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 
27 #define DRV_VERSION "2.4.0-k"
28 char igbvf_driver_name[] = "igbvf";
29 const char igbvf_driver_version[] = DRV_VERSION;
30 static const char igbvf_driver_string[] =
31 		  "Intel(R) Gigabit Virtual Function Network Driver";
32 static const char igbvf_copyright[] =
33 		  "Copyright (c) 2009 - 2012 Intel Corporation.";
34 
35 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
36 static int debug = -1;
37 module_param(debug, int, 0);
38 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
39 
40 static int igbvf_poll(struct napi_struct *napi, int budget);
41 static void igbvf_reset(struct igbvf_adapter *);
42 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
43 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
44 
45 static struct igbvf_info igbvf_vf_info = {
46 	.mac		= e1000_vfadapt,
47 	.flags		= 0,
48 	.pba		= 10,
49 	.init_ops	= e1000_init_function_pointers_vf,
50 };
51 
52 static struct igbvf_info igbvf_i350_vf_info = {
53 	.mac		= e1000_vfadapt_i350,
54 	.flags		= 0,
55 	.pba		= 10,
56 	.init_ops	= e1000_init_function_pointers_vf,
57 };
58 
59 static const struct igbvf_info *igbvf_info_tbl[] = {
60 	[board_vf]	= &igbvf_vf_info,
61 	[board_i350_vf]	= &igbvf_i350_vf_info,
62 };
63 
64 /**
65  * igbvf_desc_unused - calculate if we have unused descriptors
66  * @rx_ring: address of receive ring structure
67  **/
68 static int igbvf_desc_unused(struct igbvf_ring *ring)
69 {
70 	if (ring->next_to_clean > ring->next_to_use)
71 		return ring->next_to_clean - ring->next_to_use - 1;
72 
73 	return ring->count + ring->next_to_clean - ring->next_to_use - 1;
74 }
75 
76 /**
77  * igbvf_receive_skb - helper function to handle Rx indications
78  * @adapter: board private structure
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  **/
83 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
84 			      struct net_device *netdev,
85 			      struct sk_buff *skb,
86 			      u32 status, u16 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(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 
103 static 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  **/
133 static 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 
217 no_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  *
239  * the return value indicates whether actual cleaning was done, there
240  * is no guarantee that everything was cleaned
241  **/
242 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
243 			       int *work_done, int work_to_do)
244 {
245 	struct igbvf_ring *rx_ring = adapter->rx_ring;
246 	struct net_device *netdev = adapter->netdev;
247 	struct pci_dev *pdev = adapter->pdev;
248 	union e1000_adv_rx_desc *rx_desc, *next_rxd;
249 	struct igbvf_buffer *buffer_info, *next_buffer;
250 	struct sk_buff *skb;
251 	bool cleaned = false;
252 	int cleaned_count = 0;
253 	unsigned int total_bytes = 0, total_packets = 0;
254 	unsigned int i;
255 	u32 length, hlen, staterr;
256 
257 	i = rx_ring->next_to_clean;
258 	rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
259 	staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
260 
261 	while (staterr & E1000_RXD_STAT_DD) {
262 		if (*work_done >= work_to_do)
263 			break;
264 		(*work_done)++;
265 		rmb(); /* read descriptor and rx_buffer_info after status DD */
266 
267 		buffer_info = &rx_ring->buffer_info[i];
268 
269 		/* HW will not DMA in data larger than the given buffer, even
270 		 * if it parses the (NFS, of course) header to be larger.  In
271 		 * that case, it fills the header buffer and spills the rest
272 		 * into the page.
273 		 */
274 		hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
275 		       & E1000_RXDADV_HDRBUFLEN_MASK) >>
276 		       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 			buffer_info->dma = 0;
301 			skb_put(skb, hlen);
302 		}
303 
304 		if (length) {
305 			dma_unmap_page(&pdev->dev, buffer_info->page_dma,
306 				       PAGE_SIZE / 2,
307 				       DMA_FROM_DEVICE);
308 			buffer_info->page_dma = 0;
309 
310 			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
311 					   buffer_info->page,
312 					   buffer_info->page_offset,
313 					   length);
314 
315 			if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
316 			    (page_count(buffer_info->page) != 1))
317 				buffer_info->page = NULL;
318 			else
319 				get_page(buffer_info->page);
320 
321 			skb->len += length;
322 			skb->data_len += length;
323 			skb->truesize += PAGE_SIZE / 2;
324 		}
325 send_up:
326 		i++;
327 		if (i == rx_ring->count)
328 			i = 0;
329 		next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
330 		prefetch(next_rxd);
331 		next_buffer = &rx_ring->buffer_info[i];
332 
333 		if (!(staterr & E1000_RXD_STAT_EOP)) {
334 			buffer_info->skb = next_buffer->skb;
335 			buffer_info->dma = next_buffer->dma;
336 			next_buffer->skb = skb;
337 			next_buffer->dma = 0;
338 			goto next_desc;
339 		}
340 
341 		if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
342 			dev_kfree_skb_irq(skb);
343 			goto next_desc;
344 		}
345 
346 		total_bytes += skb->len;
347 		total_packets++;
348 
349 		igbvf_rx_checksum_adv(adapter, staterr, skb);
350 
351 		skb->protocol = eth_type_trans(skb, netdev);
352 
353 		igbvf_receive_skb(adapter, netdev, skb, staterr,
354 				  rx_desc->wb.upper.vlan);
355 
356 next_desc:
357 		rx_desc->wb.upper.status_error = 0;
358 
359 		/* return some buffers to hardware, one at a time is too slow */
360 		if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
361 			igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
362 			cleaned_count = 0;
363 		}
364 
365 		/* use prefetched values */
366 		rx_desc = next_rxd;
367 		buffer_info = next_buffer;
368 
369 		staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
370 	}
371 
372 	rx_ring->next_to_clean = i;
373 	cleaned_count = igbvf_desc_unused(rx_ring);
374 
375 	if (cleaned_count)
376 		igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
377 
378 	adapter->total_rx_packets += total_packets;
379 	adapter->total_rx_bytes += total_bytes;
380 	netdev->stats.rx_bytes += total_bytes;
381 	netdev->stats.rx_packets += total_packets;
382 	return cleaned;
383 }
384 
385 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
386 			    struct igbvf_buffer *buffer_info)
387 {
388 	if (buffer_info->dma) {
389 		if (buffer_info->mapped_as_page)
390 			dma_unmap_page(&adapter->pdev->dev,
391 				       buffer_info->dma,
392 				       buffer_info->length,
393 				       DMA_TO_DEVICE);
394 		else
395 			dma_unmap_single(&adapter->pdev->dev,
396 					 buffer_info->dma,
397 					 buffer_info->length,
398 					 DMA_TO_DEVICE);
399 		buffer_info->dma = 0;
400 	}
401 	if (buffer_info->skb) {
402 		dev_kfree_skb_any(buffer_info->skb);
403 		buffer_info->skb = NULL;
404 	}
405 	buffer_info->time_stamp = 0;
406 }
407 
408 /**
409  * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
410  * @adapter: board private structure
411  *
412  * Return 0 on success, negative on failure
413  **/
414 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
415 			     struct igbvf_ring *tx_ring)
416 {
417 	struct pci_dev *pdev = adapter->pdev;
418 	int size;
419 
420 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
421 	tx_ring->buffer_info = vzalloc(size);
422 	if (!tx_ring->buffer_info)
423 		goto err;
424 
425 	/* round up to nearest 4K */
426 	tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
427 	tx_ring->size = ALIGN(tx_ring->size, 4096);
428 
429 	tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
430 					   &tx_ring->dma, GFP_KERNEL);
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;
439 err:
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  **/
452 int 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 	if (!rx_ring->desc)
472 		goto err;
473 
474 	rx_ring->next_to_clean = 0;
475 	rx_ring->next_to_use = 0;
476 
477 	rx_ring->adapter = adapter;
478 
479 	return 0;
480 
481 err:
482 	vfree(rx_ring->buffer_info);
483 	rx_ring->buffer_info = NULL;
484 	dev_err(&adapter->pdev->dev,
485 		"Unable to allocate memory for the receive descriptor ring\n");
486 	return -ENOMEM;
487 }
488 
489 /**
490  * igbvf_clean_tx_ring - Free Tx Buffers
491  * @tx_ring: ring to be cleaned
492  **/
493 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
494 {
495 	struct igbvf_adapter *adapter = tx_ring->adapter;
496 	struct igbvf_buffer *buffer_info;
497 	unsigned long size;
498 	unsigned int i;
499 
500 	if (!tx_ring->buffer_info)
501 		return;
502 
503 	/* Free all the Tx ring sk_buffs */
504 	for (i = 0; i < tx_ring->count; i++) {
505 		buffer_info = &tx_ring->buffer_info[i];
506 		igbvf_put_txbuf(adapter, buffer_info);
507 	}
508 
509 	size = sizeof(struct igbvf_buffer) * tx_ring->count;
510 	memset(tx_ring->buffer_info, 0, size);
511 
512 	/* Zero out the descriptor ring */
513 	memset(tx_ring->desc, 0, tx_ring->size);
514 
515 	tx_ring->next_to_use = 0;
516 	tx_ring->next_to_clean = 0;
517 
518 	writel(0, adapter->hw.hw_addr + tx_ring->head);
519 	writel(0, adapter->hw.hw_addr + tx_ring->tail);
520 }
521 
522 /**
523  * igbvf_free_tx_resources - Free Tx Resources per Queue
524  * @tx_ring: ring to free resources from
525  *
526  * Free all transmit software resources
527  **/
528 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
529 {
530 	struct pci_dev *pdev = tx_ring->adapter->pdev;
531 
532 	igbvf_clean_tx_ring(tx_ring);
533 
534 	vfree(tx_ring->buffer_info);
535 	tx_ring->buffer_info = NULL;
536 
537 	dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
538 			  tx_ring->dma);
539 
540 	tx_ring->desc = NULL;
541 }
542 
543 /**
544  * igbvf_clean_rx_ring - Free Rx Buffers per Queue
545  * @adapter: board private structure
546  **/
547 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
548 {
549 	struct igbvf_adapter *adapter = rx_ring->adapter;
550 	struct igbvf_buffer *buffer_info;
551 	struct pci_dev *pdev = adapter->pdev;
552 	unsigned long size;
553 	unsigned int i;
554 
555 	if (!rx_ring->buffer_info)
556 		return;
557 
558 	/* Free all the Rx ring sk_buffs */
559 	for (i = 0; i < rx_ring->count; i++) {
560 		buffer_info = &rx_ring->buffer_info[i];
561 		if (buffer_info->dma) {
562 			if (adapter->rx_ps_hdr_size) {
563 				dma_unmap_single(&pdev->dev, buffer_info->dma,
564 						 adapter->rx_ps_hdr_size,
565 						 DMA_FROM_DEVICE);
566 			} else {
567 				dma_unmap_single(&pdev->dev, buffer_info->dma,
568 						 adapter->rx_buffer_len,
569 						 DMA_FROM_DEVICE);
570 			}
571 			buffer_info->dma = 0;
572 		}
573 
574 		if (buffer_info->skb) {
575 			dev_kfree_skb(buffer_info->skb);
576 			buffer_info->skb = NULL;
577 		}
578 
579 		if (buffer_info->page) {
580 			if (buffer_info->page_dma)
581 				dma_unmap_page(&pdev->dev,
582 					       buffer_info->page_dma,
583 					       PAGE_SIZE / 2,
584 					       DMA_FROM_DEVICE);
585 			put_page(buffer_info->page);
586 			buffer_info->page = NULL;
587 			buffer_info->page_dma = 0;
588 			buffer_info->page_offset = 0;
589 		}
590 	}
591 
592 	size = sizeof(struct igbvf_buffer) * rx_ring->count;
593 	memset(rx_ring->buffer_info, 0, size);
594 
595 	/* Zero out the descriptor ring */
596 	memset(rx_ring->desc, 0, rx_ring->size);
597 
598 	rx_ring->next_to_clean = 0;
599 	rx_ring->next_to_use = 0;
600 
601 	writel(0, adapter->hw.hw_addr + rx_ring->head);
602 	writel(0, adapter->hw.hw_addr + rx_ring->tail);
603 }
604 
605 /**
606  * igbvf_free_rx_resources - Free Rx Resources
607  * @rx_ring: ring to clean the resources from
608  *
609  * Free all receive software resources
610  **/
611 
612 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
613 {
614 	struct pci_dev *pdev = rx_ring->adapter->pdev;
615 
616 	igbvf_clean_rx_ring(rx_ring);
617 
618 	vfree(rx_ring->buffer_info);
619 	rx_ring->buffer_info = NULL;
620 
621 	dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
622 			  rx_ring->dma);
623 	rx_ring->desc = NULL;
624 }
625 
626 /**
627  * igbvf_update_itr - update the dynamic ITR value based on statistics
628  * @adapter: pointer to adapter
629  * @itr_setting: current adapter->itr
630  * @packets: the number of packets during this measurement interval
631  * @bytes: the number of bytes during this measurement interval
632  *
633  * Stores a new ITR value based on packets and byte counts during the last
634  * interrupt.  The advantage of per interrupt computation is faster updates
635  * and more accurate ITR for the current traffic pattern.  Constants in this
636  * function were computed based on theoretical maximum wire speed and thresholds
637  * were set based on testing data as well as attempting to minimize response
638  * time while increasing bulk throughput.
