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/tcp.h>
8 #include <linux/udp.h>
9 #include <linux/ip.h>
10 #include <linux/pm_runtime.h>
11 #include <net/pkt_sched.h>
12 #include <linux/bpf_trace.h>
13 #include <net/xdp_sock_drv.h>
14 #include <linux/pci.h>
15 
16 #include <net/ipv6.h>
17 
18 #include "igc.h"
19 #include "igc_hw.h"
20 #include "igc_tsn.h"
21 #include "igc_xdp.h"
22 
23 #define DRV_SUMMARY	"Intel(R) 2.5G Ethernet Linux Driver"
24 
25 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
26 
27 #define IGC_XDP_PASS		0
28 #define IGC_XDP_CONSUMED	BIT(0)
29 #define IGC_XDP_TX		BIT(1)
30 #define IGC_XDP_REDIRECT	BIT(2)
31 
32 static int debug = -1;
33 
34 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
35 MODULE_DESCRIPTION(DRV_SUMMARY);
36 MODULE_LICENSE("GPL v2");
37 module_param(debug, int, 0);
38 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
39 
40 char igc_driver_name[] = "igc";
41 static const char igc_driver_string[] = DRV_SUMMARY;
42 static const char igc_copyright[] =
43 	"Copyright(c) 2018 Intel Corporation.";
44 
45 static const struct igc_info *igc_info_tbl[] = {
46 	[board_base] = &igc_base_info,
47 };
48 
49 static const struct pci_device_id igc_pci_tbl[] = {
50 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base },
51 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base },
52 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base },
53 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base },
54 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base },
55 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K2), board_base },
56 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_K), board_base },
57 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LMVP), board_base },
58 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LMVP), board_base },
59 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_IT), board_base },
60 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LM), board_base },
61 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_V), board_base },
62 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_IT), board_base },
63 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I221_V), board_base },
64 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_BLANK_NVM), board_base },
65 	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_BLANK_NVM), board_base },
66 	/* required last entry */
67 	{0, }
68 };
69 
70 MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
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 net_device *dev = adapter->netdev;
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 		netdev_err(dev, "Error on hardware initialization\n");
109 
110 	/* Re-establish EEE setting */
111 	igc_set_eee_i225(hw, true, true, true);
112 
113 	if (!netif_running(adapter->netdev))
114 		igc_power_down_phy_copper_base(&adapter->hw);
115 
116 	/* Enable HW to recognize an 802.1Q VLAN Ethernet packet */
117 	wr32(IGC_VET, ETH_P_8021Q);
118 
119 	/* Re-enable PTP, where applicable. */
120 	igc_ptp_reset(adapter);
121 
122 	/* Re-enable TSN offloading, where applicable. */
123 	igc_tsn_reset(adapter);
124 
125 	igc_get_phy_info(hw);
126 }
127 
128 /**
129  * igc_power_up_link - Power up the phy link
130  * @adapter: address of board private structure
131  */
132 static void igc_power_up_link(struct igc_adapter *adapter)
133 {
134 	igc_reset_phy(&adapter->hw);
135 
136 	igc_power_up_phy_copper(&adapter->hw);
137 
138 	igc_setup_link(&adapter->hw);
139 }
140 
141 /**
142  * igc_release_hw_control - release control of the h/w to f/w
143  * @adapter: address of board private structure
144  *
145  * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
146  * For ASF and Pass Through versions of f/w this means that the
147  * driver is no longer loaded.
148  */
149 static void igc_release_hw_control(struct igc_adapter *adapter)
150 {
151 	struct igc_hw *hw = &adapter->hw;
152 	u32 ctrl_ext;
153 
154 	if (!pci_device_is_present(adapter->pdev))
155 		return;
156 
157 	/* Let firmware take over control of h/w */
158 	ctrl_ext = rd32(IGC_CTRL_EXT);
159 	wr32(IGC_CTRL_EXT,
160 	     ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
161 }
162 
163 /**
164  * igc_get_hw_control - get control of the h/w from f/w
165  * @adapter: address of board private structure
166  *
167  * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
168  * For ASF and Pass Through versions of f/w this means that
169  * the driver is loaded.
170  */
171 static void igc_get_hw_control(struct igc_adapter *adapter)
172 {
173 	struct igc_hw *hw = &adapter->hw;
174 	u32 ctrl_ext;
175 
176 	/* Let firmware know the driver has taken over */
177 	ctrl_ext = rd32(IGC_CTRL_EXT);
178 	wr32(IGC_CTRL_EXT,
179 	     ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
180 }
181 
182 static void igc_unmap_tx_buffer(struct device *dev, struct igc_tx_buffer *buf)
183 {
184 	dma_unmap_single(dev, dma_unmap_addr(buf, dma),
185 			 dma_unmap_len(buf, len), DMA_TO_DEVICE);
186 
187 	dma_unmap_len_set(buf, len, 0);
188 }
189 
190 /**
191  * igc_clean_tx_ring - Free Tx Buffers
192  * @tx_ring: ring to be cleaned
193  */
194 static void igc_clean_tx_ring(struct igc_ring *tx_ring)
195 {
196 	u16 i = tx_ring->next_to_clean;
197 	struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
198 	u32 xsk_frames = 0;
199 
200 	while (i != tx_ring->next_to_use) {
201 		union igc_adv_tx_desc *eop_desc, *tx_desc;
202 
203 		switch (tx_buffer->type) {
204 		case IGC_TX_BUFFER_TYPE_XSK:
205 			xsk_frames++;
206 			break;
207 		case IGC_TX_BUFFER_TYPE_XDP:
208 			xdp_return_frame(tx_buffer->xdpf);
209 			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
210 			break;
211 		case IGC_TX_BUFFER_TYPE_SKB:
212 			dev_kfree_skb_any(tx_buffer->skb);
213 			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
214 			break;
215 		default:
216 			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
217 			break;
218 		}
219 
220 		/* check for eop_desc to determine the end of the packet */
221 		eop_desc = tx_buffer->next_to_watch;
222 		tx_desc = IGC_TX_DESC(tx_ring, i);
223 
224 		/* unmap remaining buffers */
225 		while (tx_desc != eop_desc) {
226 			tx_buffer++;
227 			tx_desc++;
228 			i++;
229 			if (unlikely(i == tx_ring->count)) {
230 				i = 0;
231 				tx_buffer = tx_ring->tx_buffer_info;
232 				tx_desc = IGC_TX_DESC(tx_ring, 0);
233 			}
234 
235 			/* unmap any remaining paged data */
236 			if (dma_unmap_len(tx_buffer, len))
237 				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
238 		}
239 
240 		tx_buffer->next_to_watch = NULL;
241 
242 		/* move us one more past the eop_desc for start of next pkt */
243 		tx_buffer++;
244 		i++;
245 		if (unlikely(i == tx_ring->count)) {
246 			i = 0;
247 			tx_buffer = tx_ring->tx_buffer_info;
248 		}
249 	}
250 
251 	if (tx_ring->xsk_pool && xsk_frames)
252 		xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
253 
254 	/* reset BQL for queue */
255 	netdev_tx_reset_queue(txring_txq(tx_ring));
256 
257 	/* Zero out the buffer ring */
258 	memset(tx_ring->tx_buffer_info, 0,
259 	       sizeof(*tx_ring->tx_buffer_info) * tx_ring->count);
260 
261 	/* Zero out the descriptor ring */
262 	memset(tx_ring->desc, 0, tx_ring->size);
263 
264 	/* reset next_to_use and next_to_clean */
265 	tx_ring->next_to_use = 0;
266 	tx_ring->next_to_clean = 0;
267 }
268 
269 /**
270  * igc_free_tx_resources - Free Tx Resources per Queue
271  * @tx_ring: Tx descriptor ring for a specific queue
272  *
273  * Free all transmit software resources
274  */
275 void igc_free_tx_resources(struct igc_ring *tx_ring)
276 {
277 	igc_disable_tx_ring(tx_ring);
278 
279 	vfree(tx_ring->tx_buffer_info);
280 	tx_ring->tx_buffer_info = NULL;
281 
282 	/* if not set, then don't free */
283 	if (!tx_ring->desc)
284 		return;
285 
286 	dma_free_coherent(tx_ring->dev, tx_ring->size,
287 			  tx_ring->desc, tx_ring->dma);
288 
289 	tx_ring->desc = NULL;
290 }
291 
292 /**
293  * igc_free_all_tx_resources - Free Tx Resources for All Queues
294  * @adapter: board private structure
295  *
296  * Free all transmit software resources
297  */
298 static void igc_free_all_tx_resources(struct igc_adapter *adapter)
299 {
300 	int i;
301 
302 	for (i = 0; i < adapter->num_tx_queues; i++)
303 		igc_free_tx_resources(adapter->tx_ring[i]);
304 }
305 
306 /**
307  * igc_clean_all_tx_rings - Free Tx Buffers for all queues
308  * @adapter: board private structure
309  */
310 static void igc_clean_all_tx_rings(struct igc_adapter *adapter)
311 {
312 	int i;
313 
314 	for (i = 0; i < adapter->num_tx_queues; i++)
315 		if (adapter->tx_ring[i])
316 			igc_clean_tx_ring(adapter->tx_ring[i]);
317 }
318 
319 static void igc_disable_tx_ring_hw(struct igc_ring *ring)
320 {
321 	struct igc_hw *hw = &ring->q_vector->adapter->hw;
322 	u8 idx = ring->reg_idx;
323 	u32 txdctl;
324 
325 	txdctl = rd32(IGC_TXDCTL(idx));
326 	txdctl &= ~IGC_TXDCTL_QUEUE_ENABLE;
327 	txdctl |= IGC_TXDCTL_SWFLUSH;
328 	wr32(IGC_TXDCTL(idx), txdctl);
329 }
330 
331 /**
332  * igc_disable_all_tx_rings_hw - Disable all transmit queue operation
333  * @adapter: board private structure
334  */
335 static void igc_disable_all_tx_rings_hw(struct igc_adapter *adapter)
336 {
337 	int i;
338 
339 	for (i = 0; i < adapter->num_tx_queues; i++) {
340 		struct igc_ring *tx_ring = adapter->tx_ring[i];
341 
342 		igc_disable_tx_ring_hw(tx_ring);
343 	}
344 }
345 
346 /**
347  * igc_setup_tx_resources - allocate Tx resources (Descriptors)
348  * @tx_ring: tx descriptor ring (for a specific queue) to setup
349  *
350  * Return 0 on success, negative on failure
351  */
352 int igc_setup_tx_resources(struct igc_ring *tx_ring)
353 {
354 	struct net_device *ndev = tx_ring->netdev;
355 	struct device *dev = tx_ring->dev;
356 	int size = 0;
357 
358 	size = sizeof(struct igc_tx_buffer) * tx_ring->count;
359 	tx_ring->tx_buffer_info = vzalloc(size);
360 	if (!tx_ring->tx_buffer_info)
361 		goto err;
362 
363 	/* round up to nearest 4K */
364 	tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc);
365 	tx_ring->size = ALIGN(tx_ring->size, 4096);
366 
367 	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
368 					   &tx_ring->dma, GFP_KERNEL);
369 
370 	if (!tx_ring->desc)
371 		goto err;
372 
373 	tx_ring->next_to_use = 0;
374 	tx_ring->next_to_clean = 0;
375 
376 	return 0;
377 
378 err:
379 	vfree(tx_ring->tx_buffer_info);
380 	netdev_err(ndev, "Unable to allocate memory for Tx descriptor ring\n");
381 	return -ENOMEM;
382 }
383 
384 /**
385  * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues
386  * @adapter: board private structure
387  *
388  * Return 0 on success, negative on failure
389  */
390 static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
391 {
392 	struct net_device *dev = adapter->netdev;
393 	int i, err = 0;
394 
395 	for (i = 0; i < adapter->num_tx_queues; i++) {
396 		err = igc_setup_tx_resources(adapter->tx_ring[i]);
397 		if (err) {
398 			netdev_err(dev, "Error on Tx queue %u setup\n", i);
399 			for (i--; i >= 0; i--)
400 				igc_free_tx_resources(adapter->tx_ring[i]);
401 			break;
402 		}
403 	}
404 
405 	return err;
406 }
407 
408 static void igc_clean_rx_ring_page_shared(struct igc_ring *rx_ring)
409 {
410 	u16 i = rx_ring->next_to_clean;
411 
412 	dev_kfree_skb(rx_ring->skb);
413 	rx_ring->skb = NULL;
414 
415 	/* Free all the Rx ring sk_buffs */
416 	while (i != rx_ring->next_to_alloc) {
417 		struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
418 
419 		/* Invalidate cache lines that may have been written to by
420 		 * device so that we avoid corrupting memory.
421 		 */
422 		dma_sync_single_range_for_cpu(rx_ring->dev,
423 					      buffer_info->dma,
424 					      buffer_info->page_offset,
425 					      igc_rx_bufsz(rx_ring),
426 					      DMA_FROM_DEVICE);
427 
428 		/* free resources associated with mapping */
429 		dma_unmap_page_attrs(rx_ring->dev,
430 				     buffer_info->dma,
431 				     igc_rx_pg_size(rx_ring),
432 				     DMA_FROM_DEVICE,
433 				     IGC_RX_DMA_ATTR);
434 		__page_frag_cache_drain(buffer_info->page,
435 					buffer_info->pagecnt_bias);
436 
437 		i++;
438 		if (i == rx_ring->count)
439 			i = 0;
440 	}
441 }
442 
443 static void igc_clean_rx_ring_xsk_pool(struct igc_ring *ring)
444 {
445 	struct igc_rx_buffer *bi;
446 	u16 i;
447 
448 	for (i = 0; i < ring->count; i++) {
449 		bi = &ring->rx_buffer_info[i];
450 		if (!bi->xdp)
451 			continue;
452 
453 		xsk_buff_free(bi->xdp);
454 		bi->xdp = NULL;
455 	}
456 }
457 
458 /**
459  * igc_clean_rx_ring - Free Rx Buffers per Queue
460  * @ring: ring to free buffers from
461  */
462 static void igc_clean_rx_ring(struct igc_ring *ring)
463 {
464 	if (ring->xsk_pool)
465 		igc_clean_rx_ring_xsk_pool(ring);
466 	else
467 		igc_clean_rx_ring_page_shared(ring);
468 
469 	clear_ring_uses_large_buffer(ring);
470 
471 	ring->next_to_alloc = 0;
472 	ring->next_to_clean = 0;
473 	ring->next_to_use = 0;
474 }
475 
476 /**
477  * igc_clean_all_rx_rings - Free Rx Buffers for all queues
478  * @adapter: board private structure
479  */
480 static void igc_clean_all_rx_rings(struct igc_adapter *adapter)
481 {
482 	int i;
483 
484 	for (i = 0; i < adapter->num_rx_queues; i++)
485 		if (adapter->rx_ring[i])
486 			igc_clean_rx_ring(adapter->rx_ring[i]);
487 }
488 
489 /**
490  * igc_free_rx_resources - Free Rx Resources
491  * @rx_ring: ring to clean the resources from
492  *
493  * Free all receive software resources
494  */
495 void igc_free_rx_resources(struct igc_ring *rx_ring)
496 {
497 	igc_clean_rx_ring(rx_ring);
498 
499 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
500 
501 	vfree(rx_ring->rx_buffer_info);
502 	rx_ring->rx_buffer_info = NULL;
503 
504 	/* if not set, then don't free */
505 	if (!rx_ring->desc)
506 		return;
507 
508 	dma_free_coherent(rx_ring->dev, rx_ring->size,
509 			  rx_ring->desc, rx_ring->dma);
510 
511 	rx_ring->desc = NULL;
512 }
513 
514 /**
515  * igc_free_all_rx_resources - Free Rx Resources for All Queues
516  * @adapter: board private structure
517  *
518  * Free all receive software resources
519  */
520 static void igc_free_all_rx_resources(struct igc_adapter *adapter)
521 {
522 	int i;
523 
524 	for (i = 0; i < adapter->num_rx_queues; i++)
525 		igc_free_rx_resources(adapter->rx_ring[i]);
526 }
527 
528 /**
529  * igc_setup_rx_resources - allocate Rx resources (Descriptors)
530  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
531  *
532  * Returns 0 on success, negative on failure
533  */
534 int igc_setup_rx_resources(struct igc_ring *rx_ring)
535 {
536 	struct net_device *ndev = rx_ring->netdev;
537 	struct device *dev = rx_ring->dev;
538 	u8 index = rx_ring->queue_index;
539 	int size, desc_len, res;
540 
541 	/* XDP RX-queue info */
542 	if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
543 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
544 	res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, ndev, index,
545 			       rx_ring->q_vector->napi.napi_id);
546 	if (res < 0) {
547 		netdev_err(ndev, "Failed to register xdp_rxq index %u\n",
548 			   index);
549 		return res;
550 	}
551 
552 	size = sizeof(struct igc_rx_buffer) * rx_ring->count;
553 	rx_ring->rx_buffer_info = vzalloc(size);
554 	if (!rx_ring->rx_buffer_info)
555 		goto err;
556 
557 	desc_len = sizeof(union igc_adv_rx_desc);
558 
559 	/* Round up to nearest 4K */
560 	rx_ring->size = rx_ring->count * desc_len;
561 	rx_ring->size = ALIGN(rx_ring->size, 4096);
562 
563 	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
564 					   &rx_ring->dma, GFP_KERNEL);
565 
566 	if (!rx_ring->desc)
567 		goto err;
568 
569 	rx_ring->next_to_alloc = 0;
570 	rx_ring->next_to_clean = 0;
571 	rx_ring->next_to_use = 0;
572 
573 	return 0;
574 
575 err:
576 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
577 	vfree(rx_ring->rx_buffer_info);
578 	rx_ring->rx_buffer_info = NULL;
579 	netdev_err(ndev, "Unable to allocate memory for Rx descriptor ring\n");
580 	return -ENOMEM;
581 }
582 
583 /**
584  * igc_setup_all_rx_resources - wrapper to allocate Rx resources
585  *                                (Descriptors) for all queues
586  * @adapter: board private structure
587  *
588  * Return 0 on success, negative on failure
589  */
590 static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
591 {
592 	struct net_device *dev = adapter->netdev;
593 	int i, err = 0;
594 
595 	for (i = 0; i < adapter->num_rx_queues; i++) {
596 		err = igc_setup_rx_resources(adapter->rx_ring[i]);
597 		if (err) {
598 			netdev_err(dev, "Error on Rx queue %u setup\n", i);
599 			for (i--; i >= 0; i--)
600 				igc_free_rx_resources(adapter->rx_ring[i]);
601 			break;
602 		}
603 	}
604 
605 	return err;
606 }
607 
608 static struct xsk_buff_pool *igc_get_xsk_pool(struct igc_adapter *adapter,
609 					      struct igc_ring *ring)
610 {
611 	if (!igc_xdp_is_enabled(adapter) ||
612 	    !test_bit(IGC_RING_FLAG_AF_XDP_ZC, &ring->flags))
613 		return NULL;
614 
615 	return xsk_get_pool_from_qid(ring->netdev, ring->queue_index);
616 }
617 
618 /**
619  * igc_configure_rx_ring - Configure a receive ring after Reset
620  * @adapter: board private structure
621  * @ring: receive ring to be configured
622  *
623  * Configure the Rx unit of the MAC after a reset.
624  */
625 static void igc_configure_rx_ring(struct igc_adapter *adapter,
626 				  struct igc_ring *ring)
627 {
628 	struct igc_hw *hw = &adapter->hw;
629 	union igc_adv_rx_desc *rx_desc;
630 	int reg_idx = ring->reg_idx;
631 	u32 srrctl = 0, rxdctl = 0;
632 	u64 rdba = ring->dma;
633 	u32 buf_size;
634 
635 	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
636 	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
637 	if (ring->xsk_pool) {
638 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
639 						   MEM_TYPE_XSK_BUFF_POOL,
640 						   NULL));
641 		xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
642 	} else {
643 		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
644 						   MEM_TYPE_PAGE_SHARED,
645 						   NULL));
646 	}
647 
648 	if (igc_xdp_is_enabled(adapter))
649 		set_ring_uses_large_buffer(ring);
650 
651 	/* disable the queue */
652 	wr32(IGC_RXDCTL(reg_idx), 0);
653 
654 	/* Set DMA base address registers */
655 	wr32(IGC_RDBAL(reg_idx),
656 	     rdba & 0x00000000ffffffffULL);
657 	wr32(IGC_RDBAH(reg_idx), rdba >> 32);
658 	wr32(IGC_RDLEN(reg_idx),
659 	     ring->count * sizeof(union igc_adv_rx_desc));
660 
661 	/* initialize head and tail */
662 	ring->tail = adapter->io_addr + IGC_RDT(reg_idx);
663 	wr32(IGC_RDH(reg_idx), 0);
664 	writel(0, ring->tail);
665 
666 	/* reset next-to- use/clean to place SW in sync with hardware */
667 	ring->next_to_clean = 0;
668 	ring->next_to_use = 0;
669 
670 	if (ring->xsk_pool)
671 		buf_size = xsk_pool_get_rx_frame_size(ring->xsk_pool);
672 	else if (ring_uses_large_buffer(ring))
673 		buf_size = IGC_RXBUFFER_3072;
674 	else
675 		buf_size = IGC_RXBUFFER_2048;
676 
677 	srrctl = rd32(IGC_SRRCTL(reg_idx));
678 	srrctl &= ~(IGC_SRRCTL_BSIZEPKT_MASK | IGC_SRRCTL_BSIZEHDR_MASK |
679 		    IGC_SRRCTL_DESCTYPE_MASK);
680 	srrctl |= IGC_SRRCTL_BSIZEHDR(IGC_RX_HDR_LEN);
681 	srrctl |= IGC_SRRCTL_BSIZEPKT(buf_size);
682 	srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
683 
684 	wr32(IGC_SRRCTL(reg_idx), srrctl);
685 
686 	rxdctl |= IGC_RX_PTHRESH;
687 	rxdctl |= IGC_RX_HTHRESH << 8;
688 	rxdctl |= IGC_RX_WTHRESH << 16;
689 
690 	/* initialize rx_buffer_info */
691 	memset(ring->rx_buffer_info, 0,
692 	       sizeof(struct igc_rx_buffer) * ring->count);
693 
694 	/* initialize Rx descriptor 0 */
695 	rx_desc = IGC_RX_DESC(ring, 0);
696 	rx_desc->wb.upper.length = 0;
697 
698 	/* enable receive descriptor fetching */
699 	rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
700 
701 	wr32(IGC_RXDCTL(reg_idx), rxdctl);
702 }
703 
704 /**
705  * igc_configure_rx - Configure receive Unit after Reset
706  * @adapter: board private structure
707  *
708  * Configure the Rx unit of the MAC after a reset.
709  */
710 static void igc_configure_rx(struct igc_adapter *adapter)
711 {
712 	int i;
713 
714 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
715 	 * the Base and Length of the Rx Descriptor Ring
716 	 */
717 	for (i = 0; i < adapter->num_rx_queues; i++)
718 		igc_configure_rx_ring(adapter, adapter->rx_ring[i]);
719 }
720 
721 /**
722  * igc_configure_tx_ring - Configure transmit ring after Reset
723  * @adapter: board private structure
724  * @ring: tx ring to configure
725  *
726  * Configure a transmit ring after a reset.
727  */
728 static void igc_configure_tx_ring(struct igc_adapter *adapter,
729 				  struct igc_ring *ring)
730 {
731 	struct igc_hw *hw = &adapter->hw;
732 	int reg_idx = ring->reg_idx;
733 	u64 tdba = ring->dma;
734 	u32 txdctl = 0;
735 
736 	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
737 
738 	/* disable the queue */
739 	wr32(IGC_TXDCTL(reg_idx), 0);
740 	wrfl();
741 
742 	wr32(IGC_TDLEN(reg_idx),
743 	     ring->count * sizeof(union igc_adv_tx_desc));
744 	wr32(IGC_TDBAL(reg_idx),
745 	     tdba & 0x00000000ffffffffULL);
746 	wr32(IGC_TDBAH(reg_idx), tdba >> 32);
747 
748 	ring->tail = adapter->io_addr + IGC_TDT(reg_idx);
749 	wr32(IGC_TDH(reg_idx), 0);
750 	writel(0, ring->tail);
751 
752 	txdctl |= IGC_TX_PTHRESH;
753 	txdctl |= IGC_TX_HTHRESH << 8;
754 	txdctl |= IGC_TX_WTHRESH << 16;
755 
756 	txdctl |= IGC_TXDCTL_QUEUE_ENABLE;
757 	wr32(IGC_TXDCTL(reg_idx), txdctl);
758 }
759 
760 /**
761  * igc_configure_tx - Configure transmit Unit after Reset
762  * @adapter: board private structure
763  *
764  * Configure the Tx unit of the MAC after a reset.
765  */
766 static void igc_configure_tx(struct igc_adapter *adapter)
767 {
768 	int i;
769 
770 	for (i = 0; i < adapter->num_tx_queues; i++)
771 		igc_configure_tx_ring(adapter, adapter->tx_ring[i]);
772 }
773 
774 /**
775  * igc_setup_mrqc - configure the multiple receive queue control registers
776  * @adapter: Board private structure
777  */
778 static void igc_setup_mrqc(struct igc_adapter *adapter)
779 {
780 	struct igc_hw *hw = &adapter->hw;
781 	u32 j, num_rx_queues;
782 	u32 mrqc, rxcsum;
783 	u32 rss_key[10];
784 
785 	netdev_rss_key_fill(rss_key, sizeof(rss_key));
786 	for (j = 0; j < 10; j++)
787 		wr32(IGC_RSSRK(j), rss_key[j]);
788 
789 	num_rx_queues = adapter->rss_queues;
790 
791 	if (adapter->rss_indir_tbl_init != num_rx_queues) {
792 		for (j = 0; j < IGC_RETA_SIZE; j++)
793 			adapter->rss_indir_tbl[j] =
794 			(j * num_rx_queues) / IGC_RETA_SIZE;
795 		adapter->rss_indir_tbl_init = num_rx_queues;
796 	}
797 	igc_write_rss_indir_tbl(adapter);
798 
799 	/* Disable raw packet checksumming so that RSS hash is placed in
800 	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
801 	 * offloads as they are enabled by default
802 	 */
803 	rxcsum = rd32(IGC_RXCSUM);
804 	rxcsum |= IGC_RXCSUM_PCSD;
805 
806 	/* Enable Receive Checksum Offload for SCTP */
807 	rxcsum |= IGC_RXCSUM_CRCOFL;
808 
809 	/* Don't need to set TUOFL or IPOFL, they default to 1 */
810 	wr32(IGC_RXCSUM, rxcsum);
811 
812 	/* Generate RSS hash based on packet types, TCP/UDP
813 	 * port numbers and/or IPv4/v6 src and dst addresses
814 	 */
815 	mrqc = IGC_MRQC_RSS_FIELD_IPV4 |
816 	       IGC_MRQC_RSS_FIELD_IPV4_TCP |
817 	       IGC_MRQC_RSS_FIELD_IPV6 |
818 	       IGC_MRQC_RSS_FIELD_IPV6_TCP |
819 	       IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
820 
821 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP)
822 		mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP;
823 	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP)
824 		mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP;
825 
826 	mrqc |= IGC_MRQC_ENABLE_RSS_MQ;
827 
828 	wr32(IGC_MRQC, mrqc);
829 }
830 
831 /**
832  * igc_setup_rctl - configure the receive control registers
833  * @adapter: Board private structure
834  */
835 static void igc_setup_rctl(struct igc_adapter *adapter)
836 {
837 	struct igc_hw *hw = &adapter->hw;
838 	u32 rctl;
839 
840 	rctl = rd32(IGC_RCTL);
841 
842 	rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
843 	rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC);
844 
845 	rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF |
846 		(hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
847 
848 	/* enable stripping of CRC. Newer features require
849 	 * that the HW strips the CRC.
850 	 */
851 	rctl |= IGC_RCTL_SECRC;
852 
853 	/* disable store bad packets and clear size bits. */
854 	rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256);
855 
856 	/* enable LPE to allow for reception of jumbo frames */
857 	rctl |= IGC_RCTL_LPE;
858 
859 	/* disable queue 0 to prevent tail write w/o re-config */
860 	wr32(IGC_RXDCTL(0), 0);
861 
862 	/* This is useful for sniffing bad packets. */
863 	if (adapter->netdev->features & NETIF_F_RXALL) {
864 		/* UPE and MPE will be handled by normal PROMISC logic
865 		 * in set_rx_mode
866 		 */
867 		rctl |= (IGC_RCTL_SBP | /* Receive bad packets */
868 			 IGC_RCTL_BAM | /* RX All Bcast Pkts */
869 			 IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
870 
871 		rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */
872 			  IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */
873 	}
874 
875 	wr32(IGC_RCTL, rctl);
876 }
877 
878 /**
879  * igc_setup_tctl - configure the transmit control registers
880  * @adapter: Board private structure
881  */
882 static void igc_setup_tctl(struct igc_adapter *adapter)
883 {
884 	struct igc_hw *hw = &adapter->hw;
885 	u32 tctl;
886 
887 	/* disable queue 0 which icould be enabled by default */
888 	wr32(IGC_TXDCTL(0), 0);
889 
890 	/* Program the Transmit Control Register */
891 	tctl = rd32(IGC_TCTL);
892 	tctl &= ~IGC_TCTL_CT;
893 	tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC |
894 		(IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT);
895 
896 	/* Enable transmits */
897 	tctl |= IGC_TCTL_EN;
898 
899 	wr32(IGC_TCTL, tctl);
900 }
901 
902 /**
903  * igc_set_mac_filter_hw() - Set MAC address filter in hardware
904  * @adapter: Pointer to adapter where the filter should be set
905  * @index: Filter index
906  * @type: MAC address filter type (source or destination)
907  * @addr: MAC address
908  * @queue: If non-negative, queue assignment feature is enabled and frames
909  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
910  *         assignment is disabled.
911  */
912 static void igc_set_mac_filter_hw(struct igc_adapter *adapter, int index,
913 				  enum igc_mac_filter_type type,
914 				  const u8 *addr, int queue)
915 {
916 	struct net_device *dev = adapter->netdev;
917 	struct igc_hw *hw = &adapter->hw;
918 	u32 ral, rah;
919 
920 	if (WARN_ON(index >= hw->mac.rar_entry_count))
921 		return;
922 
923 	ral = le32_to_cpup((__le32 *)(addr));
924 	rah = le16_to_cpup((__le16 *)(addr + 4));
925 
926 	if (type == IGC_MAC_FILTER_TYPE_SRC) {
927 		rah &= ~IGC_RAH_ASEL_MASK;
928 		rah |= IGC_RAH_ASEL_SRC_ADDR;
929 	}
930 
931 	if (queue >= 0) {
932 		rah &= ~IGC_RAH_QSEL_MASK;
933 		rah |= (queue << IGC_RAH_QSEL_SHIFT);
934 		rah |= IGC_RAH_QSEL_ENABLE;
935 	}
936 
937 	rah |= IGC_RAH_AV;
938 
939 	wr32(IGC_RAL(index), ral);
940 	wr32(IGC_RAH(index), rah);
941 
942 	netdev_dbg(dev, "MAC address filter set in HW: index %d", index);
943 }
944 
945 /**
946  * igc_clear_mac_filter_hw() - Clear MAC address filter in hardware
947  * @adapter: Pointer to adapter where the filter should be cleared
948  * @index: Filter index
949  */
950 static void igc_clear_mac_filter_hw(struct igc_adapter *adapter, int index)
951 {
952 	struct net_device *dev = adapter->netdev;
953 	struct igc_hw *hw = &adapter->hw;
954 
955 	if (WARN_ON(index >= hw->mac.rar_entry_count))
956 		return;
957 
958 	wr32(IGC_RAL(index), 0);
959 	wr32(IGC_RAH(index), 0);
960 
961 	netdev_dbg(dev, "MAC address filter cleared in HW: index %d", index);
962 }
963 
964 /* Set default MAC address for the PF in the first RAR entry */
965 static void igc_set_default_mac_filter(struct igc_adapter *adapter)
966 {
967 	struct net_device *dev = adapter->netdev;
968 	u8 *addr = adapter->hw.mac.addr;
969 
970 	netdev_dbg(dev, "Set default MAC address filter: address %pM", addr);
971 
972 	igc_set_mac_filter_hw(adapter, 0, IGC_MAC_FILTER_TYPE_DST, addr, -1);
973 }
974 
975 /**
976  * igc_set_mac - Change the Ethernet Address of the NIC
977  * @netdev: network interface device structure
978  * @p: pointer to an address structure
979  *
980  * Returns 0 on success, negative on failure
981  */
982 static int igc_set_mac(struct net_device *netdev, void *p)
983 {
984 	struct igc_adapter *adapter = netdev_priv(netdev);
985 	struct igc_hw *hw = &adapter->hw;
986 	struct sockaddr *addr = p;
987 
988 	if (!is_valid_ether_addr(addr->sa_data))
989 		return -EADDRNOTAVAIL;
990 
991 	eth_hw_addr_set(netdev, addr->sa_data);
992 	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
993 
994 	/* set the correct pool for the new PF MAC address in entry 0 */
995 	igc_set_default_mac_filter(adapter);
996 
997 	return 0;
998 }
999 
1000 /**
1001  *  igc_write_mc_addr_list - write multicast addresses to MTA
1002  *  @netdev: network interface device structure
1003  *
1004  *  Writes multicast address list to the MTA hash table.
