1// SPDX-License-Identifier: GPL-2.0
2/* Copyright(c) 2009 - 2018 Intel Corporation. */
3
4#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
5
6#include <linux/module.h>
7#include <linux/types.h>
8#include <linux/init.h>
9#include <linux/pci.h>
10#include <linux/vmalloc.h>
11#include <linux/pagemap.h>
12#include <linux/delay.h>
13#include <linux/netdevice.h>
14#include <linux/tcp.h>
15#include <linux/ipv6.h>
16#include <linux/slab.h>
17#include <net/checksum.h>
18#include <net/ip6_checksum.h>
19#include <linux/mii.h>
20#include <linux/ethtool.h>
21#include <linux/if_vlan.h>
22#include <linux/prefetch.h>
23#include <linux/sctp.h>
24
25#include "igbvf.h"
26
27char igbvf_driver_name[] = "igbvf";
28static const char igbvf_driver_string[] =
29		  "Intel(R) Gigabit Virtual Function Network Driver";
30static const char igbvf_copyright[] =
31		  "Copyright (c) 2009 - 2012 Intel Corporation.";
32
33#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
34static int debug = -1;
35module_param(debug, int, 0);
36MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
37
38static int igbvf_poll(struct napi_struct *napi, int budget);
39static void igbvf_reset(struct igbvf_adapter *);
40static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
41static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
42
43static struct igbvf_info igbvf_vf_info = {
44	.mac		= e1000_vfadapt,
45	.flags		= 0,
46	.pba		= 10,
47	.init_ops	= e1000_init_function_pointers_vf,
48};
49
50static struct igbvf_info igbvf_i350_vf_info = {
51	.mac		= e1000_vfadapt_i350,
52	.flags		= 0,
53	.pba		= 10,
54	.init_ops	= e1000_init_function_pointers_vf,
55};
56
57static const struct igbvf_info *igbvf_info_tbl[] = {
58	[board_vf]	= &igbvf_vf_info,
59	[board_i350_vf]	= &igbvf_i350_vf_info,
60};
61
62/**
63 * igbvf_desc_unused - calculate if we have unused descriptors
64 * @ring: address of receive ring structure
65 **/
66static int igbvf_desc_unused(struct igbvf_ring *ring)
67{
68	if (ring->next_to_clean > ring->next_to_use)
69		return ring->next_to_clean - ring->next_to_use - 1;
70
71	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
72}
73
74/**
75 * igbvf_receive_skb - helper function to handle Rx indications
76 * @adapter: board private structure
77 * @netdev: pointer to netdev struct
78 * @skb: skb to indicate to stack
79 * @status: descriptor status field as written by hardware
80 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
81 * @skb: pointer to sk_buff to be indicated to stack
82 **/
83static void igbvf_receive_skb(struct igbvf_adapter *adapter,
84			      struct net_device *netdev,
85			      struct sk_buff *skb,
86			      u32 status, __le16 vlan)
87{
88	u16 vid;
89
90	if (status & E1000_RXD_STAT_VP) {
91		if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
92		    (status & E1000_RXDEXT_STATERR_LB))
93			vid = be16_to_cpu((__force __be16)vlan) & E1000_RXD_SPC_VLAN_MASK;
94		else
95			vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
96		if (test_bit(vid, adapter->active_vlans))
97			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
98	}
99
100	napi_gro_receive(&adapter->rx_ring->napi, skb);
101}
102
103static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
104					 u32 status_err, struct sk_buff *skb)
105{
106	skb_checksum_none_assert(skb);
107
108	/* Ignore Checksum bit is set or checksum is disabled through ethtool */
109	if ((status_err & E1000_RXD_STAT_IXSM) ||
110	    (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
111		return;
112
113	/* TCP/UDP checksum error bit is set */
114	if (status_err &
115	    (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
116		/* let the stack verify checksum errors */
117		adapter->hw_csum_err++;
118		return;
119	}
120
121	/* It must be a TCP or UDP packet with a valid checksum */
122	if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
123		skb->ip_summed = CHECKSUM_UNNECESSARY;
124
125	adapter->hw_csum_good++;
126}
127
128/**
129 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
130 * @rx_ring: address of ring structure to repopulate
131 * @cleaned_count: number of buffers to repopulate
132 **/
133static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
134				   int cleaned_count)
135{
136	struct igbvf_adapter *adapter = rx_ring->adapter;
137	struct net_device *netdev = adapter->netdev;
138	struct pci_dev *pdev = adapter->pdev;
139	union e1000_adv_rx_desc *rx_desc;
140	struct igbvf_buffer *buffer_info;
141	struct sk_buff *skb;
142	unsigned int i;
143	int bufsz;
144
145	i = rx_ring->next_to_use;
146	buffer_info = &rx_ring->buffer_info[i];
147
148	if (adapter->rx_ps_hdr_size)
149		bufsz = adapter->rx_ps_hdr_size;
150	else
151		bufsz = adapter->rx_buffer_len;
152
153	while (cleaned_count--) {
154		rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
155
156		if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
157			if (!buffer_info->page) {
158				buffer_info->page = alloc_page(GFP_ATOMIC);
159				if (!buffer_info->page) {
160					adapter->alloc_rx_buff_failed++;
161					goto no_buffers;
162				}
163				buffer_info->page_offset = 0;
164			} else {
165				buffer_info->page_offset ^= PAGE_SIZE / 2;
166			}
167			buffer_info->page_dma =
168				dma_map_page(&pdev->dev, buffer_info->page,
169					     buffer_info->page_offset,
170					     PAGE_SIZE / 2,
171					     DMA_FROM_DEVICE);
172			if (dma_mapping_error(&pdev->dev,
173					      buffer_info->page_dma)) {
174				__free_page(buffer_info->page);
175				buffer_info->page = NULL;
176				dev_err(&pdev->dev, "RX DMA map failed\n");
177				break;
178			}
179		}
180
181		if (!buffer_info->skb) {
182			skb = netdev_alloc_skb_ip_align(netdev, bufsz);
183			if (!skb) {
184				adapter->alloc_rx_buff_failed++;
185				goto no_buffers;
186			}
187
188			buffer_info->skb = skb;
189			buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
190							  bufsz,
191							  DMA_FROM_DEVICE);
192			if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
193				dev_kfree_skb(buffer_info->skb);
194				buffer_info->skb = NULL;
195				dev_err(&pdev->dev, "RX DMA map failed\n");
196				goto no_buffers;
197			}
198		}
199		/* Refresh the desc even if buffer_addrs didn't change because
200		 * each write-back erases this info.
201		 */
202		if (adapter->rx_ps_hdr_size) {
203			rx_desc->read.pkt_addr =
204			     cpu_to_le64(buffer_info->page_dma);
205			rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
206		} else {
207			rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
208			rx_desc->read.hdr_addr = 0;
209		}
210
211		i++;
212		if (i == rx_ring->count)
213			i = 0;
214		buffer_info = &rx_ring->buffer_info[i];
215	}
216
217no_buffers:
218	if (rx_ring->next_to_use != i) {
219		rx_ring->next_to_use = i;
220		if (i == 0)
221			i = (rx_ring->count - 1);
222		else
223			i--;
224
225		/* Force memory writes to complete before letting h/w
226		 * know there are new descriptors to fetch.  (Only
227		 * applicable for weak-ordered memory model archs,
228		 * such as IA-64).
229		*/
230		wmb();
231		writel(i, adapter->hw.hw_addr + rx_ring->tail);
232	}
233}
234
235/**
236 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
237 * @adapter: board private structure
238 * @work_done: output parameter used to indicate completed work
239 * @work_to_do: input parameter setting limit of work
240 *
241 * the return value indicates whether actual cleaning was done, there
242 * is no guarantee that everything was cleaned
243 **/
244static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
245			       int *work_done, int work_to_do)
246{
247	struct igbvf_ring *rx_ring = adapter->rx_ring;
248	struct net_device *netdev = adapter->netdev;
249	struct pci_dev *pdev = adapter->pdev;
250	union e1000_adv_rx_desc *rx_desc, *next_rxd;
251	struct igbvf_buffer *buffer_info, *next_buffer;
252	struct sk_buff *skb;
253	bool cleaned = false;
254	int cleaned_count = 0;
255	unsigned int total_bytes = 0, total_packets = 0;
256	unsigned int i;
257	u32 length, hlen, staterr;
258
259	i = rx_ring->next_to_clean;
260	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
261	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
262
263	while (staterr & E1000_RXD_STAT_DD) {
264		if (*work_done >= work_to_do)
265			break;
266		(*work_done)++;
267		rmb(); /* read descriptor and rx_buffer_info after status DD */
268
269		buffer_info = &rx_ring->buffer_info[i];
270
271		/* HW will not DMA in data larger than the given buffer, even
272		 * if it parses the (NFS, of course) header to be larger.  In
273		 * that case, it fills the header buffer and spills the rest
274		 * into the page.