639  **/
640 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
641 					   enum latency_range itr_setting,
642 					   int packets, int bytes)
643 {
644 	enum latency_range retval = itr_setting;
645 
646 	if (packets == 0)
647 		goto update_itr_done;
648 
649 	switch (itr_setting) {
650 	case lowest_latency:
651 		/* handle TSO and jumbo frames */
652 		if (bytes/packets > 8000)
653 			retval = bulk_latency;
654 		else if ((packets < 5) && (bytes > 512))
655 			retval = low_latency;
656 		break;
657 	case low_latency:  /* 50 usec aka 20000 ints/s */
658 		if (bytes > 10000) {
659 			/* this if handles the TSO accounting */
660 			if (bytes/packets > 8000)
661 				retval = bulk_latency;
662 			else if ((packets < 10) || ((bytes/packets) > 1200))
663 				retval = bulk_latency;
664 			else if ((packets > 35))
665 				retval = lowest_latency;
666 		} else if (bytes/packets > 2000) {
667 			retval = bulk_latency;
668 		} else if (packets <= 2 && bytes < 512) {
669 			retval = lowest_latency;
670 		}
671 		break;
672 	case bulk_latency: /* 250 usec aka 4000 ints/s */
673 		if (bytes > 25000) {
674 			if (packets > 35)
675 				retval = low_latency;
676 		} else if (bytes < 6000) {
677 			retval = low_latency;
678 		}
679 		break;
680 	default:
681 		break;
682 	}
683 
684 update_itr_done:
685 	return retval;
686 }
687 
688 static int igbvf_range_to_itr(enum latency_range current_range)
689 {
690 	int new_itr;
691 
692 	switch (current_range) {
693 	/* counts and packets in update_itr are dependent on these numbers */
694 	case lowest_latency:
695 		new_itr = IGBVF_70K_ITR;
696 		break;
697 	case low_latency:
698 		new_itr = IGBVF_20K_ITR;
699 		break;
700 	case bulk_latency:
701 		new_itr = IGBVF_4K_ITR;
702 		break;
703 	default:
704 		new_itr = IGBVF_START_ITR;
705 		break;
706 	}
707 	return new_itr;
708 }
709 
710 static void igbvf_set_itr(struct igbvf_adapter *adapter)
711 {
712 	u32 new_itr;
713 
714 	adapter->tx_ring->itr_range =
715 			igbvf_update_itr(adapter,
716 					 adapter->tx_ring->itr_val,
717 					 adapter->total_tx_packets,
718 					 adapter->total_tx_bytes);
719 
720 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
721 	if (adapter->requested_itr == 3 &&
722 	    adapter->tx_ring->itr_range == lowest_latency)
723 		adapter->tx_ring->itr_range = low_latency;
724 
725 	new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
726 
727 	if (new_itr != adapter->tx_ring->itr_val) {
728 		u32 current_itr = adapter->tx_ring->itr_val;
729 		/* this attempts to bias the interrupt rate towards Bulk
730 		 * by adding intermediate steps when interrupt rate is
731 		 * increasing
732 		 */
733 		new_itr = new_itr > current_itr ?
734 			  min(current_itr + (new_itr >> 2), new_itr) :
735 			  new_itr;
736 		adapter->tx_ring->itr_val = new_itr;
737 
738 		adapter->tx_ring->set_itr = 1;
739 	}
740 
741 	adapter->rx_ring->itr_range =
742 			igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
743 					 adapter->total_rx_packets,
744 					 adapter->total_rx_bytes);
745 	if (adapter->requested_itr == 3 &&
746 	    adapter->rx_ring->itr_range == lowest_latency)
747 		adapter->rx_ring->itr_range = low_latency;
748 
749 	new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
750 
751 	if (new_itr != adapter->rx_ring->itr_val) {
752 		u32 current_itr = adapter->rx_ring->itr_val;
753 
754 		new_itr = new_itr > current_itr ?
755 			  min(current_itr + (new_itr >> 2), new_itr) :
756 			  new_itr;
757 		adapter->rx_ring->itr_val = new_itr;
758 
759 		adapter->rx_ring->set_itr = 1;
760 	}
761 }
762 
763 /**
764  * igbvf_clean_tx_irq - Reclaim resources after transmit completes
765  * @adapter: board private structure
766  *
767  * returns true if ring is completely cleaned
768  **/
769 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
770 {
771 	struct igbvf_adapter *adapter = tx_ring->adapter;
772 	struct net_device *netdev = adapter->netdev;
773 	struct igbvf_buffer *buffer_info;
774 	struct sk_buff *skb;
775 	union e1000_adv_tx_desc *tx_desc, *eop_desc;
776 	unsigned int total_bytes = 0, total_packets = 0;
777 	unsigned int i, count = 0;
778 	bool cleaned = false;
779 
780 	i = tx_ring->next_to_clean;
781 	buffer_info = &tx_ring->buffer_info[i];
782 	eop_desc = buffer_info->next_to_watch;
783 
784 	do {
785 		/* if next_to_watch is not set then there is no work pending */
786 		if (!eop_desc)
787 			break;
788 
789 		/* prevent any other reads prior to eop_desc */
790 		smp_rmb();
791 
792 		/* if DD is not set pending work has not been completed */
793 		if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
794 			break;
795 
796 		/* clear next_to_watch to prevent false hangs */
797 		buffer_info->next_to_watch = NULL;
798 
799 		for (cleaned = false; !cleaned; count++) {
800 			tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
801 			cleaned = (tx_desc == eop_desc);
802 			skb = buffer_info->skb;
803 
804 			if (skb) {
805 				unsigned int segs, bytecount;
806 
807 				/* gso_segs is currently only valid for tcp */
808 				segs = skb_shinfo(skb)->gso_segs ?: 1;
809 				/* multiply data chunks by size of headers */
810 				bytecount = ((segs - 1) * skb_headlen(skb)) +
811 					    skb->len;
812 				total_packets += segs;
813 				total_bytes += bytecount;
814 			}
815 
816 			igbvf_put_txbuf(adapter, buffer_info);
817 			tx_desc->wb.status = 0;
818 
819 			i++;
820 			if (i == tx_ring->count)
821 				i = 0;
822 
823 			buffer_info = &tx_ring->buffer_info[i];
824 		}
825 
826 		eop_desc = buffer_info->next_to_watch;
827 	} while (count < tx_ring->count);
828 
829 	tx_ring->next_to_clean = i;
830 
831 	if (unlikely(count && netif_carrier_ok(netdev) &&
832 	    igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
833 		/* Make sure that anybody stopping the queue after this
834 		 * sees the new next_to_clean.
835 		 */
836 		smp_mb();
837 		if (netif_queue_stopped(netdev) &&
838 		    !(test_bit(__IGBVF_DOWN, &adapter->state))) {
839 			netif_wake_queue(netdev);
840 			++adapter->restart_queue;
841 		}
842 	}
843 
844 	netdev->stats.tx_bytes += total_bytes;
845 	netdev->stats.tx_packets += total_packets;
846 	return count < tx_ring->count;
847 }
848 
849 static irqreturn_t igbvf_msix_other(int irq, void *data)
850 {
851 	struct net_device *netdev = data;
852 	struct igbvf_adapter *adapter = netdev_priv(netdev);
853 	struct e1000_hw *hw = &adapter->hw;
854 
855 	adapter->int_counter1++;
856 
857 	hw->mac.get_link_status = 1;
858 	if (!test_bit(__IGBVF_DOWN, &adapter->state))
859 		mod_timer(&adapter->watchdog_timer, jiffies + 1);
860 
861 	ew32(EIMS, adapter->eims_other);
862 
863 	return IRQ_HANDLED;
864 }
865 
866 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
867 {
868 	struct net_device *netdev = data;
869 	struct igbvf_adapter *adapter = netdev_priv(netdev);
870 	struct e1000_hw *hw = &adapter->hw;
871 	struct igbvf_ring *tx_ring = adapter->tx_ring;
872 
873 	if (tx_ring->set_itr) {
874 		writel(tx_ring->itr_val,
875 		       adapter->hw.hw_addr + tx_ring->itr_register);
876 		adapter->tx_ring->set_itr = 0;
877 	}
878 
879 	adapter->total_tx_bytes = 0;
880 	adapter->total_tx_packets = 0;
881 
882 	/* auto mask will automatically re-enable the interrupt when we write
883 	 * EICS
884 	 */
885 	if (!igbvf_clean_tx_irq(tx_ring))
886 		/* Ring was not completely cleaned, so fire another interrupt */
887 		ew32(EICS, tx_ring->eims_value);
888 	else
889 		ew32(EIMS, tx_ring->eims_value);
890 
891 	return IRQ_HANDLED;
892 }
893 
894 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
895 {
896 	struct net_device *netdev = data;
897 	struct igbvf_adapter *adapter = netdev_priv(netdev);
898 
899 	adapter->int_counter0++;
900 
901 	/* Write the ITR value calculated at the end of the
902 	 * previous interrupt.
903 	 */
904 	if (adapter->rx_ring->set_itr) {
905 		writel(adapter->rx_ring->itr_val,
906 		       adapter->hw.hw_addr + adapter->rx_ring->itr_register);
907 		adapter->rx_ring->set_itr = 0;
908 	}
909 
910 	if (napi_schedule_prep(&adapter->rx_ring->napi)) {
911 		adapter->total_rx_bytes = 0;
912 		adapter->total_rx_packets = 0;
913 		__napi_schedule(&adapter->rx_ring->napi);
914 	}
915 
916 	return IRQ_HANDLED;
917 }
918 
919 #define IGBVF_NO_QUEUE -1
920 
921 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
922 				int tx_queue, int msix_vector)
923 {
924 	struct e1000_hw *hw = &adapter->hw;
925 	u32 ivar, index;
926 
927 	/* 82576 uses a table-based method for assigning vectors.
