1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c)  2018 Intel Corporation */
3 
4 #include <linux/module.h>
5 #include <linux/types.h>
6 #include <linux/if_vlan.h>
7 #include <linux/aer.h>
8 #include <linux/tcp.h>
9 #include <linux/udp.h>
10 #include <linux/ip.h>
11 #include <linux/pm_runtime.h>
12 
13 #include <net/ipv6.h>
14 
15 #include "igc.h"
16 #include "igc_hw.h"
17 
18 #define DRV_VERSION	"0.0.1-k"
19 #define DRV_SUMMARY	"Intel(R) 2.5G Ethernet Linux Driver"
20 
21 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
22 
23 static int debug = -1;
24 
25 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
26 MODULE_DESCRIPTION(DRV_SUMMARY);
27 MODULE_LICENSE("GPL v2");
28 MODULE_VERSION(DRV_VERSION);
29 module_param(debug, int, 0);
30 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
31 
32 char igc_driver_name[] = "igc";
33 char igc_driver_version[] = DRV_VERSION;
34 static const char igc_driver_string[] = DRV_SUMMARY;
35 static const char igc_copyright[] =
36 	"Copyright(c) 2018 Intel Corporation.";
37 
38 static const struct igc_info *igc_info_tbl[] = {
39 	[board_base] = &igc_base_info,
40 };
41 
42 static const struct pci_device_id igc_pci_tbl[] = {
43 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base },
44 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base },
45 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base },
46 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base },
47 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base },
48 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_BLANK_NVM), board_base },
49 	/* required last entry */
50 	{0, }
51 };
52 
53 MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
54 
55 enum latency_range {
56 	lowest_latency = 0,
57 	low_latency = 1,
58 	bulk_latency = 2,
59 	latency_invalid = 255
60 };
61 
62 /**
63  * igc_power_down_link - Power down the phy/serdes link
64  * @adapter: address of board private structure
65  */
66 static void igc_power_down_link(struct igc_adapter *adapter)
67 {
68 	if (adapter->hw.phy.media_type == igc_media_type_copper)
69 		igc_power_down_phy_copper_base(&adapter->hw);
70 }
71 
72 void igc_reset(struct igc_adapter *adapter)
73 {
74 	struct pci_dev *pdev = adapter->pdev;
75 	struct igc_hw *hw = &adapter->hw;
76 	struct igc_fc_info *fc = &hw->fc;
77 	u32 pba, hwm;
78 
79 	/* Repartition PBA for greater than 9k MTU if required */
80 	pba = IGC_PBA_34K;
81 
82 	/* flow control settings
83 	 * The high water mark must be low enough to fit one full frame
84 	 * after transmitting the pause frame.  As such we must have enough
85 	 * space to allow for us to complete our current transmit and then
86 	 * receive the frame that is in progress from the link partner.
87 	 * Set it to:
88 	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
89 	 */
90 	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
91 
92 	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
93 	fc->low_water = fc->high_water - 16;
94 	fc->pause_time = 0xFFFF;
95 	fc->send_xon = 1;
96 	fc->current_mode = fc->requested_mode;
97 
98 	hw->mac.ops.reset_hw(hw);
99 
100 	if (hw->mac.ops.init_hw(hw))
101 		dev_err(&pdev->dev, "Hardware Error\n");
102 
103 	if (!netif_running(adapter->netdev))
104 		igc_power_down_link(adapter);
105 
106 	/* Re-enable PTP, where applicable. */
107 	igc_ptp_reset(adapter);
108 
109 	igc_get_phy_info(hw);
110 }
111 
112 /**
113  * igc_power_up_link - Power up the phy link
114  * @adapter: address of board private structure
115  */
116 static void igc_power_up_link(struct igc_adapter *adapter)
117 {
118 	igc_reset_phy(&adapter->hw);
119 
120 	if (adapter->hw.phy.media_type == igc_media_type_copper)
121 		igc_power_up_phy_copper(&adapter->hw);
122 
123 	igc_setup_link(&adapter->hw);
124 }
125 
126 /**
127  * igc_release_hw_control - release control of the h/w to f/w
128  * @adapter: address of board private structure
129  *
130  * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
131  * For ASF and Pass Through versions of f/w this means that the
132  * driver is no longer loaded.
133  */
134 static void igc_release_hw_control(struct igc_adapter *adapter)
135 {
136 	struct igc_hw *hw = &adapter->hw;
137 	u32 ctrl_ext;
138 
139 	/* Let firmware take over control of h/w */
140 	ctrl_ext = rd32(IGC_CTRL_EXT);
141 	wr32(IGC_CTRL_EXT,
142 	     ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
143 }
144 
145 /**
146  * igc_get_hw_control - get control of the h/w from f/w
147  * @adapter: address of board private structure
148  *
149  * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
150  * For ASF and Pass Through versions of f/w this means that
151  * the driver is loaded.
152  */
153 static void igc_get_hw_control(struct igc_adapter *adapter)
154 {
155 	struct igc_hw *hw = &adapter->hw;
156 	u32 ctrl_ext;
157 
158 	/* Let firmware know the driver has taken over */
159 	ctrl_ext = rd32(IGC_CTRL_EXT);
160 	wr32(IGC_CTRL_EXT,
161 	     ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
162 }
163 
164 /**
165  * igc_clean_tx_ring - Free Tx Buffers
166  * @tx_ring: ring to be cleaned
167  */
168 static void igc_clean_tx_ring(struct igc_ring *tx_ring)
169 {
170 	u16 i = tx_ring->next_to_clean;
171 	struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
172 
173 	while (i != tx_ring->next_to_use) {
174 		union igc_adv_tx_desc *eop_desc, *tx_desc;
175 
176 		/* Free all the Tx ring sk_buffs */
177 		dev_kfree_skb_any(tx_buffer->skb);
178 
179 		/* unmap skb header data */
180 		dma_unmap_single(tx_ring->dev,
181 				 dma_unmap_addr(tx_buffer, dma),
182 				 dma_unmap_len(tx_buffer, len),
183 				 DMA_TO_DEVICE);
184 
185 		/* check for eop_desc to determine the end of the packet */
186 		eop_desc = tx_buffer->next_to_watch;
187 		tx_desc = IGC_TX_DESC(tx_ring, i);
188 
189 		/* unmap remaining buffers */
190 		while (tx_desc != eop_desc) {
191 			tx_buffer++;
192 			tx_desc++;
193 			i++;
194 			if (unlikely(i == tx_ring->count)) {
195 				i = 0;
196 				tx_buffer = tx_ring->tx_buffer_info;
197 				tx_desc = IGC_TX_DESC(tx_ring, 0);
198 			}
199 
200 			/* unmap any remaining paged data */
201 			if (dma_unmap_len(tx_buffer, len))
202 				dma_unmap_page(tx_ring->dev,
203 					       dma_unmap_addr(tx_buffer, dma),
204 					       dma_unmap_len(tx_buffer, len),
205 					       DMA_TO_DEVICE);
206 		}
207 
208 		/* move us one more past the eop_desc for start of next pkt */
209 		tx_buffer++;
210 		i++;
211 		if (unlikely(i == tx_ring->count)) {
212 			i = 0;
213 			tx_buffer = tx_ring->tx_buffer_info;
214 		}
215 	}
216 
217 	/* reset BQL for queue */
218 	netdev_tx_reset_queue(txring_txq(tx_ring));
219 
220 	/* reset next_to_use and next_to_clean */
221 	tx_ring->next_to_use = 0;
222 	tx_ring->next_to_clean = 0;
223 }
224 
225 /**
226  * igc_free_tx_resources - Free Tx Resources per Queue
227  * @tx_ring: Tx descriptor ring for a specific queue
228  *
229  * Free all transmit software resources
230  */
231 void igc_free_tx_resources(struct igc_ring *tx_ring)
232 {
233 	igc_clean_tx_ring(tx_ring);
234 
235 	vfree(tx_ring->tx_buffer_info);
236 	tx_ring->tx_buffer_info = NULL;
237 
238 	/* if not set, then don't free */
239 	if (!tx_ring->desc)
240 		return;
241 
242 	dma_free_coherent(tx_ring->dev, tx_ring->size,
243 			  tx_ring->desc, tx_ring->dma);
244 
245 	tx_ring->desc = NULL;
246 }
247 
248 /**
249  * igc_free_all_tx_resources - Free Tx Resources for All Queues
250  * @adapter: board private structure
251  *
252  * Free all transmit software resources
253  */
254 static void igc_free_all_tx_resources(struct igc_adapter *adapter)
255 {
256 	int i;
257 
258 	for (i = 0; i < adapter->num_tx_queues; i++)
259 		igc_free_tx_resources(adapter->tx_ring[i]);
260 }
261 
262 /**
263  * igc_clean_all_tx_rings - Free Tx Buffers for all queues
264  * @adapter: board private structure
265  */
266 static void igc_clean_all_tx_rings(struct igc_adapter *adapter)
267 {
268 	int i;
269 
270 	for (i = 0; i < adapter->num_tx_queues; i++)
271 		if (adapter->tx_ring[i])
272 			igc_clean_tx_ring(adapter->tx_ring[i]);
273 }
274 
275 /**
276  * igc_setup_tx_resources - allocate Tx resources (Descriptors)
277  * @tx_ring: tx descriptor ring (for a specific queue) to setup
278  *
279  * Return 0 on success, negative on failure
280  */
281 int igc_setup_tx_resources(struct igc_ring *tx_ring)
282 {
283 	struct device *dev = tx_ring->dev;
284 	int size = 0;
285 
286 	size = sizeof(struct igc_tx_buffer) * tx_ring->count;
287 	tx_ring->tx_buffer_info = vzalloc(size);
288 	if (!tx_ring->tx_buffer_info)
289 		goto err;
290 
291 	/* round up to nearest 4K */
292 	tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc);
293 	tx_ring->size = ALIGN(tx_ring->size, 4096);
294 
295 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
296 					   &tx_ring->dma, GFP_KERNEL);
297 
298 	if (!tx_ring->desc)
299 		goto err;
300 
301 	tx_ring->next_to_use = 0;
302 	tx_ring->next_to_clean = 0;
303 
304 	return 0;
305 
306 err:
307 	vfree(tx_ring->tx_buffer_info);
308 	dev_err(dev,
309 		"Unable to allocate memory for the transmit descriptor ring\n");
310 	return -ENOMEM;
311 }
312 
313 /**
314  * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues
315  * @adapter: board private structure
316  *
317  * Return 0 on success, negative on failure
318  */
319 static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
320 {
321 	struct pci_dev *pdev = adapter->pdev;
322 	int i, err = 0;
323 
324 	for (i = 0; i < adapter->num_tx_queues; i++) {
325 		err = igc_setup_tx_resources(adapter->tx_ring[i]);
326 		if (err) {
327 			dev_err(&pdev->dev,
328 				"Allocation for Tx Queue %u failed\n", i);
329 			for (i--; i >= 0; i--)
330 				igc_free_tx_resources(adapter->tx_ring[i]);
331 			break;
332 		}
333 	}
334 
335 	return err;
336 }
337 
338 /**
339  * igc_clean_rx_ring - Free Rx Buffers per Queue
340  * @rx_ring: ring to free buffers from
341  */
342 static void igc_clean_rx_ring(struct igc_ring *rx_ring)
343 {
344 	u16 i = rx_ring->next_to_clean;
345 
346 	dev_kfree_skb(rx_ring->skb);
347 	rx_ring->skb = NULL;
348 
349 	/* Free all the Rx ring sk_buffs */
350 	while (i != rx_ring->next_to_alloc) {
351 		struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
352 
353 		/* Invalidate cache lines that may have been written to by
354 		 * device so that we avoid corrupting memory.
355 		 */
356 		dma_sync_single_range_for_cpu(rx_ring->dev,
357 					      buffer_info->dma,
358 					      buffer_info->page_offset,
359 					      igc_rx_bufsz(rx_ring),
360 					      DMA_FROM_DEVICE);
361 
362 		/* free resources associated with mapping */
363 		dma_unmap_page_attrs(rx_ring->dev,
364 				     buffer_info->dma,
365 				     igc_rx_pg_size(rx_ring),
366 				     DMA_FROM_DEVICE,
367 				     IGC_RX_DMA_ATTR);
368 		__page_frag_cache_drain(buffer_info->page,
369 					buffer_info->pagecnt_bias);
370 
371 		i++;
372 		if (i == rx_ring->count)
373 			i = 0;
374 	}
375 
376 	rx_ring->next_to_alloc = 0;
377 	rx_ring->next_to_clean = 0;
378 	rx_ring->next_to_use = 0;
379 }
380 
381 /**
382  * igc_clean_all_rx_rings - Free Rx Buffers for all queues
383  * @adapter: board private structure
384  */
385 static void igc_clean_all_rx_rings(struct igc_adapter *adapter)
386 {
387 	int i;
388 
389 	for (i = 0; i < adapter->num_rx_queues; i++)
390 		if (adapter->rx_ring[i])
391 			igc_clean_rx_ring(adapter->rx_ring[i]);
392 }
393 
394 /**
395  * igc_free_rx_resources - Free Rx Resources
396  * @rx_ring: ring to clean the resources from
397  *
398  * Free all receive software resources
399  */
400 void igc_free_rx_resources(struct igc_ring *rx_ring)
401 {
402 	igc_clean_rx_ring(rx_ring);
403 
404 	vfree(rx_ring->rx_buffer_info);
405 	rx_ring->rx_buffer_info = NULL;
406 
407 	/* if not set, then don't free */
408 	if (!rx_ring->desc)
409 		return;
410 
411 	dma_free_coherent(rx_ring->dev, rx_ring->size,
412 			  rx_ring->desc, rx_ring->dma);
413 
414 	rx_ring->desc = NULL;
415 }
416 
417 /**
418  * igc_free_all_rx_resources - Free Rx Resources for All Queues
419  * @adapter: board private structure
420  *
421  * Free all receive software resources
422  */
423 static void igc_free_all_rx_resources(struct igc_adapter *adapter)
424 {
425 	int i;
426 
427 	for (i = 0; i < adapter->num_rx_queues; i++)
428 		igc_free_rx_resources(adapter->rx_ring[i]);
429 }
430 
431 /**
432  * igc_setup_rx_resources - allocate Rx resources (Descriptors)
433  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
434  *
435  * Returns 0 on success, negative on failure
436  */
437 int igc_setup_rx_resources(struct igc_ring *rx_ring)
438 {
439 	struct device *dev = rx_ring->dev;
440 	int size, desc_len;
441 
442 	size = sizeof(struct igc_rx_buffer) * rx_ring->count;
443 	rx_ring->rx_buffer_info = vzalloc(size);
444 	if (!rx_ring->rx_buffer_info)
445 		goto err;
446 
447 	desc_len = sizeof(union igc_adv_rx_desc);
448 
449 	/* Round up to nearest 4K */
450 	rx_ring->size = rx_ring->count * desc_len;
451 	rx_ring->size = ALIGN(rx_ring->size, 4096);
452 
453 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
454 					   &rx_ring->dma, GFP_KERNEL);
455 
456 	if (!rx_ring->desc)
457 		goto err;
458 
459 	rx_ring->next_to_alloc = 0;
460 	rx_ring->next_to_clean = 0;
461 	rx_ring->next_to_use = 0;
462 
463 	return 0;
464 
465 err:
466 	vfree(rx_ring->rx_buffer_info);
467 	rx_ring->rx_buffer_info = NULL;
468 	dev_err(dev,
469 		"Unable to allocate memory for the receive descriptor ring\n");
470 	return -ENOMEM;
471 }
472 
473 /**
474  * igc_setup_all_rx_resources - wrapper to allocate Rx resources
475  *                                (Descriptors) for all queues
476  * @adapter: board private structure
477  *
478  * Return 0 on success, negative on failure
479  */
480 static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
481 {
482 	struct pci_dev *pdev = adapter->pdev;
483 	int i, err = 0;
484 
485 	for (i = 0; i < adapter->num_rx_queues; i++) {
486 		err = igc_setup_rx_resources(adapter->rx_ring[i]);
487 		if (err) {
488 			dev_err(&pdev->dev,
489 				"Allocation for Rx Queue %u failed\n", i);
490 			for (i--; i >= 0; i--)
491 				igc_free_rx_resources(adapter->rx_ring[i]);
492 			break;
493 		}
494 	}
495 
496 	return err;
497 }
498 
499 /**
500  * igc_configure_rx_ring - Configure a receive ring after Reset
501  * @adapter: board private structure
502  * @ring: receive ring to be configured
503  *
504  * Configure the Rx unit of the MAC after a reset.
505  */
506 static void igc_configure_rx_ring(struct igc_adapter *adapter,
507 				  struct igc_ring *ring)
508 {
509 	struct igc_hw *hw = &adapter->hw;
510 	union igc_adv_rx_desc *rx_desc;
511 	int reg_idx = ring->reg_idx;
512 	u32 srrctl = 0, rxdctl = 0;
513 	u64 rdba = ring->dma;
514 
515 	/* disable the queue */
516 	wr32(IGC_RXDCTL(reg_idx), 0);
517 
518 	/* Set DMA base address registers */
519 	wr32(IGC_RDBAL(reg_idx),
520 	     rdba & 0x00000000ffffffffULL);
521 	wr32(IGC_RDBAH(reg_idx), rdba >> 32);
522 	wr32(IGC_RDLEN(reg_idx),
523 	     ring->count * sizeof(union igc_adv_rx_desc));
524 
525 	/* initialize head and tail */
526 	ring->tail = adapter->io_addr + IGC_RDT(reg_idx);
527 	wr32(IGC_RDH(reg_idx), 0);
528 	writel(0, ring->tail);
529 
530 	/* reset next-to- use/clean to place SW in sync with hardware */
531 	ring->next_to_clean = 0;
532 	ring->next_to_use = 0;
533 
534 	/* set descriptor configuration */
535 	srrctl = IGC_RX_HDR_LEN << IGC_SRRCTL_BSIZEHDRSIZE_SHIFT;
536 	if (ring_uses_large_buffer(ring))
537 		srrctl |= IGC_RXBUFFER_3072 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
538 	else
539 		srrctl |= IGC_RXBUFFER_2048 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
540 	srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
541 
542 	wr32(IGC_SRRCTL(reg_idx), srrctl);
543 
544 	rxdctl |= IGC_RX_PTHRESH;
545 	rxdctl |= IGC_RX_HTHRESH << 8;
546 	rxdctl |= IGC_RX_WTHRESH << 16;
547 
548 	/* initialize rx_buffer_info */
549 	memset(ring->rx_buffer_info, 0,
550 	       sizeof(struct igc_rx_buffer) * ring->count);
551 
552 	/* initialize Rx descriptor 0 */
553 	rx_desc = IGC_RX_DESC(ring, 0);
554 	rx_desc->wb.upper.length = 0;
555 
556 	/* enable receive descriptor fetching */
557 	rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
558 
559 	wr32(IGC_RXDCTL(reg_idx), rxdctl);
560 }
561 
562 /**
563  * igc_configure_rx - Configure receive Unit after Reset
564  * @adapter: board private structure
565  *
566  * Configure the Rx unit of the MAC after a reset.
567  */
568 static void igc_configure_rx(struct igc_adapter *adapter)
569 {
570 	int i;
571 
572 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
573 	 * the Base and Length of the Rx Descriptor Ring
574 	 */
575 	for (i = 0; i < adapter->num_rx_queues; i++)
576 		igc_configure_rx_ring(adapter, adapter->rx_ring[i]);
577 }
578 
579 /**
580  * igc_configure_tx_ring - Configure transmit ring after Reset
581  * @adapter: board private structure
582  * @ring: tx ring to configure
583  *
584  * Configure a transmit ring after a reset.
585  */
586 static void igc_configure_tx_ring(struct igc_adapter *adapter,
587 				  struct igc_ring *ring)
588 {
589 	struct igc_hw *hw = &adapter->hw;
590 	int reg_idx = ring->reg_idx;
591 	u64 tdba = ring->dma;
592 	u32 txdctl = 0;
593 
594 	/* disable the queue */
595 	wr32(IGC_TXDCTL(reg_idx), 0);
596 	wrfl();
597 	mdelay(10);
598 
599 	wr32(IGC_TDLEN(reg_idx),
600 	     ring->count * sizeof(union igc_adv_tx_desc));
601 	wr32(IGC_TDBAL(reg_idx),
602 	     tdba & 0x00000000ffffffffULL);
603 	wr32(IGC_TDBAH(reg_idx), tdba >> 32);
604 
605 	ring->tail = adapter->io_addr + IGC_TDT(reg_idx);
606 	wr32(IGC_TDH(reg_idx), 0);
607 	writel(0, ring->tail);
608 
609 	txdctl |= IGC_TX_PTHRESH;
610 	txdctl |= IGC_TX_HTHRESH << 8;
611 	txdctl |= IGC_TX_WTHRESH << 16;
612 
613 	txdctl |= IGC_TXDCTL_QUEUE_ENABLE;
614 	wr32(IGC_TXDCTL(reg_idx), txdctl);
615 }
616 
617 /**
618  * igc_configure_tx - Configure transmit Unit after Reset
619  * @adapter: board private structure
620  *
621  * Configure the Tx unit of the MAC after a reset.
622  */
623 static void igc_configure_tx(struct igc_adapter *adapter)
624 {
625 	int i;
626 
627 	for (i = 0; i < adapter->num_tx_queues; i++)
628 		igc_configure_tx_ring(adapter, adapter->tx_ring[i]);
629 }
630 
631 /**
632  * igc_setup_mrqc - configure the multiple receive queue control registers
633  * @adapter: Board private structure
634  */
635 static void igc_setup_mrqc(struct igc_adapter *adapter)
636 {
637 	struct igc_hw *hw = &adapter->hw;
638 	u32 j, num_rx_queues;
639 	u32 mrqc, rxcsum;
640 	u32 rss_key[10];
641 
642 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
643 	for (j = 0; j < 10; j++)
644 		wr32(IGC_RSSRK(j), rss_key[j]);
645 
646 	num_rx_queues = adapter->rss_queues;
647 
648 	if (adapter->rss_indir_tbl_init != num_rx_queues) {
649 		for (j = 0; j < IGC_RETA_SIZE; j++)
650 			adapter->rss_indir_tbl[j] =
651 			(j * num_rx_queues) / IGC_RETA_SIZE;
652 		adapter->rss_indir_tbl_init = num_rx_queues;
653 	}
654 	igc_write_rss_indir_tbl(adapter);
655 
656 	/* Disable raw packet checksumming so that RSS hash is placed in
657 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
658 	 * offloads as they are enabled by default
659 	 */
660 	rxcsum = rd32(IGC_RXCSUM);
661 	rxcsum |= IGC_RXCSUM_PCSD;
662 
663 	/* Enable Receive Checksum Offload for SCTP */
664 	rxcsum |= IGC_RXCSUM_CRCOFL;
665 
666 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
667 	wr32(IGC_RXCSUM, rxcsum);
668 
669 	/* Generate RSS hash based on packet types, TCP/UDP
670 	 * port numbers and/or IPv4/v6 src and dst addresses
671 	 */
672 	mrqc = IGC_MRQC_RSS_FIELD_IPV4 |
673 	       IGC_MRQC_RSS_FIELD_IPV4_TCP |
674 	       IGC_MRQC_RSS_FIELD_IPV6 |
675 	       IGC_MRQC_RSS_FIELD_IPV6_TCP |
676 	       IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
677 
678 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP)
679 		mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP;
680 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP)
681 		mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP;
682 
683 	mrqc |= IGC_MRQC_ENABLE_RSS_MQ;
684 
685 	wr32(IGC_MRQC, mrqc);
686 }
687 
688 /**
689  * igc_setup_rctl - configure the receive control registers
690  * @adapter: Board private structure
691  */
692 static void igc_setup_rctl(struct igc_adapter *adapter)
693 {
694 	struct igc_hw *hw = &adapter->hw;
695 	u32 rctl;
696 
697 	rctl = rd32(IGC_RCTL);
698 
699 	rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
700 	rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC);
701 
702 	rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF |
703 		(hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
704 
705 	/* enable stripping of CRC. Newer features require
706 	 * that the HW strips the CRC.
