xref: /linux/drivers/net/ethernet/intel/igc/igc_main.c (revision 2c63221cd9e5c0dad0424029aeb1c40faada8330)
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 
12 #include <net/ipv6.h>
13 
14 #include "igc.h"
15 #include "igc_hw.h"
16 
17 #define DRV_VERSION	"0.0.1-k"
18 #define DRV_SUMMARY	"Intel(R) 2.5G Ethernet Linux Driver"
19 
20 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
21 
22 static int debug = -1;
23 
24 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
25 MODULE_DESCRIPTION(DRV_SUMMARY);
26 MODULE_LICENSE("GPL v2");
27 MODULE_VERSION(DRV_VERSION);
28 module_param(debug, int, 0);
29 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
30 
31 char igc_driver_name[] = "igc";
32 char igc_driver_version[] = DRV_VERSION;
33 static const char igc_driver_string[] = DRV_SUMMARY;
34 static const char igc_copyright[] =
35 	"Copyright(c) 2018 Intel Corporation.";
36 
37 static const struct igc_info *igc_info_tbl[] = {
38 	[board_base] = &igc_base_info,
39 };
40 
41 static const struct pci_device_id igc_pci_tbl[] = {
42 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base },
43 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base },
44 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base },
45 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base },
46 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base },
47 	/* required last entry */
48 	{0, }
49 };
50 
51 MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
52 
53 /* forward declaration */
54 static void igc_clean_tx_ring(struct igc_ring *tx_ring);
55 static int igc_sw_init(struct igc_adapter *);
56 static void igc_configure(struct igc_adapter *adapter);
57 static void igc_power_down_link(struct igc_adapter *adapter);
58 static void igc_set_default_mac_filter(struct igc_adapter *adapter);
59 static void igc_set_rx_mode(struct net_device *netdev);
60 static void igc_write_itr(struct igc_q_vector *q_vector);
61 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector);
62 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx);
63 static void igc_set_interrupt_capability(struct igc_adapter *adapter,
64 					 bool msix);
65 static void igc_free_q_vectors(struct igc_adapter *adapter);
66 static void igc_irq_disable(struct igc_adapter *adapter);
67 static void igc_irq_enable(struct igc_adapter *adapter);
68 static void igc_configure_msix(struct igc_adapter *adapter);
69 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
70 				  struct igc_rx_buffer *bi);
71 
72 enum latency_range {
73 	lowest_latency = 0,
74 	low_latency = 1,
75 	bulk_latency = 2,
76 	latency_invalid = 255
77 };
78 
79 void igc_reset(struct igc_adapter *adapter)
80 {
81 	struct pci_dev *pdev = adapter->pdev;
82 	struct igc_hw *hw = &adapter->hw;
83 	struct igc_fc_info *fc = &hw->fc;
84 	u32 pba, hwm;
85 
86 	/* Repartition PBA for greater than 9k MTU if required */
87 	pba = IGC_PBA_34K;
88 
89 	/* flow control settings
90 	 * The high water mark must be low enough to fit one full frame
91 	 * after transmitting the pause frame.  As such we must have enough
92 	 * space to allow for us to complete our current transmit and then
93 	 * receive the frame that is in progress from the link partner.
94 	 * Set it to:
95 	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
96 	 */
97 	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
98 
99 	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
100 	fc->low_water = fc->high_water - 16;
101 	fc->pause_time = 0xFFFF;
102 	fc->send_xon = 1;
103 	fc->current_mode = fc->requested_mode;
104 
105 	hw->mac.ops.reset_hw(hw);
106 
107 	if (hw->mac.ops.init_hw(hw))
108 		dev_err(&pdev->dev, "Hardware Error\n");
109 
110 	if (!netif_running(adapter->netdev))
111 		igc_power_down_link(adapter);
112 
113 	igc_get_phy_info(hw);
114 }
115 
116 /**
117  * igc_power_up_link - Power up the phy/serdes link
118  * @adapter: address of board private structure
119  */
120 static void igc_power_up_link(struct igc_adapter *adapter)
121 {
122 	igc_reset_phy(&adapter->hw);
123 
124 	if (adapter->hw.phy.media_type == igc_media_type_copper)
125 		igc_power_up_phy_copper(&adapter->hw);
126 
127 	igc_setup_link(&adapter->hw);
128 }
129 
130 /**
131  * igc_power_down_link - Power down the phy/serdes link
132  * @adapter: address of board private structure
133  */
134 static void igc_power_down_link(struct igc_adapter *adapter)
135 {
136 	if (adapter->hw.phy.media_type == igc_media_type_copper)
137 		igc_power_down_phy_copper_base(&adapter->hw);
138 }
139 
140 /**
141  * igc_release_hw_control - release control of the h/w to f/w
142  * @adapter: address of board private structure
143  *
144  * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
145  * For ASF and Pass Through versions of f/w this means that the
146  * driver is no longer loaded.
147  */
148 static void igc_release_hw_control(struct igc_adapter *adapter)
149 {
150 	struct igc_hw *hw = &adapter->hw;
151 	u32 ctrl_ext;
152 
153 	/* Let firmware take over control of h/w */
154 	ctrl_ext = rd32(IGC_CTRL_EXT);
155 	wr32(IGC_CTRL_EXT,
156 	     ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
157 }
158 
159 /**
160  * igc_get_hw_control - get control of the h/w from f/w
161  * @adapter: address of board private structure
162  *
163  * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
164  * For ASF and Pass Through versions of f/w this means that
165  * the driver is loaded.
166  */
167 static void igc_get_hw_control(struct igc_adapter *adapter)
168 {
169 	struct igc_hw *hw = &adapter->hw;
170 	u32 ctrl_ext;
171 
172 	/* Let firmware know the driver has taken over */
173 	ctrl_ext = rd32(IGC_CTRL_EXT);
174 	wr32(IGC_CTRL_EXT,
175 	     ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
176 }
177 
178 /**
179  * igc_free_tx_resources - Free Tx Resources per Queue
180  * @tx_ring: Tx descriptor ring for a specific queue
181  *
182  * Free all transmit software resources
183  */
184 void igc_free_tx_resources(struct igc_ring *tx_ring)
185 {
186 	igc_clean_tx_ring(tx_ring);
187 
188 	vfree(tx_ring->tx_buffer_info);
189 	tx_ring->tx_buffer_info = NULL;
190 
191 	/* if not set, then don't free */
192 	if (!tx_ring->desc)
193 		return;
194 
195 	dma_free_coherent(tx_ring->dev, tx_ring->size,
196 			  tx_ring->desc, tx_ring->dma);
197 
198 	tx_ring->desc = NULL;
199 }
200 
201 /**
202  * igc_free_all_tx_resources - Free Tx Resources for All Queues
203  * @adapter: board private structure
204  *
205  * Free all transmit software resources
206  */
207 static void igc_free_all_tx_resources(struct igc_adapter *adapter)
208 {
209 	int i;
210 
211 	for (i = 0; i < adapter->num_tx_queues; i++)
212 		igc_free_tx_resources(adapter->tx_ring[i]);
213 }
214 
215 /**
216  * igc_clean_tx_ring - Free Tx Buffers
217  * @tx_ring: ring to be cleaned
218  */
219 static void igc_clean_tx_ring(struct igc_ring *tx_ring)
220 {
221 	u16 i = tx_ring->next_to_clean;
222 	struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
223 
224 	while (i != tx_ring->next_to_use) {
225 		union igc_adv_tx_desc *eop_desc, *tx_desc;
226 
227 		/* Free all the Tx ring sk_buffs */
228 		dev_kfree_skb_any(tx_buffer->skb);
229 
230 		/* unmap skb header data */
231 		dma_unmap_single(tx_ring->dev,
232 				 dma_unmap_addr(tx_buffer, dma),
233 				 dma_unmap_len(tx_buffer, len),
234 				 DMA_TO_DEVICE);
235 
236 		/* check for eop_desc to determine the end of the packet */
237 		eop_desc = tx_buffer->next_to_watch;
238 		tx_desc = IGC_TX_DESC(tx_ring, i);
239 
240 		/* unmap remaining buffers */
241 		while (tx_desc != eop_desc) {
242 			tx_buffer++;
243 			tx_desc++;
244 			i++;
245 			if (unlikely(i == tx_ring->count)) {
246 				i = 0;
247 				tx_buffer = tx_ring->tx_buffer_info;
248 				tx_desc = IGC_TX_DESC(tx_ring, 0);
249 			}
250 
251 			/* unmap any remaining paged data */
252 			if (dma_unmap_len(tx_buffer, len))
253 				dma_unmap_page(tx_ring->dev,
254 					       dma_unmap_addr(tx_buffer, dma),
255 					       dma_unmap_len(tx_buffer, len),
256 					       DMA_TO_DEVICE);
257 		}
258 
259 		/* move us one more past the eop_desc for start of next pkt */
260 		tx_buffer++;
261 		i++;
262 		if (unlikely(i == tx_ring->count)) {
263 			i = 0;
264 			tx_buffer = tx_ring->tx_buffer_info;
265 		}
266 	}
267 
268 	/* reset BQL for queue */
269 	netdev_tx_reset_queue(txring_txq(tx_ring));
270 
271 	/* reset next_to_use and next_to_clean */
272 	tx_ring->next_to_use = 0;
273 	tx_ring->next_to_clean = 0;
274 }
275 
276 /**
277  * igc_clean_all_tx_rings - Free Tx Buffers for all queues
278  * @adapter: board private structure
279  */
280 static void igc_clean_all_tx_rings(struct igc_adapter *adapter)
281 {
282 	int i;
283 
284 	for (i = 0; i < adapter->num_tx_queues; i++)
285 		if (adapter->tx_ring[i])
286 			igc_clean_tx_ring(adapter->tx_ring[i]);
287 }
288 
289 /**
290  * igc_setup_tx_resources - allocate Tx resources (Descriptors)
291  * @tx_ring: tx descriptor ring (for a specific queue) to setup
292  *
293  * Return 0 on success, negative on failure
294  */
295 int igc_setup_tx_resources(struct igc_ring *tx_ring)
296 {
297 	struct device *dev = tx_ring->dev;
298 	int size = 0;
299 
300 	size = sizeof(struct igc_tx_buffer) * tx_ring->count;
301 	tx_ring->tx_buffer_info = vzalloc(size);
302 	if (!tx_ring->tx_buffer_info)
303 		goto err;
304 
305 	/* round up to nearest 4K */
306 	tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc);
307 	tx_ring->size = ALIGN(tx_ring->size, 4096);
308 
309 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
310 					   &tx_ring->dma, GFP_KERNEL);
311 
312 	if (!tx_ring->desc)
313 		goto err;
314 
315 	tx_ring->next_to_use = 0;
316 	tx_ring->next_to_clean = 0;
317 
318 	return 0;
319 
320 err:
321 	vfree(tx_ring->tx_buffer_info);
322 	dev_err(dev,
323 		"Unable to allocate memory for the transmit descriptor ring\n");
324 	return -ENOMEM;
325 }
326 
327 /**
328  * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues
329  * @adapter: board private structure
330  *
331  * Return 0 on success, negative on failure
332  */
333 static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
334 {
335 	struct pci_dev *pdev = adapter->pdev;
336 	int i, err = 0;
337 
338 	for (i = 0; i < adapter->num_tx_queues; i++) {
339 		err = igc_setup_tx_resources(adapter->tx_ring[i]);
340 		if (err) {
341 			dev_err(&pdev->dev,
342 				"Allocation for Tx Queue %u failed\n", i);
343 			for (i--; i >= 0; i--)
344 				igc_free_tx_resources(adapter->tx_ring[i]);
345 			break;
346 		}
347 	}
348 
349 	return err;
350 }
351 
352 /**
353  * igc_clean_rx_ring - Free Rx Buffers per Queue
354  * @rx_ring: ring to free buffers from
355  */
356 static void igc_clean_rx_ring(struct igc_ring *rx_ring)
357 {
358 	u16 i = rx_ring->next_to_clean;
359 
360 	dev_kfree_skb(rx_ring->skb);
361 	rx_ring->skb = NULL;
362 
363 	/* Free all the Rx ring sk_buffs */
364 	while (i != rx_ring->next_to_alloc) {
365 		struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
366 
367 		/* Invalidate cache lines that may have been written to by
368 		 * device so that we avoid corrupting memory.
369 		 */
370 		dma_sync_single_range_for_cpu(rx_ring->dev,
371 					      buffer_info->dma,
372 					      buffer_info->page_offset,
373 					      igc_rx_bufsz(rx_ring),
374 					      DMA_FROM_DEVICE);
375 
376 		/* free resources associated with mapping */
377 		dma_unmap_page_attrs(rx_ring->dev,
378 				     buffer_info->dma,
379 				     igc_rx_pg_size(rx_ring),
380 				     DMA_FROM_DEVICE,
381 				     IGC_RX_DMA_ATTR);
382 		__page_frag_cache_drain(buffer_info->page,
383 					buffer_info->pagecnt_bias);
384 
385 		i++;
386 		if (i == rx_ring->count)
387 			i = 0;
388 	}
389 
390 	rx_ring->next_to_alloc = 0;
391 	rx_ring->next_to_clean = 0;
392 	rx_ring->next_to_use = 0;
393 }
394 
395 /**
396  * igc_clean_all_rx_rings - Free Rx Buffers for all queues
397  * @adapter: board private structure
398  */
399 static void igc_clean_all_rx_rings(struct igc_adapter *adapter)
400 {
401 	int i;
402 
403 	for (i = 0; i < adapter->num_rx_queues; i++)
404 		if (adapter->rx_ring[i])
405 			igc_clean_rx_ring(adapter->rx_ring[i]);
406 }
407 
408 /**
409  * igc_free_rx_resources - Free Rx Resources
410  * @rx_ring: ring to clean the resources from
411  *
412  * Free all receive software resources
413  */
414 void igc_free_rx_resources(struct igc_ring *rx_ring)
415 {
416 	igc_clean_rx_ring(rx_ring);
417 
418 	vfree(rx_ring->rx_buffer_info);
419 	rx_ring->rx_buffer_info = NULL;
420 
421 	/* if not set, then don't free */
422 	if (!rx_ring->desc)
423 		return;
424 
425 	dma_free_coherent(rx_ring->dev, rx_ring->size,
426 			  rx_ring->desc, rx_ring->dma);
427 
428 	rx_ring->desc = NULL;
429 }
430 
431 /**
432  * igc_free_all_rx_resources - Free Rx Resources for All Queues
433  * @adapter: board private structure
434  *
435  * Free all receive software resources
436  */
437 static void igc_free_all_rx_resources(struct igc_adapter *adapter)
438 {
439 	int i;
440 
441 	for (i = 0; i < adapter->num_rx_queues; i++)
442 		igc_free_rx_resources(adapter->rx_ring[i]);
443 }
444 
445 /**
446  * igc_setup_rx_resources - allocate Rx resources (Descriptors)
447  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
448  *
449  * Returns 0 on success, negative on failure
450  */
451 int igc_setup_rx_resources(struct igc_ring *rx_ring)
452 {
453 	struct device *dev = rx_ring->dev;
454 	int size, desc_len;
455 
456 	size = sizeof(struct igc_rx_buffer) * rx_ring->count;
457 	rx_ring->rx_buffer_info = vzalloc(size);
458 	if (!rx_ring->rx_buffer_info)
459 		goto err;
460 
461 	desc_len = sizeof(union igc_adv_rx_desc);
462 
463 	/* Round up to nearest 4K */
464 	rx_ring->size = rx_ring->count * desc_len;
465 	rx_ring->size = ALIGN(rx_ring->size, 4096);
466 
467 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
468 					   &rx_ring->dma, GFP_KERNEL);
469 
470 	if (!rx_ring->desc)
471 		goto err;
472 
473 	rx_ring->next_to_alloc = 0;
474 	rx_ring->next_to_clean = 0;
475 	rx_ring->next_to_use = 0;
476 
477 	return 0;
478 
479 err:
480 	vfree(rx_ring->rx_buffer_info);
481 	rx_ring->rx_buffer_info = NULL;
482 	dev_err(dev,
483 		"Unable to allocate memory for the receive descriptor ring\n");
484 	return -ENOMEM;
485 }
486 
487 /**
488  * igc_setup_all_rx_resources - wrapper to allocate Rx resources
489  *                                (Descriptors) for all queues
490  * @adapter: board private structure
491  *
492  * Return 0 on success, negative on failure
493  */
494 static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
495 {
496 	struct pci_dev *pdev = adapter->pdev;
497 	int i, err = 0;
498 
499 	for (i = 0; i < adapter->num_rx_queues; i++) {
500 		err = igc_setup_rx_resources(adapter->rx_ring[i]);
501 		if (err) {
502 			dev_err(&pdev->dev,
503 				"Allocation for Rx Queue %u failed\n", i);
504 			for (i--; i >= 0; i--)
505 				igc_free_rx_resources(adapter->rx_ring[i]);
506 			break;
507 		}
508 	}
509 
510 	return err;
511 }
512 
513 /**
514  * igc_configure_rx_ring - Configure a receive ring after Reset
515  * @adapter: board private structure
516  * @ring: receive ring to be configured
517  *
518  * Configure the Rx unit of the MAC after a reset.
519  */
520 static void igc_configure_rx_ring(struct igc_adapter *adapter,
521 				  struct igc_ring *ring)
522 {
523 	struct igc_hw *hw = &adapter->hw;
524 	union igc_adv_rx_desc *rx_desc;
525 	int reg_idx = ring->reg_idx;
526 	u32 srrctl = 0, rxdctl = 0;
527 	u64 rdba = ring->dma;
528 
529 	/* disable the queue */
530 	wr32(IGC_RXDCTL(reg_idx), 0);
531 
532 	/* Set DMA base address registers */
533 	wr32(IGC_RDBAL(reg_idx),
534 	     rdba & 0x00000000ffffffffULL);
535 	wr32(IGC_RDBAH(reg_idx), rdba >> 32);
536 	wr32(IGC_RDLEN(reg_idx),
537 	     ring->count * sizeof(union igc_adv_rx_desc));
538 
539 	/* initialize head and tail */
540 	ring->tail = adapter->io_addr + IGC_RDT(reg_idx);
541 	wr32(IGC_RDH(reg_idx), 0);
542 	writel(0, ring->tail);
543 
544 	/* reset next-to- use/clean to place SW in sync with hardware */
545 	ring->next_to_clean = 0;
546 	ring->next_to_use = 0;
547 
548 	/* set descriptor configuration */
549 	srrctl = IGC_RX_HDR_LEN << IGC_SRRCTL_BSIZEHDRSIZE_SHIFT;
550 	if (ring_uses_large_buffer(ring))
551 		srrctl |= IGC_RXBUFFER_3072 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
552 	else
553 		srrctl |= IGC_RXBUFFER_2048 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
554 	srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
555 
556 	wr32(IGC_SRRCTL(reg_idx), srrctl);
557 
558 	rxdctl |= IGC_RX_PTHRESH;
559 	rxdctl |= IGC_RX_HTHRESH << 8;
560 	rxdctl |= IGC_RX_WTHRESH << 16;
561 
562 	/* initialize rx_buffer_info */
563 	memset(ring->rx_buffer_info, 0,
564 	       sizeof(struct igc_rx_buffer) * ring->count);
565 
566 	/* initialize Rx descriptor 0 */
567 	rx_desc = IGC_RX_DESC(ring, 0);
568 	rx_desc->wb.upper.length = 0;
569 
570 	/* enable receive descriptor fetching */
571 	rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
572 
573 	wr32(IGC_RXDCTL(reg_idx), rxdctl);
574 }
575 
576 /**
577  * igc_configure_rx - Configure receive Unit after Reset
578  * @adapter: board private structure
579  *
580  * Configure the Rx unit of the MAC after a reset.
581  */
582 static void igc_configure_rx(struct igc_adapter *adapter)
583 {
584 	int i;
585 
586 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
587 	 * the Base and Length of the Rx Descriptor Ring
588 	 */
589 	for (i = 0; i < adapter->num_rx_queues; i++)
590 		igc_configure_rx_ring(adapter, adapter->rx_ring[i]);
591 }
592 
593 /**
594  * igc_configure_tx_ring - Configure transmit ring after Reset
595  * @adapter: board private structure
596  * @ring: tx ring to configure
597  *
598  * Configure a transmit ring after a reset.
599  */
600 static void igc_configure_tx_ring(struct igc_adapter *adapter,
601 				  struct igc_ring *ring)
602 {
603 	struct igc_hw *hw = &adapter->hw;
604 	int reg_idx = ring->reg_idx;
605 	u64 tdba = ring->dma;
606 	u32 txdctl = 0;
607 
608 	/* disable the queue */
609 	wr32(IGC_TXDCTL(reg_idx), 0);
610 	wrfl();
611 	mdelay(10);
612 
613 	wr32(IGC_TDLEN(reg_idx),
614 	     ring->count * sizeof(union igc_adv_tx_desc));
615 	wr32(IGC_TDBAL(reg_idx),
616 	     tdba & 0x00000000ffffffffULL);
617 	wr32(IGC_TDBAH(reg_idx), tdba >> 32);
618 
619 	ring->tail = adapter->io_addr + IGC_TDT(reg_idx);
620 	wr32(IGC_TDH(reg_idx), 0);
621 	writel(0, ring->tail);
622 
623 	txdctl |= IGC_TX_PTHRESH;
624 	txdctl |= IGC_TX_HTHRESH << 8;
625 	txdctl |= IGC_TX_WTHRESH << 16;
626 
627 	txdctl |= IGC_TXDCTL_QUEUE_ENABLE;
628 	wr32(IGC_TXDCTL(reg_idx), txdctl);
629 }
630 
631 /**
632  * igc_configure_tx - Configure transmit Unit after Reset
633  * @adapter: board private structure
634  *
635  * Configure the Tx unit of the MAC after a reset.