928 	 * Each queue has a single entry in the table to which we write
929 	 * a vector number along with a "valid" bit.  Sadly, the layout
930 	 * of the table is somewhat counterintuitive.
931 	 */
932 	if (rx_queue > IGBVF_NO_QUEUE) {
933 		index = (rx_queue >> 1);
934 		ivar = array_er32(IVAR0, index);
935 		if (rx_queue & 0x1) {
936 			/* vector goes into third byte of register */
937 			ivar = ivar & 0xFF00FFFF;
938 			ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
939 		} else {
940 			/* vector goes into low byte of register */
941 			ivar = ivar & 0xFFFFFF00;
942 			ivar |= msix_vector | E1000_IVAR_VALID;
943 		}
944 		adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
945 		array_ew32(IVAR0, index, ivar);
946 	}
947 	if (tx_queue > IGBVF_NO_QUEUE) {
948 		index = (tx_queue >> 1);
949 		ivar = array_er32(IVAR0, index);
950 		if (tx_queue & 0x1) {
951 			/* vector goes into high byte of register */
952 			ivar = ivar & 0x00FFFFFF;
953 			ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
954 		} else {
955 			/* vector goes into second byte of register */
956 			ivar = ivar & 0xFFFF00FF;
957 			ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
958 		}
959 		adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
960 		array_ew32(IVAR0, index, ivar);
961 	}
962 }
963 
964 /**
965  * igbvf_configure_msix - Configure MSI-X hardware
966  * @adapter: board private structure
967  *
968  * igbvf_configure_msix sets up the hardware to properly
969  * generate MSI-X interrupts.
970  **/
971 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
972 {
973 	u32 tmp;
974 	struct e1000_hw *hw = &adapter->hw;
975 	struct igbvf_ring *tx_ring = adapter->tx_ring;
976 	struct igbvf_ring *rx_ring = adapter->rx_ring;
977 	int vector = 0;
978 
979 	adapter->eims_enable_mask = 0;
980 
981 	igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
982 	adapter->eims_enable_mask |= tx_ring->eims_value;
983 	writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
984 	igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
985 	adapter->eims_enable_mask |= rx_ring->eims_value;
986 	writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
987 
988 	/* set vector for other causes, i.e. link changes */
989 
990 	tmp = (vector++ | E1000_IVAR_VALID);
991 
992 	ew32(IVAR_MISC, tmp);
993 
994 	adapter->eims_enable_mask = GENMASK(vector - 1, 0);
995 	adapter->eims_other = BIT(vector - 1);
996 	e1e_flush();
997 }
998 
999 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1000 {
1001 	if (adapter->msix_entries) {
1002 		pci_disable_msix(adapter->pdev);
1003 		kfree(adapter->msix_entries);
1004 		adapter->msix_entries = NULL;
1005 	}
1006 }
1007 
1008 /**
1009  * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1010  * @adapter: board private structure
1011  *
1012  * Attempt to configure interrupts using the best available
1013  * capabilities of the hardware and kernel.
1014  **/
1015 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1016 {
1017 	int err = -ENOMEM;
1018 	int i;
1019 
1020 	/* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1021 	adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1022 					GFP_KERNEL);
1023 	if (adapter->msix_entries) {
1024 		for (i = 0; i < 3; i++)
1025 			adapter->msix_entries[i].entry = i;
1026 
1027 		err = pci_enable_msix_range(adapter->pdev,
1028 					    adapter->msix_entries, 3, 3);
1029 	}
1030 
1031 	if (err < 0) {
1032 		/* MSI-X failed */
1033 		dev_err(&adapter->pdev->dev,
1034 			"Failed to initialize MSI-X interrupts.\n");
1035 		igbvf_reset_interrupt_capability(adapter);
1036 	}
1037 }
1038 
1039 /**
1040  * igbvf_request_msix - Initialize MSI-X interrupts
1041  * @adapter: board private structure
1042  *
1043  * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1044  * kernel.
1045  **/
1046 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1047 {
1048 	struct net_device *netdev = adapter->netdev;
1049 	int err = 0, vector = 0;
1050 
1051 	if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1052 		sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1053 		sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1054 	} else {
1055 		memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1056 		memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1057 	}
1058 
1059 	err = request_irq(adapter->msix_entries[vector].vector,
1060 			  igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1061 			  netdev);
1062 	if (err)
1063 		goto out;
1064 
1065 	adapter->tx_ring->itr_register = E1000_EITR(vector);
1066 	adapter->tx_ring->itr_val = adapter->current_itr;
1067 	vector++;
1068 
1069 	err = request_irq(adapter->msix_entries[vector].vector,
1070 			  igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1071 			  netdev);
1072 	if (err)
1073 		goto out;
1074 
1075 	adapter->rx_ring->itr_register = E1000_EITR(vector);
1076 	adapter->rx_ring->itr_val = adapter->current_itr;
1077 	vector++;
1078 
1079 	err = request_irq(adapter->msix_entries[vector].vector,
1080 			  igbvf_msix_other, 0, netdev->name, netdev);
1081 	if (err)
1082 		goto out;
1083 
1084 	igbvf_configure_msix(adapter);
1085 	return 0;
1086 out:
1087 	return err;
1088 }
1089 
1090 /**
1091  * igbvf_alloc_queues - Allocate memory for all rings
1092  * @adapter: board private structure to initialize
1093  **/
1094 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1095 {
1096 	struct net_device *netdev = adapter->netdev;
1097 
1098 	adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1099 	if (!adapter->tx_ring)
1100 		return -ENOMEM;
1101 
1102 	adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1103 	if (!adapter->rx_ring) {
1104 		kfree(adapter->tx_ring);
1105 		return -ENOMEM;
1106 	}
1107 
1108 	netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1109 
1110 	return 0;
1111 }
1112 
1113 /**
1114  * igbvf_request_irq - initialize interrupts
1115  * @adapter: board private structure
1116  *
1117  * Attempts to configure interrupts using the best available
1118  * capabilities of the hardware and kernel.
1119  **/
1120 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1121 {
1122 	int err = -1;
1123 
1124 	/* igbvf supports msi-x only */
1125 	if (adapter->msix_entries)
1126 		err = igbvf_request_msix(adapter);
1127 
1128 	if (!err)
1129 		return err;
1130 
1131 	dev_err(&adapter->pdev->dev,
1132 		"Unable to allocate interrupt, Error: %d\n", err);
1133 
1134 	return err;
1135 }
1136 
1137 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1138 {
1139 	struct net_device *netdev = adapter->netdev;
1140 	int vector;
1141 
1142 	if (adapter->msix_entries) {
1143 		for (vector = 0; vector < 3; vector++)
1144 			free_irq(adapter->msix_entries[vector].vector, netdev);
1145 	}
1146 }
1147 
1148 /**
1149  * igbvf_irq_disable - Mask off interrupt generation on the NIC
1150  * @adapter: board private structure
1151  **/
1152 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1153 {
1154 	struct e1000_hw *hw = &adapter->hw;
1155 
1156 	ew32(EIMC, ~0);
1157 
1158 	if (adapter->msix_entries)
1159 		ew32(EIAC, 0);
1160 }
1161 
1162 /**
1163  * igbvf_irq_enable - Enable default interrupt generation settings
1164  * @adapter: board private structure
1165  **/
1166 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1167 {
1168 	struct e1000_hw *hw = &adapter->hw;
1169 
1170 	ew32(EIAC, adapter->eims_enable_mask);
1171 	ew32(EIAM, adapter->eims_enable_mask);
1172 	ew32(EIMS, adapter->eims_enable_mask);
1173 }
1174 
1175 /**
1176  * igbvf_poll - NAPI Rx polling callback
1177  * @napi: struct associated with this polling callback
1178  * @budget: amount of packets driver is allowed to process this poll
1179  **/
1180 static int igbvf_poll(struct napi_struct *napi, int budget)
1181 {
1182 	struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1183 	struct igbvf_adapter *adapter = rx_ring->adapter;
1184 	struct e1000_hw *hw = &adapter->hw;
1185 	int work_done = 0;
1186 
1187 	igbvf_clean_rx_irq(adapter, &work_done, budget);
1188 
1189 	/* If not enough Rx work done, exit the polling mode */
1190 	if (work_done < budget) {
1191 		napi_complete_done(napi, work_done);
1192 
1193 		if (adapter->requested_itr & 3)
1194 			igbvf_set_itr(adapter);
1195 
1196 		if (!test_bit(__IGBVF_DOWN, &adapter->state))
1197 			ew32(EIMS, adapter->rx_ring->eims_value);
1198 	}
1199 
1200 	return work_done;
1201 }
1202 
1203 /**
1204  * igbvf_set_rlpml - set receive large packet maximum length
1205  * @adapter: board private structure
1206  *
1207  * Configure the maximum size of packets that will be received
1208  */
1209 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1210 {
1211 	int max_frame_size;
1212 	struct e1000_hw *hw = &adapter->hw;
1213 
1214 	max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1215 
1216 	spin_lock_bh(&hw->mbx_lock);
1217 
1218 	e1000_rlpml_set_vf(hw, max_frame_size);
1219 
1220 	spin_unlock_bh(&hw->mbx_lock);
1221 }
1222 
1223 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1224 				 __be16 proto, u16 vid)
1225 {
1226 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1227 	struct e1000_hw *hw = &adapter->hw;
1228 
1229 	spin_lock_bh(&hw->mbx_lock);
1230 
1231 	if (hw->mac.ops.set_vfta(hw, vid, true)) {
1232 		dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1233 		spin_unlock_bh(&hw->mbx_lock);
1234 		return -EINVAL;
1235 	}
1236 
1237 	spin_unlock_bh(&hw->mbx_lock);
1238 
1239 	set_bit(vid, adapter->active_vlans);
1240 	return 0;
1241 }
1242 
1243 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1244 				  __be16 proto, u16 vid)
1245 {
1246 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1247 	struct e1000_hw *hw = &adapter->hw;
1248 
1249 	spin_lock_bh(&hw->mbx_lock);
1250 
1251 	if (hw->mac.ops.set_vfta(hw, vid, false)) {
1252 		dev_err(&adapter->pdev->dev,
1253 			"Failed to remove vlan id %d\n", vid);
1254 		spin_unlock_bh(&hw->mbx_lock);
1255 		return -EINVAL;
1256 	}
1257 
1258 	spin_unlock_bh(&hw->mbx_lock);
1259 
1260 	clear_bit(vid, adapter->active_vlans);
1261 	return 0;
1262 }
1263 
1264 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1265 {
1266 	u16 vid;
1267 
1268 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1269 		igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1270 }
1271 
1272 /**
1273  * igbvf_configure_tx - Configure Transmit Unit after Reset
1274  * @adapter: board private structure
1275  *
1276  * Configure the Tx unit of the MAC after a reset.
1277  **/
1278 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1279 {
1280 	struct e1000_hw *hw = &adapter->hw;
1281 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1282 	u64 tdba;
1283 	u32 txdctl, dca_txctrl;
1284 
1285 	/* disable transmits */
1286 	txdctl = er32(TXDCTL(0));
1287 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1288 	e1e_flush();
1289 	msleep(10);
1290 
1291 	/* Setup the HW Tx Head and Tail descriptor pointers */
1292 	ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1293 	tdba = tx_ring->dma;
1294 	ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1295 	ew32(TDBAH(0), (tdba >> 32));
1296 	ew32(TDH(0), 0);
1297 	ew32(TDT(0), 0);
1298 	tx_ring->head = E1000_TDH(0);
1299 	tx_ring->tail = E1000_TDT(0);
1300 
1301 	/* Turn off Relaxed Ordering on head write-backs.  The writebacks
1302 	 * MUST be delivered in order or it will completely screw up
1303 	 * our bookkeeping.