707 	 */
708 	rctl |= IGC_RCTL_SECRC;
709 
710 	/* disable store bad packets and clear size bits. */
711 	rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256);
712 
713 	/* enable LPE to allow for reception of jumbo frames */
714 	rctl |= IGC_RCTL_LPE;
715 
716 	/* disable queue 0 to prevent tail write w/o re-config */
717 	wr32(IGC_RXDCTL(0), 0);
718 
719 	/* This is useful for sniffing bad packets. */
720 	if (adapter->netdev->features & NETIF_F_RXALL) {
721 		/* UPE and MPE will be handled by normal PROMISC logic
722 		 * in set_rx_mode
723 		 */
724 		rctl |= (IGC_RCTL_SBP | /* Receive bad packets */
725 			 IGC_RCTL_BAM | /* RX All Bcast Pkts */
726 			 IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
727 
728 		rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */
729 			  IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */
730 	}
731 
732 	wr32(IGC_RCTL, rctl);
733 }
734 
735 /**
736  * igc_setup_tctl - configure the transmit control registers
737  * @adapter: Board private structure
738  */
739 static void igc_setup_tctl(struct igc_adapter *adapter)
740 {
741 	struct igc_hw *hw = &adapter->hw;
742 	u32 tctl;
743 
744 	/* disable queue 0 which icould be enabled by default */
745 	wr32(IGC_TXDCTL(0), 0);
746 
747 	/* Program the Transmit Control Register */
748 	tctl = rd32(IGC_TCTL);
749 	tctl &= ~IGC_TCTL_CT;
750 	tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC |
751 		(IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT);
752 
753 	/* Enable transmits */
754 	tctl |= IGC_TCTL_EN;
755 
756 	wr32(IGC_TCTL, tctl);
757 }
758 
759 /**
760  * igc_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
761  * @adapter: address of board private structure
762  * @index: Index of the RAR entry which need to be synced with MAC table
763  */
764 static void igc_rar_set_index(struct igc_adapter *adapter, u32 index)
765 {
766 	u8 *addr = adapter->mac_table[index].addr;
767 	struct igc_hw *hw = &adapter->hw;
768 	u32 rar_low, rar_high;
769 
770 	/* HW expects these to be in network order when they are plugged
771 	 * into the registers which are little endian.  In order to guarantee
772 	 * that ordering we need to do an leXX_to_cpup here in order to be
773 	 * ready for the byteswap that occurs with writel
774 	 */
775 	rar_low = le32_to_cpup((__le32 *)(addr));
776 	rar_high = le16_to_cpup((__le16 *)(addr + 4));
777 
778 	/* Indicate to hardware the Address is Valid. */
779 	if (adapter->mac_table[index].state & IGC_MAC_STATE_IN_USE) {
780 		if (is_valid_ether_addr(addr))
781 			rar_high |= IGC_RAH_AV;
782 
783 		rar_high |= IGC_RAH_POOL_1 <<
784 			adapter->mac_table[index].queue;
785 	}
786 
787 	wr32(IGC_RAL(index), rar_low);
788 	wrfl();
789 	wr32(IGC_RAH(index), rar_high);
790 	wrfl();
791 }
792 
793 /* Set default MAC address for the PF in the first RAR entry */
794 static void igc_set_default_mac_filter(struct igc_adapter *adapter)
795 {
796 	struct igc_mac_addr *mac_table = &adapter->mac_table[0];
797 
798 	ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
799 	mac_table->state = IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
800 
801 	igc_rar_set_index(adapter, 0);
802 }
803 
804 /**
805  * igc_set_mac - Change the Ethernet Address of the NIC
806  * @netdev: network interface device structure
807  * @p: pointer to an address structure
808  *
809  * Returns 0 on success, negative on failure
810  */
811 static int igc_set_mac(struct net_device *netdev, void *p)
812 {
813 	struct igc_adapter *adapter = netdev_priv(netdev);
814 	struct igc_hw *hw = &adapter->hw;
815 	struct sockaddr *addr = p;
816 
817 	if (!is_valid_ether_addr(addr->sa_data))
818 		return -EADDRNOTAVAIL;
819 
820 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
821 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
822 
823 	/* set the correct pool for the new PF MAC address in entry 0 */
824 	igc_set_default_mac_filter(adapter);
825 
826 	return 0;
827 }
828 
829 /**
830  *  igc_write_mc_addr_list - write multicast addresses to MTA
831  *  @netdev: network interface device structure
832  *
833  *  Writes multicast address list to the MTA hash table.
834  *  Returns: -ENOMEM on failure
835  *           0 on no addresses written
836  *           X on writing X addresses to MTA
837  **/
838 static int igc_write_mc_addr_list(struct net_device *netdev)
839 {
840 	struct igc_adapter *adapter = netdev_priv(netdev);
841 	struct igc_hw *hw = &adapter->hw;
842 	struct netdev_hw_addr *ha;
843 	u8  *mta_list;
844 	int i;
845 
846 	if (netdev_mc_empty(netdev)) {
847 		/* nothing to program, so clear mc list */
848 		igc_update_mc_addr_list(hw, NULL, 0);
849 		return 0;
850 	}
851 
852 	mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
853 	if (!mta_list)
854 		return -ENOMEM;
855 
856 	/* The shared function expects a packed array of only addresses. */
857 	i = 0;
858 	netdev_for_each_mc_addr(ha, netdev)
859 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
860 
861 	igc_update_mc_addr_list(hw, mta_list, i);
862 	kfree(mta_list);
863 
864 	return netdev_mc_count(netdev);
865 }
866 
867 static void igc_tx_ctxtdesc(struct igc_ring *tx_ring,
868 			    struct igc_tx_buffer *first,
869 			    u32 vlan_macip_lens, u32 type_tucmd,
870 			    u32 mss_l4len_idx)
871 {
872 	struct igc_adv_tx_context_desc *context_desc;
873 	u16 i = tx_ring->next_to_use;
874 	struct timespec64 ts;
875 
876 	context_desc = IGC_TX_CTXTDESC(tx_ring, i);
877 
878 	i++;
879 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
880 
881 	/* set bits to identify this as an advanced context descriptor */
882 	type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
883 
884 	/* For i225, context index must be unique per ring. */
885 	if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
886 		mss_l4len_idx |= tx_ring->reg_idx << 4;
887 
888 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
889 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
890 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
891 
892 	/* We assume there is always a valid Tx time available. Invalid times
893 	 * should have been handled by the upper layers.
894 	 */
895 	if (tx_ring->launchtime_enable) {
896 		ts = ktime_to_timespec64(first->skb->tstamp);
897 		first->skb->tstamp = ktime_set(0, 0);
898 		context_desc->launch_time = cpu_to_le32(ts.tv_nsec / 32);
899 	} else {
900 		context_desc->launch_time = 0;
901 	}
902 }
903 
904 static inline bool igc_ipv6_csum_is_sctp(struct sk_buff *skb)
905 {
906 	unsigned int offset = 0;
907 
908 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
909 
910 	return offset == skb_checksum_start_offset(skb);
911 }
912 
913 static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first)
914 {
915 	struct sk_buff *skb = first->skb;
916 	u32 vlan_macip_lens = 0;
917 	u32 type_tucmd = 0;
918 
919 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
920 csum_failed:
921 		if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) &&
922 		    !tx_ring->launchtime_enable)
923 			return;
924 		goto no_csum;
925 	}
926 
927 	switch (skb->csum_offset) {
928 	case offsetof(struct tcphdr, check):
929 		type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
930 		/* fall through */
931 	case offsetof(struct udphdr, check):
932 		break;
933 	case offsetof(struct sctphdr, checksum):
934 		/* validate that this is actually an SCTP request */
935 		if ((first->protocol == htons(ETH_P_IP) &&
936 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
937 		    (first->protocol == htons(ETH_P_IPV6) &&
938 		     igc_ipv6_csum_is_sctp(skb))) {
939 			type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP;
940 			break;
941 		}
942 		/* fall through */
943 	default:
944 		skb_checksum_help(skb);
945 		goto csum_failed;
946 	}
947 
948 	/* update TX checksum flag */
949 	first->tx_flags |= IGC_TX_FLAGS_CSUM;
950 	vlan_macip_lens = skb_checksum_start_offset(skb) -
951 			  skb_network_offset(skb);
952 no_csum:
953 	vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT;
954 	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
955 
956 	igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
957 }
958 
959 static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
960 {
961 	struct net_device *netdev = tx_ring->netdev;
962 
963 	netif_stop_subqueue(netdev, tx_ring->queue_index);
964 
965 	/* memory barriier comment */
966 	smp_mb();
967 
968 	/* We need to check again in a case another CPU has just
969 	 * made room available.
970 	 */
971 	if (igc_desc_unused(tx_ring) < size)
972 		return -EBUSY;
973 
974 	/* A reprieve! */
975 	netif_wake_subqueue(netdev, tx_ring->queue_index);
976 
977 	u64_stats_update_begin(&tx_ring->tx_syncp2);
978 	tx_ring->tx_stats.restart_queue2++;
979 	u64_stats_update_end(&tx_ring->tx_syncp2);
980 
981 	return 0;
982 }
983 
984 static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
985 {
986 	if (igc_desc_unused(tx_ring) >= size)
987 		return 0;
988 	return __igc_maybe_stop_tx(tx_ring, size);
989 }
990 
991 #define IGC_SET_FLAG(_input, _flag, _result) \
992 	(((_flag) <= (_result)) ?				\
993 	 ((u32)((_input) & (_flag)) * ((_result) / (_flag))) :	\
994 	 ((u32)((_input) & (_flag)) / ((_flag) / (_result))))
995 
996 static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
997 {
998 	/* set type for advanced descriptor with frame checksum insertion */
999 	u32 cmd_type = IGC_ADVTXD_DTYP_DATA |
1000 		       IGC_ADVTXD_DCMD_DEXT |
1001 		       IGC_ADVTXD_DCMD_IFCS;
1002 
1003 	/* set segmentation bits for TSO */
1004 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO,
1005 				 (IGC_ADVTXD_DCMD_TSE));
1006 
1007 	/* set timestamp bit if present */
1008 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP,
1009 				 (IGC_ADVTXD_MAC_TSTAMP));
1010 
1011 	return cmd_type;
1012 }
1013 
1014 static void igc_tx_olinfo_status(struct igc_ring *tx_ring,
1015 				 union igc_adv_tx_desc *tx_desc,
1016 				 u32 tx_flags, unsigned int paylen)
1017 {
1018 	u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT;
1019 
1020 	/* insert L4 checksum */
1021 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_CSUM) *
1022 			  ((IGC_TXD_POPTS_TXSM << 8) /
1023 			  IGC_TX_FLAGS_CSUM);
1024 
1025 	/* insert IPv4 checksum */
1026 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_IPV4) *
1027 			  (((IGC_TXD_POPTS_IXSM << 8)) /
1028 			  IGC_TX_FLAGS_IPV4);
1029 
1030 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1031 }
1032 
1033 static int igc_tx_map(struct igc_ring *tx_ring,
1034 		      struct igc_tx_buffer *first,
1035 		      const u8 hdr_len)
1036 {
1037 	struct sk_buff *skb = first->skb;
1038 	struct igc_tx_buffer *tx_buffer;
1039 	union igc_adv_tx_desc *tx_desc;
1040 	u32 tx_flags = first->tx_flags;
1041 	skb_frag_t *frag;
1042 	u16 i = tx_ring->next_to_use;
1043 	unsigned int data_len, size;
1044 	dma_addr_t dma;
1045 	u32 cmd_type = igc_tx_cmd_type(skb, tx_flags);
1046 
1047 	tx_desc = IGC_TX_DESC(tx_ring, i);
1048 
1049 	igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
1050 
1051 	size = skb_headlen(skb);
1052 	data_len = skb->data_len;
1053 
1054 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1055 
1056 	tx_buffer = first;
1057 
1058 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1059 		if (dma_mapping_error(tx_ring->dev, dma))
1060 			goto dma_error;
1061 
1062 		/* record length, and DMA address */
1063 		dma_unmap_len_set(tx_buffer, len, size);
1064 		dma_unmap_addr_set(tx_buffer, dma, dma);
1065 
1066 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
1067 
1068 		while (unlikely(size > IGC_MAX_DATA_PER_TXD)) {
1069 			tx_desc->read.cmd_type_len =
1070 				cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD);
1071 
1072 			i++;
1073 			tx_desc++;
1074 			if (i == tx_ring->count) {
1075 				tx_desc = IGC_TX_DESC(tx_ring, 0);
1076 				i = 0;
1077 			}
1078 			tx_desc->read.olinfo_status = 0;
1079 
1080 			dma += IGC_MAX_DATA_PER_TXD;
1081 			size -= IGC_MAX_DATA_PER_TXD;
1082 
1083 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
1084 		}
1085 
1086 		if (likely(!data_len))
1087 			break;
1088 
1089 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
1090 
1091 		i++;
1092 		tx_desc++;
1093 		if (i == tx_ring->count) {
1094 			tx_desc = IGC_TX_DESC(tx_ring, 0);
1095 			i = 0;
1096 		}
1097 		tx_desc->read.olinfo_status = 0;
1098 
1099 		size = skb_frag_size(frag);
1100 		data_len -= size;
1101 
1102 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
1103 				       size, DMA_TO_DEVICE);
1104 
1105 		tx_buffer = &tx_ring->tx_buffer_info[i];
1106 	}
1107 
1108 	/* write last descriptor with RS and EOP bits */
1109 	cmd_type |= size | IGC_TXD_DCMD;
1110 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1111 
1112 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1113 
1114 	/* set the timestamp */
1115 	first->time_stamp = jiffies;
1116 
1117 	skb_tx_timestamp(skb);
1118 
1119 	/* Force memory writes to complete before letting h/w know there
1120 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1121 	 * memory model archs, such as IA-64).
1122 	 *
1123 	 * We also need this memory barrier to make certain all of the
1124 	 * status bits have been updated before next_to_watch is written.
1125 	 */
1126 	wmb();
1127 
1128 	/* set next_to_watch value indicating a packet is present */
1129 	first->next_to_watch = tx_desc;
1130 
1131 	i++;
1132 	if (i == tx_ring->count)
1133 		i = 0;
1134 
1135 	tx_ring->next_to_use = i;
1136 
1137 	/* Make sure there is space in the ring for the next send. */
1138 	igc_maybe_stop_tx(tx_ring, DESC_NEEDED);
1139 
1140 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1141 		writel(i, tx_ring->tail);
1142 	}
1143 
1144 	return 0;
1145 dma_error:
1146 	dev_err(tx_ring->dev, "TX DMA map failed\n");
1147 	tx_buffer = &tx_ring->tx_buffer_info[i];
1148 
1149 	/* clear dma mappings for failed tx_buffer_info map */
1150 	while (tx_buffer != first) {
1151 		if (dma_unmap_len(tx_buffer, len))
1152 			dma_unmap_page(tx_ring->dev,
1153 				       dma_unmap_addr(tx_buffer, dma),
1154 				       dma_unmap_len(tx_buffer, len),
1155 				       DMA_TO_DEVICE);
1156 		dma_unmap_len_set(tx_buffer, len, 0);
1157 
1158 		if (i-- == 0)
1159 			i += tx_ring->count;
1160 		tx_buffer = &tx_ring->tx_buffer_info[i];
1161 	}
1162 
1163 	if (dma_unmap_len(tx_buffer, len))
1164 		dma_unmap_single(tx_ring->dev,
1165 				 dma_unmap_addr(tx_buffer, dma),
1166 				 dma_unmap_len(tx_buffer, len),
1167 				 DMA_TO_DEVICE);
1168 	dma_unmap_len_set(tx_buffer, len, 0);
1169 
1170 	dev_kfree_skb_any(tx_buffer->skb);
1171 	tx_buffer->skb = NULL;
1172 
1173 	tx_ring->next_to_use = i;
1174 
1175 	return -1;
1176 }
1177 
1178 static int igc_tso(struct igc_ring *tx_ring,
1179 		   struct igc_tx_buffer *first,
1180 		   u8 *hdr_len)
1181 {
1182 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1183 	struct sk_buff *skb = first->skb;
1184 	union {
1185 		struct iphdr *v4;
1186 		struct ipv6hdr *v6;
1187 		unsigned char *hdr;
1188 	} ip;
1189 	union {
1190 		struct tcphdr *tcp;
1191 		struct udphdr *udp;
1192 		unsigned char *hdr;
1193 	} l4;
1194 	u32 paylen, l4_offset;
1195 	int err;
1196 
1197 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1198 		return 0;
1199 
1200 	if (!skb_is_gso(skb))
1201 		return 0;
1202 
1203 	err = skb_cow_head(skb, 0);
1204 	if (err < 0)
1205 		return err;
1206 
1207 	ip.hdr = skb_network_header(skb);
1208 	l4.hdr = skb_checksum_start(skb);
1209 
1210 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1211 	type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1212 
1213 	/* initialize outer IP header fields */
1214 	if (ip.v4->version == 4) {
1215 		unsigned char *csum_start = skb_checksum_start(skb);
1216 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
1217 
1218 		/* IP header will have to cancel out any data that
1219 		 * is not a part of the outer IP header
1220 		 */
1221 		ip.v4->check = csum_fold(csum_partial(trans_start,
1222 						      csum_start - trans_start,
1223 						      0));
1224 		type_tucmd |= IGC_ADVTXD_TUCMD_IPV4;
1225 
1226 		ip.v4->tot_len = 0;
1227 		first->tx_flags |= IGC_TX_FLAGS_TSO |
1228 				   IGC_TX_FLAGS_CSUM |
1229 				   IGC_TX_FLAGS_IPV4;
1230 	} else {
1231 		ip.v6->payload_len = 0;
1232 		first->tx_flags |= IGC_TX_FLAGS_TSO |
1233 				   IGC_TX_FLAGS_CSUM;
1234 	}
1235 
1236 	/* determine offset of inner transport header */
1237 	l4_offset = l4.hdr - skb->data;
1238 
1239 	/* remove payload length from inner checksum */
1240 	paylen = skb->len - l4_offset;
1241 	if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) {
1242 		/* compute length of segmentation header */
1243 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
1244 		csum_replace_by_diff(&l4.tcp->check,
1245 				     (__force __wsum)htonl(paylen));
1246 	} else {
1247 		/* compute length of segmentation header */
1248 		*hdr_len = sizeof(*l4.udp) + l4_offset;
1249 		csum_replace_by_diff(&l4.udp->check,
1250 				     (__force __wsum)htonl(paylen));
1251 	}
1252 
1253 	/* update gso size and bytecount with header size */
1254 	first->gso_segs = skb_shinfo(skb)->gso_segs;
1255 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
1256 
1257 	/* MSS L4LEN IDX */
1258 	mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT;
1259 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT;
1260 
1261 	/* VLAN MACLEN IPLEN */
1262 	vlan_macip_lens = l4.hdr - ip.hdr;
1263 	vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT;
1264 	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1265 
1266 	igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
1267 			type_tucmd, mss_l4len_idx);
1268 
1269 	return 1;
1270 }
1271 
1272 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
1273 				       struct igc_ring *tx_ring)
1274 {
1275 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1276 	__be16 protocol = vlan_get_protocol(skb);
1277 	struct igc_tx_buffer *first;
1278 	u32 tx_flags = 0;
1279 	unsigned short f;
1280 	u8 hdr_len = 0;
1281 	int tso = 0;
1282 
1283 	/* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
1284 	 *	+ 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
1285 	 *	+ 2 desc gap to keep tail from touching head,
1286 	 *	+ 1 desc for context descriptor,
1287 	 * otherwise try next time
1288 	 */
1289 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1290 		count += TXD_USE_COUNT(skb_frag_size(
1291 						&skb_shinfo(skb)->frags[f]));
1292 
1293 	if (igc_maybe_stop_tx(tx_ring, count + 3)) {
1294 		/* this is a hard error */
1295 		return NETDEV_TX_BUSY;
1296 	}
1297 
1298 	/* record the location of the first descriptor for this packet */
1299 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1300 	first->skb = skb;
1301 	first->bytecount = skb->len;
1302 	first->gso_segs = 1;
1303 
1304 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
1305 		struct igc_adapter *adapter = netdev_priv(tx_ring->netdev);
1306 
1307 		/* FIXME: add support for retrieving timestamps from
1308 		 * the other timer registers before skipping the
1309 		 * timestamping request.
1310 		 */
1311 		if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
1312 		    !test_and_set_bit_lock(__IGC_PTP_TX_IN_PROGRESS,
1313 					   &adapter->state)) {
1314 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1315 			tx_flags |= IGC_TX_FLAGS_TSTAMP;
1316 
1317 			adapter->ptp_tx_skb = skb_get(skb);
1318 			adapter->ptp_tx_start = jiffies;
1319 		} else {
1320 			adapter->tx_hwtstamp_skipped++;
1321 		}
1322 	}
1323 
1324 	/* record initial flags and protocol */
1325 	first->tx_flags = tx_flags;
1326 	first->protocol = protocol;
1327 
1328 	tso = igc_tso(tx_ring, first, &hdr_len);
1329 	if (tso < 0)
1330 		goto out_drop;
1331 	else if (!tso)
1332 		igc_tx_csum(tx_ring, first);
1333 
1334 	igc_tx_map(tx_ring, first, hdr_len);
1335 
1336 	return NETDEV_TX_OK;
1337 
1338 out_drop:
1339 	dev_kfree_skb_any(first->skb);
1340 	first->skb = NULL;
1341 
1342 	return NETDEV_TX_OK;
1343 }
1344 
1345 static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter,
1346 						    struct sk_buff *skb)
1347 {
1348 	unsigned int r_idx = skb->queue_mapping;
1349 
1350 	if (r_idx >= adapter->num_tx_queues)
1351 		r_idx = r_idx % adapter->num_tx_queues;
1352 
1353 	return adapter->tx_ring[r_idx];
1354 }
1355 
1356 static netdev_tx_t igc_xmit_frame(struct sk_buff *skb,
1357 				  struct net_device *netdev)
1358 {
1359 	struct igc_adapter *adapter = netdev_priv(netdev);
1360 
1361 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
1362 	 * in order to meet this minimum size requirement.