636  */
637 static void igc_configure_tx(struct igc_adapter *adapter)
638 {
639 	int i;
640 
641 	for (i = 0; i < adapter->num_tx_queues; i++)
642 		igc_configure_tx_ring(adapter, adapter->tx_ring[i]);
643 }
644 
645 /**
646  * igc_setup_mrqc - configure the multiple receive queue control registers
647  * @adapter: Board private structure
648  */
649 static void igc_setup_mrqc(struct igc_adapter *adapter)
650 {
651 	struct igc_hw *hw = &adapter->hw;
652 	u32 j, num_rx_queues;
653 	u32 mrqc, rxcsum;
654 	u32 rss_key[10];
655 
656 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
657 	for (j = 0; j < 10; j++)
658 		wr32(IGC_RSSRK(j), rss_key[j]);
659 
660 	num_rx_queues = adapter->rss_queues;
661 
662 	if (adapter->rss_indir_tbl_init != num_rx_queues) {
663 		for (j = 0; j < IGC_RETA_SIZE; j++)
664 			adapter->rss_indir_tbl[j] =
665 			(j * num_rx_queues) / IGC_RETA_SIZE;
666 		adapter->rss_indir_tbl_init = num_rx_queues;
667 	}
668 	igc_write_rss_indir_tbl(adapter);
669 
670 	/* Disable raw packet checksumming so that RSS hash is placed in
671 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
672 	 * offloads as they are enabled by default
673 	 */
674 	rxcsum = rd32(IGC_RXCSUM);
675 	rxcsum |= IGC_RXCSUM_PCSD;
676 
677 	/* Enable Receive Checksum Offload for SCTP */
678 	rxcsum |= IGC_RXCSUM_CRCOFL;
679 
680 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
681 	wr32(IGC_RXCSUM, rxcsum);
682 
683 	/* Generate RSS hash based on packet types, TCP/UDP
684 	 * port numbers and/or IPv4/v6 src and dst addresses
685 	 */
686 	mrqc = IGC_MRQC_RSS_FIELD_IPV4 |
687 	       IGC_MRQC_RSS_FIELD_IPV4_TCP |
688 	       IGC_MRQC_RSS_FIELD_IPV6 |
689 	       IGC_MRQC_RSS_FIELD_IPV6_TCP |
690 	       IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
691 
692 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP)
693 		mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP;
694 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP)
695 		mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP;
696 
697 	mrqc |= IGC_MRQC_ENABLE_RSS_MQ;
698 
699 	wr32(IGC_MRQC, mrqc);
700 }
701 
702 /**
703  * igc_setup_rctl - configure the receive control registers
704  * @adapter: Board private structure
705  */
706 static void igc_setup_rctl(struct igc_adapter *adapter)
707 {
708 	struct igc_hw *hw = &adapter->hw;
709 	u32 rctl;
710 
711 	rctl = rd32(IGC_RCTL);
712 
713 	rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
714 	rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC);
715 
716 	rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF |
717 		(hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
718 
719 	/* enable stripping of CRC. Newer features require
720 	 * that the HW strips the CRC.
721 	 */
722 	rctl |= IGC_RCTL_SECRC;
723 
724 	/* disable store bad packets and clear size bits. */
725 	rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256);
726 
727 	/* enable LPE to allow for reception of jumbo frames */
728 	rctl |= IGC_RCTL_LPE;
729 
730 	/* disable queue 0 to prevent tail write w/o re-config */
731 	wr32(IGC_RXDCTL(0), 0);
732 
733 	/* This is useful for sniffing bad packets. */
734 	if (adapter->netdev->features & NETIF_F_RXALL) {
735 		/* UPE and MPE will be handled by normal PROMISC logic
736 		 * in set_rx_mode
737 		 */
738 		rctl |= (IGC_RCTL_SBP | /* Receive bad packets */
739 			 IGC_RCTL_BAM | /* RX All Bcast Pkts */
740 			 IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
741 
742 		rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */
743 			  IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */
744 	}
745 
746 	wr32(IGC_RCTL, rctl);
747 }
748 
749 /**
750  * igc_setup_tctl - configure the transmit control registers
751  * @adapter: Board private structure
752  */
753 static void igc_setup_tctl(struct igc_adapter *adapter)
754 {
755 	struct igc_hw *hw = &adapter->hw;
756 	u32 tctl;
757 
758 	/* disable queue 0 which icould be enabled by default */
759 	wr32(IGC_TXDCTL(0), 0);
760 
761 	/* Program the Transmit Control Register */
762 	tctl = rd32(IGC_TCTL);
763 	tctl &= ~IGC_TCTL_CT;
764 	tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC |
765 		(IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT);
766 
767 	/* Enable transmits */
768 	tctl |= IGC_TCTL_EN;
769 
770 	wr32(IGC_TCTL, tctl);
771 }
772 
773 /**
774  * igc_set_mac - Change the Ethernet Address of the NIC
775  * @netdev: network interface device structure
776  * @p: pointer to an address structure
777  *
778  * Returns 0 on success, negative on failure
779  */
780 static int igc_set_mac(struct net_device *netdev, void *p)
781 {
782 	struct igc_adapter *adapter = netdev_priv(netdev);
783 	struct igc_hw *hw = &adapter->hw;
784 	struct sockaddr *addr = p;
785 
786 	if (!is_valid_ether_addr(addr->sa_data))
787 		return -EADDRNOTAVAIL;
788 
789 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
790 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
791 
792 	/* set the correct pool for the new PF MAC address in entry 0 */
793 	igc_set_default_mac_filter(adapter);
794 
795 	return 0;
796 }
797 
798 /**
799  *  igc_write_mc_addr_list - write multicast addresses to MTA
800  *  @netdev: network interface device structure
801  *
802  *  Writes multicast address list to the MTA hash table.
803  *  Returns: -ENOMEM on failure
804  *           0 on no addresses written
805  *           X on writing X addresses to MTA
806  **/
807 static int igc_write_mc_addr_list(struct net_device *netdev)
808 {
809 	struct igc_adapter *adapter = netdev_priv(netdev);
810 	struct igc_hw *hw = &adapter->hw;
811 	struct netdev_hw_addr *ha;
812 	u8  *mta_list;
813 	int i;
814 
815 	if (netdev_mc_empty(netdev)) {
816 		/* nothing to program, so clear mc list */
817 		igc_update_mc_addr_list(hw, NULL, 0);
818 		return 0;
819 	}
820 
821 	mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
822 	if (!mta_list)
823 		return -ENOMEM;
824 
825 	/* The shared function expects a packed array of only addresses. */
826 	i = 0;
827 	netdev_for_each_mc_addr(ha, netdev)
828 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
829 
830 	igc_update_mc_addr_list(hw, mta_list, i);
831 	kfree(mta_list);
832 
833 	return netdev_mc_count(netdev);
834 }
835 
836 static void igc_tx_ctxtdesc(struct igc_ring *tx_ring,
837 			    struct igc_tx_buffer *first,
838 			    u32 vlan_macip_lens, u32 type_tucmd,
839 			    u32 mss_l4len_idx)
840 {
841 	struct igc_adv_tx_context_desc *context_desc;
842 	u16 i = tx_ring->next_to_use;
843 	struct timespec64 ts;
844 
845 	context_desc = IGC_TX_CTXTDESC(tx_ring, i);
846 
847 	i++;
848 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
849 
850 	/* set bits to identify this as an advanced context descriptor */
851 	type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
852 
853 	/* For 82575, context index must be unique per ring. */
854 	if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
855 		mss_l4len_idx |= tx_ring->reg_idx << 4;
856 
857 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
858 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
859 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
860 
861 	/* We assume there is always a valid Tx time available. Invalid times
862 	 * should have been handled by the upper layers.
863 	 */
864 	if (tx_ring->launchtime_enable) {
865 		ts = ns_to_timespec64(first->skb->tstamp);
866 		first->skb->tstamp = 0;
867 		context_desc->launch_time = cpu_to_le32(ts.tv_nsec / 32);
868 	} else {
869 		context_desc->launch_time = 0;
870 	}
871 }
872 
873 static inline bool igc_ipv6_csum_is_sctp(struct sk_buff *skb)
874 {
875 	unsigned int offset = 0;
876 
877 	ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
878 
879 	return offset == skb_checksum_start_offset(skb);
880 }
881 
882 static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first)
883 {
884 	struct sk_buff *skb = first->skb;
885 	u32 vlan_macip_lens = 0;
886 	u32 type_tucmd = 0;
887 
888 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
889 csum_failed:
890 		if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) &&
891 		    !tx_ring->launchtime_enable)
892 			return;
893 		goto no_csum;
894 	}
895 
896 	switch (skb->csum_offset) {
897 	case offsetof(struct tcphdr, check):
898 		type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
899 		/* fall through */
900 	case offsetof(struct udphdr, check):
901 		break;
902 	case offsetof(struct sctphdr, checksum):
903 		/* validate that this is actually an SCTP request */
904 		if ((first->protocol == htons(ETH_P_IP) &&
905 		     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
906 		    (first->protocol == htons(ETH_P_IPV6) &&
907 		     igc_ipv6_csum_is_sctp(skb))) {
908 			type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP;
909 			break;
910 		}
911 		/* fall through */
912 	default:
913 		skb_checksum_help(skb);
914 		goto csum_failed;
915 	}
916 
917 	/* update TX checksum flag */
918 	first->tx_flags |= IGC_TX_FLAGS_CSUM;
919 	vlan_macip_lens = skb_checksum_start_offset(skb) -
920 			  skb_network_offset(skb);
921 no_csum:
922 	vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT;
923 	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
924 
925 	igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
926 }
927 
928 static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
929 {
930 	struct net_device *netdev = tx_ring->netdev;
931 
932 	netif_stop_subqueue(netdev, tx_ring->queue_index);
933 
934 	/* memory barriier comment */
935 	smp_mb();
936 
937 	/* We need to check again in a case another CPU has just
938 	 * made room available.
939 	 */
940 	if (igc_desc_unused(tx_ring) < size)
941 		return -EBUSY;
942 
943 	/* A reprieve! */
944 	netif_wake_subqueue(netdev, tx_ring->queue_index);
945 
946 	u64_stats_update_begin(&tx_ring->tx_syncp2);
947 	tx_ring->tx_stats.restart_queue2++;
948 	u64_stats_update_end(&tx_ring->tx_syncp2);
949 
950 	return 0;
951 }
952 
953 static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
954 {
955 	if (igc_desc_unused(tx_ring) >= size)
956 		return 0;
957 	return __igc_maybe_stop_tx(tx_ring, size);
958 }
959 
960 static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
961 {
962 	/* set type for advanced descriptor with frame checksum insertion */
963 	u32 cmd_type = IGC_ADVTXD_DTYP_DATA |
964 		       IGC_ADVTXD_DCMD_DEXT |
965 		       IGC_ADVTXD_DCMD_IFCS;
966 
967 	return cmd_type;
968 }
969 
970 static void igc_tx_olinfo_status(struct igc_ring *tx_ring,
971 				 union igc_adv_tx_desc *tx_desc,
972 				 u32 tx_flags, unsigned int paylen)
973 {
974 	u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT;
975 
976 	/* insert L4 checksum */
977 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_CSUM) *
978 			  ((IGC_TXD_POPTS_TXSM << 8) /
979 			  IGC_TX_FLAGS_CSUM);
980 
981 	/* insert IPv4 checksum */
982 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_IPV4) *
983 			  (((IGC_TXD_POPTS_IXSM << 8)) /
984 			  IGC_TX_FLAGS_IPV4);
985 
986 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
987 }
988 
989 static int igc_tx_map(struct igc_ring *tx_ring,
990 		      struct igc_tx_buffer *first,
991 		      const u8 hdr_len)
992 {
993 	struct sk_buff *skb = first->skb;
994 	struct igc_tx_buffer *tx_buffer;
995 	union igc_adv_tx_desc *tx_desc;
996 	u32 tx_flags = first->tx_flags;
997 	skb_frag_t *frag;
998 	u16 i = tx_ring->next_to_use;
999 	unsigned int data_len, size;
1000 	dma_addr_t dma;
1001 	u32 cmd_type = igc_tx_cmd_type(skb, tx_flags);
1002 
1003 	tx_desc = IGC_TX_DESC(tx_ring, i);
1004 
1005 	igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
1006 
1007 	size = skb_headlen(skb);
1008 	data_len = skb->data_len;
1009 
1010 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1011 
1012 	tx_buffer = first;
1013 
1014 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1015 		if (dma_mapping_error(tx_ring->dev, dma))
1016 			goto dma_error;
1017 
1018 		/* record length, and DMA address */
1019 		dma_unmap_len_set(tx_buffer, len, size);
1020 		dma_unmap_addr_set(tx_buffer, dma, dma);
1021 
1022 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
1023 
1024 		while (unlikely(size > IGC_MAX_DATA_PER_TXD)) {
1025 			tx_desc->read.cmd_type_len =
1026 				cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD);
1027 
1028 			i++;
1029 			tx_desc++;
1030 			if (i == tx_ring->count) {
1031 				tx_desc = IGC_TX_DESC(tx_ring, 0);
1032 				i = 0;
1033 			}
1034 			tx_desc->read.olinfo_status = 0;
1035 
1036 			dma += IGC_MAX_DATA_PER_TXD;
1037 			size -= IGC_MAX_DATA_PER_TXD;
1038 
1039 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
1040 		}
1041 
1042 		if (likely(!data_len))
1043 			break;
1044 
1045 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
1046 
1047 		i++;
1048 		tx_desc++;
1049 		if (i == tx_ring->count) {
1050 			tx_desc = IGC_TX_DESC(tx_ring, 0);
1051 			i = 0;
1052 		}
1053 		tx_desc->read.olinfo_status = 0;
1054 
1055 		size = skb_frag_size(frag);
1056 		data_len -= size;
1057 
1058 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
1059 				       size, DMA_TO_DEVICE);
1060 
1061 		tx_buffer = &tx_ring->tx_buffer_info[i];
1062 	}
1063 
1064 	/* write last descriptor with RS and EOP bits */
1065 	cmd_type |= size | IGC_TXD_DCMD;
1066 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1067 
1068 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1069 
1070 	/* set the timestamp */
1071 	first->time_stamp = jiffies;
1072 
1073 	skb_tx_timestamp(skb);
1074 
1075 	/* Force memory writes to complete before letting h/w know there
1076 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1077 	 * memory model archs, such as IA-64).
1078 	 *
1079 	 * We also need this memory barrier to make certain all of the
1080 	 * status bits have been updated before next_to_watch is written.
1081 	 */
1082 	wmb();
1083 
1084 	/* set next_to_watch value indicating a packet is present */
1085 	first->next_to_watch = tx_desc;
1086 
1087 	i++;
1088 	if (i == tx_ring->count)
1089 		i = 0;
1090 
1091 	tx_ring->next_to_use = i;
1092 
1093 	/* Make sure there is space in the ring for the next send. */
1094 	igc_maybe_stop_tx(tx_ring, DESC_NEEDED);
1095 
1096 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1097 		writel(i, tx_ring->tail);
1098 	}
1099 
1100 	return 0;
1101 dma_error:
1102 	dev_err(tx_ring->dev, "TX DMA map failed\n");
1103 	tx_buffer = &tx_ring->tx_buffer_info[i];
1104 
1105 	/* clear dma mappings for failed tx_buffer_info map */
1106 	while (tx_buffer != first) {
1107 		if (dma_unmap_len(tx_buffer, len))
1108 			dma_unmap_page(tx_ring->dev,
1109 				       dma_unmap_addr(tx_buffer, dma),
1110 				       dma_unmap_len(tx_buffer, len),
1111 				       DMA_TO_DEVICE);
1112 		dma_unmap_len_set(tx_buffer, len, 0);
1113 
1114 		if (i-- == 0)
1115 			i += tx_ring->count;
1116 		tx_buffer = &tx_ring->tx_buffer_info[i];
1117 	}
1118 
1119 	if (dma_unmap_len(tx_buffer, len))
1120 		dma_unmap_single(tx_ring->dev,
1121 				 dma_unmap_addr(tx_buffer, dma),
1122 				 dma_unmap_len(tx_buffer, len),
1123 				 DMA_TO_DEVICE);
1124 	dma_unmap_len_set(tx_buffer, len, 0);
1125 
1126 	dev_kfree_skb_any(tx_buffer->skb);
1127 	tx_buffer->skb = NULL;
1128 
1129 	tx_ring->next_to_use = i;
1130 
1131 	return -1;
1132 }
1133 
1134 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
1135 				       struct igc_ring *tx_ring)
1136 {
1137 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1138 	__be16 protocol = vlan_get_protocol(skb);
1139 	struct igc_tx_buffer *first;
1140 	u32 tx_flags = 0;
1141 	unsigned short f;
1142 	u8 hdr_len = 0;
1143 
1144 	/* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
1145 	 *	+ 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
1146 	 *	+ 2 desc gap to keep tail from touching head,
1147 	 *	+ 1 desc for context descriptor,
1148 	 * otherwise try next time
1149 	 */
1150 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1151 		count += TXD_USE_COUNT(skb_frag_size(
1152 						&skb_shinfo(skb)->frags[f]));
1153 
1154 	if (igc_maybe_stop_tx(tx_ring, count + 3)) {
1155 		/* this is a hard error */
1156 		return NETDEV_TX_BUSY;
1157 	}
1158 
1159 	/* record the location of the first descriptor for this packet */
1160 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1161 	first->skb = skb;
1162 	first->bytecount = skb->len;
1163 	first->gso_segs = 1;
1164 
1165 	/* record initial flags and protocol */
1166 	first->tx_flags = tx_flags;
1167 	first->protocol = protocol;
1168 
1169 	igc_tx_csum(tx_ring, first);
1170 
1171 	igc_tx_map(tx_ring, first, hdr_len);
1172 
1173 	return NETDEV_TX_OK;
1174 }
1175 
1176 static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter,
1177 						    struct sk_buff *skb)
1178 {
1179 	unsigned int r_idx = skb->queue_mapping;
1180 
1181 	if (r_idx >= adapter->num_tx_queues)
1182 		r_idx = r_idx % adapter->num_tx_queues;
1183 
1184 	return adapter->tx_ring[r_idx];
1185 }
1186 
1187 static netdev_tx_t igc_xmit_frame(struct sk_buff *skb,
1188 				  struct net_device *netdev)
1189 {
1190 	struct igc_adapter *adapter = netdev_priv(netdev);
1191 
1192 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
1193 	 * in order to meet this minimum size requirement.
1194 	 */
1195 	if (skb->len < 17) {
1196 		if (skb_padto(skb, 17))
1197 			return NETDEV_TX_OK;
1198 		skb->len = 17;
1199 	}
1200 
1201 	return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb));
1202 }
1203 
1204 static void igc_rx_checksum(struct igc_ring *ring,
1205 			    union igc_adv_rx_desc *rx_desc,
1206 			    struct sk_buff *skb)
1207 {
1208 	skb_checksum_none_assert(skb);
1209 
1210 	/* Ignore Checksum bit is set */
1211 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM))
1212 		return;
1213 
1214 	/* Rx checksum disabled via ethtool */
1215 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1216 		return;
1217 
1218 	/* TCP/UDP checksum error bit is set */
1219 	if (igc_test_staterr(rx_desc,
1220 			     IGC_RXDEXT_STATERR_TCPE |
1221 			     IGC_RXDEXT_STATERR_IPE)) {
1222 		/* work around errata with sctp packets where the TCPE aka
1223 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
1224 		 * packets (aka let the stack check the crc32c)
1225 		 */
1226 		if (!(skb->len == 60 &&
1227 		      test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
1228 			u64_stats_update_begin(&ring->rx_syncp);
1229 			ring->rx_stats.csum_err++;
1230 			u64_stats_update_end(&ring->rx_syncp);
1231 		}
1232 		/* let the stack verify checksum errors */
1233 		return;
1234 	}
1235 	/* It must be a TCP or UDP packet with a valid checksum */
1236 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS |
1237 				      IGC_RXD_STAT_UDPCS))
1238 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1239 
1240 	dev_dbg(ring->dev, "cksum success: bits %08X\n",
1241 		le32_to_cpu(rx_desc->wb.upper.status_error));
1242 }
1243 
1244 static inline void igc_rx_hash(struct igc_ring *ring,
1245 			       union igc_adv_rx_desc *rx_desc,
1246 			       struct sk_buff *skb)
1247 {
1248 	if (ring->netdev->features & NETIF_F_RXHASH)
1249 		skb_set_hash(skb,
1250 			     le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1251 			     PKT_HASH_TYPE_L3);
1252 }
1253 
1254 /**
1255  * igc_process_skb_fields - Populate skb header fields from Rx descriptor
1256  * @rx_ring: rx descriptor ring packet is being transacted on
1257  * @rx_desc: pointer to the EOP Rx descriptor
1258  * @skb: pointer to current skb being populated
1259  *
1260  * This function checks the ring, descriptor, and packet information in
1261  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1262  * other fields within the skb.