275		 */
276		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
277		       & E1000_RXDADV_HDRBUFLEN_MASK) >>
278		       E1000_RXDADV_HDRBUFLEN_SHIFT;
279		if (hlen > adapter->rx_ps_hdr_size)
280			hlen = adapter->rx_ps_hdr_size;
281
282		length = le16_to_cpu(rx_desc->wb.upper.length);
283		cleaned = true;
284		cleaned_count++;
285
286		skb = buffer_info->skb;
287		prefetch(skb->data - NET_IP_ALIGN);
288		buffer_info->skb = NULL;
289		if (!adapter->rx_ps_hdr_size) {
290			dma_unmap_single(&pdev->dev, buffer_info->dma,
291					 adapter->rx_buffer_len,
292					 DMA_FROM_DEVICE);
293			buffer_info->dma = 0;
294			skb_put(skb, length);
295			goto send_up;
296		}
297
298		if (!skb_shinfo(skb)->nr_frags) {
299			dma_unmap_single(&pdev->dev, buffer_info->dma,
300					 adapter->rx_ps_hdr_size,
301					 DMA_FROM_DEVICE);
302			buffer_info->dma = 0;
303			skb_put(skb, hlen);
304		}
305
306		if (length) {
307			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
308				       PAGE_SIZE / 2,
309				       DMA_FROM_DEVICE);
310			buffer_info->page_dma = 0;
311
312			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
313					   buffer_info->page,
314					   buffer_info->page_offset,
315					   length);
316
317			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
318			    (page_count(buffer_info->page) != 1))
319				buffer_info->page = NULL;
320			else
321				get_page(buffer_info->page);
322
323			skb->len += length;
324			skb->data_len += length;
325			skb->truesize += PAGE_SIZE / 2;
326		}
327send_up:
328		i++;
329		if (i == rx_ring->count)
330			i = 0;
331		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
332		prefetch(next_rxd);
333		next_buffer = &rx_ring->buffer_info[i];
334
335		if (!(staterr & E1000_RXD_STAT_EOP)) {
336			buffer_info->skb = next_buffer->skb;
337			buffer_info->dma = next_buffer->dma;
338			next_buffer->skb = skb;
339			next_buffer->dma = 0;
340			goto next_desc;
341		}
342
343		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
344			dev_kfree_skb_irq(skb);
345			goto next_desc;
346		}
347
348		total_bytes += skb->len;
349		total_packets++;
350
351		igbvf_rx_checksum_adv(adapter, staterr, skb);
352
353		skb->protocol = eth_type_trans(skb, netdev);
354
355		igbvf_receive_skb(adapter, netdev, skb, staterr,
356				  rx_desc->wb.upper.vlan);
357
358next_desc:
359		rx_desc->wb.upper.status_error = 0;
360
361		/* return some buffers to hardware, one at a time is too slow */
362		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
363			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
364			cleaned_count = 0;
365		}
366
367		/* use prefetched values */
368		rx_desc = next_rxd;
369		buffer_info = next_buffer;
370
371		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
372	}
373
374	rx_ring->next_to_clean = i;
375	cleaned_count = igbvf_desc_unused(rx_ring);
376
377	if (cleaned_count)
378		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
379
380	adapter->total_rx_packets += total_packets;
381	adapter->total_rx_bytes += total_bytes;
382	netdev->stats.rx_bytes += total_bytes;
383	netdev->stats.rx_packets += total_packets;
384	return cleaned;
385}
386
387static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
388			    struct igbvf_buffer *buffer_info)
389{
390	if (buffer_info->dma) {
391		if (buffer_info->mapped_as_page)
392			dma_unmap_page(&adapter->pdev->dev,
393				       buffer_info->dma,
394				       buffer_info->length,
395				       DMA_TO_DEVICE);
396		else
397			dma_unmap_single(&adapter->pdev->dev,
398					 buffer_info->dma,
399					 buffer_info->length,
400					 DMA_TO_DEVICE);
401		buffer_info->dma = 0;
402	}
403	if (buffer_info->skb) {
404		dev_kfree_skb_any(buffer_info->skb);
405		buffer_info->skb = NULL;
406	}
407	buffer_info->time_stamp = 0;
408}
409
410/**
411 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
412 * @adapter: board private structure
413 * @tx_ring: ring being initialized
414 *
415 * Return 0 on success, negative on failure
416 **/
417int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
418			     struct igbvf_ring *tx_ring)
419{
420	struct pci_dev *pdev = adapter->pdev;
421	int size;
422
423	size = sizeof(struct igbvf_buffer) * tx_ring->count;
424	tx_ring->buffer_info = vzalloc(size);
425	if (!tx_ring->buffer_info)
426		goto err;
427
428	/* round up to nearest 4K */
429	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
430	tx_ring->size = ALIGN(tx_ring->size, 4096);
431
432	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
433					   &tx_ring->dma, GFP_KERNEL);
434	if (!tx_ring->desc)
435		goto err;
436
437	tx_ring->adapter = adapter;
438	tx_ring->next_to_use = 0;
439	tx_ring->next_to_clean = 0;
440
441	return 0;
442err:
443	vfree(tx_ring->buffer_info);
444	dev_err(&adapter->pdev->dev,
445		"Unable to allocate memory for the transmit descriptor ring\n");
446	return -ENOMEM;
447}
448
449/**
450 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
451 * @adapter: board private structure
452 * @rx_ring: ring being initialized
453 *
454 * Returns 0 on success, negative on failure
455 **/
456int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
457			     struct igbvf_ring *rx_ring)
458{
459	struct pci_dev *pdev = adapter->pdev;
460	int size, desc_len;
461
462	size = sizeof(struct igbvf_buffer) * rx_ring->count;
463	rx_ring->buffer_info = vzalloc(size);
464	if (!rx_ring->buffer_info)
465		goto err;
466
467	desc_len = sizeof(union e1000_adv_rx_desc);
468
469	/* Round up to nearest 4K */
470	rx_ring->size = rx_ring->count * desc_len;
471	rx_ring->size = ALIGN(rx_ring->size, 4096);
472
473	rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
474					   &rx_ring->dma, GFP_KERNEL);
475	if (!rx_ring->desc)
476		goto err;
477
478	rx_ring->next_to_clean = 0;
479	rx_ring->next_to_use = 0;
480
481	rx_ring->adapter = adapter;
482
483	return 0;
484
485err:
486	vfree(rx_ring->buffer_info);
487	rx_ring->buffer_info = NULL;
488	dev_err(&adapter->pdev->dev,
489		"Unable to allocate memory for the receive descriptor ring\n");
490	return -ENOMEM;
491}
492
493/**
494 * igbvf_clean_tx_ring - Free Tx Buffers
495 * @tx_ring: ring to be cleaned
496 **/
497static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
498{
499	struct igbvf_adapter *adapter = tx_ring->adapter;
500	struct igbvf_buffer *buffer_info;
501	unsigned long size;
502	unsigned int i;
503
504	if (!tx_ring->buffer_info)
505		return;
506
507	/* Free all the Tx ring sk_buffs */
508	for (i = 0; i < tx_ring->count; i++) {
509		buffer_info = &tx_ring->buffer_info[i];
510		igbvf_put_txbuf(adapter, buffer_info);
511	}
512
513	size = sizeof(struct igbvf_buffer) * tx_ring->count;
514	memset(tx_ring->buffer_info, 0, size);
515
516	/* Zero out the descriptor ring */
517	memset(tx_ring->desc, 0, tx_ring->size);
518
519	tx_ring->next_to_use = 0;
520	tx_ring->next_to_clean = 0;
521
522	writel(0, adapter->hw.hw_addr + tx_ring->head);
523	writel(0, adapter->hw.hw_addr + tx_ring->tail);
524}
525
526/**
527 * igbvf_free_tx_resources - Free Tx Resources per Queue
528 * @tx_ring: ring to free resources from
529 *
530 * Free all transmit software resources
531 **/
532void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
533{
534	struct pci_dev *pdev = tx_ring->adapter->pdev;
535
536	igbvf_clean_tx_ring(tx_ring);
537
538	vfree(tx_ring->buffer_info);
539	tx_ring->buffer_info = NULL;
540
541	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
542			  tx_ring->dma);
543
544	tx_ring->desc = NULL;
545}
546
547/**
548 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
549 * @rx_ring: ring structure pointer to free buffers from
550 **/
551static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
552{
553	struct igbvf_adapter *adapter = rx_ring->adapter;
554	struct igbvf_buffer *buffer_info;
555	struct pci_dev *pdev = adapter->pdev;
556	unsigned long size;
557	unsigned int i;
558
559	if (!rx_ring->buffer_info)
560		return;
561
562	/* Free all the Rx ring sk_buffs */
563	for (i = 0; i < rx_ring->count; i++) {
564		buffer_info = &rx_ring->buffer_info[i];
565		if (buffer_info->dma) {
566			if (adapter->rx_ps_hdr_size) {
567				dma_unmap_single(&pdev->dev, buffer_info->dma,
568						 adapter->rx_ps_hdr_size,
569						 DMA_FROM_DEVICE);
570			} else {
571				dma_unmap_single(&pdev->dev, buffer_info->dma,
572						 adapter->rx_buffer_len,
573						 DMA_FROM_DEVICE);
574			}
575			buffer_info->dma = 0;
576		}
577
578		if (buffer_info->skb) {
579			dev_kfree_skb(buffer_info->skb);
580			buffer_info->skb = NULL;
581		}
582
583		if (buffer_info->page) {
584			if (buffer_info->page_dma)
585				dma_unmap_page(&pdev->dev,
586					       buffer_info->page_dma,
587					       PAGE_SIZE / 2,
588					       DMA_FROM_DEVICE);
589			put_page(buffer_info->page);
590			buffer_info->page = NULL;
591			buffer_info->page_dma = 0;
592			buffer_info->page_offset = 0;
593		}
594	}
595
596	size = sizeof(struct igbvf_buffer) * rx_ring->count;
597	memset(rx_ring->buffer_info, 0, size);
598
599	/* Zero out the descriptor ring */
600	memset(rx_ring->desc, 0, rx_ring->size);
601
602	rx_ring->next_to_clean = 0;
603	rx_ring->next_to_use = 0;
604
605	writel(0, adapter->hw.hw_addr + rx_ring->head);
606	writel(0, adapter->hw.hw_addr + rx_ring->tail);
607}
608
609/**
610 * igbvf_free_rx_resources - Free Rx Resources
611 * @rx_ring: ring to clean the resources from
612 *
613 * Free all receive software resources
614 **/
615
616void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
617{
618	struct pci_dev *pdev = rx_ring->adapter->pdev;
619
620	igbvf_clean_rx_ring(rx_ring);
621
622	vfree(rx_ring->buffer_info);
623	rx_ring->buffer_info = NULL;
624
625	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
626			  rx_ring->dma);
627	rx_ring->desc = NULL;
628}
629
630/**
631 * igbvf_update_itr - update the dynamic ITR value based on statistics
632 * @adapter: pointer to adapter
633 * @itr_setting: current adapter->itr
634 * @packets: the number of packets during this measurement interval
635 * @bytes: the number of bytes during this measurement interval
636 *
637 * Stores a new ITR value based on packets and byte counts during the last
638 * interrupt.  The advantage of per interrupt computation is faster updates
639 * and more accurate ITR for the current traffic pattern.  Constants in this
640 * function were computed based on theoretical maximum wire speed and thresholds
641 * were set based on testing data as well as attempting to minimize response
642 * time while increasing bulk throughput.
643 **/
644static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
645					   enum latency_range itr_setting,
646					   int packets, int bytes)
647{
648	enum latency_range retval = itr_setting;
649
650	if (packets == 0)
651		goto update_itr_done;
652
653	switch (itr_setting) {
654	case lowest_latency:
655		/* handle TSO and jumbo frames */
656		if (bytes/packets > 8000)
657			retval = bulk_latency;
658		else if ((packets < 5) && (bytes > 512))
659			retval = low_latency;
660		break;
661	case low_latency:  /* 50 usec aka 20000 ints/s */
662		if (bytes > 10000) {
663			/* this if handles the TSO accounting */
664			if (bytes/packets > 8000)
665				retval = bulk_latency;
666			else if ((packets < 10) || ((bytes/packets) > 1200))
667				retval = bulk_latency;
668			else if ((packets > 35))
669				retval = lowest_latency;
670		} else if (bytes/packets > 2000) {
671			retval = bulk_latency;
672		} else if (packets <= 2 && bytes < 512) {
673			retval = lowest_latency;
674		}
675		break;
676	case bulk_latency: /* 250 usec aka 4000 ints/s */
677		if (bytes > 25000) {
678			if (packets > 35)
679				retval = low_latency;
680		} else if (bytes < 6000) {
681			retval = low_latency;
682		}
683		break;
684	default:
685		break;
686	}
687
688update_itr_done:
689	return retval;
690}
691
692static int igbvf_range_to_itr(enum latency_range current_range)
693{
694	int new_itr;
695
696	switch (current_range) {
697	/* counts and packets in update_itr are dependent on these numbers */
698	case lowest_latency:
699		new_itr = IGBVF_70K_ITR;
700		break;
701	case low_latency:
702		new_itr = IGBVF_20K_ITR;
703		break;
704	case bulk_latency:
705		new_itr = IGBVF_4K_ITR;
706		break;
707	default:
708		new_itr = IGBVF_START_ITR;
709		break;
710	}
711	return new_itr;
712}
713
714static void igbvf_set_itr(struct igbvf_adapter *adapter)
715{
716	u32 new_itr;
717
718	adapter->tx_ring->itr_range =
719			igbvf_update_itr(adapter,
720					 adapter->tx_ring->itr_val,
721					 adapter->total_tx_packets,
722					 adapter->total_tx_bytes);
723
724	/* conservative mode (itr 3) eliminates the lowest_latency setting */
725	if (adapter->requested_itr == 3 &&
726	    adapter->tx_ring->itr_range == lowest_latency)
727		adapter->tx_ring->itr_range = low_latency;
728
729	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
730
731	if (new_itr != adapter->tx_ring->itr_val) {
732		u32 current_itr = adapter->tx_ring->itr_val;
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
758		new_itr = new_itr > current_itr ?