1363 	 */
1364 	if (skb->len < 17) {
1365 		if (skb_padto(skb, 17))
1366 			return NETDEV_TX_OK;
1367 		skb->len = 17;
1368 	}
1369 
1370 	return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb));
1371 }
1372 
1373 static void igc_rx_checksum(struct igc_ring *ring,
1374 			    union igc_adv_rx_desc *rx_desc,
1375 			    struct sk_buff *skb)
1376 {
1377 	skb_checksum_none_assert(skb);
1378 
1379 	/* Ignore Checksum bit is set */
1380 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM))
1381 		return;
1382 
1383 	/* Rx checksum disabled via ethtool */
1384 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1385 		return;
1386 
1387 	/* TCP/UDP checksum error bit is set */
1388 	if (igc_test_staterr(rx_desc,
1389 			     IGC_RXDEXT_STATERR_TCPE |
1390 			     IGC_RXDEXT_STATERR_IPE)) {
1391 		/* work around errata with sctp packets where the TCPE aka
1392 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
1393 		 * packets (aka let the stack check the crc32c)
1394 		 */
1395 		if (!(skb->len == 60 &&
1396 		      test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
1397 			u64_stats_update_begin(&ring->rx_syncp);
1398 			ring->rx_stats.csum_err++;
1399 			u64_stats_update_end(&ring->rx_syncp);
1400 		}
1401 		/* let the stack verify checksum errors */
1402 		return;
1403 	}
1404 	/* It must be a TCP or UDP packet with a valid checksum */
1405 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS |
1406 				      IGC_RXD_STAT_UDPCS))
1407 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1408 
1409 	dev_dbg(ring->dev, "cksum success: bits %08X\n",
1410 		le32_to_cpu(rx_desc->wb.upper.status_error));
1411 }
1412 
1413 static inline void igc_rx_hash(struct igc_ring *ring,
1414 			       union igc_adv_rx_desc *rx_desc,
1415 			       struct sk_buff *skb)
1416 {
1417 	if (ring->netdev->features & NETIF_F_RXHASH)
1418 		skb_set_hash(skb,
1419 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1420 			     PKT_HASH_TYPE_L3);
1421 }
1422 
1423 /**
1424  * igc_process_skb_fields - Populate skb header fields from Rx descriptor
1425  * @rx_ring: rx descriptor ring packet is being transacted on
1426  * @rx_desc: pointer to the EOP Rx descriptor
1427  * @skb: pointer to current skb being populated
1428  *
1429  * This function checks the ring, descriptor, and packet information in
1430  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1431  * other fields within the skb.
1432  */
1433 static void igc_process_skb_fields(struct igc_ring *rx_ring,
1434 				   union igc_adv_rx_desc *rx_desc,
1435 				   struct sk_buff *skb)
1436 {
1437 	igc_rx_hash(rx_ring, rx_desc, skb);
1438 
1439 	igc_rx_checksum(rx_ring, rx_desc, skb);
1440 
1441 	if (igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TS) &&
1442 	    !igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP))
1443 		igc_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
1444 
1445 	skb_record_rx_queue(skb, rx_ring->queue_index);
1446 
1447 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1448 }
1449 
1450 static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring,
1451 					       const unsigned int size)
1452 {
1453 	struct igc_rx_buffer *rx_buffer;
1454 
1455 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1456 	prefetchw(rx_buffer->page);
1457 
1458 	/* we are reusing so sync this buffer for CPU use */
1459 	dma_sync_single_range_for_cpu(rx_ring->dev,
1460 				      rx_buffer->dma,
1461 				      rx_buffer->page_offset,
1462 				      size,
1463 				      DMA_FROM_DEVICE);
1464 
1465 	rx_buffer->pagecnt_bias--;
1466 
1467 	return rx_buffer;
1468 }
1469 
1470 /**
1471  * igc_add_rx_frag - Add contents of Rx buffer to sk_buff
1472  * @rx_ring: rx descriptor ring to transact packets on
1473  * @rx_buffer: buffer containing page to add
1474  * @skb: sk_buff to place the data into
1475  * @size: size of buffer to be added
1476  *
1477  * This function will add the data contained in rx_buffer->page to the skb.
1478  */
1479 static void igc_add_rx_frag(struct igc_ring *rx_ring,
1480 			    struct igc_rx_buffer *rx_buffer,
1481 			    struct sk_buff *skb,
1482 			    unsigned int size)
1483 {
1484 #if (PAGE_SIZE < 8192)
1485 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1486 
1487 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1488 			rx_buffer->page_offset, size, truesize);
1489 	rx_buffer->page_offset ^= truesize;
1490 #else
1491 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1492 				SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1493 				SKB_DATA_ALIGN(size);
1494 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1495 			rx_buffer->page_offset, size, truesize);
1496 	rx_buffer->page_offset += truesize;
1497 #endif
1498 }
1499 
1500 static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring,
1501 				     struct igc_rx_buffer *rx_buffer,
1502 				     union igc_adv_rx_desc *rx_desc,
1503 				     unsigned int size)
1504 {
1505 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1506 #if (PAGE_SIZE < 8192)
1507 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1508 #else
1509 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1510 				SKB_DATA_ALIGN(IGC_SKB_PAD + size);
1511 #endif
1512 	struct sk_buff *skb;
1513 
1514 	/* prefetch first cache line of first page */
1515 	prefetch(va);
1516 #if L1_CACHE_BYTES < 128
1517 	prefetch(va + L1_CACHE_BYTES);
1518 #endif
1519 
1520 	/* build an skb around the page buffer */
1521 	skb = build_skb(va - IGC_SKB_PAD, truesize);
1522 	if (unlikely(!skb))
1523 		return NULL;
1524 
1525 	/* update pointers within the skb to store the data */
1526 	skb_reserve(skb, IGC_SKB_PAD);
1527 	__skb_put(skb, size);
1528 
1529 	/* update buffer offset */
1530 #if (PAGE_SIZE < 8192)
1531 	rx_buffer->page_offset ^= truesize;
1532 #else
1533 	rx_buffer->page_offset += truesize;
1534 #endif
1535 
1536 	return skb;
1537 }
1538 
1539 static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring,
1540 					 struct igc_rx_buffer *rx_buffer,
1541 					 union igc_adv_rx_desc *rx_desc,
1542 					 unsigned int size)
1543 {
1544 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1545 #if (PAGE_SIZE < 8192)
1546 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1547 #else
1548 	unsigned int truesize = SKB_DATA_ALIGN(size);
1549 #endif
1550 	unsigned int headlen;
1551 	struct sk_buff *skb;
1552 
1553 	/* prefetch first cache line of first page */
1554 	prefetch(va);
1555 #if L1_CACHE_BYTES < 128
1556 	prefetch(va + L1_CACHE_BYTES);
1557 #endif
1558 
1559 	/* allocate a skb to store the frags */
1560 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGC_RX_HDR_LEN);
1561 	if (unlikely(!skb))
1562 		return NULL;
1563 
1564 	if (unlikely(igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP))) {
1565 		igc_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
1566 		va += IGC_TS_HDR_LEN;
1567 		size -= IGC_TS_HDR_LEN;
1568 	}
1569 
1570 	/* Determine available headroom for copy */
1571 	headlen = size;
1572 	if (headlen > IGC_RX_HDR_LEN)
1573 		headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN);
1574 
1575 	/* align pull length to size of long to optimize memcpy performance */
1576 	memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1577 
1578 	/* update all of the pointers */
1579 	size -= headlen;
1580 	if (size) {
1581 		skb_add_rx_frag(skb, 0, rx_buffer->page,
1582 				(va + headlen) - page_address(rx_buffer->page),
1583 				size, truesize);
1584 #if (PAGE_SIZE < 8192)
1585 		rx_buffer->page_offset ^= truesize;
1586 #else
1587 		rx_buffer->page_offset += truesize;
1588 #endif
1589 	} else {
1590 		rx_buffer->pagecnt_bias++;
1591 	}
1592 
1593 	return skb;
1594 }
1595 
1596 /**
1597  * igc_reuse_rx_page - page flip buffer and store it back on the ring
1598  * @rx_ring: rx descriptor ring to store buffers on
1599  * @old_buff: donor buffer to have page reused
1600  *
1601  * Synchronizes page for reuse by the adapter
1602  */
1603 static void igc_reuse_rx_page(struct igc_ring *rx_ring,
1604 			      struct igc_rx_buffer *old_buff)
1605 {
1606 	u16 nta = rx_ring->next_to_alloc;
1607 	struct igc_rx_buffer *new_buff;
1608 
1609 	new_buff = &rx_ring->rx_buffer_info[nta];
1610 
1611 	/* update, and store next to alloc */
1612 	nta++;
1613 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1614 
1615 	/* Transfer page from old buffer to new buffer.
1616 	 * Move each member individually to avoid possible store
1617 	 * forwarding stalls.
1618 	 */
1619 	new_buff->dma		= old_buff->dma;
1620 	new_buff->page		= old_buff->page;
1621 	new_buff->page_offset	= old_buff->page_offset;
1622 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1623 }
1624 
1625 static inline bool igc_page_is_reserved(struct page *page)
1626 {
1627 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1628 }
1629 
1630 static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer)
1631 {
1632 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1633 	struct page *page = rx_buffer->page;
1634 
1635 	/* avoid re-using remote pages */
1636 	if (unlikely(igc_page_is_reserved(page)))
1637 		return false;
1638 
1639 #if (PAGE_SIZE < 8192)
1640 	/* if we are only owner of page we can reuse it */
1641 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1642 		return false;
1643 #else
1644 #define IGC_LAST_OFFSET \
1645 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048)
1646 
1647 	if (rx_buffer->page_offset > IGC_LAST_OFFSET)
1648 		return false;
1649 #endif
1650 
1651 	/* If we have drained the page fragment pool we need to update
1652 	 * the pagecnt_bias and page count so that we fully restock the
1653 	 * number of references the driver holds.
1654 	 */
1655 	if (unlikely(!pagecnt_bias)) {
1656 		page_ref_add(page, USHRT_MAX);
1657 		rx_buffer->pagecnt_bias = USHRT_MAX;
1658 	}
1659 
1660 	return true;
1661 }
1662 
1663 /**
1664  * igc_is_non_eop - process handling of non-EOP buffers
1665  * @rx_ring: Rx ring being processed
1666  * @rx_desc: Rx descriptor for current buffer
1667  *
1668  * This function updates next to clean.  If the buffer is an EOP buffer
1669  * this function exits returning false, otherwise it will place the
1670  * sk_buff in the next buffer to be chained and return true indicating
1671  * that this is in fact a non-EOP buffer.
1672  */
1673 static bool igc_is_non_eop(struct igc_ring *rx_ring,
1674 			   union igc_adv_rx_desc *rx_desc)
1675 {
1676 	u32 ntc = rx_ring->next_to_clean + 1;
1677 
1678 	/* fetch, update, and store next to clean */
1679 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1680 	rx_ring->next_to_clean = ntc;
1681 
1682 	prefetch(IGC_RX_DESC(rx_ring, ntc));
1683 
1684 	if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP)))
1685 		return false;
1686 
1687 	return true;
1688 }
1689 
1690 /**
1691  * igc_cleanup_headers - Correct corrupted or empty headers
1692  * @rx_ring: rx descriptor ring packet is being transacted on
1693  * @rx_desc: pointer to the EOP Rx descriptor
1694  * @skb: pointer to current skb being fixed
1695  *
1696  * Address the case where we are pulling data in on pages only
1697  * and as such no data is present in the skb header.
1698  *
1699  * In addition if skb is not at least 60 bytes we need to pad it so that
1700  * it is large enough to qualify as a valid Ethernet frame.
1701  *
1702  * Returns true if an error was encountered and skb was freed.
1703  */
1704 static bool igc_cleanup_headers(struct igc_ring *rx_ring,
1705 				union igc_adv_rx_desc *rx_desc,
1706 				struct sk_buff *skb)
1707 {
1708 	if (unlikely((igc_test_staterr(rx_desc,
1709 				       IGC_RXDEXT_ERR_FRAME_ERR_MASK)))) {
1710 		struct net_device *netdev = rx_ring->netdev;
1711 
1712 		if (!(netdev->features & NETIF_F_RXALL)) {
1713 			dev_kfree_skb_any(skb);
1714 			return true;
1715 		}
1716 	}
1717 
1718 	/* if eth_skb_pad returns an error the skb was freed */
1719 	if (eth_skb_pad(skb))
1720 		return true;
1721 
1722 	return false;
1723 }
1724 
1725 static void igc_put_rx_buffer(struct igc_ring *rx_ring,
1726 			      struct igc_rx_buffer *rx_buffer)
1727 {
1728 	if (igc_can_reuse_rx_page(rx_buffer)) {
1729 		/* hand second half of page back to the ring */
1730 		igc_reuse_rx_page(rx_ring, rx_buffer);
1731 	} else {
1732 		/* We are not reusing the buffer so unmap it and free
1733 		 * any references we are holding to it
1734 		 */
1735 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
1736 				     igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
1737 				     IGC_RX_DMA_ATTR);
1738 		__page_frag_cache_drain(rx_buffer->page,
1739 					rx_buffer->pagecnt_bias);
1740 	}
1741 
1742 	/* clear contents of rx_buffer */
1743 	rx_buffer->page = NULL;
1744 }
1745 
1746 static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring)
1747 {
1748 	return ring_uses_build_skb(rx_ring) ? IGC_SKB_PAD : 0;
1749 }
1750 
1751 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
1752 				  struct igc_rx_buffer *bi)
1753 {
1754 	struct page *page = bi->page;
1755 	dma_addr_t dma;
1756 
1757 	/* since we are recycling buffers we should seldom need to alloc */
1758 	if (likely(page))
1759 		return true;
1760 
1761 	/* alloc new page for storage */
1762 	page = dev_alloc_pages(igc_rx_pg_order(rx_ring));
1763 	if (unlikely(!page)) {
1764 		rx_ring->rx_stats.alloc_failed++;
1765 		return false;
1766 	}
1767 
1768 	/* map page for use */
1769 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1770 				 igc_rx_pg_size(rx_ring),
1771 				 DMA_FROM_DEVICE,
1772 				 IGC_RX_DMA_ATTR);
1773 
1774 	/* if mapping failed free memory back to system since
1775 	 * there isn't much point in holding memory we can't use
1776 	 */
1777 	if (dma_mapping_error(rx_ring->dev, dma)) {
1778 		__free_page(page);
1779 
1780 		rx_ring->rx_stats.alloc_failed++;
1781 		return false;
1782 	}
1783 
1784 	bi->dma = dma;
1785 	bi->page = page;
1786 	bi->page_offset = igc_rx_offset(rx_ring);
1787 	bi->pagecnt_bias = 1;
1788 
1789 	return true;
1790 }
1791 
1792 /**
1793  * igc_alloc_rx_buffers - Replace used receive buffers; packet split
1794  * @rx_ring: rx descriptor ring
1795  * @cleaned_count: number of buffers to clean
1796  */
1797 static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
1798 {
1799 	union igc_adv_rx_desc *rx_desc;
1800 	u16 i = rx_ring->next_to_use;
1801 	struct igc_rx_buffer *bi;
1802 	u16 bufsz;
1803 
1804 	/* nothing to do */
1805 	if (!cleaned_count)
1806 		return;
1807 
1808 	rx_desc = IGC_RX_DESC(rx_ring, i);
1809 	bi = &rx_ring->rx_buffer_info[i];
1810 	i -= rx_ring->count;
1811 
1812 	bufsz = igc_rx_bufsz(rx_ring);
1813 
1814 	do {
1815 		if (!igc_alloc_mapped_page(rx_ring, bi))
1816 			break;
1817 
1818 		/* sync the buffer for use by the device */
1819 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1820 						 bi->page_offset, bufsz,
1821 						 DMA_FROM_DEVICE);
1822 
1823 		/* Refresh the desc even if buffer_addrs didn't change
1824 		 * because each write-back erases this info.
1825 		 */
1826 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1827 
1828 		rx_desc++;
1829 		bi++;
1830 		i++;
1831 		if (unlikely(!i)) {
1832 			rx_desc = IGC_RX_DESC(rx_ring, 0);
1833 			bi = rx_ring->rx_buffer_info;
1834 			i -= rx_ring->count;
1835 		}
1836 
1837 		/* clear the length for the next_to_use descriptor */
1838 		rx_desc->wb.upper.length = 0;
1839 
1840 		cleaned_count--;
1841 	} while (cleaned_count);
1842 
1843 	i += rx_ring->count;
1844 
1845 	if (rx_ring->next_to_use != i) {
1846 		/* record the next descriptor to use */
1847 		rx_ring->next_to_use = i;
1848 
1849 		/* update next to alloc since we have filled the ring */
1850 		rx_ring->next_to_alloc = i;
1851 
1852 		/* Force memory writes to complete before letting h/w
1853 		 * know there are new descriptors to fetch.  (Only
1854 		 * applicable for weak-ordered memory model archs,
1855 		 * such as IA-64).
1856 		 */
1857 		wmb();
1858 		writel(i, rx_ring->tail);
1859 	}
1860 }
1861 
1862 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
1863 {
1864 	unsigned int total_bytes = 0, total_packets = 0;
1865 	struct igc_ring *rx_ring = q_vector->rx.ring;
1866 	struct sk_buff *skb = rx_ring->skb;
1867 	u16 cleaned_count = igc_desc_unused(rx_ring);
1868 
1869 	while (likely(total_packets < budget)) {
1870 		union igc_adv_rx_desc *rx_desc;
1871 		struct igc_rx_buffer *rx_buffer;
1872 		unsigned int size;
1873 
1874 		/* return some buffers to hardware, one at a time is too slow */
1875 		if (cleaned_count >= IGC_RX_BUFFER_WRITE) {
1876 			igc_alloc_rx_buffers(rx_ring, cleaned_count);
1877 			cleaned_count = 0;
1878 		}
1879 
1880 		rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean);
1881 		size = le16_to_cpu(rx_desc->wb.upper.length);
1882 		if (!size)
1883 			break;
1884 
1885 		/* This memory barrier is needed to keep us from reading
1886 		 * any other fields out of the rx_desc until we know the
1887 		 * descriptor has been written back
1888 		 */
1889 		dma_rmb();
1890 
1891 		rx_buffer = igc_get_rx_buffer(rx_ring, size);
1892 
1893 		/* retrieve a buffer from the ring */
1894 		if (skb)
1895 			igc_add_rx_frag(rx_ring, rx_buffer, skb, size);
1896 		else if (ring_uses_build_skb(rx_ring))
1897 			skb = igc_build_skb(rx_ring, rx_buffer, rx_desc, size);
1898 		else
1899 			skb = igc_construct_skb(rx_ring, rx_buffer,
1900 						rx_desc, size);
1901 
1902 		/* exit if we failed to retrieve a buffer */
1903 		if (!skb) {
1904 			rx_ring->rx_stats.alloc_failed++;
1905 			rx_buffer->pagecnt_bias++;
1906 			break;
1907 		}
1908 
1909 		igc_put_rx_buffer(rx_ring, rx_buffer);
1910 		cleaned_count++;
1911 
1912 		/* fetch next buffer in frame if non-eop */
1913 		if (igc_is_non_eop(rx_ring, rx_desc))
1914 			continue;
1915 
1916 		/* verify the packet layout is correct */
1917 		if (igc_cleanup_headers(rx_ring, rx_desc, skb)) {
1918 			skb = NULL;
1919 			continue;
1920 		}
1921 
1922 		/* probably a little skewed due to removing CRC */
1923 		total_bytes += skb->len;
1924 
1925 		/* populate checksum, timestamp, VLAN, and protocol */
1926 		igc_process_skb_fields(rx_ring, rx_desc, skb);
1927 
1928 		napi_gro_receive(&q_vector->napi, skb);
1929 
1930 		/* reset skb pointer */
1931 		skb = NULL;
1932 
1933 		/* update budget accounting */
1934 		total_packets++;
1935 	}
1936 
1937 	/* place incomplete frames back on ring for completion */
1938 	rx_ring->skb = skb;
1939 
1940 	u64_stats_update_begin(&rx_ring->rx_syncp);
1941 	rx_ring->rx_stats.packets += total_packets;
1942 	rx_ring->rx_stats.bytes += total_bytes;
1943 	u64_stats_update_end(&rx_ring->rx_syncp);
1944 	q_vector->rx.total_packets += total_packets;
1945 	q_vector->rx.total_bytes += total_bytes;
1946 
1947 	if (cleaned_count)
1948 		igc_alloc_rx_buffers(rx_ring, cleaned_count);
1949 
1950 	return total_packets;
1951 }
1952 
1953 /**
1954  * igc_clean_tx_irq - Reclaim resources after transmit completes
1955  * @q_vector: pointer to q_vector containing needed info
1956  * @napi_budget: Used to determine if we are in netpoll
1957  *
1958  * returns true if ring is completely cleaned
1959  */
1960 static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
1961 {
1962 	struct igc_adapter *adapter = q_vector->adapter;
1963 	unsigned int total_bytes = 0, total_packets = 0;
1964 	unsigned int budget = q_vector->tx.work_limit;
1965 	struct igc_ring *tx_ring = q_vector->tx.ring;
1966 	unsigned int i = tx_ring->next_to_clean;
1967 	struct igc_tx_buffer *tx_buffer;
1968 	union igc_adv_tx_desc *tx_desc;
1969 
1970 	if (test_bit(__IGC_DOWN, &adapter->state))
1971 		return true;
1972 
1973 	tx_buffer = &tx_ring->tx_buffer_info[i];
1974 	tx_desc = IGC_TX_DESC(tx_ring, i);
1975 	i -= tx_ring->count;
1976 
1977 	do {
1978 		union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1979 
1980 		/* if next_to_watch is not set then there is no work pending */
1981 		if (!eop_desc)
1982 			break;
1983 
1984 		/* prevent any other reads prior to eop_desc */
1985 		smp_rmb();
1986 
1987 		/* if DD is not set pending work has not been completed */
1988 		if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD)))
1989 			break;
1990 
1991 		/* clear next_to_watch to prevent false hangs */
1992 		tx_buffer->next_to_watch = NULL;
1993 
1994 		/* update the statistics for this packet */
1995 		total_bytes += tx_buffer->bytecount;
1996 		total_packets += tx_buffer->gso_segs;
1997 
1998 		/* free the skb */
1999 		napi_consume_skb(tx_buffer->skb, napi_budget);
2000 
2001 		/* unmap skb header data */
2002 		dma_unmap_single(tx_ring->dev,
2003 				 dma_unmap_addr(tx_buffer, dma),
2004 				 dma_unmap_len(tx_buffer, len),
2005 				 DMA_TO_DEVICE);
2006 
2007 		/* clear tx_buffer data */
2008 		dma_unmap_len_set(tx_buffer, len, 0);
2009 
2010 		/* clear last DMA location and unmap remaining buffers */
2011 		while (tx_desc != eop_desc) {
2012 			tx_buffer++;
2013 			tx_desc++;
2014 			i++;
2015 			if (unlikely(!i)) {
2016 				i -= tx_ring->count;
2017 				tx_buffer = tx_ring->tx_buffer_info;
2018 				tx_desc = IGC_TX_DESC(tx_ring, 0);
2019 			}
2020 
2021 			/* unmap any remaining paged data */
2022 			if (dma_unmap_len(tx_buffer, len)) {
2023 				dma_unmap_page(tx_ring->dev,
2024 					       dma_unmap_addr(tx_buffer, dma),
2025 					       dma_unmap_len(tx_buffer, len),
2026 					       DMA_TO_DEVICE);
2027 				dma_unmap_len_set(tx_buffer, len, 0);
2028 			}
2029 		}
2030 
2031 		/* move us one more past the eop_desc for start of next pkt */
2032 		tx_buffer++;
2033 		tx_desc++;
2034 		i++;
2035 		if (unlikely(!i)) {
2036 			i -= tx_ring->count;
2037 			tx_buffer = tx_ring->tx_buffer_info;
2038 			tx_desc = IGC_TX_DESC(tx_ring, 0);
2039 		}
2040 
2041 		/* issue prefetch for next Tx descriptor */
2042 		prefetch(tx_desc);
2043 
2044 		/* update budget accounting */
2045 		budget--;
2046 	} while (likely(budget));
2047 
2048 	netdev_tx_completed_queue(txring_txq(tx_ring),
2049 				  total_packets, total_bytes);
2050 
2051 	i += tx_ring->count;
2052 	tx_ring->next_to_clean = i;
2053 	u64_stats_update_begin(&tx_ring->tx_syncp);
2054 	tx_ring->tx_stats.bytes += total_bytes;
2055 	tx_ring->tx_stats.packets += total_packets;
2056 	u64_stats_update_end(&tx_ring->tx_syncp);
2057 	q_vector->tx.total_bytes += total_bytes;
2058 	q_vector->tx.total_packets += total_packets;
2059 
2060 	if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
2061 		struct igc_hw *hw = &adapter->hw;
2062 
2063 		/* Detect a transmit hang in hardware, this serializes the
2064 		 * check with the clearing of time_stamp and movement of i
2065 		 */
2066 		clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
2067 		if (tx_buffer->next_to_watch &&
2068 		    time_after(jiffies, tx_buffer->time_stamp +
2069 		    (adapter->tx_timeout_factor * HZ)) &&
2070 		    !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF)) {
2071 			/* detected Tx unit hang */
2072 			dev_err(tx_ring->dev,
2073 				"Detected Tx Unit Hang\n"
2074 				"  Tx Queue             <%d>\n"
2075 				"  TDH                  <%x>\n"
2076 				"  TDT                  <%x>\n"
2077 				"  next_to_use          <%x>\n"
2078 				"  next_to_clean        <%x>\n"
2079 				"buffer_info[next_to_clean]\n"
2080 				"  time_stamp           <%lx>\n"
2081 				"  next_to_watch        <%p>\n"
2082 				"  jiffies              <%lx>\n"
2083 				"  desc.status          <%x>\n",
2084 				tx_ring->queue_index,
2085 				rd32(IGC_TDH(tx_ring->reg_idx)),
2086 				readl(tx_ring->tail),
2087 				tx_ring->next_to_use,
2088 				tx_ring->next_to_clean,
2089 				tx_buffer->time_stamp,
2090 				tx_buffer->next_to_watch,
2091 				jiffies,
2092 				tx_buffer->next_to_watch->wb.status);
2093 			netif_stop_subqueue(tx_ring->netdev,
2094 					    tx_ring->queue_index);
2095 
2096 			/* we are about to reset, no point in enabling stuff */
2097 			return true;
2098 		}
2099 	}
2100 
2101 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
2102 	if (unlikely(total_packets &&
2103 		     netif_carrier_ok(tx_ring->netdev) &&
2104 		     igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
2105 		/* Make sure that anybody stopping the queue after this
2106 		 * sees the new next_to_clean.