1263  */
1264 static void igc_process_skb_fields(struct igc_ring *rx_ring,
1265 				   union igc_adv_rx_desc *rx_desc,
1266 				   struct sk_buff *skb)
1267 {
1268 	igc_rx_hash(rx_ring, rx_desc, skb);
1269 
1270 	igc_rx_checksum(rx_ring, rx_desc, skb);
1271 
1272 	skb_record_rx_queue(skb, rx_ring->queue_index);
1273 
1274 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1275 }
1276 
1277 static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring,
1278 					       const unsigned int size)
1279 {
1280 	struct igc_rx_buffer *rx_buffer;
1281 
1282 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1283 	prefetchw(rx_buffer->page);
1284 
1285 	/* we are reusing so sync this buffer for CPU use */
1286 	dma_sync_single_range_for_cpu(rx_ring->dev,
1287 				      rx_buffer->dma,
1288 				      rx_buffer->page_offset,
1289 				      size,
1290 				      DMA_FROM_DEVICE);
1291 
1292 	rx_buffer->pagecnt_bias--;
1293 
1294 	return rx_buffer;
1295 }
1296 
1297 /**
1298  * igc_add_rx_frag - Add contents of Rx buffer to sk_buff
1299  * @rx_ring: rx descriptor ring to transact packets on
1300  * @rx_buffer: buffer containing page to add
1301  * @skb: sk_buff to place the data into
1302  * @size: size of buffer to be added
1303  *
1304  * This function will add the data contained in rx_buffer->page to the skb.
1305  */
1306 static void igc_add_rx_frag(struct igc_ring *rx_ring,
1307 			    struct igc_rx_buffer *rx_buffer,
1308 			    struct sk_buff *skb,
1309 			    unsigned int size)
1310 {
1311 #if (PAGE_SIZE < 8192)
1312 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1313 
1314 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1315 			rx_buffer->page_offset, size, truesize);
1316 	rx_buffer->page_offset ^= truesize;
1317 #else
1318 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1319 				SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1320 				SKB_DATA_ALIGN(size);
1321 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1322 			rx_buffer->page_offset, size, truesize);
1323 	rx_buffer->page_offset += truesize;
1324 #endif
1325 }
1326 
1327 static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring,
1328 				     struct igc_rx_buffer *rx_buffer,
1329 				     union igc_adv_rx_desc *rx_desc,
1330 				     unsigned int size)
1331 {
1332 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1333 #if (PAGE_SIZE < 8192)
1334 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1335 #else
1336 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1337 				SKB_DATA_ALIGN(IGC_SKB_PAD + size);
1338 #endif
1339 	struct sk_buff *skb;
1340 
1341 	/* prefetch first cache line of first page */
1342 	prefetch(va);
1343 #if L1_CACHE_BYTES < 128
1344 	prefetch(va + L1_CACHE_BYTES);
1345 #endif
1346 
1347 	/* build an skb around the page buffer */
1348 	skb = build_skb(va - IGC_SKB_PAD, truesize);
1349 	if (unlikely(!skb))
1350 		return NULL;
1351 
1352 	/* update pointers within the skb to store the data */
1353 	skb_reserve(skb, IGC_SKB_PAD);
1354 	__skb_put(skb, size);
1355 
1356 	/* update buffer offset */
1357 #if (PAGE_SIZE < 8192)
1358 	rx_buffer->page_offset ^= truesize;
1359 #else
1360 	rx_buffer->page_offset += truesize;
1361 #endif
1362 
1363 	return skb;
1364 }
1365 
1366 static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring,
1367 					 struct igc_rx_buffer *rx_buffer,
1368 					 union igc_adv_rx_desc *rx_desc,
1369 					 unsigned int size)
1370 {
1371 	void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1372 #if (PAGE_SIZE < 8192)
1373 	unsigned int truesize = igc_rx_pg_size(rx_ring) / 2;
1374 #else
1375 	unsigned int truesize = SKB_DATA_ALIGN(size);
1376 #endif
1377 	unsigned int headlen;
1378 	struct sk_buff *skb;
1379 
1380 	/* prefetch first cache line of first page */
1381 	prefetch(va);
1382 #if L1_CACHE_BYTES < 128
1383 	prefetch(va + L1_CACHE_BYTES);
1384 #endif
1385 
1386 	/* allocate a skb to store the frags */
1387 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGC_RX_HDR_LEN);
1388 	if (unlikely(!skb))
1389 		return NULL;
1390 
1391 	/* Determine available headroom for copy */
1392 	headlen = size;
1393 	if (headlen > IGC_RX_HDR_LEN)
1394 		headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN);
1395 
1396 	/* align pull length to size of long to optimize memcpy performance */
1397 	memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1398 
1399 	/* update all of the pointers */
1400 	size -= headlen;
1401 	if (size) {
1402 		skb_add_rx_frag(skb, 0, rx_buffer->page,
1403 				(va + headlen) - page_address(rx_buffer->page),
1404 				size, truesize);
1405 #if (PAGE_SIZE < 8192)
1406 		rx_buffer->page_offset ^= truesize;
1407 #else
1408 		rx_buffer->page_offset += truesize;
1409 #endif
1410 	} else {
1411 		rx_buffer->pagecnt_bias++;
1412 	}
1413 
1414 	return skb;
1415 }
1416 
1417 /**
1418  * igc_reuse_rx_page - page flip buffer and store it back on the ring
1419  * @rx_ring: rx descriptor ring to store buffers on
1420  * @old_buff: donor buffer to have page reused
1421  *
1422  * Synchronizes page for reuse by the adapter
1423  */
1424 static void igc_reuse_rx_page(struct igc_ring *rx_ring,
1425 			      struct igc_rx_buffer *old_buff)
1426 {
1427 	u16 nta = rx_ring->next_to_alloc;
1428 	struct igc_rx_buffer *new_buff;
1429 
1430 	new_buff = &rx_ring->rx_buffer_info[nta];
1431 
1432 	/* update, and store next to alloc */
1433 	nta++;
1434 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1435 
1436 	/* Transfer page from old buffer to new buffer.
1437 	 * Move each member individually to avoid possible store
1438 	 * forwarding stalls.
1439 	 */
1440 	new_buff->dma		= old_buff->dma;
1441 	new_buff->page		= old_buff->page;
1442 	new_buff->page_offset	= old_buff->page_offset;
1443 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1444 }
1445 
1446 static inline bool igc_page_is_reserved(struct page *page)
1447 {
1448 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1449 }
1450 
1451 static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer)
1452 {
1453 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1454 	struct page *page = rx_buffer->page;
1455 
1456 	/* avoid re-using remote pages */
1457 	if (unlikely(igc_page_is_reserved(page)))
1458 		return false;
1459 
1460 #if (PAGE_SIZE < 8192)
1461 	/* if we are only owner of page we can reuse it */
1462 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1463 		return false;
1464 #else
1465 #define IGC_LAST_OFFSET \
1466 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048)
1467 
1468 	if (rx_buffer->page_offset > IGC_LAST_OFFSET)
1469 		return false;
1470 #endif
1471 
1472 	/* If we have drained the page fragment pool we need to update
1473 	 * the pagecnt_bias and page count so that we fully restock the
1474 	 * number of references the driver holds.
1475 	 */
1476 	if (unlikely(!pagecnt_bias)) {
1477 		page_ref_add(page, USHRT_MAX);
1478 		rx_buffer->pagecnt_bias = USHRT_MAX;
1479 	}
1480 
1481 	return true;
1482 }
1483 
1484 /**
1485  * igc_is_non_eop - process handling of non-EOP buffers
1486  * @rx_ring: Rx ring being processed
1487  * @rx_desc: Rx descriptor for current buffer
1488  * @skb: current socket buffer containing buffer in progress
1489  *
1490  * This function updates next to clean.  If the buffer is an EOP buffer
1491  * this function exits returning false, otherwise it will place the
1492  * sk_buff in the next buffer to be chained and return true indicating
1493  * that this is in fact a non-EOP buffer.
1494  */
1495 static bool igc_is_non_eop(struct igc_ring *rx_ring,
1496 			   union igc_adv_rx_desc *rx_desc)
1497 {
1498 	u32 ntc = rx_ring->next_to_clean + 1;
1499 
1500 	/* fetch, update, and store next to clean */
1501 	ntc = (ntc < rx_ring->count) ? ntc : 0;
1502 	rx_ring->next_to_clean = ntc;
1503 
1504 	prefetch(IGC_RX_DESC(rx_ring, ntc));
1505 
1506 	if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP)))
1507 		return false;
1508 
1509 	return true;
1510 }
1511 
1512 /**
1513  * igc_cleanup_headers - Correct corrupted or empty headers
1514  * @rx_ring: rx descriptor ring packet is being transacted on
1515  * @rx_desc: pointer to the EOP Rx descriptor
1516  * @skb: pointer to current skb being fixed
1517  *
1518  * Address the case where we are pulling data in on pages only
1519  * and as such no data is present in the skb header.
1520  *
1521  * In addition if skb is not at least 60 bytes we need to pad it so that
1522  * it is large enough to qualify as a valid Ethernet frame.
1523  *
1524  * Returns true if an error was encountered and skb was freed.
1525  */
1526 static bool igc_cleanup_headers(struct igc_ring *rx_ring,
1527 				union igc_adv_rx_desc *rx_desc,
1528 				struct sk_buff *skb)
1529 {
1530 	if (unlikely((igc_test_staterr(rx_desc,
1531 				       IGC_RXDEXT_ERR_FRAME_ERR_MASK)))) {
1532 		struct net_device *netdev = rx_ring->netdev;
1533 
1534 		if (!(netdev->features & NETIF_F_RXALL)) {
1535 			dev_kfree_skb_any(skb);
1536 			return true;
1537 		}
1538 	}
1539 
1540 	/* if eth_skb_pad returns an error the skb was freed */
1541 	if (eth_skb_pad(skb))
1542 		return true;
1543 
1544 	return false;
1545 }
1546 
1547 static void igc_put_rx_buffer(struct igc_ring *rx_ring,
1548 			      struct igc_rx_buffer *rx_buffer)
1549 {
1550 	if (igc_can_reuse_rx_page(rx_buffer)) {
1551 		/* hand second half of page back to the ring */
1552 		igc_reuse_rx_page(rx_ring, rx_buffer);
1553 	} else {
1554 		/* We are not reusing the buffer so unmap it and free
1555 		 * any references we are holding to it
1556 		 */
1557 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
1558 				     igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
1559 				     IGC_RX_DMA_ATTR);
1560 		__page_frag_cache_drain(rx_buffer->page,
1561 					rx_buffer->pagecnt_bias);
1562 	}
1563 
1564 	/* clear contents of rx_buffer */
1565 	rx_buffer->page = NULL;
1566 }
1567 
1568 /**
1569  * igc_alloc_rx_buffers - Replace used receive buffers; packet split
1570  * @adapter: address of board private structure
1571  */
1572 static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
1573 {
1574 	union igc_adv_rx_desc *rx_desc;
1575 	u16 i = rx_ring->next_to_use;
1576 	struct igc_rx_buffer *bi;
1577 	u16 bufsz;
1578 
1579 	/* nothing to do */
1580 	if (!cleaned_count)
1581 		return;
1582 
1583 	rx_desc = IGC_RX_DESC(rx_ring, i);
1584 	bi = &rx_ring->rx_buffer_info[i];
1585 	i -= rx_ring->count;
1586 
1587 	bufsz = igc_rx_bufsz(rx_ring);
1588 
1589 	do {
1590 		if (!igc_alloc_mapped_page(rx_ring, bi))
1591 			break;
1592 
1593 		/* sync the buffer for use by the device */
1594 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1595 						 bi->page_offset, bufsz,
1596 						 DMA_FROM_DEVICE);
1597 
1598 		/* Refresh the desc even if buffer_addrs didn't change
1599 		 * because each write-back erases this info.
1600 		 */
1601 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1602 
1603 		rx_desc++;
1604 		bi++;
1605 		i++;
1606 		if (unlikely(!i)) {
1607 			rx_desc = IGC_RX_DESC(rx_ring, 0);
1608 			bi = rx_ring->rx_buffer_info;
1609 			i -= rx_ring->count;
1610 		}
1611 
1612 		/* clear the length for the next_to_use descriptor */
1613 		rx_desc->wb.upper.length = 0;
1614 
1615 		cleaned_count--;
1616 	} while (cleaned_count);
1617 
1618 	i += rx_ring->count;
1619 
1620 	if (rx_ring->next_to_use != i) {
1621 		/* record the next descriptor to use */
1622 		rx_ring->next_to_use = i;
1623 
1624 		/* update next to alloc since we have filled the ring */
1625 		rx_ring->next_to_alloc = i;
1626 
1627 		/* Force memory writes to complete before letting h/w
1628 		 * know there are new descriptors to fetch.  (Only
1629 		 * applicable for weak-ordered memory model archs,
1630 		 * such as IA-64).
1631 		 */
1632 		wmb();
1633 		writel(i, rx_ring->tail);
1634 	}
1635 }
1636 
1637 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
1638 {
1639 	unsigned int total_bytes = 0, total_packets = 0;
1640 	struct igc_ring *rx_ring = q_vector->rx.ring;
1641 	struct sk_buff *skb = rx_ring->skb;
1642 	u16 cleaned_count = igc_desc_unused(rx_ring);
1643 
1644 	while (likely(total_packets < budget)) {
1645 		union igc_adv_rx_desc *rx_desc;
1646 		struct igc_rx_buffer *rx_buffer;
1647 		unsigned int size;
1648 
1649 		/* return some buffers to hardware, one at a time is too slow */
1650 		if (cleaned_count >= IGC_RX_BUFFER_WRITE) {
1651 			igc_alloc_rx_buffers(rx_ring, cleaned_count);
1652 			cleaned_count = 0;
1653 		}
1654 
1655 		rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean);
1656 		size = le16_to_cpu(rx_desc->wb.upper.length);
1657 		if (!size)
1658 			break;
1659 
1660 		/* This memory barrier is needed to keep us from reading
1661 		 * any other fields out of the rx_desc until we know the
1662 		 * descriptor has been written back
1663 		 */
1664 		dma_rmb();
1665 
1666 		rx_buffer = igc_get_rx_buffer(rx_ring, size);
1667 
1668 		/* retrieve a buffer from the ring */
1669 		if (skb)
1670 			igc_add_rx_frag(rx_ring, rx_buffer, skb, size);
1671 		else if (ring_uses_build_skb(rx_ring))
1672 			skb = igc_build_skb(rx_ring, rx_buffer, rx_desc, size);
1673 		else
1674 			skb = igc_construct_skb(rx_ring, rx_buffer,
1675 						rx_desc, size);
1676 
1677 		/* exit if we failed to retrieve a buffer */
1678 		if (!skb) {
1679 			rx_ring->rx_stats.alloc_failed++;
1680 			rx_buffer->pagecnt_bias++;
1681 			break;
1682 		}
1683 
1684 		igc_put_rx_buffer(rx_ring, rx_buffer);
1685 		cleaned_count++;
1686 
1687 		/* fetch next buffer in frame if non-eop */
1688 		if (igc_is_non_eop(rx_ring, rx_desc))
1689 			continue;
1690 
1691 		/* verify the packet layout is correct */
1692 		if (igc_cleanup_headers(rx_ring, rx_desc, skb)) {
1693 			skb = NULL;
1694 			continue;
1695 		}
1696 
1697 		/* probably a little skewed due to removing CRC */
1698 		total_bytes += skb->len;
1699 
1700 		/* populate checksum, timestamp, VLAN, and protocol */
1701 		igc_process_skb_fields(rx_ring, rx_desc, skb);
1702 
1703 		napi_gro_receive(&q_vector->napi, skb);
1704 
1705 		/* reset skb pointer */
1706 		skb = NULL;
1707 
1708 		/* update budget accounting */
1709 		total_packets++;
1710 	}
1711 
1712 	/* place incomplete frames back on ring for completion */
1713 	rx_ring->skb = skb;
1714 
1715 	u64_stats_update_begin(&rx_ring->rx_syncp);
1716 	rx_ring->rx_stats.packets += total_packets;
1717 	rx_ring->rx_stats.bytes += total_bytes;
1718 	u64_stats_update_end(&rx_ring->rx_syncp);
1719 	q_vector->rx.total_packets += total_packets;
1720 	q_vector->rx.total_bytes += total_bytes;
1721 
1722 	if (cleaned_count)
1723 		igc_alloc_rx_buffers(rx_ring, cleaned_count);
1724 
1725 	return total_packets;
1726 }
1727 
1728 static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring)
1729 {
1730 	return ring_uses_build_skb(rx_ring) ? IGC_SKB_PAD : 0;
1731 }
1732 
1733 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
1734 				  struct igc_rx_buffer *bi)
1735 {
1736 	struct page *page = bi->page;
1737 	dma_addr_t dma;
1738 
1739 	/* since we are recycling buffers we should seldom need to alloc */
1740 	if (likely(page))
1741 		return true;
1742 
1743 	/* alloc new page for storage */
1744 	page = dev_alloc_pages(igc_rx_pg_order(rx_ring));
1745 	if (unlikely(!page)) {
1746 		rx_ring->rx_stats.alloc_failed++;
1747 		return false;
1748 	}
1749 
1750 	/* map page for use */
1751 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1752 				 igc_rx_pg_size(rx_ring),
1753 				 DMA_FROM_DEVICE,
1754 				 IGC_RX_DMA_ATTR);
1755 
1756 	/* if mapping failed free memory back to system since
1757 	 * there isn't much point in holding memory we can't use
1758 	 */
1759 	if (dma_mapping_error(rx_ring->dev, dma)) {
1760 		__free_page(page);
1761 
1762 		rx_ring->rx_stats.alloc_failed++;
1763 		return false;
1764 	}
1765 
1766 	bi->dma = dma;
1767 	bi->page = page;
1768 	bi->page_offset = igc_rx_offset(rx_ring);
1769 	bi->pagecnt_bias = 1;
1770 
1771 	return true;
1772 }
1773 
1774 /**
1775  * igc_clean_tx_irq - Reclaim resources after transmit completes
1776  * @q_vector: pointer to q_vector containing needed info
1777  * @napi_budget: Used to determine if we are in netpoll
1778  *
1779  * returns true if ring is completely cleaned
1780  */
1781 static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
1782 {
1783 	struct igc_adapter *adapter = q_vector->adapter;
1784 	unsigned int total_bytes = 0, total_packets = 0;
1785 	unsigned int budget = q_vector->tx.work_limit;
1786 	struct igc_ring *tx_ring = q_vector->tx.ring;
1787 	unsigned int i = tx_ring->next_to_clean;
1788 	struct igc_tx_buffer *tx_buffer;
1789 	union igc_adv_tx_desc *tx_desc;
1790 
1791 	if (test_bit(__IGC_DOWN, &adapter->state))
1792 		return true;
1793 
1794 	tx_buffer = &tx_ring->tx_buffer_info[i];
1795 	tx_desc = IGC_TX_DESC(tx_ring, i);
1796 	i -= tx_ring->count;
1797 
1798 	do {
1799 		union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1800 
1801 		/* if next_to_watch is not set then there is no work pending */
1802 		if (!eop_desc)
1803 			break;
1804 
1805 		/* prevent any other reads prior to eop_desc */
1806 		smp_rmb();
1807 
1808 		/* if DD is not set pending work has not been completed */
1809 		if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD)))
1810 			break;
1811 
1812 		/* clear next_to_watch to prevent false hangs */
1813 		tx_buffer->next_to_watch = NULL;
1814 
1815 		/* update the statistics for this packet */
1816 		total_bytes += tx_buffer->bytecount;
1817 		total_packets += tx_buffer->gso_segs;
1818 
1819 		/* free the skb */
1820 		napi_consume_skb(tx_buffer->skb, napi_budget);
1821 
1822 		/* unmap skb header data */
1823 		dma_unmap_single(tx_ring->dev,
1824 				 dma_unmap_addr(tx_buffer, dma),
1825 				 dma_unmap_len(tx_buffer, len),
1826 				 DMA_TO_DEVICE);
1827 
1828 		/* clear tx_buffer data */
1829 		dma_unmap_len_set(tx_buffer, len, 0);
1830 
1831 		/* clear last DMA location and unmap remaining buffers */
1832 		while (tx_desc != eop_desc) {
1833 			tx_buffer++;
1834 			tx_desc++;
1835 			i++;
1836 			if (unlikely(!i)) {
1837 				i -= tx_ring->count;
1838 				tx_buffer = tx_ring->tx_buffer_info;
1839 				tx_desc = IGC_TX_DESC(tx_ring, 0);
1840 			}
1841 
1842 			/* unmap any remaining paged data */
1843 			if (dma_unmap_len(tx_buffer, len)) {
1844 				dma_unmap_page(tx_ring->dev,
1845 					       dma_unmap_addr(tx_buffer, dma),
1846 					       dma_unmap_len(tx_buffer, len),
1847 					       DMA_TO_DEVICE);
1848 				dma_unmap_len_set(tx_buffer, len, 0);
1849 			}
1850 		}
1851 
1852 		/* move us one more past the eop_desc for start of next pkt */
1853 		tx_buffer++;
1854 		tx_desc++;
1855 		i++;
1856 		if (unlikely(!i)) {
1857 			i -= tx_ring->count;
1858 			tx_buffer = tx_ring->tx_buffer_info;
1859 			tx_desc = IGC_TX_DESC(tx_ring, 0);
1860 		}
1861 
1862 		/* issue prefetch for next Tx descriptor */
1863 		prefetch(tx_desc);
1864 
1865 		/* update budget accounting */
1866 		budget--;
1867 	} while (likely(budget));
1868 
1869 	netdev_tx_completed_queue(txring_txq(tx_ring),
1870 				  total_packets, total_bytes);
1871 
1872 	i += tx_ring->count;
1873 	tx_ring->next_to_clean = i;
1874 	u64_stats_update_begin(&tx_ring->tx_syncp);
1875 	tx_ring->tx_stats.bytes += total_bytes;
1876 	tx_ring->tx_stats.packets += total_packets;
1877 	u64_stats_update_end(&tx_ring->tx_syncp);
1878 	q_vector->tx.total_bytes += total_bytes;
1879 	q_vector->tx.total_packets += total_packets;
1880 
1881 	if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
1882 		struct igc_hw *hw = &adapter->hw;
1883 
1884 		/* Detect a transmit hang in hardware, this serializes the
1885 		 * check with the clearing of time_stamp and movement of i
1886 		 */
1887 		clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
1888 		if (tx_buffer->next_to_watch &&
1889 		    time_after(jiffies, tx_buffer->time_stamp +
1890 		    (adapter->tx_timeout_factor * HZ)) &&
1891 		    !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF)) {
1892 			/* detected Tx unit hang */
1893 			dev_err(tx_ring->dev,
1894 				"Detected Tx Unit Hang\n"
1895 				"  Tx Queue             <%d>\n"
1896 				"  TDH                  <%x>\n"
1897 				"  TDT                  <%x>\n"
1898 				"  next_to_use          <%x>\n"
1899 				"  next_to_clean        <%x>\n"
1900 				"buffer_info[next_to_clean]\n"
1901 				"  time_stamp           <%lx>\n"
1902 				"  next_to_watch        <%p>\n"
1903 				"  jiffies              <%lx>\n"
1904 				"  desc.status          <%x>\n",
1905 				tx_ring->queue_index,
1906 				rd32(IGC_TDH(tx_ring->reg_idx)),
1907 				readl(tx_ring->tail),
1908 				tx_ring->next_to_use,
1909 				tx_ring->next_to_clean,
1910 				tx_buffer->time_stamp,
1911 				tx_buffer->next_to_watch,
1912 				jiffies,
1913 				tx_buffer->next_to_watch->wb.status);
1914 			netif_stop_subqueue(tx_ring->netdev,
1915 					    tx_ring->queue_index);
1916 
1917 			/* we are about to reset, no point in enabling stuff */
1918 			return true;
1919 		}
1920 	}
1921 
1922 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1923 	if (unlikely(total_packets &&
1924 		     netif_carrier_ok(tx_ring->netdev) &&
1925 		     igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
1926 		/* Make sure that anybody stopping the queue after this
1927 		 * sees the new next_to_clean.