1005  *  Returns: -ENOMEM on failure
1006  *           0 on no addresses written
1007  *           X on writing X addresses to MTA
1008  **/
1009 static int igc_write_mc_addr_list(struct net_device *netdev)
1010 {
1011 	struct igc_adapter *adapter = netdev_priv(netdev);
1012 	struct igc_hw *hw = &adapter->hw;
1013 	struct netdev_hw_addr *ha;
1014 	u8  *mta_list;
1015 	int i;
1016 
1017 	if (netdev_mc_empty(netdev)) {
1018 		/* nothing to program, so clear mc list */
1019 		igc_update_mc_addr_list(hw, NULL, 0);
1020 		return 0;
1021 	}
1022 
1023 	mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
1024 	if (!mta_list)
1025 		return -ENOMEM;
1026 
1027 	/* The shared function expects a packed array of only addresses. */
1028 	i = 0;
1029 	netdev_for_each_mc_addr(ha, netdev)
1030 		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1031 
1032 	igc_update_mc_addr_list(hw, mta_list, i);
1033 	kfree(mta_list);
1034 
1035 	return netdev_mc_count(netdev);
1036 }
1037 
1038 static __le32 igc_tx_launchtime(struct igc_ring *ring, ktime_t txtime,
1039 				bool *first_flag, bool *insert_empty)
1040 {
1041 	struct igc_adapter *adapter = netdev_priv(ring->netdev);
1042 	ktime_t cycle_time = adapter->cycle_time;
1043 	ktime_t base_time = adapter->base_time;
1044 	ktime_t now = ktime_get_clocktai();
1045 	ktime_t baset_est, end_of_cycle;
1046 	s32 launchtime;
1047 	s64 n;
1048 
1049 	n = div64_s64(ktime_sub_ns(now, base_time), cycle_time);
1050 
1051 	baset_est = ktime_add_ns(base_time, cycle_time * (n));
1052 	end_of_cycle = ktime_add_ns(baset_est, cycle_time);
1053 
1054 	if (ktime_compare(txtime, end_of_cycle) >= 0) {
1055 		if (baset_est != ring->last_ff_cycle) {
1056 			*first_flag = true;
1057 			ring->last_ff_cycle = baset_est;
1058 
1059 			if (ktime_compare(end_of_cycle, ring->last_tx_cycle) > 0)
1060 				*insert_empty = true;
1061 		}
1062 	}
1063 
1064 	/* Introducing a window at end of cycle on which packets
1065 	 * potentially not honor launchtime. Window of 5us chosen
1066 	 * considering software update the tail pointer and packets
1067 	 * are dma'ed to packet buffer.
1068 	 */
1069 	if ((ktime_sub_ns(end_of_cycle, now) < 5 * NSEC_PER_USEC))
1070 		netdev_warn(ring->netdev, "Packet with txtime=%llu may not be honoured\n",
1071 			    txtime);
1072 
1073 	ring->last_tx_cycle = end_of_cycle;
1074 
1075 	launchtime = ktime_sub_ns(txtime, baset_est);
1076 	if (launchtime > 0)
1077 		div_s64_rem(launchtime, cycle_time, &launchtime);
1078 	else
1079 		launchtime = 0;
1080 
1081 	return cpu_to_le32(launchtime);
1082 }
1083 
1084 static int igc_init_empty_frame(struct igc_ring *ring,
1085 				struct igc_tx_buffer *buffer,
1086 				struct sk_buff *skb)
1087 {
1088 	unsigned int size;
1089 	dma_addr_t dma;
1090 
1091 	size = skb_headlen(skb);
1092 
1093 	dma = dma_map_single(ring->dev, skb->data, size, DMA_TO_DEVICE);
1094 	if (dma_mapping_error(ring->dev, dma)) {
1095 		netdev_err_once(ring->netdev, "Failed to map DMA for TX\n");
1096 		return -ENOMEM;
1097 	}
1098 
1099 	buffer->skb = skb;
1100 	buffer->protocol = 0;
1101 	buffer->bytecount = skb->len;
1102 	buffer->gso_segs = 1;
1103 	buffer->time_stamp = jiffies;
1104 	dma_unmap_len_set(buffer, len, skb->len);
1105 	dma_unmap_addr_set(buffer, dma, dma);
1106 
1107 	return 0;
1108 }
1109 
1110 static int igc_init_tx_empty_descriptor(struct igc_ring *ring,
1111 					struct sk_buff *skb,
1112 					struct igc_tx_buffer *first)
1113 {
1114 	union igc_adv_tx_desc *desc;
1115 	u32 cmd_type, olinfo_status;
1116 	int err;
1117 
1118 	if (!igc_desc_unused(ring))
1119 		return -EBUSY;
1120 
1121 	err = igc_init_empty_frame(ring, first, skb);
1122 	if (err)
1123 		return err;
1124 
1125 	cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
1126 		   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
1127 		   first->bytecount;
1128 	olinfo_status = first->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
1129 
1130 	desc = IGC_TX_DESC(ring, ring->next_to_use);
1131 	desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1132 	desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1133 	desc->read.buffer_addr = cpu_to_le64(dma_unmap_addr(first, dma));
1134 
1135 	netdev_tx_sent_queue(txring_txq(ring), skb->len);
1136 
1137 	first->next_to_watch = desc;
1138 
1139 	ring->next_to_use++;
1140 	if (ring->next_to_use == ring->count)
1141 		ring->next_to_use = 0;
1142 
1143 	return 0;
1144 }
1145 
1146 #define IGC_EMPTY_FRAME_SIZE 60
1147 
1148 static void igc_tx_ctxtdesc(struct igc_ring *tx_ring,
1149 			    __le32 launch_time, bool first_flag,
1150 			    u32 vlan_macip_lens, u32 type_tucmd,
1151 			    u32 mss_l4len_idx)
1152 {
1153 	struct igc_adv_tx_context_desc *context_desc;
1154 	u16 i = tx_ring->next_to_use;
1155 
1156 	context_desc = IGC_TX_CTXTDESC(tx_ring, i);
1157 
1158 	i++;
1159 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1160 
1161 	/* set bits to identify this as an advanced context descriptor */
1162 	type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
1163 
1164 	/* For i225, context index must be unique per ring. */
1165 	if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
1166 		mss_l4len_idx |= tx_ring->reg_idx << 4;
1167 
1168 	if (first_flag)
1169 		mss_l4len_idx |= IGC_ADVTXD_TSN_CNTX_FIRST;
1170 
1171 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
1172 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
1173 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
1174 	context_desc->launch_time	= launch_time;
1175 }
1176 
1177 static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first,
1178 			__le32 launch_time, bool first_flag)
1179 {
1180 	struct sk_buff *skb = first->skb;
1181 	u32 vlan_macip_lens = 0;
1182 	u32 type_tucmd = 0;
1183 
1184 	if (skb->ip_summed != CHECKSUM_PARTIAL) {
1185 csum_failed:
1186 		if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) &&
1187 		    !tx_ring->launchtime_enable)
1188 			return;
1189 		goto no_csum;
1190 	}
1191 
1192 	switch (skb->csum_offset) {
1193 	case offsetof(struct tcphdr, check):
1194 		type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1195 		fallthrough;
1196 	case offsetof(struct udphdr, check):
1197 		break;
1198 	case offsetof(struct sctphdr, checksum):
1199 		/* validate that this is actually an SCTP request */
1200 		if (skb_csum_is_sctp(skb)) {
1201 			type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP;
1202 			break;
1203 		}
1204 		fallthrough;
1205 	default:
1206 		skb_checksum_help(skb);
1207 		goto csum_failed;
1208 	}
1209 
1210 	/* update TX checksum flag */
1211 	first->tx_flags |= IGC_TX_FLAGS_CSUM;
1212 	vlan_macip_lens = skb_checksum_start_offset(skb) -
1213 			  skb_network_offset(skb);
1214 no_csum:
1215 	vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT;
1216 	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1217 
1218 	igc_tx_ctxtdesc(tx_ring, launch_time, first_flag,
1219 			vlan_macip_lens, type_tucmd, 0);
1220 }
1221 
1222 static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1223 {
1224 	struct net_device *netdev = tx_ring->netdev;
1225 
1226 	netif_stop_subqueue(netdev, tx_ring->queue_index);
1227 
1228 	/* memory barriier comment */
1229 	smp_mb();
1230 
1231 	/* We need to check again in a case another CPU has just
1232 	 * made room available.
1233 	 */
1234 	if (igc_desc_unused(tx_ring) < size)
1235 		return -EBUSY;
1236 
1237 	/* A reprieve! */
1238 	netif_wake_subqueue(netdev, tx_ring->queue_index);
1239 
1240 	u64_stats_update_begin(&tx_ring->tx_syncp2);
1241 	tx_ring->tx_stats.restart_queue2++;
1242 	u64_stats_update_end(&tx_ring->tx_syncp2);
1243 
1244 	return 0;
1245 }
1246 
1247 static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1248 {
1249 	if (igc_desc_unused(tx_ring) >= size)
1250 		return 0;
1251 	return __igc_maybe_stop_tx(tx_ring, size);
1252 }
1253 
1254 #define IGC_SET_FLAG(_input, _flag, _result) \
1255 	(((_flag) <= (_result)) ?				\
1256 	 ((u32)((_input) & (_flag)) * ((_result) / (_flag))) :	\
1257 	 ((u32)((_input) & (_flag)) / ((_flag) / (_result))))
1258 
1259 static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
1260 {
1261 	/* set type for advanced descriptor with frame checksum insertion */
1262 	u32 cmd_type = IGC_ADVTXD_DTYP_DATA |
1263 		       IGC_ADVTXD_DCMD_DEXT |
1264 		       IGC_ADVTXD_DCMD_IFCS;
1265 
1266 	/* set HW vlan bit if vlan is present */
1267 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_VLAN,
1268 				 IGC_ADVTXD_DCMD_VLE);
1269 
1270 	/* set segmentation bits for TSO */
1271 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO,
1272 				 (IGC_ADVTXD_DCMD_TSE));
1273 
1274 	/* set timestamp bit if present, will select the register set
1275 	 * based on the _TSTAMP(_X) bit.
1276 	 */
1277 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP,
1278 				 (IGC_ADVTXD_MAC_TSTAMP));
1279 
1280 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_1,
1281 				 (IGC_ADVTXD_TSTAMP_REG_1));
1282 
1283 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_2,
1284 				 (IGC_ADVTXD_TSTAMP_REG_2));
1285 
1286 	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_3,
1287 				 (IGC_ADVTXD_TSTAMP_REG_3));
1288 
1289 	/* insert frame checksum */
1290 	cmd_type ^= IGC_SET_FLAG(skb->no_fcs, 1, IGC_ADVTXD_DCMD_IFCS);
1291 
1292 	return cmd_type;
1293 }
1294 
1295 static void igc_tx_olinfo_status(struct igc_ring *tx_ring,
1296 				 union igc_adv_tx_desc *tx_desc,
1297 				 u32 tx_flags, unsigned int paylen)
1298 {
1299 	u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT;
1300 
1301 	/* insert L4 checksum */
1302 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_CSUM) *
1303 			  ((IGC_TXD_POPTS_TXSM << 8) /
1304 			  IGC_TX_FLAGS_CSUM);
1305 
1306 	/* insert IPv4 checksum */
1307 	olinfo_status |= (tx_flags & IGC_TX_FLAGS_IPV4) *
1308 			  (((IGC_TXD_POPTS_IXSM << 8)) /
1309 			  IGC_TX_FLAGS_IPV4);
1310 
1311 	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1312 }
1313 
1314 static int igc_tx_map(struct igc_ring *tx_ring,
1315 		      struct igc_tx_buffer *first,
1316 		      const u8 hdr_len)
1317 {
1318 	struct sk_buff *skb = first->skb;
1319 	struct igc_tx_buffer *tx_buffer;
1320 	union igc_adv_tx_desc *tx_desc;
1321 	u32 tx_flags = first->tx_flags;
1322 	skb_frag_t *frag;
1323 	u16 i = tx_ring->next_to_use;
1324 	unsigned int data_len, size;
1325 	dma_addr_t dma;
1326 	u32 cmd_type;
1327 
1328 	cmd_type = igc_tx_cmd_type(skb, tx_flags);
1329 	tx_desc = IGC_TX_DESC(tx_ring, i);
1330 
1331 	igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
1332 
1333 	size = skb_headlen(skb);
1334 	data_len = skb->data_len;
1335 
1336 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1337 
1338 	tx_buffer = first;
1339 
1340 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1341 		if (dma_mapping_error(tx_ring->dev, dma))
1342 			goto dma_error;
1343 
1344 		/* record length, and DMA address */
1345 		dma_unmap_len_set(tx_buffer, len, size);
1346 		dma_unmap_addr_set(tx_buffer, dma, dma);
1347 
1348 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
1349 
1350 		while (unlikely(size > IGC_MAX_DATA_PER_TXD)) {
1351 			tx_desc->read.cmd_type_len =
1352 				cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD);
1353 
1354 			i++;
1355 			tx_desc++;
1356 			if (i == tx_ring->count) {
1357 				tx_desc = IGC_TX_DESC(tx_ring, 0);
1358 				i = 0;
1359 			}
1360 			tx_desc->read.olinfo_status = 0;
1361 
1362 			dma += IGC_MAX_DATA_PER_TXD;
1363 			size -= IGC_MAX_DATA_PER_TXD;
1364 
1365 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
1366 		}
1367 
1368 		if (likely(!data_len))
1369 			break;
1370 
1371 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
1372 
1373 		i++;
1374 		tx_desc++;
1375 		if (i == tx_ring->count) {
1376 			tx_desc = IGC_TX_DESC(tx_ring, 0);
1377 			i = 0;
1378 		}
1379 		tx_desc->read.olinfo_status = 0;
1380 
1381 		size = skb_frag_size(frag);
1382 		data_len -= size;
1383 
1384 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
1385 				       size, DMA_TO_DEVICE);
1386 
1387 		tx_buffer = &tx_ring->tx_buffer_info[i];
1388 	}
1389 
1390 	/* write last descriptor with RS and EOP bits */
1391 	cmd_type |= size | IGC_TXD_DCMD;
1392 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1393 
1394 	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1395 
1396 	/* set the timestamp */
1397 	first->time_stamp = jiffies;
1398 
1399 	skb_tx_timestamp(skb);
1400 
1401 	/* Force memory writes to complete before letting h/w know there
1402 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1403 	 * memory model archs, such as IA-64).
1404 	 *
1405 	 * We also need this memory barrier to make certain all of the
1406 	 * status bits have been updated before next_to_watch is written.
1407 	 */
1408 	wmb();
1409 
1410 	/* set next_to_watch value indicating a packet is present */
1411 	first->next_to_watch = tx_desc;
1412 
1413 	i++;
1414 	if (i == tx_ring->count)
1415 		i = 0;
1416 
1417 	tx_ring->next_to_use = i;
1418 
1419 	/* Make sure there is space in the ring for the next send. */
1420 	igc_maybe_stop_tx(tx_ring, DESC_NEEDED);
1421 
1422 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1423 		writel(i, tx_ring->tail);
1424 	}
1425 
1426 	return 0;
1427 dma_error:
1428 	netdev_err(tx_ring->netdev, "TX DMA map failed\n");
1429 	tx_buffer = &tx_ring->tx_buffer_info[i];
1430 
1431 	/* clear dma mappings for failed tx_buffer_info map */
1432 	while (tx_buffer != first) {
1433 		if (dma_unmap_len(tx_buffer, len))
1434 			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
1435 
1436 		if (i-- == 0)
1437 			i += tx_ring->count;
1438 		tx_buffer = &tx_ring->tx_buffer_info[i];
1439 	}
1440 
1441 	if (dma_unmap_len(tx_buffer, len))
1442 		igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
1443 
1444 	dev_kfree_skb_any(tx_buffer->skb);
1445 	tx_buffer->skb = NULL;
1446 
1447 	tx_ring->next_to_use = i;
1448 
1449 	return -1;
1450 }
1451 
1452 static int igc_tso(struct igc_ring *tx_ring,
1453 		   struct igc_tx_buffer *first,
1454 		   __le32 launch_time, bool first_flag,
1455 		   u8 *hdr_len)
1456 {
1457 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1458 	struct sk_buff *skb = first->skb;
1459 	union {
1460 		struct iphdr *v4;
1461 		struct ipv6hdr *v6;
1462 		unsigned char *hdr;
1463 	} ip;
1464 	union {
1465 		struct tcphdr *tcp;
1466 		struct udphdr *udp;
1467 		unsigned char *hdr;
1468 	} l4;
1469 	u32 paylen, l4_offset;
1470 	int err;
1471 
1472 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1473 		return 0;
1474 
1475 	if (!skb_is_gso(skb))
1476 		return 0;
1477 
1478 	err = skb_cow_head(skb, 0);
1479 	if (err < 0)
1480 		return err;
1481 
1482 	ip.hdr = skb_network_header(skb);
1483 	l4.hdr = skb_checksum_start(skb);
1484 
1485 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1486 	type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1487 
1488 	/* initialize outer IP header fields */
1489 	if (ip.v4->version == 4) {
1490 		unsigned char *csum_start = skb_checksum_start(skb);
1491 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
1492 
1493 		/* IP header will have to cancel out any data that
1494 		 * is not a part of the outer IP header
1495 		 */
1496 		ip.v4->check = csum_fold(csum_partial(trans_start,
1497 						      csum_start - trans_start,
1498 						      0));
1499 		type_tucmd |= IGC_ADVTXD_TUCMD_IPV4;
1500 
1501 		ip.v4->tot_len = 0;
1502 		first->tx_flags |= IGC_TX_FLAGS_TSO |
1503 				   IGC_TX_FLAGS_CSUM |
1504 				   IGC_TX_FLAGS_IPV4;
1505 	} else {
1506 		ip.v6->payload_len = 0;
1507 		first->tx_flags |= IGC_TX_FLAGS_TSO |
1508 				   IGC_TX_FLAGS_CSUM;
1509 	}
1510 
1511 	/* determine offset of inner transport header */
1512 	l4_offset = l4.hdr - skb->data;
1513 
1514 	/* remove payload length from inner checksum */
1515 	paylen = skb->len - l4_offset;
1516 	if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) {
1517 		/* compute length of segmentation header */
1518 		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
1519 		csum_replace_by_diff(&l4.tcp->check,
1520 				     (__force __wsum)htonl(paylen));
1521 	} else {
1522 		/* compute length of segmentation header */
1523 		*hdr_len = sizeof(*l4.udp) + l4_offset;
1524 		csum_replace_by_diff(&l4.udp->check,
1525 				     (__force __wsum)htonl(paylen));
1526 	}
1527 
1528 	/* update gso size and bytecount with header size */
1529 	first->gso_segs = skb_shinfo(skb)->gso_segs;
1530 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
1531 
1532 	/* MSS L4LEN IDX */
1533 	mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT;
1534 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT;
1535 
1536 	/* VLAN MACLEN IPLEN */
1537 	vlan_macip_lens = l4.hdr - ip.hdr;
1538 	vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT;
1539 	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1540 
1541 	igc_tx_ctxtdesc(tx_ring, launch_time, first_flag,
1542 			vlan_macip_lens, type_tucmd, mss_l4len_idx);
1543 
1544 	return 1;
1545 }
1546 
1547 static bool igc_request_tx_tstamp(struct igc_adapter *adapter, struct sk_buff *skb, u32 *flags)
1548 {
1549 	int i;
1550 
1551 	for (i = 0; i < IGC_MAX_TX_TSTAMP_REGS; i++) {
1552 		struct igc_tx_timestamp_request *tstamp = &adapter->tx_tstamp[i];
1553 
1554 		if (tstamp->skb)
1555 			continue;
1556 
1557 		tstamp->skb = skb_get(skb);
1558 		tstamp->start = jiffies;
1559 		*flags = tstamp->flags;
1560 
1561 		return true;
1562 	}
1563 
1564 	return false;
1565 }
1566 
1567 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
1568 				       struct igc_ring *tx_ring)
1569 {
1570 	struct igc_adapter *adapter = netdev_priv(tx_ring->netdev);
1571 	bool first_flag = false, insert_empty = false;
1572 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1573 	__be16 protocol = vlan_get_protocol(skb);
1574 	struct igc_tx_buffer *first;
1575 	__le32 launch_time = 0;
1576 	u32 tx_flags = 0;
1577 	unsigned short f;
1578 	ktime_t txtime;
1579 	u8 hdr_len = 0;
1580 	int tso = 0;
1581 
1582 	/* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
1583 	 *	+ 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
1584 	 *	+ 2 desc gap to keep tail from touching head,
1585 	 *	+ 1 desc for context descriptor,
1586 	 * otherwise try next time
1587 	 */
1588 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1589 		count += TXD_USE_COUNT(skb_frag_size(
1590 						&skb_shinfo(skb)->frags[f]));
1591 
1592 	if (igc_maybe_stop_tx(tx_ring, count + 5)) {
1593 		/* this is a hard error */
1594 		return NETDEV_TX_BUSY;
1595 	}
1596 
1597 	if (!tx_ring->launchtime_enable)
1598 		goto done;
1599 
1600 	txtime = skb->tstamp;
1601 	skb->tstamp = ktime_set(0, 0);
1602 	launch_time = igc_tx_launchtime(tx_ring, txtime, &first_flag, &insert_empty);
1603 
1604 	if (insert_empty) {
1605 		struct igc_tx_buffer *empty_info;
1606 		struct sk_buff *empty;
1607 		void *data;
1608 
1609 		empty_info = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1610 		empty = alloc_skb(IGC_EMPTY_FRAME_SIZE, GFP_ATOMIC);
1611 		if (!empty)
1612 			goto done;
1613 
1614 		data = skb_put(empty, IGC_EMPTY_FRAME_SIZE);
1615 		memset(data, 0, IGC_EMPTY_FRAME_SIZE);
1616 
1617 		igc_tx_ctxtdesc(tx_ring, 0, false, 0, 0, 0);
1618 
1619 		if (igc_init_tx_empty_descriptor(tx_ring,
1620 						 empty,
1621 						 empty_info) < 0)
1622 			dev_kfree_skb_any(empty);
1623 	}
1624 
1625 done:
1626 	/* record the location of the first descriptor for this packet */
1627 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1628 	first->type = IGC_TX_BUFFER_TYPE_SKB;
1629 	first->skb = skb;
1630 	first->bytecount = skb->len;
1631 	first->gso_segs = 1;
1632 
1633 	if (adapter->qbv_transition || tx_ring->oper_gate_closed)
1634 		goto out_drop;
1635 
1636 	if (tx_ring->max_sdu > 0 && first->bytecount > tx_ring->max_sdu) {
1637 		adapter->stats.txdrop++;
1638 		goto out_drop;
1639 	}
1640 
1641 	if (unlikely(test_bit(IGC_RING_FLAG_TX_HWTSTAMP, &tx_ring->flags) &&
1642 		     skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
1643 		/* FIXME: add support for retrieving timestamps from
1644 		 * the other timer registers before skipping the
1645 		 * timestamping request.
1646 		 */
1647 		unsigned long flags;
1648 		u32 tstamp_flags;
1649 
1650 		spin_lock_irqsave(&adapter->ptp_tx_lock, flags);
1651 		if (igc_request_tx_tstamp(adapter, skb, &tstamp_flags)) {
1652 			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1653 			tx_flags |= IGC_TX_FLAGS_TSTAMP | tstamp_flags;
1654 		} else {
1655 			adapter->tx_hwtstamp_skipped++;
1656 		}
1657 
1658 		spin_unlock_irqrestore(&adapter->ptp_tx_lock, flags);
1659 	}
1660 
1661 	if (skb_vlan_tag_present(skb)) {
1662 		tx_flags |= IGC_TX_FLAGS_VLAN;
1663 		tx_flags |= (skb_vlan_tag_get(skb) << IGC_TX_FLAGS_VLAN_SHIFT);
1664 	}
1665 
1666 	/* record initial flags and protocol */
1667 	first->tx_flags = tx_flags;
1668 	first->protocol = protocol;
1669 
1670 	tso = igc_tso(tx_ring, first, launch_time, first_flag, &hdr_len);
1671 	if (tso < 0)
1672 		goto out_drop;
1673 	else if (!tso)
1674 		igc_tx_csum(tx_ring, first, launch_time, first_flag);
1675 
1676 	igc_tx_map(tx_ring, first, hdr_len);
1677 
1678 	return NETDEV_TX_OK;
1679 
1680 out_drop:
1681 	dev_kfree_skb_any(first->skb);
1682 	first->skb = NULL;
1683 
1684 	return NETDEV_TX_OK;
1685 }
1686 
1687 static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter,
1688 						    struct sk_buff *skb)
1689 {
1690 	unsigned int r_idx = skb->queue_mapping;
1691 
1692 	if (r_idx >= adapter->num_tx_queues)
1693 		r_idx = r_idx % adapter->num_tx_queues;
1694 
1695 	return adapter->tx_ring[r_idx];
1696 }
1697 
1698 static netdev_tx_t igc_xmit_frame(struct sk_buff *skb,
1699 				  struct net_device *netdev)
1700 {
1701 	struct igc_adapter *adapter = netdev_priv(netdev);
1702 
1703 	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
1704 	 * in order to meet this minimum size requirement.
1705 	 */
1706 	if (skb->len < 17) {
1707 		if (skb_padto(skb, 17))
1708 			return NETDEV_TX_OK;
1709 		skb->len = 17;
1710 	}
1711 
1712 	return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb));
1713 }
1714 
1715 static void igc_rx_checksum(struct igc_ring *ring,
1716 			    union igc_adv_rx_desc *rx_desc,
1717 			    struct sk_buff *skb)
1718 {
1719 	skb_checksum_none_assert(skb);
1720 
1721 	/* Ignore Checksum bit is set */
1722 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM))
1723 		return;
1724 
1725 	/* Rx checksum disabled via ethtool */
1726 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1727 		return;
1728 
1729 	/* TCP/UDP checksum error bit is set */
1730 	if (igc_test_staterr(rx_desc,
1731 			     IGC_RXDEXT_STATERR_L4E |
1732 			     IGC_RXDEXT_STATERR_IPE)) {
1733 		/* work around errata with sctp packets where the TCPE aka
1734 		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
1735 		 * packets (aka let the stack check the crc32c)
1736 		 */
1737 		if (!(skb->len == 60 &&
1738 		      test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
1739 			u64_stats_update_begin(&ring->rx_syncp);
1740 			ring->rx_stats.csum_err++;
1741 			u64_stats_update_end(&ring->rx_syncp);
1742 		}
1743 		/* let the stack verify checksum errors */
1744 		return;
1745 	}
1746 	/* It must be a TCP or UDP packet with a valid checksum */
1747 	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS |
1748 				      IGC_RXD_STAT_UDPCS))
1749 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1750 
1751 	netdev_dbg(ring->netdev, "cksum success: bits %08X\n",
1752 		   le32_to_cpu(rx_desc->wb.upper.status_error));
1753 }
1754 
1755 /* Mapping HW RSS Type to enum pkt_hash_types */
1756 static const enum pkt_hash_types igc_rss_type_table[IGC_RSS_TYPE_MAX_TABLE] = {
1757 	[IGC_RSS_TYPE_NO_HASH]		= PKT_HASH_TYPE_L2,
1758 	[IGC_RSS_TYPE_HASH_TCP_IPV4]	= PKT_HASH_TYPE_L4,
1759 	[IGC_RSS_TYPE_HASH_IPV4]	= PKT_HASH_TYPE_L3,
1760 	[IGC_RSS_TYPE_HASH_TCP_IPV6]	= PKT_HASH_TYPE_L4,
1761 	[IGC_RSS_TYPE_HASH_IPV6_EX]	= PKT_HASH_TYPE_L3,
1762 	[IGC_RSS_TYPE_HASH_IPV6]	= PKT_HASH_TYPE_L3,
1763 	[IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = PKT_HASH_TYPE_L4,
1764 	[IGC_RSS_TYPE_HASH_UDP_IPV4]	= PKT_HASH_TYPE_L4,
1765 	[IGC_RSS_TYPE_HASH_UDP_IPV6]	= PKT_HASH_TYPE_L4,
1766 	[IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = PKT_HASH_TYPE_L4,
1767 	[10] = PKT_HASH_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW  */
1768 	[11] = PKT_HASH_TYPE_NONE, /* keep array sized for SW bit-mask   */
1769 	[12] = PKT_HASH_TYPE_NONE, /* to handle future HW revisons       */
1770 	[13] = PKT_HASH_TYPE_NONE,
1771 	[14] = PKT_HASH_TYPE_NONE,
1772 	[15] = PKT_HASH_TYPE_NONE,
1773 };
1774 
1775 static inline void igc_rx_hash(struct igc_ring *ring,
1776 			       union igc_adv_rx_desc *rx_desc,
1777 			       struct sk_buff *skb)
1778 {
1779 	if (ring->netdev->features & NETIF_F_RXHASH) {
1780 		u32 rss_hash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1781 		u32 rss_type = igc_rss_type(rx_desc);
1782 
1783 		skb_set_hash(skb, rss_hash, igc_rss_type_table[rss_type]);
1784 	}
1785 }
1786 
1787 static void igc_rx_vlan(struct igc_ring *rx_ring,
1788 			union igc_adv_rx_desc *rx_desc,
1789 			struct sk_buff *skb)
1790 {
1791 	struct net_device *dev = rx_ring->netdev;
1792 	u16 vid;
1793 
1794 	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1795 	    igc_test_staterr(rx_desc, IGC_RXD_STAT_VP)) {
1796 		if (igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_LB) &&
1797 		    test_bit(IGC_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
1798 			vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
1799 		else
1800 			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1801 
1802 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1803 	}
1804 }
1805 
1806 /**
1807  * igc_process_skb_fields - Populate skb header fields from Rx descriptor
1808  * @rx_ring: rx descriptor ring packet is being transacted on
1809  * @rx_desc: pointer to the EOP Rx descriptor
1810  * @skb: pointer to current skb being populated
1811  *
1812  * This function checks the ring, descriptor, and packet information in order
1813  * to populate the hash, checksum, VLAN, protocol, and other fields within the
1814  * skb.
1815  */
1816 static void igc_process_skb_fields(struct igc_ring *rx_ring,
1817 				   union igc_adv_rx_desc *rx_desc,
1818 				   struct sk_buff *skb)
1819 {
1820 	igc_rx_hash(rx_ring, rx_desc, skb);
1821 
1822 	igc_rx_checksum(rx_ring, rx_desc, skb);
1823 
1824 	igc_rx_vlan(rx_ring, rx_desc, skb);
1825 
1826 	skb_record_rx_queue(skb, rx_ring->queue_index);
1827 
1828 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1829 }
1830 
1831 static void igc_vlan_mode(struct net_device *netdev, netdev_features_t features)
1832 {
1833 	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1834 	struct igc_adapter *adapter = netdev_priv(netdev);
1835 	struct igc_hw *hw = &adapter->hw;
1836 	u32 ctrl;
1837 
1838 	ctrl = rd32(IGC_CTRL);
1839 
1840 	if (enable) {
1841 		/* enable VLAN tag insert/strip */
1842 		ctrl |= IGC_CTRL_VME;
1843 	} else {
1844 		/* disable VLAN tag insert/strip */
1845 		ctrl &= ~IGC_CTRL_VME;
1846 	}
1847 	wr32(IGC_CTRL, ctrl);
1848 }
1849 
1850 static void igc_restore_vlan(struct igc_adapter *adapter)
1851 {
1852 	igc_vlan_mode(adapter->netdev, adapter->netdev->features);
1853 }
1854 
1855 static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring,
1856 					       const unsigned int size,
1857 					       int *rx_buffer_pgcnt)
1858 {
1859 	struct igc_rx_buffer *rx_buffer;
1860 
1861 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1862 	*rx_buffer_pgcnt =
1863 #if (PAGE_SIZE < 8192)
1864 		page_count(rx_buffer->page);
1865 #else
1866 		0;
1867 #endif
1868 	prefetchw(rx_buffer->page);
1869 
1870 	/* we are reusing so sync this buffer for CPU use */
1871 	dma_sync_single_range_for_cpu(rx_ring->dev,
1872 				      rx_buffer->dma,
1873 				      rx_buffer->page_offset,
1874 				      size,
1875 				      DMA_FROM_DEVICE);
1876 
1877 	rx_buffer->pagecnt_bias--;
1878 
1879 	return rx_buffer;
1880 }
1881 
1882 static void igc_rx_buffer_flip(struct igc_rx_buffer *buffer,
1883 			       unsigned int truesize)
1884 {
1885 #if (PAGE_SIZE < 8192)
1886 	buffer->page_offset ^= truesize;
1887 #else
1888 	buffer->page_offset += truesize;
1889 #endif
1890 }
1891 
1892 static unsigned int igc_get_rx_frame_truesize(struct igc_ring *ring,
1893 					      unsigned int size)
1894 {
1895 	unsigned int truesize;
1896 
1897 #if (PAGE_SIZE < 8192)
1898 	truesize = igc_rx_pg_size(ring) / 2;
1899 #else
1900 	truesize = ring_uses_build_skb(ring) ?
1901 		   SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1902 		   SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1903 		   SKB_DATA_ALIGN(size);
1904 #endif
1905 	return truesize;
1906 }
1907 
1908 /**
1909  * igc_add_rx_frag - Add contents of Rx buffer to sk_buff
1910  * @rx_ring: rx descriptor ring to transact packets on
1911  * @rx_buffer: buffer containing page to add
1912  * @skb: sk_buff to place the data into
1913  * @size: size of buffer to be added
1914  *
1915  * This function will add the data contained in rx_buffer->page to the skb.
1916  */
1917 static void igc_add_rx_frag(struct igc_ring *rx_ring,
1918 			    struct igc_rx_buffer *rx_buffer,
1919 			    struct sk_buff *skb,
1920 			    unsigned int size)
1921 {
1922 	unsigned int truesize;
1923 
1924 #if (PAGE_SIZE < 8192)
1925 	truesize = igc_rx_pg_size(rx_ring) / 2;
1926 #else
1927 	truesize = ring_uses_build_skb(rx_ring) ?