2107 		 */
2108 		smp_mb();
2109 		if (__netif_subqueue_stopped(tx_ring->netdev,
2110 					     tx_ring->queue_index) &&
2111 		    !(test_bit(__IGC_DOWN, &adapter->state))) {
2112 			netif_wake_subqueue(tx_ring->netdev,
2113 					    tx_ring->queue_index);
2114 
2115 			u64_stats_update_begin(&tx_ring->tx_syncp);
2116 			tx_ring->tx_stats.restart_queue++;
2117 			u64_stats_update_end(&tx_ring->tx_syncp);
2118 		}
2119 	}
2120 
2121 	return !!budget;
2122 }
2123 
2124 static void igc_nfc_filter_restore(struct igc_adapter *adapter)
2125 {
2126 	struct igc_nfc_filter *rule;
2127 
2128 	spin_lock(&adapter->nfc_lock);
2129 
2130 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
2131 		igc_add_filter(adapter, rule);
2132 
2133 	spin_unlock(&adapter->nfc_lock);
2134 }
2135 
2136 /* If the filter to be added and an already existing filter express
2137  * the same address and address type, it should be possible to only
2138  * override the other configurations, for example the queue to steer
2139  * traffic.
2140  */
2141 static bool igc_mac_entry_can_be_used(const struct igc_mac_addr *entry,
2142 				      const u8 *addr, const u8 flags)
2143 {
2144 	if (!(entry->state & IGC_MAC_STATE_IN_USE))
2145 		return true;
2146 
2147 	if ((entry->state & IGC_MAC_STATE_SRC_ADDR) !=
2148 	    (flags & IGC_MAC_STATE_SRC_ADDR))
2149 		return false;
2150 
2151 	if (!ether_addr_equal(addr, entry->addr))
2152 		return false;
2153 
2154 	return true;
2155 }
2156 
2157 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
2158  * 'flags' is used to indicate what kind of match is made, match is by
2159  * default for the destination address, if matching by source address
2160  * is desired the flag IGC_MAC_STATE_SRC_ADDR can be used.
2161  */
2162 static int igc_add_mac_filter(struct igc_adapter *adapter,
2163 			      const u8 *addr, const u8 queue)
2164 {
2165 	struct igc_hw *hw = &adapter->hw;
2166 	int rar_entries = hw->mac.rar_entry_count;
2167 	int i;
2168 
2169 	if (is_zero_ether_addr(addr))
2170 		return -EINVAL;
2171 
2172 	/* Search for the first empty entry in the MAC table.
2173 	 * Do not touch entries at the end of the table reserved for the VF MAC
2174 	 * addresses.
2175 	 */
2176 	for (i = 0; i < rar_entries; i++) {
2177 		if (!igc_mac_entry_can_be_used(&adapter->mac_table[i],
2178 					       addr, 0))
2179 			continue;
2180 
2181 		ether_addr_copy(adapter->mac_table[i].addr, addr);
2182 		adapter->mac_table[i].queue = queue;
2183 		adapter->mac_table[i].state |= IGC_MAC_STATE_IN_USE;
2184 
2185 		igc_rar_set_index(adapter, i);
2186 		return i;
2187 	}
2188 
2189 	return -ENOSPC;
2190 }
2191 
2192 /* Remove a MAC filter for 'addr' directing matching traffic to
2193  * 'queue', 'flags' is used to indicate what kind of match need to be
2194  * removed, match is by default for the destination address, if
2195  * matching by source address is to be removed the flag
2196  * IGC_MAC_STATE_SRC_ADDR can be used.
2197  */
2198 static int igc_del_mac_filter(struct igc_adapter *adapter,
2199 			      const u8 *addr, const u8 queue)
2200 {
2201 	struct igc_hw *hw = &adapter->hw;
2202 	int rar_entries = hw->mac.rar_entry_count;
2203 	int i;
2204 
2205 	if (is_zero_ether_addr(addr))
2206 		return -EINVAL;
2207 
2208 	/* Search for matching entry in the MAC table based on given address
2209 	 * and queue. Do not touch entries at the end of the table reserved
2210 	 * for the VF MAC addresses.
2211 	 */
2212 	for (i = 0; i < rar_entries; i++) {
2213 		if (!(adapter->mac_table[i].state & IGC_MAC_STATE_IN_USE))
2214 			continue;
2215 		if (adapter->mac_table[i].state != 0)
2216 			continue;
2217 		if (adapter->mac_table[i].queue != queue)
2218 			continue;
2219 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
2220 			continue;
2221 
2222 		/* When a filter for the default address is "deleted",
2223 		 * we return it to its initial configuration
2224 		 */
2225 		if (adapter->mac_table[i].state & IGC_MAC_STATE_DEFAULT) {
2226 			adapter->mac_table[i].state =
2227 				IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
2228 			adapter->mac_table[i].queue = 0;
2229 		} else {
2230 			adapter->mac_table[i].state = 0;
2231 			adapter->mac_table[i].queue = 0;
2232 			memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
2233 		}
2234 
2235 		igc_rar_set_index(adapter, i);
2236 		return 0;
2237 	}
2238 
2239 	return -ENOENT;
2240 }
2241 
2242 static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr)
2243 {
2244 	struct igc_adapter *adapter = netdev_priv(netdev);
2245 	int ret;
2246 
2247 	ret = igc_add_mac_filter(adapter, addr, adapter->num_rx_queues);
2248 
2249 	return min_t(int, ret, 0);
2250 }
2251 
2252 static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr)
2253 {
2254 	struct igc_adapter *adapter = netdev_priv(netdev);
2255 
2256 	igc_del_mac_filter(adapter, addr, adapter->num_rx_queues);
2257 
2258 	return 0;
2259 }
2260 
2261 /**
2262  * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2263  * @netdev: network interface device structure
2264  *
2265  * The set_rx_mode entry point is called whenever the unicast or multicast
2266  * address lists or the network interface flags are updated.  This routine is
2267  * responsible for configuring the hardware for proper unicast, multicast,
2268  * promiscuous mode, and all-multi behavior.
2269  */
2270 static void igc_set_rx_mode(struct net_device *netdev)
2271 {
2272 	struct igc_adapter *adapter = netdev_priv(netdev);
2273 	struct igc_hw *hw = &adapter->hw;
2274 	u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
2275 	int count;
2276 
2277 	/* Check for Promiscuous and All Multicast modes */
2278 	if (netdev->flags & IFF_PROMISC) {
2279 		rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE;
2280 	} else {
2281 		if (netdev->flags & IFF_ALLMULTI) {
2282 			rctl |= IGC_RCTL_MPE;
2283 		} else {
2284 			/* Write addresses to the MTA, if the attempt fails
2285 			 * then we should just turn on promiscuous mode so
2286 			 * that we can at least receive multicast traffic
2287 			 */
2288 			count = igc_write_mc_addr_list(netdev);
2289 			if (count < 0)
2290 				rctl |= IGC_RCTL_MPE;
2291 		}
2292 	}
2293 
2294 	/* Write addresses to available RAR registers, if there is not
2295 	 * sufficient space to store all the addresses then enable
2296 	 * unicast promiscuous mode
2297 	 */
2298 	if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync))
2299 		rctl |= IGC_RCTL_UPE;
2300 
2301 	/* update state of unicast and multicast */
2302 	rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
2303 	wr32(IGC_RCTL, rctl);
2304 
2305 #if (PAGE_SIZE < 8192)
2306 	if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB)
2307 		rlpml = IGC_MAX_FRAME_BUILD_SKB;
2308 #endif
2309 	wr32(IGC_RLPML, rlpml);
2310 }
2311 
2312 /**
2313  * igc_configure - configure the hardware for RX and TX
2314  * @adapter: private board structure
2315  */
2316 static void igc_configure(struct igc_adapter *adapter)
2317 {
2318 	struct net_device *netdev = adapter->netdev;
2319 	int i = 0;
2320 
2321 	igc_get_hw_control(adapter);
2322 	igc_set_rx_mode(netdev);
2323 
2324 	igc_setup_tctl(adapter);
2325 	igc_setup_mrqc(adapter);
2326 	igc_setup_rctl(adapter);
2327 
2328 	igc_nfc_filter_restore(adapter);
2329 	igc_configure_tx(adapter);
2330 	igc_configure_rx(adapter);
2331 
2332 	igc_rx_fifo_flush_base(&adapter->hw);
2333 
2334 	/* call igc_desc_unused which always leaves
2335 	 * at least 1 descriptor unused to make sure
2336 	 * next_to_use != next_to_clean
2337 	 */
2338 	for (i = 0; i < adapter->num_rx_queues; i++) {
2339 		struct igc_ring *ring = adapter->rx_ring[i];
2340 
2341 		igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
2342 	}
2343 }
2344 
2345 /**
2346  * igc_write_ivar - configure ivar for given MSI-X vector
2347  * @hw: pointer to the HW structure
2348  * @msix_vector: vector number we are allocating to a given ring
2349  * @index: row index of IVAR register to write within IVAR table
2350  * @offset: column offset of in IVAR, should be multiple of 8
2351  *
2352  * The IVAR table consists of 2 columns,
2353  * each containing an cause allocation for an Rx and Tx ring, and a
2354  * variable number of rows depending on the number of queues supported.
2355  */
2356 static void igc_write_ivar(struct igc_hw *hw, int msix_vector,
2357 			   int index, int offset)
2358 {
2359 	u32 ivar = array_rd32(IGC_IVAR0, index);
2360 
2361 	/* clear any bits that are currently set */
2362 	ivar &= ~((u32)0xFF << offset);
2363 
2364 	/* write vector and valid bit */
2365 	ivar |= (msix_vector | IGC_IVAR_VALID) << offset;
2366 
2367 	array_wr32(IGC_IVAR0, index, ivar);
2368 }
2369 
2370 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector)
2371 {
2372 	struct igc_adapter *adapter = q_vector->adapter;
2373 	struct igc_hw *hw = &adapter->hw;
2374 	int rx_queue = IGC_N0_QUEUE;
2375 	int tx_queue = IGC_N0_QUEUE;
2376 
2377 	if (q_vector->rx.ring)
2378 		rx_queue = q_vector->rx.ring->reg_idx;
2379 	if (q_vector->tx.ring)
2380 		tx_queue = q_vector->tx.ring->reg_idx;
2381 
2382 	switch (hw->mac.type) {
2383 	case igc_i225:
2384 		if (rx_queue > IGC_N0_QUEUE)
2385 			igc_write_ivar(hw, msix_vector,
2386 				       rx_queue >> 1,
2387 				       (rx_queue & 0x1) << 4);
2388 		if (tx_queue > IGC_N0_QUEUE)
2389 			igc_write_ivar(hw, msix_vector,
2390 				       tx_queue >> 1,
2391 				       ((tx_queue & 0x1) << 4) + 8);
2392 		q_vector->eims_value = BIT(msix_vector);
2393 		break;
2394 	default:
2395 		WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n");
2396 		break;
2397 	}
2398 
2399 	/* add q_vector eims value to global eims_enable_mask */
2400 	adapter->eims_enable_mask |= q_vector->eims_value;
2401 
2402 	/* configure q_vector to set itr on first interrupt */
2403 	q_vector->set_itr = 1;
2404 }
2405 
2406 /**
2407  * igc_configure_msix - Configure MSI-X hardware
2408  * @adapter: Pointer to adapter structure
2409  *
2410  * igc_configure_msix sets up the hardware to properly
2411  * generate MSI-X interrupts.
2412  */
2413 static void igc_configure_msix(struct igc_adapter *adapter)
2414 {
2415 	struct igc_hw *hw = &adapter->hw;
2416 	int i, vector = 0;
2417 	u32 tmp;
2418 
2419 	adapter->eims_enable_mask = 0;
2420 
2421 	/* set vector for other causes, i.e. link changes */
2422 	switch (hw->mac.type) {
2423 	case igc_i225:
2424 		/* Turn on MSI-X capability first, or our settings
2425 		 * won't stick.  And it will take days to debug.
2426 		 */
2427 		wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE |
2428 		     IGC_GPIE_PBA | IGC_GPIE_EIAME |
2429 		     IGC_GPIE_NSICR);
2430 
2431 		/* enable msix_other interrupt */
2432 		adapter->eims_other = BIT(vector);
2433 		tmp = (vector++ | IGC_IVAR_VALID) << 8;
2434 
2435 		wr32(IGC_IVAR_MISC, tmp);
2436 		break;
2437 	default:
2438 		/* do nothing, since nothing else supports MSI-X */
2439 		break;
2440 	} /* switch (hw->mac.type) */
2441 
2442 	adapter->eims_enable_mask |= adapter->eims_other;
2443 
2444 	for (i = 0; i < adapter->num_q_vectors; i++)
2445 		igc_assign_vector(adapter->q_vector[i], vector++);
2446 
2447 	wrfl();
2448 }
2449 
2450 /**
2451  * igc_irq_enable - Enable default interrupt generation settings
2452  * @adapter: board private structure
2453  */
2454 static void igc_irq_enable(struct igc_adapter *adapter)
2455 {
2456 	struct igc_hw *hw = &adapter->hw;
2457 
2458 	if (adapter->msix_entries) {
2459 		u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA;
2460 		u32 regval = rd32(IGC_EIAC);
2461 
2462 		wr32(IGC_EIAC, regval | adapter->eims_enable_mask);
2463 		regval = rd32(IGC_EIAM);
2464 		wr32(IGC_EIAM, regval | adapter->eims_enable_mask);
2465 		wr32(IGC_EIMS, adapter->eims_enable_mask);
2466 		wr32(IGC_IMS, ims);
2467 	} else {
2468 		wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
2469 		wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
2470 	}
2471 }
2472 
2473 /**
2474  * igc_irq_disable - Mask off interrupt generation on the NIC
2475  * @adapter: board private structure
2476  */
2477 static void igc_irq_disable(struct igc_adapter *adapter)
2478 {
2479 	struct igc_hw *hw = &adapter->hw;
2480 
2481 	if (adapter->msix_entries) {
2482 		u32 regval = rd32(IGC_EIAM);
2483 
2484 		wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask);
2485 		wr32(IGC_EIMC, adapter->eims_enable_mask);
2486 		regval = rd32(IGC_EIAC);
2487 		wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask);
2488 	}
2489 
2490 	wr32(IGC_IAM, 0);
2491 	wr32(IGC_IMC, ~0);
2492 	wrfl();
2493 
2494 	if (adapter->msix_entries) {
2495 		int vector = 0, i;
2496 
2497 		synchronize_irq(adapter->msix_entries[vector++].vector);
2498 
2499 		for (i = 0; i < adapter->num_q_vectors; i++)
2500 			synchronize_irq(adapter->msix_entries[vector++].vector);
2501 	} else {
2502 		synchronize_irq(adapter->pdev->irq);
2503 	}
2504 }
2505 
2506 void igc_set_flag_queue_pairs(struct igc_adapter *adapter,
2507 			      const u32 max_rss_queues)
2508 {
2509 	/* Determine if we need to pair queues. */
2510 	/* If rss_queues > half of max_rss_queues, pair the queues in
2511 	 * order to conserve interrupts due to limited supply.
2512 	 */
2513 	if (adapter->rss_queues > (max_rss_queues / 2))
2514 		adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
2515 	else
2516 		adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS;
2517 }
2518 
2519 unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter)
2520 {
2521 	unsigned int max_rss_queues;
2522 
2523 	/* Determine the maximum number of RSS queues supported. */
2524 	max_rss_queues = IGC_MAX_RX_QUEUES;
2525 
2526 	return max_rss_queues;
2527 }
2528 
2529 static void igc_init_queue_configuration(struct igc_adapter *adapter)
2530 {
2531 	u32 max_rss_queues;
2532 
2533 	max_rss_queues = igc_get_max_rss_queues(adapter);
2534 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2535 
2536 	igc_set_flag_queue_pairs(adapter, max_rss_queues);
2537 }
2538 
2539 /**
2540  * igc_reset_q_vector - Reset config for interrupt vector
2541  * @adapter: board private structure to initialize
2542  * @v_idx: Index of vector to be reset
2543  *
2544  * If NAPI is enabled it will delete any references to the
2545  * NAPI struct. This is preparation for igc_free_q_vector.
2546  */
2547 static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx)
2548 {
2549 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
2550 
2551 	/* if we're coming from igc_set_interrupt_capability, the vectors are
2552 	 * not yet allocated
2553 	 */
2554 	if (!q_vector)
2555 		return;
2556 
2557 	if (q_vector->tx.ring)
2558 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
2559 
2560 	if (q_vector->rx.ring)
2561 		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
2562 
2563 	netif_napi_del(&q_vector->napi);
2564 }
2565 
2566 /**
2567  * igc_free_q_vector - Free memory allocated for specific interrupt vector
2568  * @adapter: board private structure to initialize
2569  * @v_idx: Index of vector to be freed
2570  *
2571  * This function frees the memory allocated to the q_vector.
2572  */
2573 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx)
2574 {
2575 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
2576 
2577 	adapter->q_vector[v_idx] = NULL;
2578 
2579 	/* igc_get_stats64() might access the rings on this vector,
2580 	 * we must wait a grace period before freeing it.
2581 	 */
2582 	if (q_vector)
2583 		kfree_rcu(q_vector, rcu);
2584 }
2585 
2586 /**
2587  * igc_free_q_vectors - Free memory allocated for interrupt vectors
2588  * @adapter: board private structure to initialize
2589  *
2590  * This function frees the memory allocated to the q_vectors.  In addition if
2591  * NAPI is enabled it will delete any references to the NAPI struct prior
2592  * to freeing the q_vector.
2593  */
2594 static void igc_free_q_vectors(struct igc_adapter *adapter)
2595 {
2596 	int v_idx = adapter->num_q_vectors;
2597 
2598 	adapter->num_tx_queues = 0;
2599 	adapter->num_rx_queues = 0;
2600 	adapter->num_q_vectors = 0;
2601 
2602 	while (v_idx--) {
2603 		igc_reset_q_vector(adapter, v_idx);
2604 		igc_free_q_vector(adapter, v_idx);
2605 	}
2606 }
2607 
2608 /**
2609  * igc_update_itr - update the dynamic ITR value based on statistics
2610  * @q_vector: pointer to q_vector
2611  * @ring_container: ring info to update the itr for
2612  *
2613  * Stores a new ITR value based on packets and byte
2614  * counts during the last interrupt.  The advantage of per interrupt
2615  * computation is faster updates and more accurate ITR for the current
2616  * traffic pattern.  Constants in this function were computed
2617  * based on theoretical maximum wire speed and thresholds were set based
2618  * on testing data as well as attempting to minimize response time
2619  * while increasing bulk throughput.
2620  * NOTE: These calculations are only valid when operating in a single-
2621  * queue environment.
2622  */
2623 static void igc_update_itr(struct igc_q_vector *q_vector,
2624 			   struct igc_ring_container *ring_container)
2625 {
2626 	unsigned int packets = ring_container->total_packets;
2627 	unsigned int bytes = ring_container->total_bytes;
2628 	u8 itrval = ring_container->itr;
2629 
2630 	/* no packets, exit with status unchanged */
2631 	if (packets == 0)
2632 		return;
2633 
2634 	switch (itrval) {
2635 	case lowest_latency:
2636 		/* handle TSO and jumbo frames */
2637 		if (bytes / packets > 8000)
2638 			itrval = bulk_latency;
2639 		else if ((packets < 5) && (bytes > 512))
2640 			itrval = low_latency;
2641 		break;
2642 	case low_latency:  /* 50 usec aka 20000 ints/s */
2643 		if (bytes > 10000) {
2644 			/* this if handles the TSO accounting */
2645 			if (bytes / packets > 8000)
2646 				itrval = bulk_latency;
2647 			else if ((packets < 10) || ((bytes / packets) > 1200))
2648 				itrval = bulk_latency;
2649 			else if ((packets > 35))
2650 				itrval = lowest_latency;
2651 		} else if (bytes / packets > 2000) {
2652 			itrval = bulk_latency;
2653 		} else if (packets <= 2 && bytes < 512) {
2654 			itrval = lowest_latency;
2655 		}
2656 		break;
2657 	case bulk_latency: /* 250 usec aka 4000 ints/s */
2658 		if (bytes > 25000) {
2659 			if (packets > 35)
2660 				itrval = low_latency;
2661 		} else if (bytes < 1500) {
2662 			itrval = low_latency;
2663 		}
2664 		break;
2665 	}
2666 
2667 	/* clear work counters since we have the values we need */
2668 	ring_container->total_bytes = 0;
2669 	ring_container->total_packets = 0;
2670 
2671 	/* write updated itr to ring container */
2672 	ring_container->itr = itrval;
2673 }
2674 
2675 static void igc_set_itr(struct igc_q_vector *q_vector)
2676 {
2677 	struct igc_adapter *adapter = q_vector->adapter;
2678 	u32 new_itr = q_vector->itr_val;
2679 	u8 current_itr = 0;
2680 
2681 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2682 	switch (adapter->link_speed) {
2683 	case SPEED_10:
2684 	case SPEED_100:
2685 		current_itr = 0;
2686 		new_itr = IGC_4K_ITR;
2687 		goto set_itr_now;
2688 	default:
2689 		break;
2690 	}
2691 
2692 	igc_update_itr(q_vector, &q_vector->tx);
2693 	igc_update_itr(q_vector, &q_vector->rx);
2694 
2695 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2696 
2697 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2698 	if (current_itr == lowest_latency &&
2699 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
2700 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
2701 		current_itr = low_latency;
2702 
2703 	switch (current_itr) {
2704 	/* counts and packets in update_itr are dependent on these numbers */
2705 	case lowest_latency:
2706 		new_itr = IGC_70K_ITR; /* 70,000 ints/sec */
2707 		break;
2708 	case low_latency:
2709 		new_itr = IGC_20K_ITR; /* 20,000 ints/sec */
2710 		break;
2711 	case bulk_latency:
2712 		new_itr = IGC_4K_ITR;  /* 4,000 ints/sec */
2713 		break;
2714 	default:
2715 		break;
2716 	}
2717 
2718 set_itr_now:
2719 	if (new_itr != q_vector->itr_val) {
2720 		/* this attempts to bias the interrupt rate towards Bulk
2721 		 * by adding intermediate steps when interrupt rate is
2722 		 * increasing
2723 		 */
2724 		new_itr = new_itr > q_vector->itr_val ?