1928 		 */
1929 		smp_mb();
1930 		if (__netif_subqueue_stopped(tx_ring->netdev,
1931 					     tx_ring->queue_index) &&
1932 		    !(test_bit(__IGC_DOWN, &adapter->state))) {
1933 			netif_wake_subqueue(tx_ring->netdev,
1934 					    tx_ring->queue_index);
1935 
1936 			u64_stats_update_begin(&tx_ring->tx_syncp);
1937 			tx_ring->tx_stats.restart_queue++;
1938 			u64_stats_update_end(&tx_ring->tx_syncp);
1939 		}
1940 	}
1941 
1942 	return !!budget;
1943 }
1944 
1945 /**
1946  * igc_up - Open the interface and prepare it to handle traffic
1947  * @adapter: board private structure
1948  */
1949 void igc_up(struct igc_adapter *adapter)
1950 {
1951 	struct igc_hw *hw = &adapter->hw;
1952 	int i = 0;
1953 
1954 	/* hardware has been reset, we need to reload some things */
1955 	igc_configure(adapter);
1956 
1957 	clear_bit(__IGC_DOWN, &adapter->state);
1958 
1959 	for (i = 0; i < adapter->num_q_vectors; i++)
1960 		napi_enable(&adapter->q_vector[i]->napi);
1961 
1962 	if (adapter->msix_entries)
1963 		igc_configure_msix(adapter);
1964 	else
1965 		igc_assign_vector(adapter->q_vector[0], 0);
1966 
1967 	/* Clear any pending interrupts. */
1968 	rd32(IGC_ICR);
1969 	igc_irq_enable(adapter);
1970 
1971 	netif_tx_start_all_queues(adapter->netdev);
1972 
1973 	/* start the watchdog. */
1974 	hw->mac.get_link_status = 1;
1975 	schedule_work(&adapter->watchdog_task);
1976 }
1977 
1978 /**
1979  * igc_update_stats - Update the board statistics counters
1980  * @adapter: board private structure
1981  */
1982 void igc_update_stats(struct igc_adapter *adapter)
1983 {
1984 	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
1985 	struct pci_dev *pdev = adapter->pdev;
1986 	struct igc_hw *hw = &adapter->hw;
1987 	u64 _bytes, _packets;
1988 	u64 bytes, packets;
1989 	unsigned int start;
1990 	u32 mpc;
1991 	int i;
1992 
1993 	/* Prevent stats update while adapter is being reset, or if the pci
1994 	 * connection is down.
1995 	 */
1996 	if (adapter->link_speed == 0)
1997 		return;
1998 	if (pci_channel_offline(pdev))
1999 		return;
2000 
2001 	packets = 0;
2002 	bytes = 0;
2003 
2004 	rcu_read_lock();
2005 	for (i = 0; i < adapter->num_rx_queues; i++) {
2006 		struct igc_ring *ring = adapter->rx_ring[i];
2007 		u32 rqdpc = rd32(IGC_RQDPC(i));
2008 
2009 		if (hw->mac.type >= igc_i225)
2010 			wr32(IGC_RQDPC(i), 0);
2011 
2012 		if (rqdpc) {
2013 			ring->rx_stats.drops += rqdpc;
2014 			net_stats->rx_fifo_errors += rqdpc;
2015 		}
2016 
2017 		do {
2018 			start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
2019 			_bytes = ring->rx_stats.bytes;
2020 			_packets = ring->rx_stats.packets;
2021 		} while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
2022 		bytes += _bytes;
2023 		packets += _packets;
2024 	}
2025 
2026 	net_stats->rx_bytes = bytes;
2027 	net_stats->rx_packets = packets;
2028 
2029 	packets = 0;
2030 	bytes = 0;
2031 	for (i = 0; i < adapter->num_tx_queues; i++) {
2032 		struct igc_ring *ring = adapter->tx_ring[i];
2033 
2034 		do {
2035 			start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
2036 			_bytes = ring->tx_stats.bytes;
2037 			_packets = ring->tx_stats.packets;
2038 		} while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
2039 		bytes += _bytes;
2040 		packets += _packets;
2041 	}
2042 	net_stats->tx_bytes = bytes;
2043 	net_stats->tx_packets = packets;
2044 	rcu_read_unlock();
2045 
2046 	/* read stats registers */
2047 	adapter->stats.crcerrs += rd32(IGC_CRCERRS);
2048 	adapter->stats.gprc += rd32(IGC_GPRC);
2049 	adapter->stats.gorc += rd32(IGC_GORCL);
2050 	rd32(IGC_GORCH); /* clear GORCL */
2051 	adapter->stats.bprc += rd32(IGC_BPRC);
2052 	adapter->stats.mprc += rd32(IGC_MPRC);
2053 	adapter->stats.roc += rd32(IGC_ROC);
2054 
2055 	adapter->stats.prc64 += rd32(IGC_PRC64);
2056 	adapter->stats.prc127 += rd32(IGC_PRC127);
2057 	adapter->stats.prc255 += rd32(IGC_PRC255);
2058 	adapter->stats.prc511 += rd32(IGC_PRC511);
2059 	adapter->stats.prc1023 += rd32(IGC_PRC1023);
2060 	adapter->stats.prc1522 += rd32(IGC_PRC1522);
2061 	adapter->stats.symerrs += rd32(IGC_SYMERRS);
2062 	adapter->stats.sec += rd32(IGC_SEC);
2063 
2064 	mpc = rd32(IGC_MPC);
2065 	adapter->stats.mpc += mpc;
2066 	net_stats->rx_fifo_errors += mpc;
2067 	adapter->stats.scc += rd32(IGC_SCC);
2068 	adapter->stats.ecol += rd32(IGC_ECOL);
2069 	adapter->stats.mcc += rd32(IGC_MCC);
2070 	adapter->stats.latecol += rd32(IGC_LATECOL);
2071 	adapter->stats.dc += rd32(IGC_DC);
2072 	adapter->stats.rlec += rd32(IGC_RLEC);
2073 	adapter->stats.xonrxc += rd32(IGC_XONRXC);
2074 	adapter->stats.xontxc += rd32(IGC_XONTXC);
2075 	adapter->stats.xoffrxc += rd32(IGC_XOFFRXC);
2076 	adapter->stats.xofftxc += rd32(IGC_XOFFTXC);
2077 	adapter->stats.fcruc += rd32(IGC_FCRUC);
2078 	adapter->stats.gptc += rd32(IGC_GPTC);
2079 	adapter->stats.gotc += rd32(IGC_GOTCL);
2080 	rd32(IGC_GOTCH); /* clear GOTCL */
2081 	adapter->stats.rnbc += rd32(IGC_RNBC);
2082 	adapter->stats.ruc += rd32(IGC_RUC);
2083 	adapter->stats.rfc += rd32(IGC_RFC);
2084 	adapter->stats.rjc += rd32(IGC_RJC);
2085 	adapter->stats.tor += rd32(IGC_TORH);
2086 	adapter->stats.tot += rd32(IGC_TOTH);
2087 	adapter->stats.tpr += rd32(IGC_TPR);
2088 
2089 	adapter->stats.ptc64 += rd32(IGC_PTC64);
2090 	adapter->stats.ptc127 += rd32(IGC_PTC127);
2091 	adapter->stats.ptc255 += rd32(IGC_PTC255);
2092 	adapter->stats.ptc511 += rd32(IGC_PTC511);
2093 	adapter->stats.ptc1023 += rd32(IGC_PTC1023);
2094 	adapter->stats.ptc1522 += rd32(IGC_PTC1522);
2095 
2096 	adapter->stats.mptc += rd32(IGC_MPTC);
2097 	adapter->stats.bptc += rd32(IGC_BPTC);
2098 
2099 	adapter->stats.tpt += rd32(IGC_TPT);
2100 	adapter->stats.colc += rd32(IGC_COLC);
2101 
2102 	adapter->stats.algnerrc += rd32(IGC_ALGNERRC);
2103 
2104 	adapter->stats.tsctc += rd32(IGC_TSCTC);
2105 	adapter->stats.tsctfc += rd32(IGC_TSCTFC);
2106 
2107 	adapter->stats.iac += rd32(IGC_IAC);
2108 	adapter->stats.icrxoc += rd32(IGC_ICRXOC);
2109 	adapter->stats.icrxptc += rd32(IGC_ICRXPTC);
2110 	adapter->stats.icrxatc += rd32(IGC_ICRXATC);
2111 	adapter->stats.ictxptc += rd32(IGC_ICTXPTC);
2112 	adapter->stats.ictxatc += rd32(IGC_ICTXATC);
2113 	adapter->stats.ictxqec += rd32(IGC_ICTXQEC);
2114 	adapter->stats.ictxqmtc += rd32(IGC_ICTXQMTC);
2115 	adapter->stats.icrxdmtc += rd32(IGC_ICRXDMTC);
2116 
2117 	/* Fill out the OS statistics structure */
2118 	net_stats->multicast = adapter->stats.mprc;
2119 	net_stats->collisions = adapter->stats.colc;
2120 
2121 	/* Rx Errors */
2122 
2123 	/* RLEC on some newer hardware can be incorrect so build
2124 	 * our own version based on RUC and ROC
2125 	 */
2126 	net_stats->rx_errors = adapter->stats.rxerrc +
2127 		adapter->stats.crcerrs + adapter->stats.algnerrc +
2128 		adapter->stats.ruc + adapter->stats.roc +
2129 		adapter->stats.cexterr;
2130 	net_stats->rx_length_errors = adapter->stats.ruc +
2131 				      adapter->stats.roc;
2132 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
2133 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
2134 	net_stats->rx_missed_errors = adapter->stats.mpc;
2135 
2136 	/* Tx Errors */
2137 	net_stats->tx_errors = adapter->stats.ecol +
2138 			       adapter->stats.latecol;
2139 	net_stats->tx_aborted_errors = adapter->stats.ecol;
2140 	net_stats->tx_window_errors = adapter->stats.latecol;
2141 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
2142 
2143 	/* Tx Dropped needs to be maintained elsewhere */
2144 
2145 	/* Management Stats */
2146 	adapter->stats.mgptc += rd32(IGC_MGTPTC);
2147 	adapter->stats.mgprc += rd32(IGC_MGTPRC);
2148 	adapter->stats.mgpdc += rd32(IGC_MGTPDC);
2149 }
2150 
2151 static void igc_nfc_filter_exit(struct igc_adapter *adapter)
2152 {
2153 	struct igc_nfc_filter *rule;
2154 
2155 	spin_lock(&adapter->nfc_lock);
2156 
2157 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
2158 		igc_erase_filter(adapter, rule);
2159 
2160 	hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
2161 		igc_erase_filter(adapter, rule);
2162 
2163 	spin_unlock(&adapter->nfc_lock);
2164 }
2165 
2166 static void igc_nfc_filter_restore(struct igc_adapter *adapter)
2167 {
2168 	struct igc_nfc_filter *rule;
2169 
2170 	spin_lock(&adapter->nfc_lock);
2171 
2172 	hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
2173 		igc_add_filter(adapter, rule);
2174 
2175 	spin_unlock(&adapter->nfc_lock);
2176 }
2177 
2178 /**
2179  * igc_down - Close the interface
2180  * @adapter: board private structure
2181  */
2182 void igc_down(struct igc_adapter *adapter)
2183 {
2184 	struct net_device *netdev = adapter->netdev;
2185 	struct igc_hw *hw = &adapter->hw;
2186 	u32 tctl, rctl;
2187 	int i = 0;
2188 
2189 	set_bit(__IGC_DOWN, &adapter->state);
2190 
2191 	/* disable receives in the hardware */
2192 	rctl = rd32(IGC_RCTL);
2193 	wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN);
2194 	/* flush and sleep below */
2195 
2196 	igc_nfc_filter_exit(adapter);
2197 
2198 	/* set trans_start so we don't get spurious watchdogs during reset */
2199 	netif_trans_update(netdev);
2200 
2201 	netif_carrier_off(netdev);
2202 	netif_tx_stop_all_queues(netdev);
2203 
2204 	/* disable transmits in the hardware */
2205 	tctl = rd32(IGC_TCTL);
2206 	tctl &= ~IGC_TCTL_EN;
2207 	wr32(IGC_TCTL, tctl);
2208 	/* flush both disables and wait for them to finish */
2209 	wrfl();
2210 	usleep_range(10000, 20000);
2211 
2212 	igc_irq_disable(adapter);
2213 
2214 	adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
2215 
2216 	for (i = 0; i < adapter->num_q_vectors; i++) {
2217 		if (adapter->q_vector[i]) {
2218 			napi_synchronize(&adapter->q_vector[i]->napi);
2219 			napi_disable(&adapter->q_vector[i]->napi);
2220 		}
2221 	}
2222 
2223 	del_timer_sync(&adapter->watchdog_timer);
2224 	del_timer_sync(&adapter->phy_info_timer);
2225 
2226 	/* record the stats before reset*/
2227 	spin_lock(&adapter->stats64_lock);
2228 	igc_update_stats(adapter);
2229 	spin_unlock(&adapter->stats64_lock);
2230 
2231 	adapter->link_speed = 0;
2232 	adapter->link_duplex = 0;
2233 
2234 	if (!pci_channel_offline(adapter->pdev))
2235 		igc_reset(adapter);
2236 
2237 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
2238 	adapter->flags &= ~IGC_FLAG_VLAN_PROMISC;
2239 
2240 	igc_clean_all_tx_rings(adapter);
2241 	igc_clean_all_rx_rings(adapter);
2242 }
2243 
2244 void igc_reinit_locked(struct igc_adapter *adapter)
2245 {
2246 	WARN_ON(in_interrupt());
2247 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
2248 		usleep_range(1000, 2000);
2249 	igc_down(adapter);
2250 	igc_up(adapter);
2251 	clear_bit(__IGC_RESETTING, &adapter->state);
2252 }
2253 
2254 static void igc_reset_task(struct work_struct *work)
2255 {
2256 	struct igc_adapter *adapter;
2257 
2258 	adapter = container_of(work, struct igc_adapter, reset_task);
2259 
2260 	netdev_err(adapter->netdev, "Reset adapter\n");
2261 	igc_reinit_locked(adapter);
2262 }
2263 
2264 /**
2265  * igc_change_mtu - Change the Maximum Transfer Unit
2266  * @netdev: network interface device structure
2267  * @new_mtu: new value for maximum frame size
2268  *
2269  * Returns 0 on success, negative on failure
2270  */
2271 static int igc_change_mtu(struct net_device *netdev, int new_mtu)
2272 {
2273 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
2274 	struct igc_adapter *adapter = netdev_priv(netdev);
2275 	struct pci_dev *pdev = adapter->pdev;
2276 
2277 	/* adjust max frame to be at least the size of a standard frame */
2278 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
2279 		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
2280 
2281 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
2282 		usleep_range(1000, 2000);
2283 
2284 	/* igc_down has a dependency on max_frame_size */
2285 	adapter->max_frame_size = max_frame;
2286 
2287 	if (netif_running(netdev))
2288 		igc_down(adapter);
2289 
2290 	dev_info(&pdev->dev, "changing MTU from %d to %d\n",
2291 		 netdev->mtu, new_mtu);
2292 	netdev->mtu = new_mtu;
2293 
2294 	if (netif_running(netdev))
2295 		igc_up(adapter);
2296 	else
2297 		igc_reset(adapter);
2298 
2299 	clear_bit(__IGC_RESETTING, &adapter->state);
2300 
2301 	return 0;
2302 }
2303 
2304 /**
2305  * igc_get_stats - Get System Network Statistics
2306  * @netdev: network interface device structure
2307  *
2308  * Returns the address of the device statistics structure.
2309  * The statistics are updated here and also from the timer callback.
2310  */
2311 static struct net_device_stats *igc_get_stats(struct net_device *netdev)
2312 {
2313 	struct igc_adapter *adapter = netdev_priv(netdev);
2314 
2315 	if (!test_bit(__IGC_RESETTING, &adapter->state))
2316 		igc_update_stats(adapter);
2317 
2318 	/* only return the current stats */
2319 	return &netdev->stats;
2320 }
2321 
2322 static netdev_features_t igc_fix_features(struct net_device *netdev,
2323 					  netdev_features_t features)
2324 {
2325 	/* Since there is no support for separate Rx/Tx vlan accel
2326 	 * enable/disable make sure Tx flag is always in same state as Rx.
