvf_configure(struct igbvf_adapter *adapter)
1403{
1404	igbvf_set_multi(adapter->netdev);
1405
1406	igbvf_restore_vlan(adapter);
1407
1408	igbvf_configure_tx(adapter);
1409	igbvf_setup_srrctl(adapter);
1410	igbvf_configure_rx(adapter);
1411	igbvf_alloc_rx_buffers(adapter->rx_ring,
1412	                       igbvf_desc_unused(adapter->rx_ring));
1413}
1414
1415/* igbvf_reset - bring the hardware into a known good state
1416 *
1417 * This function boots the hardware and enables some settings that
1418 * require a configuration cycle of the hardware - those cannot be
1419 * set/changed during runtime. After reset the device needs to be
1420 * properly configured for Rx, Tx etc.
1421 */
1422static void igbvf_reset(struct igbvf_adapter *adapter)
1423{
1424	struct e1000_mac_info *mac = &adapter->hw.mac;
1425	struct net_device *netdev = adapter->netdev;
1426	struct e1000_hw *hw = &adapter->hw;
1427
1428	/* Allow time for pending master requests to run */
1429	if (mac->ops.reset_hw(hw))
1430		dev_err(&adapter->pdev->dev, "PF still resetting\n");
1431
1432	mac->ops.init_hw(hw);
1433
1434	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1435		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1436		       netdev->addr_len);
1437		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1438		       netdev->addr_len);
1439	}
1440
1441	adapter->last_reset = jiffies;
1442}
1443
1444int igbvf_up(struct igbvf_adapter *adapter)
1445{
1446	struct e1000_hw *hw = &adapter->hw;
1447
1448	/* hardware has been reset, we need to reload some things */
1449	igbvf_configure(adapter);
1450
1451	clear_bit(__IGBVF_DOWN, &adapter->state);
1452
1453	napi_enable(&adapter->rx_ring->napi);
1454	if (adapter->msix_entries)
1455		igbvf_configure_msix(adapter);
1456
1457	/* Clear any pending interrupts. */
1458	er32(EICR);
1459	igbvf_irq_enable(adapter);
1460
1461	/* start the watchdog */
1462	hw->mac.get_link_status = 1;
1463	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1464
1465
1466	return 0;
1467}
1468
1469void igbvf_down(struct igbvf_adapter *adapter)
1470{
1471	struct net_device *netdev = adapter->netdev;
1472	struct e1000_hw *hw = &adapter->hw;
1473	u32 rxdctl, txdctl;
1474
1475	/*
1476	 * signal that we're down so the interrupt handler does not
1477	 * reschedule our watchdog timer
1478	 */
1479	set_bit(__IGBVF_DOWN, &adapter->state);
1480
1481	/* disable receives in the hardware */
1482	rxdctl = er32(RXDCTL(0));
1483	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1484
1485	netif_stop_queue(netdev);
1486
1487	/* disable transmits in the hardware */
1488	txdctl = er32(TXDCTL(0));
1489	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1490
1491	/* flush both disables and wait for them to finish */
1492	e1e_flush();
1493	msleep(10);
1494
1495	napi_disable(&adapter->rx_ring->napi);
1496
1497	igbvf_irq_disable(adapter);
1498
1499	del_timer_sync(&adapter->watchdog_timer);
1500
1501	netif_carrier_off(netdev);
1502
1503	/* record the stats before reset*/
1504	igbvf_update_stats(adapter);
1505
1506	adapter->link_speed = 0;
1507	adapter->link_duplex = 0;
1508
1509	igbvf_reset(adapter);
1510	igbvf_clean_tx_ring(adapter->tx_ring);
1511	igbvf_clean_rx_ring(adapter->rx_ring);
1512}
1513
1514void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1515{
1516	might_sleep();
1517	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1518		msleep(1);
1519	igbvf_down(adapter);
1520	igbvf_up(adapter);
1521	clear_bit(__IGBVF_RESETTING, &adapter->state);
1522}
1523
1524/**
1525 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1526 * @adapter: board private structure to initialize
1527 *
1528 * igbvf_sw_init initializes the Adapter private data structure.
1529 * Fields are initialized based on PCI device information and
1530 * OS network device settings (MTU size).
1531 **/
1532static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1533{
1534	struct net_device *netdev = adapter->netdev;
1535	s32 rc;
1536
1537	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1538	adapter->rx_ps_hdr_size = 0;
1539	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1540	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1541
1542	adapter->tx_int_delay = 8;
1543	adapter->tx_abs_int_delay = 32;
1544	adapter->rx_int_delay = 0;
1545	adapter->rx_abs_int_delay = 8;
1546	adapter->requested_itr = 3;
1547	adapter->current_itr = IGBVF_START_ITR;
1548
1549	/* Set various function pointers */
1550	adapter->ei->init_ops(&adapter->hw);
1551
1552	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1553	if (rc)
1554		return rc;
1555
1556	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1557	if (rc)
1558		return rc;
1559
1560	igbvf_set_interrupt_capability(adapter);
1561
1562	if (igbvf_alloc_queues(adapter))
1563		return -ENOMEM;
1564
1565	spin_lock_init(&adapter->tx_queue_lock);
1566
1567	/* Explicitly disable IRQ since the NIC can be in any state. */
1568	igbvf_irq_disable(adapter);
1569
1570	spin_lock_init(&adapter->stats_lock);
1571
1572	set_bit(__IGBVF_DOWN, &adapter->state);
1573	return 0;
1574}
1575
1576static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1577{
1578	struct e1000_hw *hw = &adapter->hw;
1579
1580	adapter->stats.last_gprc = er32(VFGPRC);
1581	adapter->stats.last_gorc = er32(VFGORC);
1582	adapter->stats.last_gptc = er32(VFGPTC);
1583	adapter->stats.last_gotc = er32(VFGOTC);
1584	adapter->stats.last_mprc = er32(VFMPRC);
1585	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1586	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1587	adapter->stats.last_gorlbc = er32(VFGORLBC);
1588	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1589
1590	adapter->stats.base_gprc = er32(VFGPRC);
1591	adapter->stats.base_gorc = er32(VFGORC);
1592	adapter->stats.base_gptc = er32(VFGPTC);
1593	adapter->stats.base_gotc = er32(VFGOTC);
1594	adapter->stats.base_mprc = er32(VFMPRC);
1595	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1596	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1597	adapter->stats.base_gorlbc = er32(VFGORLBC);
1598	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1599}
1600
1601/**
1602 * igbvf_open - Called when a network interface is made active
1603 * @netdev: network interface device structure
1604 *
1605 * Returns 0 on success, negative value on failure
1606 *
1607 * The open entry point is called when a network interface is made
1608 * active by the system (IFF_UP).  At this point all resources needed
1609 * for transmit and receive operations are allocated, the interrupt
1610 * handler is registered with the OS, the watchdog timer is started,
1611 * and the stack is notified that the interface is ready.