2725 			  max((new_itr * q_vector->itr_val) /
2726 			  (new_itr + (q_vector->itr_val >> 2)),
2727 			  new_itr) : new_itr;
2728 		/* Don't write the value here; it resets the adapter's
2729 		 * internal timer, and causes us to delay far longer than
2730 		 * we should between interrupts.  Instead, we write the ITR
2731 		 * value at the beginning of the next interrupt so the timing
2732 		 * ends up being correct.
2733 		 */
2734 		q_vector->itr_val = new_itr;
2735 		q_vector->set_itr = 1;
2736 	}
2737 }
2738 
2739 static void igc_reset_interrupt_capability(struct igc_adapter *adapter)
2740 {
2741 	int v_idx = adapter->num_q_vectors;
2742 
2743 	if (adapter->msix_entries) {
2744 		pci_disable_msix(adapter->pdev);
2745 		kfree(adapter->msix_entries);
2746 		adapter->msix_entries = NULL;
2747 	} else if (adapter->flags & IGC_FLAG_HAS_MSI) {
2748 		pci_disable_msi(adapter->pdev);
2749 	}
2750 
2751 	while (v_idx--)
2752 		igc_reset_q_vector(adapter, v_idx);
2753 }
2754 
2755 /**
2756  * igc_set_interrupt_capability - set MSI or MSI-X if supported
2757  * @adapter: Pointer to adapter structure
2758  * @msix: boolean value for MSI-X capability
2759  *
2760  * Attempt to configure interrupts using the best available
2761  * capabilities of the hardware and kernel.
2762  */
2763 static void igc_set_interrupt_capability(struct igc_adapter *adapter,
2764 					 bool msix)
2765 {
2766 	int numvecs, i;
2767 	int err;
2768 
2769 	if (!msix)
2770 		goto msi_only;
2771 	adapter->flags |= IGC_FLAG_HAS_MSIX;
2772 
2773 	/* Number of supported queues. */
2774 	adapter->num_rx_queues = adapter->rss_queues;
2775 
2776 	adapter->num_tx_queues = adapter->rss_queues;
2777 
2778 	/* start with one vector for every Rx queue */
2779 	numvecs = adapter->num_rx_queues;
2780 
2781 	/* if Tx handler is separate add 1 for every Tx queue */
2782 	if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS))
2783 		numvecs += adapter->num_tx_queues;
2784 
2785 	/* store the number of vectors reserved for queues */
2786 	adapter->num_q_vectors = numvecs;
2787 
2788 	/* add 1 vector for link status interrupts */
2789 	numvecs++;
2790 
2791 	adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
2792 					GFP_KERNEL);
2793 
2794 	if (!adapter->msix_entries)
2795 		return;
2796 
2797 	/* populate entry values */
2798 	for (i = 0; i < numvecs; i++)
2799 		adapter->msix_entries[i].entry = i;
2800 
2801 	err = pci_enable_msix_range(adapter->pdev,
2802 				    adapter->msix_entries,
2803 				    numvecs,
2804 				    numvecs);
2805 	if (err > 0)
2806 		return;
2807 
2808 	kfree(adapter->msix_entries);
2809 	adapter->msix_entries = NULL;
2810 
2811 	igc_reset_interrupt_capability(adapter);
2812 
2813 msi_only:
2814 	adapter->flags &= ~IGC_FLAG_HAS_MSIX;
2815 
2816 	adapter->rss_queues = 1;
2817 	adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
2818 	adapter->num_rx_queues = 1;
2819 	adapter->num_tx_queues = 1;
2820 	adapter->num_q_vectors = 1;
2821 	if (!pci_enable_msi(adapter->pdev))
2822 		adapter->flags |= IGC_FLAG_HAS_MSI;
2823 }
2824 
2825 /**
2826  * igc_update_ring_itr - update the dynamic ITR value based on packet size
2827  * @q_vector: pointer to q_vector
2828  *
2829  * Stores a new ITR value based on strictly on packet size.  This
2830  * algorithm is less sophisticated than that used in igc_update_itr,
2831  * due to the difficulty of synchronizing statistics across multiple
2832  * receive rings.  The divisors and thresholds used by this function
2833  * were determined based on theoretical maximum wire speed and testing
2834  * data, in order to minimize response time while increasing bulk
2835  * throughput.
2836  * NOTE: This function is called only when operating in a multiqueue
2837  * receive environment.
2838  */
2839 static void igc_update_ring_itr(struct igc_q_vector *q_vector)
2840 {
2841 	struct igc_adapter *adapter = q_vector->adapter;
2842 	int new_val = q_vector->itr_val;
2843 	int avg_wire_size = 0;
2844 	unsigned int packets;
2845 
2846 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
2847 	 * ints/sec - ITR timer value of 120 ticks.
2848 	 */
2849 	switch (adapter->link_speed) {
2850 	case SPEED_10:
2851 	case SPEED_100:
2852 		new_val = IGC_4K_ITR;
2853 		goto set_itr_val;
2854 	default:
2855 		break;
2856 	}
2857 
2858 	packets = q_vector->rx.total_packets;
2859 	if (packets)
2860 		avg_wire_size = q_vector->rx.total_bytes / packets;
2861 
2862 	packets = q_vector->tx.total_packets;
2863 	if (packets)
2864 		avg_wire_size = max_t(u32, avg_wire_size,
2865 				      q_vector->tx.total_bytes / packets);
2866 
2867 	/* if avg_wire_size isn't set no work was done */
2868 	if (!avg_wire_size)
2869 		goto clear_counts;
2870 
2871 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
2872 	avg_wire_size += 24;
2873 
2874 	/* Don't starve jumbo frames */
2875 	avg_wire_size = min(avg_wire_size, 3000);
2876 
2877 	/* Give a little boost to mid-size frames */
2878 	if (avg_wire_size > 300 && avg_wire_size < 1200)
2879 		new_val = avg_wire_size / 3;
2880 	else
2881 		new_val = avg_wire_size / 2;
2882 
2883 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
2884 	if (new_val < IGC_20K_ITR &&
2885 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
2886 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
2887 		new_val = IGC_20K_ITR;
2888 
2889 set_itr_val:
2890 	if (new_val != q_vector->itr_val) {
2891 		q_vector->itr_val = new_val;
2892 		q_vector->set_itr = 1;
2893 	}
2894 clear_counts:
2895 	q_vector->rx.total_bytes = 0;
2896 	q_vector->rx.total_packets = 0;
2897 	q_vector->tx.total_bytes = 0;
2898 	q_vector->tx.total_packets = 0;
2899 }
2900 
2901 static void igc_ring_irq_enable(struct igc_q_vector *q_vector)
2902 {
2903 	struct igc_adapter *adapter = q_vector->adapter;
2904 	struct igc_hw *hw = &adapter->hw;
2905 
2906 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
2907 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
2908 		if (adapter->num_q_vectors == 1)
2909 			igc_set_itr(q_vector);
2910 		else
2911 			igc_update_ring_itr(q_vector);
2912 	}
2913 
2914 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
2915 		if (adapter->msix_entries)
2916 			wr32(IGC_EIMS, q_vector->eims_value);
2917 		else
2918 			igc_irq_enable(adapter);
2919 	}
2920 }
2921 
2922 static void igc_add_ring(struct igc_ring *ring,
2923 			 struct igc_ring_container *head)
2924 {
2925 	head->ring = ring;
2926 	head->count++;
2927 }
2928 
2929 /**
2930  * igc_cache_ring_register - Descriptor ring to register mapping
2931  * @adapter: board private structure to initialize
2932  *
2933  * Once we know the feature-set enabled for the device, we'll cache
2934  * the register offset the descriptor ring is assigned to.
2935  */
2936 static void igc_cache_ring_register(struct igc_adapter *adapter)
2937 {
2938 	int i = 0, j = 0;
2939 
2940 	switch (adapter->hw.mac.type) {
2941 	case igc_i225:
2942 	/* Fall through */
2943 	default:
2944 		for (; i < adapter->num_rx_queues; i++)
2945 			adapter->rx_ring[i]->reg_idx = i;
2946 		for (; j < adapter->num_tx_queues; j++)
2947 			adapter->tx_ring[j]->reg_idx = j;
2948 		break;
2949 	}
2950 }
2951 
2952 /**
2953  * igc_poll - NAPI Rx polling callback
2954  * @napi: napi polling structure
2955  * @budget: count of how many packets we should handle
2956  */
2957 static int igc_poll(struct napi_struct *napi, int budget)
2958 {
2959 	struct igc_q_vector *q_vector = container_of(napi,
2960 						     struct igc_q_vector,
2961 						     napi);
2962 	bool clean_complete = true;
2963 	int work_done = 0;
2964 
2965 	if (q_vector->tx.ring)
2966 		clean_complete = igc_clean_tx_irq(q_vector, budget);
2967 
2968 	if (q_vector->rx.ring) {
2969 		int cleaned = igc_clean_rx_irq(q_vector, budget);
2970 
2971 		work_done += cleaned;
2972 		if (cleaned >= budget)
2973 			clean_complete = false;
2974 	}
2975 
2976 	/* If all work not completed, return budget and keep polling */
2977 	if (!clean_complete)
2978 		return budget;
2979 
2980 	/* Exit the polling mode, but don't re-enable interrupts if stack might
2981 	 * poll us due to busy-polling
2982 	 */
2983 	if (likely(napi_complete_done(napi, work_done)))
2984 		igc_ring_irq_enable(q_vector);
2985 
2986 	return min(work_done, budget - 1);
2987 }
2988 
2989 /**
2990  * igc_alloc_q_vector - Allocate memory for a single interrupt vector
2991  * @adapter: board private structure to initialize
2992  * @v_count: q_vectors allocated on adapter, used for ring interleaving
2993  * @v_idx: index of vector in adapter struct
2994  * @txr_count: total number of Tx rings to allocate
2995  * @txr_idx: index of first Tx ring to allocate
2996  * @rxr_count: total number of Rx rings to allocate
2997  * @rxr_idx: index of first Rx ring to allocate
2998  *
2999  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
3000  */
3001 static int igc_alloc_q_vector(struct igc_adapter *adapter,
3002 			      unsigned int v_count, unsigned int v_idx,
3003 			      unsigned int txr_count, unsigned int txr_idx,
3004 			      unsigned int rxr_count, unsigned int rxr_idx)
3005 {
3006 	struct igc_q_vector *q_vector;
3007 	struct igc_ring *ring;
3008 	int ring_count;
3009 
3010 	/* igc only supports 1 Tx and/or 1 Rx queue per vector */
3011 	if (txr_count > 1 || rxr_count > 1)
3012 		return -ENOMEM;
3013 
3014 	ring_count = txr_count + rxr_count;
3015 
3016 	/* allocate q_vector and rings */
3017 	q_vector = adapter->q_vector[v_idx];
3018 	if (!q_vector)
3019 		q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
3020 				   GFP_KERNEL);
3021 	else
3022 		memset(q_vector, 0, struct_size(q_vector, ring, ring_count));
3023 	if (!q_vector)
3024 		return -ENOMEM;
3025 
3026 	/* initialize NAPI */
3027 	netif_napi_add(adapter->netdev, &q_vector->napi,
3028 		       igc_poll, 64);
3029 
3030 	/* tie q_vector and adapter together */
3031 	adapter->q_vector[v_idx] = q_vector;
3032 	q_vector->adapter = adapter;
3033 
3034 	/* initialize work limits */
3035 	q_vector->tx.work_limit = adapter->tx_work_limit;
3036 
3037 	/* initialize ITR configuration */
3038 	q_vector->itr_register = adapter->io_addr + IGC_EITR(0);
3039 	q_vector->itr_val = IGC_START_ITR;
3040 
3041 	/* initialize pointer to rings */
3042 	ring = q_vector->ring;
3043 
3044 	/* initialize ITR */
3045 	if (rxr_count) {
3046 		/* rx or rx/tx vector */
3047 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
3048 			q_vector->itr_val = adapter->rx_itr_setting;
3049 	} else {
3050 		/* tx only vector */
3051 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
3052 			q_vector->itr_val = adapter->tx_itr_setting;
3053 	}
3054 
3055 	if (txr_count) {
3056 		/* assign generic ring traits */
3057 		ring->dev = &adapter->pdev->dev;
3058 		ring->netdev = adapter->netdev;
3059 
3060 		/* configure backlink on ring */
3061 		ring->q_vector = q_vector;
3062 
3063 		/* update q_vector Tx values */
3064 		igc_add_ring(ring, &q_vector->tx);
3065 
3066 		/* apply Tx specific ring traits */
3067 		ring->count = adapter->tx_ring_count;
3068 		ring->queue_index = txr_idx;
3069 
3070 		/* assign ring to adapter */
3071 		adapter->tx_ring[txr_idx] = ring;
3072 
3073 		/* push pointer to next ring */
3074 		ring++;
3075 	}
3076 
3077 	if (rxr_count) {
3078 		/* assign generic ring traits */
3079 		ring->dev = &adapter->pdev->dev;
3080 		ring->netdev = adapter->netdev;
3081 
3082 		/* configure backlink on ring */
3083 		ring->q_vector = q_vector;
3084 
3085 		/* update q_vector Rx values */
3086 		igc_add_ring(ring, &q_vector->rx);
3087 
3088 		/* apply Rx specific ring traits */
3089 		ring->count = adapter->rx_ring_count;
3090 		ring->queue_index = rxr_idx;
3091 
3092 		/* assign ring to adapter */
3093 		adapter->rx_ring[rxr_idx] = ring;
3094 	}
3095 
3096 	return 0;
3097 }
3098 
3099 /**
3100  * igc_alloc_q_vectors - Allocate memory for interrupt vectors
3101  * @adapter: board private structure to initialize
3102  *
3103  * We allocate one q_vector per queue interrupt.  If allocation fails we
3104  * return -ENOMEM.
3105  */
3106 static int igc_alloc_q_vectors(struct igc_adapter *adapter)
3107 {
3108 	int rxr_remaining = adapter->num_rx_queues;
3109 	int txr_remaining = adapter->num_tx_queues;
3110 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
3111 	int q_vectors = adapter->num_q_vectors;
3112 	int err;
3113 
3114 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
3115 		for (; rxr_remaining; v_idx++) {
3116 			err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
3117 						 0, 0, 1, rxr_idx);
3118 
3119 			if (err)
3120 				goto err_out;
3121 
3122 			/* update counts and index */
3123 			rxr_remaining--;
3124 			rxr_idx++;
3125 		}
3126 	}
3127 
3128 	for (; v_idx < q_vectors; v_idx++) {
3129 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
3130 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
3131 
3132 		err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
3133 					 tqpv, txr_idx, rqpv, rxr_idx);
3134 
3135 		if (err)
3136 			goto err_out;
3137 
3138 		/* update counts and index */
3139 		rxr_remaining -= rqpv;
3140 		txr_remaining -= tqpv;
3141 		rxr_idx++;
3142 		txr_idx++;
3143 	}
3144 
3145 	return 0;
3146 
3147 err_out:
3148 	adapter->num_tx_queues = 0;
3149 	adapter->num_rx_queues = 0;
3150 	adapter->num_q_vectors = 0;
3151 
3152 	while (v_idx--)
3153 		igc_free_q_vector(adapter, v_idx);
3154 
3155 	return -ENOMEM;
3156 }
3157 
3158 /**
3159  * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
3160  * @adapter: Pointer to adapter structure
3161  * @msix: boolean for MSI-X capability
3162  *
3163  * This function initializes the interrupts and allocates all of the queues.
3164  */
3165 static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix)
3166 {
3167 	struct pci_dev *pdev = adapter->pdev;
3168 	int err = 0;
3169 
3170 	igc_set_interrupt_capability(adapter, msix);
3171 
3172 	err = igc_alloc_q_vectors(adapter);
3173 	if (err) {
3174 		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
3175 		goto err_alloc_q_vectors;
3176 	}
3177 
3178 	igc_cache_ring_register(adapter);
3179 
3180 	return 0;
3181 
3182 err_alloc_q_vectors:
3183 	igc_reset_interrupt_capability(adapter);
3184 	return err;
3185 }
3186 
3187 /**
3188  * igc_sw_init - Initialize general software structures (struct igc_adapter)
3189  * @adapter: board private structure to initialize
3190  *
3191  * igc_sw_init initializes the Adapter private data structure.
3192  * Fields are initialized based on PCI device information and
3193  * OS network device settings (MTU size).
3194  */
3195 static int igc_sw_init(struct igc_adapter *adapter)
3196 {
3197 	struct net_device *netdev = adapter->netdev;
3198 	struct pci_dev *pdev = adapter->pdev;
3199 	struct igc_hw *hw = &adapter->hw;
3200 
3201 	int size = sizeof(struct igc_mac_addr) * hw->mac.rar_entry_count;
3202 
3203 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3204 
3205 	/* set default ring sizes */
3206 	adapter->tx_ring_count = IGC_DEFAULT_TXD;
3207 	adapter->rx_ring_count = IGC_DEFAULT_RXD;
3208 
3209 	/* set default ITR values */
3210 	adapter->rx_itr_setting = IGC_DEFAULT_ITR;
3211 	adapter->tx_itr_setting = IGC_DEFAULT_ITR;
3212 
3213 	/* set default work limits */
3214 	adapter->tx_work_limit = IGC_DEFAULT_TX_WORK;
3215 
3216 	/* adjust max frame to be at least the size of a standard frame */
3217 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3218 				VLAN_HLEN;
3219 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3220 
3221 	spin_lock_init(&adapter->nfc_lock);
3222 	spin_lock_init(&adapter->stats64_lock);
3223 	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
3224 	adapter->flags |= IGC_FLAG_HAS_MSIX;
3225 
3226 	adapter->mac_table = kzalloc(size, GFP_ATOMIC);
3227 	if (!adapter->mac_table)
3228 		return -ENOMEM;
3229 
3230 	igc_init_queue_configuration(adapter);
3231 
3232 	/* This call may decrease the number of queues */
3233 	if (igc_init_interrupt_scheme(adapter, true)) {
3234 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3235 		return -ENOMEM;
3236 	}
3237 
3238 	/* Explicitly disable IRQ since the NIC can be in any state. */
3239 	igc_irq_disable(adapter);
3240 
3241 	set_bit(__IGC_DOWN, &adapter->state);
3242 
3243 	return 0;
3244 }
3245 
3246 /**
3247  * igc_up - Open the interface and prepare it to handle traffic
3248  * @adapter: board private structure
3249  */
3250 void igc_up(struct igc_adapter *adapter)
3251 {
3252 	struct igc_hw *hw = &adapter->hw;
3253 	int i = 0;
3254 
3255 	/* hardware has been reset, we need to reload some things */
3256 	igc_configure(adapter);
3257 
3258 	clear_bit(__IGC_DOWN, &adapter->state);
3259 
3260 	for (i = 0; i < adapter->num_q_vectors; i++)
3261 		napi_enable(&adapter->q_vector[i]->napi);
3262 
3263 	if (adapter->msix_entries)
3264 		igc_configure_msix(adapter);
3265 	else
3266 		igc_assign_vector(adapter->q_vector[0], 0);
3267 
3268 	/* Clear any pending interrupts. */
3269 	rd32(IGC_ICR);
3270 	igc_irq_enable(adapter);
3271 
3272 	netif_tx_start_all_queues(adapter->netdev);
3273 
3274 	/* start the watchdog. */
3275 	hw->mac.get_link_status = 1;
3276 	schedule_work(&adapter->watchdog_task);
3277 }
3278 
3279 /**
3280  * igc_update_stats - Update the board statistics counters
3281  * @adapter: board private structure
3282  */
3283 void igc_update_stats(struct igc_adapter *adapter)
3284 {
3285 	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
3286 	struct pci_dev *pdev = adapter->pdev;
3287 	struct igc_hw *hw = &adapter->hw;
3288 	u64 _bytes, _packets;
3289 	u64 bytes, packets;
3290 	unsigned int start;
3291 	u32 mpc;
3292 	int i;
3293 
3294 	/* Prevent stats update while adapter is being reset, or if the pci
3295 	 * connection is down.