1304 	 */
1305 	dca_txctrl = er32(DCA_TXCTRL(0));
1306 	dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1307 	ew32(DCA_TXCTRL(0), dca_txctrl);
1308 
1309 	/* enable transmits */
1310 	txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1311 	ew32(TXDCTL(0), txdctl);
1312 
1313 	/* Setup Transmit Descriptor Settings for eop descriptor */
1314 	adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1315 
1316 	/* enable Report Status bit */
1317 	adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1318 }
1319 
1320 /**
1321  * igbvf_setup_srrctl - configure the receive control registers
1322  * @adapter: Board private structure
1323  **/
1324 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1325 {
1326 	struct e1000_hw *hw = &adapter->hw;
1327 	u32 srrctl = 0;
1328 
1329 	srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1330 		    E1000_SRRCTL_BSIZEHDR_MASK |
1331 		    E1000_SRRCTL_BSIZEPKT_MASK);
1332 
1333 	/* Enable queue drop to avoid head of line blocking */
1334 	srrctl |= E1000_SRRCTL_DROP_EN;
1335 
1336 	/* Setup buffer sizes */
1337 	srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1338 		  E1000_SRRCTL_BSIZEPKT_SHIFT;
1339 
1340 	if (adapter->rx_buffer_len < 2048) {
1341 		adapter->rx_ps_hdr_size = 0;
1342 		srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1343 	} else {
1344 		adapter->rx_ps_hdr_size = 128;
1345 		srrctl |= adapter->rx_ps_hdr_size <<
1346 			  E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1347 		srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1348 	}
1349 
1350 	ew32(SRRCTL(0), srrctl);
1351 }
1352 
1353 /**
1354  * igbvf_configure_rx - Configure Receive Unit after Reset
1355  * @adapter: board private structure
1356  *
1357  * Configure the Rx unit of the MAC after a reset.
1358  **/
1359 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1360 {
1361 	struct e1000_hw *hw = &adapter->hw;
1362 	struct igbvf_ring *rx_ring = adapter->rx_ring;
1363 	u64 rdba;
1364 	u32 rxdctl;
1365 
1366 	/* disable receives */
1367 	rxdctl = er32(RXDCTL(0));
1368 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1369 	e1e_flush();
1370 	msleep(10);
1371 
1372 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
1373 	 * the Base and Length of the Rx Descriptor Ring
1374 	 */
1375 	rdba = rx_ring->dma;
1376 	ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1377 	ew32(RDBAH(0), (rdba >> 32));
1378 	ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1379 	rx_ring->head = E1000_RDH(0);
1380 	rx_ring->tail = E1000_RDT(0);
1381 	ew32(RDH(0), 0);
1382 	ew32(RDT(0), 0);
1383 
1384 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1385 	rxdctl &= 0xFFF00000;
1386 	rxdctl |= IGBVF_RX_PTHRESH;
1387 	rxdctl |= IGBVF_RX_HTHRESH << 8;
1388 	rxdctl |= IGBVF_RX_WTHRESH << 16;
1389 
1390 	igbvf_set_rlpml(adapter);
1391 
1392 	/* enable receives */
1393 	ew32(RXDCTL(0), rxdctl);
1394 }
1395 
1396 /**
1397  * igbvf_set_multi - Multicast and Promiscuous mode set
1398  * @netdev: network interface device structure
1399  *
1400  * The set_multi entry point is called whenever the multicast address
1401  * list or the network interface flags are updated.  This routine is
1402  * responsible for configuring the hardware for proper multicast,
1403  * promiscuous mode, and all-multi behavior.
1404  **/
1405 static void igbvf_set_multi(struct net_device *netdev)
1406 {
1407 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1408 	struct e1000_hw *hw = &adapter->hw;
1409 	struct netdev_hw_addr *ha;
1410 	u8  *mta_list = NULL;
1411 	int i;
1412 
1413 	if (!netdev_mc_empty(netdev)) {
1414 		mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1415 					 GFP_ATOMIC);
1416 		if (!mta_list)
1417 			return;
1418 	}
1419 
1420 	/* prepare a packed array of only addresses. */
1421 	i = 0;
1422 	netdev_for_each_mc_addr(ha, netdev)
1423 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1424 
1425 	spin_lock_bh(&hw->mbx_lock);
1426 
1427 	hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1428 
1429 	spin_unlock_bh(&hw->mbx_lock);
1430 	kfree(mta_list);
1431 }
1432 
1433 /**
1434  * igbvf_set_uni - Configure unicast MAC filters
1435  * @netdev: network interface device structure
1436  *
1437  * This routine is responsible for configuring the hardware for proper
1438  * unicast filters.
1439  **/
1440 static int igbvf_set_uni(struct net_device *netdev)
1441 {
1442 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1443 	struct e1000_hw *hw = &adapter->hw;
1444 
1445 	if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1446 		pr_err("Too many unicast filters - No Space\n");
1447 		return -ENOSPC;
1448 	}
1449 
1450 	spin_lock_bh(&hw->mbx_lock);
1451 
1452 	/* Clear all unicast MAC filters */
1453 	hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1454 
1455 	spin_unlock_bh(&hw->mbx_lock);
1456 
1457 	if (!netdev_uc_empty(netdev)) {
1458 		struct netdev_hw_addr *ha;
1459 
1460 		/* Add MAC filters one by one */
1461 		netdev_for_each_uc_addr(ha, netdev) {
1462 			spin_lock_bh(&hw->mbx_lock);
1463 
1464 			hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1465 						ha->addr);
1466 
1467 			spin_unlock_bh(&hw->mbx_lock);
1468 			udelay(200);
1469 		}
1470 	}
1471 
1472 	return 0;
1473 }
1474 
1475 static void igbvf_set_rx_mode(struct net_device *netdev)
1476 {
1477 	igbvf_set_multi(netdev);
1478 	igbvf_set_uni(netdev);
1479 }
1480 
1481 /**
1482  * igbvf_configure - configure the hardware for Rx and Tx
1483  * @adapter: private board structure
1484  **/
1485 static void igbvf_configure(struct igbvf_adapter *adapter)
1486 {
1487 	igbvf_set_rx_mode(adapter->netdev);
1488 
1489 	igbvf_restore_vlan(adapter);
1490 
1491 	igbvf_configure_tx(adapter);
1492 	igbvf_setup_srrctl(adapter);
1493 	igbvf_configure_rx(adapter);
1494 	igbvf_alloc_rx_buffers(adapter->rx_ring,
1495 			       igbvf_desc_unused(adapter->rx_ring));
1496 }
1497 
1498 /* igbvf_reset - bring the hardware into a known good state
1499  * @adapter: private board structure
1500  *
1501  * This function boots the hardware and enables some settings that
1502  * require a configuration cycle of the hardware - those cannot be
1503  * set/changed during runtime. After reset the device needs to be
1504  * properly configured for Rx, Tx etc.
1505  */
1506 static void igbvf_reset(struct igbvf_adapter *adapter)
1507 {
1508 	struct e1000_mac_info *mac = &adapter->hw.mac;
1509 	struct net_device *netdev = adapter->netdev;
1510 	struct e1000_hw *hw = &adapter->hw;
1511 
1512 	spin_lock_bh(&hw->mbx_lock);
1513 
1514 	/* Allow time for pending master requests to run */
1515 	if (mac->ops.reset_hw(hw))
1516 		dev_err(&adapter->pdev->dev, "PF still resetting\n");
1517 
1518 	mac->ops.init_hw(hw);
1519 
1520 	spin_unlock_bh(&hw->mbx_lock);
1521 
1522 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1523 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1524 		       netdev->addr_len);
1525 		memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1526 		       netdev->addr_len);
1527 	}
1528 
1529 	adapter->last_reset = jiffies;
1530 }
1531 
1532 int igbvf_up(struct igbvf_adapter *adapter)
1533 {
1534 	struct e1000_hw *hw = &adapter->hw;
1535 
1536 	/* hardware has been reset, we need to reload some things */
1537 	igbvf_configure(adapter);
1538 
1539 	clear_bit(__IGBVF_DOWN, &adapter->state);
1540 
1541 	napi_enable(&adapter->rx_ring->napi);
1542 	if (adapter->msix_entries)
1543 		igbvf_configure_msix(adapter);
1544 
1545 	/* Clear any pending interrupts. */
1546 	er32(EICR);
1547 	igbvf_irq_enable(adapter);
1548 
1549 	/* start the watchdog */
1550 	hw->mac.get_link_status = 1;
1551 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1552 
1553 	return 0;
1554 }
1555 
1556 void igbvf_down(struct igbvf_adapter *adapter)
1557 {
1558 	struct net_device *netdev = adapter->netdev;
1559 	struct e1000_hw *hw = &adapter->hw;
1560 	u32 rxdctl, txdctl;
1561 
1562 	/* signal that we're down so the interrupt handler does not
1563 	 * reschedule our watchdog timer
1564 	 */
1565 	set_bit(__IGBVF_DOWN, &adapter->state);
1566 
1567 	/* disable receives in the hardware */
1568 	rxdctl = er32(RXDCTL(0));
1569 	ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1570 
1571 	netif_carrier_off(netdev);
1572 	netif_stop_queue(netdev);
1573 
1574 	/* disable transmits in the hardware */
1575 	txdctl = er32(TXDCTL(0));
1576 	ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1577 
1578 	/* flush both disables and wait for them to finish */
1579 	e1e_flush();
1580 	msleep(10);
1581 
1582 	napi_disable(&adapter->rx_ring->napi);
1583 
1584 	igbvf_irq_disable(adapter);
1585 
1586 	del_timer_sync(&adapter->watchdog_timer);
1587 
1588 	/* record the stats before reset*/
1589 	igbvf_update_stats(adapter);
1590 
1591 	adapter->link_speed = 0;
1592 	adapter->link_duplex = 0;
1593 
1594 	igbvf_reset(adapter);
1595 	igbvf_clean_tx_ring(adapter->tx_ring);
1596 	igbvf_clean_rx_ring(adapter->rx_ring);
1597 }
1598 
1599 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1600 {
1601 	might_sleep();
1602 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1603 		usleep_range(1000, 2000);
1604 	igbvf_down(adapter);
1605 	igbvf_up(adapter);
1606 	clear_bit(__IGBVF_RESETTING, &adapter->state);
1607 }
1608 
1609 /**
1610  * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1611  * @adapter: board private structure to initialize
1612  *
1613  * igbvf_sw_init initializes the Adapter private data structure.
1614  * Fields are initialized based on PCI device information and
1615  * OS network device settings (MTU size).