1612 **/
1613static int igbvf_open(struct net_device *netdev)
1614{
1615	struct igbvf_adapter *adapter = netdev_priv(netdev);
1616	struct e1000_hw *hw = &adapter->hw;
1617	int err;
1618
1619	/* disallow open during test */
1620	if (test_bit(__IGBVF_TESTING, &adapter->state))
1621		return -EBUSY;
1622
1623	/* allocate transmit descriptors */
1624	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1625	if (err)
1626		goto err_setup_tx;
1627
1628	/* allocate receive descriptors */
1629	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1630	if (err)
1631		goto err_setup_rx;
1632
1633	/*
1634	 * before we allocate an interrupt, we must be ready to handle it.
1635	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1636	 * as soon as we call pci_request_irq, so we have to setup our
1637	 * clean_rx handler before we do so.
1638	 */
1639	igbvf_configure(adapter);
1640
1641	err = igbvf_request_irq(adapter);
1642	if (err)
1643		goto err_req_irq;
1644
1645	/* From here on the code is the same as igbvf_up() */
1646	clear_bit(__IGBVF_DOWN, &adapter->state);
1647
1648	napi_enable(&adapter->rx_ring->napi);
1649
1650	/* clear any pending interrupts */
1651	er32(EICR);
1652
1653	igbvf_irq_enable(adapter);
1654
1655	/* start the watchdog */
1656	hw->mac.get_link_status = 1;
1657	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1658
1659	return 0;
1660
1661err_req_irq:
1662	igbvf_free_rx_resources(adapter->rx_ring);
1663err_setup_rx:
1664	igbvf_free_tx_resources(adapter->tx_ring);
1665err_setup_tx:
1666	igbvf_reset(adapter);
1667
1668	return err;
1669}
1670
1671/**
1672 * igbvf_close - Disables a network interface
1673 * @netdev: network interface device structure
1674 *
1675 * Returns 0, this is not allowed to fail
1676 *
1677 * The close entry point is called when an interface is de-activated
1678 * by the OS.  The hardware is still under the drivers control, but
1679 * needs to be disabled.  A global MAC reset is issued to stop the
1680 * hardware, and all transmit and receive resources are freed.
1681 **/
1682static int igbvf_close(struct net_device *netdev)
1683{
1684	struct igbvf_adapter *adapter = netdev_priv(netdev);
1685
1686	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1687	igbvf_down(adapter);
1688
1689	igbvf_free_irq(adapter);
1690
1691	igbvf_free_tx_resources(adapter->tx_ring);
1692	igbvf_free_rx_resources(adapter->rx_ring);
1693
1694	return 0;
1695}
1696/**
1697 * igbvf_set_mac - Change the Ethernet Address of the NIC
1698 * @netdev: network interface device structure
1699 * @p: pointer to an address structure
1700 *
1701 * Returns 0 on success, negative on failure
1702 **/
1703static int igbvf_set_mac(struct net_device *netdev, void *p)
1704{
1705	struct igbvf_adapter *adapter = netdev_priv(netdev);
1706	struct e1000_hw *hw = &adapter->hw;
1707	struct sockaddr *addr = p;
1708
1709	if (!is_valid_ether_addr(addr->sa_data))
1710		return -EADDRNOTAVAIL;
1711
1712	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1713
1714	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1715
1716	if (memcmp(addr->sa_data, hw->mac.addr, 6))
1717		return -EADDRNOTAVAIL;
1718
1719	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1720	netdev->addr_assign_type &= ~NET_ADDR_RANDOM;
1721
1722	return 0;
1723}
1724
1725#define UPDATE_VF_COUNTER(reg, name)                                    \
1726	{                                                               \
1727		u32 current_counter = er32(reg);                        \
1728		if (current_counter < adapter->stats.last_##name)       \
1729			adapter->stats.name += 0x100000000LL;           \
1730		adapter->stats.last_##name = current_counter;           \
1731		adapter->stats.name &= 0xFFFFFFFF00000000LL;            \
1732		adapter->stats.name |= current_counter;                 \
1733	}
1734
1735/**
1736 * igbvf_update_stats - Update the board statistics counters
1737 * @adapter: board private structure
1738**/
1739void igbvf_update_stats(struct igbvf_adapter *adapter)
1740{
1741	struct e1000_hw *hw = &adapter->hw;
1742	struct pci_dev *pdev = adapter->pdev;
1743
1744	/*
1745	 * Prevent stats update while adapter is being reset, link is down
1746	 * or if the pci connection is down.