3296 	 */
3297 	if (adapter->link_speed == 0)
3298 		return;
3299 	if (pci_channel_offline(pdev))
3300 		return;
3301 
3302 	packets = 0;
3303 	bytes = 0;
3304 
3305 	rcu_read_lock();
3306 	for (i = 0; i < adapter->num_rx_queues; i++) {
3307 		struct igc_ring *ring = adapter->rx_ring[i];
3308 		u32 rqdpc = rd32(IGC_RQDPC(i));
3309 
3310 		if (hw->mac.type >= igc_i225)
3311 			wr32(IGC_RQDPC(i), 0);
3312 
3313 		if (rqdpc) {
3314 			ring->rx_stats.drops += rqdpc;
3315 			net_stats->rx_fifo_errors += rqdpc;
3316 		}
3317 
3318 		do {
3319 			start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
3320 			_bytes = ring->rx_stats.bytes;
3321 			_packets = ring->rx_stats.packets;
3322 		} while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
3323 		bytes += _bytes;
3324 		packets += _packets;
3325 	}
3326 
3327 	net_stats->rx_bytes = bytes;
3328 	net_stats->rx_packets = packets;
3329 
3330 	packets = 0;
3331 	bytes = 0;
3332 	for (i = 0; i < adapter->num_tx_queues; i++) {
3333 		struct igc_ring *ring = adapter->tx_ring[i];
3334 
3335 		do {
3336 			start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
3337 			_bytes = ring->tx_stats.bytes;
3338 			_packets = ring->tx_stats.packets;
3339 		} while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
3340 		bytes += _bytes;
3341 		packets += _packets;
3342 	}
3343 	net_stats->tx_bytes = bytes;
3344 	net_stats->tx_packets = packets;
3345 	rcu_read_unlock();
3346 
3347 	/* read stats registers */
3348 	adapter->stats.crcerrs += rd32(IGC_CRCERRS);
3349 	adapter->stats.gprc += rd32(IGC_GPRC);
3350 	adapter->stats.gorc += rd32(IGC_GORCL);
3351 	rd32(IGC_GORCH); /* clear GORCL */
3352 	adapter->stats.bprc += rd32(IGC_BPRC);
3353 	adapter->stats.mprc += rd32(IGC_MPRC);
3354 	adapter->stats.roc += rd32(IGC_ROC);
3355 
3356 	adapter->stats.prc64 += rd32(IGC_PRC64);
3357 	adapter->stats.prc127 += rd32(IGC_PRC127);
3358 	adapter->stats.prc255 += rd32(IGC_PRC255);
3359 	adapter->stats.prc511 += rd32(IGC_PRC511);
3360 	adapter->stats.prc1023 += rd32(IGC_PRC1023);
3361 	adapter->stats.prc1522 += rd32(IGC_PRC1522);
3362 	adapter->stats.symerrs += rd32(IGC_SYMERRS);
3363 	adapter->stats.sec += rd32(IGC_SEC);
3364 
3365 	mpc = rd32(IGC_MPC);
3366 	adapter->stats.mpc += mpc;
3367 	net_stats->rx_fifo_errors += mpc;
3368 	adapter->stats.scc += rd32(IGC_SCC);
3369 	adapter->stats.ecol += rd32(IGC_ECOL);
3370 	adapter->stats.mcc += rd32(IGC_MCC);
3371 	adapter->stats.latecol += rd32(IGC_LATECOL);
3372 	adapter->stats.dc += rd32(IGC_DC);
3373 	adapter->stats.rlec += rd32(IGC_RLEC);
3374 	adapter->stats.xonrxc += rd32(IGC_XONRXC);
3375 	adapter->stats.xontxc += rd32(IGC_XONTXC);
3376 	adapter->stats.xoffrxc += rd32(IGC_XOFFRXC);
3377 	adapter->stats.xofftxc += rd32(IGC_XOFFTXC);
3378 	adapter->stats.fcruc += rd32(IGC_FCRUC);
3379 	adapter->stats.gptc += rd32(IGC_GPTC);
3380 	adapter->stats.gotc += rd32(IGC_GOTCL);
3381 	rd32(IGC_GOTCH); /* clear GOTCL */
3382 	adapter->stats.rnbc += rd32(IGC_RNBC);
3383 	adapter->stats.ruc += rd32(IGC_RUC);
3384 	adapter->stats.rfc += rd32(IGC_RFC);
3385 	adapter->stats.rjc += rd32(IGC_RJC);
3386 	adapter->stats.tor += rd32(IGC_TORH);
3387 	adapter->stats.tot += rd32(IGC_TOTH);
3388 	adapter->stats.tpr += rd32(IGC_TPR);
3389 
3390 	adapter->stats.ptc64 += rd32(IGC_PTC64);
3391 	adapter->stats.ptc127 += rd32(IGC_PTC127);
3392 	adapter->stats.ptc255 += rd32(IGC_PTC255);
3393 	adapter->stats.ptc511 += rd32(IGC_PTC511);
3394 	adapter->stats.ptc1023 += rd32(IGC_PTC1023);
3395 	adapter->stats.ptc1522 += rd32(IGC_PTC1522);
3396 
3397 	adapter->stats.mptc += rd32(IGC_MPTC);
3398 	adapter->stats.bptc += rd32(IGC_BPTC);
3399 
3400 	adapter->stats.tpt += rd32(IGC_TPT);
3401 	adapter->stats.colc += rd32(IGC_COLC);
3402 
3403 	adapter->stats.algnerrc += rd32(IGC_ALGNERRC);
3404 
3405 	adapter->stats.tsctc += rd32(IGC_TSCTC);
3406 	adapter->stats.tsctfc += rd32(IGC_TSCTFC);
3407 
3408 	adapter->stats.iac += rd32(IGC_IAC);
3409 	adapter->stats.icrxoc += rd32(IGC_ICRXOC);
3410 	adapter->stats.icrxptc += rd32(IGC_ICRXPTC);
3411 	adapter->stats.icrxatc += rd32(IGC_ICRXATC);
3412 	adapter->stats.ictxptc += rd32(IGC_ICTXPTC);
3413 	adapter->stats.ictxatc += rd32(IGC_ICTXATC);
3414 	adapter->stats.ictxqec += rd32(IGC_ICTXQEC);
3415 	adapter->stats.ictxqmtc += rd32(IGC_ICTXQMTC);
3416 	adapter->stats.icrxdmtc += rd32(IGC_ICRXDMTC);
3417 
3418 	/* Fill out the OS statistics structure */
3419 	net_stats->multicast = adapter->stats.mprc;
3420 	net_stats->collisions = adapter->stats.colc;
3421 
3422 	/* Rx Errors */
3423 
3424 	/* RLEC on some newer hardware can be incorrect so build
3425 	 * our own version based on RUC and ROC
3426 	 */
3427 	net_stats->rx_errors = adapter->stats.rxerrc +
3428 		adapter->stats.crcerrs + adapter->stats.algnerrc +
3429 		adapter->stats.ruc + adapter->stats.roc +
3430 		adapter->stats.cexterr;
3431 	net_stats->rx_length_errors = adapter->stats.ruc +
3432 				      adapter->stats.roc;
3433 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
3434 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
3435 	net_stats->rx_missed_errors = adapter->stats.mpc;
3436 
3437 	/* Tx Errors */
3438 	net_stats->tx_errors = adapter->stats.ecol +
3439 			       adapter->stats.latecol;
3440 	net_stats->tx_aborted_errors = adapter->stats.ecol;
3441 	net_stats->tx_window_errors = adapter->stats.latecol;
3442 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
3443 
3444 	/* Tx Dropped needs to be maintained elsewhere */
3445 
3446 	/* Management Stats */
3447 	adapter->stats.mgptc += rd32(IGC_MGTPTC);
3448 	adapter->stats.mgprc += rd32(IGC_MGTPRC);
3449 	adapter->stats.mgpdc += rd32(IGC_MGTPDC);
3450 }
3451 
3452 static void igc_nfc_filter_exit(struct igc_adapter *adapter)
3453 {
3454 	struct igc_nfc_filter *rule;
3455 
3456 	spin_lock(&adapter->nfc_lock);
3457 
3458 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
3459 		igc_erase_filter(adapter, rule);
3460 
3461 	hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
3462 		igc_erase_filter(adapter, rule);
3463 
3464 	spin_unlock(&adapter->nfc_lock);
3465 }
3466 
3467 /**
3468  * igc_down - Close the interface
3469  * @adapter: board private structure
3470  */
3471 void igc_down(struct igc_adapter *adapter)
3472 {
3473 	struct net_device *netdev = adapter->netdev;
3474 	struct igc_hw *hw = &adapter->hw;
3475 	u32 tctl, rctl;
3476 	int i = 0;
3477 
3478 	set_bit(__IGC_DOWN, &adapter->state);
3479 
3480 	/* disable receives in the hardware */
3481 	rctl = rd32(IGC_RCTL);
3482 	wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN);
3483 	/* flush and sleep below */
3484 
3485 	igc_nfc_filter_exit(adapter);
3486 
3487 	/* set trans_start so we don't get spurious watchdogs during reset */
3488 	netif_trans_update(netdev);
3489 
3490 	netif_carrier_off(netdev);
3491 	netif_tx_stop_all_queues(netdev);
3492 
3493 	/* disable transmits in the hardware */
3494 	tctl = rd32(IGC_TCTL);
3495 	tctl &= ~IGC_TCTL_EN;
3496 	wr32(IGC_TCTL, tctl);
3497 	/* flush both disables and wait for them to finish */
3498 	wrfl();
3499 	usleep_range(10000, 20000);
3500 
3501 	igc_irq_disable(adapter);
3502 
3503 	adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
3504 
3505 	for (i = 0; i < adapter->num_q_vectors; i++) {
3506 		if (adapter->q_vector[i]) {
3507 			napi_synchronize(&adapter->q_vector[i]->napi);
3508 			napi_disable(&adapter->q_vector[i]->napi);
3509 		}
3510 	}
3511 
3512 	del_timer_sync(&adapter->watchdog_timer);
3513 	del_timer_sync(&adapter->phy_info_timer);
3514 
3515 	/* record the stats before reset*/
3516 	spin_lock(&adapter->stats64_lock);
3517 	igc_update_stats(adapter);
3518 	spin_unlock(&adapter->stats64_lock);
3519 
3520 	adapter->link_speed = 0;
3521 	adapter->link_duplex = 0;
3522 
3523 	if (!pci_channel_offline(adapter->pdev))
3524 		igc_reset(adapter);
3525 
3526 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
3527 	adapter->flags &= ~IGC_FLAG_VLAN_PROMISC;
3528 
3529 	igc_clean_all_tx_rings(adapter);
3530 	igc_clean_all_rx_rings(adapter);
3531 }
3532 
3533 void igc_reinit_locked(struct igc_adapter *adapter)
3534 {
3535 	WARN_ON(in_interrupt());
3536 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
3537 		usleep_range(1000, 2000);
3538 	igc_down(adapter);
3539 	igc_up(adapter);
3540 	clear_bit(__IGC_RESETTING, &adapter->state);
3541 }
3542 
3543 static void igc_reset_task(struct work_struct *work)
3544 {
3545 	struct igc_adapter *adapter;
3546 
3547 	adapter = container_of(work, struct igc_adapter, reset_task);
3548 
3549 	igc_rings_dump(adapter);
3550 	igc_regs_dump(adapter);
3551 	netdev_err(adapter->netdev, "Reset adapter\n");
3552 	igc_reinit_locked(adapter);
3553 }
3554 
3555 /**
3556  * igc_change_mtu - Change the Maximum Transfer Unit
3557  * @netdev: network interface device structure
3558  * @new_mtu: new value for maximum frame size
3559  *
3560  * Returns 0 on success, negative on failure
3561  */
3562 static int igc_change_mtu(struct net_device *netdev, int new_mtu)
3563 {
3564 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
3565 	struct igc_adapter *adapter = netdev_priv(netdev);
3566 
3567 	/* adjust max frame to be at least the size of a standard frame */
3568 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3569 		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
3570 
3571 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
3572 		usleep_range(1000, 2000);
3573 
3574 	/* igc_down has a dependency on max_frame_size */
3575 	adapter->max_frame_size = max_frame;
3576 
3577 	if (netif_running(netdev))
3578 		igc_down(adapter);
3579 
3580 	netdev_dbg(netdev, "changing MTU from %d to %d\n",
3581 		   netdev->mtu, new_mtu);
3582 	netdev->mtu = new_mtu;
3583 
3584 	if (netif_running(netdev))
3585 		igc_up(adapter);
3586 	else
3587 		igc_reset(adapter);
3588 
3589 	clear_bit(__IGC_RESETTING, &adapter->state);
3590 
3591 	return 0;
3592 }
3593 
3594 /**
3595  * igc_get_stats - Get System Network Statistics
3596  * @netdev: network interface device structure
3597  *
3598  * Returns the address of the device statistics structure.
3599  * The statistics are updated here and also from the timer callback.
3600  */
3601 static struct net_device_stats *igc_get_stats(struct net_device *netdev)
3602 {
3603 	struct igc_adapter *adapter = netdev_priv(netdev);
3604 
3605 	if (!test_bit(__IGC_RESETTING, &adapter->state))
3606 		igc_update_stats(adapter);
3607 
3608 	/* only return the current stats */
3609 	return &netdev->stats;
3610 }
3611 
3612 static netdev_features_t igc_fix_features(struct net_device *netdev,
3613 					  netdev_features_t features)
3614 {
3615 	/* Since there is no support for separate Rx/Tx vlan accel
3616 	 * enable/disable make sure Tx flag is always in same state as Rx.
3617 	 */
3618 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
3619 		features |= NETIF_F_HW_VLAN_CTAG_TX;
3620 	else
3621 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
3622 
3623 	return features;
3624 }
3625 
3626 static int igc_set_features(struct net_device *netdev,
3627 			    netdev_features_t features)
3628 {
3629 	netdev_features_t changed = netdev->features ^ features;
3630 	struct igc_adapter *adapter = netdev_priv(netdev);
3631 
3632 	/* Add VLAN support */
3633 	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
3634 		return 0;
3635 
3636 	if (!(features & NETIF_F_NTUPLE)) {
3637 		struct hlist_node *node2;
3638 		struct igc_nfc_filter *rule;
3639 
3640 		spin_lock(&adapter->nfc_lock);
3641 		hlist_for_each_entry_safe(rule, node2,
3642 					  &adapter->nfc_filter_list, nfc_node) {
3643 			igc_erase_filter(adapter, rule);
3644 			hlist_del(&rule->nfc_node);
3645 			kfree(rule);
3646 		}
3647 		spin_unlock(&adapter->nfc_lock);
3648 		adapter->nfc_filter_count = 0;
3649 	}
3650 
3651 	netdev->features = features;
3652 
3653 	if (netif_running(netdev))
3654 		igc_reinit_locked(adapter);
3655 	else
3656 		igc_reset(adapter);
3657 
3658 	return 1;
3659 }
3660 
3661 static netdev_features_t
3662 igc_features_check(struct sk_buff *skb, struct net_device *dev,
3663 		   netdev_features_t features)
3664 {
3665 	unsigned int network_hdr_len, mac_hdr_len;
3666 
3667 	/* Make certain the headers can be described by a context descriptor */
3668 	mac_hdr_len = skb_network_header(skb) - skb->data;
3669 	if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN))
3670 		return features & ~(NETIF_F_HW_CSUM |
3671 				    NETIF_F_SCTP_CRC |
3672 				    NETIF_F_HW_VLAN_CTAG_TX |
3673 				    NETIF_F_TSO |
3674 				    NETIF_F_TSO6);
3675 
3676 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
3677 	if (unlikely(network_hdr_len >  IGC_MAX_NETWORK_HDR_LEN))
3678 		return features & ~(NETIF_F_HW_CSUM |
3679 				    NETIF_F_SCTP_CRC |
3680 				    NETIF_F_TSO |
3681 				    NETIF_F_TSO6);
3682 
3683 	/* We can only support IPv4 TSO in tunnels if we can mangle the
3684 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
3685 	 */
3686 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
3687 		features &= ~NETIF_F_TSO;
3688 
3689 	return features;
3690 }
3691 
3692 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
3693  * 'flags' is used to indicate what kind of match is made, match is by
3694  * default for the destination address, if matching by source address
3695  * is desired the flag IGC_MAC_STATE_SRC_ADDR can be used.
3696  */
3697 static int igc_add_mac_filter_flags(struct igc_adapter *adapter,
3698 				    const u8 *addr, const u8 queue,
3699 				    const u8 flags)
3700 {
3701 	struct igc_hw *hw = &adapter->hw;
3702 	int rar_entries = hw->mac.rar_entry_count;
3703 	int i;
3704 
3705 	if (is_zero_ether_addr(addr))
3706 		return -EINVAL;
3707 
3708 	/* Search for the first empty entry in the MAC table.
3709 	 * Do not touch entries at the end of the table reserved for the VF MAC
3710 	 * addresses.
3711 	 */
3712 	for (i = 0; i < rar_entries; i++) {
3713 		if (!igc_mac_entry_can_be_used(&adapter->mac_table[i],
3714 					       addr, flags))
3715 			continue;
3716 
3717 		ether_addr_copy(adapter->mac_table[i].addr, addr);
3718 		adapter->mac_table[i].queue = queue;
3719 		adapter->mac_table[i].state |= IGC_MAC_STATE_IN_USE | flags;
3720 
3721 		igc_rar_set_index(adapter, i);
3722 		return i;
3723 	}
3724 
3725 	return -ENOSPC;
3726 }
3727 
3728 int igc_add_mac_steering_filter(struct igc_adapter *adapter,
3729 				const u8 *addr, u8 queue, u8 flags)
3730 {
3731 	return igc_add_mac_filter_flags(adapter, addr, queue,
3732 					IGC_MAC_STATE_QUEUE_STEERING | flags);
3733 }
3734 
3735 /* Remove a MAC filter for 'addr' directing matching traffic to
3736  * 'queue', 'flags' is used to indicate what kind of match need to be
3737  * removed, match is by default for the destination address, if
3738  * matching by source address is to be removed the flag
3739  * IGC_MAC_STATE_SRC_ADDR can be used.
3740  */
3741 static int igc_del_mac_filter_flags(struct igc_adapter *adapter,
3742 				    const u8 *addr, const u8 queue,
3743 				    const u8 flags)
3744 {
3745 	struct igc_hw *hw = &adapter->hw;
3746 	int rar_entries = hw->mac.rar_entry_count;
3747 	int i;
3748 
3749 	if (is_zero_ether_addr(addr))
3750 		return -EINVAL;
3751 
3752 	/* Search for matching entry in the MAC table based on given address
3753 	 * and queue. Do not touch entries at the end of the table reserved
3754 	 * for the VF MAC addresses.
3755 	 */
3756 	for (i = 0; i < rar_entries; i++) {
3757 		if (!(adapter->mac_table[i].state & IGC_MAC_STATE_IN_USE))
3758 			continue;
3759 		if ((adapter->mac_table[i].state & flags) != flags)
3760 			continue;
3761 		if (adapter->mac_table[i].queue != queue)
3762 			continue;
3763 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
3764 			continue;
3765 
3766 		/* When a filter for the default address is "deleted",
3767 		 * we return it to its initial configuration
3768 		 */
3769 		if (adapter->mac_table[i].state & IGC_MAC_STATE_DEFAULT) {
3770 			adapter->mac_table[i].state =
3771 				IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
3772 		} else {
3773 			adapter->mac_table[i].state = 0;
3774 			adapter->mac_table[i].queue = 0;
3775 			memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
3776 		}
3777 
3778 		igc_rar_set_index(adapter, i);
3779 		return 0;
3780 	}
3781 
3782 	return -ENOENT;
3783 }
3784 
3785 int igc_del_mac_steering_filter(struct igc_adapter *adapter,
3786 				const u8 *addr, u8 queue, u8 flags)
3787 {
3788 	return igc_del_mac_filter_flags(adapter, addr, queue,
3789 					IGC_MAC_STATE_QUEUE_STEERING | flags);
3790 }
3791 
3792 static void igc_tsync_interrupt(struct igc_adapter *adapter)
3793 {
3794 	struct igc_hw *hw = &adapter->hw;
3795 	u32 tsicr = rd32(IGC_TSICR);
3796 	u32 ack = 0;
3797 
3798 	if (tsicr & IGC_TSICR_TXTS) {
3799 		/* retrieve hardware timestamp */
3800 		schedule_work(&adapter->ptp_tx_work);
3801 		ack |= IGC_TSICR_TXTS;
3802 	}
3803 
3804 	/* acknowledge the interrupts */
3805 	wr32(IGC_TSICR, ack);
3806 }
3807 
3808 /**
3809  * igc_msix_other - msix other interrupt handler
3810  * @irq: interrupt number
3811  * @data: pointer to a q_vector
3812  */
3813 static irqreturn_t igc_msix_other(int irq, void *data)
3814 {
3815 	struct igc_adapter *adapter = data;
3816 	struct igc_hw *hw = &adapter->hw;
3817 	u32 icr = rd32(IGC_ICR);
3818 
3819 	/* reading ICR causes bit 31 of EICR to be cleared */
3820 	if (icr & IGC_ICR_DRSTA)
3821 		schedule_work(&adapter->reset_task);
3822 
3823 	if (icr & IGC_ICR_DOUTSYNC) {
3824 		/* HW is reporting DMA is out of sync */
3825 		adapter->stats.doosync++;
3826 	}
3827 
3828 	if (icr & IGC_ICR_LSC) {
3829 		hw->mac.get_link_status = 1;
3830 		/* guard against interrupt when we're going down */
3831 		if (!test_bit(__IGC_DOWN, &adapter->state))
3832 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
3833 	}
3834 
3835 	if (icr & IGC_ICR_TS)
3836 		igc_tsync_interrupt(adapter);
3837 
3838 	wr32(IGC_EIMS, adapter->eims_other);
3839 
3840 	return IRQ_HANDLED;
3841 }
3842 
3843 static void igc_write_itr(struct igc_q_vector *q_vector)
3844 {
3845 	u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK;
3846 
3847 	if (!q_vector->set_itr)
3848 		return;
3849 
3850 	if (!itr_val)
3851 		itr_val = IGC_ITR_VAL_MASK;
3852 
3853 	itr_val |= IGC_EITR_CNT_IGNR;
3854 
3855 	writel(itr_val, q_vector->itr_register);
3856 	q_vector->set_itr = 0;
3857 }
3858 
3859 static irqreturn_t igc_msix_ring(int irq, void *data)
3860 {
3861 	struct igc_q_vector *q_vector = data;
3862 
3863 	/* Write the ITR value calculated from the previous interrupt. */
3864 	igc_write_itr(q_vector);
3865 
3866 	napi_schedule(&q_vector->napi);
3867 
3868 	return IRQ_HANDLED;
3869 }
3870 
3871 /**
3872  * igc_request_msix - Initialize MSI-X interrupts
3873  * @adapter: Pointer to adapter structure
3874  *
3875  * igc_request_msix allocates MSI-X vectors and requests interrupts from the
3876  * kernel.
3877  */
3878 static int igc_request_msix(struct igc_adapter *adapter)
3879 {
3880 	int i = 0, err = 0, vector = 0, free_vector = 0;
3881 	struct net_device *netdev = adapter->netdev;
3882 
3883 	err = request_irq(adapter->msix_entries[vector].vector,
3884 			  &igc_msix_other, 0, netdev->name, adapter);
3885 	if (err)
3886 		goto err_out;
3887 
3888 	for (i = 0; i < adapter->num_q_vectors; i++) {
3889 		struct igc_q_vector *q_vector = adapter->q_vector[i];
3890 
3891 		vector++;
3892 
3893 		q_vector->itr_register = adapter->io_addr + IGC_EITR(vector);
3894 
3895 		if (q_vector->rx.ring && q_vector->tx.ring)
3896 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
3897 				q_vector->rx.ring->queue_index);
3898 		else if (q_vector->tx.ring)
3899 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
3900 				q_vector->tx.ring->queue_index);
3901 		else if (q_vector->rx.ring)
3902 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
3903 				q_vector->rx.ring->queue_index);
3904 		else
3905 			sprintf(q_vector->name, "%s-unused", netdev->name);
3906 
3907 		err = request_irq(adapter->msix_entries[vector].vector,
3908 				  igc_msix_ring, 0, q_vector->name,
3909 				  q_vector);
3910 		if (err)
3911 			goto err_free;
3912 	}
3913 
3914 	igc_configure_msix(adapter);
3915 	return 0;
3916 
3917 err_free:
3918 	/* free already assigned IRQs */
3919 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
3920 
3921 	vector--;
3922 	for (i = 0; i < vector; i++) {
3923 		free_irq(adapter->msix_entries[free_vector++].vector,
3924 			 adapter->q_vector[i]);
3925 	}
3926 err_out:
3927 	return err;
3928 }
3929 
3930 /**
3931  * igc_clear_interrupt_scheme - reset the device to a state of no interrupts
3932  * @adapter: Pointer to adapter structure
3933  *
3934  * This function resets the device so that it has 0 rx queues, tx queues, and
3935  * MSI-X interrupts allocated.