1616  **/
1617 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1618 {
1619 	struct net_device *netdev = adapter->netdev;
1620 	s32 rc;
1621 
1622 	adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1623 	adapter->rx_ps_hdr_size = 0;
1624 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1625 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1626 
1627 	adapter->tx_int_delay = 8;
1628 	adapter->tx_abs_int_delay = 32;
1629 	adapter->rx_int_delay = 0;
1630 	adapter->rx_abs_int_delay = 8;
1631 	adapter->requested_itr = 3;
1632 	adapter->current_itr = IGBVF_START_ITR;
1633 
1634 	/* Set various function pointers */
1635 	adapter->ei->init_ops(&adapter->hw);
1636 
1637 	rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1638 	if (rc)
1639 		return rc;
1640 
1641 	rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1642 	if (rc)
1643 		return rc;
1644 
1645 	igbvf_set_interrupt_capability(adapter);
1646 
1647 	if (igbvf_alloc_queues(adapter))
1648 		return -ENOMEM;
1649 
1650 	spin_lock_init(&adapter->tx_queue_lock);
1651 
1652 	/* Explicitly disable IRQ since the NIC can be in any state. */
1653 	igbvf_irq_disable(adapter);
1654 
1655 	spin_lock_init(&adapter->stats_lock);
1656 	spin_lock_init(&adapter->hw.mbx_lock);
1657 
1658 	set_bit(__IGBVF_DOWN, &adapter->state);
1659 	return 0;
1660 }
1661 
1662 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1663 {
1664 	struct e1000_hw *hw = &adapter->hw;
1665 
1666 	adapter->stats.last_gprc = er32(VFGPRC);
1667 	adapter->stats.last_gorc = er32(VFGORC);
1668 	adapter->stats.last_gptc = er32(VFGPTC);
1669 	adapter->stats.last_gotc = er32(VFGOTC);
1670 	adapter->stats.last_mprc = er32(VFMPRC);
1671 	adapter->stats.last_gotlbc = er32(VFGOTLBC);
1672 	adapter->stats.last_gptlbc = er32(VFGPTLBC);
1673 	adapter->stats.last_gorlbc = er32(VFGORLBC);
1674 	adapter->stats.last_gprlbc = er32(VFGPRLBC);
1675 
1676 	adapter->stats.base_gprc = er32(VFGPRC);
1677 	adapter->stats.base_gorc = er32(VFGORC);
1678 	adapter->stats.base_gptc = er32(VFGPTC);
1679 	adapter->stats.base_gotc = er32(VFGOTC);
1680 	adapter->stats.base_mprc = er32(VFMPRC);
1681 	adapter->stats.base_gotlbc = er32(VFGOTLBC);
1682 	adapter->stats.base_gptlbc = er32(VFGPTLBC);
1683 	adapter->stats.base_gorlbc = er32(VFGORLBC);
1684 	adapter->stats.base_gprlbc = er32(VFGPRLBC);
1685 }
1686 
1687 /**
1688  * igbvf_open - Called when a network interface is made active
1689  * @netdev: network interface device structure
1690  *
1691  * Returns 0 on success, negative value on failure
1692  *
1693  * The open entry point is called when a network interface is made
1694  * active by the system (IFF_UP).  At this point all resources needed
1695  * for transmit and receive operations are allocated, the interrupt
1696  * handler is registered with the OS, the watchdog timer is started,
1697  * and the stack is notified that the interface is ready.
1698  **/
1699 static int igbvf_open(struct net_device *netdev)
1700 {
1701 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1702 	struct e1000_hw *hw = &adapter->hw;
1703 	int err;
1704 
1705 	/* disallow open during test */
1706 	if (test_bit(__IGBVF_TESTING, &adapter->state))
1707 		return -EBUSY;
1708 
1709 	/* allocate transmit descriptors */
1710 	err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1711 	if (err)
1712 		goto err_setup_tx;
1713 
1714 	/* allocate receive descriptors */
1715 	err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1716 	if (err)
1717 		goto err_setup_rx;
1718 
1719 	/* before we allocate an interrupt, we must be ready to handle it.
1720 	 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1721 	 * as soon as we call pci_request_irq, so we have to setup our
1722 	 * clean_rx handler before we do so.
1723 	 */
1724 	igbvf_configure(adapter);
1725 
1726 	err = igbvf_request_irq(adapter);
1727 	if (err)
1728 		goto err_req_irq;
1729 
1730 	/* From here on the code is the same as igbvf_up() */
1731 	clear_bit(__IGBVF_DOWN, &adapter->state);
1732 
1733 	napi_enable(&adapter->rx_ring->napi);
1734 
1735 	/* clear any pending interrupts */
1736 	er32(EICR);
1737 
1738 	igbvf_irq_enable(adapter);
1739 
1740 	/* start the watchdog */
1741 	hw->mac.get_link_status = 1;
1742 	mod_timer(&adapter->watchdog_timer, jiffies + 1);
1743 
1744 	return 0;
1745 
1746 err_req_irq:
1747 	igbvf_free_rx_resources(adapter->rx_ring);
1748 err_setup_rx:
1749 	igbvf_free_tx_resources(adapter->tx_ring);
1750 err_setup_tx:
1751 	igbvf_reset(adapter);
1752 
1753 	return err;
1754 }
1755 
1756 /**
1757  * igbvf_close - Disables a network interface
1758  * @netdev: network interface device structure
1759  *
1760  * Returns 0, this is not allowed to fail
1761  *
1762  * The close entry point is called when an interface is de-activated
1763  * by the OS.  The hardware is still under the drivers control, but
1764  * needs to be disabled.  A global MAC reset is issued to stop the
1765  * hardware, and all transmit and receive resources are freed.
1766  **/
1767 static int igbvf_close(struct net_device *netdev)
1768 {
1769 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1770 
1771 	WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1772 	igbvf_down(adapter);
1773 
1774 	igbvf_free_irq(adapter);
1775 
1776 	igbvf_free_tx_resources(adapter->tx_ring);
1777 	igbvf_free_rx_resources(adapter->rx_ring);
1778 
1779 	return 0;
1780 }
1781 
1782 /**
1783  * igbvf_set_mac - Change the Ethernet Address of the NIC
1784  * @netdev: network interface device structure
1785  * @p: pointer to an address structure
1786  *
1787  * Returns 0 on success, negative on failure
1788  **/
1789 static int igbvf_set_mac(struct net_device *netdev, void *p)
1790 {
1791 	struct igbvf_adapter *adapter = netdev_priv(netdev);
1792 	struct e1000_hw *hw = &adapter->hw;
1793 	struct sockaddr *addr = p;
1794 
1795 	if (!is_valid_ether_addr(addr->sa_data))
1796 		return -EADDRNOTAVAIL;
1797 
1798 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1799 
1800 	spin_lock_bh(&hw->mbx_lock);
1801 
1802 	hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1803 
1804 	spin_unlock_bh(&hw->mbx_lock);
1805 
1806 	if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1807 		return -EADDRNOTAVAIL;
1808 
1809 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1810 
1811 	return 0;
1812 }
1813 
1814 #define UPDATE_VF_COUNTER(reg, name) \
1815 { \
1816 	u32 current_counter = er32(reg); \
1817 	if (current_counter < adapter->stats.last_##name) \
1818 		adapter->stats.name += 0x100000000LL; \
1819 	adapter->stats.last_##name = current_counter; \
1820 	adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1821 	adapter->stats.name |= current_counter; \
1822 }
1823 
1824 /**
1825  * igbvf_update_stats - Update the board statistics counters
1826  * @adapter: board private structure
1827 **/
1828 void igbvf_update_stats(struct igbvf_adapter *adapter)
1829 {
1830 	struct e1000_hw *hw = &adapter->hw;
1831 	struct pci_dev *pdev = adapter->pdev;
1832 
1833 	/* Prevent stats update while adapter is being reset, link is down
1834 	 * or if the pci connection is down.
1835 	 */
1836 	if (adapter->link_speed == 0)
1837 		return;
1838 
1839 	if (test_bit(__IGBVF_RESETTING, &adapter->state))
1840 		return;
1841 
1842 	if (pci_channel_offline(pdev))
1843 		return;
1844 
1845 	UPDATE_VF_COUNTER(VFGPRC, gprc);
1846 	UPDATE_VF_COUNTER(VFGORC, gorc);
1847 	UPDATE_VF_COUNTER(VFGPTC, gptc);
1848 	UPDATE_VF_COUNTER(VFGOTC, gotc);
1849 	UPDATE_VF_COUNTER(VFMPRC, mprc);
1850 	UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1851 	UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1852 	UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1853 	UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1854 
1855 	/* Fill out the OS statistics structure */
1856 	adapter->netdev->stats.multicast = adapter->stats.mprc;
1857 }
1858 
1859 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1860 {
1861 	dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1862 		 adapter->link_speed,
1863 		 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1864 }
1865 
1866 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1867 {
1868 	struct e1000_hw *hw = &adapter->hw;
1869 	s32 ret_val = E1000_SUCCESS;
1870 	bool link_active;
1871 
1872 	/* If interface is down, stay link down */
1873 	if (test_bit(__IGBVF_DOWN, &adapter->state))
1874 		return false;
1875 
1876 	spin_lock_bh(&hw->mbx_lock);
1877 
1878 	ret_val = hw->mac.ops.check_for_link(hw);
1879 
1880 	spin_unlock_bh(&hw->mbx_lock);
1881 
1882 	link_active = !hw->mac.get_link_status;
1883 
1884 	/* if check for link returns error we will need to reset */
1885 	if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1886 		schedule_work(&adapter->reset_task);
1887 
1888 	return link_active;
1889 }
1890 
1891 /**
1892  * igbvf_watchdog - Timer Call-back
1893  * @data: pointer to adapter cast into an unsigned long
1894  **/
1895 static void igbvf_watchdog(struct timer_list *t)
1896 {
1897 	struct igbvf_adapter *adapter = from_timer(adapter, t, watchdog_timer);
1898 
1899 	/* Do the rest outside of interrupt context */
1900 	schedule_work(&adapter->watchdog_task);
1901 }
1902 
1903 static void igbvf_watchdog_task(struct work_struct *work)
1904 {
1905 	struct igbvf_adapter *adapter = container_of(work,
1906 						     struct igbvf_adapter,
1907 						     watchdog_task);
1908 	struct net_device *netdev = adapter->netdev;
1909 	struct e1000_mac_info *mac = &adapter->hw.mac;
1910 	struct igbvf_ring *tx_ring = adapter->tx_ring;
1911 	struct e1000_hw *hw = &adapter->hw;
1912 	u32 link;
1913 	int tx_pending = 0;
1914 
1915 	link = igbvf_has_link(adapter);
1916 
1917 	if (link) {
1918 		if (!netif_carrier_ok(netdev)) {
1919 			mac->ops.get_link_up_info(&adapter->hw,
1920 						  &adapter->link_speed,
1921 						  &adapter->link_duplex);
1922 			igbvf_print_link_info(adapter);
1923 
1924 			netif_carrier_on(netdev);
1925 			netif_wake_queue(netdev);
1926 		}
1927 	} else {
1928 		if (netif_carrier_ok(netdev)) {
1929 			adapter->link_speed = 0;
1930 			adapter->link_duplex = 0;
1931 			dev_info(&adapter->pdev->dev, "Link is Down\n");
1932 			netif_carrier_off(netdev);
1933 			netif_stop_queue(netdev);
1934 		}
1935 	}
1936 
1937 	if (netif_carrier_ok(netdev)) {
1938 		igbvf_update_stats(adapter);
1939 	} else {
1940 		tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1941 			      tx_ring->count);
1942 		if (tx_pending) {
1943 			/* We've lost link, so the controller stops DMA,
1944 			 * but we've got queued Tx work that's never going
1945 			 * to get done, so reset controller to flush Tx.
1946 			 * (Do the reset outside of interrupt context).