2327 	 */
2328 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
2329 		features |= NETIF_F_HW_VLAN_CTAG_TX;
2330 	else
2331 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2332 
2333 	return features;
2334 }
2335 
2336 static int igc_set_features(struct net_device *netdev,
2337 			    netdev_features_t features)
2338 {
2339 	netdev_features_t changed = netdev->features ^ features;
2340 	struct igc_adapter *adapter = netdev_priv(netdev);
2341 
2342 	/* Add VLAN support */
2343 	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2344 		return 0;
2345 
2346 	if (!(features & NETIF_F_NTUPLE)) {
2347 		struct hlist_node *node2;
2348 		struct igc_nfc_filter *rule;
2349 
2350 		spin_lock(&adapter->nfc_lock);
2351 		hlist_for_each_entry_safe(rule, node2,
2352 					  &adapter->nfc_filter_list, nfc_node) {
2353 			igc_erase_filter(adapter, rule);
2354 			hlist_del(&rule->nfc_node);
2355 			kfree(rule);
2356 		}
2357 		spin_unlock(&adapter->nfc_lock);
2358 		adapter->nfc_filter_count = 0;
2359 	}
2360 
2361 	netdev->features = features;
2362 
2363 	if (netif_running(netdev))
2364 		igc_reinit_locked(adapter);
2365 	else
2366 		igc_reset(adapter);
2367 
2368 	return 1;
2369 }
2370 
2371 static netdev_features_t
2372 igc_features_check(struct sk_buff *skb, struct net_device *dev,
2373 		   netdev_features_t features)
2374 {
2375 	unsigned int network_hdr_len, mac_hdr_len;
2376 
2377 	/* Make certain the headers can be described by a context descriptor */
2378 	mac_hdr_len = skb_network_header(skb) - skb->data;
2379 	if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN))
2380 		return features & ~(NETIF_F_HW_CSUM |
2381 				    NETIF_F_SCTP_CRC |
2382 				    NETIF_F_HW_VLAN_CTAG_TX |
2383 				    NETIF_F_TSO |
2384 				    NETIF_F_TSO6);
2385 
2386 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2387 	if (unlikely(network_hdr_len >  IGC_MAX_NETWORK_HDR_LEN))
2388 		return features & ~(NETIF_F_HW_CSUM |
2389 				    NETIF_F_SCTP_CRC |
2390 				    NETIF_F_TSO |
2391 				    NETIF_F_TSO6);
2392 
2393 	/* We can only support IPv4 TSO in tunnels if we can mangle the
2394 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2395 	 */
2396 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2397 		features &= ~NETIF_F_TSO;
2398 
2399 	return features;
2400 }
2401 
2402 /**
2403  * igc_configure - configure the hardware for RX and TX
2404  * @adapter: private board structure
2405  */
2406 static void igc_configure(struct igc_adapter *adapter)
2407 {
2408 	struct net_device *netdev = adapter->netdev;
2409 	int i = 0;
2410 
2411 	igc_get_hw_control(adapter);
2412 	igc_set_rx_mode(netdev);
2413 
2414 	igc_setup_tctl(adapter);
2415 	igc_setup_mrqc(adapter);
2416 	igc_setup_rctl(adapter);
2417 
2418 	igc_nfc_filter_restore(adapter);
2419 	igc_configure_tx(adapter);
2420 	igc_configure_rx(adapter);
2421 
2422 	igc_rx_fifo_flush_base(&adapter->hw);
2423 
2424 	/* call igc_desc_unused which always leaves
2425 	 * at least 1 descriptor unused to make sure
2426 	 * next_to_use != next_to_clean
2427 	 */
2428 	for (i = 0; i < adapter->num_rx_queues; i++) {
2429 		struct igc_ring *ring = adapter->rx_ring[i];
2430 
2431 		igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
2432 	}
2433 }
2434 
2435 /**
2436  * igc_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
2437  * @adapter: address of board private structure
2438  * @index: Index of the RAR entry which need to be synced with MAC table
2439  */
2440 static void igc_rar_set_index(struct igc_adapter *adapter, u32 index)
2441 {
2442 	u8 *addr = adapter->mac_table[index].addr;
2443 	struct igc_hw *hw = &adapter->hw;
2444 	u32 rar_low, rar_high;
2445 
2446 	/* HW expects these to be in network order when they are plugged
2447 	 * into the registers which are little endian.  In order to guarantee
2448 	 * that ordering we need to do an leXX_to_cpup here in order to be
2449 	 * ready for the byteswap that occurs with writel
2450 	 */
2451 	rar_low = le32_to_cpup((__le32 *)(addr));
2452 	rar_high = le16_to_cpup((__le16 *)(addr + 4));
2453 
2454 	/* Indicate to hardware the Address is Valid. */
2455 	if (adapter->mac_table[index].state & IGC_MAC_STATE_IN_USE) {
2456 		if (is_valid_ether_addr(addr))
2457 			rar_high |= IGC_RAH_AV;
2458 
2459 		rar_high |= IGC_RAH_POOL_1 <<
2460 			adapter->mac_table[index].queue;
2461 	}
2462 
2463 	wr32(IGC_RAL(index), rar_low);
2464 	wrfl();
2465 	wr32(IGC_RAH(index), rar_high);
2466 	wrfl();
2467 }
2468 
2469 /* Set default MAC address for the PF in the first RAR entry */
2470 static void igc_set_default_mac_filter(struct igc_adapter *adapter)
2471 {
2472 	struct igc_mac_addr *mac_table = &adapter->mac_table[0];
2473 
2474 	ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
2475 	mac_table->state = IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
2476 
2477 	igc_rar_set_index(adapter, 0);
2478 }
2479 
2480 /* If the filter to be added and an already existing filter express
2481  * the same address and address type, it should be possible to only
2482  * override the other configurations, for example the queue to steer
2483  * traffic.
2484  */
2485 static bool igc_mac_entry_can_be_used(const struct igc_mac_addr *entry,
2486 				      const u8 *addr, const u8 flags)
2487 {
2488 	if (!(entry->state & IGC_MAC_STATE_IN_USE))
2489 		return true;
2490 
2491 	if ((entry->state & IGC_MAC_STATE_SRC_ADDR) !=
2492 	    (flags & IGC_MAC_STATE_SRC_ADDR))
2493 		return false;
2494 
2495 	if (!ether_addr_equal(addr, entry->addr))
2496 		return false;
2497 
2498 	return true;
2499 }
2500 
2501 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
2502  * 'flags' is used to indicate what kind of match is made, match is by
2503  * default for the destination address, if matching by source address
2504  * is desired the flag IGC_MAC_STATE_SRC_ADDR can be used.
2505  */
2506 static int igc_add_mac_filter_flags(struct igc_adapter *adapter,
2507 				    const u8 *addr, const u8 queue,
2508 				    const u8 flags)
2509 {
2510 	struct igc_hw *hw = &adapter->hw;
2511 	int rar_entries = hw->mac.rar_entry_count;
2512 	int i;
2513 
2514 	if (is_zero_ether_addr(addr))
2515 		return -EINVAL;
2516 
2517 	/* Search for the first empty entry in the MAC table.
2518 	 * Do not touch entries at the end of the table reserved for the VF MAC
2519 	 * addresses.
2520 	 */
2521 	for (i = 0; i < rar_entries; i++) {
2522 		if (!igc_mac_entry_can_be_used(&adapter->mac_table[i],
2523 					       addr, flags))
2524 			continue;
2525 
2526 		ether_addr_copy(adapter->mac_table[i].addr, addr);
2527 		adapter->mac_table[i].queue = queue;
2528 		adapter->mac_table[i].state |= IGC_MAC_STATE_IN_USE | flags;
2529 
2530 		igc_rar_set_index(adapter, i);
2531 		return i;
2532 	}
2533 
2534 	return -ENOSPC;
2535 }
2536 
2537 int igc_add_mac_steering_filter(struct igc_adapter *adapter,
2538 				const u8 *addr, u8 queue, u8 flags)
2539 {
2540 	return igc_add_mac_filter_flags(adapter, addr, queue,
2541 					IGC_MAC_STATE_QUEUE_STEERING | flags);
2542 }
2543 
2544 /* Remove a MAC filter for 'addr' directing matching traffic to
2545  * 'queue', 'flags' is used to indicate what kind of match need to be
2546  * removed, match is by default for the destination address, if
2547  * matching by source address is to be removed the flag
2548  * IGC_MAC_STATE_SRC_ADDR can be used.
2549  */
2550 static int igc_del_mac_filter_flags(struct igc_adapter *adapter,
2551 				    const u8 *addr, const u8 queue,
2552 				    const u8 flags)
2553 {
2554 	struct igc_hw *hw = &adapter->hw;
2555 	int rar_entries = hw->mac.rar_entry_count;
2556 	int i;
2557 
2558 	if (is_zero_ether_addr(addr))
2559 		return -EINVAL;
2560 
2561 	/* Search for matching entry in the MAC table based on given address
2562 	 * and queue. Do not touch entries at the end of the table reserved
2563 	 * for the VF MAC addresses.
2564 	 */
2565 	for (i = 0; i < rar_entries; i++) {
2566 		if (!(adapter->mac_table[i].state & IGC_MAC_STATE_IN_USE))
2567 			continue;
2568 		if ((adapter->mac_table[i].state & flags) != flags)
2569 			continue;
2570 		if (adapter->mac_table[i].queue != queue)
2571 			continue;
2572 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
2573 			continue;
2574 
2575 		/* When a filter for the default address is "deleted",
2576 		 * we return it to its initial configuration
2577 		 */
2578 		if (adapter->mac_table[i].state & IGC_MAC_STATE_DEFAULT) {
2579 			adapter->mac_table[i].state =
2580 				IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
2581 		} else {
2582 			adapter->mac_table[i].state = 0;
2583 			adapter->mac_table[i].queue = 0;
2584 			memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
2585 		}
2586 
2587 		igc_rar_set_index(adapter, i);
2588 		return 0;
2589 	}
2590 
2591 	return -ENOENT;
2592 }
2593 
2594 int igc_del_mac_steering_filter(struct igc_adapter *adapter,
2595 				const u8 *addr, u8 queue, u8 flags)
2596 {
2597 	return igc_del_mac_filter_flags(adapter, addr, queue,
2598 					IGC_MAC_STATE_QUEUE_STEERING | flags);
2599 }
2600 
2601 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
2602  * 'flags' is used to indicate what kind of match is made, match is by
2603  * default for the destination address, if matching by source address
2604  * is desired the flag IGC_MAC_STATE_SRC_ADDR can be used.
2605  */
2606 static int igc_add_mac_filter(struct igc_adapter *adapter,
2607 			      const u8 *addr, const u8 queue)
2608 {
2609 	struct igc_hw *hw = &adapter->hw;
2610 	int rar_entries = hw->mac.rar_entry_count;
2611 	int i;
2612 
2613 	if (is_zero_ether_addr(addr))
2614 		return -EINVAL;
2615 
2616 	/* Search for the first empty entry in the MAC table.
2617 	 * Do not touch entries at the end of the table reserved for the VF MAC
2618 	 * addresses.
2619 	 */
2620 	for (i = 0; i < rar_entries; i++) {
2621 		if (!igc_mac_entry_can_be_used(&adapter->mac_table[i],
2622 					       addr, 0))
2623 			continue;
2624 
2625 		ether_addr_copy(adapter->mac_table[i].addr, addr);
2626 		adapter->mac_table[i].queue = queue;
2627 		adapter->mac_table[i].state |= IGC_MAC_STATE_IN_USE;
2628 
2629 		igc_rar_set_index(adapter, i);
2630 		return i;
2631 	}
2632 
2633 	return -ENOSPC;
2634 }
2635 
2636 /* Remove a MAC filter for 'addr' directing matching traffic to
2637  * 'queue', 'flags' is used to indicate what kind of match need to be
2638  * removed, match is by default for the destination address, if
2639  * matching by source address is to be removed the flag
2640  * IGC_MAC_STATE_SRC_ADDR can be used.
2641  */
2642 static int igc_del_mac_filter(struct igc_adapter *adapter,
2643 			      const u8 *addr, const u8 queue)
2644 {
2645 	struct igc_hw *hw = &adapter->hw;
2646 	int rar_entries = hw->mac.rar_entry_count;
2647 	int i;
2648 
2649 	if (is_zero_ether_addr(addr))
2650 		return -EINVAL;
2651 
2652 	/* Search for matching entry in the MAC table based on given address
2653 	 * and queue. Do not touch entries at the end of the table reserved
2654 	 * for the VF MAC addresses.
2655 	 */
2656 	for (i = 0; i < rar_entries; i++) {
2657 		if (!(adapter->mac_table[i].state & IGC_MAC_STATE_IN_USE))
2658 			continue;
2659 		if (adapter->mac_table[i].state != 0)
2660 			continue;
2661 		if (adapter->mac_table[i].queue != queue)
2662 			continue;
2663 		if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
2664 			continue;
2665 
2666 		/* When a filter for the default address is "deleted",
2667 		 * we return it to its initial configuration
2668 		 */
2669 		if (adapter->mac_table[i].state & IGC_MAC_STATE_DEFAULT) {
2670 			adapter->mac_table[i].state =
2671 				IGC_MAC_STATE_DEFAULT | IGC_MAC_STATE_IN_USE;
2672 			adapter->mac_table[i].queue = 0;
2673 		} else {
2674 			adapter->mac_table[i].state = 0;
2675 			adapter->mac_table[i].queue = 0;
2676 			memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
2677 		}
2678 
2679 		igc_rar_set_index(adapter, i);
2680 		return 0;
2681 	}
2682 
2683 	return -ENOENT;
2684 }
2685 
2686 static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr)
2687 {
2688 	struct igc_adapter *adapter = netdev_priv(netdev);
2689 	int ret;
2690 
2691 	ret = igc_add_mac_filter(adapter, addr, adapter->num_rx_queues);
2692 
2693 	return min_t(int, ret, 0);
2694 }
2695 
2696 static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr)
2697 {
2698 	struct igc_adapter *adapter = netdev_priv(netdev);
2699 
2700 	igc_del_mac_filter(adapter, addr, adapter->num_rx_queues);
2701 
2702 	return 0;
2703 }
2704 
2705 /**
2706  * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2707  * @netdev: network interface device structure
2708  *
2709  * The set_rx_mode entry point is called whenever the unicast or multicast
2710  * address lists or the network interface flags are updated.  This routine is
2711  * responsible for configuring the hardware for proper unicast, multicast,
2712  * promiscuous mode, and all-multi behavior.
2713  */
2714 static void igc_set_rx_mode(struct net_device *netdev)
2715 {
2716 	struct igc_adapter *adapter = netdev_priv(netdev);
2717 	struct igc_hw *hw = &adapter->hw;
2718 	u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
2719 	int count;
2720 
2721 	/* Check for Promiscuous and All Multicast modes */
2722 	if (netdev->flags & IFF_PROMISC) {
2723 		rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE;
2724 	} else {
2725 		if (netdev->flags & IFF_ALLMULTI) {
2726 			rctl |= IGC_RCTL_MPE;
2727 		} else {
2728 			/* Write addresses to the MTA, if the attempt fails
2729 			 * then we should just turn on promiscuous mode so
2730 			 * that we can at least receive multicast traffic
2731 			 */
2732 			count = igc_write_mc_addr_list(netdev);
2733 			if (count < 0)
2734 				rctl |= IGC_RCTL_MPE;
2735 		}
2736 	}
2737 
2738 	/* Write addresses to available RAR registers, if there is not
2739 	 * sufficient space to store all the addresses then enable
2740 	 * unicast promiscuous mode
2741 	 */
2742 	if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync))
2743 		rctl |= IGC_RCTL_UPE;
2744 
2745 	/* update state of unicast and multicast */
2746 	rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
2747 	wr32(IGC_RCTL, rctl);
2748 
2749 #if (PAGE_SIZE < 8192)
2750 	if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB)
2751 		rlpml = IGC_MAX_FRAME_BUILD_SKB;
2752 #endif
2753 	wr32(IGC_RLPML, rlpml);
2754 }
2755 
2756 /**
2757  * igc_msix_other - msix other interrupt handler
2758  * @irq: interrupt number
2759  * @data: pointer to a q_vector
2760  */
2761 static irqreturn_t igc_msix_other(int irq, void *data)
2762 {
2763 	struct igc_adapter *adapter = data;
2764 	struct igc_hw *hw = &adapter->hw;
2765 	u32 icr = rd32(IGC_ICR);
2766 
2767 	/* reading ICR causes bit 31 of EICR to be cleared */
2768 	if (icr & IGC_ICR_DRSTA)
2769 		schedule_work(&adapter->reset_task);
2770 
2771 	if (icr & IGC_ICR_DOUTSYNC) {
2772 		/* HW is reporting DMA is out of sync */
2773 		adapter->stats.doosync++;
2774 	}
2775 
2776 	if (icr & IGC_ICR_LSC) {
2777 		hw->mac.get_link_status = 1;
2778 		/* guard against interrupt when we're going down */
2779 		if (!test_bit(__IGC_DOWN, &adapter->state))
2780 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
2781 	}
2782 
2783 	wr32(IGC_EIMS, adapter->eims_other);
2784 
2785 	return IRQ_HANDLED;
2786 }
2787 
2788 /**
2789  * igc_write_ivar - configure ivar for given MSI-X vector
2790  * @hw: pointer to the HW structure
2791  * @msix_vector: vector number we are allocating to a given ring
2792  * @index: row index of IVAR register to write within IVAR table
2793  * @offset: column offset of in IVAR, should be multiple of 8
2794  *
2795  * The IVAR table consists of 2 columns,
2796  * each containing an cause allocation for an Rx and Tx ring, and a
2797  * variable number of rows depending on the number of queues supported.
2798  */
2799 static void igc_write_ivar(struct igc_hw *hw, int msix_vector,
2800 			   int index, int offset)
2801 {
2802 	u32 ivar = array_rd32(IGC_IVAR0, index);
2803 
2804 	/* clear any bits that are currently set */
2805 	ivar &= ~((u32)0xFF << offset);
2806 
2807 	/* write vector and valid bit */
2808 	ivar |= (msix_vector | IGC_IVAR_VALID) << offset;
2809 
2810 	array_wr32(IGC_IVAR0, index, ivar);
2811 }
2812 
2813 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector)
2814 {
2815 	struct igc_adapter *adapter = q_vector->adapter;
2816 	struct igc_hw *hw = &adapter->hw;
2817 	int rx_queue = IGC_N0_QUEUE;
2818 	int tx_queue = IGC_N0_QUEUE;
2819 
2820 	if (q_vector->rx.ring)
2821 		rx_queue = q_vector->rx.ring->reg_idx;
2822 	if (q_vector->tx.ring)
2823 		tx_queue = q_vector->tx.ring->reg_idx;
2824 
2825 	switch (hw->mac.type) {
2826 	case igc_i225:
2827 		if (rx_queue > IGC_N0_QUEUE)
2828 			igc_write_ivar(hw, msix_vector,
2829 				       rx_queue >> 1,
2830 				       (rx_queue & 0x1) << 4);
2831 		if (tx_queue > IGC_N0_QUEUE)
2832 			igc_write_ivar(hw, msix_vector,
2833 				       tx_queue >> 1,
2834 				       ((tx_queue & 0x1) << 4) + 8);
2835 		q_vector->eims_value = BIT(msix_vector);
2836 		break;
2837 	default:
2838 		WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n");
2839 		break;
2840 	}
2841 
2842 	/* add q_vector eims value to global eims_enable_mask */
2843 	adapter->eims_enable_mask |= q_vector->eims_value;
2844 
2845 	/* configure q_vector to set itr on first interrupt */
2846 	q_vector->set_itr = 1;
2847 }
2848 
2849 /**
2850  * igc_configure_msix - Configure MSI-X hardware
2851  * @adapter: Pointer to adapter structure
2852  *
2853  * igc_configure_msix sets up the hardware to properly
2854  * generate MSI-X interrupts.
2855  */
2856 static void igc_configure_msix(struct igc_adapter *adapter)
2857 {
2858 	struct igc_hw *hw = &adapter->hw;
2859 	int i, vector = 0;
2860 	u32 tmp;
2861 
2862 	adapter->eims_enable_mask = 0;
2863 
2864 	/* set vector for other causes, i.e. link changes */
2865 	switch (hw->mac.type) {
2866 	case igc_i225:
2867 		/* Turn on MSI-X capability first, or our settings
2868 		 * won't stick.  And it will take days to debug.
2869 		 */
2870 		wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE |
2871 		     IGC_GPIE_PBA | IGC_GPIE_EIAME |
2872 		     IGC_GPIE_NSICR);
2873 
2874 		/* enable msix_other interrupt */
2875 		adapter->eims_other = BIT(vector);
2876 		tmp = (vector++ | IGC_IVAR_VALID) << 8;
2877 
2878 		wr32(IGC_IVAR_MISC, tmp);
2879 		break;
2880 	default:
2881 		/* do nothing, since nothing else supports MSI-X */
2882 		break;
2883 	} /* switch (hw->mac.type) */
2884 
2885 	adapter->eims_enable_mask |= adapter->eims_other;
2886 
2887 	for (i = 0; i < adapter->num_q_vectors; i++)
2888 		igc_assign_vector(adapter->q_vector[i], vector++);
2889 
2890 	wrfl();
2891 }
2892 
2893 static irqreturn_t igc_msix_ring(int irq, void *data)
2894 {
2895 	struct igc_q_vector *q_vector = data;
2896 
2897 	/* Write the ITR value calculated from the previous interrupt. */
2898 	igc_write_itr(q_vector);
2899 
2900 	napi_schedule(&q_vector->napi);
2901 
2902 	return IRQ_HANDLED;
2903 }
2904 
2905 /**
2906  * igc_request_msix - Initialize MSI-X interrupts
2907  * @adapter: Pointer to adapter structure
2908  *
2909  * igc_request_msix allocates MSI-X vectors and requests interrupts from the
2910  * kernel.