759			  min(current_itr + (new_itr >> 2), new_itr) :
760			  new_itr;
761		adapter->rx_ring->itr_val = new_itr;
762
763		adapter->rx_ring->set_itr = 1;
764	}
765}
766
767/**
768 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
769 * @tx_ring: ring structure to clean descriptors from
770 *
771 * returns true if ring is completely cleaned
772 **/
773static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
774{
775	struct igbvf_adapter *adapter = tx_ring->adapter;
776	struct net_device *netdev = adapter->netdev;
777	struct igbvf_buffer *buffer_info;
778	struct sk_buff *skb;
779	union e1000_adv_tx_desc *tx_desc, *eop_desc;
780	unsigned int total_bytes = 0, total_packets = 0;
781	unsigned int i, count = 0;
782	bool cleaned = false;
783
784	i = tx_ring->next_to_clean;
785	buffer_info = &tx_ring->buffer_info[i];
786	eop_desc = buffer_info->next_to_watch;
787
788	do {
789		/* if next_to_watch is not set then there is no work pending */
790		if (!eop_desc)
791			break;
792
793		/* prevent any other reads prior to eop_desc */
794		smp_rmb();
795
796		/* if DD is not set pending work has not been completed */
797		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
798			break;
799
800		/* clear next_to_watch to prevent false hangs */
801		buffer_info->next_to_watch = NULL;
802
803		for (cleaned = false; !cleaned; count++) {
804			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
805			cleaned = (tx_desc == eop_desc);
806			skb = buffer_info->skb;
807
808			if (skb) {
809				unsigned int segs, bytecount;
810
811				/* gso_segs is currently only valid for tcp */
812				segs = skb_shinfo(skb)->gso_segs ?: 1;
813				/* multiply data chunks by size of headers */
814				bytecount = ((segs - 1) * skb_headlen(skb)) +
815					    skb->len;
816				total_packets += segs;
817				total_bytes += bytecount;
818			}
819
820			igbvf_put_txbuf(adapter, buffer_info);
821			tx_desc->wb.status = 0;
822
823			i++;
824			if (i == tx_ring->count)
825				i = 0;
826
827			buffer_info = &tx_ring->buffer_info[i];
828		}
829
830		eop_desc = buffer_info->next_to_watch;
831	} while (count < tx_ring->count);
832
833	tx_ring->next_to_clean = i;
834
835	if (unlikely(count && netif_carrier_ok(netdev) &&
836	    igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
837		/* Make sure that anybody stopping the queue after this
838		 * sees the new next_to_clean.
839		 */
840		smp_mb();
841		if (netif_queue_stopped(netdev) &&
842		    !(test_bit(__IGBVF_DOWN, &adapter->state))) {
843			netif_wake_queue(netdev);
844			++adapter->restart_queue;
845		}
846	}
847
848	netdev->stats.tx_bytes += total_bytes;
849	netdev->stats.tx_packets += total_packets;
850	return count < tx_ring->count;
851}
852
853static irqreturn_t igbvf_msix_other(int irq, void *data)
854{
855	struct net_device *netdev = data;
856	struct igbvf_adapter *adapter = netdev_priv(netdev);
857	struct e1000_hw *hw = &adapter->hw;
858
859	adapter->int_counter1++;
860
861	hw->mac.get_link_status = 1;
862	if (!test_bit(__IGBVF_DOWN, &adapter->state))
863		mod_timer(&adapter->watchdog_timer, jiffies + 1);
864
865	ew32(EIMS, adapter->eims_other);
866
867	return IRQ_HANDLED;
868}
869
870static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
871{
872	struct net_device *netdev = data;
873	struct igbvf_adapter *adapter = netdev_priv(netdev);
874	struct e1000_hw *hw = &adapter->hw;
875	struct igbvf_ring *tx_ring = adapter->tx_ring;
876
877	if (tx_ring->set_itr) {
878		writel(tx_ring->itr_val,
879		       adapter->hw.hw_addr + tx_ring->itr_register);
880		adapter->tx_ring->set_itr = 0;
881	}
882
883	adapter->total_tx_bytes = 0;
884	adapter->total_tx_packets = 0;
885
886	/* auto mask will automatically re-enable the interrupt when we write
887	 * EICS
888	 */
889	if (!igbvf_clean_tx_irq(tx_ring))
890		/* Ring was not completely cleaned, so fire another interrupt */
891		ew32(EICS, tx_ring->eims_value);
892	else
893		ew32(EIMS, tx_ring->eims_value);
894
895	return IRQ_HANDLED;
896}
897
898static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
899{
900	struct net_device *netdev = data;
901	struct igbvf_adapter *adapter = netdev_priv(netdev);
902
903	adapter->int_counter0++;
904
905	/* Write the ITR value calculated at the end of the
906	 * previous interrupt.
907	 */
908	if (adapter->rx_ring->set_itr) {
909		writel(adapter->rx_ring->itr_val,
910		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
911		adapter->rx_ring->set_itr = 0;
912	}
913
914	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
915		adapter->total_rx_bytes = 0;
916		adapter->total_rx_packets = 0;
917		__napi_schedule(&adapter->rx_ring->napi);
918	}
919
920	return IRQ_HANDLED;
921}
922
923#define IGBVF_NO_QUEUE -1
924
925static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
926				int tx_queue, int msix_vector)
927{
928	struct e1000_hw *hw = &adapter->hw;
929	u32 ivar, index;
930
931	/* 82576 uses a table-based method for assigning vectors.
932	 * Each queue has a single entry in the table to which we write
933	 * a vector number along with a "valid" bit.  Sadly, the layout
934	 * of the table is somewhat counterintuitive.
935	 */
936	if (rx_queue > IGBVF_NO_QUEUE) {
937		index = (rx_queue >> 1);
938		ivar = array_er32(IVAR0, index);
939		if (rx_queue & 0x1) {
940			/* vector goes into third byte of register */
941			ivar = ivar & 0xFF00FFFF;
942			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
943		} else {
944			/* vector goes into low byte of register */
945			ivar = ivar & 0xFFFFFF00;
946			ivar |= msix_vector | E1000_IVAR_VALID;
947		}
948		adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
949		array_ew32(IVAR0, index, ivar);
950	}
951	if (tx_queue > IGBVF_NO_QUEUE) {
952		index = (tx_queue >> 1);
953		ivar = array_er32(IVAR0, index);
954		if (tx_queue & 0x1) {
955			/* vector goes into high byte of register */
956			ivar = ivar & 0x00FFFFFF;
957			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
958		} else {
959			/* vector goes into second byte of register */
960			ivar = ivar & 0xFFFF00FF;
961			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
962		}
963		adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
964		array_ew32(IVAR0, index, ivar);
965	}
966}
967
968/**
969 * igbvf_configure_msix - Configure MSI-X hardware
970 * @adapter: board private structure
971 *
972 * igbvf_configure_msix sets up the hardware to properly
973 * generate MSI-X interrupts.
974 **/
975static void igbvf_configure_msix(struct igbvf_adapter *adapter)
976{
977	u32 tmp;
978	struct e1000_hw *hw = &adapter->hw;
979	struct igbvf_ring *tx_ring = adapter->tx_ring;
980	struct igbvf_ring *rx_ring = adapter->rx_ring;
981	int vector = 0;
982
983	adapter->eims_enable_mask = 0;
984
985	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
986	adapter->eims_enable_mask |= tx_ring->eims_value;
987	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
988	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
989	adapter->eims_enable_mask |= rx_ring->eims_value;
990	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
991
992	/* set vector for other causes, i.e. link changes */
993
994	tmp = (vector++ | E1000_IVAR_VALID);
995
996	ew32(IVAR_MISC, tmp);
997
998	adapter->eims_enable_mask = GENMASK(vector - 1, 0);
999	adapter->eims_other = BIT(vector - 1);
1000	e1e_flush();
1001}
1002
1003static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1004{
1005	if (adapter->msix_entries) {
1006		pci_disable_msix(adapter->pdev);
1007		kfree(adapter->msix_entries);
1008		adapter->msix_entries = NULL;
1009	}
1010}
1011
1012/**
1013 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1014 * @adapter: board private structure
1015 *
1016 * Attempt to configure interrupts using the best available
1017 * capabilities of the hardware and kernel.
1018 **/
1019static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1020{
1021	int err = -ENOMEM;
1022	int i;
1023
1024	/* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1025	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1026					GFP_KERNEL);
1027	if (adapter->msix_entries) {
1028		for (i = 0; i < 3; i++)
1029			adapter->msix_entries[i].entry = i;
1030
1031		err = pci_enable_msix_range(adapter->pdev,
1032					    adapter->msix_entries, 3, 3);
1033	}
1034
1035	if (err < 0) {
1036		/* MSI-X failed */
1037		dev_err(&adapter->pdev->dev,
1038			"Failed to initialize MSI-X interrupts.\n");
1039		igbvf_reset_interrupt_capability(adapter);
1040	}
1041}
1042
1043/**
1044 * igbvf_request_msix - Initialize MSI-X interrupts
1045 * @adapter: board private structure
1046 *
1047 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1048 * kernel.
1049 **/
1050static int igbvf_request_msix(struct igbvf_adapter *adapter)
1051{
1052	struct net_device *netdev = adapter->netdev;
1053	int err = 0, vector = 0;
1054
1055	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1056		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1057		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1058	} else {
1059		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1060		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1061	}
1062
1063	err = request_irq(adapter->msix_entries[vector].vector,
1064			  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1065			  netdev);
1066	if (err)
1067		goto out;
1068
1069	adapter->tx_ring->itr_register = E1000_EITR(vector);
1070	adapter->tx_ring->itr_val = adapter->current_itr;
1071	vector++;
1072
1073	err = request_irq(adapter->msix_entries[vector].vector,
1074			  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1075			  netdev);
1076	if (err)
1077		goto free_irq_tx;
1078
1079	adapter->rx_ring->itr_register = E1000_EITR(vector);
1080	adapter->rx_ring->itr_val = adapter->current_itr;
1081	vector++;
1082
1083	err = request_irq(adapter->msix_entries[vector].vector,
1084			  igbvf_msix_other, 0, netdev->name, netdev);
1085	if (err)
1086		goto free_irq_rx;
1087
1088	igbvf_configure_msix(adapter);
1089	return 0;
1090free_irq_rx:
1091	free_irq(adapter->msix_entries[--vector].vector, netdev);
1092free_irq_tx:
1093	free_irq(adapter->msix_entries[--vector].vector, netdev);
1094out:
1095	return err;
1096}
1097
1098/**
1099 * igbvf_alloc_queues - Allocate memory for all rings
1100 * @adapter: board private structure to initialize
1101 **/
1102static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1103{
1104	struct net_device *netdev = adapter->netdev;
1105
1106	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1107	if (!adapter->tx_ring)
1108		return -ENOMEM;
1109
1110	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1111	if (!adapter->rx_ring) {
1112		kfree(adapter->tx_ring);
1113		return -ENOMEM;
1114	}
1115
1116	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1117
1118	return 0;
1119}
1120
1121/**
1122 * igbvf_request_irq - initialize interrupts
1123 * @adapter: board private structure
1124 *
1125 * Attempts to configure interrupts using the best available
1126 * capabilities of the hardware and kernel.