1947 			 */
1948 			adapter->tx_timeout_count++;
1949 			schedule_work(&adapter->reset_task);
1950 		}
1951 	}
1952 
1953 	/* Cause software interrupt to ensure Rx ring is cleaned */
1954 	ew32(EICS, adapter->rx_ring->eims_value);
1955 
1956 	/* Reset the timer */
1957 	if (!test_bit(__IGBVF_DOWN, &adapter->state))
1958 		mod_timer(&adapter->watchdog_timer,
1959 			  round_jiffies(jiffies + (2 * HZ)));
1960 }
1961 
1962 #define IGBVF_TX_FLAGS_CSUM		0x00000001
1963 #define IGBVF_TX_FLAGS_VLAN		0x00000002
1964 #define IGBVF_TX_FLAGS_TSO		0x00000004
1965 #define IGBVF_TX_FLAGS_IPV4		0x00000008
1966 #define IGBVF_TX_FLAGS_VLAN_MASK	0xffff0000
1967 #define IGBVF_TX_FLAGS_VLAN_SHIFT	16
1968 
1969 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1970 			      u32 type_tucmd, u32 mss_l4len_idx)
1971 {
1972 	struct e1000_adv_tx_context_desc *context_desc;
1973 	struct igbvf_buffer *buffer_info;
1974 	u16 i = tx_ring->next_to_use;
1975 
1976 	context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1977 	buffer_info = &tx_ring->buffer_info[i];
1978 
1979 	i++;
1980 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1981 
1982 	/* set bits to identify this as an advanced context descriptor */
1983 	type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1984 
1985 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
1986 	context_desc->seqnum_seed	= 0;
1987 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
1988 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
1989 
1990 	buffer_info->time_stamp = jiffies;
1991 	buffer_info->dma = 0;
1992 }
1993 
1994 static int igbvf_tso(struct igbvf_ring *tx_ring,
1995 		     struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1996 {
1997 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1998 	union {
1999 		struct iphdr *v4;
2000 		struct ipv6hdr *v6;
2001 		unsigned char *hdr;
2002 	} ip;
2003 	union {
2004 		struct tcphdr *tcp;
2005 		unsigned char *hdr;
2006 	} l4;
2007 	u32 paylen, l4_offset;
2008 	int err;
2009 
2010 	if (skb->ip_summed != CHECKSUM_PARTIAL)
2011 		return 0;
2012 
2013 	if (!skb_is_gso(skb))
2014 		return 0;
2015 
2016 	err = skb_cow_head(skb, 0);
2017 	if (err < 0)
2018 		return err;
2019 
2020 	ip.hdr = skb_network_header(skb);
2021 	l4.hdr = skb_checksum_start(skb);
2022 
2023 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2024 	type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2025 
2026 	/* initialize outer IP header fields */
2027 	if (ip.v4->version == 4) {
2028 		unsigned char *csum_start = skb_checksum_start(skb);
2029 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2030 
2031 		/* IP header will have to cancel out any data that
2032 		 * is not a part of the outer IP header
2033 		 */
2034 		ip.v4->check = csum_fold(csum_partial(trans_start,
2035 						      csum_start - trans_start,
2036 						      0));
2037 		type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2038 
2039 		ip.v4->tot_len = 0;
2040 	} else {
2041 		ip.v6->payload_len = 0;
2042 	}
2043 
2044 	/* determine offset of inner transport header */
2045 	l4_offset = l4.hdr - skb->data;
2046 
2047 	/* compute length of segmentation header */
2048 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2049 
2050 	/* remove payload length from inner checksum */
2051 	paylen = skb->len - l4_offset;
2052 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2053 
2054 	/* MSS L4LEN IDX */
2055 	mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2056 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2057 
2058 	/* VLAN MACLEN IPLEN */
2059 	vlan_macip_lens = l4.hdr - ip.hdr;
2060 	vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2061 	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2062 
2063 	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2064 
2065 	return 1;
2066 }
2067 
2068 static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb)
2069 {
2070 	unsigned int offset = 0;
2071 
2072 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
2073 
2074 	return offset == skb_checksum_start_offset(skb);
2075 }
2076 
2077 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2078 			  u32 tx_flags, __be16 protocol)
2079 {
2080 	u32 vlan_macip_lens = 0;
2081 	u32 type_tucmd = 0;
2082 
2083 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
2084 csum_failed:
2085 		if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2086 			return false;
2087 		goto no_csum;
2088 	}
2089 
2090 	switch (skb->csum_offset) {
2091 	case offsetof(struct tcphdr, check):
2092 		type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2093 		/* fall through */
2094 	case offsetof(struct udphdr, check):
2095 		break;
2096 	case offsetof(struct sctphdr, checksum):
2097 		/* validate that this is actually an SCTP request */
2098 		if (((protocol == htons(ETH_P_IP)) &&
2099 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
2100 		    ((protocol == htons(ETH_P_IPV6)) &&
2101 		     igbvf_ipv6_csum_is_sctp(skb))) {
2102 			type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2103 			break;
2104 		}
2105 		/* fall through */
2106 	default:
2107 		skb_checksum_help(skb);
2108 		goto csum_failed;
2109 	}
2110 
2111 	vlan_macip_lens = skb_checksum_start_offset(skb) -
2112 			  skb_network_offset(skb);
2113 no_csum:
2114 	vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2115 	vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2116 
2117 	igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2118 	return true;
2119 }
2120 
2121 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2122 {
2123 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2124 
2125 	/* there is enough descriptors then we don't need to worry  */
2126 	if (igbvf_desc_unused(adapter->tx_ring) >= size)
2127 		return 0;
2128 
2129 	netif_stop_queue(netdev);
2130 
2131 	/* Herbert's original patch had:
2132 	 *  smp_mb__after_netif_stop_queue();
2133 	 * but since that doesn't exist yet, just open code it.
2134 	 */
2135 	smp_mb();
2136 
2137 	/* We need to check again just in case room has been made available */
2138 	if (igbvf_desc_unused(adapter->tx_ring) < size)
2139 		return -EBUSY;
2140 
2141 	netif_wake_queue(netdev);
2142 
2143 	++adapter->restart_queue;
2144 	return 0;
2145 }
2146 
2147 #define IGBVF_MAX_TXD_PWR	16
2148 #define IGBVF_MAX_DATA_PER_TXD	(1u << IGBVF_MAX_TXD_PWR)
2149 
2150 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2151 				   struct igbvf_ring *tx_ring,
2152 				   struct sk_buff *skb)
2153 {
2154 	struct igbvf_buffer *buffer_info;
2155 	struct pci_dev *pdev = adapter->pdev;
2156 	unsigned int len = skb_headlen(skb);
2157 	unsigned int count = 0, i;
2158 	unsigned int f;
2159 
2160 	i = tx_ring->next_to_use;
2161 
2162 	buffer_info = &tx_ring->buffer_info[i];
2163 	BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2164 	buffer_info->length = len;
2165 	/* set time_stamp *before* dma to help avoid a possible race */
2166 	buffer_info->time_stamp = jiffies;
2167 	buffer_info->mapped_as_page = false;
2168 	buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2169 					  DMA_TO_DEVICE);
2170 	if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2171 		goto dma_error;
2172 
2173 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2174 		const struct skb_frag_struct *frag;
2175 
2176 		count++;
2177 		i++;
2178 		if (i == tx_ring->count)
2179 			i = 0;
2180 
2181 		frag = &skb_shinfo(skb)->frags[f];
2182 		len = skb_frag_size(frag);
2183 
2184 		buffer_info = &tx_ring->buffer_info[i];
2185 		BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2186 		buffer_info->length = len;
2187 		buffer_info->time_stamp = jiffies;
2188 		buffer_info->mapped_as_page = true;
2189 		buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2190 						    DMA_TO_DEVICE);
2191 		if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2192 			goto dma_error;
2193 	}
2194 
2195 	tx_ring->buffer_info[i].skb = skb;
2196 
2197 	return ++count;
2198 
2199 dma_error:
2200 	dev_err(&pdev->dev, "TX DMA map failed\n");
2201 
2202 	/* clear timestamp and dma mappings for failed buffer_info mapping */
2203 	buffer_info->dma = 0;
2204 	buffer_info->time_stamp = 0;
2205 	buffer_info->length = 0;
2206 	buffer_info->mapped_as_page = false;
2207 	if (count)
2208 		count--;
2209 
2210 	/* clear timestamp and dma mappings for remaining portion of packet */
2211 	while (count--) {
2212 		if (i == 0)
2213 			i += tx_ring->count;
2214 		i--;
2215 		buffer_info = &tx_ring->buffer_info[i];
2216 		igbvf_put_txbuf(adapter, buffer_info);
2217 	}
2218 
2219 	return 0;
2220 }
2221 
2222 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2223 				      struct igbvf_ring *tx_ring,
2224 				      int tx_flags, int count,
2225 				      unsigned int first, u32 paylen,
2226 				      u8 hdr_len)
2227 {
2228 	union e1000_adv_tx_desc *tx_desc = NULL;
2229 	struct igbvf_buffer *buffer_info;
2230 	u32 olinfo_status = 0, cmd_type_len;
2231 	unsigned int i;
2232 
2233 	cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2234 			E1000_ADVTXD_DCMD_DEXT);
2235 
2236 	if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2237 		cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2238 
2239 	if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2240 		cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2241 
2242 		/* insert tcp checksum */
2243 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2244 
2245 		/* insert ip checksum */
2246 		if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2247 			olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2248 
2249 	} else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2250 		olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2251 	}
2252 
2253 	olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2254 
2255 	i = tx_ring->next_to_use;
2256 	while (count--) {
2257 		buffer_info = &tx_ring->buffer_info[i];
2258 		tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2259 		tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2260 		tx_desc->read.cmd_type_len =
2261 			 cpu_to_le32(cmd_type_len | buffer_info->length);
2262 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2263 		i++;
2264 		if (i == tx_ring->count)
2265 			i = 0;
2266 	}
2267 
2268 	tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2269 	/* Force memory writes to complete before letting h/w
2270 	 * know there are new descriptors to fetch.  (Only
2271 	 * applicable for weak-ordered memory model archs,
2272 	 * such as IA-64).