1747	 */
1748	if (adapter->link_speed == 0)
1749		return;
1750
1751	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1752		return;
1753
1754	if (pci_channel_offline(pdev))
1755		return;
1756
1757	UPDATE_VF_COUNTER(VFGPRC, gprc);
1758	UPDATE_VF_COUNTER(VFGORC, gorc);
1759	UPDATE_VF_COUNTER(VFGPTC, gptc);
1760	UPDATE_VF_COUNTER(VFGOTC, gotc);
1761	UPDATE_VF_COUNTER(VFMPRC, mprc);
1762	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1763	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1764	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1765	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1766
1767	/* Fill out the OS statistics structure */
1768	adapter->net_stats.multicast = adapter->stats.mprc;
1769}
1770
1771static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1772{
1773	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1774		 adapter->link_speed,
1775		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1776}
1777
1778static bool igbvf_has_link(struct igbvf_adapter *adapter)
1779{
1780	struct e1000_hw *hw = &adapter->hw;
1781	s32 ret_val = E1000_SUCCESS;
1782	bool link_active;
1783
1784	/* If interface is down, stay link down */
1785	if (test_bit(__IGBVF_DOWN, &adapter->state))
1786		return false;
1787
1788	ret_val = hw->mac.ops.check_for_link(hw);
1789	link_active = !hw->mac.get_link_status;
1790
1791	/* if check for link returns error we will need to reset */
1792	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1793		schedule_work(&adapter->reset_task);
1794
1795	return link_active;
1796}
1797
1798/**
1799 * igbvf_watchdog - Timer Call-back
1800 * @data: pointer to adapter cast into an unsigned long
1801 **/
1802static void igbvf_watchdog(unsigned long data)
1803{
1804	struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1805
1806	/* Do the rest outside of interrupt context */
1807	schedule_work(&adapter->watchdog_task);
1808}
1809
1810static void igbvf_watchdog_task(struct work_struct *work)
1811{
1812	struct igbvf_adapter *adapter = container_of(work,
1813	                                             struct igbvf_adapter,
1814	                                             watchdog_task);
1815	struct net_device *netdev = adapter->netdev;
1816	struct e1000_mac_info *mac = &adapter->hw.mac;
1817	struct igbvf_ring *tx_ring = adapter->tx_ring;
1818	struct e1000_hw *hw = &adapter->hw;
1819	u32 link;
1820	int tx_pending = 0;
1821
1822	link = igbvf_has_link(adapter);
1823
1824	if (link) {
1825		if (!netif_carrier_ok(netdev)) {
1826			mac->ops.get_link_up_info(&adapter->hw,
1827			                          &adapter->link_speed,
1828			                          &adapter->link_duplex);
1829			igbvf_print_link_info(adapter);
1830
1831			netif_carrier_on(netdev);
1832			netif_wake_queue(netdev);
1833		}
1834	} else {
1835		if (netif_carrier_ok(netdev)) {
1836			adapter->link_speed = 0;
1837			adapter->link_duplex = 0;
1838			dev_info(&adapter->pdev->dev, "Link is Down\n");
1839			netif_carrier_off(netdev);
1840			netif_stop_queue(netdev);
1841		}
1842	}
1843
1844	if (netif_carrier_ok(netdev)) {
1845		igbvf_update_stats(adapter);
1846	} else {
1847		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1848		              tx_ring->count);
1849		if (tx_pending) {
1850			/*
1851			 * We've lost link, so the controller stops DMA,
1852			 * but we've got queued Tx work that's never going
1853			 * to get done, so reset controller to flush Tx.
1854			 * (Do the reset outside of interrupt context).
1855			 */
1856			adapter->tx_timeout_count++;
1857			schedule_work(&adapter->reset_task);
1858		}
1859	}
1860
1861	/* Cause software interrupt to ensure Rx ring is cleaned */
1862	ew32(EICS, adapter->rx_ring->eims_value);
1863
1864	/* Reset the timer */
1865	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1866		mod_timer(&adapter->watchdog_timer,
1867			  round_jiffies(jiffies + (2 * HZ)));
1868}
1869
1870#define IGBVF_TX_FLAGS_CSUM             0x00000001
1871#define IGBVF_TX_FLAGS_VLAN             0x00000002
1872#define IGBVF_TX_FLAGS_TSO              0x00000004
1873#define IGBVF_TX_FLAGS_IPV4             0x00000008
1874#define IGBVF_TX_FLAGS_VLAN_MASK        0xffff0000
1875#define IGBVF_TX_FLAGS_VLAN_SHIFT       16
1876
1877static int igbvf_tso(struct igbvf_adapter *adapter,
1878                     struct igbvf_ring *tx_ring,
1879                     struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1880{
1881	struct e1000_adv_tx_context_desc *context_desc;
1882	unsigned int i;
1883	int err;
1884	struct igbvf_buffer *buffer_info;
1885	u32 info = 0, tu_cmd = 0;
1886	u32 mss_l4len_idx, l4len;
1887	*hdr_len = 0;
1888
1889	if (skb_header_cloned(skb)) {
1890		err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1891		if (err) {
1892			dev_err(&adapter->pdev->dev,
1893			        "igbvf_tso returning an error\n");
1894			return err;
1895		}
1896	}
1897
1898	l4len = tcp_hdrlen(skb);
1899	*hdr_len += l4len;
1900
1901	if (skb->protocol == htons(ETH_P_IP)) {
1902		struct iphdr *iph = ip_hdr(skb);
1903		iph->tot_len = 0;
1904		iph->check = 0;
1905		tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1906		                                         iph->daddr, 0,
1907		                                         IPPROTO_TCP,
1908		                                         0);
1909	} else if (skb_is_gso_v6(skb)) {
1910		ipv6_hdr(skb)->payload_len = 0;
1911		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1912		                                       &ipv6_hdr(skb)->daddr,
1913		                                       0, IPPROTO_TCP, 0);
1914	}
1915
1916	i = tx_ring->next_to_use;
1917
1918	buffer_info = &tx_ring->buffer_info[i];
1919	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1920	/* VLAN MACLEN IPLEN */
1921	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1922		info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1923	info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1924	*hdr_len += skb_network_offset(skb);
1925	info |= (skb_transport_header(skb) - skb_network_header(skb));
1926	*hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1927	context_desc->vlan_macip_lens = cpu_to_le32(info);
1928
1929	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1930	tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1931
1932	if (skb->protocol == htons(ETH_P_IP))
1933		tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1934	tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1935