1928 		   SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1929 		   SKB_DATA_ALIGN(size);
1930 #endif
1931 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1932 			rx_buffer->page_offset, size, truesize);
1933 
1934 	igc_rx_buffer_flip(rx_buffer, truesize);
1935 }
1936 
1937 static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring,
1938 				     struct igc_rx_buffer *rx_buffer,
1939 				     struct xdp_buff *xdp)
1940 {
1941 	unsigned int size = xdp->data_end - xdp->data;
1942 	unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1943 	unsigned int metasize = xdp->data - xdp->data_meta;
1944 	struct sk_buff *skb;
1945 
1946 	/* prefetch first cache line of first page */
1947 	net_prefetch(xdp->data_meta);
1948 
1949 	/* build an skb around the page buffer */
1950 	skb = napi_build_skb(xdp->data_hard_start, truesize);
1951 	if (unlikely(!skb))
1952 		return NULL;
1953 
1954 	/* update pointers within the skb to store the data */
1955 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
1956 	__skb_put(skb, size);
1957 	if (metasize)
1958 		skb_metadata_set(skb, metasize);
1959 
1960 	igc_rx_buffer_flip(rx_buffer, truesize);
1961 	return skb;
1962 }
1963 
1964 static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring,
1965 					 struct igc_rx_buffer *rx_buffer,
1966 					 struct xdp_buff *xdp,
1967 					 ktime_t timestamp)
1968 {
1969 	unsigned int metasize = xdp->data - xdp->data_meta;
1970 	unsigned int size = xdp->data_end - xdp->data;
1971 	unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1972 	void *va = xdp->data;
1973 	unsigned int headlen;
1974 	struct sk_buff *skb;
1975 
1976 	/* prefetch first cache line of first page */
1977 	net_prefetch(xdp->data_meta);
1978 
1979 	/* allocate a skb to store the frags */
1980 	skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1981 			     IGC_RX_HDR_LEN + metasize);
1982 	if (unlikely(!skb))
1983 		return NULL;
1984 
1985 	if (timestamp)
1986 		skb_hwtstamps(skb)->hwtstamp = timestamp;
1987 
1988 	/* Determine available headroom for copy */
1989 	headlen = size;
1990 	if (headlen > IGC_RX_HDR_LEN)
1991 		headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN);
1992 
1993 	/* align pull length to size of long to optimize memcpy performance */
1994 	memcpy(__skb_put(skb, headlen + metasize), xdp->data_meta,
1995 	       ALIGN(headlen + metasize, sizeof(long)));
1996 
1997 	if (metasize) {
1998 		skb_metadata_set(skb, metasize);
1999 		__skb_pull(skb, metasize);
2000 	}
2001 
2002 	/* update all of the pointers */
2003 	size -= headlen;
2004 	if (size) {
2005 		skb_add_rx_frag(skb, 0, rx_buffer->page,
2006 				(va + headlen) - page_address(rx_buffer->page),
2007 				size, truesize);
2008 		igc_rx_buffer_flip(rx_buffer, truesize);
2009 	} else {
2010 		rx_buffer->pagecnt_bias++;
2011 	}
2012 
2013 	return skb;
2014 }
2015 
2016 /**
2017  * igc_reuse_rx_page - page flip buffer and store it back on the ring
2018  * @rx_ring: rx descriptor ring to store buffers on
2019  * @old_buff: donor buffer to have page reused
2020  *
2021  * Synchronizes page for reuse by the adapter
2022  */
2023 static void igc_reuse_rx_page(struct igc_ring *rx_ring,
2024 			      struct igc_rx_buffer *old_buff)
2025 {
2026 	u16 nta = rx_ring->next_to_alloc;
2027 	struct igc_rx_buffer *new_buff;
2028 
2029 	new_buff = &rx_ring->rx_buffer_info[nta];
2030 
2031 	/* update, and store next to alloc */
2032 	nta++;
2033 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
2034 
2035 	/* Transfer page from old buffer to new buffer.
2036 	 * Move each member individually to avoid possible store
2037 	 * forwarding stalls.
2038 	 */
2039 	new_buff->dma		= old_buff->dma;
2040 	new_buff->page		= old_buff->page;
2041 	new_buff->page_offset	= old_buff->page_offset;
2042 	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
2043 }
2044 
2045 static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer,
2046 				  int rx_buffer_pgcnt)
2047 {
2048 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
2049 	struct page *page = rx_buffer->page;
2050 
2051 	/* avoid re-using remote and pfmemalloc pages */
2052 	if (!dev_page_is_reusable(page))
2053 		return false;
2054 
2055 #if (PAGE_SIZE < 8192)
2056 	/* if we are only owner of page we can reuse it */
2057 	if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
2058 		return false;
2059 #else
2060 #define IGC_LAST_OFFSET \
2061 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048)
2062 
2063 	if (rx_buffer->page_offset > IGC_LAST_OFFSET)
2064 		return false;
2065 #endif
2066 
2067 	/* If we have drained the page fragment pool we need to update
2068 	 * the pagecnt_bias and page count so that we fully restock the
2069 	 * number of references the driver holds.
2070 	 */
2071 	if (unlikely(pagecnt_bias == 1)) {
2072 		page_ref_add(page, USHRT_MAX - 1);
2073 		rx_buffer->pagecnt_bias = USHRT_MAX;
2074 	}
2075 
2076 	return true;
2077 }
2078 
2079 /**
2080  * igc_is_non_eop - process handling of non-EOP buffers
2081  * @rx_ring: Rx ring being processed
2082  * @rx_desc: Rx descriptor for current buffer
2083  *
2084  * This function updates next to clean.  If the buffer is an EOP buffer
2085  * this function exits returning false, otherwise it will place the
2086  * sk_buff in the next buffer to be chained and return true indicating
2087  * that this is in fact a non-EOP buffer.
2088  */
2089 static bool igc_is_non_eop(struct igc_ring *rx_ring,
2090 			   union igc_adv_rx_desc *rx_desc)
2091 {
2092 	u32 ntc = rx_ring->next_to_clean + 1;
2093 
2094 	/* fetch, update, and store next to clean */
2095 	ntc = (ntc < rx_ring->count) ? ntc : 0;
2096 	rx_ring->next_to_clean = ntc;
2097 
2098 	prefetch(IGC_RX_DESC(rx_ring, ntc));
2099 
2100 	if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP)))
2101 		return false;
2102 
2103 	return true;
2104 }
2105 
2106 /**
2107  * igc_cleanup_headers - Correct corrupted or empty headers
2108  * @rx_ring: rx descriptor ring packet is being transacted on
2109  * @rx_desc: pointer to the EOP Rx descriptor
2110  * @skb: pointer to current skb being fixed
2111  *
2112  * Address the case where we are pulling data in on pages only
2113  * and as such no data is present in the skb header.
2114  *
2115  * In addition if skb is not at least 60 bytes we need to pad it so that
2116  * it is large enough to qualify as a valid Ethernet frame.
2117  *
2118  * Returns true if an error was encountered and skb was freed.
2119  */
2120 static bool igc_cleanup_headers(struct igc_ring *rx_ring,
2121 				union igc_adv_rx_desc *rx_desc,
2122 				struct sk_buff *skb)
2123 {
2124 	/* XDP packets use error pointer so abort at this point */
2125 	if (IS_ERR(skb))
2126 		return true;
2127 
2128 	if (unlikely(igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_RXE))) {
2129 		struct net_device *netdev = rx_ring->netdev;
2130 
2131 		if (!(netdev->features & NETIF_F_RXALL)) {
2132 			dev_kfree_skb_any(skb);
2133 			return true;
2134 		}
2135 	}
2136 
2137 	/* if eth_skb_pad returns an error the skb was freed */
2138 	if (eth_skb_pad(skb))
2139 		return true;
2140 
2141 	return false;
2142 }
2143 
2144 static void igc_put_rx_buffer(struct igc_ring *rx_ring,
2145 			      struct igc_rx_buffer *rx_buffer,
2146 			      int rx_buffer_pgcnt)
2147 {
2148 	if (igc_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2149 		/* hand second half of page back to the ring */
2150 		igc_reuse_rx_page(rx_ring, rx_buffer);
2151 	} else {
2152 		/* We are not reusing the buffer so unmap it and free
2153 		 * any references we are holding to it
2154 		 */
2155 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2156 				     igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
2157 				     IGC_RX_DMA_ATTR);
2158 		__page_frag_cache_drain(rx_buffer->page,
2159 					rx_buffer->pagecnt_bias);
2160 	}
2161 
2162 	/* clear contents of rx_buffer */
2163 	rx_buffer->page = NULL;
2164 }
2165 
2166 static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring)
2167 {
2168 	struct igc_adapter *adapter = rx_ring->q_vector->adapter;
2169 
2170 	if (ring_uses_build_skb(rx_ring))
2171 		return IGC_SKB_PAD;
2172 	if (igc_xdp_is_enabled(adapter))
2173 		return XDP_PACKET_HEADROOM;
2174 
2175 	return 0;
2176 }
2177 
2178 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
2179 				  struct igc_rx_buffer *bi)
2180 {
2181 	struct page *page = bi->page;
2182 	dma_addr_t dma;
2183 
2184 	/* since we are recycling buffers we should seldom need to alloc */
2185 	if (likely(page))
2186 		return true;
2187 
2188 	/* alloc new page for storage */
2189 	page = dev_alloc_pages(igc_rx_pg_order(rx_ring));
2190 	if (unlikely(!page)) {
2191 		rx_ring->rx_stats.alloc_failed++;
2192 		return false;
2193 	}
2194 
2195 	/* map page for use */
2196 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
2197 				 igc_rx_pg_size(rx_ring),
2198 				 DMA_FROM_DEVICE,
2199 				 IGC_RX_DMA_ATTR);
2200 
2201 	/* if mapping failed free memory back to system since
2202 	 * there isn't much point in holding memory we can't use
2203 	 */
2204 	if (dma_mapping_error(rx_ring->dev, dma)) {
2205 		__free_page(page);
2206 
2207 		rx_ring->rx_stats.alloc_failed++;
2208 		return false;
2209 	}
2210 
2211 	bi->dma = dma;
2212 	bi->page = page;
2213 	bi->page_offset = igc_rx_offset(rx_ring);
2214 	page_ref_add(page, USHRT_MAX - 1);
2215 	bi->pagecnt_bias = USHRT_MAX;
2216 
2217 	return true;
2218 }
2219 
2220 /**
2221  * igc_alloc_rx_buffers - Replace used receive buffers; packet split
2222  * @rx_ring: rx descriptor ring
2223  * @cleaned_count: number of buffers to clean
2224  */
2225 static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
2226 {
2227 	union igc_adv_rx_desc *rx_desc;
2228 	u16 i = rx_ring->next_to_use;
2229 	struct igc_rx_buffer *bi;
2230 	u16 bufsz;
2231 
2232 	/* nothing to do */
2233 	if (!cleaned_count)
2234 		return;
2235 
2236 	rx_desc = IGC_RX_DESC(rx_ring, i);
2237 	bi = &rx_ring->rx_buffer_info[i];
2238 	i -= rx_ring->count;
2239 
2240 	bufsz = igc_rx_bufsz(rx_ring);
2241 
2242 	do {
2243 		if (!igc_alloc_mapped_page(rx_ring, bi))
2244 			break;
2245 
2246 		/* sync the buffer for use by the device */
2247 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
2248 						 bi->page_offset, bufsz,
2249 						 DMA_FROM_DEVICE);
2250 
2251 		/* Refresh the desc even if buffer_addrs didn't change
2252 		 * because each write-back erases this info.
2253 		 */
2254 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
2255 
2256 		rx_desc++;
2257 		bi++;
2258 		i++;
2259 		if (unlikely(!i)) {
2260 			rx_desc = IGC_RX_DESC(rx_ring, 0);
2261 			bi = rx_ring->rx_buffer_info;
2262 			i -= rx_ring->count;
2263 		}
2264 
2265 		/* clear the length for the next_to_use descriptor */
2266 		rx_desc->wb.upper.length = 0;
2267 
2268 		cleaned_count--;
2269 	} while (cleaned_count);
2270 
2271 	i += rx_ring->count;
2272 
2273 	if (rx_ring->next_to_use != i) {
2274 		/* record the next descriptor to use */
2275 		rx_ring->next_to_use = i;
2276 
2277 		/* update next to alloc since we have filled the ring */
2278 		rx_ring->next_to_alloc = i;
2279 
2280 		/* Force memory writes to complete before letting h/w
2281 		 * know there are new descriptors to fetch.  (Only
2282 		 * applicable for weak-ordered memory model archs,
2283 		 * such as IA-64).
2284 		 */
2285 		wmb();
2286 		writel(i, rx_ring->tail);
2287 	}
2288 }
2289 
2290 static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count)
2291 {
2292 	union igc_adv_rx_desc *desc;
2293 	u16 i = ring->next_to_use;
2294 	struct igc_rx_buffer *bi;
2295 	dma_addr_t dma;
2296 	bool ok = true;
2297 
2298 	if (!count)
2299 		return ok;
2300 
2301 	XSK_CHECK_PRIV_TYPE(struct igc_xdp_buff);
2302 
2303 	desc = IGC_RX_DESC(ring, i);
2304 	bi = &ring->rx_buffer_info[i];
2305 	i -= ring->count;
2306 
2307 	do {
2308 		bi->xdp = xsk_buff_alloc(ring->xsk_pool);
2309 		if (!bi->xdp) {
2310 			ok = false;
2311 			break;
2312 		}
2313 
2314 		dma = xsk_buff_xdp_get_dma(bi->xdp);
2315 		desc->read.pkt_addr = cpu_to_le64(dma);
2316 
2317 		desc++;
2318 		bi++;
2319 		i++;
2320 		if (unlikely(!i)) {
2321 			desc = IGC_RX_DESC(ring, 0);
2322 			bi = ring->rx_buffer_info;
2323 			i -= ring->count;
2324 		}
2325 
2326 		/* Clear the length for the next_to_use descriptor. */
2327 		desc->wb.upper.length = 0;
2328 
2329 		count--;
2330 	} while (count);
2331 
2332 	i += ring->count;
2333 
2334 	if (ring->next_to_use != i) {
2335 		ring->next_to_use = i;
2336 
2337 		/* Force memory writes to complete before letting h/w
2338 		 * know there are new descriptors to fetch.  (Only
2339 		 * applicable for weak-ordered memory model archs,
2340 		 * such as IA-64).
2341 		 */
2342 		wmb();
2343 		writel(i, ring->tail);
2344 	}
2345 
2346 	return ok;
2347 }
2348 
2349 /* This function requires __netif_tx_lock is held by the caller. */
2350 static int igc_xdp_init_tx_descriptor(struct igc_ring *ring,
2351 				      struct xdp_frame *xdpf)
2352 {
2353 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
2354 	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
2355 	u16 count, index = ring->next_to_use;
2356 	struct igc_tx_buffer *head = &ring->tx_buffer_info[index];
2357 	struct igc_tx_buffer *buffer = head;
2358 	union igc_adv_tx_desc *desc = IGC_TX_DESC(ring, index);
2359 	u32 olinfo_status, len = xdpf->len, cmd_type;
2360 	void *data = xdpf->data;
2361 	u16 i;
2362 
2363 	count = TXD_USE_COUNT(len);
2364 	for (i = 0; i < nr_frags; i++)
2365 		count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
2366 
2367 	if (igc_maybe_stop_tx(ring, count + 3)) {
2368 		/* this is a hard error */
2369 		return -EBUSY;
2370 	}
2371 
2372 	i = 0;
2373 	head->bytecount = xdp_get_frame_len(xdpf);
2374 	head->type = IGC_TX_BUFFER_TYPE_XDP;
2375 	head->gso_segs = 1;
2376 	head->xdpf = xdpf;
2377 
2378 	olinfo_status = head->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
2379 	desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2380 
2381 	for (;;) {
2382 		dma_addr_t dma;
2383 
2384 		dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE);
2385 		if (dma_mapping_error(ring->dev, dma)) {
2386 			netdev_err_once(ring->netdev,
2387 					"Failed to map DMA for TX\n");
2388 			goto unmap;
2389 		}
2390 
2391 		dma_unmap_len_set(buffer, len, len);
2392 		dma_unmap_addr_set(buffer, dma, dma);
2393 
2394 		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2395 			   IGC_ADVTXD_DCMD_IFCS | len;
2396 
2397 		desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2398 		desc->read.buffer_addr = cpu_to_le64(dma);
2399 
2400 		buffer->protocol = 0;
2401 
2402 		if (++index == ring->count)
2403 			index = 0;
2404 
2405 		if (i == nr_frags)
2406 			break;
2407 
2408 		buffer = &ring->tx_buffer_info[index];
2409 		desc = IGC_TX_DESC(ring, index);
2410 		desc->read.olinfo_status = 0;
2411 
2412 		data = skb_frag_address(&sinfo->frags[i]);
2413 		len = skb_frag_size(&sinfo->frags[i]);
2414 		i++;
2415 	}
2416 	desc->read.cmd_type_len |= cpu_to_le32(IGC_TXD_DCMD);
2417 
2418 	netdev_tx_sent_queue(txring_txq(ring), head->bytecount);
2419 	/* set the timestamp */
2420 	head->time_stamp = jiffies;
2421 	/* set next_to_watch value indicating a packet is present */
2422 	head->next_to_watch = desc;
2423 	ring->next_to_use = index;
2424 
2425 	return 0;
2426 
2427 unmap:
2428 	for (;;) {
2429 		buffer = &ring->tx_buffer_info[index];
2430 		if (dma_unmap_len(buffer, len))
2431 			dma_unmap_page(ring->dev,
2432 				       dma_unmap_addr(buffer, dma),
2433 				       dma_unmap_len(buffer, len),
2434 				       DMA_TO_DEVICE);
2435 		dma_unmap_len_set(buffer, len, 0);
2436 		if (buffer == head)
2437 			break;
2438 
2439 		if (!index)
2440 			index += ring->count;
2441 		index--;
2442 	}
2443 
2444 	return -ENOMEM;
2445 }
2446 
2447 static struct igc_ring *igc_xdp_get_tx_ring(struct igc_adapter *adapter,
2448 					    int cpu)
2449 {
2450 	int index = cpu;
2451 
2452 	if (unlikely(index < 0))
2453 		index = 0;
2454 
2455 	while (index >= adapter->num_tx_queues)
2456 		index -= adapter->num_tx_queues;
2457 
2458 	return adapter->tx_ring[index];
2459 }
2460 
2461 static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)
2462 {
2463 	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2464 	int cpu = smp_processor_id();
2465 	struct netdev_queue *nq;
2466 	struct igc_ring *ring;
2467 	int res;
2468 
2469 	if (unlikely(!xdpf))
2470 		return -EFAULT;
2471 
2472 	ring = igc_xdp_get_tx_ring(adapter, cpu);
2473 	nq = txring_txq(ring);
2474 
2475 	__netif_tx_lock(nq, cpu);
2476 	/* Avoid transmit queue timeout since we share it with the slow path */
2477 	txq_trans_cond_update(nq);
2478 	res = igc_xdp_init_tx_descriptor(ring, xdpf);
2479 	__netif_tx_unlock(nq);
2480 	return res;
2481 }
2482 
2483 /* This function assumes rcu_read_lock() is held by the caller. */
2484 static int __igc_xdp_run_prog(struct igc_adapter *adapter,
2485 			      struct bpf_prog *prog,
2486 			      struct xdp_buff *xdp)
2487 {
2488 	u32 act = bpf_prog_run_xdp(prog, xdp);
2489 
2490 	switch (act) {
2491 	case XDP_PASS:
2492 		return IGC_XDP_PASS;
2493 	case XDP_TX:
2494 		if (igc_xdp_xmit_back(adapter, xdp) < 0)
2495 			goto out_failure;
2496 		return IGC_XDP_TX;
2497 	case XDP_REDIRECT:
2498 		if (xdp_do_redirect(adapter->netdev, xdp, prog) < 0)
2499 			goto out_failure;
2500 		return IGC_XDP_REDIRECT;
2501 		break;
2502 	default:
2503 		bpf_warn_invalid_xdp_action(adapter->netdev, prog, act);
2504 		fallthrough;
2505 	case XDP_ABORTED:
2506 out_failure:
2507 		trace_xdp_exception(adapter->netdev, prog, act);
2508 		fallthrough;
2509 	case XDP_DROP:
2510 		return IGC_XDP_CONSUMED;
2511 	}
2512 }
2513 
2514 static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
2515 					struct xdp_buff *xdp)
2516 {
2517 	struct bpf_prog *prog;
2518 	int res;
2519 
2520 	prog = READ_ONCE(adapter->xdp_prog);
2521 	if (!prog) {
2522 		res = IGC_XDP_PASS;
2523 		goto out;
2524 	}
2525 
2526 	res = __igc_xdp_run_prog(adapter, prog, xdp);
2527 
2528 out:
2529 	return ERR_PTR(-res);
2530 }
2531 
2532 /* This function assumes __netif_tx_lock is held by the caller. */
2533 static void igc_flush_tx_descriptors(struct igc_ring *ring)
2534 {
2535 	/* Once tail pointer is updated, hardware can fetch the descriptors
2536 	 * any time so we issue a write membar here to ensure all memory
2537 	 * writes are complete before the tail pointer is updated.
2538 	 */
2539 	wmb();
2540 	writel(ring->next_to_use, ring->tail);
2541 }
2542 
2543 static void igc_finalize_xdp(struct igc_adapter *adapter, int status)
2544 {
2545 	int cpu = smp_processor_id();
2546 	struct netdev_queue *nq;
2547 	struct igc_ring *ring;
2548 
2549 	if (status & IGC_XDP_TX) {
2550 		ring = igc_xdp_get_tx_ring(adapter, cpu);
2551 		nq = txring_txq(ring);
2552 
2553 		__netif_tx_lock(nq, cpu);
2554 		igc_flush_tx_descriptors(ring);
2555 		__netif_tx_unlock(nq);
2556 	}
2557 
2558 	if (status & IGC_XDP_REDIRECT)
2559 		xdp_do_flush();
2560 }
2561 
2562 static void igc_update_rx_stats(struct igc_q_vector *q_vector,
2563 				unsigned int packets, unsigned int bytes)
2564 {
2565 	struct igc_ring *ring = q_vector->rx.ring;
2566 
2567 	u64_stats_update_begin(&ring->rx_syncp);
2568 	ring->rx_stats.packets += packets;
2569 	ring->rx_stats.bytes += bytes;
2570 	u64_stats_update_end(&ring->rx_syncp);
2571 
2572 	q_vector->rx.total_packets += packets;
2573 	q_vector->rx.total_bytes += bytes;
2574 }
2575 
2576 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
2577 {
2578 	unsigned int total_bytes = 0, total_packets = 0;
2579 	struct igc_adapter *adapter = q_vector->adapter;
2580 	struct igc_ring *rx_ring = q_vector->rx.ring;
2581 	struct sk_buff *skb = rx_ring->skb;
2582 	u16 cleaned_count = igc_desc_unused(rx_ring);
2583 	int xdp_status = 0, rx_buffer_pgcnt;
2584 
2585 	while (likely(total_packets < budget)) {
2586 		union igc_adv_rx_desc *rx_desc;
2587 		struct igc_rx_buffer *rx_buffer;
2588 		unsigned int size, truesize;
2589 		struct igc_xdp_buff ctx;
2590 		ktime_t timestamp = 0;
2591 		int pkt_offset = 0;
2592 		void *pktbuf;
2593 
2594 		/* return some buffers to hardware, one at a time is too slow */
2595 		if (cleaned_count >= IGC_RX_BUFFER_WRITE) {
2596 			igc_alloc_rx_buffers(rx_ring, cleaned_count);
2597 			cleaned_count = 0;
2598 		}
2599 
2600 		rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean);
2601 		size = le16_to_cpu(rx_desc->wb.upper.length);
2602 		if (!size)
2603 			break;
2604 
2605 		/* This memory barrier is needed to keep us from reading
2606 		 * any other fields out of the rx_desc until we know the
2607 		 * descriptor has been written back
2608 		 */
2609 		dma_rmb();
2610 
2611 		rx_buffer = igc_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
2612 		truesize = igc_get_rx_frame_truesize(rx_ring, size);
2613 
2614 		pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
2615 
2616 		if (igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP)) {
2617 			timestamp = igc_ptp_rx_pktstamp(q_vector->adapter,
2618 							pktbuf);
2619 			ctx.rx_ts = timestamp;
2620 			pkt_offset = IGC_TS_HDR_LEN;
2621 			size -= IGC_TS_HDR_LEN;
2622 		}
2623 
2624 		if (!skb) {
2625 			xdp_init_buff(&ctx.xdp, truesize, &rx_ring->xdp_rxq);
2626 			xdp_prepare_buff(&ctx.xdp, pktbuf - igc_rx_offset(rx_ring),
2627 					 igc_rx_offset(rx_ring) + pkt_offset,
2628 					 size, true);
2629 			xdp_buff_clear_frags_flag(&ctx.xdp);
2630 			ctx.rx_desc = rx_desc;
2631 
2632 			skb = igc_xdp_run_prog(adapter, &ctx.xdp);
2633 		}
2634 
2635 		if (IS_ERR(skb)) {
2636 			unsigned int xdp_res = -PTR_ERR(skb);
2637 
2638 			switch (xdp_res) {
2639 			case IGC_XDP_CONSUMED:
2640 				rx_buffer->pagecnt_bias++;
2641 				break;
2642 			case IGC_XDP_TX:
2643 			case IGC_XDP_REDIRECT:
2644 				igc_rx_buffer_flip(rx_buffer, truesize);
2645 				xdp_status |= xdp_res;
2646 				break;
2647 			}
2648 
2649 			total_packets++;
2650 			total_bytes += size;
2651 		} else if (skb)
2652 			igc_add_rx_frag(rx_ring, rx_buffer, skb, size);
2653 		else if (ring_uses_build_skb(rx_ring))
2654 			skb = igc_build_skb(rx_ring, rx_buffer, &ctx.xdp);
2655 		else
2656 			skb = igc_construct_skb(rx_ring, rx_buffer, &ctx.xdp,
2657 						timestamp);
2658 
2659 		/* exit if we failed to retrieve a buffer */
2660 		if (!skb) {
2661 			rx_ring->rx_stats.alloc_failed++;
2662 			rx_buffer->pagecnt_bias++;
2663 			break;
2664 		}
2665 
2666 		igc_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
2667 		cleaned_count++;
2668 
2669 		/* fetch next buffer in frame if non-eop */
2670 		if (igc_is_non_eop(rx_ring, rx_desc))
2671 			continue;
2672 
2673 		/* verify the packet layout is correct */
2674 		if (igc_cleanup_headers(rx_ring, rx_desc, skb)) {
2675 			skb = NULL;
2676 			continue;
2677 		}
2678 
2679 		/* probably a little skewed due to removing CRC */
2680 		total_bytes += skb->len;
2681 
2682 		/* populate checksum, VLAN, and protocol */
2683 		igc_process_skb_fields(rx_ring, rx_desc, skb);
2684 
2685 		napi_gro_receive(&q_vector->napi, skb);
2686 
2687 		/* reset skb pointer */
2688 		skb = NULL;
2689 
2690 		/* update budget accounting */
2691 		total_packets++;
2692 	}
2693 
2694 	if (xdp_status)
2695 		igc_finalize_xdp(adapter, xdp_status);
2696 
2697 	/* place incomplete frames back on ring for completion */
2698 	rx_ring->skb = skb;
2699 
2700 	igc_update_rx_stats(q_vector, total_packets, total_bytes);
2701 
2702 	if (cleaned_count)
2703 		igc_alloc_rx_buffers(rx_ring, cleaned_count);
2704 
2705 	return total_packets;
2706 }
2707 
2708 static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring,
2709 					    struct xdp_buff *xdp)
2710 {
2711 	unsigned int totalsize = xdp->data_end - xdp->data_meta;
2712 	unsigned int metasize = xdp->data - xdp->data_meta;
2713 	struct sk_buff *skb;
2714 
2715 	net_prefetch(xdp->data_meta);
2716 
2717 	skb = __napi_alloc_skb(&ring->q_vector->napi, totalsize,
2718 			       GFP_ATOMIC | __GFP_NOWARN);
2719 	if (unlikely(!skb))
2720 		return NULL;
2721 
2722 	memcpy(__skb_put(skb, totalsize), xdp->data_meta,
2723 	       ALIGN(totalsize, sizeof(long)));
2724 
2725 	if (metasize) {
2726 		skb_metadata_set(skb, metasize);
2727 		__skb_pull(skb, metasize);
2728 	}
2729 
2730 	return skb;
2731 }
2732 
2733 static void igc_dispatch_skb_zc(struct igc_q_vector *q_vector,
2734 				union igc_adv_rx_desc *desc,
2735 				struct xdp_buff *xdp,
2736 				ktime_t timestamp)
2737 {
2738 	struct igc_ring *ring = q_vector->rx.ring;
2739 	struct sk_buff *skb;
2740 
2741 	skb = igc_construct_skb_zc(ring, xdp);
2742 	if (!skb) {
2743 		ring->rx_stats.alloc_failed++;
2744 		return;
2745 	}
2746 
2747 	if (timestamp)
2748 		skb_hwtstamps(skb)->hwtstamp = timestamp;
2749 
2750 	if (igc_cleanup_headers(ring, desc, skb))
2751 		return;
2752 
2753 	igc_process_skb_fields(ring, desc, skb);
2754 	napi_gro_receive(&q_vector->napi, skb);
2755 }
2756 
2757 static struct igc_xdp_buff *xsk_buff_to_igc_ctx(struct xdp_buff *xdp)
2758 {
2759 	/* xdp_buff pointer used by ZC code path is alloc as xdp_buff_xsk. The
2760 	 * igc_xdp_buff shares its layout with xdp_buff_xsk and private
2761 	 * igc_xdp_buff fields fall into xdp_buff_xsk->cb
2762 	 */
2763        return (struct igc_xdp_buff *)xdp;
2764 }
2765 
2766 static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget)
2767 {
2768 	struct igc_adapter *adapter = q_vector->adapter;
2769 	struct igc_ring *ring = q_vector->rx.ring;
2770 	u16 cleaned_count = igc_desc_unused(ring);
2771 	int total_bytes = 0, total_packets = 0;
2772 	u16 ntc = ring->next_to_clean;
2773 	struct bpf_prog *prog;
2774 	bool failure = false;
2775 	int xdp_status = 0;
2776 
2777 	rcu_read_lock();
2778 
2779 	prog = READ_ONCE(adapter->xdp_prog);
2780 
2781 	while (likely(total_packets < budget)) {
2782 		union igc_adv_rx_desc *desc;
2783 		struct igc_rx_buffer *bi;
2784 		struct igc_xdp_buff *ctx;
2785 		ktime_t timestamp = 0;
2786 		unsigned int size;
2787 		int res;
2788 
2789 		desc = IGC_RX_DESC(ring, ntc);
2790 		size = le16_to_cpu(desc->wb.upper.length);
2791 		if (!size)
2792 			break;
2793 
2794 		/* This memory barrier is needed to keep us from reading
2795 		 * any other fields out of the rx_desc until we know the
2796 		 * descriptor has been written back
2797 		 */
2798 		dma_rmb();
2799 
2800 		bi = &ring->rx_buffer_info[ntc];
2801 
2802 		ctx = xsk_buff_to_igc_ctx(bi->xdp);
2803 		ctx->rx_desc = desc;
2804 
2805 		if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) {
2806 			timestamp = igc_ptp_rx_pktstamp(q_vector->adapter,
2807 							bi->xdp->data);
2808 			ctx->rx_ts = timestamp;
2809 
2810 			bi->xdp->data += IGC_TS_HDR_LEN;
2811 
2812 			/* HW timestamp has been copied into local variable. Metadata
2813 			 * length when XDP program is called should be 0.