3936  */
3937 static void igc_clear_interrupt_scheme(struct igc_adapter *adapter)
3938 {
3939 	igc_free_q_vectors(adapter);
3940 	igc_reset_interrupt_capability(adapter);
3941 }
3942 
3943 /* Need to wait a few seconds after link up to get diagnostic information from
3944  * the phy
3945  */
3946 static void igc_update_phy_info(struct timer_list *t)
3947 {
3948 	struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer);
3949 
3950 	igc_get_phy_info(&adapter->hw);
3951 }
3952 
3953 /**
3954  * igc_has_link - check shared code for link and determine up/down
3955  * @adapter: pointer to driver private info
3956  */
3957 bool igc_has_link(struct igc_adapter *adapter)
3958 {
3959 	struct igc_hw *hw = &adapter->hw;
3960 	bool link_active = false;
3961 
3962 	/* get_link_status is set on LSC (link status) interrupt or
3963 	 * rx sequence error interrupt.  get_link_status will stay
3964 	 * false until the igc_check_for_link establishes link
3965 	 * for copper adapters ONLY
3966 	 */
3967 	switch (hw->phy.media_type) {
3968 	case igc_media_type_copper:
3969 		if (!hw->mac.get_link_status)
3970 			return true;
3971 		hw->mac.ops.check_for_link(hw);
3972 		link_active = !hw->mac.get_link_status;
3973 		break;
3974 	default:
3975 	case igc_media_type_unknown:
3976 		break;
3977 	}
3978 
3979 	if (hw->mac.type == igc_i225 &&
3980 	    hw->phy.id == I225_I_PHY_ID) {
3981 		if (!netif_carrier_ok(adapter->netdev)) {
3982 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
3983 		} else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) {
3984 			adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE;
3985 			adapter->link_check_timeout = jiffies;
3986 		}
3987 	}
3988 
3989 	return link_active;
3990 }
3991 
3992 /**
3993  * igc_watchdog - Timer Call-back
3994  * @t: timer for the watchdog
3995  */
3996 static void igc_watchdog(struct timer_list *t)
3997 {
3998 	struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer);
3999 	/* Do the rest outside of interrupt context */
4000 	schedule_work(&adapter->watchdog_task);
4001 }
4002 
4003 static void igc_watchdog_task(struct work_struct *work)
4004 {
4005 	struct igc_adapter *adapter = container_of(work,
4006 						   struct igc_adapter,
4007 						   watchdog_task);
4008 	struct net_device *netdev = adapter->netdev;
4009 	struct igc_hw *hw = &adapter->hw;
4010 	struct igc_phy_info *phy = &hw->phy;
4011 	u16 phy_data, retry_count = 20;
4012 	u32 connsw;
4013 	u32 link;
4014 	int i;
4015 
4016 	link = igc_has_link(adapter);
4017 
4018 	if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) {
4019 		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4020 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
4021 		else
4022 			link = false;
4023 	}
4024 
4025 	/* Force link down if we have fiber to swap to */
4026 	if (adapter->flags & IGC_FLAG_MAS_ENABLE) {
4027 		if (hw->phy.media_type == igc_media_type_copper) {
4028 			connsw = rd32(IGC_CONNSW);
4029 			if (!(connsw & IGC_CONNSW_AUTOSENSE_EN))
4030 				link = 0;
4031 		}
4032 	}
4033 	if (link) {
4034 		/* Cancel scheduled suspend requests. */
4035 		pm_runtime_resume(netdev->dev.parent);
4036 
4037 		if (!netif_carrier_ok(netdev)) {
4038 			u32 ctrl;
4039 
4040 			hw->mac.ops.get_speed_and_duplex(hw,
4041 							 &adapter->link_speed,
4042 							 &adapter->link_duplex);
4043 
4044 			ctrl = rd32(IGC_CTRL);
4045 			/* Link status message must follow this format */
4046 			netdev_info(netdev,
4047 				    "igc: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4048 				    netdev->name,
4049 				    adapter->link_speed,
4050 				    adapter->link_duplex == FULL_DUPLEX ?
4051 				    "Full" : "Half",
4052 				    (ctrl & IGC_CTRL_TFCE) &&
4053 				    (ctrl & IGC_CTRL_RFCE) ? "RX/TX" :
4054 				    (ctrl & IGC_CTRL_RFCE) ?  "RX" :
4055 				    (ctrl & IGC_CTRL_TFCE) ?  "TX" : "None");
4056 
4057 			/* check if SmartSpeed worked */
4058 			igc_check_downshift(hw);
4059 			if (phy->speed_downgraded)
4060 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4061 
4062 			/* adjust timeout factor according to speed/duplex */
4063 			adapter->tx_timeout_factor = 1;
4064 			switch (adapter->link_speed) {
4065 			case SPEED_10:
4066 				adapter->tx_timeout_factor = 14;
4067 				break;
4068 			case SPEED_100:
4069 				/* maybe add some timeout factor ? */
4070 				break;
4071 			}
4072 
4073 			if (adapter->link_speed != SPEED_1000)
4074 				goto no_wait;
4075 
4076 			/* wait for Remote receiver status OK */
4077 retry_read_status:
4078 			if (!igc_read_phy_reg(hw, PHY_1000T_STATUS,
4079 					      &phy_data)) {
4080 				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
4081 				    retry_count) {
4082 					msleep(100);
4083 					retry_count--;
4084 					goto retry_read_status;
4085 				} else if (!retry_count) {
4086 					dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
4087 				}
4088 			} else {
4089 				dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
4090 			}
4091 no_wait:
4092 			netif_carrier_on(netdev);
4093 
4094 			/* link state has changed, schedule phy info update */
4095 			if (!test_bit(__IGC_DOWN, &adapter->state))
4096 				mod_timer(&adapter->phy_info_timer,
4097 					  round_jiffies(jiffies + 2 * HZ));
4098 		}
4099 	} else {
4100 		if (netif_carrier_ok(netdev)) {
4101 			adapter->link_speed = 0;
4102 			adapter->link_duplex = 0;
4103 
4104 			/* Links status message must follow this format */
4105 			netdev_info(netdev, "igc: %s NIC Link is Down\n",
4106 				    netdev->name);
4107 			netif_carrier_off(netdev);
4108 
4109 			/* link state has changed, schedule phy info update */
4110 			if (!test_bit(__IGC_DOWN, &adapter->state))
4111 				mod_timer(&adapter->phy_info_timer,
4112 					  round_jiffies(jiffies + 2 * HZ));
4113 
4114 			/* link is down, time to check for alternate media */
4115 			if (adapter->flags & IGC_FLAG_MAS_ENABLE) {
4116 				if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
4117 					schedule_work(&adapter->reset_task);
4118 					/* return immediately */
4119 					return;
4120 				}
4121 			}
4122 			pm_schedule_suspend(netdev->dev.parent,
4123 					    MSEC_PER_SEC * 5);
4124 
4125 		/* also check for alternate media here */
4126 		} else if (!netif_carrier_ok(netdev) &&
4127 			   (adapter->flags & IGC_FLAG_MAS_ENABLE)) {
4128 			if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
4129 				schedule_work(&adapter->reset_task);
4130 				/* return immediately */
4131 				return;
4132 			}
4133 		}
4134 	}
4135 
4136 	spin_lock(&adapter->stats64_lock);
4137 	igc_update_stats(adapter);
4138 	spin_unlock(&adapter->stats64_lock);
4139 
4140 	for (i = 0; i < adapter->num_tx_queues; i++) {
4141 		struct igc_ring *tx_ring = adapter->tx_ring[i];
4142 
4143 		if (!netif_carrier_ok(netdev)) {
4144 			/* We've lost link, so the controller stops DMA,
4145 			 * but we've got queued Tx work that's never going
4146 			 * to get done, so reset controller to flush Tx.
4147 			 * (Do the reset outside of interrupt context).
4148 			 */
4149 			if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) {
4150 				adapter->tx_timeout_count++;
4151 				schedule_work(&adapter->reset_task);
4152 				/* return immediately since reset is imminent */
4153 				return;
4154 			}
4155 		}
4156 
4157 		/* Force detection of hung controller every watchdog period */
4158 		set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4159 	}
4160 
4161 	/* Cause software interrupt to ensure Rx ring is cleaned */
4162 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
4163 		u32 eics = 0;
4164 
4165 		for (i = 0; i < adapter->num_q_vectors; i++)
4166 			eics |= adapter->q_vector[i]->eims_value;
4167 		wr32(IGC_EICS, eics);
4168 	} else {
4169 		wr32(IGC_ICS, IGC_ICS_RXDMT0);
4170 	}
4171 
4172 	igc_ptp_tx_hang(adapter);
4173 
4174 	/* Reset the timer */
4175 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
4176 		if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)
4177 			mod_timer(&adapter->watchdog_timer,
4178 				  round_jiffies(jiffies +  HZ));
4179 		else
4180 			mod_timer(&adapter->watchdog_timer,
4181 				  round_jiffies(jiffies + 2 * HZ));
4182 	}
4183 }
4184 
4185 /**
4186  * igc_intr_msi - Interrupt Handler
4187  * @irq: interrupt number
4188  * @data: pointer to a network interface device structure
4189  */
4190 static irqreturn_t igc_intr_msi(int irq, void *data)
4191 {
4192 	struct igc_adapter *adapter = data;
4193 	struct igc_q_vector *q_vector = adapter->q_vector[0];
4194 	struct igc_hw *hw = &adapter->hw;
4195 	/* read ICR disables interrupts using IAM */
4196 	u32 icr = rd32(IGC_ICR);
4197 
4198 	igc_write_itr(q_vector);
4199 
4200 	if (icr & IGC_ICR_DRSTA)
4201 		schedule_work(&adapter->reset_task);
4202 
4203 	if (icr & IGC_ICR_DOUTSYNC) {
4204 		/* HW is reporting DMA is out of sync */
4205 		adapter->stats.doosync++;
4206 	}
4207 
4208 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
4209 		hw->mac.get_link_status = 1;
4210 		if (!test_bit(__IGC_DOWN, &adapter->state))
4211 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
4212 	}
4213 
4214 	napi_schedule(&q_vector->napi);
4215 
4216 	return IRQ_HANDLED;
4217 }
4218 
4219 /**
4220  * igc_intr - Legacy Interrupt Handler
4221  * @irq: interrupt number
4222  * @data: pointer to a network interface device structure
4223  */
4224 static irqreturn_t igc_intr(int irq, void *data)
4225 {
4226 	struct igc_adapter *adapter = data;
4227 	struct igc_q_vector *q_vector = adapter->q_vector[0];
4228 	struct igc_hw *hw = &adapter->hw;
4229 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
4230 	 * need for the IMC write
4231 	 */
4232 	u32 icr = rd32(IGC_ICR);
4233 
4234 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4235 	 * not set, then the adapter didn't send an interrupt
4236 	 */
4237 	if (!(icr & IGC_ICR_INT_ASSERTED))
4238 		return IRQ_NONE;
4239 
4240 	igc_write_itr(q_vector);
4241 
4242 	if (icr & IGC_ICR_DRSTA)
4243 		schedule_work(&adapter->reset_task);
4244 
4245 	if (icr & IGC_ICR_DOUTSYNC) {
4246 		/* HW is reporting DMA is out of sync */
4247 		adapter->stats.doosync++;
4248 	}
4249 
4250 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
4251 		hw->mac.get_link_status = 1;
4252 		/* guard against interrupt when we're going down */
4253 		if (!test_bit(__IGC_DOWN, &adapter->state))
4254 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
4255 	}
4256 
4257 	napi_schedule(&q_vector->napi);
4258 
4259 	return IRQ_HANDLED;
4260 }
4261 
4262 static void igc_free_irq(struct igc_adapter *adapter)
4263 {
4264 	if (adapter->msix_entries) {
4265 		int vector = 0, i;
4266 
4267 		free_irq(adapter->msix_entries[vector++].vector, adapter);
4268 
4269 		for (i = 0; i < adapter->num_q_vectors; i++)
4270 			free_irq(adapter->msix_entries[vector++].vector,
4271 				 adapter->q_vector[i]);
4272 	} else {
4273 		free_irq(adapter->pdev->irq, adapter);
4274 	}
4275 }
4276 
4277 /**
4278  * igc_request_irq - initialize interrupts
4279  * @adapter: Pointer to adapter structure
4280  *
4281  * Attempts to configure interrupts using the best available
4282  * capabilities of the hardware and kernel.
4283  */
4284 static int igc_request_irq(struct igc_adapter *adapter)
4285 {
4286 	struct net_device *netdev = adapter->netdev;
4287 	struct pci_dev *pdev = adapter->pdev;
4288 	int err = 0;
4289 
4290 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
4291 		err = igc_request_msix(adapter);
4292 		if (!err)
4293 			goto request_done;
4294 		/* fall back to MSI */
4295 		igc_free_all_tx_resources(adapter);
4296 		igc_free_all_rx_resources(adapter);
4297 
4298 		igc_clear_interrupt_scheme(adapter);
4299 		err = igc_init_interrupt_scheme(adapter, false);
4300 		if (err)
4301 			goto request_done;
4302 		igc_setup_all_tx_resources(adapter);
4303 		igc_setup_all_rx_resources(adapter);
4304 		igc_configure(adapter);
4305 	}
4306 
4307 	igc_assign_vector(adapter->q_vector[0], 0);
4308 
4309 	if (adapter->flags & IGC_FLAG_HAS_MSI) {
4310 		err = request_irq(pdev->irq, &igc_intr_msi, 0,
4311 				  netdev->name, adapter);
4312 		if (!err)
4313 			goto request_done;
4314 
4315 		/* fall back to legacy interrupts */
4316 		igc_reset_interrupt_capability(adapter);
4317 		adapter->flags &= ~IGC_FLAG_HAS_MSI;
4318 	}
4319 
4320 	err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED,
4321 			  netdev->name, adapter);
4322 
4323 	if (err)
4324 		dev_err(&pdev->dev, "Error %d getting interrupt\n",
4325 			err);
4326 
4327 request_done:
4328 	return err;
4329 }
4330 
4331 /**
4332  * __igc_open - Called when a network interface is made active
4333  * @netdev: network interface device structure
4334  * @resuming: boolean indicating if the device is resuming
4335  *
4336  * Returns 0 on success, negative value on failure
4337  *
4338  * The open entry point is called when a network interface is made
4339  * active by the system (IFF_UP).  At this point all resources needed
4340  * for transmit and receive operations are allocated, the interrupt
4341  * handler is registered with the OS, the watchdog timer is started,
4342  * and the stack is notified that the interface is ready.
4343  */
4344 static int __igc_open(struct net_device *netdev, bool resuming)
4345 {
4346 	struct igc_adapter *adapter = netdev_priv(netdev);
4347 	struct pci_dev *pdev = adapter->pdev;
4348 	struct igc_hw *hw = &adapter->hw;
4349 	int err = 0;
4350 	int i = 0;
4351 
4352 	/* disallow open during test */
4353 
4354 	if (test_bit(__IGC_TESTING, &adapter->state)) {
4355 		WARN_ON(resuming);
4356 		return -EBUSY;
4357 	}
4358 
4359 	if (!resuming)
4360 		pm_runtime_get_sync(&pdev->dev);
4361 
4362 	netif_carrier_off(netdev);
4363 
4364 	/* allocate transmit descriptors */
4365 	err = igc_setup_all_tx_resources(adapter);
4366 	if (err)
4367 		goto err_setup_tx;
4368 
4369 	/* allocate receive descriptors */
4370 	err = igc_setup_all_rx_resources(adapter);
4371 	if (err)
4372 		goto err_setup_rx;
4373 
4374 	igc_power_up_link(adapter);
4375 
4376 	igc_configure(adapter);
4377 
4378 	err = igc_request_irq(adapter);
4379 	if (err)
4380 		goto err_req_irq;
4381 
4382 	/* Notify the stack of the actual queue counts. */
4383 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
4384 	if (err)
4385 		goto err_set_queues;
4386 
4387 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
4388 	if (err)
4389 		goto err_set_queues;
4390 
4391 	clear_bit(__IGC_DOWN, &adapter->state);
4392 
4393 	for (i = 0; i < adapter->num_q_vectors; i++)
4394 		napi_enable(&adapter->q_vector[i]->napi);
4395 
4396 	/* Clear any pending interrupts. */
4397 	rd32(IGC_ICR);
4398 	igc_irq_enable(adapter);
4399 
4400 	if (!resuming)
4401 		pm_runtime_put(&pdev->dev);
4402 
4403 	netif_tx_start_all_queues(netdev);
4404 
4405 	/* start the watchdog. */
4406 	hw->mac.get_link_status = 1;
4407 	schedule_work(&adapter->watchdog_task);
4408 
4409 	return IGC_SUCCESS;
4410 
4411 err_set_queues:
4412 	igc_free_irq(adapter);
4413 err_req_irq:
4414 	igc_release_hw_control(adapter);
4415 	igc_power_down_link(adapter);
4416 	igc_free_all_rx_resources(adapter);
4417 err_setup_rx:
4418 	igc_free_all_tx_resources(adapter);
4419 err_setup_tx:
4420 	igc_reset(adapter);
4421 	if (!resuming)
4422 		pm_runtime_put(&pdev->dev);
4423 
4424 	return err;
4425 }
4426 
4427 static int igc_open(struct net_device *netdev)
4428 {
4429 	return __igc_open(netdev, false);
4430 }
4431 
4432 /**
4433  * __igc_close - Disables a network interface
4434  * @netdev: network interface device structure
4435  * @suspending: boolean indicating the device is suspending
4436  *
4437  * Returns 0, this is not allowed to fail
4438  *
4439  * The close entry point is called when an interface is de-activated
4440  * by the OS.  The hardware is still under the driver's control, but
4441  * needs to be disabled.  A global MAC reset is issued to stop the
4442  * hardware, and all transmit and receive resources are freed.
4443  */
4444 static int __igc_close(struct net_device *netdev, bool suspending)
4445 {
4446 	struct igc_adapter *adapter = netdev_priv(netdev);
4447 	struct pci_dev *pdev = adapter->pdev;
4448 
4449 	WARN_ON(test_bit(__IGC_RESETTING, &adapter->state));
4450 
4451 	if (!suspending)
4452 		pm_runtime_get_sync(&pdev->dev);
4453 
4454 	igc_down(adapter);
4455 
4456 	igc_release_hw_control(adapter);
4457 
4458 	igc_free_irq(adapter);
4459 
4460 	igc_free_all_tx_resources(adapter);
4461 	igc_free_all_rx_resources(adapter);
4462 
4463 	if (!suspending)
4464 		pm_runtime_put_sync(&pdev->dev);
4465 
4466 	return 0;
4467 }
4468 
4469 static int igc_close(struct net_device *netdev)
4470 {
4471 	if (netif_device_present(netdev) || netdev->dismantle)
4472 		return __igc_close(netdev, false);
4473 	return 0;
4474 }
4475 
4476 /**
4477  * igc_ioctl - Access the hwtstamp interface
4478  * @netdev: network interface device structure
4479  * @ifreq: interface request data
4480  * @cmd: ioctl command
4481  **/
4482 static int igc_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4483 {
4484 	switch (cmd) {
4485 	case SIOCGHWTSTAMP:
4486 		return igc_ptp_get_ts_config(netdev, ifr);
4487 	case SIOCSHWTSTAMP:
4488 		return igc_ptp_set_ts_config(netdev, ifr);
4489 	default:
4490 		return -EOPNOTSUPP;
4491 	}
4492 }
4493 
4494 static const struct net_device_ops igc_netdev_ops = {
4495 	.ndo_open		= igc_open,
4496 	.ndo_stop		= igc_close,
4497 	.ndo_start_xmit		= igc_xmit_frame,
4498 	.ndo_set_rx_mode	= igc_set_rx_mode,
4499 	.ndo_set_mac_address	= igc_set_mac,
4500 	.ndo_change_mtu		= igc_change_mtu,
4501 	.ndo_get_stats		= igc_get_stats,
4502 	.ndo_fix_features	= igc_fix_features,
4503 	.ndo_set_features	= igc_set_features,
4504 	.ndo_features_check	= igc_features_check,
4505 	.ndo_do_ioctl		= igc_ioctl,
4506 };
4507 
4508 /* PCIe configuration access */
4509 void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
4510 {
4511 	struct igc_adapter *adapter = hw->back;
4512 
4513 	pci_read_config_word(adapter->pdev, reg, value);
4514 }
4515 
4516 void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
4517 {
4518 	struct igc_adapter *adapter = hw->back;
4519 
4520 	pci_write_config_word(adapter->pdev, reg, *value);
4521 }
4522 
4523 s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
4524 {
4525 	struct igc_adapter *adapter = hw->back;
4526 
4527 	if (!pci_is_pcie(adapter->pdev))
4528 		return -IGC_ERR_CONFIG;
4529 
4530 	pcie_capability_read_word(adapter->pdev, reg, value);
4531 
4532 	return IGC_SUCCESS;
4533 }
4534 
4535 s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
4536 {
4537 	struct igc_adapter *adapter = hw->back;
4538 
4539 	if (!pci_is_pcie(adapter->pdev))
4540 		return -IGC_ERR_CONFIG;
4541 
4542 	pcie_capability_write_word(adapter->pdev, reg, *value);
4543 
4544 	return IGC_SUCCESS;
4545 }
4546 
4547 u32 igc_rd32(struct igc_hw *hw, u32 reg)
4548 {
4549 	struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw);
4550 	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
4551 	u32 value = 0;
4552 
4553 	if (IGC_REMOVED(hw_addr))
4554 		return ~value;
4555 
4556 	value = readl(&hw_addr[reg]);
4557 
4558 	/* reads should not return all F's */
4559 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
4560 		struct net_device *netdev = igc->netdev;
4561 
4562 		hw->hw_addr = NULL;
4563 		netif_device_detach(netdev);
4564 		netdev_err(netdev, "PCIe link lost, device now detached\n");
4565 		WARN(pci_device_is_present(igc->pdev),
4566 		     "igc: Failed to read reg 0x%x!\n", reg);
4567 	}
4568 
4569 	return value;
4570 }
4571 
4572 int igc_set_spd_dplx(struct igc_adapter *adapter, u32 spd, u8 dplx)
4573 {
4574 	struct pci_dev *pdev = adapter->pdev;
4575 	struct igc_mac_info *mac = &adapter->hw.mac;
4576 
4577 	mac->autoneg = 0;
4578 
4579 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
4580 	 * for the switch() below to work
4581 	 */
4582 	if ((spd & 1) || (dplx & ~1))
4583 		goto err_inval;
4584 
4585 	switch (spd + dplx) {
4586 	case SPEED_10 + DUPLEX_HALF:
4587 		mac->forced_speed_duplex = ADVERTISE_10_HALF;
4588 		break;
4589 	case SPEED_10 + DUPLEX_FULL:
4590 		mac->forced_speed_duplex = ADVERTISE_10_FULL;
4591 		break;
4592 	case SPEED_100 + DUPLEX_HALF:
4593 		mac->forced_speed_duplex = ADVERTISE_100_HALF;
4594 		break;
4595 	case SPEED_100 + DUPLEX_FULL:
4596 		mac->forced_speed_duplex = ADVERTISE_100_FULL;
4597 		break;
4598 	case SPEED_1000 + DUPLEX_FULL:
4599 		mac->autoneg = 1;
4600 		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
4601 		break;
4602 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
4603 		goto err_inval;
4604 	case SPEED_2500 + DUPLEX_FULL:
4605 		mac->autoneg = 1;
4606 		adapter->hw.phy.autoneg_advertised = ADVERTISE_2500_FULL;
4607 		break;
4608 	case SPEED_2500 + DUPLEX_HALF: /* not supported */
4609 	default:
4610 		goto err_inval;
4611 	}
4612 
4613 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
4614 	adapter->hw.phy.mdix = AUTO_ALL_MODES;
4615 
4616 	return 0;
4617 
4618 err_inval:
4619 	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
4620 	return -EINVAL;
4621 }
4622 
4623 /**
4624  * igc_probe - Device Initialization Routine
4625  * @pdev: PCI device information struct
4626  * @ent: entry in igc_pci_tbl
4627  *
4628  * Returns 0 on success, negative on failure
4629  *
4630  * igc_probe initializes an adapter identified by a pci_dev structure.