2273 	 */
2274 	wmb();
2275 
2276 	tx_ring->buffer_info[first].next_to_watch = tx_desc;
2277 	tx_ring->next_to_use = i;
2278 	writel(i, adapter->hw.hw_addr + tx_ring->tail);
2279 	/* we need this if more than one processor can write to our tail
2280 	 * at a time, it synchronizes IO on IA64/Altix systems
2281 	 */
2282 	mmiowb();
2283 }
2284 
2285 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2286 					     struct net_device *netdev,
2287 					     struct igbvf_ring *tx_ring)
2288 {
2289 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2290 	unsigned int first, tx_flags = 0;
2291 	u8 hdr_len = 0;
2292 	int count = 0;
2293 	int tso = 0;
2294 	__be16 protocol = vlan_get_protocol(skb);
2295 
2296 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2297 		dev_kfree_skb_any(skb);
2298 		return NETDEV_TX_OK;
2299 	}
2300 
2301 	if (skb->len <= 0) {
2302 		dev_kfree_skb_any(skb);
2303 		return NETDEV_TX_OK;
2304 	}
2305 
2306 	/* need: count + 4 desc gap to keep tail from touching
2307 	 *       + 2 desc gap to keep tail from touching head,
2308 	 *       + 1 desc for skb->data,
2309 	 *       + 1 desc for context descriptor,
2310 	 * head, otherwise try next time
2311 	 */
2312 	if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2313 		/* this is a hard error */
2314 		return NETDEV_TX_BUSY;
2315 	}
2316 
2317 	if (skb_vlan_tag_present(skb)) {
2318 		tx_flags |= IGBVF_TX_FLAGS_VLAN;
2319 		tx_flags |= (skb_vlan_tag_get(skb) <<
2320 			     IGBVF_TX_FLAGS_VLAN_SHIFT);
2321 	}
2322 
2323 	if (protocol == htons(ETH_P_IP))
2324 		tx_flags |= IGBVF_TX_FLAGS_IPV4;
2325 
2326 	first = tx_ring->next_to_use;
2327 
2328 	tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2329 	if (unlikely(tso < 0)) {
2330 		dev_kfree_skb_any(skb);
2331 		return NETDEV_TX_OK;
2332 	}
2333 
2334 	if (tso)
2335 		tx_flags |= IGBVF_TX_FLAGS_TSO;
2336 	else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2337 		 (skb->ip_summed == CHECKSUM_PARTIAL))
2338 		tx_flags |= IGBVF_TX_FLAGS_CSUM;
2339 
2340 	/* count reflects descriptors mapped, if 0 then mapping error
2341 	 * has occurred and we need to rewind the descriptor queue
2342 	 */
2343 	count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2344 
2345 	if (count) {
2346 		igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2347 				   first, skb->len, hdr_len);
2348 		/* Make sure there is space in the ring for the next send. */
2349 		igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2350 	} else {
2351 		dev_kfree_skb_any(skb);
2352 		tx_ring->buffer_info[first].time_stamp = 0;
2353 		tx_ring->next_to_use = first;
2354 	}
2355 
2356 	return NETDEV_TX_OK;
2357 }
2358 
2359 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2360 				    struct net_device *netdev)
2361 {
2362 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2363 	struct igbvf_ring *tx_ring;
2364 
2365 	if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2366 		dev_kfree_skb_any(skb);
2367 		return NETDEV_TX_OK;
2368 	}
2369 
2370 	tx_ring = &adapter->tx_ring[0];
2371 
2372 	return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2373 }
2374 
2375 /**
2376  * igbvf_tx_timeout - Respond to a Tx Hang
2377  * @netdev: network interface device structure
2378  **/
2379 static void igbvf_tx_timeout(struct net_device *netdev)
2380 {
2381 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2382 
2383 	/* Do the reset outside of interrupt context */
2384 	adapter->tx_timeout_count++;
2385 	schedule_work(&adapter->reset_task);
2386 }
2387 
2388 static void igbvf_reset_task(struct work_struct *work)
2389 {
2390 	struct igbvf_adapter *adapter;
2391 
2392 	adapter = container_of(work, struct igbvf_adapter, reset_task);
2393 
2394 	igbvf_reinit_locked(adapter);
2395 }
2396 
2397 /**
2398  * igbvf_change_mtu - Change the Maximum Transfer Unit
2399  * @netdev: network interface device structure
2400  * @new_mtu: new value for maximum frame size
2401  *
2402  * Returns 0 on success, negative on failure
2403  **/
2404 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2405 {
2406 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2407 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2408 
2409 	while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2410 		usleep_range(1000, 2000);
2411 	/* igbvf_down has a dependency on max_frame_size */
2412 	adapter->max_frame_size = max_frame;
2413 	if (netif_running(netdev))
2414 		igbvf_down(adapter);
2415 
2416 	/* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2417 	 * means we reserve 2 more, this pushes us to allocate from the next
2418 	 * larger slab size.
2419 	 * i.e. RXBUFFER_2048 --> size-4096 slab
2420 	 * However with the new *_jumbo_rx* routines, jumbo receives will use
2421 	 * fragmented skbs
2422 	 */
2423 
2424 	if (max_frame <= 1024)
2425 		adapter->rx_buffer_len = 1024;
2426 	else if (max_frame <= 2048)
2427 		adapter->rx_buffer_len = 2048;
2428 	else
2429 #if (PAGE_SIZE / 2) > 16384
2430 		adapter->rx_buffer_len = 16384;
2431 #else
2432 		adapter->rx_buffer_len = PAGE_SIZE / 2;
2433 #endif
2434 
2435 	/* adjust allocation if LPE protects us, and we aren't using SBP */
2436 	if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2437 	    (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2438 		adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2439 					 ETH_FCS_LEN;
2440 
2441 	dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2442 		 netdev->mtu, new_mtu);
2443 	netdev->mtu = new_mtu;
2444 
2445 	if (netif_running(netdev))
2446 		igbvf_up(adapter);
2447 	else
2448 		igbvf_reset(adapter);
2449 
2450 	clear_bit(__IGBVF_RESETTING, &adapter->state);
2451 
2452 	return 0;
2453 }
2454 
2455 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2456 {
2457 	switch (cmd) {
2458 	default:
2459 		return -EOPNOTSUPP;
2460 	}
2461 }
2462 
2463 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2464 {
2465 	struct net_device *netdev = pci_get_drvdata(pdev);
2466 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2467 #ifdef CONFIG_PM
2468 	int retval = 0;
2469 #endif
2470 
2471 	netif_device_detach(netdev);
2472 
2473 	if (netif_running(netdev)) {
2474 		WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2475 		igbvf_down(adapter);
2476 		igbvf_free_irq(adapter);
2477 	}
2478 
2479 #ifdef CONFIG_PM
2480 	retval = pci_save_state(pdev);
2481 	if (retval)
2482 		return retval;
2483 #endif
2484 
2485 	pci_disable_device(pdev);
2486 
2487 	return 0;
2488 }
2489 
2490 #ifdef CONFIG_PM
2491 static int igbvf_resume(struct pci_dev *pdev)
2492 {
2493 	struct net_device *netdev = pci_get_drvdata(pdev);
2494 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2495 	u32 err;
2496 
2497 	pci_restore_state(pdev);
2498 	err = pci_enable_device_mem(pdev);
2499 	if (err) {
2500 		dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2501 		return err;
2502 	}
2503 
2504 	pci_set_master(pdev);
2505 
2506 	if (netif_running(netdev)) {
2507 		err = igbvf_request_irq(adapter);
2508 		if (err)
2509 			return err;
2510 	}
2511 
2512 	igbvf_reset(adapter);
2513 
2514 	if (netif_running(netdev))
2515 		igbvf_up(adapter);
2516 
2517 	netif_device_attach(netdev);
2518 
2519 	return 0;
2520 }
2521 #endif
2522 
2523 static void igbvf_shutdown(struct pci_dev *pdev)
2524 {
2525 	igbvf_suspend(pdev, PMSG_SUSPEND);
2526 }
2527 
2528 #ifdef CONFIG_NET_POLL_CONTROLLER
2529 /* Polling 'interrupt' - used by things like netconsole to send skbs
2530  * without having to re-enable interrupts. It's not called while
2531  * the interrupt routine is executing.
2532  */
2533 static void igbvf_netpoll(struct net_device *netdev)
2534 {
2535 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2536 
2537 	disable_irq(adapter->pdev->irq);
2538 
2539 	igbvf_clean_tx_irq(adapter->tx_ring);
2540 
2541 	enable_irq(adapter->pdev->irq);
2542 }
2543 #endif
2544 
2545 /**
2546  * igbvf_io_error_detected - called when PCI error is detected
2547  * @pdev: Pointer to PCI device
2548  * @state: The current pci connection state
2549  *
2550  * This function is called after a PCI bus error affecting
2551  * this device has been detected.
2552  */
2553 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2554 						pci_channel_state_t state)
2555 {
2556 	struct net_device *netdev = pci_get_drvdata(pdev);
2557 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2558 
2559 	netif_device_detach(netdev);
2560 
2561 	if (state == pci_channel_io_perm_failure)
2562 		return PCI_ERS_RESULT_DISCONNECT;
2563 
2564 	if (netif_running(netdev))
2565 		igbvf_down(adapter);
2566 	pci_disable_device(pdev);
2567 
2568 	/* Request a slot slot reset. */
2569 	return PCI_ERS_RESULT_NEED_RESET;
2570 }
2571 
2572 /**
2573  * igbvf_io_slot_reset - called after the pci bus has been reset.
2574  * @pdev: Pointer to PCI device
2575  *
2576  * Restart the card from scratch, as if from a cold-boot. Implementation
2577  * resembles the first-half of the igbvf_resume routine.
2578  */
2579 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2580 {
2581 	struct net_device *netdev = pci_get_drvdata(pdev);
2582 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2583 
2584 	if (pci_enable_device_mem(pdev)) {
2585 		dev_err(&pdev->dev,
2586 			"Cannot re-enable PCI device after reset.\n");
2587 		return PCI_ERS_RESULT_DISCONNECT;
2588 	}
2589 	pci_set_master(pdev);
2590 
2591 	igbvf_reset(adapter);
2592 
2593 	return PCI_ERS_RESULT_RECOVERED;
2594 }
2595 
2596 /**
2597  * igbvf_io_resume - called when traffic can start flowing again.
2598  * @pdev: Pointer to PCI device
2599  *
2600  * This callback is called when the error recovery driver tells us that
2601  * its OK to resume normal operation. Implementation resembles the
2602  * second-half of the igbvf_resume routine.
2603  */
2604 static void igbvf_io_resume(struct pci_dev *pdev)
2605 {
2606 	struct net_device *netdev = pci_get_drvdata(pdev);
2607 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2608 
2609 	if (netif_running(netdev)) {
2610 		if (igbvf_up(adapter)) {
2611 			dev_err(&pdev->dev,
2612 				"can't bring device back up after reset\n");
2613 			return;
2614 		}
2615 	}
2616 
2617 	netif_device_attach(netdev);
2618 }
2619 
2620 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2621 {
2622 	struct e1000_hw *hw = &adapter->hw;
2623 	struct net_device *netdev = adapter->netdev;
2624 	struct pci_dev *pdev = adapter->pdev;
2625 
2626 	if (hw->mac.type == e1000_vfadapt_i350)
2627 		dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2628 	else
2629 		dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2630 	dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2631 }
2632 
2633 static int igbvf_set_features(struct net_device *netdev,
2634 			      netdev_features_t features)
2635 {
2636 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2637 
2638 	if (features & NETIF_F_RXCSUM)
2639 		adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2640 	else
2641 		adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2642 
2643 	return 0;
2644 }
2645 
2646 #define IGBVF_MAX_MAC_HDR_LEN		127
2647 #define IGBVF_MAX_NETWORK_HDR_LEN	511
2648 
2649 static netdev_features_t
2650 igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2651 		     netdev_features_t features)
2652 {
2653 	unsigned int network_hdr_len, mac_hdr_len;
2654 
2655 	/* Make certain the headers can be described by a context descriptor */
2656 	mac_hdr_len = skb_network_header(skb) - skb->data;
2657 	if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2658 		return features & ~(NETIF_F_HW_CSUM |
2659 				    NETIF_F_SCTP_CRC |
2660 				    NETIF_F_HW_VLAN_CTAG_TX |
2661 				    NETIF_F_TSO |
2662 				    NETIF_F_TSO6);
2663 
2664 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2665 	if (unlikely(network_hdr_len >  IGBVF_MAX_NETWORK_HDR_LEN))
2666 		return features & ~(NETIF_F_HW_CSUM |
2667 				    NETIF_F_SCTP_CRC |
2668 				    NETIF_F_TSO |
2669 				    NETIF_F_TSO6);
2670 
2671 	/* We can only support IPV4 TSO in tunnels if we can mangle the
2672 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2673 	 */
2674 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2675 		features &= ~NETIF_F_TSO;
2676 
2677 	return features;
2678 }
2679 
2680 static const struct net_device_ops igbvf_netdev_ops = {
2681 	.ndo_open		= igbvf_open,
2682 	.ndo_stop		= igbvf_close,
2683 	.ndo_start_xmit		= igbvf_xmit_frame,
2684 	.ndo_set_rx_mode	= igbvf_set_rx_mode,
2685 	.ndo_set_mac_address	= igbvf_set_mac,
2686 	.ndo_change_mtu		= igbvf_change_mtu,
2687 	.ndo_do_ioctl		= igbvf_ioctl,
2688 	.ndo_tx_timeout		= igbvf_tx_timeout,
2689 	.ndo_vlan_rx_add_vid	= igbvf_vlan_rx_add_vid,
2690 	.ndo_vlan_rx_kill_vid	= igbvf_vlan_rx_kill_vid,
2691 #ifdef CONFIG_NET_POLL_CONTROLLER
2692 	.ndo_poll_controller	= igbvf_netpoll,
2693 #endif
2694 	.ndo_set_features	= igbvf_set_features,
2695 	.ndo_features_check	= igbvf_features_check,
2696 };
2697 
2698 /**
2699  * igbvf_probe - Device Initialization Routine
2700  * @pdev: PCI device information struct
2701  * @ent: entry in igbvf_pci_tbl
2702  *
2703  * Returns 0 on success, negative on failure
2704  *
2705  * igbvf_probe initializes an adapter identified by a pci_dev structure.