2814 			 */
2815 			bi->xdp->data_meta += IGC_TS_HDR_LEN;
2816 			size -= IGC_TS_HDR_LEN;
2817 		}
2818 
2819 		bi->xdp->data_end = bi->xdp->data + size;
2820 		xsk_buff_dma_sync_for_cpu(bi->xdp, ring->xsk_pool);
2821 
2822 		res = __igc_xdp_run_prog(adapter, prog, bi->xdp);
2823 		switch (res) {
2824 		case IGC_XDP_PASS:
2825 			igc_dispatch_skb_zc(q_vector, desc, bi->xdp, timestamp);
2826 			fallthrough;
2827 		case IGC_XDP_CONSUMED:
2828 			xsk_buff_free(bi->xdp);
2829 			break;
2830 		case IGC_XDP_TX:
2831 		case IGC_XDP_REDIRECT:
2832 			xdp_status |= res;
2833 			break;
2834 		}
2835 
2836 		bi->xdp = NULL;
2837 		total_bytes += size;
2838 		total_packets++;
2839 		cleaned_count++;
2840 		ntc++;
2841 		if (ntc == ring->count)
2842 			ntc = 0;
2843 	}
2844 
2845 	ring->next_to_clean = ntc;
2846 	rcu_read_unlock();
2847 
2848 	if (cleaned_count >= IGC_RX_BUFFER_WRITE)
2849 		failure = !igc_alloc_rx_buffers_zc(ring, cleaned_count);
2850 
2851 	if (xdp_status)
2852 		igc_finalize_xdp(adapter, xdp_status);
2853 
2854 	igc_update_rx_stats(q_vector, total_packets, total_bytes);
2855 
2856 	if (xsk_uses_need_wakeup(ring->xsk_pool)) {
2857 		if (failure || ring->next_to_clean == ring->next_to_use)
2858 			xsk_set_rx_need_wakeup(ring->xsk_pool);
2859 		else
2860 			xsk_clear_rx_need_wakeup(ring->xsk_pool);
2861 		return total_packets;
2862 	}
2863 
2864 	return failure ? budget : total_packets;
2865 }
2866 
2867 static void igc_update_tx_stats(struct igc_q_vector *q_vector,
2868 				unsigned int packets, unsigned int bytes)
2869 {
2870 	struct igc_ring *ring = q_vector->tx.ring;
2871 
2872 	u64_stats_update_begin(&ring->tx_syncp);
2873 	ring->tx_stats.bytes += bytes;
2874 	ring->tx_stats.packets += packets;
2875 	u64_stats_update_end(&ring->tx_syncp);
2876 
2877 	q_vector->tx.total_bytes += bytes;
2878 	q_vector->tx.total_packets += packets;
2879 }
2880 
2881 static void igc_xdp_xmit_zc(struct igc_ring *ring)
2882 {
2883 	struct xsk_buff_pool *pool = ring->xsk_pool;
2884 	struct netdev_queue *nq = txring_txq(ring);
2885 	union igc_adv_tx_desc *tx_desc = NULL;
2886 	int cpu = smp_processor_id();
2887 	struct xdp_desc xdp_desc;
2888 	u16 budget, ntu;
2889 
2890 	if (!netif_carrier_ok(ring->netdev))
2891 		return;
2892 
2893 	__netif_tx_lock(nq, cpu);
2894 
2895 	/* Avoid transmit queue timeout since we share it with the slow path */
2896 	txq_trans_cond_update(nq);
2897 
2898 	ntu = ring->next_to_use;
2899 	budget = igc_desc_unused(ring);
2900 
2901 	while (xsk_tx_peek_desc(pool, &xdp_desc) && budget--) {
2902 		u32 cmd_type, olinfo_status;
2903 		struct igc_tx_buffer *bi;
2904 		dma_addr_t dma;
2905 
2906 		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2907 			   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
2908 			   xdp_desc.len;
2909 		olinfo_status = xdp_desc.len << IGC_ADVTXD_PAYLEN_SHIFT;
2910 
2911 		dma = xsk_buff_raw_get_dma(pool, xdp_desc.addr);
2912 		xsk_buff_raw_dma_sync_for_device(pool, dma, xdp_desc.len);
2913 
2914 		tx_desc = IGC_TX_DESC(ring, ntu);
2915 		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2916 		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2917 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
2918 
2919 		bi = &ring->tx_buffer_info[ntu];
2920 		bi->type = IGC_TX_BUFFER_TYPE_XSK;
2921 		bi->protocol = 0;
2922 		bi->bytecount = xdp_desc.len;
2923 		bi->gso_segs = 1;
2924 		bi->time_stamp = jiffies;
2925 		bi->next_to_watch = tx_desc;
2926 
2927 		netdev_tx_sent_queue(txring_txq(ring), xdp_desc.len);
2928 
2929 		ntu++;
2930 		if (ntu == ring->count)
2931 			ntu = 0;
2932 	}
2933 
2934 	ring->next_to_use = ntu;
2935 	if (tx_desc) {
2936 		igc_flush_tx_descriptors(ring);
2937 		xsk_tx_release(pool);
2938 	}
2939 
2940 	__netif_tx_unlock(nq);
2941 }
2942 
2943 /**
2944  * igc_clean_tx_irq - Reclaim resources after transmit completes
2945  * @q_vector: pointer to q_vector containing needed info
2946  * @napi_budget: Used to determine if we are in netpoll
2947  *
2948  * returns true if ring is completely cleaned
2949  */
2950 static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
2951 {
2952 	struct igc_adapter *adapter = q_vector->adapter;
2953 	unsigned int total_bytes = 0, total_packets = 0;
2954 	unsigned int budget = q_vector->tx.work_limit;
2955 	struct igc_ring *tx_ring = q_vector->tx.ring;
2956 	unsigned int i = tx_ring->next_to_clean;
2957 	struct igc_tx_buffer *tx_buffer;
2958 	union igc_adv_tx_desc *tx_desc;
2959 	u32 xsk_frames = 0;
2960 
2961 	if (test_bit(__IGC_DOWN, &adapter->state))
2962 		return true;
2963 
2964 	tx_buffer = &tx_ring->tx_buffer_info[i];
2965 	tx_desc = IGC_TX_DESC(tx_ring, i);
2966 	i -= tx_ring->count;
2967 
2968 	do {
2969 		union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
2970 
2971 		/* if next_to_watch is not set then there is no work pending */
2972 		if (!eop_desc)
2973 			break;
2974 
2975 		/* prevent any other reads prior to eop_desc */
2976 		smp_rmb();
2977 
2978 		/* if DD is not set pending work has not been completed */
2979 		if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD)))
2980 			break;
2981 
2982 		/* clear next_to_watch to prevent false hangs */
2983 		tx_buffer->next_to_watch = NULL;
2984 
2985 		/* update the statistics for this packet */
2986 		total_bytes += tx_buffer->bytecount;
2987 		total_packets += tx_buffer->gso_segs;
2988 
2989 		switch (tx_buffer->type) {
2990 		case IGC_TX_BUFFER_TYPE_XSK:
2991 			xsk_frames++;
2992 			break;
2993 		case IGC_TX_BUFFER_TYPE_XDP:
2994 			xdp_return_frame(tx_buffer->xdpf);
2995 			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
2996 			break;
2997 		case IGC_TX_BUFFER_TYPE_SKB:
2998 			napi_consume_skb(tx_buffer->skb, napi_budget);
2999 			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
3000 			break;
3001 		default:
3002 			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
3003 			break;
3004 		}
3005 
3006 		/* clear last DMA location and unmap remaining buffers */
3007 		while (tx_desc != eop_desc) {
3008 			tx_buffer++;
3009 			tx_desc++;
3010 			i++;
3011 			if (unlikely(!i)) {
3012 				i -= tx_ring->count;
3013 				tx_buffer = tx_ring->tx_buffer_info;
3014 				tx_desc = IGC_TX_DESC(tx_ring, 0);
3015 			}
3016 
3017 			/* unmap any remaining paged data */
3018 			if (dma_unmap_len(tx_buffer, len))
3019 				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
3020 		}
3021 
3022 		/* move us one more past the eop_desc for start of next pkt */
3023 		tx_buffer++;
3024 		tx_desc++;
3025 		i++;
3026 		if (unlikely(!i)) {
3027 			i -= tx_ring->count;
3028 			tx_buffer = tx_ring->tx_buffer_info;
3029 			tx_desc = IGC_TX_DESC(tx_ring, 0);
3030 		}
3031 
3032 		/* issue prefetch for next Tx descriptor */
3033 		prefetch(tx_desc);
3034 
3035 		/* update budget accounting */
3036 		budget--;
3037 	} while (likely(budget));
3038 
3039 	netdev_tx_completed_queue(txring_txq(tx_ring),
3040 				  total_packets, total_bytes);
3041 
3042 	i += tx_ring->count;
3043 	tx_ring->next_to_clean = i;
3044 
3045 	igc_update_tx_stats(q_vector, total_packets, total_bytes);
3046 
3047 	if (tx_ring->xsk_pool) {
3048 		if (xsk_frames)
3049 			xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
3050 		if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
3051 			xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
3052 		igc_xdp_xmit_zc(tx_ring);
3053 	}
3054 
3055 	if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
3056 		struct igc_hw *hw = &adapter->hw;
3057 
3058 		/* Detect a transmit hang in hardware, this serializes the
3059 		 * check with the clearing of time_stamp and movement of i
3060 		 */
3061 		clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3062 		if (tx_buffer->next_to_watch &&
3063 		    time_after(jiffies, tx_buffer->time_stamp +
3064 		    (adapter->tx_timeout_factor * HZ)) &&
3065 		    !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF) &&
3066 		    (rd32(IGC_TDH(tx_ring->reg_idx)) != readl(tx_ring->tail)) &&
3067 		    !tx_ring->oper_gate_closed) {
3068 			/* detected Tx unit hang */
3069 			netdev_err(tx_ring->netdev,
3070 				   "Detected Tx Unit Hang\n"
3071 				   "  Tx Queue             <%d>\n"
3072 				   "  TDH                  <%x>\n"
3073 				   "  TDT                  <%x>\n"
3074 				   "  next_to_use          <%x>\n"
3075 				   "  next_to_clean        <%x>\n"
3076 				   "buffer_info[next_to_clean]\n"
3077 				   "  time_stamp           <%lx>\n"
3078 				   "  next_to_watch        <%p>\n"
3079 				   "  jiffies              <%lx>\n"
3080 				   "  desc.status          <%x>\n",
3081 				   tx_ring->queue_index,
3082 				   rd32(IGC_TDH(tx_ring->reg_idx)),
3083 				   readl(tx_ring->tail),
3084 				   tx_ring->next_to_use,
3085 				   tx_ring->next_to_clean,
3086 				   tx_buffer->time_stamp,
3087 				   tx_buffer->next_to_watch,
3088 				   jiffies,
3089 				   tx_buffer->next_to_watch->wb.status);
3090 			netif_stop_subqueue(tx_ring->netdev,
3091 					    tx_ring->queue_index);
3092 
3093 			/* we are about to reset, no point in enabling stuff */
3094 			return true;
3095 		}
3096 	}
3097 
3098 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
3099 	if (unlikely(total_packets &&
3100 		     netif_carrier_ok(tx_ring->netdev) &&
3101 		     igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
3102 		/* Make sure that anybody stopping the queue after this
3103 		 * sees the new next_to_clean.
3104 		 */
3105 		smp_mb();
3106 		if (__netif_subqueue_stopped(tx_ring->netdev,
3107 					     tx_ring->queue_index) &&
3108 		    !(test_bit(__IGC_DOWN, &adapter->state))) {
3109 			netif_wake_subqueue(tx_ring->netdev,
3110 					    tx_ring->queue_index);
3111 
3112 			u64_stats_update_begin(&tx_ring->tx_syncp);
3113 			tx_ring->tx_stats.restart_queue++;
3114 			u64_stats_update_end(&tx_ring->tx_syncp);
3115 		}
3116 	}
3117 
3118 	return !!budget;
3119 }
3120 
3121 static int igc_find_mac_filter(struct igc_adapter *adapter,
3122 			       enum igc_mac_filter_type type, const u8 *addr)
3123 {
3124 	struct igc_hw *hw = &adapter->hw;
3125 	int max_entries = hw->mac.rar_entry_count;
3126 	u32 ral, rah;
3127 	int i;
3128 
3129 	for (i = 0; i < max_entries; i++) {
3130 		ral = rd32(IGC_RAL(i));
3131 		rah = rd32(IGC_RAH(i));
3132 
3133 		if (!(rah & IGC_RAH_AV))
3134 			continue;
3135 		if (!!(rah & IGC_RAH_ASEL_SRC_ADDR) != type)
3136 			continue;
3137 		if ((rah & IGC_RAH_RAH_MASK) !=
3138 		    le16_to_cpup((__le16 *)(addr + 4)))
3139 			continue;
3140 		if (ral != le32_to_cpup((__le32 *)(addr)))
3141 			continue;
3142 
3143 		return i;
3144 	}
3145 
3146 	return -1;
3147 }
3148 
3149 static int igc_get_avail_mac_filter_slot(struct igc_adapter *adapter)
3150 {
3151 	struct igc_hw *hw = &adapter->hw;
3152 	int max_entries = hw->mac.rar_entry_count;
3153 	u32 rah;
3154 	int i;
3155 
3156 	for (i = 0; i < max_entries; i++) {
3157 		rah = rd32(IGC_RAH(i));
3158 
3159 		if (!(rah & IGC_RAH_AV))
3160 			return i;
3161 	}
3162 
3163 	return -1;
3164 }
3165 
3166 /**
3167  * igc_add_mac_filter() - Add MAC address filter
3168  * @adapter: Pointer to adapter where the filter should be added
3169  * @type: MAC address filter type (source or destination)
3170  * @addr: MAC address
3171  * @queue: If non-negative, queue assignment feature is enabled and frames
3172  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
3173  *         assignment is disabled.
3174  *
3175  * Return: 0 in case of success, negative errno code otherwise.
3176  */
3177 static int igc_add_mac_filter(struct igc_adapter *adapter,
3178 			      enum igc_mac_filter_type type, const u8 *addr,
3179 			      int queue)
3180 {
3181 	struct net_device *dev = adapter->netdev;
3182 	int index;
3183 
3184 	index = igc_find_mac_filter(adapter, type, addr);
3185 	if (index >= 0)
3186 		goto update_filter;
3187 
3188 	index = igc_get_avail_mac_filter_slot(adapter);
3189 	if (index < 0)
3190 		return -ENOSPC;
3191 
3192 	netdev_dbg(dev, "Add MAC address filter: index %d type %s address %pM queue %d\n",
3193 		   index, type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
3194 		   addr, queue);
3195 
3196 update_filter:
3197 	igc_set_mac_filter_hw(adapter, index, type, addr, queue);
3198 	return 0;
3199 }
3200 
3201 /**
3202  * igc_del_mac_filter() - Delete MAC address filter
3203  * @adapter: Pointer to adapter where the filter should be deleted from
3204  * @type: MAC address filter type (source or destination)
3205  * @addr: MAC address
3206  */
3207 static void igc_del_mac_filter(struct igc_adapter *adapter,
3208 			       enum igc_mac_filter_type type, const u8 *addr)
3209 {
3210 	struct net_device *dev = adapter->netdev;
3211 	int index;
3212 
3213 	index = igc_find_mac_filter(adapter, type, addr);
3214 	if (index < 0)
3215 		return;
3216 
3217 	if (index == 0) {
3218 		/* If this is the default filter, we don't actually delete it.
3219 		 * We just reset to its default value i.e. disable queue
3220 		 * assignment.
3221 		 */
3222 		netdev_dbg(dev, "Disable default MAC filter queue assignment");
3223 
3224 		igc_set_mac_filter_hw(adapter, 0, type, addr, -1);
3225 	} else {
3226 		netdev_dbg(dev, "Delete MAC address filter: index %d type %s address %pM\n",
3227 			   index,
3228 			   type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
3229 			   addr);
3230 
3231 		igc_clear_mac_filter_hw(adapter, index);
3232 	}
3233 }
3234 
3235 /**
3236  * igc_add_vlan_prio_filter() - Add VLAN priority filter
3237  * @adapter: Pointer to adapter where the filter should be added
3238  * @prio: VLAN priority value
3239  * @queue: Queue number which matching frames are assigned to
3240  *
3241  * Return: 0 in case of success, negative errno code otherwise.
3242  */
3243 static int igc_add_vlan_prio_filter(struct igc_adapter *adapter, int prio,
3244 				    int queue)
3245 {
3246 	struct net_device *dev = adapter->netdev;
3247 	struct igc_hw *hw = &adapter->hw;
3248 	u32 vlanpqf;
3249 
3250 	vlanpqf = rd32(IGC_VLANPQF);
3251 
3252 	if (vlanpqf & IGC_VLANPQF_VALID(prio)) {
3253 		netdev_dbg(dev, "VLAN priority filter already in use\n");
3254 		return -EEXIST;
3255 	}
3256 
3257 	vlanpqf |= IGC_VLANPQF_QSEL(prio, queue);
3258 	vlanpqf |= IGC_VLANPQF_VALID(prio);
3259 
3260 	wr32(IGC_VLANPQF, vlanpqf);
3261 
3262 	netdev_dbg(dev, "Add VLAN priority filter: prio %d queue %d\n",
3263 		   prio, queue);
3264 	return 0;
3265 }
3266 
3267 /**
3268  * igc_del_vlan_prio_filter() - Delete VLAN priority filter
3269  * @adapter: Pointer to adapter where the filter should be deleted from
3270  * @prio: VLAN priority value
3271  */
3272 static void igc_del_vlan_prio_filter(struct igc_adapter *adapter, int prio)
3273 {
3274 	struct igc_hw *hw = &adapter->hw;
3275 	u32 vlanpqf;
3276 
3277 	vlanpqf = rd32(IGC_VLANPQF);
3278 
3279 	vlanpqf &= ~IGC_VLANPQF_VALID(prio);
3280 	vlanpqf &= ~IGC_VLANPQF_QSEL(prio, IGC_VLANPQF_QUEUE_MASK);
3281 
3282 	wr32(IGC_VLANPQF, vlanpqf);
3283 
3284 	netdev_dbg(adapter->netdev, "Delete VLAN priority filter: prio %d\n",
3285 		   prio);
3286 }
3287 
3288 static int igc_get_avail_etype_filter_slot(struct igc_adapter *adapter)
3289 {
3290 	struct igc_hw *hw = &adapter->hw;
3291 	int i;
3292 
3293 	for (i = 0; i < MAX_ETYPE_FILTER; i++) {
3294 		u32 etqf = rd32(IGC_ETQF(i));
3295 
3296 		if (!(etqf & IGC_ETQF_FILTER_ENABLE))
3297 			return i;
3298 	}
3299 
3300 	return -1;
3301 }
3302 
3303 /**
3304  * igc_add_etype_filter() - Add ethertype filter
3305  * @adapter: Pointer to adapter where the filter should be added
3306  * @etype: Ethertype value
3307  * @queue: If non-negative, queue assignment feature is enabled and frames
3308  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
3309  *         assignment is disabled.
3310  *
3311  * Return: 0 in case of success, negative errno code otherwise.
3312  */
3313 static int igc_add_etype_filter(struct igc_adapter *adapter, u16 etype,
3314 				int queue)
3315 {
3316 	struct igc_hw *hw = &adapter->hw;
3317 	int index;
3318 	u32 etqf;
3319 
3320 	index = igc_get_avail_etype_filter_slot(adapter);
3321 	if (index < 0)
3322 		return -ENOSPC;
3323 
3324 	etqf = rd32(IGC_ETQF(index));
3325 
3326 	etqf &= ~IGC_ETQF_ETYPE_MASK;
3327 	etqf |= etype;
3328 
3329 	if (queue >= 0) {
3330 		etqf &= ~IGC_ETQF_QUEUE_MASK;
3331 		etqf |= (queue << IGC_ETQF_QUEUE_SHIFT);
3332 		etqf |= IGC_ETQF_QUEUE_ENABLE;
3333 	}
3334 
3335 	etqf |= IGC_ETQF_FILTER_ENABLE;
3336 
3337 	wr32(IGC_ETQF(index), etqf);
3338 
3339 	netdev_dbg(adapter->netdev, "Add ethertype filter: etype %04x queue %d\n",
3340 		   etype, queue);
3341 	return 0;
3342 }
3343 
3344 static int igc_find_etype_filter(struct igc_adapter *adapter, u16 etype)
3345 {
3346 	struct igc_hw *hw = &adapter->hw;
3347 	int i;
3348 
3349 	for (i = 0; i < MAX_ETYPE_FILTER; i++) {
3350 		u32 etqf = rd32(IGC_ETQF(i));
3351 
3352 		if ((etqf & IGC_ETQF_ETYPE_MASK) == etype)
3353 			return i;
3354 	}
3355 
3356 	return -1;
3357 }
3358 
3359 /**
3360  * igc_del_etype_filter() - Delete ethertype filter
3361  * @adapter: Pointer to adapter where the filter should be deleted from
3362  * @etype: Ethertype value
3363  */
3364 static void igc_del_etype_filter(struct igc_adapter *adapter, u16 etype)
3365 {
3366 	struct igc_hw *hw = &adapter->hw;
3367 	int index;
3368 
3369 	index = igc_find_etype_filter(adapter, etype);
3370 	if (index < 0)
3371 		return;
3372 
3373 	wr32(IGC_ETQF(index), 0);
3374 
3375 	netdev_dbg(adapter->netdev, "Delete ethertype filter: etype %04x\n",
3376 		   etype);
3377 }
3378 
3379 static int igc_flex_filter_select(struct igc_adapter *adapter,
3380 				  struct igc_flex_filter *input,
3381 				  u32 *fhft)
3382 {
3383 	struct igc_hw *hw = &adapter->hw;
3384 	u8 fhft_index;
3385 	u32 fhftsl;
3386 
3387 	if (input->index >= MAX_FLEX_FILTER) {
3388 		dev_err(&adapter->pdev->dev, "Wrong Flex Filter index selected!\n");
3389 		return -EINVAL;
3390 	}
3391 
3392 	/* Indirect table select register */
3393 	fhftsl = rd32(IGC_FHFTSL);
3394 	fhftsl &= ~IGC_FHFTSL_FTSL_MASK;
3395 	switch (input->index) {
3396 	case 0 ... 7:
3397 		fhftsl |= 0x00;
3398 		break;
3399 	case 8 ... 15:
3400 		fhftsl |= 0x01;
3401 		break;
3402 	case 16 ... 23:
3403 		fhftsl |= 0x02;
3404 		break;
3405 	case 24 ... 31:
3406 		fhftsl |= 0x03;
3407 		break;
3408 	}
3409 	wr32(IGC_FHFTSL, fhftsl);
3410 
3411 	/* Normalize index down to host table register */
3412 	fhft_index = input->index % 8;
3413 
3414 	*fhft = (fhft_index < 4) ? IGC_FHFT(fhft_index) :
3415 		IGC_FHFT_EXT(fhft_index - 4);
3416 
3417 	return 0;
3418 }
3419 
3420 static int igc_write_flex_filter_ll(struct igc_adapter *adapter,
3421 				    struct igc_flex_filter *input)
3422 {
3423 	struct device *dev = &adapter->pdev->dev;
3424 	struct igc_hw *hw = &adapter->hw;
3425 	u8 *data = input->data;
3426 	u8 *mask = input->mask;
3427 	u32 queuing;
3428 	u32 fhft;
3429 	u32 wufc;
3430 	int ret;
3431 	int i;
3432 
3433 	/* Length has to be aligned to 8. Otherwise the filter will fail. Bail
3434 	 * out early to avoid surprises later.
3435 	 */
3436 	if (input->length % 8 != 0) {
3437 		dev_err(dev, "The length of a flex filter has to be 8 byte aligned!\n");
3438 		return -EINVAL;
3439 	}
3440 
3441 	/* Select corresponding flex filter register and get base for host table. */
3442 	ret = igc_flex_filter_select(adapter, input, &fhft);
3443 	if (ret)
3444 		return ret;
3445 
3446 	/* When adding a filter globally disable flex filter feature. That is
3447 	 * recommended within the datasheet.
3448 	 */
3449 	wufc = rd32(IGC_WUFC);
3450 	wufc &= ~IGC_WUFC_FLEX_HQ;
3451 	wr32(IGC_WUFC, wufc);
3452 
3453 	/* Configure filter */
3454 	queuing = input->length & IGC_FHFT_LENGTH_MASK;
3455 	queuing |= (input->rx_queue << IGC_FHFT_QUEUE_SHIFT) & IGC_FHFT_QUEUE_MASK;
3456 	queuing |= (input->prio << IGC_FHFT_PRIO_SHIFT) & IGC_FHFT_PRIO_MASK;
3457 
3458 	if (input->immediate_irq)
3459 		queuing |= IGC_FHFT_IMM_INT;
3460 
3461 	if (input->drop)
3462 		queuing |= IGC_FHFT_DROP;
3463 
3464 	wr32(fhft + 0xFC, queuing);
3465 
3466 	/* Write data (128 byte) and mask (128 bit) */
3467 	for (i = 0; i < 16; ++i) {
3468 		const size_t data_idx = i * 8;
3469 		const size_t row_idx = i * 16;
3470 		u32 dw0 =
3471 			(data[data_idx + 0] << 0) |
3472 			(data[data_idx + 1] << 8) |
3473 			(data[data_idx + 2] << 16) |
3474 			(data[data_idx + 3] << 24);
3475 		u32 dw1 =
3476 			(data[data_idx + 4] << 0) |
3477 			(data[data_idx + 5] << 8) |
3478 			(data[data_idx + 6] << 16) |
3479 			(data[data_idx + 7] << 24);
3480 		u32 tmp;
3481 
3482 		/* Write row: dw0, dw1 and mask */
3483 		wr32(fhft + row_idx, dw0);
3484 		wr32(fhft + row_idx + 4, dw1);
3485 
3486 		/* mask is only valid for MASK(7, 0) */
3487 		tmp = rd32(fhft + row_idx + 8);
3488 		tmp &= ~GENMASK(7, 0);
3489 		tmp |= mask[i];
3490 		wr32(fhft + row_idx + 8, tmp);
3491 	}
3492 
3493 	/* Enable filter. */
3494 	wufc |= IGC_WUFC_FLEX_HQ;
3495 	if (input->index > 8) {
3496 		/* Filter 0-7 are enabled via WUFC. The other 24 filters are not. */
3497 		u32 wufc_ext = rd32(IGC_WUFC_EXT);
3498 
3499 		wufc_ext |= (IGC_WUFC_EXT_FLX8 << (input->index - 8));
3500 
3501 		wr32(IGC_WUFC_EXT, wufc_ext);
3502 	} else {
3503 		wufc |= (IGC_WUFC_FLX0 << input->index);
3504 	}
3505 	wr32(IGC_WUFC, wufc);
3506 
3507 	dev_dbg(&adapter->pdev->dev, "Added flex filter %u to HW.\n",
3508 		input->index);
3509 
3510 	return 0;
3511 }
3512 
3513 static void igc_flex_filter_add_field(struct igc_flex_filter *flex,
3514 				      const void *src, unsigned int offset,
3515 				      size_t len, const void *mask)
3516 {
3517 	int i;
3518 
3519 	/* data */
3520 	memcpy(&flex->data[offset], src, len);
3521 
3522 	/* mask */
3523 	for (i = 0; i < len; ++i) {
3524 		const unsigned int idx = i + offset;
3525 		const u8 *ptr = mask;
3526 
3527 		if (mask) {
3528 			if (ptr[i] & 0xff)
3529 				flex->mask[idx / 8] |= BIT(idx % 8);
3530 
3531 			continue;
3532 		}
3533 
3534 		flex->mask[idx / 8] |= BIT(idx % 8);
3535 	}
3536 }
3537 
3538 static int igc_find_avail_flex_filter_slot(struct igc_adapter *adapter)
3539 {
3540 	struct igc_hw *hw = &adapter->hw;
3541 	u32 wufc, wufc_ext;
3542 	int i;
3543 
3544 	wufc = rd32(IGC_WUFC);
3545 	wufc_ext = rd32(IGC_WUFC_EXT);
3546 
3547 	for (i = 0; i < MAX_FLEX_FILTER; i++) {
3548 		if (i < 8) {
3549 			if (!(wufc & (IGC_WUFC_FLX0 << i)))
3550 				return i;
3551 		} else {
3552 			if (!(wufc_ext & (IGC_WUFC_EXT_FLX8 << (i - 8))))
3553 				return i;
3554 		}
3555 	}
3556 
3557 	return -ENOSPC;
3558 }
3559 
3560 static bool igc_flex_filter_in_use(struct igc_adapter *adapter)
3561 {
3562 	struct igc_hw *hw = &adapter->hw;
3563 	u32 wufc, wufc_ext;
3564 
3565 	wufc = rd32(IGC_WUFC);
3566 	wufc_ext = rd32(IGC_WUFC_EXT);
3567 
3568 	if (wufc & IGC_WUFC_FILTER_MASK)
3569 		return true;
3570 
3571 	if (wufc_ext & IGC_WUFC_EXT_FILTER_MASK)
3572 		return true;
3573 
3574 	return false;
3575 }
3576 
3577 static int igc_add_flex_filter(struct igc_adapter *adapter,
3578 			       struct igc_nfc_rule *rule)
3579 {
3580 	struct igc_flex_filter flex = { };
3581 	struct igc_nfc_filter *filter = &rule->filter;
3582 	unsigned int eth_offset, user_offset;
3583 	int ret, index;
3584 	bool vlan;
3585 
3586 	index = igc_find_avail_flex_filter_slot(adapter);
3587 	if (index < 0)
3588 		return -ENOSPC;
3589 
3590 	/* Construct the flex filter:
3591 	 *  -> dest_mac [6]
3592 	 *  -> src_mac [6]
3593 	 *  -> tpid [2]
3594 	 *  -> vlan tci [2]
3595 	 *  -> ether type [2]
3596 	 *  -> user data [8]
3597 	 *  -> = 26 bytes => 32 length
3598 	 */
3599 	flex.index    = index;
3600 	flex.length   = 32;
3601 	flex.rx_queue = rule->action;
3602 
3603 	vlan = rule->filter.vlan_tci || rule->filter.vlan_etype;
3604 	eth_offset = vlan ? 16 : 12;
3605 	user_offset = vlan ? 18 : 14;
3606 
3607 	/* Add destination MAC  */
3608 	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3609 		igc_flex_filter_add_field(&flex, &filter->dst_addr, 0,
3610 					  ETH_ALEN, NULL);
3611 
3612 	/* Add source MAC */
3613 	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3614 		igc_flex_filter_add_field(&flex, &filter->src_addr, 6,
3615 					  ETH_ALEN, NULL);
3616 
3617 	/* Add VLAN etype */
3618 	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_ETYPE)
3619 		igc_flex_filter_add_field(&flex, &filter->vlan_etype, 12,
3620 					  sizeof(filter->vlan_etype),
3621 					  NULL);
3622 
3623 	/* Add VLAN TCI */
3624 	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI)
3625 		igc_flex_filter_add_field(&flex, &filter->vlan_tci, 14,
3626 					  sizeof(filter->vlan_tci), NULL);
3627 
3628 	/* Add Ether type */
3629 	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3630 		__be16 etype = cpu_to_be16(filter->etype);
3631 
3632 		igc_flex_filter_add_field(&flex, &etype, eth_offset,
3633 					  sizeof(etype), NULL);
3634 	}
3635 
3636 	/* Add user data */
3637 	if (rule->filter.match_flags & IGC_FILTER_FLAG_USER_DATA)
3638 		igc_flex_filter_add_field(&flex, &filter->user_data,
3639 					  user_offset,
3640 					  sizeof(filter->user_data),
3641 					  filter->user_mask);
3642 
3643 	/* Add it down to the hardware and enable it. */
3644 	ret = igc_write_flex_filter_ll(adapter, &flex);
3645 	if (ret)
3646 		return ret;
3647 
3648 	filter->flex_index = index;
3649 
3650 	return 0;
3651 }
3652 
3653 static void igc_del_flex_filter(struct igc_adapter *adapter,
3654 				u16 reg_index)
3655 {
3656 	struct igc_hw *hw = &adapter->hw;
3657 	u32 wufc;
3658 
3659 	/* Just disable the filter. The filter table itself is kept
3660 	 * intact. Another flex_filter_add() should override the "old" data
3661 	 * then.
3662 	 */
3663 	if (reg_index > 8) {
3664 		u32 wufc_ext = rd32(IGC_WUFC_EXT);
3665 
3666 		wufc_ext &= ~(IGC_WUFC_EXT_FLX8 << (reg_index - 8));
3667 		wr32(IGC_WUFC_EXT, wufc_ext);
3668 	} else {
3669 		wufc = rd32(IGC_WUFC);
3670 
3671 		wufc &= ~(IGC_WUFC_FLX0 << reg_index);
3672 		wr32(IGC_WUFC, wufc);
3673 	}
3674 
3675 	if (igc_flex_filter_in_use(adapter))
3676 		return;
3677 
3678 	/* No filters are in use, we may disable flex filters */
3679 	wufc = rd32(IGC_WUFC);
3680 	wufc &= ~IGC_WUFC_FLEX_HQ;
3681 	wr32(IGC_WUFC, wufc);
3682 }
3683 
3684 static int igc_enable_nfc_rule(struct igc_adapter *adapter,
3685 			       struct igc_nfc_rule *rule)
3686 {
3687 	int err;
3688 
3689 	if (rule->flex) {
3690 		return igc_add_flex_filter(adapter, rule);
3691 	}
3692 
3693 	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3694 		err = igc_add_etype_filter(adapter, rule->filter.etype,
3695 					   rule->action);
3696 		if (err)
3697 			return err;
3698 	}
3699 
3700 	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) {
3701 		err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3702 					 rule->filter.src_addr, rule->action);
3703 		if (err)
3704 			return err;
3705 	}
3706 
3707 	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) {
3708 		err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3709 					 rule->filter.dst_addr, rule->action);
3710 		if (err)
3711 			return err;
3712 	}
3713 
3714 	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3715 		int prio = (rule->filter.vlan_tci & VLAN_PRIO_MASK) >>
3716 			   VLAN_PRIO_SHIFT;
3717 
3718 		err = igc_add_vlan_prio_filter(adapter, prio, rule->action);
3719 		if (err)
3720 			return err;
3721 	}
3722 
3723 	return 0;
3724 }
3725 
3726 static void igc_disable_nfc_rule(struct igc_adapter *adapter,
3727 				 const struct igc_nfc_rule *rule)
3728 {
3729 	if (rule->flex) {
3730 		igc_del_flex_filter(adapter, rule->filter.flex_index);
3731 		return;
3732 	}
3733 
3734 	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE)
3735 		igc_del_etype_filter(adapter, rule->filter.etype);
3736 
3737 	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3738 		int prio = (rule->filter.vlan_tci & VLAN_PRIO_MASK) >>
3739 			   VLAN_PRIO_SHIFT;
3740 
3741 		igc_del_vlan_prio_filter(adapter, prio);
3742 	}
3743 
3744 	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3745 		igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3746 				   rule->filter.src_addr);
3747 
3748 	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3749 		igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3750 				   rule->filter.dst_addr);
3751 }
3752 
3753 /**
3754  * igc_get_nfc_rule() - Get NFC rule
3755  * @adapter: Pointer to adapter
3756  * @location: Rule location
3757  *
3758  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3759  *
3760  * Return: Pointer to NFC rule at @location. If not found, NULL.
3761  */
3762 struct igc_nfc_rule *igc_get_nfc_rule(struct igc_adapter *adapter,
3763 				      u32 location)
3764 {
3765 	struct igc_nfc_rule *rule;
3766 
3767 	list_for_each_entry(rule, &adapter->nfc_rule_list, list) {
3768 		if (rule->location == location)
3769 			return rule;
3770 		if (rule->location > location)
3771 			break;
3772 	}
3773 
3774 	return NULL;
3775 }
3776 
3777 /**
3778  * igc_del_nfc_rule() - Delete NFC rule
3779  * @adapter: Pointer to adapter
3780  * @rule: Pointer to rule to be deleted
3781  *
3782  * Disable NFC rule in hardware and delete it from adapter.