4631  * The OS initialization, configuring the adapter private structure,
4632  * and a hardware reset occur.
4633  */
4634 static int igc_probe(struct pci_dev *pdev,
4635 		     const struct pci_device_id *ent)
4636 {
4637 	struct igc_adapter *adapter;
4638 	struct net_device *netdev;
4639 	struct igc_hw *hw;
4640 	const struct igc_info *ei = igc_info_tbl[ent->driver_data];
4641 	int err, pci_using_dac;
4642 
4643 	err = pci_enable_device_mem(pdev);
4644 	if (err)
4645 		return err;
4646 
4647 	pci_using_dac = 0;
4648 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4649 	if (!err) {
4650 		pci_using_dac = 1;
4651 	} else {
4652 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4653 		if (err) {
4654 			dev_err(&pdev->dev,
4655 				"No usable DMA configuration, aborting\n");
4656 			goto err_dma;
4657 		}
4658 	}
4659 
4660 	err = pci_request_mem_regions(pdev, igc_driver_name);
4661 	if (err)
4662 		goto err_pci_reg;
4663 
4664 	pci_enable_pcie_error_reporting(pdev);
4665 
4666 	pci_set_master(pdev);
4667 
4668 	err = -ENOMEM;
4669 	netdev = alloc_etherdev_mq(sizeof(struct igc_adapter),
4670 				   IGC_MAX_TX_QUEUES);
4671 
4672 	if (!netdev)
4673 		goto err_alloc_etherdev;
4674 
4675 	SET_NETDEV_DEV(netdev, &pdev->dev);
4676 
4677 	pci_set_drvdata(pdev, netdev);
4678 	adapter = netdev_priv(netdev);
4679 	adapter->netdev = netdev;
4680 	adapter->pdev = pdev;
4681 	hw = &adapter->hw;
4682 	hw->back = adapter;
4683 	adapter->port_num = hw->bus.func;
4684 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4685 
4686 	err = pci_save_state(pdev);
4687 	if (err)
4688 		goto err_ioremap;
4689 
4690 	err = -EIO;
4691 	adapter->io_addr = ioremap(pci_resource_start(pdev, 0),
4692 				   pci_resource_len(pdev, 0));
4693 	if (!adapter->io_addr)
4694 		goto err_ioremap;
4695 
4696 	/* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */
4697 	hw->hw_addr = adapter->io_addr;
4698 
4699 	netdev->netdev_ops = &igc_netdev_ops;
4700 	igc_set_ethtool_ops(netdev);
4701 	netdev->watchdog_timeo = 5 * HZ;
4702 
4703 	netdev->mem_start = pci_resource_start(pdev, 0);
4704 	netdev->mem_end = pci_resource_end(pdev, 0);
4705 
4706 	/* PCI config space info */
4707 	hw->vendor_id = pdev->vendor;
4708 	hw->device_id = pdev->device;
4709 	hw->revision_id = pdev->revision;
4710 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4711 	hw->subsystem_device_id = pdev->subsystem_device;
4712 
4713 	/* Copy the default MAC and PHY function pointers */
4714 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4715 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4716 
4717 	/* Initialize skew-specific constants */
4718 	err = ei->get_invariants(hw);
4719 	if (err)
4720 		goto err_sw_init;
4721 
4722 	/* Add supported features to the features list*/
4723 	netdev->features |= NETIF_F_SG;
4724 	netdev->features |= NETIF_F_TSO;
4725 	netdev->features |= NETIF_F_TSO6;
4726 	netdev->features |= NETIF_F_RXCSUM;
4727 	netdev->features |= NETIF_F_HW_CSUM;
4728 	netdev->features |= NETIF_F_SCTP_CRC;
4729 
4730 	/* setup the private structure */
4731 	err = igc_sw_init(adapter);
4732 	if (err)
4733 		goto err_sw_init;
4734 
4735 	/* copy netdev features into list of user selectable features */
4736 	netdev->hw_features |= NETIF_F_NTUPLE;
4737 	netdev->hw_features |= netdev->features;
4738 
4739 	if (pci_using_dac)
4740 		netdev->features |= NETIF_F_HIGHDMA;
4741 
4742 	/* MTU range: 68 - 9216 */
4743 	netdev->min_mtu = ETH_MIN_MTU;
4744 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
4745 
4746 	/* before reading the NVM, reset the controller to put the device in a
4747 	 * known good starting state
4748 	 */
4749 	hw->mac.ops.reset_hw(hw);
4750 
4751 	if (igc_get_flash_presence_i225(hw)) {
4752 		if (hw->nvm.ops.validate(hw) < 0) {
4753 			dev_err(&pdev->dev,
4754 				"The NVM Checksum Is Not Valid\n");
4755 			err = -EIO;
4756 			goto err_eeprom;
4757 		}
4758 	}
4759 
4760 	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
4761 		/* copy the MAC address out of the NVM */
4762 		if (hw->mac.ops.read_mac_addr(hw))
4763 			dev_err(&pdev->dev, "NVM Read Error\n");
4764 	}
4765 
4766 	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
4767 
4768 	if (!is_valid_ether_addr(netdev->dev_addr)) {
4769 		dev_err(&pdev->dev, "Invalid MAC Address\n");
4770 		err = -EIO;
4771 		goto err_eeprom;
4772 	}
4773 
4774 	/* configure RXPBSIZE and TXPBSIZE */
4775 	wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT);
4776 	wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT);
4777 
4778 	timer_setup(&adapter->watchdog_timer, igc_watchdog, 0);
4779 	timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0);
4780 
4781 	INIT_WORK(&adapter->reset_task, igc_reset_task);
4782 	INIT_WORK(&adapter->watchdog_task, igc_watchdog_task);
4783 
4784 	/* Initialize link properties that are user-changeable */
4785 	adapter->fc_autoneg = true;
4786 	hw->mac.autoneg = true;
4787 	hw->phy.autoneg_advertised = 0xaf;
4788 
4789 	hw->fc.requested_mode = igc_fc_default;
4790 	hw->fc.current_mode = igc_fc_default;
4791 
4792 	/* By default, support wake on port A */
4793 	adapter->flags |= IGC_FLAG_WOL_SUPPORTED;
4794 
4795 	/* initialize the wol settings based on the eeprom settings */
4796 	if (adapter->flags & IGC_FLAG_WOL_SUPPORTED)
4797 		adapter->wol |= IGC_WUFC_MAG;
4798 
4799 	device_set_wakeup_enable(&adapter->pdev->dev,
4800 				 adapter->flags & IGC_FLAG_WOL_SUPPORTED);
4801 
4802 	/* reset the hardware with the new settings */
4803 	igc_reset(adapter);
4804 
4805 	/* let the f/w know that the h/w is now under the control of the
4806 	 * driver.
4807 	 */
4808 	igc_get_hw_control(adapter);
4809 
4810 	strncpy(netdev->name, "eth%d", IFNAMSIZ);
4811 	err = register_netdev(netdev);
4812 	if (err)
4813 		goto err_register;
4814 
4815 	 /* carrier off reporting is important to ethtool even BEFORE open */
4816 	netif_carrier_off(netdev);
4817 
4818 	/* do hw tstamp init after resetting */
4819 	igc_ptp_init(adapter);
4820 
4821 	/* Check if Media Autosense is enabled */
4822 	adapter->ei = *ei;
4823 
4824 	/* print pcie link status and MAC address */
4825 	pcie_print_link_status(pdev);
4826 	netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr);
4827 
4828 	dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NEVER_SKIP);
4829 
4830 	pm_runtime_put_noidle(&pdev->dev);
4831 
4832 	return 0;
4833 
4834 err_register:
4835 	igc_release_hw_control(adapter);
4836 err_eeprom:
4837 	if (!igc_check_reset_block(hw))
4838 		igc_reset_phy(hw);
4839 err_sw_init:
4840 	igc_clear_interrupt_scheme(adapter);
4841 	iounmap(adapter->io_addr);
4842 err_ioremap:
4843 	free_netdev(netdev);
4844 err_alloc_etherdev:
4845 	pci_release_mem_regions(pdev);
4846 err_pci_reg:
4847 err_dma:
4848 	pci_disable_device(pdev);
4849 	return err;
4850 }
4851 
4852 /**
4853  * igc_remove - Device Removal Routine
4854  * @pdev: PCI device information struct
4855  *
4856  * igc_remove is called by the PCI subsystem to alert the driver
4857  * that it should release a PCI device.  This could be caused by a
4858  * Hot-Plug event, or because the driver is going to be removed from
4859  * memory.
4860  */
4861 static void igc_remove(struct pci_dev *pdev)
4862 {
4863 	struct net_device *netdev = pci_get_drvdata(pdev);
4864 	struct igc_adapter *adapter = netdev_priv(netdev);
4865 
4866 	pm_runtime_get_noresume(&pdev->dev);
4867 
4868 	igc_ptp_stop(adapter);
4869 
4870 	set_bit(__IGC_DOWN, &adapter->state);
4871 
4872 	del_timer_sync(&adapter->watchdog_timer);
4873 	del_timer_sync(&adapter->phy_info_timer);
4874 
4875 	cancel_work_sync(&adapter->reset_task);
4876 	cancel_work_sync(&adapter->watchdog_task);
4877 
4878 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
4879 	 * would have already happened in close and is redundant.
4880 	 */
4881 	igc_release_hw_control(adapter);
4882 	unregister_netdev(netdev);
4883 
4884 	igc_clear_interrupt_scheme(adapter);
4885 	pci_iounmap(pdev, adapter->io_addr);
4886 	pci_release_mem_regions(pdev);
4887 
4888 	kfree(adapter->mac_table);
4889 	free_netdev(netdev);
4890 
4891 	pci_disable_pcie_error_reporting(pdev);
4892 
4893 	pci_disable_device(pdev);
4894 }
4895 
4896 static int __igc_shutdown(struct pci_dev *pdev, bool *enable_wake,
4897 			  bool runtime)
4898 {
4899 	struct net_device *netdev = pci_get_drvdata(pdev);
4900 	struct igc_adapter *adapter = netdev_priv(netdev);
4901 	u32 wufc = runtime ? IGC_WUFC_LNKC : adapter->wol;
4902 	struct igc_hw *hw = &adapter->hw;
4903 	u32 ctrl, rctl, status;
4904 	bool wake;
4905 
4906 	rtnl_lock();
4907 	netif_device_detach(netdev);
4908 
4909 	if (netif_running(netdev))
4910 		__igc_close(netdev, true);
4911 
4912 	igc_ptp_suspend(adapter);
4913 
4914 	igc_clear_interrupt_scheme(adapter);
4915 	rtnl_unlock();
4916 
4917 	status = rd32(IGC_STATUS);
4918 	if (status & IGC_STATUS_LU)
4919 		wufc &= ~IGC_WUFC_LNKC;
4920 
4921 	if (wufc) {
4922 		igc_setup_rctl(adapter);
4923 		igc_set_rx_mode(netdev);
4924 
4925 		/* turn on all-multi mode if wake on multicast is enabled */
4926 		if (wufc & IGC_WUFC_MC) {
4927 			rctl = rd32(IGC_RCTL);
4928 			rctl |= IGC_RCTL_MPE;
4929 			wr32(IGC_RCTL, rctl);
4930 		}
4931 
4932 		ctrl = rd32(IGC_CTRL);
4933 		ctrl |= IGC_CTRL_ADVD3WUC;
4934 		wr32(IGC_CTRL, ctrl);
4935 
4936 		/* Allow time for pending master requests to run */
4937 		igc_disable_pcie_master(hw);
4938 
4939 		wr32(IGC_WUC, IGC_WUC_PME_EN);
4940 		wr32(IGC_WUFC, wufc);
4941 	} else {
4942 		wr32(IGC_WUC, 0);
4943 		wr32(IGC_WUFC, 0);
4944 	}
4945 
4946 	wake = wufc || adapter->en_mng_pt;
4947 	if (!wake)
4948 		igc_power_down_link(adapter);
4949 	else
4950 		igc_power_up_link(adapter);
4951 
4952 	if (enable_wake)
4953 		*enable_wake = wake;
4954 
4955 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
4956 	 * would have already happened in close and is redundant.
4957 	 */
4958 	igc_release_hw_control(adapter);
4959 
4960 	pci_disable_device(pdev);
4961 
4962 	return 0;
4963 }
4964 
4965 #ifdef CONFIG_PM
4966 static int __maybe_unused igc_runtime_suspend(struct device *dev)
4967 {
4968 	return __igc_shutdown(to_pci_dev(dev), NULL, 1);
4969 }
4970 
4971 static void igc_deliver_wake_packet(struct net_device *netdev)
4972 {
4973 	struct igc_adapter *adapter = netdev_priv(netdev);
4974 	struct igc_hw *hw = &adapter->hw;
4975 	struct sk_buff *skb;
4976 	u32 wupl;
4977 
4978 	wupl = rd32(IGC_WUPL) & IGC_WUPL_MASK;
4979 
4980 	/* WUPM stores only the first 128 bytes of the wake packet.
4981 	 * Read the packet only if we have the whole thing.
4982 	 */
4983 	if (wupl == 0 || wupl > IGC_WUPM_BYTES)
4984 		return;
4985 
4986 	skb = netdev_alloc_skb_ip_align(netdev, IGC_WUPM_BYTES);
4987 	if (!skb)
4988 		return;
4989 
4990 	skb_put(skb, wupl);
4991 
4992 	/* Ensure reads are 32-bit aligned */
4993 	wupl = roundup(wupl, 4);
4994 
4995 	memcpy_fromio(skb->data, hw->hw_addr + IGC_WUPM_REG(0), wupl);
4996 
4997 	skb->protocol = eth_type_trans(skb, netdev);
4998 	netif_rx(skb);
4999 }
5000 
5001 static int __maybe_unused igc_resume(struct device *dev)
5002 {
5003 	struct pci_dev *pdev = to_pci_dev(dev);
5004 	struct net_device *netdev = pci_get_drvdata(pdev);
5005 	struct igc_adapter *adapter = netdev_priv(netdev);
5006 	struct igc_hw *hw = &adapter->hw;
5007 	u32 err, val;
5008 
5009 	pci_set_power_state(pdev, PCI_D0);
5010 	pci_restore_state(pdev);
5011 	pci_save_state(pdev);
5012 
5013 	if (!pci_device_is_present(pdev))
5014 		return -ENODEV;
5015 	err = pci_enable_device_mem(pdev);
5016 	if (err) {
5017 		dev_err(&pdev->dev,
5018 			"igc: Cannot enable PCI device from suspend\n");
5019 		return err;
5020 	}
5021 	pci_set_master(pdev);
5022 
5023 	pci_enable_wake(pdev, PCI_D3hot, 0);
5024 	pci_enable_wake(pdev, PCI_D3cold, 0);
5025 
5026 	if (igc_init_interrupt_scheme(adapter, true)) {
5027 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5028 		return -ENOMEM;
5029 	}
5030 
5031 	igc_reset(adapter);
5032 
5033 	/* let the f/w know that the h/w is now under the control of the
5034 	 * driver.
5035 	 */
5036 	igc_get_hw_control(adapter);
5037 
5038 	val = rd32(IGC_WUS);
5039 	if (val & WAKE_PKT_WUS)
5040 		igc_deliver_wake_packet(netdev);
5041 
5042 	wr32(IGC_WUS, ~0);
5043 
5044 	rtnl_lock();
5045 	if (!err && netif_running(netdev))
5046 		err = __igc_open(netdev, true);
5047 
5048 	if (!err)
5049 		netif_device_attach(netdev);
5050 	rtnl_unlock();
5051 
5052 	return err;
5053 }
5054 
5055 static int __maybe_unused igc_runtime_resume(struct device *dev)
5056 {
5057 	return igc_resume(dev);
5058 }
5059 
5060 static int __maybe_unused igc_suspend(struct device *dev)
5061 {
5062 	return __igc_shutdown(to_pci_dev(dev), NULL, 0);
5063 }
5064 
5065 static int __maybe_unused igc_runtime_idle(struct device *dev)
5066 {
5067 	struct net_device *netdev = dev_get_drvdata(dev);
5068 	struct igc_adapter *adapter = netdev_priv(netdev);
5069 
5070 	if (!igc_has_link(adapter))
5071 		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
5072 
5073 	return -EBUSY;
5074 }
5075 #endif /* CONFIG_PM */
5076 
5077 static void igc_shutdown(struct pci_dev *pdev)
5078 {
5079 	bool wake;
5080 
5081 	__igc_shutdown(pdev, &wake, 0);
5082 
5083 	if (system_state == SYSTEM_POWER_OFF) {
5084 		pci_wake_from_d3(pdev, wake);
5085 		pci_set_power_state(pdev, PCI_D3hot);
5086 	}
5087 }
5088 
5089 /**
5090  *  igc_io_error_detected - called when PCI error is detected
5091  *  @pdev: Pointer to PCI device
5092  *  @state: The current PCI connection state
5093  *
5094  *  This function is called after a PCI bus error affecting
5095  *  this device has been detected.
5096  **/
5097 static pci_ers_result_t igc_io_error_detected(struct pci_dev *pdev,
5098 					      pci_channel_state_t state)
5099 {
5100 	struct net_device *netdev = pci_get_drvdata(pdev);
5101 	struct igc_adapter *adapter = netdev_priv(netdev);
5102 
5103 	netif_device_detach(netdev);
5104 
5105 	if (state == pci_channel_io_perm_failure)
5106 		return PCI_ERS_RESULT_DISCONNECT;
5107 
5108 	if (netif_running(netdev))
5109 		igc_down(adapter);
5110 	pci_disable_device(pdev);
5111 
5112 	/* Request a slot reset. */
5113 	return PCI_ERS_RESULT_NEED_RESET;
5114 }
5115 
5116 /**
5117  *  igc_io_slot_reset - called after the PCI bus has been reset.
5118  *  @pdev: Pointer to PCI device
5119  *
5120  *  Restart the card from scratch, as if from a cold-boot. Implementation
5121  *  resembles the first-half of the igc_resume routine.
5122  **/
5123 static pci_ers_result_t igc_io_slot_reset(struct pci_dev *pdev)
5124 {
5125 	struct net_device *netdev = pci_get_drvdata(pdev);
5126 	struct igc_adapter *adapter = netdev_priv(netdev);
5127 	struct igc_hw *hw = &adapter->hw;
5128 	pci_ers_result_t result;
5129 
5130 	if (pci_enable_device_mem(pdev)) {
5131 		dev_err(&pdev->dev,
5132 			"Could not re-enable PCI device after reset.\n");
5133 		result = PCI_ERS_RESULT_DISCONNECT;
5134 	} else {
5135 		pci_set_master(pdev);
5136 		pci_restore_state(pdev);
5137 		pci_save_state(pdev);
5138 
5139 		pci_enable_wake(pdev, PCI_D3hot, 0);
5140 		pci_enable_wake(pdev, PCI_D3cold, 0);
5141 
5142 		/* In case of PCI error, adapter loses its HW address
5143 		 * so we should re-assign it here.
5144 		 */
5145 		hw->hw_addr = adapter->io_addr;
5146 
5147 		igc_reset(adapter);
5148 		wr32(IGC_WUS, ~0);
5149 		result = PCI_ERS_RESULT_RECOVERED;
5150 	}
5151 
5152 	return result;
5153 }
5154 
5155 /**
5156  *  igc_io_resume - called when traffic can start to flow again.
5157  *  @pdev: Pointer to PCI device
5158  *
5159  *  This callback is called when the error recovery driver tells us that
5160  *  its OK to resume normal operation. Implementation resembles the
5161  *  second-half of the igc_resume routine.
5162  */
5163 static void igc_io_resume(struct pci_dev *pdev)
5164 {
5165 	struct net_device *netdev = pci_get_drvdata(pdev);
5166 	struct igc_adapter *adapter = netdev_priv(netdev);
5167 
5168 	rtnl_lock();
5169 	if (netif_running(netdev)) {
5170 		if (igc_open(netdev)) {
5171 			dev_err(&pdev->dev, "igc_open failed after reset\n");
5172 			return;
5173 		}
5174 	}
5175 
5176 	netif_device_attach(netdev);
5177 
5178 	/* let the f/w know that the h/w is now under the control of the
5179 	 * driver.
5180 	 */
5181 	igc_get_hw_control(adapter);
5182 	rtnl_unlock();
5183 }
5184 
5185 static const struct pci_error_handlers igc_err_handler = {
5186 	.error_detected = igc_io_error_detected,
5187 	.slot_reset = igc_io_slot_reset,
5188 	.resume = igc_io_resume,
5189 };
5190 
5191 #ifdef CONFIG_PM
5192 static const struct dev_pm_ops igc_pm_ops = {
5193 	SET_SYSTEM_SLEEP_PM_OPS(igc_suspend, igc_resume)
5194 	SET_RUNTIME_PM_OPS(igc_runtime_suspend, igc_runtime_resume,
5195 			   igc_runtime_idle)
5196 };
5197 #endif
5198 
5199 static struct pci_driver igc_driver = {
5200 	.name     = igc_driver_name,
5201 	.id_table = igc_pci_tbl,
5202 	.probe    = igc_probe,
5203 	.remove   = igc_remove,
5204 #ifdef CONFIG_PM
5205 	.driver.pm = &igc_pm_ops,
5206 #endif
5207 	.shutdown = igc_shutdown,
5208 	.err_handler = &igc_err_handler,
5209 };
5210 
5211 /**
5212  * igc_reinit_queues - return error
5213  * @adapter: pointer to adapter structure
5214  */
5215 int igc_reinit_queues(struct igc_adapter *adapter)
5216 {
5217 	struct net_device *netdev = adapter->netdev;
5218 	struct pci_dev *pdev = adapter->pdev;
5219 	int err = 0;
5220 
5221 	if (netif_running(netdev))
5222 		igc_close(netdev);
5223 
5224 	igc_reset_interrupt_capability(adapter);
5225 
5226 	if (igc_init_interrupt_scheme(adapter, true)) {
5227 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5228 		return -ENOMEM;
5229 	}
5230 
5231 	if (netif_running(netdev))
5232 		err = igc_open(netdev);
5233 
5234 	return err;
5235 }
5236 
5237 /**
5238  * igc_get_hw_dev - return device
5239  * @hw: pointer to hardware structure
5240  *
5241  * used by hardware layer to print debugging information
5242  */
5243 struct net_device *igc_get_hw_dev(struct igc_hw *hw)
5244 {
5245 	struct igc_adapter *adapter = hw->back;
5246 
5247 	return adapter->netdev;
5248 }
5249 
5250 /**
5251  * igc_init_module - Driver Registration Routine
5252  *
5253  * igc_init_module is the first routine called when the driver is
5254  * loaded. All it does is register with the PCI subsystem.
5255  */
5256 static int __init igc_init_module(void)
5257 {
5258 	int ret;
5259 
5260 	pr_info("%s - version %s\n",
5261 		igc_driver_string, igc_driver_version);
5262 
5263 	pr_info("%s\n", igc_copyright);
5264 
5265 	ret = pci_register_driver(&igc_driver);
5266 	return ret;
5267 }
5268 
5269 module_init(igc_init_module);
5270 
5271 /**
5272  * igc_exit_module - Driver Exit Cleanup Routine
5273  *
5274  * igc_exit_module is called just before the driver is removed
5275  * from memory.
5276  */
5277 static void __exit igc_exit_module(void)
5278 {
5279 	pci_unregister_driver(&igc_driver);
5280 }
5281 
5282 module_exit(igc_exit_module);
5283 /* igc_main.c */
5284