1127 **/
1128static int igbvf_request_irq(struct igbvf_adapter *adapter)
1129{
1130	int err = -1;
1131
1132	/* igbvf supports msi-x only */
1133	if (adapter->msix_entries)
1134		err = igbvf_request_msix(adapter);
1135
1136	if (!err)
1137		return err;
1138
1139	dev_err(&adapter->pdev->dev,
1140		"Unable to allocate interrupt, Error: %d\n", err);
1141
1142	return err;
1143}
1144
1145static void igbvf_free_irq(struct igbvf_adapter *adapter)
1146{
1147	struct net_device *netdev = adapter->netdev;
1148	int vector;
1149
1150	if (adapter->msix_entries) {
1151		for (vector = 0; vector < 3; vector++)
1152			free_irq(adapter->msix_entries[vector].vector, netdev);
1153	}
1154}
1155
1156/**
1157 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1158 * @adapter: board private structure
1159 **/
1160static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1161{
1162	struct e1000_hw *hw = &adapter->hw;
1163
1164	ew32(EIMC, ~0);
1165
1166	if (adapter->msix_entries)
1167		ew32(EIAC, 0);
1168}
1169
1170/**
1171 * igbvf_irq_enable - Enable default interrupt generation settings
1172 * @adapter: board private structure
1173 **/
1174static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1175{
1176	struct e1000_hw *hw = &adapter->hw;
1177
1178	ew32(EIAC, adapter->eims_enable_mask);
1179	ew32(EIAM, adapter->eims_enable_mask);
1180	ew32(EIMS, adapter->eims_enable_mask);
1181}
1182
1183/**
1184 * igbvf_poll - NAPI Rx polling callback
1185 * @napi: struct associated with this polling callback
1186 * @budget: amount of packets driver is allowed to process this poll
1187 **/
1188static int igbvf_poll(struct napi_struct *napi, int budget)
1189{
1190	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1191	struct igbvf_adapter *adapter = rx_ring->adapter;
1192	struct e1000_hw *hw = &adapter->hw;
1193	int work_done = 0;
1194
1195	igbvf_clean_rx_irq(adapter, &work_done, budget);
1196
1197	if (work_done == budget)
1198		return budget;
1199
1200	/* Exit the polling mode, but don't re-enable interrupts if stack might
1201	 * poll us due to busy-polling
1202	 */
1203	if (likely(napi_complete_done(napi, work_done))) {
1204		if (adapter->requested_itr & 3)
1205			igbvf_set_itr(adapter);
1206
1207		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1208			ew32(EIMS, adapter->rx_ring->eims_value);
1209	}
1210
1211	return work_done;
1212}
1213
1214/**
1215 * igbvf_set_rlpml - set receive large packet maximum length
1216 * @adapter: board private structure
1217 *
1218 * Configure the maximum size of packets that will be received
1219 */
1220static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1221{
1222	int max_frame_size;
1223	struct e1000_hw *hw = &adapter->hw;
1224
1225	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1226
1227	spin_lock_bh(&hw->mbx_lock);
1228
1229	e1000_rlpml_set_vf(hw, max_frame_size);
1230
1231	spin_unlock_bh(&hw->mbx_lock);
1232}
1233
1234static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1235				 __be16 proto, u16 vid)
1236{
1237	struct igbvf_adapter *adapter = netdev_priv(netdev);
1238	struct e1000_hw *hw = &adapter->hw;
1239
1240	spin_lock_bh(&hw->mbx_lock);
1241
1242	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1243		dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1244		spin_unlock_bh(&hw->mbx_lock);
1245		return -EINVAL;
1246	}
1247
1248	spin_unlock_bh(&hw->mbx_lock);
1249
1250	set_bit(vid, adapter->active_vlans);
1251	return 0;
1252}
1253
1254static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1255				  __be16 proto, u16 vid)
1256{
1257	struct igbvf_adapter *adapter = netdev_priv(netdev);
1258	struct e1000_hw *hw = &adapter->hw;
1259
1260	spin_lock_bh(&hw->mbx_lock);
1261
1262	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1263		dev_err(&adapter->pdev->dev,
1264			"Failed to remove vlan id %d\n", vid);
1265		spin_unlock_bh(&hw->mbx_lock);
1266		return -EINVAL;
1267	}
1268
1269	spin_unlock_bh(&hw->mbx_lock);
1270
1271	clear_bit(vid, adapter->active_vlans);
1272	return 0;
1273}
1274
1275static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1276{
1277	u16 vid;
1278
1279	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1280		igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1281}
1282
1283/**
1284 * igbvf_configure_tx - Configure Transmit Unit after Reset
1285 * @adapter: board private structure
1286 *
1287 * Configure the Tx unit of the MAC after a reset.
1288 **/
1289static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1290{
1291	struct e1000_hw *hw = &adapter->hw;
1292	struct igbvf_ring *tx_ring = adapter->tx_ring;
1293	u64 tdba;
1294	u32 txdctl, dca_txctrl;
1295
1296	/* disable transmits */
1297	txdctl = er32(TXDCTL(0));
1298	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1299	e1e_flush();
1300	msleep(10);
1301
1302	/* Setup the HW Tx Head and Tail descriptor pointers */
1303	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1304	tdba = tx_ring->dma;
1305	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1306	ew32(TDBAH(0), (tdba >> 32));
1307	ew32(TDH(0), 0);
1308	ew32(TDT(0), 0);
1309	tx_ring->head = E1000_TDH(0);
1310	tx_ring->tail = E1000_TDT(0);
1311
1312	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1313	 * MUST be delivered in order or it will completely screw up
1314	 * our bookkeeping.
1315	 */
1316	dca_txctrl = er32(DCA_TXCTRL(0));
1317	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1318	ew32(DCA_TXCTRL(0), dca_txctrl);
1319
1320	/* enable transmits */
1321	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1322	ew32(TXDCTL(0), txdctl);
1323
1324	/* Setup Transmit Descriptor Settings for eop descriptor */
1325	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1326
1327	/* enable Report Status bit */
1328	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1329}
1330
1331/**
1332 * igbvf_setup_srrctl - configure the receive control registers
1333 * @adapter: Board private structure
1334 **/
1335static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1336{
1337	struct e1000_hw *hw = &adapter->hw;
1338	u32 srrctl = 0;
1339
1340	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1341		    E1000_SRRCTL_BSIZEHDR_MASK |
1342		    E1000_SRRCTL_BSIZEPKT_MASK);
1343
1344	/* Enable queue drop to avoid head of line blocking */
1345	srrctl |= E1000_SRRCTL_DROP_EN;
1346
1347	/* Setup buffer sizes */
1348	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1349		  E1000_SRRCTL_BSIZEPKT_SHIFT;
1350
1351	if (adapter->rx_buffer_len < 2048) {
1352		adapter->rx_ps_hdr_size = 0;
1353		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1354	} else {
1355		adapter->rx_ps_hdr_size = 128;
1356		srrctl |= adapter->rx_ps_hdr_size <<
1357			  E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1358		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1359	}
1360
1361	ew32(SRRCTL(0), srrctl);
1362}
1363
1364/**
1365 * igbvf_configure_rx - Configure Receive Unit after Reset
1366 * @adapter: board private structure
1367 *
1368 * Configure the Rx unit of the MAC after a reset.
1369 **/
1370static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1371{
1372	struct e1000_hw *hw = &adapter->hw;
1373	struct igbvf_ring *rx_ring = adapter->rx_ring;
1374	u64 rdba;
1375	u32 rxdctl;
1376
1377	/* disable receives */
1378	rxdctl = er32(RXDCTL(0));
1379	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1380	e1e_flush();
1381	msleep(10);
1382
1383	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1384	 * the Base and Length of the Rx Descriptor Ring
1385	 */
1386	rdba = rx_ring->dma;
1387	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1388	ew32(RDBAH(0), (rdba >> 32));
1389	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1390	rx_ring->head = E1000_RDH(0);
1391	rx_ring->tail = E1000_RDT(0);
1392	ew32(RDH(0), 0);
1393	ew32(RDT(0), 0);
1394
1395	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1396	rxdctl &= 0xFFF00000;
1397	rxdctl |= IGBVF_RX_PTHRESH;
1398	rxdctl |= IGBVF_RX_HTHRESH << 8;
1399	rxdctl |= IGBVF_RX_WTHRESH << 16;
1400
1401	igbvf_set_rlpml(adapter);
1402
1403	/* enable receives */
1404	ew32(RXDCTL(0), rxdctl);
1405}
1406
1407/**
1408 * igbvf_set_multi - Multicast and Promiscuous mode set
1409 * @netdev: network interface device structure
1410 *
1411 * The set_multi entry point is called whenever the multicast address
1412 * list or the network interface flags are updated.  This routine is
1413 * responsible for configuring the hardware for proper multicast,
1414 * promiscuous mode, and all-multi behavior.
1415 **/
1416static void igbvf_set_multi(struct net_device *netdev)
1417{
1418	struct igbvf_adapter *adapter = netdev_priv(netdev);
1419	struct e1000_hw *hw = &adapter->hw;
1420	struct netdev_hw_addr *ha;
1421	u8  *mta_list = NULL;
1422	int i;
1423
1424	if (!netdev_mc_empty(netdev)) {
1425		mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1426					 GFP_ATOMIC);
1427		if (!mta_list)
1428			return;
1429	}
1430
1431	/* prepare a packed array of only addresses. */
1432	i = 0;
1433	netdev_for_each_mc_addr(ha, netdev)
1434		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1435
1436	spin_lock_bh(&hw->mbx_lock);
1437
1438	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1439
1440	spin_unlock_bh(&hw->mbx_lock);
1441	kfree(mta_list);
1442}
1443
1444/**
1445 * igbvf_set_uni - Configure unicast MAC filters
1446 * @netdev: network interface device structure
1447 *
1448 * This routine is responsible for configuring the hardware for proper
1449 * unicast filters.
1450 **/
1451static int igbvf_set_uni(struct net_device *netdev)
1452{
1453	struct igbvf_adapter *adapter = netdev_priv(netdev);
1454	struct e1000_hw *hw = &adapter->hw;
1455
1456	if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1457		pr_err("Too many unicast filters - No Space\n");
1458		return -ENOSPC;
1459	}
1460
1461	spin_lock_bh(&hw->mbx_lock);
1462
1463	/* Clear all unicast MAC filters */
1464	hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1465
1466	spin_unlock_bh(&hw->mbx_lock);
1467
1468	if (!netdev_uc_empty(netdev)) {
1469		struct netdev_hw_addr *ha;
1470
1471		/* Add MAC filters one by one */
1472		netdev_for_each_uc_addr(ha, netdev) {
1473			spin_lock_bh(&hw->mbx_lock);
1474
1475			hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1476						ha->addr);
1477
1478			spin_unlock_bh(&hw->mbx_lock);
1479			udelay(200);
1480		}
1481	}
1482
1483	return 0;
1484}
1485
1486static void igbvf_set_rx_mode(struct net_device *netdev)
1487{
1488	igbvf_set_multi(netdev);
1489	igbvf_set_uni(netdev);
1490}
1491
1492/**
1493 * igbvf_configure - configure the hardware for Rx and Tx
1494 * @adapter: private board structure
1495 **/
1496static void igbvf_configure(struct igbvf_adapter *adapter)
1497{
1498	igbvf_set_rx_mode(adapter->netdev);
1499
1500	igbvf_restore_vlan(adapter);
1501
1502	igbvf_configure_tx(adapter);
1503	igbvf_setup_srrctl(adapter);
1504	igbvf_configure_rx(adapter);
1505	igbvf_alloc_rx_buffers(adapter->rx_ring,
1506			       igbvf_desc_unused(adapter->rx_ring));
1507}
1508
1509/* igbvf_reset - bring the hardware into a known good state
1510 * @adapter: private board structure
1511 *
1512 * This function boots the hardware and enables some settings that
1513 * require a configuration cycle of the hardware - those cannot be
1514 * set/changed during runtime. After reset the device needs to be
1515 * properly configured for Rx, Tx etc.