3783  *
3784  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3785  */
3786 void igc_del_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3787 {
3788 	igc_disable_nfc_rule(adapter, rule);
3789 
3790 	list_del(&rule->list);
3791 	adapter->nfc_rule_count--;
3792 
3793 	kfree(rule);
3794 }
3795 
3796 static void igc_flush_nfc_rules(struct igc_adapter *adapter)
3797 {
3798 	struct igc_nfc_rule *rule, *tmp;
3799 
3800 	mutex_lock(&adapter->nfc_rule_lock);
3801 
3802 	list_for_each_entry_safe(rule, tmp, &adapter->nfc_rule_list, list)
3803 		igc_del_nfc_rule(adapter, rule);
3804 
3805 	mutex_unlock(&adapter->nfc_rule_lock);
3806 }
3807 
3808 /**
3809  * igc_add_nfc_rule() - Add NFC rule
3810  * @adapter: Pointer to adapter
3811  * @rule: Pointer to rule to be added
3812  *
3813  * Enable NFC rule in hardware and add it to adapter.
3814  *
3815  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3816  *
3817  * Return: 0 on success, negative errno on failure.
3818  */
3819 int igc_add_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3820 {
3821 	struct igc_nfc_rule *pred, *cur;
3822 	int err;
3823 
3824 	err = igc_enable_nfc_rule(adapter, rule);
3825 	if (err)
3826 		return err;
3827 
3828 	pred = NULL;
3829 	list_for_each_entry(cur, &adapter->nfc_rule_list, list) {
3830 		if (cur->location >= rule->location)
3831 			break;
3832 		pred = cur;
3833 	}
3834 
3835 	list_add(&rule->list, pred ? &pred->list : &adapter->nfc_rule_list);
3836 	adapter->nfc_rule_count++;
3837 	return 0;
3838 }
3839 
3840 static void igc_restore_nfc_rules(struct igc_adapter *adapter)
3841 {
3842 	struct igc_nfc_rule *rule;
3843 
3844 	mutex_lock(&adapter->nfc_rule_lock);
3845 
3846 	list_for_each_entry_reverse(rule, &adapter->nfc_rule_list, list)
3847 		igc_enable_nfc_rule(adapter, rule);
3848 
3849 	mutex_unlock(&adapter->nfc_rule_lock);
3850 }
3851 
3852 static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr)
3853 {
3854 	struct igc_adapter *adapter = netdev_priv(netdev);
3855 
3856 	return igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr, -1);
3857 }
3858 
3859 static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr)
3860 {
3861 	struct igc_adapter *adapter = netdev_priv(netdev);
3862 
3863 	igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr);
3864 	return 0;
3865 }
3866 
3867 /**
3868  * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3869  * @netdev: network interface device structure
3870  *
3871  * The set_rx_mode entry point is called whenever the unicast or multicast
3872  * address lists or the network interface flags are updated.  This routine is
3873  * responsible for configuring the hardware for proper unicast, multicast,
3874  * promiscuous mode, and all-multi behavior.
3875  */
3876 static void igc_set_rx_mode(struct net_device *netdev)
3877 {
3878 	struct igc_adapter *adapter = netdev_priv(netdev);
3879 	struct igc_hw *hw = &adapter->hw;
3880 	u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
3881 	int count;
3882 
3883 	/* Check for Promiscuous and All Multicast modes */
3884 	if (netdev->flags & IFF_PROMISC) {
3885 		rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE;
3886 	} else {
3887 		if (netdev->flags & IFF_ALLMULTI) {
3888 			rctl |= IGC_RCTL_MPE;
3889 		} else {
3890 			/* Write addresses to the MTA, if the attempt fails
3891 			 * then we should just turn on promiscuous mode so
3892 			 * that we can at least receive multicast traffic
3893 			 */
3894 			count = igc_write_mc_addr_list(netdev);
3895 			if (count < 0)
3896 				rctl |= IGC_RCTL_MPE;
3897 		}
3898 	}
3899 
3900 	/* Write addresses to available RAR registers, if there is not
3901 	 * sufficient space to store all the addresses then enable
3902 	 * unicast promiscuous mode
3903 	 */
3904 	if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync))
3905 		rctl |= IGC_RCTL_UPE;
3906 
3907 	/* update state of unicast and multicast */
3908 	rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
3909 	wr32(IGC_RCTL, rctl);
3910 
3911 #if (PAGE_SIZE < 8192)
3912 	if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB)
3913 		rlpml = IGC_MAX_FRAME_BUILD_SKB;
3914 #endif
3915 	wr32(IGC_RLPML, rlpml);
3916 }
3917 
3918 /**
3919  * igc_configure - configure the hardware for RX and TX
3920  * @adapter: private board structure
3921  */
3922 static void igc_configure(struct igc_adapter *adapter)
3923 {
3924 	struct net_device *netdev = adapter->netdev;
3925 	int i = 0;
3926 
3927 	igc_get_hw_control(adapter);
3928 	igc_set_rx_mode(netdev);
3929 
3930 	igc_restore_vlan(adapter);
3931 
3932 	igc_setup_tctl(adapter);
3933 	igc_setup_mrqc(adapter);
3934 	igc_setup_rctl(adapter);
3935 
3936 	igc_set_default_mac_filter(adapter);
3937 	igc_restore_nfc_rules(adapter);
3938 
3939 	igc_configure_tx(adapter);
3940 	igc_configure_rx(adapter);
3941 
3942 	igc_rx_fifo_flush_base(&adapter->hw);
3943 
3944 	/* call igc_desc_unused which always leaves
3945 	 * at least 1 descriptor unused to make sure
3946 	 * next_to_use != next_to_clean
3947 	 */
3948 	for (i = 0; i < adapter->num_rx_queues; i++) {
3949 		struct igc_ring *ring = adapter->rx_ring[i];
3950 
3951 		if (ring->xsk_pool)
3952 			igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
3953 		else
3954 			igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
3955 	}
3956 }
3957 
3958 /**
3959  * igc_write_ivar - configure ivar for given MSI-X vector
3960  * @hw: pointer to the HW structure
3961  * @msix_vector: vector number we are allocating to a given ring
3962  * @index: row index of IVAR register to write within IVAR table
3963  * @offset: column offset of in IVAR, should be multiple of 8
3964  *
3965  * The IVAR table consists of 2 columns,
3966  * each containing an cause allocation for an Rx and Tx ring, and a
3967  * variable number of rows depending on the number of queues supported.
3968  */
3969 static void igc_write_ivar(struct igc_hw *hw, int msix_vector,
3970 			   int index, int offset)
3971 {
3972 	u32 ivar = array_rd32(IGC_IVAR0, index);
3973 
3974 	/* clear any bits that are currently set */
3975 	ivar &= ~((u32)0xFF << offset);
3976 
3977 	/* write vector and valid bit */
3978 	ivar |= (msix_vector | IGC_IVAR_VALID) << offset;
3979 
3980 	array_wr32(IGC_IVAR0, index, ivar);
3981 }
3982 
3983 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector)
3984 {
3985 	struct igc_adapter *adapter = q_vector->adapter;
3986 	struct igc_hw *hw = &adapter->hw;
3987 	int rx_queue = IGC_N0_QUEUE;
3988 	int tx_queue = IGC_N0_QUEUE;
3989 
3990 	if (q_vector->rx.ring)
3991 		rx_queue = q_vector->rx.ring->reg_idx;
3992 	if (q_vector->tx.ring)
3993 		tx_queue = q_vector->tx.ring->reg_idx;
3994 
3995 	switch (hw->mac.type) {
3996 	case igc_i225:
3997 		if (rx_queue > IGC_N0_QUEUE)
3998 			igc_write_ivar(hw, msix_vector,
3999 				       rx_queue >> 1,
4000 				       (rx_queue & 0x1) << 4);
4001 		if (tx_queue > IGC_N0_QUEUE)
4002 			igc_write_ivar(hw, msix_vector,
4003 				       tx_queue >> 1,
4004 				       ((tx_queue & 0x1) << 4) + 8);
4005 		q_vector->eims_value = BIT(msix_vector);
4006 		break;
4007 	default:
4008 		WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n");
4009 		break;
4010 	}
4011 
4012 	/* add q_vector eims value to global eims_enable_mask */
4013 	adapter->eims_enable_mask |= q_vector->eims_value;
4014 
4015 	/* configure q_vector to set itr on first interrupt */
4016 	q_vector->set_itr = 1;
4017 }
4018 
4019 /**
4020  * igc_configure_msix - Configure MSI-X hardware
4021  * @adapter: Pointer to adapter structure
4022  *
4023  * igc_configure_msix sets up the hardware to properly
4024  * generate MSI-X interrupts.
4025  */
4026 static void igc_configure_msix(struct igc_adapter *adapter)
4027 {
4028 	struct igc_hw *hw = &adapter->hw;
4029 	int i, vector = 0;
4030 	u32 tmp;
4031 
4032 	adapter->eims_enable_mask = 0;
4033 
4034 	/* set vector for other causes, i.e. link changes */
4035 	switch (hw->mac.type) {
4036 	case igc_i225:
4037 		/* Turn on MSI-X capability first, or our settings
4038 		 * won't stick.  And it will take days to debug.
4039 		 */
4040 		wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE |
4041 		     IGC_GPIE_PBA | IGC_GPIE_EIAME |
4042 		     IGC_GPIE_NSICR);
4043 
4044 		/* enable msix_other interrupt */
4045 		adapter->eims_other = BIT(vector);
4046 		tmp = (vector++ | IGC_IVAR_VALID) << 8;
4047 
4048 		wr32(IGC_IVAR_MISC, tmp);
4049 		break;
4050 	default:
4051 		/* do nothing, since nothing else supports MSI-X */
4052 		break;
4053 	} /* switch (hw->mac.type) */
4054 
4055 	adapter->eims_enable_mask |= adapter->eims_other;
4056 
4057 	for (i = 0; i < adapter->num_q_vectors; i++)
4058 		igc_assign_vector(adapter->q_vector[i], vector++);
4059 
4060 	wrfl();
4061 }
4062 
4063 /**
4064  * igc_irq_enable - Enable default interrupt generation settings
4065  * @adapter: board private structure
4066  */
4067 static void igc_irq_enable(struct igc_adapter *adapter)
4068 {
4069 	struct igc_hw *hw = &adapter->hw;
4070 
4071 	if (adapter->msix_entries) {
4072 		u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA;
4073 		u32 regval = rd32(IGC_EIAC);
4074 
4075 		wr32(IGC_EIAC, regval | adapter->eims_enable_mask);
4076 		regval = rd32(IGC_EIAM);
4077 		wr32(IGC_EIAM, regval | adapter->eims_enable_mask);
4078 		wr32(IGC_EIMS, adapter->eims_enable_mask);
4079 		wr32(IGC_IMS, ims);
4080 	} else {
4081 		wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4082 		wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4083 	}
4084 }
4085 
4086 /**
4087  * igc_irq_disable - Mask off interrupt generation on the NIC
4088  * @adapter: board private structure
4089  */
4090 static void igc_irq_disable(struct igc_adapter *adapter)
4091 {
4092 	struct igc_hw *hw = &adapter->hw;
4093 
4094 	if (adapter->msix_entries) {
4095 		u32 regval = rd32(IGC_EIAM);
4096 
4097 		wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask);
4098 		wr32(IGC_EIMC, adapter->eims_enable_mask);
4099 		regval = rd32(IGC_EIAC);
4100 		wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask);
4101 	}
4102 
4103 	wr32(IGC_IAM, 0);
4104 	wr32(IGC_IMC, ~0);
4105 	wrfl();
4106 
4107 	if (adapter->msix_entries) {
4108 		int vector = 0, i;
4109 
4110 		synchronize_irq(adapter->msix_entries[vector++].vector);
4111 
4112 		for (i = 0; i < adapter->num_q_vectors; i++)
4113 			synchronize_irq(adapter->msix_entries[vector++].vector);
4114 	} else {
4115 		synchronize_irq(adapter->pdev->irq);
4116 	}
4117 }
4118 
4119 void igc_set_flag_queue_pairs(struct igc_adapter *adapter,
4120 			      const u32 max_rss_queues)
4121 {
4122 	/* Determine if we need to pair queues. */
4123 	/* If rss_queues > half of max_rss_queues, pair the queues in
4124 	 * order to conserve interrupts due to limited supply.
4125 	 */
4126 	if (adapter->rss_queues > (max_rss_queues / 2))
4127 		adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4128 	else
4129 		adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS;
4130 }
4131 
4132 unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter)
4133 {
4134 	return IGC_MAX_RX_QUEUES;
4135 }
4136 
4137 static void igc_init_queue_configuration(struct igc_adapter *adapter)
4138 {
4139 	u32 max_rss_queues;
4140 
4141 	max_rss_queues = igc_get_max_rss_queues(adapter);
4142 	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
4143 
4144 	igc_set_flag_queue_pairs(adapter, max_rss_queues);
4145 }
4146 
4147 /**
4148  * igc_reset_q_vector - Reset config for interrupt vector
4149  * @adapter: board private structure to initialize
4150  * @v_idx: Index of vector to be reset
4151  *
4152  * If NAPI is enabled it will delete any references to the
4153  * NAPI struct. This is preparation for igc_free_q_vector.
4154  */
4155 static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx)
4156 {
4157 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
4158 
4159 	/* if we're coming from igc_set_interrupt_capability, the vectors are
4160 	 * not yet allocated
4161 	 */
4162 	if (!q_vector)
4163 		return;
4164 
4165 	if (q_vector->tx.ring)
4166 		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
4167 
4168 	if (q_vector->rx.ring)
4169 		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
4170 
4171 	netif_napi_del(&q_vector->napi);
4172 }
4173 
4174 /**
4175  * igc_free_q_vector - Free memory allocated for specific interrupt vector
4176  * @adapter: board private structure to initialize
4177  * @v_idx: Index of vector to be freed
4178  *
4179  * This function frees the memory allocated to the q_vector.
4180  */
4181 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx)
4182 {
4183 	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
4184 
4185 	adapter->q_vector[v_idx] = NULL;
4186 
4187 	/* igc_get_stats64() might access the rings on this vector,
4188 	 * we must wait a grace period before freeing it.
4189 	 */
4190 	if (q_vector)
4191 		kfree_rcu(q_vector, rcu);
4192 }
4193 
4194 /**
4195  * igc_free_q_vectors - Free memory allocated for interrupt vectors
4196  * @adapter: board private structure to initialize
4197  *
4198  * This function frees the memory allocated to the q_vectors.  In addition if
4199  * NAPI is enabled it will delete any references to the NAPI struct prior
4200  * to freeing the q_vector.
4201  */
4202 static void igc_free_q_vectors(struct igc_adapter *adapter)
4203 {
4204 	int v_idx = adapter->num_q_vectors;
4205 
4206 	adapter->num_tx_queues = 0;
4207 	adapter->num_rx_queues = 0;
4208 	adapter->num_q_vectors = 0;
4209 
4210 	while (v_idx--) {
4211 		igc_reset_q_vector(adapter, v_idx);
4212 		igc_free_q_vector(adapter, v_idx);
4213 	}
4214 }
4215 
4216 /**
4217  * igc_update_itr - update the dynamic ITR value based on statistics
4218  * @q_vector: pointer to q_vector
4219  * @ring_container: ring info to update the itr for
4220  *
4221  * Stores a new ITR value based on packets and byte
4222  * counts during the last interrupt.  The advantage of per interrupt
4223  * computation is faster updates and more accurate ITR for the current
4224  * traffic pattern.  Constants in this function were computed
4225  * based on theoretical maximum wire speed and thresholds were set based
4226  * on testing data as well as attempting to minimize response time
4227  * while increasing bulk throughput.
4228  * NOTE: These calculations are only valid when operating in a single-
4229  * queue environment.
4230  */
4231 static void igc_update_itr(struct igc_q_vector *q_vector,
4232 			   struct igc_ring_container *ring_container)
4233 {
4234 	unsigned int packets = ring_container->total_packets;
4235 	unsigned int bytes = ring_container->total_bytes;
4236 	u8 itrval = ring_container->itr;
4237 
4238 	/* no packets, exit with status unchanged */
4239 	if (packets == 0)
4240 		return;
4241 
4242 	switch (itrval) {
4243 	case lowest_latency:
4244 		/* handle TSO and jumbo frames */
4245 		if (bytes / packets > 8000)
4246 			itrval = bulk_latency;
4247 		else if ((packets < 5) && (bytes > 512))
4248 			itrval = low_latency;
4249 		break;
4250 	case low_latency:  /* 50 usec aka 20000 ints/s */
4251 		if (bytes > 10000) {
4252 			/* this if handles the TSO accounting */
4253 			if (bytes / packets > 8000)
4254 				itrval = bulk_latency;
4255 			else if ((packets < 10) || ((bytes / packets) > 1200))
4256 				itrval = bulk_latency;
4257 			else if ((packets > 35))
4258 				itrval = lowest_latency;
4259 		} else if (bytes / packets > 2000) {
4260 			itrval = bulk_latency;
4261 		} else if (packets <= 2 && bytes < 512) {
4262 			itrval = lowest_latency;
4263 		}
4264 		break;
4265 	case bulk_latency: /* 250 usec aka 4000 ints/s */
4266 		if (bytes > 25000) {
4267 			if (packets > 35)
4268 				itrval = low_latency;
4269 		} else if (bytes < 1500) {
4270 			itrval = low_latency;
4271 		}
4272 		break;
4273 	}
4274 
4275 	/* clear work counters since we have the values we need */
4276 	ring_container->total_bytes = 0;
4277 	ring_container->total_packets = 0;
4278 
4279 	/* write updated itr to ring container */
4280 	ring_container->itr = itrval;
4281 }
4282 
4283 static void igc_set_itr(struct igc_q_vector *q_vector)
4284 {
4285 	struct igc_adapter *adapter = q_vector->adapter;
4286 	u32 new_itr = q_vector->itr_val;
4287 	u8 current_itr = 0;
4288 
4289 	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4290 	switch (adapter->link_speed) {
4291 	case SPEED_10:
4292 	case SPEED_100:
4293 		current_itr = 0;
4294 		new_itr = IGC_4K_ITR;
4295 		goto set_itr_now;
4296 	default:
4297 		break;
4298 	}
4299 
4300 	igc_update_itr(q_vector, &q_vector->tx);
4301 	igc_update_itr(q_vector, &q_vector->rx);
4302 
4303 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4304 
4305 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4306 	if (current_itr == lowest_latency &&
4307 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4308 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4309 		current_itr = low_latency;
4310 
4311 	switch (current_itr) {
4312 	/* counts and packets in update_itr are dependent on these numbers */
4313 	case lowest_latency:
4314 		new_itr = IGC_70K_ITR; /* 70,000 ints/sec */
4315 		break;
4316 	case low_latency:
4317 		new_itr = IGC_20K_ITR; /* 20,000 ints/sec */
4318 		break;
4319 	case bulk_latency:
4320 		new_itr = IGC_4K_ITR;  /* 4,000 ints/sec */
4321 		break;
4322 	default:
4323 		break;
4324 	}
4325 
4326 set_itr_now:
4327 	if (new_itr != q_vector->itr_val) {
4328 		/* this attempts to bias the interrupt rate towards Bulk
4329 		 * by adding intermediate steps when interrupt rate is
4330 		 * increasing
4331 		 */
4332 		new_itr = new_itr > q_vector->itr_val ?
4333 			  max((new_itr * q_vector->itr_val) /
4334 			  (new_itr + (q_vector->itr_val >> 2)),
4335 			  new_itr) : new_itr;
4336 		/* Don't write the value here; it resets the adapter's
4337 		 * internal timer, and causes us to delay far longer than
4338 		 * we should between interrupts.  Instead, we write the ITR
4339 		 * value at the beginning of the next interrupt so the timing
4340 		 * ends up being correct.
4341 		 */
4342 		q_vector->itr_val = new_itr;
4343 		q_vector->set_itr = 1;
4344 	}
4345 }
4346 
4347 static void igc_reset_interrupt_capability(struct igc_adapter *adapter)
4348 {
4349 	int v_idx = adapter->num_q_vectors;
4350 
4351 	if (adapter->msix_entries) {
4352 		pci_disable_msix(adapter->pdev);
4353 		kfree(adapter->msix_entries);
4354 		adapter->msix_entries = NULL;
4355 	} else if (adapter->flags & IGC_FLAG_HAS_MSI) {
4356 		pci_disable_msi(adapter->pdev);
4357 	}
4358 
4359 	while (v_idx--)
4360 		igc_reset_q_vector(adapter, v_idx);
4361 }
4362 
4363 /**
4364  * igc_set_interrupt_capability - set MSI or MSI-X if supported
4365  * @adapter: Pointer to adapter structure
4366  * @msix: boolean value for MSI-X capability
4367  *
4368  * Attempt to configure interrupts using the best available
4369  * capabilities of the hardware and kernel.
4370  */
4371 static void igc_set_interrupt_capability(struct igc_adapter *adapter,
4372 					 bool msix)
4373 {
4374 	int numvecs, i;
4375 	int err;
4376 
4377 	if (!msix)
4378 		goto msi_only;
4379 	adapter->flags |= IGC_FLAG_HAS_MSIX;
4380 
4381 	/* Number of supported queues. */
4382 	adapter->num_rx_queues = adapter->rss_queues;
4383 
4384 	adapter->num_tx_queues = adapter->rss_queues;
4385 
4386 	/* start with one vector for every Rx queue */
4387 	numvecs = adapter->num_rx_queues;
4388 
4389 	/* if Tx handler is separate add 1 for every Tx queue */
4390 	if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS))
4391 		numvecs += adapter->num_tx_queues;
4392 
4393 	/* store the number of vectors reserved for queues */
4394 	adapter->num_q_vectors = numvecs;
4395 
4396 	/* add 1 vector for link status interrupts */
4397 	numvecs++;
4398 
4399 	adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
4400 					GFP_KERNEL);
4401 
4402 	if (!adapter->msix_entries)
4403 		return;
4404 
4405 	/* populate entry values */
4406 	for (i = 0; i < numvecs; i++)
4407 		adapter->msix_entries[i].entry = i;
4408 
4409 	err = pci_enable_msix_range(adapter->pdev,
4410 				    adapter->msix_entries,
4411 				    numvecs,
4412 				    numvecs);
4413 	if (err > 0)
4414 		return;
4415 
4416 	kfree(adapter->msix_entries);
4417 	adapter->msix_entries = NULL;
4418 
4419 	igc_reset_interrupt_capability(adapter);
4420 
4421 msi_only:
4422 	adapter->flags &= ~IGC_FLAG_HAS_MSIX;
4423 
4424 	adapter->rss_queues = 1;
4425 	adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4426 	adapter->num_rx_queues = 1;
4427 	adapter->num_tx_queues = 1;
4428 	adapter->num_q_vectors = 1;
4429 	if (!pci_enable_msi(adapter->pdev))
4430 		adapter->flags |= IGC_FLAG_HAS_MSI;
4431 }
4432 
4433 /**
4434  * igc_update_ring_itr - update the dynamic ITR value based on packet size
4435  * @q_vector: pointer to q_vector
4436  *
4437  * Stores a new ITR value based on strictly on packet size.  This
4438  * algorithm is less sophisticated than that used in igc_update_itr,
4439  * due to the difficulty of synchronizing statistics across multiple
4440  * receive rings.  The divisors and thresholds used by this function
4441  * were determined based on theoretical maximum wire speed and testing
4442  * data, in order to minimize response time while increasing bulk
4443  * throughput.
4444  * NOTE: This function is called only when operating in a multiqueue
4445  * receive environment.
4446  */
4447 static void igc_update_ring_itr(struct igc_q_vector *q_vector)
4448 {
4449 	struct igc_adapter *adapter = q_vector->adapter;
4450 	int new_val = q_vector->itr_val;
4451 	int avg_wire_size = 0;
4452 	unsigned int packets;
4453 
4454 	/* For non-gigabit speeds, just fix the interrupt rate at 4000
4455 	 * ints/sec - ITR timer value of 120 ticks.
4456 	 */
4457 	switch (adapter->link_speed) {
4458 	case SPEED_10:
4459 	case SPEED_100:
4460 		new_val = IGC_4K_ITR;
4461 		goto set_itr_val;
4462 	default:
4463 		break;
4464 	}
4465 
4466 	packets = q_vector->rx.total_packets;
4467 	if (packets)
4468 		avg_wire_size = q_vector->rx.total_bytes / packets;
4469 
4470 	packets = q_vector->tx.total_packets;
4471 	if (packets)
4472 		avg_wire_size = max_t(u32, avg_wire_size,
4473 				      q_vector->tx.total_bytes / packets);
4474 
4475 	/* if avg_wire_size isn't set no work was done */
4476 	if (!avg_wire_size)
4477 		goto clear_counts;
4478 
4479 	/* Add 24 bytes to size to account for CRC, preamble, and gap */
4480 	avg_wire_size += 24;
4481 
4482 	/* Don't starve jumbo frames */
4483 	avg_wire_size = min(avg_wire_size, 3000);
4484 
4485 	/* Give a little boost to mid-size frames */
4486 	if (avg_wire_size > 300 && avg_wire_size < 1200)
4487 		new_val = avg_wire_size / 3;
4488 	else
4489 		new_val = avg_wire_size / 2;
4490 
4491 	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4492 	if (new_val < IGC_20K_ITR &&
4493 	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4494 	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4495 		new_val = IGC_20K_ITR;
4496 
4497 set_itr_val:
4498 	if (new_val != q_vector->itr_val) {
4499 		q_vector->itr_val = new_val;
4500 		q_vector->set_itr = 1;
4501 	}
4502 clear_counts:
4503 	q_vector->rx.total_bytes = 0;
4504 	q_vector->rx.total_packets = 0;
4505 	q_vector->tx.total_bytes = 0;
4506 	q_vector->tx.total_packets = 0;
4507 }
4508 
4509 static void igc_ring_irq_enable(struct igc_q_vector *q_vector)
4510 {
4511 	struct igc_adapter *adapter = q_vector->adapter;
4512 	struct igc_hw *hw = &adapter->hw;
4513 
4514 	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
4515 	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
4516 		if (adapter->num_q_vectors == 1)
4517 			igc_set_itr(q_vector);
4518 		else
4519 			igc_update_ring_itr(q_vector);
4520 	}
4521 
4522 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
4523 		if (adapter->msix_entries)
4524 			wr32(IGC_EIMS, q_vector->eims_value);
4525 		else
4526 			igc_irq_enable(adapter);
4527 	}
4528 }
4529 
4530 static void igc_add_ring(struct igc_ring *ring,
4531 			 struct igc_ring_container *head)
4532 {
4533 	head->ring = ring;
4534 	head->count++;
4535 }
4536 
4537 /**
4538  * igc_cache_ring_register - Descriptor ring to register mapping
4539  * @adapter: board private structure to initialize
4540  *
4541  * Once we know the feature-set enabled for the device, we'll cache
4542  * the register offset the descriptor ring is assigned to.
4543  */
4544 static void igc_cache_ring_register(struct igc_adapter *adapter)
4545 {
4546 	int i = 0, j = 0;
4547 
4548 	switch (adapter->hw.mac.type) {
4549 	case igc_i225:
4550 	default:
4551 		for (; i < adapter->num_rx_queues; i++)
4552 			adapter->rx_ring[i]->reg_idx = i;
4553 		for (; j < adapter->num_tx_queues; j++)
4554 			adapter->tx_ring[j]->reg_idx = j;
4555 		break;
4556 	}
4557 }
4558 
4559 /**
4560  * igc_poll - NAPI Rx polling callback
4561  * @napi: napi polling structure
4562  * @budget: count of how many packets we should handle
4563  */
4564 static int igc_poll(struct napi_struct *napi, int budget)
4565 {
4566 	struct igc_q_vector *q_vector = container_of(napi,
4567 						     struct igc_q_vector,
4568 						     napi);
4569 	struct igc_ring *rx_ring = q_vector->rx.ring;
4570 	bool clean_complete = true;
4571 	int work_done = 0;
4572 
4573 	if (q_vector->tx.ring)
4574 		clean_complete = igc_clean_tx_irq(q_vector, budget);
4575 
4576 	if (rx_ring) {
4577 		int cleaned = rx_ring->xsk_pool ?
4578 			      igc_clean_rx_irq_zc(q_vector, budget) :
4579 			      igc_clean_rx_irq(q_vector, budget);
4580 
4581 		work_done += cleaned;
4582 		if (cleaned >= budget)
4583 			clean_complete = false;
4584 	}
4585 
4586 	/* If all work not completed, return budget and keep polling */
4587 	if (!clean_complete)
4588 		return budget;
4589 
4590 	/* Exit the polling mode, but don't re-enable interrupts if stack might
4591 	 * poll us due to busy-polling
4592 	 */
4593 	if (likely(napi_complete_done(napi, work_done)))
4594 		igc_ring_irq_enable(q_vector);
4595 
4596 	return min(work_done, budget - 1);
4597 }
4598 
4599 /**
4600  * igc_alloc_q_vector - Allocate memory for a single interrupt vector
4601  * @adapter: board private structure to initialize
4602  * @v_count: q_vectors allocated on adapter, used for ring interleaving
4603  * @v_idx: index of vector in adapter struct
4604  * @txr_count: total number of Tx rings to allocate
4605  * @txr_idx: index of first Tx ring to allocate
4606  * @rxr_count: total number of Rx rings to allocate
4607  * @rxr_idx: index of first Rx ring to allocate
4608  *
4609  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
4610  */
4611 static int igc_alloc_q_vector(struct igc_adapter *adapter,
4612 			      unsigned int v_count, unsigned int v_idx,
4613 			      unsigned int txr_count, unsigned int txr_idx,
4614 			      unsigned int rxr_count, unsigned int rxr_idx)
4615 {
4616 	struct igc_q_vector *q_vector;
4617 	struct igc_ring *ring;
4618 	int ring_count;
4619 
4620 	/* igc only supports 1 Tx and/or 1 Rx queue per vector */
4621 	if (txr_count > 1 || rxr_count > 1)
4622 		return -ENOMEM;
4623 
4624 	ring_count = txr_count + rxr_count;
4625 
4626 	/* allocate q_vector and rings */
4627 	q_vector = adapter->q_vector[v_idx];
4628 	if (!q_vector)
4629 		q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
4630 				   GFP_KERNEL);
4631 	else
4632 		memset(q_vector, 0, struct_size(q_vector, ring, ring_count));
4633 	if (!q_vector)
4634 		return -ENOMEM;
4635 
4636 	/* initialize NAPI */
4637 	netif_napi_add(adapter->netdev, &q_vector->napi, igc_poll);
4638 
4639 	/* tie q_vector and adapter together */
4640 	adapter->q_vector[v_idx] = q_vector;
4641 	q_vector->adapter = adapter;
4642 
4643 	/* initialize work limits */
4644 	q_vector->tx.work_limit = adapter->tx_work_limit;
4645 
4646 	/* initialize ITR configuration */
4647 	q_vector->itr_register = adapter->io_addr + IGC_EITR(0);
4648 	q_vector->itr_val = IGC_START_ITR;
4649 
4650 	/* initialize pointer to rings */
4651 	ring = q_vector->ring;
4652 
4653 	/* initialize ITR */
4654 	if (rxr_count) {
4655 		/* rx or rx/tx vector */
4656 		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
4657 			q_vector->itr_val = adapter->rx_itr_setting;
4658 	} else {
4659 		/* tx only vector */
4660 		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
4661 			q_vector->itr_val = adapter->tx_itr_setting;
4662 	}
4663 
4664 	if (txr_count) {
4665 		/* assign generic ring traits */
4666 		ring->dev = &adapter->pdev->dev;
4667 		ring->netdev = adapter->netdev;
4668 
4669 		/* configure backlink on ring */
4670 		ring->q_vector = q_vector;
4671 
4672 		/* update q_vector Tx values */
4673 		igc_add_ring(ring, &q_vector->tx);
4674 
4675 		/* apply Tx specific ring traits */
4676 		ring->count = adapter->tx_ring_count;
4677 		ring->queue_index = txr_idx;
4678 
4679 		/* assign ring to adapter */
4680 		adapter->tx_ring[txr_idx] = ring;
4681 
4682 		/* push pointer to next ring */
4683 		ring++;
4684 	}
4685 
4686 	if (rxr_count) {
4687 		/* assign generic ring traits */
4688 		ring->dev = &adapter->pdev->dev;
4689 		ring->netdev = adapter->netdev;
4690 
4691 		/* configure backlink on ring */
4692 		ring->q_vector = q_vector;
4693 
4694 		/* update q_vector Rx values */
4695 		igc_add_ring(ring, &q_vector->rx);
4696 
4697 		/* apply Rx specific ring traits */
4698 		ring->count = adapter->rx_ring_count;
4699 		ring->queue_index = rxr_idx;
4700 
4701 		/* assign ring to adapter */
4702 		adapter->rx_ring[rxr_idx] = ring;
4703 	}
4704 
4705 	return 0;
4706 }
4707 
4708 /**
4709  * igc_alloc_q_vectors - Allocate memory for interrupt vectors
4710  * @adapter: board private structure to initialize
4711  *
4712  * We allocate one q_vector per queue interrupt.  If allocation fails we
4713  * return -ENOMEM.
4714  */
4715 static int igc_alloc_q_vectors(struct igc_adapter *adapter)
4716 {
4717 	int rxr_remaining = adapter->num_rx_queues;
4718 	int txr_remaining = adapter->num_tx_queues;
4719 	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
4720 	int q_vectors = adapter->num_q_vectors;
4721 	int err;
4722 
4723 	if (q_vectors >= (rxr_remaining + txr_remaining)) {
4724 		for (; rxr_remaining; v_idx++) {
4725 			err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4726 						 0, 0, 1, rxr_idx);
4727 
4728 			if (err)
4729 				goto err_out;
4730 
4731 			/* update counts and index */
4732 			rxr_remaining--;
4733 			rxr_idx++;
4734 		}
4735 	}
4736 
4737 	for (; v_idx < q_vectors; v_idx++) {
4738 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
4739 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
4740 
4741 		err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4742 					 tqpv, txr_idx, rqpv, rxr_idx);
4743 
4744 		if (err)
4745 			goto err_out;
4746 
4747 		/* update counts and index */
4748 		rxr_remaining -= rqpv;
4749 		txr_remaining -= tqpv;
4750 		rxr_idx++;
4751 		txr_idx++;
4752 	}
4753 
4754 	return 0;
4755 
4756 err_out:
4757 	adapter->num_tx_queues = 0;
4758 	adapter->num_rx_queues = 0;
4759 	adapter->num_q_vectors = 0;
4760 
4761 	while (v_idx--)
4762 		igc_free_q_vector(adapter, v_idx);
4763 
4764 	return -ENOMEM;
4765 }
4766 
4767 /**
4768  * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
4769  * @adapter: Pointer to adapter structure
4770  * @msix: boolean for MSI-X capability
4771  *
4772  * This function initializes the interrupts and allocates all of the queues.