2911  */
2912 static int igc_request_msix(struct igc_adapter *adapter)
2913 {
2914 	int i = 0, err = 0, vector = 0, free_vector = 0;
2915 	struct net_device *netdev = adapter->netdev;
2916 
2917 	err = request_irq(adapter->msix_entries[vector].vector,
2918 			  &igc_msix_other, 0, netdev->name, adapter);
2919 	if (err)
2920 		goto err_out;
2921 
2922 	for (i = 0; i < adapter->num_q_vectors; i++) {
2923 		struct igc_q_vector *q_vector = adapter->q_vector[i];
2924 
2925 		vector++;
2926 
2927 		q_vector->itr_register = adapter->io_addr + IGC_EITR(vector);
2928 
2929 		if (q_vector->rx.ring && q_vector->tx.ring)
2930 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
2931 				q_vector->rx.ring->queue_index);
2932 		else if (q_vector->tx.ring)
2933 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
2934 				q_vector->tx.ring->queue_index);
2935 		else if (q_vector->rx.ring)
2936 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
2937 				q_vector->rx.ring->queue_index);
2938 		else
2939 			sprintf(q_vector->name, "%s-unused", netdev->name);
2940 
2941 		err = request_irq(adapter->msix_entries[vector].vector,
2942 				  igc_msix_ring, 0, q_vector->name,
2943 				  q_vector);
2944 		if (err)
2945 			goto err_free;
2946 	}
2947 
2948 	igc_configure_msix(adapter);
2949 	return 0;
2950 
2951 err_free:
2952 	/* free already assigned IRQs */
2953 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
2954 
2955 	vector--;
2956 	for (i = 0; i < vector; i++) {
2957 		free_irq(adapter->msix_entries[free_vector++].vector,
2958 			 adapter->q_vector[i]);
2959 	}
2960 err_out:
2961 	return err;
2962 }
2963 
2964 /**
2965  * igc_reset_q_vector - Reset config for interrupt vector
2966  * @adapter: board private structure to initialize
2967  * @v_idx: Index of vector to be reset
2968  *
2969  * If NAPI is enabled it will delete any references to the
2970  * NAPI struct. This is preparation for igc_free_q_vector.
2971  */
2972 static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx)
2973 {
2974 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
2975 
2976 	/* if we're coming from igc_set_interrupt_capability, the vectors are
2977 	 * not yet allocated
2978 	 */
2979 	if (!q_vector)
2980 		return;
2981 
2982 	if (q_vector->tx.ring)
2983 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
2984 
2985 	if (q_vector->rx.ring)
2986 		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
2987 
2988 	netif_napi_del(&q_vector->napi);
2989 }
2990 
2991 static void igc_reset_interrupt_capability(struct igc_adapter *adapter)
2992 {
2993 	int v_idx = adapter->num_q_vectors;
2994 
2995 	if (adapter->msix_entries) {
2996 		pci_disable_msix(adapter->pdev);
2997 		kfree(adapter->msix_entries);
2998 		adapter->msix_entries = NULL;
2999 	} else if (adapter->flags & IGC_FLAG_HAS_MSI) {
3000 		pci_disable_msi(adapter->pdev);
3001 	}
3002 
3003 	while (v_idx--)
3004 		igc_reset_q_vector(adapter, v_idx);
3005 }
3006 
3007 /**
3008  * igc_clear_interrupt_scheme - reset the device to a state of no interrupts
3009  * @adapter: Pointer to adapter structure
3010  *
3011  * This function resets the device so that it has 0 rx queues, tx queues, and
3012  * MSI-X interrupts allocated.
3013  */
3014 static void igc_clear_interrupt_scheme(struct igc_adapter *adapter)
3015 {
3016 	igc_free_q_vectors(adapter);
3017 	igc_reset_interrupt_capability(adapter);
3018 }
3019 
3020 /**
3021  * igc_free_q_vectors - Free memory allocated for interrupt vectors
3022  * @adapter: board private structure to initialize
3023  *
3024  * This function frees the memory allocated to the q_vectors.  In addition if
3025  * NAPI is enabled it will delete any references to the NAPI struct prior
3026  * to freeing the q_vector.
3027  */
3028 static void igc_free_q_vectors(struct igc_adapter *adapter)
3029 {
3030 	int v_idx = adapter->num_q_vectors;
3031 
3032 	adapter->num_tx_queues = 0;
3033 	adapter->num_rx_queues = 0;
3034 	adapter->num_q_vectors = 0;
3035 
3036 	while (v_idx--) {
3037 		igc_reset_q_vector(adapter, v_idx);
3038 		igc_free_q_vector(adapter, v_idx);
3039 	}
3040 }
3041 
3042 /**
3043  * igc_free_q_vector - Free memory allocated for specific interrupt vector
3044  * @adapter: board private structure to initialize
3045  * @v_idx: Index of vector to be freed
3046  *
3047  * This function frees the memory allocated to the q_vector.
3048  */
3049 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx)
3050 {
3051 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
3052 
3053 	adapter->q_vector[v_idx] = NULL;
3054 
3055 	/* igc_get_stats64() might access the rings on this vector,
3056 	 * we must wait a grace period before freeing it.
3057 	 */
3058 	if (q_vector)
3059 		kfree_rcu(q_vector, rcu);
3060 }
3061 
3062 /* Need to wait a few seconds after link up to get diagnostic information from
3063  * the phy
3064  */
3065 static void igc_update_phy_info(struct timer_list *t)
3066 {
3067 	struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer);
3068 
3069 	igc_get_phy_info(&adapter->hw);
3070 }
3071 
3072 /**
3073  * igc_has_link - check shared code for link and determine up/down
3074  * @adapter: pointer to driver private info
3075  */
3076 bool igc_has_link(struct igc_adapter *adapter)
3077 {
3078 	struct igc_hw *hw = &adapter->hw;
3079 	bool link_active = false;
3080 
3081 	/* get_link_status is set on LSC (link status) interrupt or
3082 	 * rx sequence error interrupt.  get_link_status will stay
3083 	 * false until the igc_check_for_link establishes link
3084 	 * for copper adapters ONLY
3085 	 */
3086 	switch (hw->phy.media_type) {
3087 	case igc_media_type_copper:
3088 		if (!hw->mac.get_link_status)
3089 			return true;
3090 		hw->mac.ops.check_for_link(hw);
3091 		link_active = !hw->mac.get_link_status;
3092 		break;
3093 	default:
3094 	case igc_media_type_unknown:
3095 		break;
3096 	}
3097 
3098 	if (hw->mac.type == igc_i225 &&
3099 	    hw->phy.id == I225_I_PHY_ID) {
3100 		if (!netif_carrier_ok(adapter->netdev)) {
3101 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
3102 		} else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) {
3103 			adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE;
3104 			adapter->link_check_timeout = jiffies;
3105 		}
3106 	}
3107 
3108 	return link_active;
3109 }
3110 
3111 /**
3112  * igc_watchdog - Timer Call-back
3113  * @data: pointer to adapter cast into an unsigned long
3114  */
3115 static void igc_watchdog(struct timer_list *t)
3116 {
3117 	struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer);
3118 	/* Do the rest outside of interrupt context */
3119 	schedule_work(&adapter->watchdog_task);
3120 }
3121 
3122 static void igc_watchdog_task(struct work_struct *work)
3123 {
3124 	struct igc_adapter *adapter = container_of(work,
3125 						   struct igc_adapter,
3126 						   watchdog_task);
3127 	struct net_device *netdev = adapter->netdev;
3128 	struct igc_hw *hw = &adapter->hw;
3129 	struct igc_phy_info *phy = &hw->phy;
3130 	u16 phy_data, retry_count = 20;
3131 	u32 connsw;
3132 	u32 link;
3133 	int i;
3134 
3135 	link = igc_has_link(adapter);
3136 
3137 	if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) {
3138 		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
3139 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
3140 		else
3141 			link = false;
3142 	}
3143 
3144 	/* Force link down if we have fiber to swap to */
3145 	if (adapter->flags & IGC_FLAG_MAS_ENABLE) {
3146 		if (hw->phy.media_type == igc_media_type_copper) {
3147 			connsw = rd32(IGC_CONNSW);
3148 			if (!(connsw & IGC_CONNSW_AUTOSENSE_EN))
3149 				link = 0;
3150 		}
3151 	}
3152 	if (link) {
3153 		if (!netif_carrier_ok(netdev)) {
3154 			u32 ctrl;
3155 
3156 			hw->mac.ops.get_speed_and_duplex(hw,
3157 							 &adapter->link_speed,
3158 							 &adapter->link_duplex);
3159 
3160 			ctrl = rd32(IGC_CTRL);
3161 			/* Link status message must follow this format */
3162 			netdev_info(netdev,
3163 				    "igc: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
3164 				    netdev->name,
3165 				    adapter->link_speed,
3166 				    adapter->link_duplex == FULL_DUPLEX ?
3167 				    "Full" : "Half",
3168 				    (ctrl & IGC_CTRL_TFCE) &&
3169 				    (ctrl & IGC_CTRL_RFCE) ? "RX/TX" :
3170 				    (ctrl & IGC_CTRL_RFCE) ?  "RX" :
3171 				    (ctrl & IGC_CTRL_TFCE) ?  "TX" : "None");
3172 
3173 			/* check if SmartSpeed worked */
3174 			igc_check_downshift(hw);
3175 			if (phy->speed_downgraded)
3176 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
3177 
3178 			/* adjust timeout factor according to speed/duplex */
3179 			adapter->tx_timeout_factor = 1;
3180 			switch (adapter->link_speed) {
3181 			case SPEED_10:
3182 				adapter->tx_timeout_factor = 14;
3183 				break;
3184 			case SPEED_100:
3185 				/* maybe add some timeout factor ? */
3186 				break;
3187 			}
3188 
3189 			if (adapter->link_speed != SPEED_1000)
3190 				goto no_wait;
3191 
3192 			/* wait for Remote receiver status OK */
3193 retry_read_status:
3194 			if (!igc_read_phy_reg(hw, PHY_1000T_STATUS,
3195 					      &phy_data)) {
3196 				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
3197 				    retry_count) {
3198 					msleep(100);
3199 					retry_count--;
3200 					goto retry_read_status;
3201 				} else if (!retry_count) {
3202 					dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
3203 				}
3204 			} else {
3205 				dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
3206 			}
3207 no_wait:
3208 			netif_carrier_on(netdev);
3209 
3210 			/* link state has changed, schedule phy info update */
3211 			if (!test_bit(__IGC_DOWN, &adapter->state))
3212 				mod_timer(&adapter->phy_info_timer,
3213 					  round_jiffies(jiffies + 2 * HZ));
3214 		}
3215 	} else {
3216 		if (netif_carrier_ok(netdev)) {
3217 			adapter->link_speed = 0;
3218 			adapter->link_duplex = 0;
3219 
3220 			/* Links status message must follow this format */
3221 			netdev_info(netdev, "igc: %s NIC Link is Down\n",
3222 				    netdev->name);
3223 			netif_carrier_off(netdev);
3224 
3225 			/* link state has changed, schedule phy info update */
3226 			if (!test_bit(__IGC_DOWN, &adapter->state))
3227 				mod_timer(&adapter->phy_info_timer,
3228 					  round_jiffies(jiffies + 2 * HZ));
3229 
3230 			/* link is down, time to check for alternate media */
3231 			if (adapter->flags & IGC_FLAG_MAS_ENABLE) {
3232 				if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
3233 					schedule_work(&adapter->reset_task);
3234 					/* return immediately */
3235 					return;
3236 				}
3237 			}
3238 
3239 		/* also check for alternate media here */
3240 		} else if (!netif_carrier_ok(netdev) &&
3241 			   (adapter->flags & IGC_FLAG_MAS_ENABLE)) {
3242 			if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
3243 				schedule_work(&adapter->reset_task);
3244 				/* return immediately */
3245 				return;
3246 			}
3247 		}
3248 	}
3249 
3250 	spin_lock(&adapter->stats64_lock);
3251 	igc_update_stats(adapter);
3252 	spin_unlock(&adapter->stats64_lock);
3253 
3254 	for (i = 0; i < adapter->num_tx_queues; i++) {
3255 		struct igc_ring *tx_ring = adapter->tx_ring[i];
3256 
3257 		if (!netif_carrier_ok(netdev)) {
3258 			/* We've lost link, so the controller stops DMA,
3259 			 * but we've got queued Tx work that's never going
3260 			 * to get done, so reset controller to flush Tx.
3261 			 * (Do the reset outside of interrupt context).
3262 			 */
3263 			if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) {
3264 				adapter->tx_timeout_count++;
3265 				schedule_work(&adapter->reset_task);
3266 				/* return immediately since reset is imminent */
3267 				return;
3268 			}
3269 		}
3270 
3271 		/* Force detection of hung controller every watchdog period */
3272 		set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3273 	}
3274 
3275 	/* Cause software interrupt to ensure Rx ring is cleaned */
3276 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
3277 		u32 eics = 0;
3278 
3279 		for (i = 0; i < adapter->num_q_vectors; i++)
3280 			eics |= adapter->q_vector[i]->eims_value;
3281 		wr32(IGC_EICS, eics);
3282 	} else {
3283 		wr32(IGC_ICS, IGC_ICS_RXDMT0);
3284 	}
3285 
3286 	/* Reset the timer */
3287 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
3288 		if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)
3289 			mod_timer(&adapter->watchdog_timer,
3290 				  round_jiffies(jiffies +  HZ));
3291 		else
3292 			mod_timer(&adapter->watchdog_timer,
3293 				  round_jiffies(jiffies + 2 * HZ));
3294 	}
3295 }
3296 
3297 /**
3298  * igc_update_ring_itr - update the dynamic ITR value based on packet size
3299  * @q_vector: pointer to q_vector
3300  *
3301  * Stores a new ITR value based on strictly on packet size.  This
3302  * algorithm is less sophisticated than that used in igc_update_itr,
3303  * due to the difficulty of synchronizing statistics across multiple
3304  * receive rings.  The divisors and thresholds used by this function
3305  * were determined based on theoretical maximum wire speed and testing
3306  * data, in order to minimize response time while increasing bulk
3307  * throughput.
3308  * NOTE: This function is called only when operating in a multiqueue
3309  * receive environment.
3310  */
3311 static void igc_update_ring_itr(struct igc_q_vector *q_vector)
3312 {
3313 	struct igc_adapter *adapter = q_vector->adapter;
3314 	int new_val = q_vector->itr_val;
3315 	int avg_wire_size = 0;
3316 	unsigned int packets;
3317 
3318 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
3319 	 * ints/sec - ITR timer value of 120 ticks.
3320 	 */
3321 	switch (adapter->link_speed) {
3322 	case SPEED_10:
3323 	case SPEED_100:
3324 		new_val = IGC_4K_ITR;
3325 		goto set_itr_val;
3326 	default:
3327 		break;
3328 	}
3329 
3330 	packets = q_vector->rx.total_packets;
3331 	if (packets)
3332 		avg_wire_size = q_vector->rx.total_bytes / packets;
3333 
3334 	packets = q_vector->tx.total_packets;
3335 	if (packets)
3336 		avg_wire_size = max_t(u32, avg_wire_size,
3337 				      q_vector->tx.total_bytes / packets);
3338 
3339 	/* if avg_wire_size isn't set no work was done */
3340 	if (!avg_wire_size)
3341 		goto clear_counts;
3342 
3343 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
3344 	avg_wire_size += 24;
3345 
3346 	/* Don't starve jumbo frames */
3347 	avg_wire_size = min(avg_wire_size, 3000);
3348 
3349 	/* Give a little boost to mid-size frames */
3350 	if (avg_wire_size > 300 && avg_wire_size < 1200)
3351 		new_val = avg_wire_size / 3;
3352 	else
3353 		new_val = avg_wire_size / 2;
3354 
3355 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
3356 	if (new_val < IGC_20K_ITR &&
3357 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3358 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3359 		new_val = IGC_20K_ITR;
3360 
3361 set_itr_val:
3362 	if (new_val != q_vector->itr_val) {
3363 		q_vector->itr_val = new_val;
3364 		q_vector->set_itr = 1;
3365 	}
3366 clear_counts:
3367 	q_vector->rx.total_bytes = 0;
3368 	q_vector->rx.total_packets = 0;
3369 	q_vector->tx.total_bytes = 0;
3370 	q_vector->tx.total_packets = 0;
3371 }
3372 
3373 /**
3374  * igc_update_itr - update the dynamic ITR value based on statistics
3375  * @q_vector: pointer to q_vector
3376  * @ring_container: ring info to update the itr for
3377  *
3378  * Stores a new ITR value based on packets and byte
3379  * counts during the last interrupt.  The advantage of per interrupt
3380  * computation is faster updates and more accurate ITR for the current
3381  * traffic pattern.  Constants in this function were computed
3382  * based on theoretical maximum wire speed and thresholds were set based
3383  * on testing data as well as attempting to minimize response time
3384  * while increasing bulk throughput.
3385  * NOTE: These calculations are only valid when operating in a single-
3386  * queue environment.
3387  */
3388 static void igc_update_itr(struct igc_q_vector *q_vector,
3389 			   struct igc_ring_container *ring_container)
3390 {
3391 	unsigned int packets = ring_container->total_packets;
3392 	unsigned int bytes = ring_container->total_bytes;
3393 	u8 itrval = ring_container->itr;
3394 
3395 	/* no packets, exit with status unchanged */
3396 	if (packets == 0)
3397 		return;
3398 
3399 	switch (itrval) {
3400 	case lowest_latency:
3401 		/* handle TSO and jumbo frames */
3402 		if (bytes / packets > 8000)
3403 			itrval = bulk_latency;
3404 		else if ((packets < 5) && (bytes > 512))
3405 			itrval = low_latency;
3406 		break;
3407 	case low_latency:  /* 50 usec aka 20000 ints/s */
3408 		if (bytes > 10000) {
3409 			/* this if handles the TSO accounting */
3410 			if (bytes / packets > 8000)
3411 				itrval = bulk_latency;
3412 			else if ((packets < 10) || ((bytes / packets) > 1200))
3413 				itrval = bulk_latency;
3414 			else if ((packets > 35))
3415 				itrval = lowest_latency;
3416 		} else if (bytes / packets > 2000) {
3417 			itrval = bulk_latency;
3418 		} else if (packets <= 2 && bytes < 512) {
3419 			itrval = lowest_latency;
3420 		}
3421 		break;
3422 	case bulk_latency: /* 250 usec aka 4000 ints/s */
3423 		if (bytes > 25000) {
3424 			if (packets > 35)
3425 				itrval = low_latency;
3426 		} else if (bytes < 1500) {
3427 			itrval = low_latency;
3428 		}
3429 		break;
3430 	}
3431 
3432 	/* clear work counters since we have the values we need */
3433 	ring_container->total_bytes = 0;
3434 	ring_container->total_packets = 0;
3435 
3436 	/* write updated itr to ring container */
3437 	ring_container->itr = itrval;
3438 }
3439 
3440 /**
3441  * igc_intr_msi - Interrupt Handler
3442  * @irq: interrupt number
3443  * @data: pointer to a network interface device structure
3444  */
3445 static irqreturn_t igc_intr_msi(int irq, void *data)
3446 {
3447 	struct igc_adapter *adapter = data;
3448 	struct igc_q_vector *q_vector = adapter->q_vector[0];
3449 	struct igc_hw *hw = &adapter->hw;
3450 	/* read ICR disables interrupts using IAM */
3451 	u32 icr = rd32(IGC_ICR);
3452 
3453 	igc_write_itr(q_vector);
3454 
3455 	if (icr & IGC_ICR_DRSTA)
3456 		schedule_work(&adapter->reset_task);
3457 
3458 	if (icr & IGC_ICR_DOUTSYNC) {
3459 		/* HW is reporting DMA is out of sync */
3460 		adapter->stats.doosync++;
3461 	}
3462 
3463 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
3464 		hw->mac.get_link_status = 1;
3465 		if (!test_bit(__IGC_DOWN, &adapter->state))
3466 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
3467 	}
3468 
3469 	napi_schedule(&q_vector->napi);
3470 
3471 	return IRQ_HANDLED;
3472 }
3473 
3474 /**
3475  * igc_intr - Legacy Interrupt Handler
3476  * @irq: interrupt number
3477  * @data: pointer to a network interface device structure
3478  */
3479 static irqreturn_t igc_intr(int irq, void *data)
3480 {
3481 	struct igc_adapter *adapter = data;
3482 	struct igc_q_vector *q_vector = adapter->q_vector[0];
3483 	struct igc_hw *hw = &adapter->hw;
3484 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3485 	 * need for the IMC write
3486 	 */
3487 	u32 icr = rd32(IGC_ICR);
3488 
3489 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3490 	 * not set, then the adapter didn't send an interrupt
3491 	 */
3492 	if (!(icr & IGC_ICR_INT_ASSERTED))
3493 		return IRQ_NONE;
3494 
3495 	igc_write_itr(q_vector);
3496 
3497 	if (icr & IGC_ICR_DRSTA)
3498 		schedule_work(&adapter->reset_task);
3499 
3500 	if (icr & IGC_ICR_DOUTSYNC) {
3501 		/* HW is reporting DMA is out of sync */
3502 		adapter->stats.doosync++;
3503 	}
3504 
3505 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
3506 		hw->mac.get_link_status = 1;
3507 		/* guard against interrupt when we're going down */
3508 		if (!test_bit(__IGC_DOWN, &adapter->state))
3509 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
3510 	}
3511 
3512 	napi_schedule(&q_vector->napi);
3513 
3514 	return IRQ_HANDLED;
3515 }
3516 
3517 static void igc_set_itr(struct igc_q_vector *q_vector)
3518 {
3519 	struct igc_adapter *adapter = q_vector->adapter;
3520 	u32 new_itr = q_vector->itr_val;
3521 	u8 current_itr = 0;
3522 
3523 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3524 	switch (adapter->link_speed) {
3525 	case SPEED_10:
3526 	case SPEED_100:
3527 		current_itr = 0;
3528 		new_itr = IGC_4K_ITR;
3529 		goto set_itr_now;
3530 	default:
3531 		break;
3532 	}
3533 
3534 	igc_update_itr(q_vector, &q_vector->tx);
3535 	igc_update_itr(q_vector, &q_vector->rx);
3536 
3537 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
3538 
3539 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
3540 	if (current_itr == lowest_latency &&
3541 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
3542 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
3543 		current_itr = low_latency;
3544 
3545 	switch (current_itr) {
3546 	/* counts and packets in update_itr are dependent on these numbers */
3547 	case lowest_latency:
3548 		new_itr = IGC_70K_ITR; /* 70,000 ints/sec */
3549 		break;
3550 	case low_latency:
3551 		new_itr = IGC_20K_ITR; /* 20,000 ints/sec */
3552 		break;
3553 	case bulk_latency:
3554 		new_itr = IGC_4K_ITR;  /* 4,000 ints/sec */
3555 		break;
3556 	default:
3557 		break;
3558 	}
3559 
3560 set_itr_now:
3561 	if (new_itr != q_vector->itr_val) {
3562 		/* this attempts to bias the interrupt rate towards Bulk
3563 		 * by adding intermediate steps when interrupt rate is
3564 		 * increasing
3565 		 */
3566 		new_itr = new_itr > q_vector->itr_val ?