1516 */
1517static void igbvf_reset(struct igbvf_adapter *adapter)
1518{
1519	struct e1000_mac_info *mac = &adapter->hw.mac;
1520	struct net_device *netdev = adapter->netdev;
1521	struct e1000_hw *hw = &adapter->hw;
1522
1523	spin_lock_bh(&hw->mbx_lock);
1524
1525	/* Allow time for pending master requests to run */
1526	if (mac->ops.reset_hw(hw))
1527		dev_err(&adapter->pdev->dev, "PF still resetting\n");
1528
1529	mac->ops.init_hw(hw);
1530
1531	spin_unlock_bh(&hw->mbx_lock);
1532
1533	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1534		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1535		       netdev->addr_len);
1536		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1537		       netdev->addr_len);
1538	}
1539
1540	adapter->last_reset = jiffies;
1541}
1542
1543int igbvf_up(struct igbvf_adapter *adapter)
1544{
1545	struct e1000_hw *hw = &adapter->hw;
1546
1547	/* hardware has been reset, we need to reload some things */
1548	igbvf_configure(adapter);
1549
1550	clear_bit(__IGBVF_DOWN, &adapter->state);
1551
1552	napi_enable(&adapter->rx_ring->napi);
1553	if (adapter->msix_entries)
1554		igbvf_configure_msix(adapter);
1555
1556	/* Clear any pending interrupts. */
1557	er32(EICR);
1558	igbvf_irq_enable(adapter);
1559
1560	/* start the watchdog */
1561	hw->mac.get_link_status = 1;
1562	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1563
1564	return 0;
1565}
1566
1567void igbvf_down(struct igbvf_adapter *adapter)
1568{
1569	struct net_device *netdev = adapter->netdev;
1570	struct e1000_hw *hw = &adapter->hw;
1571	u32 rxdctl, txdctl;
1572
1573	/* signal that we're down so the interrupt handler does not
1574	 * reschedule our watchdog timer
1575	 */
1576	set_bit(__IGBVF_DOWN, &adapter->state);
1577
1578	/* disable receives in the hardware */
1579	rxdctl = er32(RXDCTL(0));
1580	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1581
1582	netif_carrier_off(netdev);
1583	netif_stop_queue(netdev);
1584
1585	/* disable transmits in the hardware */
1586	txdctl = er32(TXDCTL(0));
1587	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1588
1589	/* flush both disables and wait for them to finish */
1590	e1e_flush();
1591	msleep(10);
1592
1593	napi_disable(&adapter->rx_ring->napi);
1594
1595	igbvf_irq_disable(adapter);
1596
1597	del_timer_sync(&adapter->watchdog_timer);
1598
1599	/* record the stats before reset*/
1600	igbvf_update_stats(adapter);
1601
1602	adapter->link_speed = 0;
1603	adapter->link_duplex = 0;
1604
1605	igbvf_reset(adapter);
1606	igbvf_clean_tx_ring(adapter->tx_ring);
1607	igbvf_clean_rx_ring(adapter->rx_ring);
1608}
1609
1610void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1611{
1612	might_sleep();
1613	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1614		usleep_range(1000, 2000);
1615	igbvf_down(adapter);
1616	igbvf_up(adapter);
1617	clear_bit(__IGBVF_RESETTING, &adapter->state);
1618}
1619
1620/**
1621 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1622 * @adapter: board private structure to initialize
1623 *
1624 * igbvf_sw_init initializes the Adapter private data structure.
1625 * Fields are initialized based on PCI device information and
1626 * OS network device settings (MTU size).
1627 **/
1628static int igbvf_sw_init(struct igbvf_adapter *adapter)
1629{
1630	struct net_device *netdev = adapter->netdev;
1631	s32 rc;
1632
1633	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1634	adapter->rx_ps_hdr_size = 0;
1635	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1636	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1637
1638	adapter->tx_int_delay = 8;
1639	adapter->tx_abs_int_delay = 32;
1640	adapter->rx_int_delay = 0;
1641	adapter->rx_abs_int_delay = 8;
1642	adapter->requested_itr = 3;
1643	adapter->current_itr = IGBVF_START_ITR;
1644
1645	/* Set various function pointers */
1646	adapter->ei->init_ops(&adapter->hw);
1647
1648	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1649	if (rc)
1650		return rc;
1651
1652	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1653	if (rc)
1654		return rc;
1655
1656	igbvf_set_interrupt_capability(adapter);
1657
1658	if (igbvf_alloc_queues(adapter))
1659		return -ENOMEM;
1660
1661	spin_lock_init(&adapter->tx_queue_lock);
1662
1663	/* Explicitly disable IRQ since the NIC can be in any state. */
1664	igbvf_irq_disable(adapter);
1665
1666	spin_lock_init(&adapter->stats_lock);
1667	spin_lock_init(&adapter->hw.mbx_lock);
1668
1669	set_bit(__IGBVF_DOWN, &adapter->state);
1670	return 0;
1671}
1672
1673static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1674{
1675	struct e1000_hw *hw = &adapter->hw;
1676
1677	adapter->stats.last_gprc = er32(VFGPRC);
1678	adapter->stats.last_gorc = er32(VFGORC);
1679	adapter->stats.last_gptc = er32(VFGPTC);
1680	adapter->stats.last_gotc = er32(VFGOTC);
1681	adapter->stats.last_mprc = er32(VFMPRC);
1682	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1683	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1684	adapter->stats.last_gorlbc = er32(VFGORLBC);
1685	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1686
1687	adapter->stats.base_gprc = er32(VFGPRC);
1688	adapter->stats.base_gorc = er32(VFGORC);
1689	adapter->stats.base_gptc = er32(VFGPTC);
1690	adapter->stats.base_gotc = er32(VFGOTC);
1691	adapter->stats.base_mprc = er32(VFMPRC);
1692	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1693	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1694	adapter->stats.base_gorlbc = er32(VFGORLBC);
1695	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1696}
1697
1698/**
1699 * igbvf_open - Called when a network interface is made active
1700 * @netdev: network interface device structure
1701 *
1702 * Returns 0 on success, negative value on failure
1703 *
1704 * The open entry point is called when a network interface is made
1705 * active by the system (IFF_UP).  At this point all resources needed
1706 * for transmit and receive operations are allocated, the interrupt
1707 * handler is registered with the OS, the watchdog timer is started,
1708 * and the stack is notified that the interface is ready.
1709 **/
1710static int igbvf_open(struct net_device *netdev)
1711{
1712	struct igbvf_adapter *adapter = netdev_priv(netdev);
1713	struct e1000_hw *hw = &adapter->hw;
1714	int err;
1715
1716	/* disallow open during test */
1717	if (test_bit(__IGBVF_TESTING, &adapter->state))
1718		return -EBUSY;
1719
1720	/* allocate transmit descriptors */
1721	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1722	if (err)
1723		goto err_setup_tx;
1724
1725	/* allocate receive descriptors */
1726	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1727	if (err)
1728		goto err_setup_rx;
1729
1730	/* before we allocate an interrupt, we must be ready to handle it.
1731	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1732	 * as soon as we call pci_request_irq, so we have to setup our
1733	 * clean_rx handler before we do so.
1734	 */
1735	igbvf_configure(adapter);
1736
1737	err = igbvf_request_irq(adapter);
1738	if (err)
1739		goto err_req_irq;
1740
1741	/* From here on the code is the same as igbvf_up() */
1742	clear_bit(__IGBVF_DOWN, &adapter->state);
1743
1744	napi_enable(&adapter->rx_ring->napi);
1745
1746	/* clear any pending interrupts */
1747	er32(EICR);
1748
1749	igbvf_irq_enable(adapter);
1750
1751	/* start the watchdog */
1752	hw->mac.get_link_status = 1;
1753	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1754
1755	return 0;
1756
1757err_req_irq:
1758	igbvf_free_rx_resources(adapter->rx_ring);
1759err_setup_rx:
1760	igbvf_free_tx_resources(adapter->tx_ring);
1761err_setup_tx:
1762	igbvf_reset(adapter);
1763
1764	return err;
1765}
1766
1767/**
1768 * igbvf_close - Disables a network interface
1769 * @netdev: network interface device structure
1770 *
1771 * Returns 0, this is not allowed to fail
1772 *
1773 * The close entry point is called when an interface is de-activated
1774 * by the OS.  The hardware is still under the drivers control, but
1775 * needs to be disabled.  A global MAC reset is issued to stop the
1776 * hardware, and all transmit and receive resources are freed.
1777 **/
1778static int igbvf_close(struct net_device *netdev)
1779{
1780	struct igbvf_adapter *adapter = netdev_priv(netdev);
1781
1782	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1783	igbvf_down(adapter);
1784
1785	igbvf_free_irq(adapter);
1786
1787	igbvf_free_tx_resources(adapter->tx_ring);
1788	igbvf_free_rx_resources(adapter->rx_ring);
1789
1790	return 0;
1791}
1792
1793/**
1794 * igbvf_set_mac - Change the Ethernet Address of the NIC
1795 * @netdev: network interface device structure
1796 * @p: pointer to an address structure
1797 *
1798 * Returns 0 on success, negative on failure
1799 **/
1800static int igbvf_set_mac(struct net_device *netdev, void *p)
1801{
1802	struct igbvf_adapter *adapter = netdev_priv(netdev);
1803	struct e1000_hw *hw = &adapter->hw;
1804	struct sockaddr *addr = p;
1805
1806	if (!is_valid_ether_addr(addr->sa_data))
1807		return -EADDRNOTAVAIL;
1808
1809	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1810
1811	spin_lock_bh(&hw->mbx_lock);
1812
1813	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1814
1815	spin_unlock_bh(&hw->mbx_lock);
1816
1817	if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1818		return -EADDRNOTAVAIL;
1819
1820	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1821
1822	return 0;
1823}
1824
1825#define UPDATE_VF_COUNTER(reg, name) \
1826{ \
1827	u32 current_counter = er32(reg); \
1828	if (current_counter < adapter->stats.last_##name) \
1829		adapter->stats.name += 0x100000000LL; \
1830	adapter->stats.last_##name = current_counter; \
1831	adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1832	adapter->stats.name |= current_counter; \
1833}
1834
1835/**
1836 * igbvf_update_stats - Update the board statistics counters
1837 * @adapter: board private structure
1838**/
1839void igbvf_update_stats(struct igbvf_adapter *adapter)
1840{
1841	struct e1000_hw *hw = &adapter->hw;
1842	struct pci_dev *pdev = adapter->pdev;
1843
1844	/* Prevent stats update while adapter is being reset, link is down
1845	 * or if the pci connection is down.