4773  */
4774 static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix)
4775 {
4776 	struct net_device *dev = adapter->netdev;
4777 	int err = 0;
4778 
4779 	igc_set_interrupt_capability(adapter, msix);
4780 
4781 	err = igc_alloc_q_vectors(adapter);
4782 	if (err) {
4783 		netdev_err(dev, "Unable to allocate memory for vectors\n");
4784 		goto err_alloc_q_vectors;
4785 	}
4786 
4787 	igc_cache_ring_register(adapter);
4788 
4789 	return 0;
4790 
4791 err_alloc_q_vectors:
4792 	igc_reset_interrupt_capability(adapter);
4793 	return err;
4794 }
4795 
4796 /**
4797  * igc_sw_init - Initialize general software structures (struct igc_adapter)
4798  * @adapter: board private structure to initialize
4799  *
4800  * igc_sw_init initializes the Adapter private data structure.
4801  * Fields are initialized based on PCI device information and
4802  * OS network device settings (MTU size).
4803  */
4804 static int igc_sw_init(struct igc_adapter *adapter)
4805 {
4806 	struct net_device *netdev = adapter->netdev;
4807 	struct pci_dev *pdev = adapter->pdev;
4808 	struct igc_hw *hw = &adapter->hw;
4809 
4810 	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4811 
4812 	/* set default ring sizes */
4813 	adapter->tx_ring_count = IGC_DEFAULT_TXD;
4814 	adapter->rx_ring_count = IGC_DEFAULT_RXD;
4815 
4816 	/* set default ITR values */
4817 	adapter->rx_itr_setting = IGC_DEFAULT_ITR;
4818 	adapter->tx_itr_setting = IGC_DEFAULT_ITR;
4819 
4820 	/* set default work limits */
4821 	adapter->tx_work_limit = IGC_DEFAULT_TX_WORK;
4822 
4823 	/* adjust max frame to be at least the size of a standard frame */
4824 	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
4825 				VLAN_HLEN;
4826 	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4827 
4828 	mutex_init(&adapter->nfc_rule_lock);
4829 	INIT_LIST_HEAD(&adapter->nfc_rule_list);
4830 	adapter->nfc_rule_count = 0;
4831 
4832 	spin_lock_init(&adapter->stats64_lock);
4833 	spin_lock_init(&adapter->qbv_tx_lock);
4834 	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
4835 	adapter->flags |= IGC_FLAG_HAS_MSIX;
4836 
4837 	igc_init_queue_configuration(adapter);
4838 
4839 	/* This call may decrease the number of queues */
4840 	if (igc_init_interrupt_scheme(adapter, true)) {
4841 		netdev_err(netdev, "Unable to allocate memory for queues\n");
4842 		return -ENOMEM;
4843 	}
4844 
4845 	/* Explicitly disable IRQ since the NIC can be in any state. */
4846 	igc_irq_disable(adapter);
4847 
4848 	set_bit(__IGC_DOWN, &adapter->state);
4849 
4850 	return 0;
4851 }
4852 
4853 /**
4854  * igc_up - Open the interface and prepare it to handle traffic
4855  * @adapter: board private structure
4856  */
4857 void igc_up(struct igc_adapter *adapter)
4858 {
4859 	struct igc_hw *hw = &adapter->hw;
4860 	int i = 0;
4861 
4862 	/* hardware has been reset, we need to reload some things */
4863 	igc_configure(adapter);
4864 
4865 	clear_bit(__IGC_DOWN, &adapter->state);
4866 
4867 	for (i = 0; i < adapter->num_q_vectors; i++)
4868 		napi_enable(&adapter->q_vector[i]->napi);
4869 
4870 	if (adapter->msix_entries)
4871 		igc_configure_msix(adapter);
4872 	else
4873 		igc_assign_vector(adapter->q_vector[0], 0);
4874 
4875 	/* Clear any pending interrupts. */
4876 	rd32(IGC_ICR);
4877 	igc_irq_enable(adapter);
4878 
4879 	netif_tx_start_all_queues(adapter->netdev);
4880 
4881 	/* start the watchdog. */
4882 	hw->mac.get_link_status = true;
4883 	schedule_work(&adapter->watchdog_task);
4884 }
4885 
4886 /**
4887  * igc_update_stats - Update the board statistics counters
4888  * @adapter: board private structure
4889  */
4890 void igc_update_stats(struct igc_adapter *adapter)
4891 {
4892 	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
4893 	struct pci_dev *pdev = adapter->pdev;
4894 	struct igc_hw *hw = &adapter->hw;
4895 	u64 _bytes, _packets;
4896 	u64 bytes, packets;
4897 	unsigned int start;
4898 	u32 mpc;
4899 	int i;
4900 
4901 	/* Prevent stats update while adapter is being reset, or if the pci
4902 	 * connection is down.
4903 	 */
4904 	if (adapter->link_speed == 0)
4905 		return;
4906 	if (pci_channel_offline(pdev))
4907 		return;
4908 
4909 	packets = 0;
4910 	bytes = 0;
4911 
4912 	rcu_read_lock();
4913 	for (i = 0; i < adapter->num_rx_queues; i++) {
4914 		struct igc_ring *ring = adapter->rx_ring[i];
4915 		u32 rqdpc = rd32(IGC_RQDPC(i));
4916 
4917 		if (hw->mac.type >= igc_i225)
4918 			wr32(IGC_RQDPC(i), 0);
4919 
4920 		if (rqdpc) {
4921 			ring->rx_stats.drops += rqdpc;
4922 			net_stats->rx_fifo_errors += rqdpc;
4923 		}
4924 
4925 		do {
4926 			start = u64_stats_fetch_begin(&ring->rx_syncp);
4927 			_bytes = ring->rx_stats.bytes;
4928 			_packets = ring->rx_stats.packets;
4929 		} while (u64_stats_fetch_retry(&ring->rx_syncp, start));
4930 		bytes += _bytes;
4931 		packets += _packets;
4932 	}
4933 
4934 	net_stats->rx_bytes = bytes;
4935 	net_stats->rx_packets = packets;
4936 
4937 	packets = 0;
4938 	bytes = 0;
4939 	for (i = 0; i < adapter->num_tx_queues; i++) {
4940 		struct igc_ring *ring = adapter->tx_ring[i];
4941 
4942 		do {
4943 			start = u64_stats_fetch_begin(&ring->tx_syncp);
4944 			_bytes = ring->tx_stats.bytes;
4945 			_packets = ring->tx_stats.packets;
4946 		} while (u64_stats_fetch_retry(&ring->tx_syncp, start));
4947 		bytes += _bytes;
4948 		packets += _packets;
4949 	}
4950 	net_stats->tx_bytes = bytes;
4951 	net_stats->tx_packets = packets;
4952 	rcu_read_unlock();
4953 
4954 	/* read stats registers */
4955 	adapter->stats.crcerrs += rd32(IGC_CRCERRS);
4956 	adapter->stats.gprc += rd32(IGC_GPRC);
4957 	adapter->stats.gorc += rd32(IGC_GORCL);
4958 	rd32(IGC_GORCH); /* clear GORCL */
4959 	adapter->stats.bprc += rd32(IGC_BPRC);
4960 	adapter->stats.mprc += rd32(IGC_MPRC);
4961 	adapter->stats.roc += rd32(IGC_ROC);
4962 
4963 	adapter->stats.prc64 += rd32(IGC_PRC64);
4964 	adapter->stats.prc127 += rd32(IGC_PRC127);
4965 	adapter->stats.prc255 += rd32(IGC_PRC255);
4966 	adapter->stats.prc511 += rd32(IGC_PRC511);
4967 	adapter->stats.prc1023 += rd32(IGC_PRC1023);
4968 	adapter->stats.prc1522 += rd32(IGC_PRC1522);
4969 	adapter->stats.tlpic += rd32(IGC_TLPIC);
4970 	adapter->stats.rlpic += rd32(IGC_RLPIC);
4971 	adapter->stats.hgptc += rd32(IGC_HGPTC);
4972 
4973 	mpc = rd32(IGC_MPC);
4974 	adapter->stats.mpc += mpc;
4975 	net_stats->rx_fifo_errors += mpc;
4976 	adapter->stats.scc += rd32(IGC_SCC);
4977 	adapter->stats.ecol += rd32(IGC_ECOL);
4978 	adapter->stats.mcc += rd32(IGC_MCC);
4979 	adapter->stats.latecol += rd32(IGC_LATECOL);
4980 	adapter->stats.dc += rd32(IGC_DC);
4981 	adapter->stats.rlec += rd32(IGC_RLEC);
4982 	adapter->stats.xonrxc += rd32(IGC_XONRXC);
4983 	adapter->stats.xontxc += rd32(IGC_XONTXC);
4984 	adapter->stats.xoffrxc += rd32(IGC_XOFFRXC);
4985 	adapter->stats.xofftxc += rd32(IGC_XOFFTXC);
4986 	adapter->stats.fcruc += rd32(IGC_FCRUC);
4987 	adapter->stats.gptc += rd32(IGC_GPTC);
4988 	adapter->stats.gotc += rd32(IGC_GOTCL);
4989 	rd32(IGC_GOTCH); /* clear GOTCL */
4990 	adapter->stats.rnbc += rd32(IGC_RNBC);
4991 	adapter->stats.ruc += rd32(IGC_RUC);
4992 	adapter->stats.rfc += rd32(IGC_RFC);
4993 	adapter->stats.rjc += rd32(IGC_RJC);
4994 	adapter->stats.tor += rd32(IGC_TORH);
4995 	adapter->stats.tot += rd32(IGC_TOTH);
4996 	adapter->stats.tpr += rd32(IGC_TPR);
4997 
4998 	adapter->stats.ptc64 += rd32(IGC_PTC64);
4999 	adapter->stats.ptc127 += rd32(IGC_PTC127);
5000 	adapter->stats.ptc255 += rd32(IGC_PTC255);
5001 	adapter->stats.ptc511 += rd32(IGC_PTC511);
5002 	adapter->stats.ptc1023 += rd32(IGC_PTC1023);
5003 	adapter->stats.ptc1522 += rd32(IGC_PTC1522);
5004 
5005 	adapter->stats.mptc += rd32(IGC_MPTC);
5006 	adapter->stats.bptc += rd32(IGC_BPTC);
5007 
5008 	adapter->stats.tpt += rd32(IGC_TPT);
5009 	adapter->stats.colc += rd32(IGC_COLC);
5010 	adapter->stats.colc += rd32(IGC_RERC);
5011 
5012 	adapter->stats.algnerrc += rd32(IGC_ALGNERRC);
5013 
5014 	adapter->stats.tsctc += rd32(IGC_TSCTC);
5015 
5016 	adapter->stats.iac += rd32(IGC_IAC);
5017 
5018 	/* Fill out the OS statistics structure */
5019 	net_stats->multicast = adapter->stats.mprc;
5020 	net_stats->collisions = adapter->stats.colc;
5021 
5022 	/* Rx Errors */
5023 
5024 	/* RLEC on some newer hardware can be incorrect so build
5025 	 * our own version based on RUC and ROC
5026 	 */
5027 	net_stats->rx_errors = adapter->stats.rxerrc +
5028 		adapter->stats.crcerrs + adapter->stats.algnerrc +
5029 		adapter->stats.ruc + adapter->stats.roc +
5030 		adapter->stats.cexterr;
5031 	net_stats->rx_length_errors = adapter->stats.ruc +
5032 				      adapter->stats.roc;
5033 	net_stats->rx_crc_errors = adapter->stats.crcerrs;
5034 	net_stats->rx_frame_errors = adapter->stats.algnerrc;
5035 	net_stats->rx_missed_errors = adapter->stats.mpc;
5036 
5037 	/* Tx Errors */
5038 	net_stats->tx_errors = adapter->stats.ecol +
5039 			       adapter->stats.latecol;
5040 	net_stats->tx_aborted_errors = adapter->stats.ecol;
5041 	net_stats->tx_window_errors = adapter->stats.latecol;
5042 	net_stats->tx_carrier_errors = adapter->stats.tncrs;
5043 
5044 	/* Tx Dropped */
5045 	net_stats->tx_dropped = adapter->stats.txdrop;
5046 
5047 	/* Management Stats */
5048 	adapter->stats.mgptc += rd32(IGC_MGTPTC);
5049 	adapter->stats.mgprc += rd32(IGC_MGTPRC);
5050 	adapter->stats.mgpdc += rd32(IGC_MGTPDC);
5051 }
5052 
5053 /**
5054  * igc_down - Close the interface
5055  * @adapter: board private structure
5056  */
5057 void igc_down(struct igc_adapter *adapter)
5058 {
5059 	struct net_device *netdev = adapter->netdev;
5060 	struct igc_hw *hw = &adapter->hw;
5061 	u32 tctl, rctl;
5062 	int i = 0;
5063 
5064 	set_bit(__IGC_DOWN, &adapter->state);
5065 
5066 	igc_ptp_suspend(adapter);
5067 
5068 	if (pci_device_is_present(adapter->pdev)) {
5069 		/* disable receives in the hardware */
5070 		rctl = rd32(IGC_RCTL);
5071 		wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN);
5072 		/* flush and sleep below */
5073 	}
5074 	/* set trans_start so we don't get spurious watchdogs during reset */
5075 	netif_trans_update(netdev);
5076 
5077 	netif_carrier_off(netdev);
5078 	netif_tx_stop_all_queues(netdev);
5079 
5080 	if (pci_device_is_present(adapter->pdev)) {
5081 		/* disable transmits in the hardware */
5082 		tctl = rd32(IGC_TCTL);
5083 		tctl &= ~IGC_TCTL_EN;
5084 		wr32(IGC_TCTL, tctl);
5085 		/* flush both disables and wait for them to finish */
5086 		wrfl();
5087 		usleep_range(10000, 20000);
5088 
5089 		igc_irq_disable(adapter);
5090 	}
5091 
5092 	adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5093 
5094 	for (i = 0; i < adapter->num_q_vectors; i++) {
5095 		if (adapter->q_vector[i]) {
5096 			napi_synchronize(&adapter->q_vector[i]->napi);
5097 			napi_disable(&adapter->q_vector[i]->napi);
5098 		}
5099 	}
5100 
5101 	del_timer_sync(&adapter->watchdog_timer);
5102 	del_timer_sync(&adapter->phy_info_timer);
5103 
5104 	/* record the stats before reset*/
5105 	spin_lock(&adapter->stats64_lock);
5106 	igc_update_stats(adapter);
5107 	spin_unlock(&adapter->stats64_lock);
5108 
5109 	adapter->link_speed = 0;
5110 	adapter->link_duplex = 0;
5111 
5112 	if (!pci_channel_offline(adapter->pdev))
5113 		igc_reset(adapter);
5114 
5115 	/* clear VLAN promisc flag so VFTA will be updated if necessary */
5116 	adapter->flags &= ~IGC_FLAG_VLAN_PROMISC;
5117 
5118 	igc_disable_all_tx_rings_hw(adapter);
5119 	igc_clean_all_tx_rings(adapter);
5120 	igc_clean_all_rx_rings(adapter);
5121 }
5122 
5123 void igc_reinit_locked(struct igc_adapter *adapter)
5124 {
5125 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
5126 		usleep_range(1000, 2000);
5127 	igc_down(adapter);
5128 	igc_up(adapter);
5129 	clear_bit(__IGC_RESETTING, &adapter->state);
5130 }
5131 
5132 static void igc_reset_task(struct work_struct *work)
5133 {
5134 	struct igc_adapter *adapter;
5135 
5136 	adapter = container_of(work, struct igc_adapter, reset_task);
5137 
5138 	rtnl_lock();
5139 	/* If we're already down or resetting, just bail */
5140 	if (test_bit(__IGC_DOWN, &adapter->state) ||
5141 	    test_bit(__IGC_RESETTING, &adapter->state)) {
5142 		rtnl_unlock();
5143 		return;
5144 	}
5145 
5146 	igc_rings_dump(adapter);
5147 	igc_regs_dump(adapter);
5148 	netdev_err(adapter->netdev, "Reset adapter\n");
5149 	igc_reinit_locked(adapter);
5150 	rtnl_unlock();
5151 }
5152 
5153 /**
5154  * igc_change_mtu - Change the Maximum Transfer Unit
5155  * @netdev: network interface device structure
5156  * @new_mtu: new value for maximum frame size
5157  *
5158  * Returns 0 on success, negative on failure
5159  */
5160 static int igc_change_mtu(struct net_device *netdev, int new_mtu)
5161 {
5162 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5163 	struct igc_adapter *adapter = netdev_priv(netdev);
5164 
5165 	if (igc_xdp_is_enabled(adapter) && new_mtu > ETH_DATA_LEN) {
5166 		netdev_dbg(netdev, "Jumbo frames not supported with XDP");
5167 		return -EINVAL;
5168 	}
5169 
5170 	/* adjust max frame to be at least the size of a standard frame */
5171 	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5172 		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5173 
5174 	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
5175 		usleep_range(1000, 2000);
5176 
5177 	/* igc_down has a dependency on max_frame_size */
5178 	adapter->max_frame_size = max_frame;
5179 
5180 	if (netif_running(netdev))
5181 		igc_down(adapter);
5182 
5183 	netdev_dbg(netdev, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5184 	netdev->mtu = new_mtu;
5185 
5186 	if (netif_running(netdev))
5187 		igc_up(adapter);
5188 	else
5189 		igc_reset(adapter);
5190 
5191 	clear_bit(__IGC_RESETTING, &adapter->state);
5192 
5193 	return 0;
5194 }
5195 
5196 /**
5197  * igc_tx_timeout - Respond to a Tx Hang
5198  * @netdev: network interface device structure
5199  * @txqueue: queue number that timed out
5200  **/
5201 static void igc_tx_timeout(struct net_device *netdev,
5202 			   unsigned int __always_unused txqueue)
5203 {
5204 	struct igc_adapter *adapter = netdev_priv(netdev);
5205 	struct igc_hw *hw = &adapter->hw;
5206 
5207 	/* Do the reset outside of interrupt context */
5208 	adapter->tx_timeout_count++;
5209 	schedule_work(&adapter->reset_task);
5210 	wr32(IGC_EICS,
5211 	     (adapter->eims_enable_mask & ~adapter->eims_other));
5212 }
5213 
5214 /**
5215  * igc_get_stats64 - Get System Network Statistics
5216  * @netdev: network interface device structure
5217  * @stats: rtnl_link_stats64 pointer
5218  *
5219  * Returns the address of the device statistics structure.
5220  * The statistics are updated here and also from the timer callback.
5221  */
5222 static void igc_get_stats64(struct net_device *netdev,
5223 			    struct rtnl_link_stats64 *stats)
5224 {
5225 	struct igc_adapter *adapter = netdev_priv(netdev);
5226 
5227 	spin_lock(&adapter->stats64_lock);
5228 	if (!test_bit(__IGC_RESETTING, &adapter->state))
5229 		igc_update_stats(adapter);
5230 	memcpy(stats, &adapter->stats64, sizeof(*stats));
5231 	spin_unlock(&adapter->stats64_lock);
5232 }
5233 
5234 static netdev_features_t igc_fix_features(struct net_device *netdev,
5235 					  netdev_features_t features)
5236 {
5237 	/* Since there is no support for separate Rx/Tx vlan accel
5238 	 * enable/disable make sure Tx flag is always in same state as Rx.
5239 	 */
5240 	if (features & NETIF_F_HW_VLAN_CTAG_RX)
5241 		features |= NETIF_F_HW_VLAN_CTAG_TX;
5242 	else
5243 		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
5244 
5245 	return features;
5246 }
5247 
5248 static int igc_set_features(struct net_device *netdev,
5249 			    netdev_features_t features)
5250 {
5251 	netdev_features_t changed = netdev->features ^ features;
5252 	struct igc_adapter *adapter = netdev_priv(netdev);
5253 
5254 	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
5255 		igc_vlan_mode(netdev, features);
5256 
5257 	/* Add VLAN support */
5258 	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
5259 		return 0;
5260 
5261 	if (!(features & NETIF_F_NTUPLE))
5262 		igc_flush_nfc_rules(adapter);
5263 
5264 	netdev->features = features;
5265 
5266 	if (netif_running(netdev))
5267 		igc_reinit_locked(adapter);
5268 	else
5269 		igc_reset(adapter);
5270 
5271 	return 1;
5272 }
5273 
5274 static netdev_features_t
5275 igc_features_check(struct sk_buff *skb, struct net_device *dev,
5276 		   netdev_features_t features)
5277 {
5278 	unsigned int network_hdr_len, mac_hdr_len;
5279 
5280 	/* Make certain the headers can be described by a context descriptor */
5281 	mac_hdr_len = skb_network_header(skb) - skb->data;
5282 	if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN))
5283 		return features & ~(NETIF_F_HW_CSUM |
5284 				    NETIF_F_SCTP_CRC |
5285 				    NETIF_F_HW_VLAN_CTAG_TX |
5286 				    NETIF_F_TSO |
5287 				    NETIF_F_TSO6);
5288 
5289 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
5290 	if (unlikely(network_hdr_len >  IGC_MAX_NETWORK_HDR_LEN))
5291 		return features & ~(NETIF_F_HW_CSUM |
5292 				    NETIF_F_SCTP_CRC |
5293 				    NETIF_F_TSO |
5294 				    NETIF_F_TSO6);
5295 
5296 	/* We can only support IPv4 TSO in tunnels if we can mangle the
5297 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
5298 	 */
5299 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
5300 		features &= ~NETIF_F_TSO;
5301 
5302 	return features;
5303 }
5304 
5305 static void igc_tsync_interrupt(struct igc_adapter *adapter)
5306 {
5307 	u32 ack, tsauxc, sec, nsec, tsicr;
5308 	struct igc_hw *hw = &adapter->hw;
5309 	struct ptp_clock_event event;
5310 	struct timespec64 ts;
5311 
5312 	tsicr = rd32(IGC_TSICR);
5313 	ack = 0;
5314 
5315 	if (tsicr & IGC_TSICR_SYS_WRAP) {
5316 		event.type = PTP_CLOCK_PPS;
5317 		if (adapter->ptp_caps.pps)
5318 			ptp_clock_event(adapter->ptp_clock, &event);
5319 		ack |= IGC_TSICR_SYS_WRAP;
5320 	}
5321 
5322 	if (tsicr & IGC_TSICR_TXTS) {
5323 		/* retrieve hardware timestamp */
5324 		igc_ptp_tx_tstamp_event(adapter);
5325 		ack |= IGC_TSICR_TXTS;
5326 	}
5327 
5328 	if (tsicr & IGC_TSICR_TT0) {
5329 		spin_lock(&adapter->tmreg_lock);
5330 		ts = timespec64_add(adapter->perout[0].start,
5331 				    adapter->perout[0].period);
5332 		wr32(IGC_TRGTTIML0, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5333 		wr32(IGC_TRGTTIMH0, (u32)ts.tv_sec);
5334 		tsauxc = rd32(IGC_TSAUXC);
5335 		tsauxc |= IGC_TSAUXC_EN_TT0;
5336 		wr32(IGC_TSAUXC, tsauxc);
5337 		adapter->perout[0].start = ts;
5338 		spin_unlock(&adapter->tmreg_lock);
5339 		ack |= IGC_TSICR_TT0;
5340 	}
5341 
5342 	if (tsicr & IGC_TSICR_TT1) {
5343 		spin_lock(&adapter->tmreg_lock);
5344 		ts = timespec64_add(adapter->perout[1].start,
5345 				    adapter->perout[1].period);
5346 		wr32(IGC_TRGTTIML1, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5347 		wr32(IGC_TRGTTIMH1, (u32)ts.tv_sec);
5348 		tsauxc = rd32(IGC_TSAUXC);
5349 		tsauxc |= IGC_TSAUXC_EN_TT1;
5350 		wr32(IGC_TSAUXC, tsauxc);
5351 		adapter->perout[1].start = ts;
5352 		spin_unlock(&adapter->tmreg_lock);
5353 		ack |= IGC_TSICR_TT1;
5354 	}
5355 
5356 	if (tsicr & IGC_TSICR_AUTT0) {
5357 		nsec = rd32(IGC_AUXSTMPL0);
5358 		sec  = rd32(IGC_AUXSTMPH0);
5359 		event.type = PTP_CLOCK_EXTTS;
5360 		event.index = 0;
5361 		event.timestamp = sec * NSEC_PER_SEC + nsec;
5362 		ptp_clock_event(adapter->ptp_clock, &event);
5363 		ack |= IGC_TSICR_AUTT0;
5364 	}
5365 
5366 	if (tsicr & IGC_TSICR_AUTT1) {
5367 		nsec = rd32(IGC_AUXSTMPL1);
5368 		sec  = rd32(IGC_AUXSTMPH1);
5369 		event.type = PTP_CLOCK_EXTTS;
5370 		event.index = 1;
5371 		event.timestamp = sec * NSEC_PER_SEC + nsec;
5372 		ptp_clock_event(adapter->ptp_clock, &event);
5373 		ack |= IGC_TSICR_AUTT1;
5374 	}
5375 
5376 	/* acknowledge the interrupts */
5377 	wr32(IGC_TSICR, ack);
5378 }
5379 
5380 /**
5381  * igc_msix_other - msix other interrupt handler
5382  * @irq: interrupt number
5383  * @data: pointer to a q_vector
5384  */
5385 static irqreturn_t igc_msix_other(int irq, void *data)
5386 {
5387 	struct igc_adapter *adapter = data;
5388 	struct igc_hw *hw = &adapter->hw;
5389 	u32 icr = rd32(IGC_ICR);
5390 
5391 	/* reading ICR causes bit 31 of EICR to be cleared */
5392 	if (icr & IGC_ICR_DRSTA)
5393 		schedule_work(&adapter->reset_task);
5394 
5395 	if (icr & IGC_ICR_DOUTSYNC) {
5396 		/* HW is reporting DMA is out of sync */
5397 		adapter->stats.doosync++;
5398 	}
5399 
5400 	if (icr & IGC_ICR_LSC) {
5401 		hw->mac.get_link_status = true;
5402 		/* guard against interrupt when we're going down */
5403 		if (!test_bit(__IGC_DOWN, &adapter->state))
5404 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5405 	}
5406 
5407 	if (icr & IGC_ICR_TS)
5408 		igc_tsync_interrupt(adapter);
5409 
5410 	wr32(IGC_EIMS, adapter->eims_other);
5411 
5412 	return IRQ_HANDLED;
5413 }
5414 
5415 static void igc_write_itr(struct igc_q_vector *q_vector)
5416 {
5417 	u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK;
5418 
5419 	if (!q_vector->set_itr)
5420 		return;
5421 
5422 	if (!itr_val)
5423 		itr_val = IGC_ITR_VAL_MASK;
5424 
5425 	itr_val |= IGC_EITR_CNT_IGNR;
5426 
5427 	writel(itr_val, q_vector->itr_register);
5428 	q_vector->set_itr = 0;
5429 }
5430 
5431 static irqreturn_t igc_msix_ring(int irq, void *data)
5432 {
5433 	struct igc_q_vector *q_vector = data;
5434 
5435 	/* Write the ITR value calculated from the previous interrupt. */
5436 	igc_write_itr(q_vector);
5437 
5438 	napi_schedule(&q_vector->napi);
5439 
5440 	return IRQ_HANDLED;
5441 }
5442 
5443 /**
5444  * igc_request_msix - Initialize MSI-X interrupts
5445  * @adapter: Pointer to adapter structure
5446  *
5447  * igc_request_msix allocates MSI-X vectors and requests interrupts from the
5448  * kernel.
5449  */
5450 static int igc_request_msix(struct igc_adapter *adapter)
5451 {
5452 	unsigned int num_q_vectors = adapter->num_q_vectors;
5453 	int i = 0, err = 0, vector = 0, free_vector = 0;
5454 	struct net_device *netdev = adapter->netdev;
5455 
5456 	err = request_irq(adapter->msix_entries[vector].vector,
5457 			  &igc_msix_other, 0, netdev->name, adapter);
5458 	if (err)
5459 		goto err_out;
5460 
5461 	if (num_q_vectors > MAX_Q_VECTORS) {
5462 		num_q_vectors = MAX_Q_VECTORS;
5463 		dev_warn(&adapter->pdev->dev,
5464 			 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
5465 			 adapter->num_q_vectors, MAX_Q_VECTORS);
5466 	}
5467 	for (i = 0; i < num_q_vectors; i++) {
5468 		struct igc_q_vector *q_vector = adapter->q_vector[i];
5469 
5470 		vector++;
5471 
5472 		q_vector->itr_register = adapter->io_addr + IGC_EITR(vector);
5473 
5474 		if (q_vector->rx.ring && q_vector->tx.ring)
5475 			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
5476 				q_vector->rx.ring->queue_index);
5477 		else if (q_vector->tx.ring)
5478 			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
5479 				q_vector->tx.ring->queue_index);
5480 		else if (q_vector->rx.ring)
5481 			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
5482 				q_vector->rx.ring->queue_index);
5483 		else
5484 			sprintf(q_vector->name, "%s-unused", netdev->name);
5485 
5486 		err = request_irq(adapter->msix_entries[vector].vector,
5487 				  igc_msix_ring, 0, q_vector->name,
5488 				  q_vector);
5489 		if (err)
5490 			goto err_free;
5491 	}
5492 
5493 	igc_configure_msix(adapter);
5494 	return 0;
5495 
5496 err_free:
5497 	/* free already assigned IRQs */
5498 	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
5499 
5500 	vector--;
5501 	for (i = 0; i < vector; i++) {
5502 		free_irq(adapter->msix_entries[free_vector++].vector,
5503 			 adapter->q_vector[i]);
5504 	}
5505 err_out:
5506 	return err;
5507 }
5508 
5509 /**
5510  * igc_clear_interrupt_scheme - reset the device to a state of no interrupts
5511  * @adapter: Pointer to adapter structure
5512  *
5513  * This function resets the device so that it has 0 rx queues, tx queues, and
5514  * MSI-X interrupts allocated.
5515  */
5516 static void igc_clear_interrupt_scheme(struct igc_adapter *adapter)
5517 {
5518 	igc_free_q_vectors(adapter);
5519 	igc_reset_interrupt_capability(adapter);
5520 }
5521 
5522 /* Need to wait a few seconds after link up to get diagnostic information from
5523  * the phy
5524  */
5525 static void igc_update_phy_info(struct timer_list *t)
5526 {
5527 	struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5528 
5529 	igc_get_phy_info(&adapter->hw);
5530 }
5531 
5532 /**
5533  * igc_has_link - check shared code for link and determine up/down
5534  * @adapter: pointer to driver private info
5535  */
5536 bool igc_has_link(struct igc_adapter *adapter)
5537 {
5538 	struct igc_hw *hw = &adapter->hw;
5539 	bool link_active = false;
5540 
5541 	/* get_link_status is set on LSC (link status) interrupt or
5542 	 * rx sequence error interrupt.  get_link_status will stay
5543 	 * false until the igc_check_for_link establishes link
5544 	 * for copper adapters ONLY
5545 	 */
5546 	if (!hw->mac.get_link_status)
5547 		return true;
5548 	hw->mac.ops.check_for_link(hw);
5549 	link_active = !hw->mac.get_link_status;
5550 
5551 	if (hw->mac.type == igc_i225) {
5552 		if (!netif_carrier_ok(adapter->netdev)) {
5553 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5554 		} else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) {
5555 			adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE;
5556 			adapter->link_check_timeout = jiffies;
5557 		}
5558 	}
5559 
5560 	return link_active;
5561 }
5562 
5563 /**
5564  * igc_watchdog - Timer Call-back
5565  * @t: timer for the watchdog
5566  */
5567 static void igc_watchdog(struct timer_list *t)
5568 {
5569 	struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5570 	/* Do the rest outside of interrupt context */
5571 	schedule_work(&adapter->watchdog_task);
5572 }
5573 
5574 static void igc_watchdog_task(struct work_struct *work)
5575 {
5576 	struct igc_adapter *adapter = container_of(work,
5577 						   struct igc_adapter,
5578 						   watchdog_task);
5579 	struct net_device *netdev = adapter->netdev;
5580 	struct igc_hw *hw = &adapter->hw;
5581 	struct igc_phy_info *phy = &hw->phy;
5582 	u16 phy_data, retry_count = 20;
5583 	u32 link;
5584 	int i;
5585 
5586 	link = igc_has_link(adapter);
5587 
5588 	if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) {
5589 		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5590 			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5591 		else
5592 			link = false;
5593 	}
5594 
5595 	if (link) {
5596 		/* Cancel scheduled suspend requests. */
5597 		pm_runtime_resume(netdev->dev.parent);
5598 
5599 		if (!netif_carrier_ok(netdev)) {
5600 			u32 ctrl;
5601 
5602 			hw->mac.ops.get_speed_and_duplex(hw,
5603 							 &adapter->link_speed,
5604 							 &adapter->link_duplex);
5605 
5606 			ctrl = rd32(IGC_CTRL);
5607 			/* Link status message must follow this format */
5608 			netdev_info(netdev,
5609 				    "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5610 				    adapter->link_speed,
5611 				    adapter->link_duplex == FULL_DUPLEX ?