1936	context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1937
1938	/* MSS L4LEN IDX */
1939	mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1940	mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1941
1942	context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1943	context_desc->seqnum_seed = 0;
1944
1945	buffer_info->time_stamp = jiffies;
1946	buffer_info->next_to_watch = i;
1947	buffer_info->dma = 0;
1948	i++;
1949	if (i == tx_ring->count)
1950		i = 0;
1951
1952	tx_ring->next_to_use = i;
1953
1954	return true;
1955}
1956
1957static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1958                                 struct igbvf_ring *tx_ring,
1959                                 struct sk_buff *skb, u32 tx_flags)
1960{
1961	struct e1000_adv_tx_context_desc *context_desc;
1962	unsigned int i;
1963	struct igbvf_buffer *buffer_info;
1964	u32 info = 0, tu_cmd = 0;
1965
1966	if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1967	    (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1968		i = tx_ring->next_to_use;
1969		buffer_info = &tx_ring->buffer_info[i];
1970		context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1971
1972		if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1973			info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1974
1975		info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1976		if (skb->ip_summed == CHECKSUM_PARTIAL)
1977			info |= (skb_transport_header(skb) -
1978			         skb_network_header(skb));
1979
1980
1981		context_desc->vlan_macip_lens = cpu_to_le32(info);
1982
1983		tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1984
1985		if (skb->ip_summed == CHECKSUM_PARTIAL) {
1986			switch (skb->protocol) {
1987			case __constant_htons(ETH_P_IP):
1988				tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1989				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1990					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1991				break;
1992			case __constant_htons(ETH_P_IPV6):
1993				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1994					tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1995				break;
1996			default:
1997				break;
1998			}
1999		}
2000
2001		context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2002		context_desc->seqnum_seed = 0;
2003		context_desc->mss_l4len_idx = 0;
2004
2005		buffer_info->time_stamp = jiffies;
2006		buffer_info->next_to_watch = i;
2007		buffer_info->dma = 0;
2008		i++;
2009		if (i == tx_ring->count)
2010			i = 0;
2011		tx_ring->next_to_use = i;
2012
2013		return true;
2014	}
2015
2016	return false;
2017}
2018
2019static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2020{
2021	struct igbvf_adapter *adapter = netdev_priv(netdev);
2022
2023	/* there is enough descriptors then we don't need to worry  */
2024	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2025		return 0;
2026
2027	netif_stop_queue(netdev);
2028
2029	smp_mb();
2030
2031	/* We need to check again just in case room has been made available */
2032	if (igbvf_desc_unused(adapter->tx_ring) < size)
2033		return -EBUSY;
2034
2035	netif_wake_queue(netdev);
2036
2037	++adapter->restart_queue;
2038	return 0;
2039}
2040
2041#define IGBVF_MAX_TXD_PWR       16
2042#define IGBVF_MAX_DATA_PER_TXD  (1 << IGBVF_MAX_TXD_PWR)
2043
2044static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2045                                   struct igbvf_ring *tx_ring,
2046                                   struct sk_buff *skb,
2047                                   unsigned int first)
2048{
2049	struct igbvf_buffer *buffer_info;
2050	struct pci_dev *pdev = adapter->pdev;
2051	unsigned int len = skb_headlen(skb);
2052	unsigned int count = 0, i;
2053	unsigned int f;
2054
2055	i = tx_ring->next_to_use;
2056
2057	buffer_info = &tx_ring->buffer_info[i];
2058	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2059	buffer_info->length = len;
2060	/* set time_stamp *before* dma to help avoid a possible race */
2061	buffer_info->time_stamp = jiffies;
2062	buffer_info->next_to_watch = i;
2063	buffer_info->mapped_as_page = false;
2064	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2065					  DMA_TO_DEVICE);
2066	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2067		goto dma_error;
2068
2069
2070	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2071		const struct skb_frag_struct *frag;
2072
2073		count++;
2074		i++;
2075		if (i == tx_ring->count)
2076			i = 0;
2077
2078		frag = &skb_shinfo(skb)->frags[f];
2079		len = skb_frag_size(frag);
2080
2081		buffer_info = &tx_ring->buffer_info[i];
2082		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2083		buffer_info->length = len;
2084		buffer_info->time_stamp = jiffies;
2085		buffer_info->next_to_watch = i;
2086		buffer_info->mapped_as_page = true;
2087		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2088						DMA_TO_DEVICE);
2089		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2090			goto dma_error;
2091	}
2092
2093	tx_ring->buffer_info[i].skb = skb;
2094	tx_ring->buffer_info[first].next_to_watch = i;
2095
2096	return ++count;
2097
2098dma_error:
2099	dev_err(&pdev->dev, "TX DMA map failed\n");
2100
2101	/* clear timestamp and dma mappings for failed buffer_info mapping */
2102	buffer_info->dma = 0;
2103	buffer_info->time_stamp = 0;
2104	buffer_info->length = 0;
2105	buffer_info->next_to_watch = 0;
2106	buffer_info->mapped_as_page = false;
2107	if (count)
2108		count--;
2109
2110	/* clear timestamp and dma mappings for remaining portion of packet */
2111	while (count--) {
2112		if (i==0)
2113			i += tx_ring->count;
2114		i--;
2115		buffer_info = &tx_ring->buffer_info[i];
2116		igbvf_put_txbuf(adapter, buffer_info);
2117	}
2118
2119	return 0;
2120}
2121
2122static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2123                                      struct igbvf_ring *tx_ring,
2124                                      int tx_flags, int count, u32 paylen,
2125                                      u8 hdr_len)
2126{
2127	union e1000_adv_tx_desc *tx_desc = NULL;
2128	struct igbvf_buffer *buffer_info;
2129	u32 olinfo_status = 0, cmd_type_len;
2130	unsigned int i;
2131
2132	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2133	                E1000_ADVTXD_DCMD_DEXT);
2134
2135	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2136		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2137
2138	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2139		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2140