1846	 */
1847	if (adapter->link_speed == 0)
1848		return;
1849
1850	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1851		return;
1852
1853	if (pci_channel_offline(pdev))
1854		return;
1855
1856	UPDATE_VF_COUNTER(VFGPRC, gprc);
1857	UPDATE_VF_COUNTER(VFGORC, gorc);
1858	UPDATE_VF_COUNTER(VFGPTC, gptc);
1859	UPDATE_VF_COUNTER(VFGOTC, gotc);
1860	UPDATE_VF_COUNTER(VFMPRC, mprc);
1861	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1862	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1863	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1864	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1865
1866	/* Fill out the OS statistics structure */
1867	adapter->netdev->stats.multicast = adapter->stats.mprc;
1868}
1869
1870static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1871{
1872	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1873		 adapter->link_speed,
1874		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1875}
1876
1877static bool igbvf_has_link(struct igbvf_adapter *adapter)
1878{
1879	struct e1000_hw *hw = &adapter->hw;
1880	s32 ret_val = E1000_SUCCESS;
1881	bool link_active;
1882
1883	/* If interface is down, stay link down */
1884	if (test_bit(__IGBVF_DOWN, &adapter->state))
1885		return false;
1886
1887	spin_lock_bh(&hw->mbx_lock);
1888
1889	ret_val = hw->mac.ops.check_for_link(hw);
1890
1891	spin_unlock_bh(&hw->mbx_lock);
1892
1893	link_active = !hw->mac.get_link_status;
1894
1895	/* if check for link returns error we will need to reset */
1896	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1897		schedule_work(&adapter->reset_task);
1898
1899	return link_active;
1900}
1901
1902/**
1903 * igbvf_watchdog - Timer Call-back
1904 * @t: timer list pointer containing private struct
1905 **/
1906static void igbvf_watchdog(struct timer_list *t)
1907{
1908	struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer);
1909
1910	/* Do the rest outside of interrupt context */
1911	schedule_work(&adapter->watchdog_task);
1912}
1913
1914static void igbvf_watchdog_task(struct work_struct *work)
1915{
1916	struct igbvf_adapter *adapter = container_of(work,
1917						     struct igbvf_adapter,
1918						     watchdog_task);
1919	struct net_device *netdev = adapter->netdev;
1920	struct e1000_mac_info *mac = &adapter->hw.mac;
1921	struct igbvf_ring *tx_ring = adapter->tx_ring;
1922	struct e1000_hw *hw = &adapter->hw;
1923	u32 link;
1924	int tx_pending = 0;
1925
1926	link = igbvf_has_link(adapter);
1927
1928	if (link) {
1929		if (!netif_carrier_ok(netdev)) {
1930			mac->ops.get_link_up_info(&adapter->hw,
1931						  &adapter->link_speed,
1932						  &adapter->link_duplex);
1933			igbvf_print_link_info(adapter);
1934
1935			netif_carrier_on(netdev);
1936			netif_wake_queue(netdev);
1937		}
1938	} else {
1939		if (netif_carrier_ok(netdev)) {
1940			adapter->link_speed = 0;
1941			adapter->link_duplex = 0;
1942			dev_info(&adapter->pdev->dev, "Link is Down\n");
1943			netif_carrier_off(netdev);
1944			netif_stop_queue(netdev);
1945		}
1946	}
1947
1948	if (netif_carrier_ok(netdev)) {
1949		igbvf_update_stats(adapter);
1950	} else {
1951		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1952			      tx_ring->count);
1953		if (tx_pending) {
1954			/* We've lost link, so the controller stops DMA,
1955			 * but we've got queued Tx work that's never going
1956			 * to get done, so reset controller to flush Tx.
1957			 * (Do the reset outside of interrupt context).
1958			 */
1959			adapter->tx_timeout_count++;
1960			schedule_work(&adapter->reset_task);
1961		}
1962	}
1963
1964	/* Cause software interrupt to ensure Rx ring is cleaned */
1965	ew32(EICS, adapter->rx_ring->eims_value);
1966
1967	/* Reset the timer */
1968	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1969		mod_timer(&adapter->watchdog_timer,
1970			  round_jiffies(jiffies + (2 * HZ)));
1971}
1972
1973#define IGBVF_TX_FLAGS_CSUM		0x00000001
1974#define IGBVF_TX_FLAGS_VLAN		0x00000002
1975#define IGBVF_TX_FLAGS_TSO		0x00000004
1976#define IGBVF_TX_FLAGS_IPV4		0x00000008
1977#define IGBVF_TX_FLAGS_VLAN_MASK	0xffff0000
1978#define IGBVF_TX_FLAGS_VLAN_SHIFT	16
1979
1980static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1981			      u32 type_tucmd, u32 mss_l4len_idx)
1982{
1983	struct e1000_adv_tx_context_desc *context_desc;
1984	struct igbvf_buffer *buffer_info;
1985	u16 i = tx_ring->next_to_use;
1986
1987	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1988	buffer_info = &tx_ring->buffer_info[i];
1989
1990	i++;
1991	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1992
1993	/* set bits to identify this as an advanced context descriptor */
1994	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1995
1996	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
1997	context_desc->seqnum_seed	= 0;
1998	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
1999	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
2000
2001	buffer_info->time_stamp = jiffies;
2002	buffer_info->dma = 0;
2003}
2004
2005static int igbvf_tso(struct igbvf_ring *tx_ring,
2006		     struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2007{
2008	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
2009	union {
2010		struct iphdr *v4;
2011		struct ipv6hdr *v6;
2012		unsigned char *hdr;
2013	} ip;
2014	union {
2015		struct tcphdr *tcp;
2016		unsigned char *hdr;
2017	} l4;
2018	u32 paylen, l4_offset;
2019	int err;
2020
2021	if (skb->ip_summed != CHECKSUM_PARTIAL)
2022		return 0;
2023
2024	if (!skb_is_gso(skb))
2025		return 0;
2026
2027	err = skb_cow_head(skb, 0);
2028	if (err < 0)
2029		return err;
2030
2031	ip.hdr = skb_network_header(skb);
2032	l4.hdr = skb_checksum_start(skb);
2033
2034	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2035	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2036
2037	/* initialize outer IP header fields */
2038	if (ip.v4->version == 4) {
2039		unsigned char *csum_start = skb_checksum_start(skb);
2040		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2041
2042		/* IP header will have to cancel out any data that
2043		 * is not a part of the outer IP header
2044		 */
2045		ip.v4->check = csum_fold(csum_partial(trans_start,
2046						      csum_start - trans_start,
2047						      0));
2048		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2049
2050		ip.v4->tot_len = 0;
2051	} else {
2052		ip.v6->payload_len = 0;
2053	}
2054
2055	/* determine offset of inner transport header */
2056	l4_offset = l4.hdr - skb->data;
2057
2058	/* compute length of segmentation header */
2059	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2060
2061	/* remove payload length from inner checksum */
2062	paylen = skb->len - l4_offset;
2063	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2064
2065	/* MSS L4LEN IDX */
2066	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2067	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2068
2069	/* VLAN MACLEN IPLEN */
2070	vlan_macip_lens = l4.hdr - ip.hdr;
2071	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2072	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2073
2074	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2075
2076	return 1;
2077}
2078
2079static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb)
2080{
2081	unsigned int offset = 0;
2082
2083	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
2084
2085	return offset == skb_checksum_start_offset(skb);
2086}
2087
2088static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2089			  u32 tx_flags, __be16 protocol)
2090{
2091	u32 vlan_macip_lens = 0;
2092	u32 type_tucmd = 0;
2093
2094	if (skb->ip_summed != CHECKSUM_PARTIAL) {
2095csum_failed:
2096		if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2097			return false;
2098		goto no_csum;
2099	}
2100
2101	switch (skb->csum_offset) {
2102	case offsetof(struct tcphdr, check):
2103		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2104		fallthrough;
2105	case offsetof(struct udphdr, check):
2106		break;
2107	case offsetof(struct sctphdr, checksum):
2108		/* validate that this is actually an SCTP request */
2109		if (((protocol == htons(ETH_P_IP)) &&
2110		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
2111		    ((protocol == htons(ETH_P_IPV6)) &&
2112		     igbvf_ipv6_csum_is_sctp(skb))) {
2113			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2114			break;
2115		}
2116		fallthrough;
2117	default:
2118		skb_checksum_help(skb);
2119		goto csum_failed;
2120	}
2121
2122	vlan_macip_lens = skb_checksum_start_offset(skb) -
2123			  skb_network_offset(skb);
2124no_csum:
2125	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2126	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2127
2128	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2129	return true;
2130}
2131
2132static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2133{
2134	struct igbvf_adapter *adapter = netdev_priv(netdev);
2135
2136	/* there is enough descriptors then we don't need to worry  */
2137	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2138		return 0;
2139
2140	netif_stop_queue(netdev);
2141
2142	/* Herbert's original patch had:
2143	 *  smp_mb__after_netif_stop_queue();
2144	 * but since that doesn't exist yet, just open code it.
2145	 */
2146	smp_mb();
2147
2148	/* We need to check again just in case room has been made available */
2149	if (igbvf_desc_unused(adapter->tx_ring) < size)
2150		return -EBUSY;
2151
2152	netif_wake_queue(netdev);
2153
2154	++adapter->restart_queue;
2155	return 0;
2156}
2157
2158#define IGBVF_MAX_TXD_PWR	16
2159#define IGBVF_MAX_DATA_PER_TXD	(1u << IGBVF_MAX_TXD_PWR)
2160
2161static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2162				   struct igbvf_ring *tx_ring,
2163				   struct sk_buff *skb)
2164{
2165	struct igbvf_buffer *buffer_info;
2166	struct pci_dev *pdev = adapter->pdev;
2167	unsigned int len = skb_headlen(skb);
2168	unsigned int count = 0, i;
2169	unsigned int f;
2170
2171	i = tx_ring->next_to_use;
2172
2173	buffer_info = &tx_ring->buffer_info[i];
2174	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2175	buffer_info->length = len;
2176	/* set time_stamp *before* dma to help avoid a possible race */
2177	buffer_info->time_stamp = jiffies;
2178	buffer_info->mapped_as_page = false;
2179	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2180					  DMA_TO_DEVICE);
2181	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2182		goto dma_error;
2183
2184	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2185		const skb_frag_t *frag;
2186
2187		count++;
2188		i++;
2189		if (i == tx_ring->count)
2190			i = 0;
2191
2192		frag = &skb_shinfo(skb)->frags[f];
2193		len = skb_frag_size(frag);
2194
2195		buffer_info = &tx_ring->buffer_info[i];
2196		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2197		buffer_info->length = len;
2198		buffer_info->time_stamp = jiffies;
2199		buffer_info->mapped_as_page = true;
2200		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2201						    DMA_TO_DEVICE);
2202		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2203			goto dma_error;
2204	}
2205
2206	tx_ring->buffer_info[i].skb = skb;
2207
2208	return ++count;
2209
2210dma_error:
2211	dev_err(&pdev->dev, "TX DMA map failed\n");
2212
2213	/* clear timestamp and dma mappings for failed buffer_info mapping */
2214	buffer_info->dma = 0;
2215	buffer_info->time_stamp = 0;
2216	buffer_info->length = 0;
2217	buffer_info->mapped_as_page = false;
2218	if (count)
2219		count--;
2220
2221	/* clear timestamp and dma mappings for remaining portion of packet */
2222	while (count--) {
2223		if (i == 0)
2224			i += tx_ring->count;
2225		i--;
2226		buffer_info = &tx_ring->buffer_info[i];
2227		igbvf_put_txbuf(adapter, buffer_info);
2228	}
2229
2230	return 0;
2231}
2232
2233static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2234				      struct igbvf_ring *tx_ring,
2235				      int tx_flags, int count,
2236				      unsigned int first, u32 paylen,
2237				      u8 hdr_len)
2238{
2239	union e1000_adv_tx_desc *tx_desc = NULL;
2240	struct igbvf_buffer *buffer_info;
2241	u32 olinfo_status = 0, cmd_type_len;
2242	unsigned int i;
2243
2244	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2245			E1000_ADVTXD_DCMD_DEXT);
2246
2247	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2248		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2249
2250	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2251		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2252
2253		/* insert tcp checksum */
2254		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2255
2256		/* insert ip checksum */
2257		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2258			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2259
2260	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2261		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2262	}
2263
2264	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2265
2266	i = tx_ring->next_to_use;
2267	while (count--) {
2268		buffer_info = &tx_ring->buffer_info[i];
2269		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2270		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2271		tx_desc->read.cmd_type_len =
2272			 cpu_to_le32(cmd_type_len | buffer_info->length);
2273		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2274		i++;
2275		if (i == tx_ring->count)
2276			i = 0;
2277	}
2278
2279	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2280	/* Force memory writes to complete before letting h/w
2281	 * know there are new descriptors to fetch.  (Only
2282	 * applicable for weak-ordered memory model archs,
2283	 * such as IA-64).