5612 				    "Full" : "Half",
5613 				    (ctrl & IGC_CTRL_TFCE) &&
5614 				    (ctrl & IGC_CTRL_RFCE) ? "RX/TX" :
5615 				    (ctrl & IGC_CTRL_RFCE) ?  "RX" :
5616 				    (ctrl & IGC_CTRL_TFCE) ?  "TX" : "None");
5617 
5618 			/* disable EEE if enabled */
5619 			if ((adapter->flags & IGC_FLAG_EEE) &&
5620 			    adapter->link_duplex == HALF_DUPLEX) {
5621 				netdev_info(netdev,
5622 					    "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex\n");
5623 				adapter->hw.dev_spec._base.eee_enable = false;
5624 				adapter->flags &= ~IGC_FLAG_EEE;
5625 			}
5626 
5627 			/* check if SmartSpeed worked */
5628 			igc_check_downshift(hw);
5629 			if (phy->speed_downgraded)
5630 				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5631 
5632 			/* adjust timeout factor according to speed/duplex */
5633 			adapter->tx_timeout_factor = 1;
5634 			switch (adapter->link_speed) {
5635 			case SPEED_10:
5636 				adapter->tx_timeout_factor = 14;
5637 				break;
5638 			case SPEED_100:
5639 			case SPEED_1000:
5640 			case SPEED_2500:
5641 				adapter->tx_timeout_factor = 1;
5642 				break;
5643 			}
5644 
5645 			/* Once the launch time has been set on the wire, there
5646 			 * is a delay before the link speed can be determined
5647 			 * based on link-up activity. Write into the register
5648 			 * as soon as we know the correct link speed.
5649 			 */
5650 			igc_tsn_adjust_txtime_offset(adapter);
5651 
5652 			if (adapter->link_speed != SPEED_1000)
5653 				goto no_wait;
5654 
5655 			/* wait for Remote receiver status OK */
5656 retry_read_status:
5657 			if (!igc_read_phy_reg(hw, PHY_1000T_STATUS,
5658 					      &phy_data)) {
5659 				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5660 				    retry_count) {
5661 					msleep(100);
5662 					retry_count--;
5663 					goto retry_read_status;
5664 				} else if (!retry_count) {
5665 					netdev_err(netdev, "exceed max 2 second\n");
5666 				}
5667 			} else {
5668 				netdev_err(netdev, "read 1000Base-T Status Reg\n");
5669 			}
5670 no_wait:
5671 			netif_carrier_on(netdev);
5672 
5673 			/* link state has changed, schedule phy info update */
5674 			if (!test_bit(__IGC_DOWN, &adapter->state))
5675 				mod_timer(&adapter->phy_info_timer,
5676 					  round_jiffies(jiffies + 2 * HZ));
5677 		}
5678 	} else {
5679 		if (netif_carrier_ok(netdev)) {
5680 			adapter->link_speed = 0;
5681 			adapter->link_duplex = 0;
5682 
5683 			/* Links status message must follow this format */
5684 			netdev_info(netdev, "NIC Link is Down\n");
5685 			netif_carrier_off(netdev);
5686 
5687 			/* link state has changed, schedule phy info update */
5688 			if (!test_bit(__IGC_DOWN, &adapter->state))
5689 				mod_timer(&adapter->phy_info_timer,
5690 					  round_jiffies(jiffies + 2 * HZ));
5691 
5692 			pm_schedule_suspend(netdev->dev.parent,
5693 					    MSEC_PER_SEC * 5);
5694 		}
5695 	}
5696 
5697 	spin_lock(&adapter->stats64_lock);
5698 	igc_update_stats(adapter);
5699 	spin_unlock(&adapter->stats64_lock);
5700 
5701 	for (i = 0; i < adapter->num_tx_queues; i++) {
5702 		struct igc_ring *tx_ring = adapter->tx_ring[i];
5703 
5704 		if (!netif_carrier_ok(netdev)) {
5705 			/* We've lost link, so the controller stops DMA,
5706 			 * but we've got queued Tx work that's never going
5707 			 * to get done, so reset controller to flush Tx.
5708 			 * (Do the reset outside of interrupt context).
5709 			 */
5710 			if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) {
5711 				adapter->tx_timeout_count++;
5712 				schedule_work(&adapter->reset_task);
5713 				/* return immediately since reset is imminent */
5714 				return;
5715 			}
5716 		}
5717 
5718 		/* Force detection of hung controller every watchdog period */
5719 		set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5720 	}
5721 
5722 	/* Cause software interrupt to ensure Rx ring is cleaned */
5723 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5724 		u32 eics = 0;
5725 
5726 		for (i = 0; i < adapter->num_q_vectors; i++)
5727 			eics |= adapter->q_vector[i]->eims_value;
5728 		wr32(IGC_EICS, eics);
5729 	} else {
5730 		wr32(IGC_ICS, IGC_ICS_RXDMT0);
5731 	}
5732 
5733 	igc_ptp_tx_hang(adapter);
5734 
5735 	/* Reset the timer */
5736 	if (!test_bit(__IGC_DOWN, &adapter->state)) {
5737 		if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)
5738 			mod_timer(&adapter->watchdog_timer,
5739 				  round_jiffies(jiffies +  HZ));
5740 		else
5741 			mod_timer(&adapter->watchdog_timer,
5742 				  round_jiffies(jiffies + 2 * HZ));
5743 	}
5744 }
5745 
5746 /**
5747  * igc_intr_msi - Interrupt Handler
5748  * @irq: interrupt number
5749  * @data: pointer to a network interface device structure
5750  */
5751 static irqreturn_t igc_intr_msi(int irq, void *data)
5752 {
5753 	struct igc_adapter *adapter = data;
5754 	struct igc_q_vector *q_vector = adapter->q_vector[0];
5755 	struct igc_hw *hw = &adapter->hw;
5756 	/* read ICR disables interrupts using IAM */
5757 	u32 icr = rd32(IGC_ICR);
5758 
5759 	igc_write_itr(q_vector);
5760 
5761 	if (icr & IGC_ICR_DRSTA)
5762 		schedule_work(&adapter->reset_task);
5763 
5764 	if (icr & IGC_ICR_DOUTSYNC) {
5765 		/* HW is reporting DMA is out of sync */
5766 		adapter->stats.doosync++;
5767 	}
5768 
5769 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5770 		hw->mac.get_link_status = true;
5771 		if (!test_bit(__IGC_DOWN, &adapter->state))
5772 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5773 	}
5774 
5775 	if (icr & IGC_ICR_TS)
5776 		igc_tsync_interrupt(adapter);
5777 
5778 	napi_schedule(&q_vector->napi);
5779 
5780 	return IRQ_HANDLED;
5781 }
5782 
5783 /**
5784  * igc_intr - Legacy Interrupt Handler
5785  * @irq: interrupt number
5786  * @data: pointer to a network interface device structure
5787  */
5788 static irqreturn_t igc_intr(int irq, void *data)
5789 {
5790 	struct igc_adapter *adapter = data;
5791 	struct igc_q_vector *q_vector = adapter->q_vector[0];
5792 	struct igc_hw *hw = &adapter->hw;
5793 	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5794 	 * need for the IMC write
5795 	 */
5796 	u32 icr = rd32(IGC_ICR);
5797 
5798 	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5799 	 * not set, then the adapter didn't send an interrupt
5800 	 */
5801 	if (!(icr & IGC_ICR_INT_ASSERTED))
5802 		return IRQ_NONE;
5803 
5804 	igc_write_itr(q_vector);
5805 
5806 	if (icr & IGC_ICR_DRSTA)
5807 		schedule_work(&adapter->reset_task);
5808 
5809 	if (icr & IGC_ICR_DOUTSYNC) {
5810 		/* HW is reporting DMA is out of sync */
5811 		adapter->stats.doosync++;
5812 	}
5813 
5814 	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5815 		hw->mac.get_link_status = true;
5816 		/* guard against interrupt when we're going down */
5817 		if (!test_bit(__IGC_DOWN, &adapter->state))
5818 			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5819 	}
5820 
5821 	if (icr & IGC_ICR_TS)
5822 		igc_tsync_interrupt(adapter);
5823 
5824 	napi_schedule(&q_vector->napi);
5825 
5826 	return IRQ_HANDLED;
5827 }
5828 
5829 static void igc_free_irq(struct igc_adapter *adapter)
5830 {
5831 	if (adapter->msix_entries) {
5832 		int vector = 0, i;
5833 
5834 		free_irq(adapter->msix_entries[vector++].vector, adapter);
5835 
5836 		for (i = 0; i < adapter->num_q_vectors; i++)
5837 			free_irq(adapter->msix_entries[vector++].vector,
5838 				 adapter->q_vector[i]);
5839 	} else {
5840 		free_irq(adapter->pdev->irq, adapter);
5841 	}
5842 }
5843 
5844 /**
5845  * igc_request_irq - initialize interrupts
5846  * @adapter: Pointer to adapter structure
5847  *
5848  * Attempts to configure interrupts using the best available
5849  * capabilities of the hardware and kernel.
5850  */
5851 static int igc_request_irq(struct igc_adapter *adapter)
5852 {
5853 	struct net_device *netdev = adapter->netdev;
5854 	struct pci_dev *pdev = adapter->pdev;
5855 	int err = 0;
5856 
5857 	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5858 		err = igc_request_msix(adapter);
5859 		if (!err)
5860 			goto request_done;
5861 		/* fall back to MSI */
5862 		igc_free_all_tx_resources(adapter);
5863 		igc_free_all_rx_resources(adapter);
5864 
5865 		igc_clear_interrupt_scheme(adapter);
5866 		err = igc_init_interrupt_scheme(adapter, false);
5867 		if (err)
5868 			goto request_done;
5869 		igc_setup_all_tx_resources(adapter);
5870 		igc_setup_all_rx_resources(adapter);
5871 		igc_configure(adapter);
5872 	}
5873 
5874 	igc_assign_vector(adapter->q_vector[0], 0);
5875 
5876 	if (adapter->flags & IGC_FLAG_HAS_MSI) {
5877 		err = request_irq(pdev->irq, &igc_intr_msi, 0,
5878 				  netdev->name, adapter);
5879 		if (!err)
5880 			goto request_done;
5881 
5882 		/* fall back to legacy interrupts */
5883 		igc_reset_interrupt_capability(adapter);
5884 		adapter->flags &= ~IGC_FLAG_HAS_MSI;
5885 	}
5886 
5887 	err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED,
5888 			  netdev->name, adapter);
5889 
5890 	if (err)
5891 		netdev_err(netdev, "Error %d getting interrupt\n", err);
5892 
5893 request_done:
5894 	return err;
5895 }
5896 
5897 /**
5898  * __igc_open - Called when a network interface is made active
5899  * @netdev: network interface device structure
5900  * @resuming: boolean indicating if the device is resuming
5901  *
5902  * Returns 0 on success, negative value on failure
5903  *
5904  * The open entry point is called when a network interface is made
5905  * active by the system (IFF_UP).  At this point all resources needed
5906  * for transmit and receive operations are allocated, the interrupt
5907  * handler is registered with the OS, the watchdog timer is started,
5908  * and the stack is notified that the interface is ready.
5909  */
5910 static int __igc_open(struct net_device *netdev, bool resuming)
5911 {
5912 	struct igc_adapter *adapter = netdev_priv(netdev);
5913 	struct pci_dev *pdev = adapter->pdev;
5914 	struct igc_hw *hw = &adapter->hw;
5915 	int err = 0;
5916 	int i = 0;
5917 
5918 	/* disallow open during test */
5919 
5920 	if (test_bit(__IGC_TESTING, &adapter->state)) {
5921 		WARN_ON(resuming);
5922 		return -EBUSY;
5923 	}
5924 
5925 	if (!resuming)
5926 		pm_runtime_get_sync(&pdev->dev);
5927 
5928 	netif_carrier_off(netdev);
5929 
5930 	/* allocate transmit descriptors */
5931 	err = igc_setup_all_tx_resources(adapter);
5932 	if (err)
5933 		goto err_setup_tx;
5934 
5935 	/* allocate receive descriptors */
5936 	err = igc_setup_all_rx_resources(adapter);
5937 	if (err)
5938 		goto err_setup_rx;
5939 
5940 	igc_power_up_link(adapter);
5941 
5942 	igc_configure(adapter);
5943 
5944 	err = igc_request_irq(adapter);
5945 	if (err)
5946 		goto err_req_irq;
5947 
5948 	/* Notify the stack of the actual queue counts. */
5949 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
5950 	if (err)
5951 		goto err_set_queues;
5952 
5953 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
5954 	if (err)
5955 		goto err_set_queues;
5956 
5957 	clear_bit(__IGC_DOWN, &adapter->state);
5958 
5959 	for (i = 0; i < adapter->num_q_vectors; i++)
5960 		napi_enable(&adapter->q_vector[i]->napi);
5961 
5962 	/* Clear any pending interrupts. */
5963 	rd32(IGC_ICR);
5964 	igc_irq_enable(adapter);
5965 
5966 	if (!resuming)
5967 		pm_runtime_put(&pdev->dev);
5968 
5969 	netif_tx_start_all_queues(netdev);
5970 
5971 	/* start the watchdog. */
5972 	hw->mac.get_link_status = true;
5973 	schedule_work(&adapter->watchdog_task);
5974 
5975 	return IGC_SUCCESS;
5976 
5977 err_set_queues:
5978 	igc_free_irq(adapter);
5979 err_req_irq:
5980 	igc_release_hw_control(adapter);
5981 	igc_power_down_phy_copper_base(&adapter->hw);
5982 	igc_free_all_rx_resources(adapter);
5983 err_setup_rx:
5984 	igc_free_all_tx_resources(adapter);
5985 err_setup_tx:
5986 	igc_reset(adapter);
5987 	if (!resuming)
5988 		pm_runtime_put(&pdev->dev);
5989 
5990 	return err;
5991 }
5992 
5993 int igc_open(struct net_device *netdev)
5994 {
5995 	return __igc_open(netdev, false);
5996 }
5997 
5998 /**
5999  * __igc_close - Disables a network interface
6000  * @netdev: network interface device structure
6001  * @suspending: boolean indicating the device is suspending
6002  *
6003  * Returns 0, this is not allowed to fail
6004  *
6005  * The close entry point is called when an interface is de-activated
6006  * by the OS.  The hardware is still under the driver's control, but
6007  * needs to be disabled.  A global MAC reset is issued to stop the
6008  * hardware, and all transmit and receive resources are freed.
6009  */
6010 static int __igc_close(struct net_device *netdev, bool suspending)
6011 {
6012 	struct igc_adapter *adapter = netdev_priv(netdev);
6013 	struct pci_dev *pdev = adapter->pdev;
6014 
6015 	WARN_ON(test_bit(__IGC_RESETTING, &adapter->state));
6016 
6017 	if (!suspending)
6018 		pm_runtime_get_sync(&pdev->dev);
6019 
6020 	igc_down(adapter);
6021 
6022 	igc_release_hw_control(adapter);
6023 
6024 	igc_free_irq(adapter);
6025 
6026 	igc_free_all_tx_resources(adapter);
6027 	igc_free_all_rx_resources(adapter);
6028 
6029 	if (!suspending)
6030 		pm_runtime_put_sync(&pdev->dev);
6031 
6032 	return 0;
6033 }
6034 
6035 int igc_close(struct net_device *netdev)
6036 {
6037 	if (netif_device_present(netdev) || netdev->dismantle)
6038 		return __igc_close(netdev, false);
6039 	return 0;
6040 }
6041 
6042 /**
6043  * igc_ioctl - Access the hwtstamp interface
6044  * @netdev: network interface device structure
6045  * @ifr: interface request data
6046  * @cmd: ioctl command
6047  **/
6048 static int igc_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6049 {
6050 	switch (cmd) {
6051 	case SIOCGHWTSTAMP:
6052 		return igc_ptp_get_ts_config(netdev, ifr);
6053 	case SIOCSHWTSTAMP:
6054 		return igc_ptp_set_ts_config(netdev, ifr);
6055 	default:
6056 		return -EOPNOTSUPP;
6057 	}
6058 }
6059 
6060 static int igc_save_launchtime_params(struct igc_adapter *adapter, int queue,
6061 				      bool enable)
6062 {
6063 	struct igc_ring *ring;
6064 
6065 	if (queue < 0 || queue >= adapter->num_tx_queues)
6066 		return -EINVAL;
6067 
6068 	ring = adapter->tx_ring[queue];
6069 	ring->launchtime_enable = enable;
6070 
6071 	return 0;
6072 }
6073 
6074 static bool is_base_time_past(ktime_t base_time, const struct timespec64 *now)
6075 {
6076 	struct timespec64 b;
6077 
6078 	b = ktime_to_timespec64(base_time);
6079 
6080 	return timespec64_compare(now, &b) > 0;
6081 }
6082 
6083 static bool validate_schedule(struct igc_adapter *adapter,
6084 			      const struct tc_taprio_qopt_offload *qopt)
6085 {
6086 	int queue_uses[IGC_MAX_TX_QUEUES] = { };
6087 	struct igc_hw *hw = &adapter->hw;
6088 	struct timespec64 now;
6089 	size_t n;
6090 
6091 	if (qopt->cycle_time_extension)
6092 		return false;
6093 
6094 	igc_ptp_read(adapter, &now);
6095 
6096 	/* If we program the controller's BASET registers with a time
6097 	 * in the future, it will hold all the packets until that
6098 	 * time, causing a lot of TX Hangs, so to avoid that, we
6099 	 * reject schedules that would start in the future.
6100 	 * Note: Limitation above is no longer in i226.
6101 	 */
6102 	if (!is_base_time_past(qopt->base_time, &now) &&
6103 	    igc_is_device_id_i225(hw))
6104 		return false;
6105 
6106 	for (n = 0; n < qopt->num_entries; n++) {
6107 		const struct tc_taprio_sched_entry *e, *prev;
6108 		int i;
6109 
6110 		prev = n ? &qopt->entries[n - 1] : NULL;
6111 		e = &qopt->entries[n];
6112 
6113 		/* i225 only supports "global" frame preemption
6114 		 * settings.
6115 		 */
6116 		if (e->command != TC_TAPRIO_CMD_SET_GATES)
6117 			return false;
6118 
6119 		for (i = 0; i < adapter->num_tx_queues; i++)
6120 			if (e->gate_mask & BIT(i)) {
6121 				queue_uses[i]++;
6122 
6123 				/* There are limitations: A single queue cannot
6124 				 * be opened and closed multiple times per cycle
6125 				 * unless the gate stays open. Check for it.
6126 				 */
6127 				if (queue_uses[i] > 1 &&
6128 				    !(prev->gate_mask & BIT(i)))
6129 					return false;
6130 			}
6131 	}
6132 
6133 	return true;
6134 }
6135 
6136 static int igc_tsn_enable_launchtime(struct igc_adapter *adapter,
6137 				     struct tc_etf_qopt_offload *qopt)
6138 {
6139 	struct igc_hw *hw = &adapter->hw;
6140 	int err;
6141 
6142 	if (hw->mac.type != igc_i225)
6143 		return -EOPNOTSUPP;
6144 
6145 	err = igc_save_launchtime_params(adapter, qopt->queue, qopt->enable);
6146 	if (err)
6147 		return err;
6148 
6149 	return igc_tsn_offload_apply(adapter);
6150 }
6151 
6152 static int igc_qbv_clear_schedule(struct igc_adapter *adapter)
6153 {
6154 	unsigned long flags;
6155 	int i;
6156 
6157 	adapter->base_time = 0;
6158 	adapter->cycle_time = NSEC_PER_SEC;
6159 	adapter->taprio_offload_enable = false;
6160 	adapter->qbv_config_change_errors = 0;
6161 	adapter->qbv_count = 0;
6162 
6163 	for (i = 0; i < adapter->num_tx_queues; i++) {
6164 		struct igc_ring *ring = adapter->tx_ring[i];
6165 
6166 		ring->start_time = 0;
6167 		ring->end_time = NSEC_PER_SEC;
6168 		ring->max_sdu = 0;
6169 	}
6170 
6171 	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6172 
6173 	adapter->qbv_transition = false;
6174 
6175 	for (i = 0; i < adapter->num_tx_queues; i++) {
6176 		struct igc_ring *ring = adapter->tx_ring[i];
6177 
6178 		ring->oper_gate_closed = false;
6179 		ring->admin_gate_closed = false;
6180 	}
6181 
6182 	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6183 
6184 	return 0;
6185 }
6186 
6187 static int igc_tsn_clear_schedule(struct igc_adapter *adapter)
6188 {
6189 	igc_qbv_clear_schedule(adapter);
6190 
6191 	return 0;
6192 }
6193 
6194 static void igc_taprio_stats(struct net_device *dev,
6195 			     struct tc_taprio_qopt_stats *stats)
6196 {
6197 	/* When Strict_End is enabled, the tx_overruns counter
6198 	 * will always be zero.
6199 	 */
6200 	stats->tx_overruns = 0;
6201 }
6202 
6203 static void igc_taprio_queue_stats(struct net_device *dev,
6204 				   struct tc_taprio_qopt_queue_stats *queue_stats)
6205 {
6206 	struct tc_taprio_qopt_stats *stats = &queue_stats->stats;
6207 
6208 	/* When Strict_End is enabled, the tx_overruns counter
6209 	 * will always be zero.
6210 	 */
6211 	stats->tx_overruns = 0;
6212 }
6213 
6214 static int igc_save_qbv_schedule(struct igc_adapter *adapter,
6215 				 struct tc_taprio_qopt_offload *qopt)
6216 {
6217 	bool queue_configured[IGC_MAX_TX_QUEUES] = { };
6218 	struct igc_hw *hw = &adapter->hw;
6219 	u32 start_time = 0, end_time = 0;
6220 	struct timespec64 now;
6221 	unsigned long flags;
6222 	size_t n;
6223 	int i;
6224 
6225 	switch (qopt->cmd) {
6226 	case TAPRIO_CMD_REPLACE:
6227 		break;
6228 	case TAPRIO_CMD_DESTROY:
6229 		return igc_tsn_clear_schedule(adapter);
6230 	case TAPRIO_CMD_STATS:
6231 		igc_taprio_stats(adapter->netdev, &qopt->stats);
6232 		return 0;
6233 	case TAPRIO_CMD_QUEUE_STATS:
6234 		igc_taprio_queue_stats(adapter->netdev, &qopt->queue_stats);
6235 		return 0;
6236 	default:
6237 		return -EOPNOTSUPP;
6238 	}
6239 
6240 	if (qopt->base_time < 0)
6241 		return -ERANGE;
6242 
6243 	if (igc_is_device_id_i225(hw) && adapter->taprio_offload_enable)
6244 		return -EALREADY;
6245 
6246 	if (!validate_schedule(adapter, qopt))
6247 		return -EINVAL;
6248 
6249 	adapter->cycle_time = qopt->cycle_time;
6250 	adapter->base_time = qopt->base_time;
6251 	adapter->taprio_offload_enable = true;
6252 
6253 	igc_ptp_read(adapter, &now);
6254 
6255 	for (n = 0; n < qopt->num_entries; n++) {
6256 		struct tc_taprio_sched_entry *e = &qopt->entries[n];
6257 
6258 		end_time += e->interval;
6259 
6260 		/* If any of the conditions below are true, we need to manually
6261 		 * control the end time of the cycle.
6262 		 * 1. Qbv users can specify a cycle time that is not equal
6263 		 * to the total GCL intervals. Hence, recalculation is
6264 		 * necessary here to exclude the time interval that
6265 		 * exceeds the cycle time.
6266 		 * 2. According to IEEE Std. 802.1Q-2018 section 8.6.9.2,
6267 		 * once the end of the list is reached, it will switch
6268 		 * to the END_OF_CYCLE state and leave the gates in the
6269 		 * same state until the next cycle is started.
6270 		 */
6271 		if (end_time > adapter->cycle_time ||
6272 		    n + 1 == qopt->num_entries)
6273 			end_time = adapter->cycle_time;
6274 
6275 		for (i = 0; i < adapter->num_tx_queues; i++) {
6276 			struct igc_ring *ring = adapter->tx_ring[i];
6277 
6278 			if (!(e->gate_mask & BIT(i)))
6279 				continue;
6280 
6281 			/* Check whether a queue stays open for more than one
6282 			 * entry. If so, keep the start and advance the end
6283 			 * time.
6284 			 */
6285 			if (!queue_configured[i])
6286 				ring->start_time = start_time;
6287 			ring->end_time = end_time;
6288 
6289 			if (ring->start_time >= adapter->cycle_time)
6290 				queue_configured[i] = false;
6291 			else
6292 				queue_configured[i] = true;
6293 		}
6294 
6295 		start_time += e->interval;
6296 	}
6297 
6298 	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6299 
6300 	/* Check whether a queue gets configured.
6301 	 * If not, set the start and end time to be end time.
6302 	 */
6303 	for (i = 0; i < adapter->num_tx_queues; i++) {
6304 		struct igc_ring *ring = adapter->tx_ring[i];
6305 
6306 		if (!is_base_time_past(qopt->base_time, &now)) {
6307 			ring->admin_gate_closed = false;
6308 		} else {
6309 			ring->oper_gate_closed = false;
6310 			ring->admin_gate_closed = false;
6311 		}
6312 
6313 		if (!queue_configured[i]) {
6314 			if (!is_base_time_past(qopt->base_time, &now))
6315 				ring->admin_gate_closed = true;
6316 			else
6317 				ring->oper_gate_closed = true;
6318 
6319 			ring->start_time = end_time;
6320 			ring->end_time = end_time;
6321 		}
6322 	}
6323 
6324 	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6325 
6326 	for (i = 0; i < adapter->num_tx_queues; i++) {
6327 		struct igc_ring *ring = adapter->tx_ring[i];
6328 		struct net_device *dev = adapter->netdev;
6329 
6330 		if (qopt->max_sdu[i])
6331 			ring->max_sdu = qopt->max_sdu[i] + dev->hard_header_len - ETH_TLEN;
6332 		else
6333 			ring->max_sdu = 0;
6334 	}
6335 
6336 	return 0;
6337 }
6338 
6339 static int igc_tsn_enable_qbv_scheduling(struct igc_adapter *adapter,
6340 					 struct tc_taprio_qopt_offload *qopt)
6341 {
6342 	struct igc_hw *hw = &adapter->hw;
6343 	int err;
6344 
6345 	if (hw->mac.type != igc_i225)
6346 		return -EOPNOTSUPP;
6347 
6348 	err = igc_save_qbv_schedule(adapter, qopt);
6349 	if (err)
6350 		return err;
6351 
6352 	return igc_tsn_offload_apply(adapter);
6353 }
6354 
6355 static int igc_save_cbs_params(struct igc_adapter *adapter, int queue,
6356 			       bool enable, int idleslope, int sendslope,
6357 			       int hicredit, int locredit)
6358 {
6359 	bool cbs_status[IGC_MAX_SR_QUEUES] = { false };
6360 	struct net_device *netdev = adapter->netdev;
6361 	struct igc_ring *ring;
6362 	int i;
6363 
6364 	/* i225 has two sets of credit-based shaper logic.
6365 	 * Supporting it only on the top two priority queues
6366 	 */
6367 	if (queue < 0 || queue > 1)
6368 		return -EINVAL;
6369 
6370 	ring = adapter->tx_ring[queue];
6371 
6372 	for (i = 0; i < IGC_MAX_SR_QUEUES; i++)
6373 		if (adapter->tx_ring[i])
6374 			cbs_status[i] = adapter->tx_ring[i]->cbs_enable;
6375 
6376 	/* CBS should be enabled on the highest priority queue first in order
6377 	 * for the CBS algorithm to operate as intended.