2141		/* insert tcp checksum */
2142		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2143
2144		/* insert ip checksum */
2145		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2146			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2147
2148	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2149		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2150	}
2151
2152	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2153
2154	i = tx_ring->next_to_use;
2155	while (count--) {
2156		buffer_info = &tx_ring->buffer_info[i];
2157		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2158		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2159		tx_desc->read.cmd_type_len =
2160		         cpu_to_le32(cmd_type_len | buffer_info->length);
2161		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2162		i++;
2163		if (i == tx_ring->count)
2164			i = 0;
2165	}
2166
2167	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2168	/* Force memory writes to complete before letting h/w
2169	 * know there are new descriptors to fetch.  (Only
2170	 * applicable for weak-ordered memory model archs,
2171	 * such as IA-64). */
2172	wmb();
2173
2174	tx_ring->next_to_use = i;
2175	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2176	/* we need this if more than one processor can write to our tail
2177	 * at a time, it syncronizes IO on IA64/Altix systems */
2178	mmiowb();
2179}
2180
2181static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2182					     struct net_device *netdev,
2183					     struct igbvf_ring *tx_ring)
2184{
2185	struct igbvf_adapter *adapter = netdev_priv(netdev);
2186	unsigned int first, tx_flags = 0;
2187	u8 hdr_len = 0;
2188	int count = 0;
2189	int tso = 0;
2190
2191	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2192		dev_kfree_skb_any(skb);
2193		return NETDEV_TX_OK;
2194	}
2195
2196	if (skb->len <= 0) {
2197		dev_kfree_skb_any(skb);
2198		return NETDEV_TX_OK;
2199	}
2200
2201	/*
2202	 * need: count + 4 desc gap to keep tail from touching
2203         *       + 2 desc gap to keep tail from touching head,
2204         *       + 1 desc for skb->data,
2205         *       + 1 desc for context descriptor,
2206	 * head, otherwise try next time
2207	 */
2208	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2209		/* this is a hard error */
2210		return NETDEV_TX_BUSY;
2211	}
2212
2213	if (vlan_tx_tag_present(skb)) {
2214		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2215		tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2216	}
2217
2218	if (skb->protocol == htons(ETH_P_IP))
2219		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2220
2221	first = tx_ring->next_to_use;
2222
2223	tso = skb_is_gso(skb) ?
2224		igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2225	if (unlikely(tso < 0)) {
2226		dev_kfree_skb_any(skb);
2227		return NETDEV_TX_OK;
2228	}
2229
2230	if (tso)
2231		tx_flags |= IGBVF_TX_FLAGS_TSO;
2232	else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2233	         (skb->ip_summed == CHECKSUM_PARTIAL))
2234		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2235
2236	/*
2237	 * count reflects descriptors mapped, if 0 then mapping error
2238	 * has occurred and we need to rewind the descriptor queue
2239	 */
2240	count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2241
2242	if (count) {
2243		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2244		                   skb->len, hdr_len);
2245		/* Make sure there is space in the ring for the next send. */
2246		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2247	} else {
2248		dev_kfree_skb_any(skb);
2249		tx_ring->buffer_info[first].time_stamp = 0;
2250		tx_ring->next_to_use = first;
2251	}
2252
2253	return NETDEV_TX_OK;
2254}
2255
2256static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2257				    struct net_device *netdev)
2258{
2259	struct igbvf_adapter *adapter = netdev_priv(netdev);
2260	struct igbvf_ring *tx_ring;
2261
2262	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2263		dev_kfree_skb_any(skb);
2264		return NETDEV_TX_OK;
2265	}
2266
2267	tx_ring = &adapter->tx_ring[0];
2268
2269	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2270}
2271
2272/**
2273 * igbvf_tx_timeout - Respond to a Tx Hang
2274 * @netdev: network interface device structure
2275 **/
2276static void igbvf_tx_timeout(struct net_device *netdev)
2277{
2278	struct igbvf_adapter *adapter = netdev_priv(netdev);
2279
2280	/* Do the reset outside of interrupt context */
2281	adapter->tx_timeout_count++;
2282	schedule_work(&adapter->reset_task);
2283}
2284
2285static void igbvf_reset_task(struct work_struct *work)
2286{
2287	struct igbvf_adapter *adapter;
2288	adapter = container_of(work, struct igbvf_adapter, reset_task);
2289
2290	igbvf_reinit_locked(adapter);
2291}
2292
2293/**
2294 * igbvf_get_stats - Get System Network Statistics
2295 * @netdev: network interface device structure
2296 *
2297 * Returns the address of the device statistics structure.
2298 * The statistics are actually updated from the timer callback.
2299 **/
2300static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2301{
2302	struct igbvf_adapter *adapter = netdev_priv(netdev);
2303
2304	/* only return the current stats */
2305	return &adapter->net_stats;
2306}
2307
2308/**
2309 * igbvf_change_mtu - Change the Maximum Transfer Unit
2310 * @netdev: network interface device structure
2311 * @new_mtu: new value for maximum frame size
2312 *
2313 * Returns 0 on success, negative on failure
2314 **/
2315static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2316{
2317	struct igbvf_adapter *adapter = netdev_priv(netdev);
2318	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2319
2320	if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2321		dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2322		return -EINVAL;
2323	}
2324
2325#define MAX_STD_JUMBO_FRAME_SIZE 9234
2326	if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2327		dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2328		return -EINVAL;
2329	}
2330
2331	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2332		msleep(1);
2333	/* igbvf_down has a dependency on max_frame_size */
2334	adapter->max_frame_size = max_frame;
2335	if (netif_running(netdev))
2336		igbvf_down(adapter);
2337
2338	/*
2339	 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2340	 * means we reserve 2 more, this pushes us to allocate from the next
2341	 * larger slab size.