3567 			  max((new_itr * q_vector->itr_val) /
3568 			  (new_itr + (q_vector->itr_val >> 2)),
3569 			  new_itr) : new_itr;
3570 		/* Don't write the value here; it resets the adapter's
3571 		 * internal timer, and causes us to delay far longer than
3572 		 * we should between interrupts.  Instead, we write the ITR
3573 		 * value at the beginning of the next interrupt so the timing
3574 		 * ends up being correct.
3575 		 */
3576 		q_vector->itr_val = new_itr;
3577 		q_vector->set_itr = 1;
3578 	}
3579 }
3580 
3581 static void igc_ring_irq_enable(struct igc_q_vector *q_vector)
3582 {
3583 	struct igc_adapter *adapter = q_vector->adapter;
3584 	struct igc_hw *hw = &adapter->hw;
3585 
3586 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
3587 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
3588 		if (adapter->num_q_vectors == 1)
3589 			igc_set_itr(q_vector);
3590 		else
3591 			igc_update_ring_itr(q_vector);
3592 	}
3593 
3594 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
3595 		if (adapter->msix_entries)
3596 			wr32(IGC_EIMS, q_vector->eims_value);
3597 		else
3598 			igc_irq_enable(adapter);
3599 	}
3600 }
3601 
3602 /**
3603  * igc_poll - NAPI Rx polling callback
3604  * @napi: napi polling structure
3605  * @budget: count of how many packets we should handle
3606  */
3607 static int igc_poll(struct napi_struct *napi, int budget)
3608 {
3609 	struct igc_q_vector *q_vector = container_of(napi,
3610 						     struct igc_q_vector,
3611 						     napi);
3612 	bool clean_complete = true;
3613 	int work_done = 0;
3614 
3615 	if (q_vector->tx.ring)
3616 		clean_complete = igc_clean_tx_irq(q_vector, budget);
3617 
3618 	if (q_vector->rx.ring) {
3619 		int cleaned = igc_clean_rx_irq(q_vector, budget);
3620 
3621 		work_done += cleaned;
3622 		if (cleaned >= budget)
3623 			clean_complete = false;
3624 	}
3625 
3626 	/* If all work not completed, return budget and keep polling */
3627 	if (!clean_complete)
3628 		return budget;
3629 
3630 	/* Exit the polling mode, but don't re-enable interrupts if stack might
3631 	 * poll us due to busy-polling
3632 	 */
3633 	if (likely(napi_complete_done(napi, work_done)))
3634 		igc_ring_irq_enable(q_vector);
3635 
3636 	return min(work_done, budget - 1);
3637 }
3638 
3639 /**
3640  * igc_set_interrupt_capability - set MSI or MSI-X if supported
3641  * @adapter: Pointer to adapter structure
3642  *
3643  * Attempt to configure interrupts using the best available
3644  * capabilities of the hardware and kernel.
3645  */
3646 static void igc_set_interrupt_capability(struct igc_adapter *adapter,
3647 					 bool msix)
3648 {
3649 	int numvecs, i;
3650 	int err;
3651 
3652 	if (!msix)
3653 		goto msi_only;
3654 	adapter->flags |= IGC_FLAG_HAS_MSIX;
3655 
3656 	/* Number of supported queues. */
3657 	adapter->num_rx_queues = adapter->rss_queues;
3658 
3659 	adapter->num_tx_queues = adapter->rss_queues;
3660 
3661 	/* start with one vector for every Rx queue */
3662 	numvecs = adapter->num_rx_queues;
3663 
3664 	/* if Tx handler is separate add 1 for every Tx queue */
3665 	if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS))
3666 		numvecs += adapter->num_tx_queues;
3667 
3668 	/* store the number of vectors reserved for queues */
3669 	adapter->num_q_vectors = numvecs;
3670 
3671 	/* add 1 vector for link status interrupts */
3672 	numvecs++;
3673 
3674 	adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
3675 					GFP_KERNEL);
3676 
3677 	if (!adapter->msix_entries)
3678 		return;
3679 
3680 	/* populate entry values */
3681 	for (i = 0; i < numvecs; i++)
3682 		adapter->msix_entries[i].entry = i;
3683 
3684 	err = pci_enable_msix_range(adapter->pdev,
3685 				    adapter->msix_entries,
3686 				    numvecs,
3687 				    numvecs);
3688 	if (err > 0)
3689 		return;
3690 
3691 	kfree(adapter->msix_entries);
3692 	adapter->msix_entries = NULL;
3693 
3694 	igc_reset_interrupt_capability(adapter);
3695 
3696 msi_only:
3697 	adapter->flags &= ~IGC_FLAG_HAS_MSIX;
3698 
3699 	adapter->rss_queues = 1;
3700 	adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
3701 	adapter->num_rx_queues = 1;
3702 	adapter->num_tx_queues = 1;
3703 	adapter->num_q_vectors = 1;
3704 	if (!pci_enable_msi(adapter->pdev))
3705 		adapter->flags |= IGC_FLAG_HAS_MSI;
3706 }
3707 
3708 static void igc_add_ring(struct igc_ring *ring,
3709 			 struct igc_ring_container *head)
3710 {
3711 	head->ring = ring;
3712 	head->count++;
3713 }
3714 
3715 /**
3716  * igc_alloc_q_vector - Allocate memory for a single interrupt vector
3717  * @adapter: board private structure to initialize
3718  * @v_count: q_vectors allocated on adapter, used for ring interleaving
3719  * @v_idx: index of vector in adapter struct
3720  * @txr_count: total number of Tx rings to allocate
3721  * @txr_idx: index of first Tx ring to allocate
3722  * @rxr_count: total number of Rx rings to allocate
3723  * @rxr_idx: index of first Rx ring to allocate
3724  *
3725  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
3726  */
3727 static int igc_alloc_q_vector(struct igc_adapter *adapter,
3728 			      unsigned int v_count, unsigned int v_idx,
3729 			      unsigned int txr_count, unsigned int txr_idx,
3730 			      unsigned int rxr_count, unsigned int rxr_idx)
3731 {
3732 	struct igc_q_vector *q_vector;
3733 	struct igc_ring *ring;
3734 	int ring_count;
3735 
3736 	/* igc only supports 1 Tx and/or 1 Rx queue per vector */
3737 	if (txr_count > 1 || rxr_count > 1)
3738 		return -ENOMEM;
3739 
3740 	ring_count = txr_count + rxr_count;
3741 
3742 	/* allocate q_vector and rings */
3743 	q_vector = adapter->q_vector[v_idx];
3744 	if (!q_vector)
3745 		q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
3746 				   GFP_KERNEL);
3747 	else
3748 		memset(q_vector, 0, struct_size(q_vector, ring, ring_count));
3749 	if (!q_vector)
3750 		return -ENOMEM;
3751 
3752 	/* initialize NAPI */
3753 	netif_napi_add(adapter->netdev, &q_vector->napi,
3754 		       igc_poll, 64);
3755 
3756 	/* tie q_vector and adapter together */
3757 	adapter->q_vector[v_idx] = q_vector;
3758 	q_vector->adapter = adapter;
3759 
3760 	/* initialize work limits */
3761 	q_vector->tx.work_limit = adapter->tx_work_limit;
3762 
3763 	/* initialize ITR configuration */
3764 	q_vector->itr_register = adapter->io_addr + IGC_EITR(0);
3765 	q_vector->itr_val = IGC_START_ITR;
3766 
3767 	/* initialize pointer to rings */
3768 	ring = q_vector->ring;
3769 
3770 	/* initialize ITR */
3771 	if (rxr_count) {
3772 		/* rx or rx/tx vector */
3773 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
3774 			q_vector->itr_val = adapter->rx_itr_setting;
3775 	} else {
3776 		/* tx only vector */
3777 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
3778 			q_vector->itr_val = adapter->tx_itr_setting;
3779 	}
3780 
3781 	if (txr_count) {
3782 		/* assign generic ring traits */
3783 		ring->dev = &adapter->pdev->dev;
3784 		ring->netdev = adapter->netdev;
3785 
3786 		/* configure backlink on ring */
3787 		ring->q_vector = q_vector;
3788 
3789 		/* update q_vector Tx values */
3790 		igc_add_ring(ring, &q_vector->tx);
3791 
3792 		/* apply Tx specific ring traits */
3793 		ring->count = adapter->tx_ring_count;
3794 		ring->queue_index = txr_idx;
3795 
3796 		/* assign ring to adapter */
3797 		adapter->tx_ring[txr_idx] = ring;
3798 
3799 		/* push pointer to next ring */
3800 		ring++;
3801 	}
3802 
3803 	if (rxr_count) {
3804 		/* assign generic ring traits */
3805 		ring->dev = &adapter->pdev->dev;
3806 		ring->netdev = adapter->netdev;
3807 
3808 		/* configure backlink on ring */
3809 		ring->q_vector = q_vector;
3810 
3811 		/* update q_vector Rx values */
3812 		igc_add_ring(ring, &q_vector->rx);
3813 
3814 		/* apply Rx specific ring traits */
3815 		ring->count = adapter->rx_ring_count;
3816 		ring->queue_index = rxr_idx;
3817 
3818 		/* assign ring to adapter */
3819 		adapter->rx_ring[rxr_idx] = ring;
3820 	}
3821 
3822 	return 0;
3823 }
3824 
3825 /**
3826  * igc_alloc_q_vectors - Allocate memory for interrupt vectors
3827  * @adapter: board private structure to initialize
3828  *
3829  * We allocate one q_vector per queue interrupt.  If allocation fails we
3830  * return -ENOMEM.
3831  */
3832 static int igc_alloc_q_vectors(struct igc_adapter *adapter)
3833 {
3834 	int rxr_remaining = adapter->num_rx_queues;
3835 	int txr_remaining = adapter->num_tx_queues;
3836 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
3837 	int q_vectors = adapter->num_q_vectors;
3838 	int err;
3839 
3840 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
3841 		for (; rxr_remaining; v_idx++) {
3842 			err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
3843 						 0, 0, 1, rxr_idx);
3844 
3845 			if (err)
3846 				goto err_out;
3847 
3848 			/* update counts and index */
3849 			rxr_remaining--;
3850 			rxr_idx++;
3851 		}
3852 	}
3853 
3854 	for (; v_idx < q_vectors; v_idx++) {
3855 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
3856 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
3857 
3858 		err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
3859 					 tqpv, txr_idx, rqpv, rxr_idx);
3860 
3861 		if (err)
3862 			goto err_out;
3863 
3864 		/* update counts and index */
3865 		rxr_remaining -= rqpv;
3866 		txr_remaining -= tqpv;
3867 		rxr_idx++;
3868 		txr_idx++;
3869 	}
3870 
3871 	return 0;
3872 
3873 err_out:
3874 	adapter->num_tx_queues = 0;
3875 	adapter->num_rx_queues = 0;
3876 	adapter->num_q_vectors = 0;
3877 
3878 	while (v_idx--)
3879 		igc_free_q_vector(adapter, v_idx);
3880 
3881 	return -ENOMEM;
3882 }
3883 
3884 /**
3885  * igc_cache_ring_register - Descriptor ring to register mapping
3886  * @adapter: board private structure to initialize
3887  *
3888  * Once we know the feature-set enabled for the device, we'll cache
3889  * the register offset the descriptor ring is assigned to.
3890  */
3891 static void igc_cache_ring_register(struct igc_adapter *adapter)
3892 {
3893 	int i = 0, j = 0;
3894 
3895 	switch (adapter->hw.mac.type) {
3896 	case igc_i225:
3897 	/* Fall through */
3898 	default:
3899 		for (; i < adapter->num_rx_queues; i++)
3900 			adapter->rx_ring[i]->reg_idx = i;
3901 		for (; j < adapter->num_tx_queues; j++)
3902 			adapter->tx_ring[j]->reg_idx = j;
3903 		break;
3904 	}
3905 }
3906 
3907 /**
3908  * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
3909  * @adapter: Pointer to adapter structure
3910  *
3911  * This function initializes the interrupts and allocates all of the queues.
3912  */
3913 static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix)
3914 {
3915 	struct pci_dev *pdev = adapter->pdev;
3916 	int err = 0;
3917 
3918 	igc_set_interrupt_capability(adapter, msix);
3919 
3920 	err = igc_alloc_q_vectors(adapter);
3921 	if (err) {
3922 		dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
3923 		goto err_alloc_q_vectors;
3924 	}
3925 
3926 	igc_cache_ring_register(adapter);
3927 
3928 	return 0;
3929 
3930 err_alloc_q_vectors:
3931 	igc_reset_interrupt_capability(adapter);
3932 	return err;
3933 }
3934 
3935 static void igc_free_irq(struct igc_adapter *adapter)
3936 {
3937 	if (adapter->msix_entries) {
3938 		int vector = 0, i;
3939 
3940 		free_irq(adapter->msix_entries[vector++].vector, adapter);
3941 
3942 		for (i = 0; i < adapter->num_q_vectors; i++)
3943 			free_irq(adapter->msix_entries[vector++].vector,
3944 				 adapter->q_vector[i]);
3945 	} else {
3946 		free_irq(adapter->pdev->irq, adapter);
3947 	}
3948 }
3949 
3950 /**
3951  * igc_irq_disable - Mask off interrupt generation on the NIC
3952  * @adapter: board private structure
3953  */
3954 static void igc_irq_disable(struct igc_adapter *adapter)
3955 {
3956 	struct igc_hw *hw = &adapter->hw;
3957 
3958 	if (adapter->msix_entries) {
3959 		u32 regval = rd32(IGC_EIAM);
3960 
3961 		wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask);
3962 		wr32(IGC_EIMC, adapter->eims_enable_mask);
3963 		regval = rd32(IGC_EIAC);
3964 		wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask);
3965 	}
3966 
3967 	wr32(IGC_IAM, 0);
3968 	wr32(IGC_IMC, ~0);
3969 	wrfl();
3970 
3971 	if (adapter->msix_entries) {
3972 		int vector = 0, i;
3973 
3974 		synchronize_irq(adapter->msix_entries[vector++].vector);
3975 
3976 		for (i = 0; i < adapter->num_q_vectors; i++)
3977 			synchronize_irq(adapter->msix_entries[vector++].vector);
3978 	} else {
3979 		synchronize_irq(adapter->pdev->irq);
3980 	}
3981 }
3982 
3983 /**
3984  * igc_irq_enable - Enable default interrupt generation settings
3985  * @adapter: board private structure
3986  */
3987 static void igc_irq_enable(struct igc_adapter *adapter)
3988 {
3989 	struct igc_hw *hw = &adapter->hw;
3990 
3991 	if (adapter->msix_entries) {
3992 		u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA;
3993 		u32 regval = rd32(IGC_EIAC);
3994 
3995 		wr32(IGC_EIAC, regval | adapter->eims_enable_mask);
3996 		regval = rd32(IGC_EIAM);
3997 		wr32(IGC_EIAM, regval | adapter->eims_enable_mask);
3998 		wr32(IGC_EIMS, adapter->eims_enable_mask);
3999 		wr32(IGC_IMS, ims);
4000 	} else {
4001 		wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4002 		wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4003 	}
4004 }
4005 
4006 /**
4007  * igc_request_irq - initialize interrupts
4008  * @adapter: Pointer to adapter structure
4009  *
4010  * Attempts to configure interrupts using the best available
4011  * capabilities of the hardware and kernel.
4012  */
4013 static int igc_request_irq(struct igc_adapter *adapter)
4014 {
4015 	struct net_device *netdev = adapter->netdev;
4016 	struct pci_dev *pdev = adapter->pdev;
4017 	int err = 0;
4018 
4019 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
4020 		err = igc_request_msix(adapter);
4021 		if (!err)
4022 			goto request_done;
4023 		/* fall back to MSI */
4024 		igc_free_all_tx_resources(adapter);
4025 		igc_free_all_rx_resources(adapter);
4026 
4027 		igc_clear_interrupt_scheme(adapter);
4028 		err = igc_init_interrupt_scheme(adapter, false);
4029 		if (err)
4030 			goto request_done;
4031 		igc_setup_all_tx_resources(adapter);
4032 		igc_setup_all_rx_resources(adapter);
4033 		igc_configure(adapter);
4034 	}
4035 
4036 	igc_assign_vector(adapter->q_vector[0], 0);
4037 
4038 	if (adapter->flags & IGC_FLAG_HAS_MSI) {
4039 		err = request_irq(pdev->irq, &igc_intr_msi, 0,
4040 				  netdev->name, adapter);
4041 		if (!err)
4042 			goto request_done;
4043 
4044 		/* fall back to legacy interrupts */
4045 		igc_reset_interrupt_capability(adapter);
4046 		adapter->flags &= ~IGC_FLAG_HAS_MSI;
4047 	}
4048 
4049 	err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED,
4050 			  netdev->name, adapter);
4051 
4052 	if (err)
4053 		dev_err(&pdev->dev, "Error %d getting interrupt\n",
4054 			err);
4055 
4056 request_done:
4057 	return err;
4058 }
4059 
4060 static void igc_write_itr(struct igc_q_vector *q_vector)
4061 {
4062 	u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK;
4063 
4064 	if (!q_vector->set_itr)
4065 		return;
4066 
4067 	if (!itr_val)
4068 		itr_val = IGC_ITR_VAL_MASK;
4069 
4070 	itr_val |= IGC_EITR_CNT_IGNR;
4071 
4072 	writel(itr_val, q_vector->itr_register);
4073 	q_vector->set_itr = 0;
4074 }
4075 
4076 /**
4077  * igc_open - Called when a network interface is made active
4078  * @netdev: network interface device structure
4079  *
4080  * Returns 0 on success, negative value on failure
4081  *
4082  * The open entry point is called when a network interface is made
4083  * active by the system (IFF_UP).  At this point all resources needed
4084  * for transmit and receive operations are allocated, the interrupt
4085  * handler is registered with the OS, the watchdog timer is started,
4086  * and the stack is notified that the interface is ready.
4087  */
4088 static int __igc_open(struct net_device *netdev, bool resuming)
4089 {
4090 	struct igc_adapter *adapter = netdev_priv(netdev);
4091 	struct igc_hw *hw = &adapter->hw;
4092 	int err = 0;
4093 	int i = 0;
4094 
4095 	/* disallow open during test */
4096 
4097 	if (test_bit(__IGC_TESTING, &adapter->state)) {
4098 		WARN_ON(resuming);
4099 		return -EBUSY;
4100 	}
4101 
4102 	netif_carrier_off(netdev);
4103 
4104 	/* allocate transmit descriptors */
4105 	err = igc_setup_all_tx_resources(adapter);
4106 	if (err)
4107 		goto err_setup_tx;
4108 
4109 	/* allocate receive descriptors */
4110 	err = igc_setup_all_rx_resources(adapter);
4111 	if (err)
4112 		goto err_setup_rx;
4113 
4114 	igc_power_up_link(adapter);
4115 
4116 	igc_configure(adapter);
4117 
4118 	err = igc_request_irq(adapter);
4119 	if (err)
4120 		goto err_req_irq;
4121 
4122 	/* Notify the stack of the actual queue counts. */
4123 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
4124 	if (err)
4125 		goto err_set_queues;
4126 
4127 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
4128 	if (err)
4129 		goto err_set_queues;
4130 
4131 	clear_bit(__IGC_DOWN, &adapter->state);
4132 
4133 	for (i = 0; i < adapter->num_q_vectors; i++)
4134 		napi_enable(&adapter->q_vector[i]->napi);
4135 
4136 	/* Clear any pending interrupts. */
4137 	rd32(IGC_ICR);
4138 	igc_irq_enable(adapter);
4139 
4140 	netif_tx_start_all_queues(netdev);
4141 
4142 	/* start the watchdog. */
4143 	hw->mac.get_link_status = 1;
4144 	schedule_work(&adapter->watchdog_task);
4145 
4146 	return IGC_SUCCESS;
4147 
4148 err_set_queues:
4149 	igc_free_irq(adapter);
4150 err_req_irq:
4151 	igc_release_hw_control(adapter);
4152 	igc_power_down_link(adapter);
4153 	igc_free_all_rx_resources(adapter);
4154 err_setup_rx:
4155 	igc_free_all_tx_resources(adapter);
4156 err_setup_tx:
4157 	igc_reset(adapter);
4158 
4159 	return err;
4160 }
4161 
4162 static int igc_open(struct net_device *netdev)
4163 {
4164 	return __igc_open(netdev, false);
4165 }
4166 
4167 /**
4168  * igc_close - Disables a network interface
4169  * @netdev: network interface device structure
4170  *
4171  * Returns 0, this is not allowed to fail
4172  *
4173  * The close entry point is called when an interface is de-activated
4174  * by the OS.  The hardware is still under the driver's control, but
4175  * needs to be disabled.  A global MAC reset is issued to stop the
4176  * hardware, and all transmit and receive resources are freed.