2284	 */
2285	wmb();
2286
2287	tx_ring->buffer_info[first].next_to_watch = tx_desc;
2288	tx_ring->next_to_use = i;
2289	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2290}
2291
2292static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2293					     struct net_device *netdev,
2294					     struct igbvf_ring *tx_ring)
2295{
2296	struct igbvf_adapter *adapter = netdev_priv(netdev);
2297	unsigned int first, tx_flags = 0;
2298	u8 hdr_len = 0;
2299	int count = 0;
2300	int tso = 0;
2301	__be16 protocol = vlan_get_protocol(skb);
2302
2303	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2304		dev_kfree_skb_any(skb);
2305		return NETDEV_TX_OK;
2306	}
2307
2308	if (skb->len <= 0) {
2309		dev_kfree_skb_any(skb);
2310		return NETDEV_TX_OK;
2311	}
2312
2313	/* need: count + 4 desc gap to keep tail from touching
2314	 *       + 2 desc gap to keep tail from touching head,
2315	 *       + 1 desc for skb->data,
2316	 *       + 1 desc for context descriptor,
2317	 * head, otherwise try next time
2318	 */
2319	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2320		/* this is a hard error */
2321		return NETDEV_TX_BUSY;
2322	}
2323
2324	if (skb_vlan_tag_present(skb)) {
2325		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2326		tx_flags |= (skb_vlan_tag_get(skb) <<
2327			     IGBVF_TX_FLAGS_VLAN_SHIFT);
2328	}
2329
2330	if (protocol == htons(ETH_P_IP))
2331		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2332
2333	first = tx_ring->next_to_use;
2334
2335	tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2336	if (unlikely(tso < 0)) {
2337		dev_kfree_skb_any(skb);
2338		return NETDEV_TX_OK;
2339	}
2340
2341	if (tso)
2342		tx_flags |= IGBVF_TX_FLAGS_TSO;
2343	else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2344		 (skb->ip_summed == CHECKSUM_PARTIAL))
2345		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2346
2347	/* count reflects descriptors mapped, if 0 then mapping error
2348	 * has occurred and we need to rewind the descriptor queue
2349	 */
2350	count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2351
2352	if (count) {
2353		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2354				   first, skb->len, hdr_len);
2355		/* Make sure there is space in the ring for the next send. */
2356		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2357	} else {
2358		dev_kfree_skb_any(skb);
2359		tx_ring->buffer_info[first].time_stamp = 0;
2360		tx_ring->next_to_use = first;
2361	}
2362
2363	return NETDEV_TX_OK;
2364}
2365
2366static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2367				    struct net_device *netdev)
2368{
2369	struct igbvf_adapter *adapter = netdev_priv(netdev);
2370	struct igbvf_ring *tx_ring;
2371
2372	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2373		dev_kfree_skb_any(skb);
2374		return NETDEV_TX_OK;
2375	}
2376
2377	tx_ring = &adapter->tx_ring[0];
2378
2379	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2380}
2381
2382/**
2383 * igbvf_tx_timeout - Respond to a Tx Hang
2384 * @netdev: network interface device structure
2385 * @txqueue: queue timing out (unused)
2386 **/
2387static void igbvf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
2388{
2389	struct igbvf_adapter *adapter = netdev_priv(netdev);
2390
2391	/* Do the reset outside of interrupt context */
2392	adapter->tx_timeout_count++;
2393	schedule_work(&adapter->reset_task);
2394}
2395
2396static void igbvf_reset_task(struct work_struct *work)
2397{
2398	struct igbvf_adapter *adapter;
2399
2400	adapter = container_of(work, struct igbvf_adapter, reset_task);
2401
2402	igbvf_reinit_locked(adapter);
2403}
2404
2405/**
2406 * igbvf_change_mtu - Change the Maximum Transfer Unit
2407 * @netdev: network interface device structure
2408 * @new_mtu: new value for maximum frame size
2409 *
2410 * Returns 0 on success, negative on failure
2411 **/
2412static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2413{
2414	struct igbvf_adapter *adapter = netdev_priv(netdev);
2415	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2416
2417	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2418		usleep_range(1000, 2000);
2419	/* igbvf_down has a dependency on max_frame_size */
2420	adapter->max_frame_size = max_frame;
2421	if (netif_running(netdev))
2422		igbvf_down(adapter);
2423
2424	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2425	 * means we reserve 2 more, this pushes us to allocate from the next
2426	 * larger slab size.
2427	 * i.e. RXBUFFER_2048 --> size-4096 slab
2428	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2429	 * fragmented skbs
2430	 */
2431
2432	if (max_frame <= 1024)
2433		adapter->rx_buffer_len = 1024;
2434	else if (max_frame <= 2048)
2435		adapter->rx_buffer_len = 2048;
2436	else
2437#if (PAGE_SIZE / 2) > 16384
2438		adapter->rx_buffer_len = 16384;
2439#else
2440		adapter->rx_buffer_len = PAGE_SIZE / 2;
2441#endif
2442
2443	/* adjust allocation if LPE protects us, and we aren't using SBP */
2444	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2445	    (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2446		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2447					 ETH_FCS_LEN;
2448
2449	netdev_dbg(netdev, "changing MTU from %d to %d\n",
2450		   netdev->mtu, new_mtu);
2451	netdev->mtu = new_mtu;
2452
2453	if (netif_running(netdev))
2454		igbvf_up(adapter);
2455	else
2456		igbvf_reset(adapter);
2457
2458	clear_bit(__IGBVF_RESETTING, &adapter->state);
2459
2460	return 0;
2461}
2462
2463static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2464{
2465	switch (cmd) {
2466	default:
2467		return -EOPNOTSUPP;
2468	}
2469}
2470
2471static int igbvf_suspend(struct device *dev_d)
2472{
2473	struct net_device *netdev = dev_get_drvdata(dev_d);
2474	struct igbvf_adapter *adapter = netdev_priv(netdev);
2475
2476	netif_device_detach(netdev);
2477
2478	if (netif_running(netdev)) {
2479		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2480		igbvf_down(adapter);
2481		igbvf_free_irq(adapter);
2482	}
2483
2484	return 0;
2485}
2486
2487static int __maybe_unused igbvf_resume(struct device *dev_d)
2488{
2489	struct pci_dev *pdev = to_pci_dev(dev_d);
2490	struct net_device *netdev = pci_get_drvdata(pdev);
2491	struct igbvf_adapter *adapter = netdev_priv(netdev);
2492	u32 err;
2493
2494	pci_set_master(pdev);
2495
2496	if (netif_running(netdev)) {
2497		err = igbvf_request_irq(adapter);
2498		if (err)
2499			return err;
2500	}
2501
2502	igbvf_reset(adapter);
2503
2504	if (netif_running(netdev))
2505		igbvf_up(adapter);
2506
2507	netif_device_attach(netdev);
2508
2509	return 0;
2510}
2511
2512static void igbvf_shutdown(struct pci_dev *pdev)
2513{
2514	igbvf_suspend(&pdev->dev);
2515}
2516
2517#ifdef CONFIG_NET_POLL_CONTROLLER
2518/* Polling 'interrupt' - used by things like netconsole to send skbs
2519 * without having to re-enable interrupts. It's not called while
2520 * the interrupt routine is executing.
2521 */
2522static void igbvf_netpoll(struct net_device *netdev)
2523{
2524	struct igbvf_adapter *adapter = netdev_priv(netdev);
2525
2526	disable_irq(adapter->pdev->irq);
2527
2528	igbvf_clean_tx_irq(adapter->tx_ring);
2529
2530	enable_irq(adapter->pdev->irq);
2531}
2532#endif
2533
2534/**
2535 * igbvf_io_error_detected - called when PCI error is detected
2536 * @pdev: Pointer to PCI device
2537 * @state: The current pci connection state
2538 *
2539 * This function is called after a PCI bus error affecting
2540 * this device has been detected.
2541 */
2542static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2543						pci_channel_state_t state)
2544{
2545	struct net_device *netdev = pci_get_drvdata(pdev);
2546	struct igbvf_adapter *adapter = netdev_priv(netdev);
2547
2548	netif_device_detach(netdev);
2549
2550	if (state == pci_channel_io_perm_failure)
2551		return PCI_ERS_RESULT_DISCONNECT;
2552
2553	if (netif_running(netdev))
2554		igbvf_down(adapter);
2555	pci_disable_device(pdev);
2556
2557	/* Request a slot slot reset. */
2558	return PCI_ERS_RESULT_NEED_RESET;
2559}
2560
2561/**
2562 * igbvf_io_slot_reset - called after the pci bus has been reset.
2563 * @pdev: Pointer to PCI device
2564 *
2565 * Restart the card from scratch, as if from a cold-boot. Implementation
2566 * resembles the first-half of the igbvf_resume routine.
2567 */
2568static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2569{
2570	struct net_device *netdev = pci_get_drvdata(pdev);
2571	struct igbvf_adapter *adapter = netdev_priv(netdev);
2572
2573	if (pci_enable_device_mem(pdev)) {
2574		dev_err(&pdev->dev,
2575			"Cannot re-enable PCI device after reset.\n");
2576		return PCI_ERS_RESULT_DISCONNECT;
2577	}
2578	pci_set_master(pdev);
2579
2580	igbvf_reset(adapter);
2581
2582	return PCI_ERS_RESULT_RECOVERED;
2583}
2584
2585/**
2586 * igbvf_io_resume - called when traffic can start flowing again.
2587 * @pdev: Pointer to PCI device
2588 *
2589 * This callback is called when the error recovery driver tells us that
2590 * its OK to resume normal operation. Implementation resembles the
2591 * second-half of the igbvf_resume routine.
2592 */
2593static void igbvf_io_resume(struct pci_dev *pdev)
2594{
2595	struct net_device *netdev = pci_get_drvdata(pdev);
2596	struct igbvf_adapter *adapter = netdev_priv(netdev);
2597
2598	if (netif_running(netdev)) {
2599		if (igbvf_up(adapter)) {
2600			dev_err(&pdev->dev,
2601				"can't bring device back up after reset\n");
2602			return;
2603		}
2604	}
2605
2606	netif_device_attach(netdev);
2607}
2608
2609static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2610{
2611	struct e1000_hw *hw = &adapter->hw;
2612	struct net_device *netdev = adapter->netdev;
2613	struct pci_dev *pdev = adapter->pdev;
2614
2615	if (hw->mac.type == e1000_vfadapt_i350)
2616		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2617	else
2618		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2619	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2620}
2621
2622static int igbvf_set_features(struct net_device *netdev,
2623			      netdev_features_t features)
2624{
2625	struct igbvf_adapter *adapter = netdev_priv(netdev);
2626
2627	if (features & NETIF_F_RXCSUM)
2628		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2629	else
2630		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2631
2632	return 0;
2633}
2634
2635#define IGBVF_MAX_MAC_HDR_LEN		127
2636#define IGBVF_MAX_NETWORK_HDR_LEN	511
2637
2638static netdev_features_t
2639igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2640		     netdev_features_t features)
2641{
2642	unsigned int network_hdr_len, mac_hdr_len;
2643
2644	/* Make certain the headers can be described by a context descriptor */
2645	mac_hdr_len = skb_network_header(skb) - skb->data;
2646	if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2647		return features & ~(NETIF_F_HW_CSUM |
2648				    NETIF_F_SCTP_CRC |
2649				    NETIF_F_HW_VLAN_CTAG_TX |
2650				    NETIF_F_TSO |
2651				    NETIF_F_TSO6);
2652
2653	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2654	if (unlikely(network_hdr_len >  IGBVF_MAX_NETWORK_HDR_LEN))
2655		return features & ~(NETIF_F_HW_CSUM |
2656				    NETIF_F_SCTP_CRC |
2657				    NETIF_F_TSO |
2658				    NETIF_F_TSO6);
2659
2660	/* We can only support IPV4 TSO in tunnels if we can mangle the
2661	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2662	 */
2663	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2664		features &= ~NETIF_F_TSO;
2665
2666	return features;
2667}
2668
2669static const struct net_device_ops igbvf_netdev_ops = {
2670	.ndo_open		= igbvf_open,
2671	.ndo_stop		= igbvf_close,
2672	.ndo_start_xmit		= igbvf_xmit_frame,
2673	.ndo_set_rx_mode	= igbvf_set_rx_mode,
2674	.ndo_set_mac_address	= igbvf_set_mac,
2675	.ndo_change_mtu		= igbvf_change_mtu,
2676	.ndo_do_ioctl		= igbvf_ioctl,
2677	.ndo_tx_timeout		= igbvf_tx_timeout,
2678	.ndo_vlan_rx_add_vid	= igbvf_vlan_rx_add_vid,
2679	.ndo_vlan_rx_kill_vid	= igbvf_vlan_rx_kill_vid,
2680#ifdef CONFIG_NET_POLL_CONTROLLER
2681	.ndo_poll_controller	= igbvf_netpoll,
2682#endif
2683	.ndo_set_features	= igbvf_set_features,
2684	.ndo_features_check	= igbvf_features_check,
2685};
2686
2687/**
2688 * igbvf_probe - Device Initialization Routine
2689 * @pdev: PCI device information struct
2690 * @ent: entry in igbvf_pci_tbl
2691 *
2692 * Returns 0 on success, negative on failure
2693 *
2694 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2695 * The OS initialization, configuring of the adapter private structure,
2696 * and a hardware reset occur.