2342	 * i.e. RXBUFFER_2048 --> size-4096 slab
2343	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2344	 * fragmented skbs
2345	 */
2346
2347	if (max_frame <= 1024)
2348		adapter->rx_buffer_len = 1024;
2349	else if (max_frame <= 2048)
2350		adapter->rx_buffer_len = 2048;
2351	else
2352#if (PAGE_SIZE / 2) > 16384
2353		adapter->rx_buffer_len = 16384;
2354#else
2355		adapter->rx_buffer_len = PAGE_SIZE / 2;
2356#endif
2357
2358
2359	/* adjust allocation if LPE protects us, and we aren't using SBP */
2360	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2361	     (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2362		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2363		                         ETH_FCS_LEN;
2364
2365	dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2366	         netdev->mtu, new_mtu);
2367	netdev->mtu = new_mtu;
2368
2369	if (netif_running(netdev))
2370		igbvf_up(adapter);
2371	else
2372		igbvf_reset(adapter);
2373
2374	clear_bit(__IGBVF_RESETTING, &adapter->state);
2375
2376	return 0;
2377}
2378
2379static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2380{
2381	switch (cmd) {
2382	default:
2383		return -EOPNOTSUPP;
2384	}
2385}
2386
2387static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2388{
2389	struct net_device *netdev = pci_get_drvdata(pdev);
2390	struct igbvf_adapter *adapter = netdev_priv(netdev);
2391#ifdef CONFIG_PM
2392	int retval = 0;
2393#endif
2394
2395	netif_device_detach(netdev);
2396
2397	if (netif_running(netdev)) {
2398		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2399		igbvf_down(adapter);
2400		igbvf_free_irq(adapter);
2401	}
2402
2403#ifdef CONFIG_PM
2404	retval = pci_save_state(pdev);
2405	if (retval)
2406		return retval;
2407#endif
2408
2409	pci_disable_device(pdev);
2410
2411	return 0;
2412}
2413
2414#ifdef CONFIG_PM
2415static int igbvf_resume(struct pci_dev *pdev)
2416{
2417	struct net_device *netdev = pci_get_drvdata(pdev);
2418	struct igbvf_adapter *adapter = netdev_priv(netdev);
2419	u32 err;
2420
2421	pci_restore_state(pdev);
2422	err = pci_enable_device_mem(pdev);
2423	if (err) {
2424		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2425		return err;
2426	}
2427
2428	pci_set_master(pdev);
2429
2430	if (netif_running(netdev)) {
2431		err = igbvf_request_irq(adapter);
2432		if (err)
2433			return err;
2434	}
2435
2436	igbvf_reset(adapter);
2437
2438	if (netif_running(netdev))
2439		igbvf_up(adapter);
2440
2441	netif_device_attach(netdev);
2442
2443	return 0;
2444}
2445#endif
2446
2447static void igbvf_shutdown(struct pci_dev *pdev)
2448{
2449	igbvf_suspend(pdev, PMSG_SUSPEND);
2450}
2451
2452#ifdef CONFIG_NET_POLL_CONTROLLER
2453/*
2454 * Polling 'interrupt' - used by things like netconsole to send skbs
2455 * without having to re-enable interrupts. It's not called while
2456 * the interrupt routine is executing.
2457 */
2458static void igbvf_netpoll(struct net_device *netdev)
2459{
2460	struct igbvf_adapter *adapter = netdev_priv(netdev);
2461
2462	disable_irq(adapter->pdev->irq);
2463
2464	igbvf_clean_tx_irq(adapter->tx_ring);
2465
2466	enable_irq(adapter->pdev->irq);
2467}
2468#endif
2469
2470/**
2471 * igbvf_io_error_detected - called when PCI error is detected
2472 * @pdev: Pointer to PCI device
2473 * @state: The current pci connection state
2474 *
2475 * This function is called after a PCI bus error affecting
2476 * this device has been detected.
2477 */
2478static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2479                                                pci_channel_state_t state)
2480{
2481	struct net_device *netdev = pci_get_drvdata(pdev);
2482	struct igbvf_adapter *adapter = netdev_priv(netdev);
2483
2484	netif_device_detach(netdev);
2485
2486	if (state == pci_channel_io_perm_failure)
2487		return PCI_ERS_RESULT_DISCONNECT;
2488
2489	if (netif_running(netdev))
2490		igbvf_down(adapter);
2491	pci_disable_device(pdev);
2492
2493	/* Request a slot slot reset. */
2494	return PCI_ERS_RESULT_NEED_RESET;
2495}
2496
2497/**
2498 * igbvf_io_slot_reset - called after the pci bus has been reset.
2499 * @pdev: Pointer to PCI device
2500 *
2501 * Restart the card from scratch, as if from a cold-boot. Implementation
2502 * resembles the first-half of the igbvf_resume routine.
2503 */
2504static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2505{
2506	struct net_device *netdev = pci_get_drvdata(pdev);
2507	struct igbvf_adapter *adapter = netdev_priv(netdev);
2508
2509	if (pci_enable_device_mem(pdev)) {
2510		dev_err(&pdev->dev,
2511			"Cannot re-enable PCI device after reset.\n");
2512		return PCI_ERS_RESULT_DISCONNECT;
2513	}
2514	pci_set_master(pdev);
2515
2516	igbvf_reset(adapter);
2517
2518	return PCI_ERS_RESULT_RECOVERED;
2519}
2520
2521/**
2522 * igbvf_io_resume - called when traffic can start flowing again.
2523 * @pdev: Pointer to PCI device
2524 *
2525 * This callback is called when the error recovery driver tells us that
2526 * its OK to resume normal operation. Implementation resembles the
2527 * second-half of the igbvf_resume routine.