4177  */
4178 static int __igc_close(struct net_device *netdev, bool suspending)
4179 {
4180 	struct igc_adapter *adapter = netdev_priv(netdev);
4181 
4182 	WARN_ON(test_bit(__IGC_RESETTING, &adapter->state));
4183 
4184 	igc_down(adapter);
4185 
4186 	igc_release_hw_control(adapter);
4187 
4188 	igc_free_irq(adapter);
4189 
4190 	igc_free_all_tx_resources(adapter);
4191 	igc_free_all_rx_resources(adapter);
4192 
4193 	return 0;
4194 }
4195 
4196 static int igc_close(struct net_device *netdev)
4197 {
4198 	if (netif_device_present(netdev) || netdev->dismantle)
4199 		return __igc_close(netdev, false);
4200 	return 0;
4201 }
4202 
4203 static const struct net_device_ops igc_netdev_ops = {
4204 	.ndo_open		= igc_open,
4205 	.ndo_stop		= igc_close,
4206 	.ndo_start_xmit		= igc_xmit_frame,
4207 	.ndo_set_rx_mode	= igc_set_rx_mode,
4208 	.ndo_set_mac_address	= igc_set_mac,
4209 	.ndo_change_mtu		= igc_change_mtu,
4210 	.ndo_get_stats		= igc_get_stats,
4211 	.ndo_fix_features	= igc_fix_features,
4212 	.ndo_set_features	= igc_set_features,
4213 	.ndo_features_check	= igc_features_check,
4214 };
4215 
4216 /* PCIe configuration access */
4217 void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
4218 {
4219 	struct igc_adapter *adapter = hw->back;
4220 
4221 	pci_read_config_word(adapter->pdev, reg, value);
4222 }
4223 
4224 void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
4225 {
4226 	struct igc_adapter *adapter = hw->back;
4227 
4228 	pci_write_config_word(adapter->pdev, reg, *value);
4229 }
4230 
4231 s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
4232 {
4233 	struct igc_adapter *adapter = hw->back;
4234 
4235 	if (!pci_is_pcie(adapter->pdev))
4236 		return -IGC_ERR_CONFIG;
4237 
4238 	pcie_capability_read_word(adapter->pdev, reg, value);
4239 
4240 	return IGC_SUCCESS;
4241 }
4242 
4243 s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
4244 {
4245 	struct igc_adapter *adapter = hw->back;
4246 
4247 	if (!pci_is_pcie(adapter->pdev))
4248 		return -IGC_ERR_CONFIG;
4249 
4250 	pcie_capability_write_word(adapter->pdev, reg, *value);
4251 
4252 	return IGC_SUCCESS;
4253 }
4254 
4255 u32 igc_rd32(struct igc_hw *hw, u32 reg)
4256 {
4257 	struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw);
4258 	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
4259 	u32 value = 0;
4260 
4261 	if (IGC_REMOVED(hw_addr))
4262 		return ~value;
4263 
4264 	value = readl(&hw_addr[reg]);
4265 
4266 	/* reads should not return all F's */
4267 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
4268 		struct net_device *netdev = igc->netdev;
4269 
4270 		hw->hw_addr = NULL;
4271 		netif_device_detach(netdev);
4272 		netdev_err(netdev, "PCIe link lost, device now detached\n");
4273 		WARN(pci_device_is_present(igc->pdev),
4274 		     "igc: Failed to read reg 0x%x!\n", reg);
4275 	}
4276 
4277 	return value;
4278 }
4279 
4280 int igc_set_spd_dplx(struct igc_adapter *adapter, u32 spd, u8 dplx)
4281 {
4282 	struct pci_dev *pdev = adapter->pdev;
4283 	struct igc_mac_info *mac = &adapter->hw.mac;
4284 
4285 	mac->autoneg = 0;
4286 
4287 	/* Make sure dplx is at most 1 bit and lsb of speed is not set
4288 	 * for the switch() below to work
4289 	 */
4290 	if ((spd & 1) || (dplx & ~1))
4291 		goto err_inval;
4292 
4293 	switch (spd + dplx) {
4294 	case SPEED_10 + DUPLEX_HALF:
4295 		mac->forced_speed_duplex = ADVERTISE_10_HALF;
4296 		break;
4297 	case SPEED_10 + DUPLEX_FULL:
4298 		mac->forced_speed_duplex = ADVERTISE_10_FULL;
4299 		break;
4300 	case SPEED_100 + DUPLEX_HALF:
4301 		mac->forced_speed_duplex = ADVERTISE_100_HALF;
4302 		break;
4303 	case SPEED_100 + DUPLEX_FULL:
4304 		mac->forced_speed_duplex = ADVERTISE_100_FULL;
4305 		break;
4306 	case SPEED_1000 + DUPLEX_FULL:
4307 		mac->autoneg = 1;
4308 		adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
4309 		break;
4310 	case SPEED_1000 + DUPLEX_HALF: /* not supported */
4311 		goto err_inval;
4312 	case SPEED_2500 + DUPLEX_FULL:
4313 		mac->autoneg = 1;
4314 		adapter->hw.phy.autoneg_advertised = ADVERTISE_2500_FULL;
4315 		break;
4316 	case SPEED_2500 + DUPLEX_HALF: /* not supported */
4317 	default:
4318 		goto err_inval;
4319 	}
4320 
4321 	/* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
4322 	adapter->hw.phy.mdix = AUTO_ALL_MODES;
4323 
4324 	return 0;
4325 
4326 err_inval:
4327 	dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
4328 	return -EINVAL;
4329 }
4330 
4331 /**
4332  * igc_probe - Device Initialization Routine
4333  * @pdev: PCI device information struct
4334  * @ent: entry in igc_pci_tbl
4335  *
4336  * Returns 0 on success, negative on failure
4337  *
4338  * igc_probe initializes an adapter identified by a pci_dev structure.
4339  * The OS initialization, configuring the adapter private structure,
4340  * and a hardware reset occur.
4341  */
4342 static int igc_probe(struct pci_dev *pdev,
4343 		     const struct pci_device_id *ent)
4344 {
4345 	struct igc_adapter *adapter;
4346 	struct net_device *netdev;
4347 	struct igc_hw *hw;
4348 	const struct igc_info *ei = igc_info_tbl[ent->driver_data];
4349 	int err;
4350 
4351 	err = pci_enable_device_mem(pdev);
4352 	if (err)
4353 		return err;
4354 
4355 	err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
4356 	if (!err) {
4357 		err = dma_set_coherent_mask(&pdev->dev,
4358 					    DMA_BIT_MASK(64));
4359 	} else {
4360 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
4361 		if (err) {
4362 			err = dma_set_coherent_mask(&pdev->dev,
4363 						    DMA_BIT_MASK(32));
4364 			if (err) {
4365 				dev_err(&pdev->dev, "igc: Wrong DMA config\n");
4366 				goto err_dma;
4367 			}
4368 		}
4369 	}
4370 
4371 	err = pci_request_selected_regions(pdev,
4372 					   pci_select_bars(pdev,
4373 							   IORESOURCE_MEM),
4374 					   igc_driver_name);
4375 	if (err)
4376 		goto err_pci_reg;
4377 
4378 	pci_enable_pcie_error_reporting(pdev);
4379 
4380 	pci_set_master(pdev);
4381 
4382 	err = -ENOMEM;
4383 	netdev = alloc_etherdev_mq(sizeof(struct igc_adapter),
4384 				   IGC_MAX_TX_QUEUES);
4385 
4386 	if (!netdev)
4387 		goto err_alloc_etherdev;
4388 
4389 	SET_NETDEV_DEV(netdev, &pdev->dev);
4390 
4391 	pci_set_drvdata(pdev, netdev);
4392 	adapter = netdev_priv(netdev);
4393 	adapter->netdev = netdev;
4394 	adapter->pdev = pdev;
4395 	hw = &adapter->hw;
4396 	hw->back = adapter;
4397 	adapter->port_num = hw->bus.func;
4398 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4399 
4400 	err = pci_save_state(pdev);
4401 	if (err)
4402 		goto err_ioremap;
4403 
4404 	err = -EIO;
4405 	adapter->io_addr = ioremap(pci_resource_start(pdev, 0),
4406 				   pci_resource_len(pdev, 0));
4407 	if (!adapter->io_addr)
4408 		goto err_ioremap;
4409 
4410 	/* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */
4411 	hw->hw_addr = adapter->io_addr;
4412 
4413 	netdev->netdev_ops = &igc_netdev_ops;
4414 	igc_set_ethtool_ops(netdev);
4415 	netdev->watchdog_timeo = 5 * HZ;
4416 
4417 	netdev->mem_start = pci_resource_start(pdev, 0);
4418 	netdev->mem_end = pci_resource_end(pdev, 0);
4419 
4420 	/* PCI config space info */
4421 	hw->vendor_id = pdev->vendor;
4422 	hw->device_id = pdev->device;
4423 	hw->revision_id = pdev->revision;
4424 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
4425 	hw->subsystem_device_id = pdev->subsystem_device;
4426 
4427 	/* Copy the default MAC and PHY function pointers */
4428 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
4429 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
4430 
4431 	/* Initialize skew-specific constants */
4432 	err = ei->get_invariants(hw);
4433 	if (err)
4434 		goto err_sw_init;
4435 
4436 	/* Add supported features to the features list*/
4437 	netdev->features |= NETIF_F_RXCSUM;
4438 	netdev->features |= NETIF_F_HW_CSUM;
4439 	netdev->features |= NETIF_F_SCTP_CRC;
4440 
4441 	/* setup the private structure */
4442 	err = igc_sw_init(adapter);
4443 	if (err)
4444 		goto err_sw_init;
4445 
4446 	/* copy netdev features into list of user selectable features */
4447 	netdev->hw_features |= NETIF_F_NTUPLE;
4448 	netdev->hw_features |= netdev->features;
4449 
4450 	/* MTU range: 68 - 9216 */
4451 	netdev->min_mtu = ETH_MIN_MTU;
4452 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
4453 
4454 	/* before reading the NVM, reset the controller to put the device in a
4455 	 * known good starting state
4456 	 */
4457 	hw->mac.ops.reset_hw(hw);
4458 
4459 	if (igc_get_flash_presence_i225(hw)) {
4460 		if (hw->nvm.ops.validate(hw) < 0) {
4461 			dev_err(&pdev->dev,
4462 				"The NVM Checksum Is Not Valid\n");
4463 			err = -EIO;
4464 			goto err_eeprom;
4465 		}
4466 	}
4467 
4468 	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
4469 		/* copy the MAC address out of the NVM */
4470 		if (hw->mac.ops.read_mac_addr(hw))
4471 			dev_err(&pdev->dev, "NVM Read Error\n");
4472 	}
4473 
4474 	memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
4475 
4476 	if (!is_valid_ether_addr(netdev->dev_addr)) {
4477 		dev_err(&pdev->dev, "Invalid MAC Address\n");
4478 		err = -EIO;
4479 		goto err_eeprom;
4480 	}
4481 
4482 	/* configure RXPBSIZE and TXPBSIZE */
4483 	wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT);
4484 	wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT);
4485 
4486 	timer_setup(&adapter->watchdog_timer, igc_watchdog, 0);
4487 	timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0);
4488 
4489 	INIT_WORK(&adapter->reset_task, igc_reset_task);
4490 	INIT_WORK(&adapter->watchdog_task, igc_watchdog_task);
4491 
4492 	/* Initialize link properties that are user-changeable */
4493 	adapter->fc_autoneg = true;
4494 	hw->mac.autoneg = true;
4495 	hw->phy.autoneg_advertised = 0xaf;
4496 
4497 	hw->fc.requested_mode = igc_fc_default;
4498 	hw->fc.current_mode = igc_fc_default;
4499 
4500 	/* reset the hardware with the new settings */
4501 	igc_reset(adapter);
4502 
4503 	/* let the f/w know that the h/w is now under the control of the
4504 	 * driver.
4505 	 */
4506 	igc_get_hw_control(adapter);
4507 
4508 	strncpy(netdev->name, "eth%d", IFNAMSIZ);
4509 	err = register_netdev(netdev);
4510 	if (err)
4511 		goto err_register;
4512 
4513 	 /* carrier off reporting is important to ethtool even BEFORE open */
4514 	netif_carrier_off(netdev);
4515 
4516 	/* Check if Media Autosense is enabled */
4517 	adapter->ei = *ei;
4518 
4519 	/* print pcie link status and MAC address */
4520 	pcie_print_link_status(pdev);
4521 	netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr);
4522 
4523 	return 0;
4524 
4525 err_register:
4526 	igc_release_hw_control(adapter);
4527 err_eeprom:
4528 	if (!igc_check_reset_block(hw))
4529 		igc_reset_phy(hw);
4530 err_sw_init:
4531 	igc_clear_interrupt_scheme(adapter);
4532 	iounmap(adapter->io_addr);
4533 err_ioremap:
4534 	free_netdev(netdev);
4535 err_alloc_etherdev:
4536 	pci_release_selected_regions(pdev,
4537 				     pci_select_bars(pdev, IORESOURCE_MEM));
4538 err_pci_reg:
4539 err_dma:
4540 	pci_disable_device(pdev);
4541 	return err;
4542 }
4543 
4544 /**
4545  * igc_remove - Device Removal Routine
4546  * @pdev: PCI device information struct
4547  *
4548  * igc_remove is called by the PCI subsystem to alert the driver
4549  * that it should release a PCI device.  This could be caused by a
4550  * Hot-Plug event, or because the driver is going to be removed from
4551  * memory.
4552  */
4553 static void igc_remove(struct pci_dev *pdev)
4554 {
4555 	struct net_device *netdev = pci_get_drvdata(pdev);
4556 	struct igc_adapter *adapter = netdev_priv(netdev);
4557 
4558 	set_bit(__IGC_DOWN, &adapter->state);
4559 
4560 	del_timer_sync(&adapter->watchdog_timer);
4561 	del_timer_sync(&adapter->phy_info_timer);
4562 
4563 	cancel_work_sync(&adapter->reset_task);
4564 	cancel_work_sync(&adapter->watchdog_task);
4565 
4566 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
4567 	 * would have already happened in close and is redundant.
4568 	 */
4569 	igc_release_hw_control(adapter);
4570 	unregister_netdev(netdev);
4571 
4572 	igc_clear_interrupt_scheme(adapter);
4573 	pci_iounmap(pdev, adapter->io_addr);
4574 	pci_release_mem_regions(pdev);
4575 
4576 	kfree(adapter->mac_table);
4577 	free_netdev(netdev);
4578 
4579 	pci_disable_pcie_error_reporting(pdev);
4580 
4581 	pci_disable_device(pdev);
4582 }
4583 
4584 static struct pci_driver igc_driver = {
4585 	.name     = igc_driver_name,
4586 	.id_table = igc_pci_tbl,
4587 	.probe    = igc_probe,
4588 	.remove   = igc_remove,
4589 };
4590 
4591 void igc_set_flag_queue_pairs(struct igc_adapter *adapter,
4592 			      const u32 max_rss_queues)
4593 {
4594 	/* Determine if we need to pair queues. */
4595 	/* If rss_queues > half of max_rss_queues, pair the queues in
4596 	 * order to conserve interrupts due to limited supply.
4597 	 */
4598 	if (adapter->rss_queues > (max_rss_queues / 2))
4599 		adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4600 	else
4601 		adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS;
4602 }
4603 
4604 unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter)
4605 {
4606 	unsigned int max_rss_queues;
4607 
4608 	/* Determine the maximum number of RSS queues supported. */
4609 	max_rss_queues = IGC_MAX_RX_QUEUES;
4610 
4611 	return max_rss_queues;
4612 }
4613 
4614 static void igc_init_queue_configuration(struct igc_adapter *adapter)
4615 {
4616 	u32 max_rss_queues;
4617 
4618 	max_rss_queues = igc_get_max_rss_queues(adapter);
4619 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
4620 
4621 	igc_set_flag_queue_pairs(adapter, max_rss_queues);
4622 }
4623 
4624 /**
4625  * igc_sw_init - Initialize general software structures (struct igc_adapter)
4626  * @adapter: board private structure to initialize
4627  *
4628  * igc_sw_init initializes the Adapter private data structure.
4629  * Fields are initialized based on PCI device information and
4630  * OS network device settings (MTU size).
4631  */
4632 static int igc_sw_init(struct igc_adapter *adapter)
4633 {
4634 	struct net_device *netdev = adapter->netdev;
4635 	struct pci_dev *pdev = adapter->pdev;
4636 	struct igc_hw *hw = &adapter->hw;
4637 
4638 	int size = sizeof(struct igc_mac_addr) * hw->mac.rar_entry_count;
4639 
4640 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4641 
4642 	/* set default ring sizes */
4643 	adapter->tx_ring_count = IGC_DEFAULT_TXD;
4644 	adapter->rx_ring_count = IGC_DEFAULT_RXD;
4645 
4646 	/* set default ITR values */
4647 	adapter->rx_itr_setting = IGC_DEFAULT_ITR;
4648 	adapter->tx_itr_setting = IGC_DEFAULT_ITR;
4649 
4650 	/* set default work limits */
4651 	adapter->tx_work_limit = IGC_DEFAULT_TX_WORK;
4652 
4653 	/* adjust max frame to be at least the size of a standard frame */
4654 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
4655 				VLAN_HLEN;
4656 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4657 
4658 	spin_lock_init(&adapter->nfc_lock);
4659 	spin_lock_init(&adapter->stats64_lock);
4660 	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
4661 	adapter->flags |= IGC_FLAG_HAS_MSIX;
4662 
4663 	adapter->mac_table = kzalloc(size, GFP_ATOMIC);
4664 	if (!adapter->mac_table)
4665 		return -ENOMEM;
4666 
4667 	igc_init_queue_configuration(adapter);
4668 
4669 	/* This call may decrease the number of queues */
4670 	if (igc_init_interrupt_scheme(adapter, true)) {
4671 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4672 		return -ENOMEM;
4673 	}
4674 
4675 	/* Explicitly disable IRQ since the NIC can be in any state. */
4676 	igc_irq_disable(adapter);
4677 
4678 	set_bit(__IGC_DOWN, &adapter->state);
4679 
4680 	return 0;
4681 }
4682 
4683 /**
4684  * igc_reinit_queues - return error
4685  * @adapter: pointer to adapter structure
4686  */
4687 int igc_reinit_queues(struct igc_adapter *adapter)
4688 {
4689 	struct net_device *netdev = adapter->netdev;
4690 	struct pci_dev *pdev = adapter->pdev;
4691 	int err = 0;
4692 
4693 	if (netif_running(netdev))
4694 		igc_close(netdev);
4695 
4696 	igc_reset_interrupt_capability(adapter);
4697 
4698 	if (igc_init_interrupt_scheme(adapter, true)) {
4699 		dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4700 		return -ENOMEM;
4701 	}
4702 
4703 	if (netif_running(netdev))
4704 		err = igc_open(netdev);
4705 
4706 	return err;
4707 }
4708 
4709 /**
4710  * igc_get_hw_dev - return device
4711  * @hw: pointer to hardware structure
4712  *
4713  * used by hardware layer to print debugging information
4714  */
4715 struct net_device *igc_get_hw_dev(struct igc_hw *hw)
4716 {
4717 	struct igc_adapter *adapter = hw->back;
4718 
4719 	return adapter->netdev;
4720 }
4721 
4722 /**
4723  * igc_init_module - Driver Registration Routine
4724  *
4725  * igc_init_module is the first routine called when the driver is
4726  * loaded. All it does is register with the PCI subsystem.
4727  */
4728 static int __init igc_init_module(void)
4729 {
4730 	int ret;
4731 
4732 	pr_info("%s - version %s\n",
4733 		igc_driver_string, igc_driver_version);
4734 
4735 	pr_info("%s\n", igc_copyright);
4736 
4737 	ret = pci_register_driver(&igc_driver);
4738 	return ret;
4739 }
4740 
4741 module_init(igc_init_module);
4742 
4743 /**
4744  * igc_exit_module - Driver Exit Cleanup Routine
4745  *
4746  * igc_exit_module is called just before the driver is removed
4747  * from memory.
4748  */
4749 static void __exit igc_exit_module(void)
4750 {
4751 	pci_unregister_driver(&igc_driver);
4752 }
4753 
4754 module_exit(igc_exit_module);
4755 /* igc_main.c */
4756