2697 **/
2698static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2699{
2700	struct net_device *netdev;
2701	struct igbvf_adapter *adapter;
2702	struct e1000_hw *hw;
2703	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2704
2705	static int cards_found;
2706	int err, pci_using_dac;
2707
2708	err = pci_enable_device_mem(pdev);
2709	if (err)
2710		return err;
2711
2712	pci_using_dac = 0;
2713	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2714	if (!err) {
2715		pci_using_dac = 1;
2716	} else {
2717		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2718		if (err) {
2719			dev_err(&pdev->dev,
2720				"No usable DMA configuration, aborting\n");
2721			goto err_dma;
2722		}
2723	}
2724
2725	err = pci_request_regions(pdev, igbvf_driver_name);
2726	if (err)
2727		goto err_pci_reg;
2728
2729	pci_set_master(pdev);
2730
2731	err = -ENOMEM;
2732	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2733	if (!netdev)
2734		goto err_alloc_etherdev;
2735
2736	SET_NETDEV_DEV(netdev, &pdev->dev);
2737
2738	pci_set_drvdata(pdev, netdev);
2739	adapter = netdev_priv(netdev);
2740	hw = &adapter->hw;
2741	adapter->netdev = netdev;
2742	adapter->pdev = pdev;
2743	adapter->ei = ei;
2744	adapter->pba = ei->pba;
2745	adapter->flags = ei->flags;
2746	adapter->hw.back = adapter;
2747	adapter->hw.mac.type = ei->mac;
2748	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2749
2750	/* PCI config space info */
2751
2752	hw->vendor_id = pdev->vendor;
2753	hw->device_id = pdev->device;
2754	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2755	hw->subsystem_device_id = pdev->subsystem_device;
2756	hw->revision_id = pdev->revision;
2757
2758	err = -EIO;
2759	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2760				      pci_resource_len(pdev, 0));
2761
2762	if (!adapter->hw.hw_addr)
2763		goto err_ioremap;
2764
2765	if (ei->get_variants) {
2766		err = ei->get_variants(adapter);
2767		if (err)
2768			goto err_get_variants;
2769	}
2770
2771	/* setup adapter struct */
2772	err = igbvf_sw_init(adapter);
2773	if (err)
2774		goto err_sw_init;
2775
2776	/* construct the net_device struct */
2777	netdev->netdev_ops = &igbvf_netdev_ops;
2778
2779	igbvf_set_ethtool_ops(netdev);
2780	netdev->watchdog_timeo = 5 * HZ;
2781	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2782
2783	adapter->bd_number = cards_found++;
2784
2785	netdev->hw_features = NETIF_F_SG |
2786			      NETIF_F_TSO |
2787			      NETIF_F_TSO6 |
2788			      NETIF_F_RXCSUM |
2789			      NETIF_F_HW_CSUM |
2790			      NETIF_F_SCTP_CRC;
2791
2792#define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2793				    NETIF_F_GSO_GRE_CSUM | \
2794				    NETIF_F_GSO_IPXIP4 | \
2795				    NETIF_F_GSO_IPXIP6 | \
2796				    NETIF_F_GSO_UDP_TUNNEL | \
2797				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
2798
2799	netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2800	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2801			       IGBVF_GSO_PARTIAL_FEATURES;
2802
2803	netdev->features = netdev->hw_features;
2804
2805	if (pci_using_dac)
2806		netdev->features |= NETIF_F_HIGHDMA;
2807
2808	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2809	netdev->mpls_features |= NETIF_F_HW_CSUM;
2810	netdev->hw_enc_features |= netdev->vlan_features;
2811
2812	/* set this bit last since it cannot be part of vlan_features */
2813	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2814			    NETIF_F_HW_VLAN_CTAG_RX |
2815			    NETIF_F_HW_VLAN_CTAG_TX;
2816
2817	/* MTU range: 68 - 9216 */
2818	netdev->min_mtu = ETH_MIN_MTU;
2819	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2820
2821	spin_lock_bh(&hw->mbx_lock);
2822
2823	/*reset the controller to put the device in a known good state */
2824	err = hw->mac.ops.reset_hw(hw);
2825	if (err) {
2826		dev_info(&pdev->dev,
2827			 "PF still in reset state. Is the PF interface up?\n");
2828	} else {
2829		err = hw->mac.ops.read_mac_addr(hw);
2830		if (err)
2831			dev_info(&pdev->dev, "Error reading MAC address.\n");
2832		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2833			dev_info(&pdev->dev,
2834				 "MAC address not assigned by administrator.\n");
2835		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2836		       netdev->addr_len);
2837	}
2838
2839	spin_unlock_bh(&hw->mbx_lock);
2840
2841	if (!is_valid_ether_addr(netdev->dev_addr)) {
2842		dev_info(&pdev->dev, "Assigning random MAC address.\n");
2843		eth_hw_addr_random(netdev);
2844		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2845		       netdev->addr_len);
2846	}
2847
2848	timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0);
2849
2850	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2851	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2852
2853	/* ring size defaults */
2854	adapter->rx_ring->count = 1024;
2855	adapter->tx_ring->count = 1024;
2856
2857	/* reset the hardware with the new settings */
2858	igbvf_reset(adapter);
2859
2860	/* set hardware-specific flags */
2861	if (adapter->hw.mac.type == e1000_vfadapt_i350)
2862		adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2863
2864	strcpy(netdev->name, "eth%d");
2865	err = register_netdev(netdev);
2866	if (err)
2867		goto err_hw_init;
2868
2869	/* tell the stack to leave us alone until igbvf_open() is called */
2870	netif_carrier_off(netdev);
2871	netif_stop_queue(netdev);
2872
2873	igbvf_print_device_info(adapter);
2874
2875	igbvf_initialize_last_counter_stats(adapter);
2876
2877	return 0;
2878
2879err_hw_init:
2880	netif_napi_del(&adapter->rx_ring->napi);
2881	kfree(adapter->tx_ring);
2882	kfree(adapter->rx_ring);
2883err_sw_init:
2884	igbvf_reset_interrupt_capability(adapter);
2885err_get_variants:
2886	iounmap(adapter->hw.hw_addr);
2887err_ioremap:
2888	free_netdev(netdev);
2889err_alloc_etherdev:
2890	pci_release_regions(pdev);
2891err_pci_reg:
2892err_dma:
2893	pci_disable_device(pdev);
2894	return err;
2895}
2896
2897/**
2898 * igbvf_remove - Device Removal Routine
2899 * @pdev: PCI device information struct
2900 *
2901 * igbvf_remove is called by the PCI subsystem to alert the driver
2902 * that it should release a PCI device.  The could be caused by a
2903 * Hot-Plug event, or because the driver is going to be removed from
2904 * memory.
2905 **/
2906static void igbvf_remove(struct pci_dev *pdev)
2907{
2908	struct net_device *netdev = pci_get_drvdata(pdev);
2909	struct igbvf_adapter *adapter = netdev_priv(netdev);
2910	struct e1000_hw *hw = &adapter->hw;
2911
2912	/* The watchdog timer may be rescheduled, so explicitly
2913	 * disable it from being rescheduled.
2914	 */
2915	set_bit(__IGBVF_DOWN, &adapter->state);
2916	del_timer_sync(&adapter->watchdog_timer);
2917
2918	cancel_work_sync(&adapter->reset_task);
2919	cancel_work_sync(&adapter->watchdog_task);
2920
2921	unregister_netdev(netdev);
2922
2923	igbvf_reset_interrupt_capability(adapter);
2924
2925	/* it is important to delete the NAPI struct prior to freeing the
2926	 * Rx ring so that you do not end up with null pointer refs
2927	 */
2928	netif_napi_del(&adapter->rx_ring->napi);
2929	kfree(adapter->tx_ring);
2930	kfree(adapter->rx_ring);
2931
2932	iounmap(hw->hw_addr);
2933	if (hw->flash_address)
2934		iounmap(hw->flash_address);
2935	pci_release_regions(pdev);
2936
2937	free_netdev(netdev);
2938
2939	pci_disable_device(pdev);
2940}
2941
2942/* PCI Error Recovery (ERS) */
2943static const struct pci_error_handlers igbvf_err_handler = {
2944	.error_detected = igbvf_io_error_detected,
2945	.slot_reset = igbvf_io_slot_reset,
2946	.resume = igbvf_io_resume,
2947};
2948
2949static const struct pci_device_id igbvf_pci_tbl[] = {
2950	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2951	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2952	{ } /* terminate list */
2953};
2954MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2955
2956static SIMPLE_DEV_PM_OPS(igbvf_pm_ops, igbvf_suspend, igbvf_resume);
2957
2958/* PCI Device API Driver */
2959static struct pci_driver igbvf_driver = {
2960	.name		= igbvf_driver_name,
2961	.id_table	= igbvf_pci_tbl,
2962	.probe		= igbvf_probe,
2963	.remove		= igbvf_remove,
2964	.driver.pm	= &igbvf_pm_ops,
2965	.shutdown	= igbvf_shutdown,
2966	.err_handler	= &igbvf_err_handler
2967};
2968
2969/**
2970 * igbvf_init_module - Driver Registration Routine
2971 *
2972 * igbvf_init_module is the first routine called when the driver is
2973 * loaded. All it does is register with the PCI subsystem.
2974 **/
2975static int __init igbvf_init_module(void)
2976{
2977	int ret;
2978
2979	pr_info("%s\n", igbvf_driver_string);
2980	pr_info("%s\n", igbvf_copyright);
2981
2982	ret = pci_register_driver(&igbvf_driver);
2983
2984	return ret;
2985}
2986module_init(igbvf_init_module);
2987
2988/**
2989 * igbvf_exit_module - Driver Exit Cleanup Routine
2990 *
2991 * igbvf_exit_module is called just before the driver is removed
2992 * from memory.
2993 **/
2994static void __exit igbvf_exit_module(void)
2995{
2996	pci_unregister_driver(&igbvf_driver);
2997}
2998module_exit(igbvf_exit_module);
2999
3000MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
3001MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
3002MODULE_LICENSE("GPL v2");
3003
3004/* netdev.c */
3005