2528 */
2529static void igbvf_io_resume(struct pci_dev *pdev)
2530{
2531	struct net_device *netdev = pci_get_drvdata(pdev);
2532	struct igbvf_adapter *adapter = netdev_priv(netdev);
2533
2534	if (netif_running(netdev)) {
2535		if (igbvf_up(adapter)) {
2536			dev_err(&pdev->dev,
2537				"can't bring device back up after reset\n");
2538			return;
2539		}
2540	}
2541
2542	netif_device_attach(netdev);
2543}
2544
2545static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2546{
2547	struct e1000_hw *hw = &adapter->hw;
2548	struct net_device *netdev = adapter->netdev;
2549	struct pci_dev *pdev = adapter->pdev;
2550
2551	if (hw->mac.type == e1000_vfadapt_i350)
2552		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2553	else
2554		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2555	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2556}
2557
2558static int igbvf_set_features(struct net_device *netdev,
2559	netdev_features_t features)
2560{
2561	struct igbvf_adapter *adapter = netdev_priv(netdev);
2562
2563	if (features & NETIF_F_RXCSUM)
2564		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2565	else
2566		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2567
2568	return 0;
2569}
2570
2571static const struct net_device_ops igbvf_netdev_ops = {
2572	.ndo_open                       = igbvf_open,
2573	.ndo_stop                       = igbvf_close,
2574	.ndo_start_xmit                 = igbvf_xmit_frame,
2575	.ndo_get_stats                  = igbvf_get_stats,
2576	.ndo_set_rx_mode		= igbvf_set_multi,
2577	.ndo_set_mac_address            = igbvf_set_mac,
2578	.ndo_change_mtu                 = igbvf_change_mtu,
2579	.ndo_do_ioctl                   = igbvf_ioctl,
2580	.ndo_tx_timeout                 = igbvf_tx_timeout,
2581	.ndo_vlan_rx_add_vid            = igbvf_vlan_rx_add_vid,
2582	.ndo_vlan_rx_kill_vid           = igbvf_vlan_rx_kill_vid,
2583#ifdef CONFIG_NET_POLL_CONTROLLER
2584	.ndo_poll_controller            = igbvf_netpoll,
2585#endif
2586	.ndo_set_features               = igbvf_set_features,
2587};
2588
2589/**
2590 * igbvf_probe - Device Initialization Routine
2591 * @pdev: PCI device information struct
2592 * @ent: entry in igbvf_pci_tbl
2593 *
2594 * Returns 0 on success, negative on failure
2595 *
2596 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2597 * The OS initialization, configuring of the adapter private structure,
2598 * and a hardware reset occur.
2599 **/
2600static int __devinit igbvf_probe(struct pci_dev *pdev,
2601                                 const struct pci_device_id *ent)
2602{
2603	struct net_device *netdev;
2604	struct igbvf_adapter *adapter;
2605	struct e1000_hw *hw;
2606	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2607
2608	static int cards_found;
2609	int err, pci_using_dac;
2610
2611	err = pci_enable_device_mem(pdev);
2612	if (err)
2613		return err;
2614
2615	pci_using_dac = 0;
2616	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2617	if (!err) {
2618		err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2619		if (!err)
2620			pci_using_dac = 1;
2621	} else {
2622		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2623		if (err) {
2624			err = dma_set_coherent_mask(&pdev->dev,
2625						    DMA_BIT_MASK(32));
2626			if (err) {
2627				dev_err(&pdev->dev, "No usable DMA "
2628				        "configuration, aborting\n");
2629				goto err_dma;
2630			}
2631		}
2632	}
2633
2634	err = pci_request_regions(pdev, igbvf_driver_name);
2635	if (err)
2636		goto err_pci_reg;
2637
2638	pci_set_master(pdev);
2639
2640	err = -ENOMEM;
2641	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2642	if (!netdev)
2643		goto err_alloc_etherdev;
2644
2645	SET_NETDEV_DEV(netdev, &pdev->dev);
2646
2647	pci_set_drvdata(pdev, netdev);
2648	adapter = netdev_priv(netdev);
2649	hw = &adapter->hw;
2650	adapter->netdev = netdev;
2651	adapter->pdev = pdev;
2652	adapter->ei = ei;
2653	adapter->pba = ei->pba;
2654	adapter->flags = ei->flags;
2655	adapter->hw.back = adapter;
2656	adapter->hw.mac.type = ei->mac;
2657	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2658
2659	/* PCI config space info */
2660
2661	hw->vendor_id = pdev->vendor;
2662	hw->device_id = pdev->device;
2663	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2664	hw->subsystem_device_id = pdev->subsystem_device;
2665	hw->revision_id = pdev->revision;
2666
2667	err = -EIO;
2668	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2669	                              pci_resource_len(pdev, 0));
2670
2671	if (!adapter->hw.hw_addr)
2672		goto err_ioremap;
2673
2674	if (ei->get_variants) {
2675		err = ei->get_variants(adapter);
2676		if (err)
2677			goto err_ioremap;
2678	}
2679
2680	/* setup adapter struct */
2681	err = igbvf_sw_init(adapter);
2682	if (err)
2683		goto err_sw_init;
2684
2685	/* construct the net_device struct */
2686	netdev->netdev_ops = &igbvf_netdev_ops;
2687
2688	igbvf_set_ethtool_ops(netdev);
2689	netdev->watchdog_timeo = 5 * HZ;
2690	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2691
2692	adapter->bd_number = cards_found++;
2693
2694	netdev->hw_features = NETIF_F_SG |
2695	                   NETIF_F_IP_CSUM |
2696			   NETIF_F_IPV6_CSUM |
2697			   NETIF_F_TSO |
2698			   NETIF_F_TSO6 |
2699			   NETIF_F_RXCSUM;
2700
2701	netdev->features = netdev->hw_features |
2702	                   NETIF_F_HW_VLAN_TX |
2703	                   NETIF_F_HW_VLAN_RX |
2704	                   NETIF_F_HW_VLAN_FILTER;
2705
2706	if (pci_using_dac)
2707		netdev->features |= NETIF_F_HIGHDMA;
2708
2709	netdev->vlan_features |= NETIF_F_TSO;
2710	netdev->vlan_features |= NETIF_F_TSO6;
2711	netdev->vlan_features |= NETIF_F_IP_CSUM;
2712	netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2713	netdev->vlan_features |= NETIF_F_SG;
2714
2715	/*reset the controller to put the device in a known good state */
2716	err = hw->mac.ops.reset_hw(hw);
2717	if (err) {
2718		dev_info(&pdev->dev,
2719			 "PF still in reset state, assigning new address."