2706  * The OS initialization, configuring of the adapter private structure,
2707  * and a hardware reset occur.
2708  **/
2709 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2710 {
2711 	struct net_device *netdev;
2712 	struct igbvf_adapter *adapter;
2713 	struct e1000_hw *hw;
2714 	const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2715 
2716 	static int cards_found;
2717 	int err, pci_using_dac;
2718 
2719 	err = pci_enable_device_mem(pdev);
2720 	if (err)
2721 		return err;
2722 
2723 	pci_using_dac = 0;
2724 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2725 	if (!err) {
2726 		pci_using_dac = 1;
2727 	} else {
2728 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2729 		if (err) {
2730 			dev_err(&pdev->dev,
2731 				"No usable DMA configuration, aborting\n");
2732 			goto err_dma;
2733 		}
2734 	}
2735 
2736 	err = pci_request_regions(pdev, igbvf_driver_name);
2737 	if (err)
2738 		goto err_pci_reg;
2739 
2740 	pci_set_master(pdev);
2741 
2742 	err = -ENOMEM;
2743 	netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2744 	if (!netdev)
2745 		goto err_alloc_etherdev;
2746 
2747 	SET_NETDEV_DEV(netdev, &pdev->dev);
2748 
2749 	pci_set_drvdata(pdev, netdev);
2750 	adapter = netdev_priv(netdev);
2751 	hw = &adapter->hw;
2752 	adapter->netdev = netdev;
2753 	adapter->pdev = pdev;
2754 	adapter->ei = ei;
2755 	adapter->pba = ei->pba;
2756 	adapter->flags = ei->flags;
2757 	adapter->hw.back = adapter;
2758 	adapter->hw.mac.type = ei->mac;
2759 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2760 
2761 	/* PCI config space info */
2762 
2763 	hw->vendor_id = pdev->vendor;
2764 	hw->device_id = pdev->device;
2765 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
2766 	hw->subsystem_device_id = pdev->subsystem_device;
2767 	hw->revision_id = pdev->revision;
2768 
2769 	err = -EIO;
2770 	adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2771 				      pci_resource_len(pdev, 0));
2772 
2773 	if (!adapter->hw.hw_addr)
2774 		goto err_ioremap;
2775 
2776 	if (ei->get_variants) {
2777 		err = ei->get_variants(adapter);
2778 		if (err)
2779 			goto err_get_variants;
2780 	}
2781 
2782 	/* setup adapter struct */
2783 	err = igbvf_sw_init(adapter);
2784 	if (err)
2785 		goto err_sw_init;
2786 
2787 	/* construct the net_device struct */
2788 	netdev->netdev_ops = &igbvf_netdev_ops;
2789 
2790 	igbvf_set_ethtool_ops(netdev);
2791 	netdev->watchdog_timeo = 5 * HZ;
2792 	strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2793 
2794 	adapter->bd_number = cards_found++;
2795 
2796 	netdev->hw_features = NETIF_F_SG |
2797 			      NETIF_F_TSO |
2798 			      NETIF_F_TSO6 |
2799 			      NETIF_F_RXCSUM |
2800 			      NETIF_F_HW_CSUM |
2801 			      NETIF_F_SCTP_CRC;
2802 
2803 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2804 				    NETIF_F_GSO_GRE_CSUM | \
2805 				    NETIF_F_GSO_IPXIP4 | \
2806 				    NETIF_F_GSO_IPXIP6 | \
2807 				    NETIF_F_GSO_UDP_TUNNEL | \
2808 				    NETIF_F_GSO_UDP_TUNNEL_CSUM)
2809 
2810 	netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2811 	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2812 			       IGBVF_GSO_PARTIAL_FEATURES;
2813 
2814 	netdev->features = netdev->hw_features;
2815 
2816 	if (pci_using_dac)
2817 		netdev->features |= NETIF_F_HIGHDMA;
2818 
2819 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2820 	netdev->mpls_features |= NETIF_F_HW_CSUM;
2821 	netdev->hw_enc_features |= netdev->vlan_features;
2822 
2823 	/* set this bit last since it cannot be part of vlan_features */
2824 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2825 			    NETIF_F_HW_VLAN_CTAG_RX |
2826 			    NETIF_F_HW_VLAN_CTAG_TX;
2827 
2828 	/* MTU range: 68 - 9216 */
2829 	netdev->min_mtu = ETH_MIN_MTU;
2830 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2831 
2832 	spin_lock_bh(&hw->mbx_lock);
2833 
2834 	/*reset the controller to put the device in a known good state */
2835 	err = hw->mac.ops.reset_hw(hw);
2836 	if (err) {
2837 		dev_info(&pdev->dev,
2838 			 "PF still in reset state. Is the PF interface up?\n");
2839 	} else {
2840 		err = hw->mac.ops.read_mac_addr(hw);
2841 		if (err)
2842 			dev_info(&pdev->dev, "Error reading MAC address.\n");
2843 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
2844 			dev_info(&pdev->dev,
2845 				 "MAC address not assigned by administrator.\n");
2846 		memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2847 		       netdev->addr_len);
2848 	}
2849 
2850 	spin_unlock_bh(&hw->mbx_lock);
2851 
2852 	if (!is_valid_ether_addr(netdev->dev_addr)) {
2853 		dev_info(&pdev->dev, "Assigning random MAC address.\n");
2854 		eth_hw_addr_random(netdev);
2855 		memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2856 		       netdev->addr_len);
2857 	}
2858 
2859 	timer_setup(&adapter->watchdog_timer, igbvf_watchdog, 0);
2860 
2861 	INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2862 	INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2863 
2864 	/* ring size defaults */
2865 	adapter->rx_ring->count = 1024;
2866 	adapter->tx_ring->count = 1024;
2867 
2868 	/* reset the hardware with the new settings */
2869 	igbvf_reset(adapter);
2870 
2871 	/* set hardware-specific flags */
2872 	if (adapter->hw.mac.type == e1000_vfadapt_i350)
2873 		adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2874 
2875 	strcpy(netdev->name, "eth%d");
2876 	err = register_netdev(netdev);
2877 	if (err)
2878 		goto err_hw_init;
2879 
2880 	/* tell the stack to leave us alone until igbvf_open() is called */
2881 	netif_carrier_off(netdev);
2882 	netif_stop_queue(netdev);
2883 
2884 	igbvf_print_device_info(adapter);
2885 
2886 	igbvf_initialize_last_counter_stats(adapter);
2887 
2888 	return 0;
2889 
2890 err_hw_init:
2891 	kfree(adapter->tx_ring);
2892 	kfree(adapter->rx_ring);
2893 err_sw_init:
2894 	igbvf_reset_interrupt_capability(adapter);
2895 err_get_variants:
2896 	iounmap(adapter->hw.hw_addr);
2897 err_ioremap:
2898 	free_netdev(netdev);
2899 err_alloc_etherdev:
2900 	pci_release_regions(pdev);
2901 err_pci_reg:
2902 err_dma:
2903 	pci_disable_device(pdev);
2904 	return err;
2905 }
2906 
2907 /**
2908  * igbvf_remove - Device Removal Routine
2909  * @pdev: PCI device information struct
2910  *
2911  * igbvf_remove is called by the PCI subsystem to alert the driver
2912  * that it should release a PCI device.  The could be caused by a
2913  * Hot-Plug event, or because the driver is going to be removed from
2914  * memory.
2915  **/
2916 static void igbvf_remove(struct pci_dev *pdev)
2917 {
2918 	struct net_device *netdev = pci_get_drvdata(pdev);
2919 	struct igbvf_adapter *adapter = netdev_priv(netdev);
2920 	struct e1000_hw *hw = &adapter->hw;
2921 
2922 	/* The watchdog timer may be rescheduled, so explicitly
2923 	 * disable it from being rescheduled.
2924 	 */
2925 	set_bit(__IGBVF_DOWN, &adapter->state);
2926 	del_timer_sync(&adapter->watchdog_timer);
2927 
2928 	cancel_work_sync(&adapter->reset_task);
2929 	cancel_work_sync(&adapter->watchdog_task);
2930 
2931 	unregister_netdev(netdev);
2932 
2933 	igbvf_reset_interrupt_capability(adapter);
2934 
2935 	/* it is important to delete the NAPI struct prior to freeing the
2936 	 * Rx ring so that you do not end up with null pointer refs
2937 	 */
2938 	netif_napi_del(&adapter->rx_ring->napi);
2939 	kfree(adapter->tx_ring);
2940 	kfree(adapter->rx_ring);
2941 
2942 	iounmap(hw->hw_addr);
2943 	if (hw->flash_address)
2944 		iounmap(hw->flash_address);
2945 	pci_release_regions(pdev);
2946 
2947 	free_netdev(netdev);
2948 
2949 	pci_disable_device(pdev);
2950 }
2951 
2952 /* PCI Error Recovery (ERS) */
2953 static const struct pci_error_handlers igbvf_err_handler = {
2954 	.error_detected = igbvf_io_error_detected,
2955 	.slot_reset = igbvf_io_slot_reset,
2956 	.resume = igbvf_io_resume,
2957 };
2958 
2959 static const struct pci_device_id igbvf_pci_tbl[] = {
2960 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2961 	{ PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2962 	{ } /* terminate list */
2963 };
2964 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2965 
2966 /* PCI Device API Driver */
2967 static struct pci_driver igbvf_driver = {
2968 	.name		= igbvf_driver_name,
2969 	.id_table	= igbvf_pci_tbl,
2970 	.probe		= igbvf_probe,
2971 	.remove		= igbvf_remove,
2972 #ifdef CONFIG_PM
2973 	/* Power Management Hooks */
2974 	.suspend	= igbvf_suspend,
2975 	.resume		= igbvf_resume,
2976 #endif
2977 	.shutdown	= igbvf_shutdown,
2978 	.err_handler	= &igbvf_err_handler
2979 };
2980 
2981 /**
2982  * igbvf_init_module - Driver Registration Routine
2983  *
2984  * igbvf_init_module is the first routine called when the driver is
2985  * loaded. All it does is register with the PCI subsystem.
2986  **/
2987 static int __init igbvf_init_module(void)
2988 {
2989 	int ret;
2990 
2991 	pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2992 	pr_info("%s\n", igbvf_copyright);
2993 
2994 	ret = pci_register_driver(&igbvf_driver);
2995 
2996 	return ret;
2997 }
2998 module_init(igbvf_init_module);
2999 
3000 /**
3001  * igbvf_exit_module - Driver Exit Cleanup Routine
3002  *
3003  * igbvf_exit_module is called just before the driver is removed
3004  * from memory.
3005  **/
3006 static void __exit igbvf_exit_module(void)
3007 {
3008 	pci_unregister_driver(&igbvf_driver);
3009 }
3010 module_exit(igbvf_exit_module);
3011 
3012 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
3013 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
3014 MODULE_LICENSE("GPL");
3015 MODULE_VERSION(DRV_VERSION);
3016 
3017 /* netdev.c */
3018