6378 	 */
6379 	if (enable) {
6380 		if (queue == 1 && !cbs_status[0]) {
6381 			netdev_err(netdev,
6382 				   "Enabling CBS on queue1 before queue0\n");
6383 			return -EINVAL;
6384 		}
6385 	} else {
6386 		if (queue == 0 && cbs_status[1]) {
6387 			netdev_err(netdev,
6388 				   "Disabling CBS on queue0 before queue1\n");
6389 			return -EINVAL;
6390 		}
6391 	}
6392 
6393 	ring->cbs_enable = enable;
6394 	ring->idleslope = idleslope;
6395 	ring->sendslope = sendslope;
6396 	ring->hicredit = hicredit;
6397 	ring->locredit = locredit;
6398 
6399 	return 0;
6400 }
6401 
6402 static int igc_tsn_enable_cbs(struct igc_adapter *adapter,
6403 			      struct tc_cbs_qopt_offload *qopt)
6404 {
6405 	struct igc_hw *hw = &adapter->hw;
6406 	int err;
6407 
6408 	if (hw->mac.type != igc_i225)
6409 		return -EOPNOTSUPP;
6410 
6411 	if (qopt->queue < 0 || qopt->queue > 1)
6412 		return -EINVAL;
6413 
6414 	err = igc_save_cbs_params(adapter, qopt->queue, qopt->enable,
6415 				  qopt->idleslope, qopt->sendslope,
6416 				  qopt->hicredit, qopt->locredit);
6417 	if (err)
6418 		return err;
6419 
6420 	return igc_tsn_offload_apply(adapter);
6421 }
6422 
6423 static int igc_tc_query_caps(struct igc_adapter *adapter,
6424 			     struct tc_query_caps_base *base)
6425 {
6426 	struct igc_hw *hw = &adapter->hw;
6427 
6428 	switch (base->type) {
6429 	case TC_SETUP_QDISC_TAPRIO: {
6430 		struct tc_taprio_caps *caps = base->caps;
6431 
6432 		caps->broken_mqprio = true;
6433 
6434 		if (hw->mac.type == igc_i225) {
6435 			caps->supports_queue_max_sdu = true;
6436 			caps->gate_mask_per_txq = true;
6437 		}
6438 
6439 		return 0;
6440 	}
6441 	default:
6442 		return -EOPNOTSUPP;
6443 	}
6444 }
6445 
6446 static int igc_setup_tc(struct net_device *dev, enum tc_setup_type type,
6447 			void *type_data)
6448 {
6449 	struct igc_adapter *adapter = netdev_priv(dev);
6450 
6451 	adapter->tc_setup_type = type;
6452 
6453 	switch (type) {
6454 	case TC_QUERY_CAPS:
6455 		return igc_tc_query_caps(adapter, type_data);
6456 	case TC_SETUP_QDISC_TAPRIO:
6457 		return igc_tsn_enable_qbv_scheduling(adapter, type_data);
6458 
6459 	case TC_SETUP_QDISC_ETF:
6460 		return igc_tsn_enable_launchtime(adapter, type_data);
6461 
6462 	case TC_SETUP_QDISC_CBS:
6463 		return igc_tsn_enable_cbs(adapter, type_data);
6464 
6465 	default:
6466 		return -EOPNOTSUPP;
6467 	}
6468 }
6469 
6470 static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf)
6471 {
6472 	struct igc_adapter *adapter = netdev_priv(dev);
6473 
6474 	switch (bpf->command) {
6475 	case XDP_SETUP_PROG:
6476 		return igc_xdp_set_prog(adapter, bpf->prog, bpf->extack);
6477 	case XDP_SETUP_XSK_POOL:
6478 		return igc_xdp_setup_pool(adapter, bpf->xsk.pool,
6479 					  bpf->xsk.queue_id);
6480 	default:
6481 		return -EOPNOTSUPP;
6482 	}
6483 }
6484 
6485 static int igc_xdp_xmit(struct net_device *dev, int num_frames,
6486 			struct xdp_frame **frames, u32 flags)
6487 {
6488 	struct igc_adapter *adapter = netdev_priv(dev);
6489 	int cpu = smp_processor_id();
6490 	struct netdev_queue *nq;
6491 	struct igc_ring *ring;
6492 	int i, drops;
6493 
6494 	if (unlikely(!netif_carrier_ok(dev)))
6495 		return -ENETDOWN;
6496 
6497 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
6498 		return -EINVAL;
6499 
6500 	ring = igc_xdp_get_tx_ring(adapter, cpu);
6501 	nq = txring_txq(ring);
6502 
6503 	__netif_tx_lock(nq, cpu);
6504 
6505 	/* Avoid transmit queue timeout since we share it with the slow path */
6506 	txq_trans_cond_update(nq);
6507 
6508 	drops = 0;
6509 	for (i = 0; i < num_frames; i++) {
6510 		int err;
6511 		struct xdp_frame *xdpf = frames[i];
6512 
6513 		err = igc_xdp_init_tx_descriptor(ring, xdpf);
6514 		if (err) {
6515 			xdp_return_frame_rx_napi(xdpf);
6516 			drops++;
6517 		}
6518 	}
6519 
6520 	if (flags & XDP_XMIT_FLUSH)
6521 		igc_flush_tx_descriptors(ring);
6522 
6523 	__netif_tx_unlock(nq);
6524 
6525 	return num_frames - drops;
6526 }
6527 
6528 static void igc_trigger_rxtxq_interrupt(struct igc_adapter *adapter,
6529 					struct igc_q_vector *q_vector)
6530 {
6531 	struct igc_hw *hw = &adapter->hw;
6532 	u32 eics = 0;
6533 
6534 	eics |= q_vector->eims_value;
6535 	wr32(IGC_EICS, eics);
6536 }
6537 
6538 int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
6539 {
6540 	struct igc_adapter *adapter = netdev_priv(dev);
6541 	struct igc_q_vector *q_vector;
6542 	struct igc_ring *ring;
6543 
6544 	if (test_bit(__IGC_DOWN, &adapter->state))
6545 		return -ENETDOWN;
6546 
6547 	if (!igc_xdp_is_enabled(adapter))
6548 		return -ENXIO;
6549 
6550 	if (queue_id >= adapter->num_rx_queues)
6551 		return -EINVAL;
6552 
6553 	ring = adapter->rx_ring[queue_id];
6554 
6555 	if (!ring->xsk_pool)
6556 		return -ENXIO;
6557 
6558 	q_vector = adapter->q_vector[queue_id];
6559 	if (!napi_if_scheduled_mark_missed(&q_vector->napi))
6560 		igc_trigger_rxtxq_interrupt(adapter, q_vector);
6561 
6562 	return 0;
6563 }
6564 
6565 static const struct net_device_ops igc_netdev_ops = {
6566 	.ndo_open		= igc_open,
6567 	.ndo_stop		= igc_close,
6568 	.ndo_start_xmit		= igc_xmit_frame,
6569 	.ndo_set_rx_mode	= igc_set_rx_mode,
6570 	.ndo_set_mac_address	= igc_set_mac,
6571 	.ndo_change_mtu		= igc_change_mtu,
6572 	.ndo_tx_timeout		= igc_tx_timeout,
6573 	.ndo_get_stats64	= igc_get_stats64,
6574 	.ndo_fix_features	= igc_fix_features,
6575 	.ndo_set_features	= igc_set_features,
6576 	.ndo_features_check	= igc_features_check,
6577 	.ndo_eth_ioctl		= igc_ioctl,
6578 	.ndo_setup_tc		= igc_setup_tc,
6579 	.ndo_bpf		= igc_bpf,
6580 	.ndo_xdp_xmit		= igc_xdp_xmit,
6581 	.ndo_xsk_wakeup		= igc_xsk_wakeup,
6582 };
6583 
6584 /* PCIe configuration access */
6585 void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6586 {
6587 	struct igc_adapter *adapter = hw->back;
6588 
6589 	pci_read_config_word(adapter->pdev, reg, value);
6590 }
6591 
6592 void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6593 {
6594 	struct igc_adapter *adapter = hw->back;
6595 
6596 	pci_write_config_word(adapter->pdev, reg, *value);
6597 }
6598 
6599 s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6600 {
6601 	struct igc_adapter *adapter = hw->back;
6602 
6603 	if (!pci_is_pcie(adapter->pdev))
6604 		return -IGC_ERR_CONFIG;
6605 
6606 	pcie_capability_read_word(adapter->pdev, reg, value);
6607 
6608 	return IGC_SUCCESS;
6609 }
6610 
6611 s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6612 {
6613 	struct igc_adapter *adapter = hw->back;
6614 
6615 	if (!pci_is_pcie(adapter->pdev))
6616 		return -IGC_ERR_CONFIG;
6617 
6618 	pcie_capability_write_word(adapter->pdev, reg, *value);
6619 
6620 	return IGC_SUCCESS;
6621 }
6622 
6623 u32 igc_rd32(struct igc_hw *hw, u32 reg)
6624 {
6625 	struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw);
6626 	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
6627 	u32 value = 0;
6628 
6629 	if (IGC_REMOVED(hw_addr))
6630 		return ~value;
6631 
6632 	value = readl(&hw_addr[reg]);
6633 
6634 	/* reads should not return all F's */
6635 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
6636 		struct net_device *netdev = igc->netdev;
6637 
6638 		hw->hw_addr = NULL;
6639 		netif_device_detach(netdev);
6640 		netdev_err(netdev, "PCIe link lost, device now detached\n");
6641 		WARN(pci_device_is_present(igc->pdev),
6642 		     "igc: Failed to read reg 0x%x!\n", reg);
6643 	}
6644 
6645 	return value;
6646 }
6647 
6648 /* Mapping HW RSS Type to enum xdp_rss_hash_type */
6649 static enum xdp_rss_hash_type igc_xdp_rss_type[IGC_RSS_TYPE_MAX_TABLE] = {
6650 	[IGC_RSS_TYPE_NO_HASH]		= XDP_RSS_TYPE_L2,
6651 	[IGC_RSS_TYPE_HASH_TCP_IPV4]	= XDP_RSS_TYPE_L4_IPV4_TCP,
6652 	[IGC_RSS_TYPE_HASH_IPV4]	= XDP_RSS_TYPE_L3_IPV4,
6653 	[IGC_RSS_TYPE_HASH_TCP_IPV6]	= XDP_RSS_TYPE_L4_IPV6_TCP,
6654 	[IGC_RSS_TYPE_HASH_IPV6_EX]	= XDP_RSS_TYPE_L3_IPV6_EX,
6655 	[IGC_RSS_TYPE_HASH_IPV6]	= XDP_RSS_TYPE_L3_IPV6,
6656 	[IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_TCP_EX,
6657 	[IGC_RSS_TYPE_HASH_UDP_IPV4]	= XDP_RSS_TYPE_L4_IPV4_UDP,
6658 	[IGC_RSS_TYPE_HASH_UDP_IPV6]	= XDP_RSS_TYPE_L4_IPV6_UDP,
6659 	[IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_UDP_EX,
6660 	[10] = XDP_RSS_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW  */
6661 	[11] = XDP_RSS_TYPE_NONE, /* keep array sized for SW bit-mask   */
6662 	[12] = XDP_RSS_TYPE_NONE, /* to handle future HW revisons       */
6663 	[13] = XDP_RSS_TYPE_NONE,
6664 	[14] = XDP_RSS_TYPE_NONE,
6665 	[15] = XDP_RSS_TYPE_NONE,
6666 };
6667 
6668 static int igc_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash,
6669 			   enum xdp_rss_hash_type *rss_type)
6670 {
6671 	const struct igc_xdp_buff *ctx = (void *)_ctx;
6672 
6673 	if (!(ctx->xdp.rxq->dev->features & NETIF_F_RXHASH))
6674 		return -ENODATA;
6675 
6676 	*hash = le32_to_cpu(ctx->rx_desc->wb.lower.hi_dword.rss);
6677 	*rss_type = igc_xdp_rss_type[igc_rss_type(ctx->rx_desc)];
6678 
6679 	return 0;
6680 }
6681 
6682 static int igc_xdp_rx_timestamp(const struct xdp_md *_ctx, u64 *timestamp)
6683 {
6684 	const struct igc_xdp_buff *ctx = (void *)_ctx;
6685 
6686 	if (igc_test_staterr(ctx->rx_desc, IGC_RXDADV_STAT_TSIP)) {
6687 		*timestamp = ctx->rx_ts;
6688 
6689 		return 0;
6690 	}
6691 
6692 	return -ENODATA;
6693 }
6694 
6695 static const struct xdp_metadata_ops igc_xdp_metadata_ops = {
6696 	.xmo_rx_hash			= igc_xdp_rx_hash,
6697 	.xmo_rx_timestamp		= igc_xdp_rx_timestamp,
6698 };
6699 
6700 static enum hrtimer_restart igc_qbv_scheduling_timer(struct hrtimer *timer)
6701 {
6702 	struct igc_adapter *adapter = container_of(timer, struct igc_adapter,
6703 						   hrtimer);
6704 	unsigned long flags;
6705 	unsigned int i;
6706 
6707 	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6708 
6709 	adapter->qbv_transition = true;
6710 	for (i = 0; i < adapter->num_tx_queues; i++) {
6711 		struct igc_ring *tx_ring = adapter->tx_ring[i];
6712 
6713 		if (tx_ring->admin_gate_closed) {
6714 			tx_ring->admin_gate_closed = false;
6715 			tx_ring->oper_gate_closed = true;
6716 		} else {
6717 			tx_ring->oper_gate_closed = false;
6718 		}
6719 	}
6720 	adapter->qbv_transition = false;
6721 
6722 	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6723 
6724 	return HRTIMER_NORESTART;
6725 }
6726 
6727 /**
6728  * igc_probe - Device Initialization Routine
6729  * @pdev: PCI device information struct
6730  * @ent: entry in igc_pci_tbl
6731  *
6732  * Returns 0 on success, negative on failure
6733  *
6734  * igc_probe initializes an adapter identified by a pci_dev structure.
6735  * The OS initialization, configuring the adapter private structure,
6736  * and a hardware reset occur.
6737  */
6738 static int igc_probe(struct pci_dev *pdev,
6739 		     const struct pci_device_id *ent)
6740 {
6741 	struct igc_adapter *adapter;
6742 	struct net_device *netdev;
6743 	struct igc_hw *hw;
6744 	const struct igc_info *ei = igc_info_tbl[ent->driver_data];
6745 	int err;
6746 
6747 	err = pci_enable_device_mem(pdev);
6748 	if (err)
6749 		return err;
6750 
6751 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6752 	if (err) {
6753 		dev_err(&pdev->dev,
6754 			"No usable DMA configuration, aborting\n");
6755 		goto err_dma;
6756 	}
6757 
6758 	err = pci_request_mem_regions(pdev, igc_driver_name);
6759 	if (err)
6760 		goto err_pci_reg;
6761 
6762 	err = pci_enable_ptm(pdev, NULL);
6763 	if (err < 0)
6764 		dev_info(&pdev->dev, "PCIe PTM not supported by PCIe bus/controller\n");
6765 
6766 	pci_set_master(pdev);
6767 
6768 	err = -ENOMEM;
6769 	netdev = alloc_etherdev_mq(sizeof(struct igc_adapter),
6770 				   IGC_MAX_TX_QUEUES);
6771 
6772 	if (!netdev)
6773 		goto err_alloc_etherdev;
6774 
6775 	SET_NETDEV_DEV(netdev, &pdev->dev);
6776 
6777 	pci_set_drvdata(pdev, netdev);
6778 	adapter = netdev_priv(netdev);
6779 	adapter->netdev = netdev;
6780 	adapter->pdev = pdev;
6781 	hw = &adapter->hw;
6782 	hw->back = adapter;
6783 	adapter->port_num = hw->bus.func;
6784 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6785 
6786 	err = pci_save_state(pdev);
6787 	if (err)
6788 		goto err_ioremap;
6789 
6790 	err = -EIO;
6791 	adapter->io_addr = ioremap(pci_resource_start(pdev, 0),
6792 				   pci_resource_len(pdev, 0));
6793 	if (!adapter->io_addr)
6794 		goto err_ioremap;
6795 
6796 	/* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */
6797 	hw->hw_addr = adapter->io_addr;
6798 
6799 	netdev->netdev_ops = &igc_netdev_ops;
6800 	netdev->xdp_metadata_ops = &igc_xdp_metadata_ops;
6801 	igc_ethtool_set_ops(netdev);
6802 	netdev->watchdog_timeo = 5 * HZ;
6803 
6804 	netdev->mem_start = pci_resource_start(pdev, 0);
6805 	netdev->mem_end = pci_resource_end(pdev, 0);
6806 
6807 	/* PCI config space info */
6808 	hw->vendor_id = pdev->vendor;
6809 	hw->device_id = pdev->device;
6810 	hw->revision_id = pdev->revision;
6811 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6812 	hw->subsystem_device_id = pdev->subsystem_device;
6813 
6814 	/* Copy the default MAC and PHY function pointers */
6815 	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6816 	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6817 
6818 	/* Initialize skew-specific constants */
6819 	err = ei->get_invariants(hw);
6820 	if (err)
6821 		goto err_sw_init;
6822 
6823 	/* Add supported features to the features list*/
6824 	netdev->features |= NETIF_F_SG;
6825 	netdev->features |= NETIF_F_TSO;
6826 	netdev->features |= NETIF_F_TSO6;
6827 	netdev->features |= NETIF_F_TSO_ECN;
6828 	netdev->features |= NETIF_F_RXHASH;
6829 	netdev->features |= NETIF_F_RXCSUM;
6830 	netdev->features |= NETIF_F_HW_CSUM;
6831 	netdev->features |= NETIF_F_SCTP_CRC;
6832 	netdev->features |= NETIF_F_HW_TC;
6833 
6834 #define IGC_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
6835 				  NETIF_F_GSO_GRE_CSUM | \
6836 				  NETIF_F_GSO_IPXIP4 | \
6837 				  NETIF_F_GSO_IPXIP6 | \
6838 				  NETIF_F_GSO_UDP_TUNNEL | \
6839 				  NETIF_F_GSO_UDP_TUNNEL_CSUM)
6840 
6841 	netdev->gso_partial_features = IGC_GSO_PARTIAL_FEATURES;
6842 	netdev->features |= NETIF_F_GSO_PARTIAL | IGC_GSO_PARTIAL_FEATURES;
6843 
6844 	/* setup the private structure */
6845 	err = igc_sw_init(adapter);
6846 	if (err)
6847 		goto err_sw_init;
6848 
6849 	/* copy netdev features into list of user selectable features */
6850 	netdev->hw_features |= NETIF_F_NTUPLE;
6851 	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
6852 	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
6853 	netdev->hw_features |= netdev->features;
6854 
6855 	netdev->features |= NETIF_F_HIGHDMA;
6856 
6857 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
6858 	netdev->mpls_features |= NETIF_F_HW_CSUM;
6859 	netdev->hw_enc_features |= netdev->vlan_features;
6860 
6861 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
6862 			       NETDEV_XDP_ACT_XSK_ZEROCOPY;
6863 
6864 	/* MTU range: 68 - 9216 */
6865 	netdev->min_mtu = ETH_MIN_MTU;
6866 	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
6867 
6868 	/* before reading the NVM, reset the controller to put the device in a
6869 	 * known good starting state
6870 	 */
6871 	hw->mac.ops.reset_hw(hw);
6872 
6873 	if (igc_get_flash_presence_i225(hw)) {
6874 		if (hw->nvm.ops.validate(hw) < 0) {
6875 			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6876 			err = -EIO;
6877 			goto err_eeprom;
6878 		}
6879 	}
6880 
6881 	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
6882 		/* copy the MAC address out of the NVM */
6883 		if (hw->mac.ops.read_mac_addr(hw))
6884 			dev_err(&pdev->dev, "NVM Read Error\n");
6885 	}
6886 
6887 	eth_hw_addr_set(netdev, hw->mac.addr);
6888 
6889 	if (!is_valid_ether_addr(netdev->dev_addr)) {
6890 		dev_err(&pdev->dev, "Invalid MAC Address\n");
6891 		err = -EIO;
6892 		goto err_eeprom;
6893 	}
6894 
6895 	/* configure RXPBSIZE and TXPBSIZE */
6896 	wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT);
6897 	wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT);
6898 
6899 	timer_setup(&adapter->watchdog_timer, igc_watchdog, 0);
6900 	timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0);
6901 
6902 	INIT_WORK(&adapter->reset_task, igc_reset_task);
6903 	INIT_WORK(&adapter->watchdog_task, igc_watchdog_task);
6904 
6905 	hrtimer_init(&adapter->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
6906 	adapter->hrtimer.function = &igc_qbv_scheduling_timer;
6907 
6908 	/* Initialize link properties that are user-changeable */
6909 	adapter->fc_autoneg = true;
6910 	hw->mac.autoneg = true;
6911 	hw->phy.autoneg_advertised = 0xaf;
6912 
6913 	hw->fc.requested_mode = igc_fc_default;
6914 	hw->fc.current_mode = igc_fc_default;
6915 
6916 	/* By default, support wake on port A */
6917 	adapter->flags |= IGC_FLAG_WOL_SUPPORTED;
6918 
6919 	/* initialize the wol settings based on the eeprom settings */
6920 	if (adapter->flags & IGC_FLAG_WOL_SUPPORTED)
6921 		adapter->wol |= IGC_WUFC_MAG;
6922 
6923 	device_set_wakeup_enable(&adapter->pdev->dev,
6924 				 adapter->flags & IGC_FLAG_WOL_SUPPORTED);
6925 
6926 	igc_ptp_init(adapter);
6927 
6928 	igc_tsn_clear_schedule(adapter);
6929 
6930 	/* reset the hardware with the new settings */
6931 	igc_reset(adapter);
6932 
6933 	/* let the f/w know that the h/w is now under the control of the
6934 	 * driver.
6935 	 */
6936 	igc_get_hw_control(adapter);
6937 
6938 	strncpy(netdev->name, "eth%d", IFNAMSIZ);
6939 	err = register_netdev(netdev);
6940 	if (err)
6941 		goto err_register;
6942 
6943 	 /* carrier off reporting is important to ethtool even BEFORE open */
6944 	netif_carrier_off(netdev);
6945 
6946 	/* Check if Media Autosense is enabled */
6947 	adapter->ei = *ei;
6948 
6949 	/* print pcie link status and MAC address */
6950 	pcie_print_link_status(pdev);
6951 	netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr);
6952 
6953 	dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
6954 	/* Disable EEE for internal PHY devices */
6955 	hw->dev_spec._base.eee_enable = false;
6956 	adapter->flags &= ~IGC_FLAG_EEE;
6957 	igc_set_eee_i225(hw, false, false, false);
6958 
6959 	pm_runtime_put_noidle(&pdev->dev);
6960 
6961 	return 0;
6962 
6963 err_register:
6964 	igc_release_hw_control(adapter);
6965 err_eeprom:
6966 	if (!igc_check_reset_block(hw))
6967 		igc_reset_phy(hw);
6968 err_sw_init:
6969 	igc_clear_interrupt_scheme(adapter);
6970 	iounmap(adapter->io_addr);
6971 err_ioremap:
6972 	free_netdev(netdev);
6973 err_alloc_etherdev:
6974 	pci_release_mem_regions(pdev);
6975 err_pci_reg:
6976 err_dma:
6977 	pci_disable_device(pdev);
6978 	return err;
6979 }
6980 
6981 /**
6982  * igc_remove - Device Removal Routine
6983  * @pdev: PCI device information struct
6984  *
6985  * igc_remove is called by the PCI subsystem to alert the driver
6986  * that it should release a PCI device.  This could be caused by a
6987  * Hot-Plug event, or because the driver is going to be removed from
6988  * memory.
6989  */
6990 static void igc_remove(struct pci_dev *pdev)
6991 {
6992 	struct net_device *netdev = pci_get_drvdata(pdev);
6993 	struct igc_adapter *adapter = netdev_priv(netdev);
6994 
6995 	pm_runtime_get_noresume(&pdev->dev);
6996 
6997 	igc_flush_nfc_rules(adapter);
6998 
6999 	igc_ptp_stop(adapter);
7000 
7001 	pci_disable_ptm(pdev);
7002 	pci_clear_master(pdev);
7003 
7004 	set_bit(__IGC_DOWN, &adapter->state);
7005 
7006 	del_timer_sync(&adapter->watchdog_timer);
7007 	del_timer_sync(&adapter->phy_info_timer);
7008 
7009 	cancel_work_sync(&adapter->reset_task);
7010 	cancel_work_sync(&adapter->watchdog_task);
7011 	hrtimer_cancel(&adapter->hrtimer);
7012 
7013 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7014 	 * would have already happened in close and is redundant.
7015 	 */
7016 	igc_release_hw_control(adapter);
7017 	unregister_netdev(netdev);
7018 
7019 	igc_clear_interrupt_scheme(adapter);
7020 	pci_iounmap(pdev, adapter->io_addr);
7021 	pci_release_mem_regions(pdev);
7022 
7023 	free_netdev(netdev);
7024 
7025 	pci_disable_device(pdev);
7026 }
7027 
7028 static int __igc_shutdown(struct pci_dev *pdev, bool *enable_wake,
7029 			  bool runtime)
7030 {
7031 	struct net_device *netdev = pci_get_drvdata(pdev);
7032 	struct igc_adapter *adapter = netdev_priv(netdev);
7033 	u32 wufc = runtime ? IGC_WUFC_LNKC : adapter->wol;
7034 	struct igc_hw *hw = &adapter->hw;
7035 	u32 ctrl, rctl, status;
7036 	bool wake;
7037 
7038 	rtnl_lock();
7039 	netif_device_detach(netdev);
7040 
7041 	if (netif_running(netdev))
7042 		__igc_close(netdev, true);
7043 
7044 	igc_ptp_suspend(adapter);
7045 
7046 	igc_clear_interrupt_scheme(adapter);
7047 	rtnl_unlock();
7048 
7049 	status = rd32(IGC_STATUS);
7050 	if (status & IGC_STATUS_LU)
7051 		wufc &= ~IGC_WUFC_LNKC;
7052 
7053 	if (wufc) {
7054 		igc_setup_rctl(adapter);
7055 		igc_set_rx_mode(netdev);
7056 
7057 		/* turn on all-multi mode if wake on multicast is enabled */
7058 		if (wufc & IGC_WUFC_MC) {
7059 			rctl = rd32(IGC_RCTL);
7060 			rctl |= IGC_RCTL_MPE;
7061 			wr32(IGC_RCTL, rctl);
7062 		}
7063 
7064 		ctrl = rd32(IGC_CTRL);
7065 		ctrl |= IGC_CTRL_ADVD3WUC;
7066 		wr32(IGC_CTRL, ctrl);
7067 
7068 		/* Allow time for pending master requests to run */
7069 		igc_disable_pcie_master(hw);
7070 
7071 		wr32(IGC_WUC, IGC_WUC_PME_EN);
7072 		wr32(IGC_WUFC, wufc);
7073 	} else {
7074 		wr32(IGC_WUC, 0);
7075 		wr32(IGC_WUFC, 0);
7076 	}
7077 
7078 	wake = wufc || adapter->en_mng_pt;
7079 	if (!wake)
7080 		igc_power_down_phy_copper_base(&adapter->hw);
7081 	else
7082 		igc_power_up_link(adapter);
7083 
7084 	if (enable_wake)
7085 		*enable_wake = wake;
7086 
7087 	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7088 	 * would have already happened in close and is redundant.
7089 	 */
7090 	igc_release_hw_control(adapter);
7091 
7092 	pci_disable_device(pdev);
7093 
7094 	return 0;
7095 }
7096 
7097 #ifdef CONFIG_PM
7098 static int __maybe_unused igc_runtime_suspend(struct device *dev)
7099 {
7100 	return __igc_shutdown(to_pci_dev(dev), NULL, 1);
7101 }
7102 
7103 static void igc_deliver_wake_packet(struct net_device *netdev)
7104 {
7105 	struct igc_adapter *adapter = netdev_priv(netdev);
7106 	struct igc_hw *hw = &adapter->hw;
7107 	struct sk_buff *skb;
7108 	u32 wupl;
7109 
7110 	wupl = rd32(IGC_WUPL) & IGC_WUPL_MASK;
7111 
7112 	/* WUPM stores only the first 128 bytes of the wake packet.
7113 	 * Read the packet only if we have the whole thing.
7114 	 */
7115 	if (wupl == 0 || wupl > IGC_WUPM_BYTES)
7116 		return;
7117 
7118 	skb = netdev_alloc_skb_ip_align(netdev, IGC_WUPM_BYTES);
7119 	if (!skb)
7120 		return;
7121 
7122 	skb_put(skb, wupl);
7123 
7124 	/* Ensure reads are 32-bit aligned */
7125 	wupl = roundup(wupl, 4);
7126 
7127 	memcpy_fromio(skb->data, hw->hw_addr + IGC_WUPM_REG(0), wupl);
7128 
7129 	skb->protocol = eth_type_trans(skb, netdev);
7130 	netif_rx(skb);
7131 }
7132 
7133 static int __maybe_unused igc_resume(struct device *dev)
7134 {
7135 	struct pci_dev *pdev = to_pci_dev(dev);
7136 	struct net_device *netdev = pci_get_drvdata(pdev);
7137 	struct igc_adapter *adapter = netdev_priv(netdev);
7138 	struct igc_hw *hw = &adapter->hw;
7139 	u32 err, val;
7140 
7141 	pci_set_power_state(pdev, PCI_D0);
7142 	pci_restore_state(pdev);
7143 	pci_save_state(pdev);
7144 
7145 	if (!pci_device_is_present(pdev))
7146 		return -ENODEV;
7147 	err = pci_enable_device_mem(pdev);
7148 	if (err) {
7149 		netdev_err(netdev, "Cannot enable PCI device from suspend\n");
7150 		return err;
7151 	}
7152 	pci_set_master(pdev);
7153 
7154 	pci_enable_wake(pdev, PCI_D3hot, 0);
7155 	pci_enable_wake(pdev, PCI_D3cold, 0);
7156 
7157 	if (igc_init_interrupt_scheme(adapter, true)) {
7158 		netdev_err(netdev, "Unable to allocate memory for queues\n");
7159 		return -ENOMEM;
7160 	}
7161 
7162 	igc_reset(adapter);
7163 
7164 	/* let the f/w know that the h/w is now under the control of the
7165 	 * driver.
7166 	 */
7167 	igc_get_hw_control(adapter);
7168 
7169 	val = rd32(IGC_WUS);
7170 	if (val & WAKE_PKT_WUS)
7171 		igc_deliver_wake_packet(netdev);
7172 
7173 	wr32(IGC_WUS, ~0);
7174 
7175 	rtnl_lock();
7176 	if (!err && netif_running(netdev))
7177 		err = __igc_open(netdev, true);
7178 
7179 	if (!err)
7180 		netif_device_attach(netdev);
7181 	rtnl_unlock();
7182 
7183 	return err;
7184 }
7185 
7186 static int __maybe_unused igc_runtime_resume(struct device *dev)
7187 {
7188 	return igc_resume(dev);
7189 }
7190 
7191 static int __maybe_unused igc_suspend(struct device *dev)
7192 {
7193 	return __igc_shutdown(to_pci_dev(dev), NULL, 0);
7194 }
7195 
7196 static int __maybe_unused igc_runtime_idle(struct device *dev)
7197 {
7198 	struct net_device *netdev = dev_get_drvdata(dev);
7199 	struct igc_adapter *adapter = netdev_priv(netdev);
7200 
7201 	if (!igc_has_link(adapter))
7202 		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7203 
7204 	return -EBUSY;
7205 }
7206 #endif /* CONFIG_PM */
7207 
7208 static void igc_shutdown(struct pci_dev *pdev)
7209 {
7210 	bool wake;
7211 
7212 	__igc_shutdown(pdev, &wake, 0);
7213 
7214 	if (system_state == SYSTEM_POWER_OFF) {
7215 		pci_wake_from_d3(pdev, wake);
7216 		pci_set_power_state(pdev, PCI_D3hot);
7217 	}
7218 }
7219 
7220 /**
7221  *  igc_io_error_detected - called when PCI error is detected
7222  *  @pdev: Pointer to PCI device
7223  *  @state: The current PCI connection state
7224  *
7225  *  This function is called after a PCI bus error affecting
7226  *  this device has been detected.
7227  **/
7228 static pci_ers_result_t igc_io_error_detected(struct pci_dev *pdev,
7229 					      pci_channel_state_t state)
7230 {
7231 	struct net_device *netdev = pci_get_drvdata(pdev);
7232 	struct igc_adapter *adapter = netdev_priv(netdev);
7233 
7234 	netif_device_detach(netdev);
7235 
7236 	if (state == pci_channel_io_perm_failure)
7237 		return PCI_ERS_RESULT_DISCONNECT;
7238 
7239 	if (netif_running(netdev))
7240 		igc_down(adapter);
7241 	pci_disable_device(pdev);
7242 
7243 	/* Request a slot reset. */
7244 	return PCI_ERS_RESULT_NEED_RESET;
7245 }
7246 
7247 /**
7248  *  igc_io_slot_reset - called after the PCI bus has been reset.
7249  *  @pdev: Pointer to PCI device
7250  *
7251  *  Restart the card from scratch, as if from a cold-boot. Implementation
7252  *  resembles the first-half of the igc_resume routine.
7253  **/
7254 static pci_ers_result_t igc_io_slot_reset(struct pci_dev *pdev)
7255 {
7256 	struct net_device *netdev = pci_get_drvdata(pdev);
7257 	struct igc_adapter *adapter = netdev_priv(netdev);
7258 	struct igc_hw *hw = &adapter->hw;
7259 	pci_ers_result_t result;
7260 
7261 	if (pci_enable_device_mem(pdev)) {
7262 		netdev_err(netdev, "Could not re-enable PCI device after reset\n");
7263 		result = PCI_ERS_RESULT_DISCONNECT;
7264 	} else {
7265 		pci_set_master(pdev);
7266 		pci_restore_state(pdev);
7267 		pci_save_state(pdev);
7268 
7269 		pci_enable_wake(pdev, PCI_D3hot, 0);
7270 		pci_enable_wake(pdev, PCI_D3cold, 0);
7271 
7272 		/* In case of PCI error, adapter loses its HW address
7273 		 * so we should re-assign it here.
7274 		 */
7275 		hw->hw_addr = adapter->io_addr;
7276 
7277 		igc_reset(adapter);
7278 		wr32(IGC_WUS, ~0);
7279 		result = PCI_ERS_RESULT_RECOVERED;
7280 	}
7281 
7282 	return result;
7283 }
7284 
7285 /**
7286  *  igc_io_resume - called when traffic can start to flow again.
7287  *  @pdev: Pointer to PCI device
7288  *
7289  *  This callback is called when the error recovery driver tells us that
7290  *  its OK to resume normal operation. Implementation resembles the
7291  *  second-half of the igc_resume routine.
7292  */
7293 static void igc_io_resume(struct pci_dev *pdev)
7294 {
7295 	struct net_device *netdev = pci_get_drvdata(pdev);
7296 	struct igc_adapter *adapter = netdev_priv(netdev);
7297 
7298 	rtnl_lock();
7299 	if (netif_running(netdev)) {
7300 		if (igc_open(netdev)) {
7301 			netdev_err(netdev, "igc_open failed after reset\n");
7302 			return;
7303 		}
7304 	}
7305 
7306 	netif_device_attach(netdev);
7307 
7308 	/* let the f/w know that the h/w is now under the control of the
7309 	 * driver.
7310 	 */
7311 	igc_get_hw_control(adapter);
7312 	rtnl_unlock();
7313 }
7314 
7315 static const struct pci_error_handlers igc_err_handler = {
7316 	.error_detected = igc_io_error_detected,
7317 	.slot_reset = igc_io_slot_reset,
7318 	.resume = igc_io_resume,
7319 };
7320 
7321 #ifdef CONFIG_PM
7322 static const struct dev_pm_ops igc_pm_ops = {
7323 	SET_SYSTEM_SLEEP_PM_OPS(igc_suspend, igc_resume)
7324 	SET_RUNTIME_PM_OPS(igc_runtime_suspend, igc_runtime_resume,
7325 			   igc_runtime_idle)
7326 };
7327 #endif
7328 
7329 static struct pci_driver igc_driver = {
7330 	.name     = igc_driver_name,
7331 	.id_table = igc_pci_tbl,
7332 	.probe    = igc_probe,
7333 	.remove   = igc_remove,
7334 #ifdef CONFIG_PM
7335 	.driver.pm = &igc_pm_ops,
7336 #endif
7337 	.shutdown = igc_shutdown,
7338 	.err_handler = &igc_err_handler,
7339 };
7340 
7341 /**
7342  * igc_reinit_queues - return error
7343  * @adapter: pointer to adapter structure
7344  */
7345 int igc_reinit_queues(struct igc_adapter *adapter)
7346 {
7347 	struct net_device *netdev = adapter->netdev;
7348 	int err = 0;
7349 
7350 	if (netif_running(netdev))
7351 		igc_close(netdev);
7352 
7353 	igc_reset_interrupt_capability(adapter);
7354 
7355 	if (igc_init_interrupt_scheme(adapter, true)) {
7356 		netdev_err(netdev, "Unable to allocate memory for queues\n");
7357 		return -ENOMEM;
7358 	}
7359 
7360 	if (netif_running(netdev))
7361 		err = igc_open(netdev);
7362 
7363 	return err;
7364 }
7365 
7366 /**
7367  * igc_get_hw_dev - return device
7368  * @hw: pointer to hardware structure
7369  *
7370  * used by hardware layer to print debugging information
7371  */
7372 struct net_device *igc_get_hw_dev(struct igc_hw *hw)
7373 {
7374 	struct igc_adapter *adapter = hw->back;
7375 
7376 	return adapter->netdev;
7377 }
7378 
7379 static void igc_disable_rx_ring_hw(struct igc_ring *ring)
7380 {
7381 	struct igc_hw *hw = &ring->q_vector->adapter->hw;
7382 	u8 idx = ring->reg_idx;
7383 	u32 rxdctl;
7384 
7385 	rxdctl = rd32(IGC_RXDCTL(idx));
7386 	rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE;
7387 	rxdctl |= IGC_RXDCTL_SWFLUSH;
7388 	wr32(IGC_RXDCTL(idx), rxdctl);
7389 }
7390 
7391 void igc_disable_rx_ring(struct igc_ring *ring)
7392 {
7393 	igc_disable_rx_ring_hw(ring);
7394 	igc_clean_rx_ring(ring);
7395 }
7396 
7397 void igc_enable_rx_ring(struct igc_ring *ring)
7398 {
7399 	struct igc_adapter *adapter = ring->q_vector->adapter;
7400 
7401 	igc_configure_rx_ring(adapter, ring);
7402 
7403 	if (ring->xsk_pool)
7404 		igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
7405 	else
7406 		igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
7407 }
7408 
7409 void igc_disable_tx_ring(struct igc_ring *ring)
7410 {
7411 	igc_disable_tx_ring_hw(ring);
7412 	igc_clean_tx_ring(ring);
7413 }
7414 
7415 void igc_enable_tx_ring(struct igc_ring *ring)
7416 {
7417 	struct igc_adapter *adapter = ring->q_vector->adapter;
7418 
7419 	igc_configure_tx_ring(adapter, ring);
7420 }
7421 
7422 /**
7423  * igc_init_module - Driver Registration Routine
7424  *
7425  * igc_init_module is the first routine called when the driver is
7426  * loaded. All it does is register with the PCI subsystem.
7427  */
7428 static int __init igc_init_module(void)
7429 {
7430 	int ret;
7431 
7432 	pr_info("%s\n", igc_driver_string);
7433 	pr_info("%s\n", igc_copyright);
7434 
7435 	ret = pci_register_driver(&igc_driver);
7436 	return ret;
7437 }
7438 
7439 module_init(igc_init_module);
7440 
7441 /**
7442  * igc_exit_module - Driver Exit Cleanup Routine
7443  *
7444  * igc_exit_module is called just before the driver is removed
7445  * from memory.
7446  */
7447 static void __exit igc_exit_module(void)
7448 {
7449 	pci_unregister_driver(&igc_driver);
7450 }
7451 
7452 module_exit(igc_exit_module);
7453 /* igc_main.c */
7454