2720			 " Is the PF interface up?\n");
2721		eth_hw_addr_random(netdev);
2722		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2723			netdev->addr_len);
2724	} else {
2725		err = hw->mac.ops.read_mac_addr(hw);
2726		if (err) {
2727			dev_err(&pdev->dev, "Error reading MAC address\n");
2728			goto err_hw_init;
2729		}
2730		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2731			netdev->addr_len);
2732	}
2733
2734	if (!is_valid_ether_addr(netdev->dev_addr)) {
2735		dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2736		        netdev->dev_addr);
2737		err = -EIO;
2738		goto err_hw_init;
2739	}
2740
2741	memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2742
2743	setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2744	            (unsigned long) adapter);
2745
2746	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2747	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2748
2749	/* ring size defaults */
2750	adapter->rx_ring->count = 1024;
2751	adapter->tx_ring->count = 1024;
2752
2753	/* reset the hardware with the new settings */
2754	igbvf_reset(adapter);
2755
2756	strcpy(netdev->name, "eth%d");
2757	err = register_netdev(netdev);
2758	if (err)
2759		goto err_hw_init;
2760
2761	/* tell the stack to leave us alone until igbvf_open() is called */
2762	netif_carrier_off(netdev);
2763	netif_stop_queue(netdev);
2764
2765	igbvf_print_device_info(adapter);
2766
2767	igbvf_initialize_last_counter_stats(adapter);
2768
2769	return 0;
2770
2771err_hw_init:
2772	kfree(adapter->tx_ring);
2773	kfree(adapter->rx_ring);
2774err_sw_init:
2775	igbvf_reset_interrupt_capability(adapter);
2776	iounmap(adapter->hw.hw_addr);
2777err_ioremap:
2778	free_netdev(netdev);
2779err_alloc_etherdev:
2780	pci_release_regions(pdev);
2781err_pci_reg:
2782err_dma:
2783	pci_disable_device(pdev);
2784	return err;
2785}
2786
2787/**
2788 * igbvf_remove - Device Removal Routine
2789 * @pdev: PCI device information struct
2790 *
2791 * igbvf_remove is called by the PCI subsystem to alert the driver
2792 * that it should release a PCI device.  The could be caused by a
2793 * Hot-Plug event, or because the driver is going to be removed from
2794 * memory.
2795 **/
2796static void __devexit igbvf_remove(struct pci_dev *pdev)
2797{
2798	struct net_device *netdev = pci_get_drvdata(pdev);
2799	struct igbvf_adapter *adapter = netdev_priv(netdev);
2800	struct e1000_hw *hw = &adapter->hw;
2801
2802	/*
2803	 * The watchdog timer may be rescheduled, so explicitly
2804	 * disable it from being rescheduled.
2805	 */
2806	set_bit(__IGBVF_DOWN, &adapter->state);
2807	del_timer_sync(&adapter->watchdog_timer);
2808
2809	cancel_work_sync(&adapter->reset_task);
2810	cancel_work_sync(&adapter->watchdog_task);
2811
2812	unregister_netdev(netdev);
2813
2814	igbvf_reset_interrupt_capability(adapter);
2815
2816	/*
2817	 * it is important to delete the napi struct prior to freeing the
2818	 * rx ring so that you do not end up with null pointer refs
2819	 */
2820	netif_napi_del(&adapter->rx_ring->napi);
2821	kfree(adapter->tx_ring);
2822	kfree(adapter->rx_ring);
2823
2824	iounmap(hw->hw_addr);
2825	if (hw->flash_address)
2826		iounmap(hw->flash_address);
2827	pci_release_regions(pdev);
2828
2829	free_netdev(netdev);
2830
2831	pci_disable_device(pdev);
2832}
2833
2834/* PCI Error Recovery (ERS) */
2835static struct pci_error_handlers igbvf_err_handler = {
2836	.error_detected = igbvf_io_error_detected,
2837	.slot_reset = igbvf_io_slot_reset,
2838	.resume = igbvf_io_resume,
2839};
2840
2841static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2842	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2843	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2844	{ } /* terminate list */
2845};
2846MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2847
2848/* PCI Device API Driver */
2849static struct pci_driver igbvf_driver = {
2850	.name     = igbvf_driver_name,
2851	.id_table = igbvf_pci_tbl,
2852	.probe    = igbvf_probe,
2853	.remove   = __devexit_p(igbvf_remove),
2854#ifdef CONFIG_PM
2855	/* Power Management Hooks */
2856	.suspend  = igbvf_suspend,
2857	.resume   = igbvf_resume,
2858#endif
2859	.shutdown = igbvf_shutdown,
2860	.err_handler = &igbvf_err_handler
2861};
2862
2863/**
2864 * igbvf_init_module - Driver Registration Routine
2865 *
2866 * igbvf_init_module is the first routine called when the driver is
2867 * loaded. All it does is register with the PCI subsystem.
2868 **/
2869static int __init igbvf_init_module(void)
2870{
2871	int ret;
2872	pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2873	pr_info("%s\n", igbvf_copyright);
2874
2875	ret = pci_register_driver(&igbvf_driver);
2876
2877	return ret;
2878}
2879module_init(igbvf_init_module);
2880
2881/**
2882 * igbvf_exit_module - Driver Exit Cleanup Routine
2883 *
2884 * igbvf_exit_module is called just before the driver is removed
2885 * from memory.
2886 **/
2887static void __exit igbvf_exit_module(void)
2888{
2889	pci_unregister_driver(&igbvf_driver);
2890}
2891module_exit(igbvf_exit_module);
2892
2893
2894MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2895MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2896MODULE_LICENSE("GPL");
2897MODULE_VERSION(DRV_VERSION);
2898
2899/* netdev.c */
2900