1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2018 Intel Corporation. */
3 
4 /******************************************************************************
5  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
6 ******************************************************************************/
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
10 #include <linux/types.h>
11 #include <linux/bitops.h>
12 #include <linux/module.h>
13 #include <linux/pci.h>
14 #include <linux/netdevice.h>
15 #include <linux/vmalloc.h>
16 #include <linux/string.h>
17 #include <linux/in.h>
18 #include <linux/ip.h>
19 #include <linux/tcp.h>
20 #include <linux/sctp.h>
21 #include <linux/ipv6.h>
22 #include <linux/slab.h>
23 #include <net/checksum.h>
24 #include <net/ip6_checksum.h>
25 #include <linux/ethtool.h>
26 #include <linux/if.h>
27 #include <linux/if_vlan.h>
28 #include <linux/prefetch.h>
29 #include <net/mpls.h>
30 #include <linux/bpf.h>
31 #include <linux/bpf_trace.h>
32 #include <linux/atomic.h>
33 #include <net/xfrm.h>
34 
35 #include "ixgbevf.h"
36 
37 const char ixgbevf_driver_name[] = "ixgbevf";
38 static const char ixgbevf_driver_string[] =
39 	"Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
40 
41 static char ixgbevf_copyright[] =
42 	"Copyright (c) 2009 - 2018 Intel Corporation.";
43 
44 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
45 	[board_82599_vf]	= &ixgbevf_82599_vf_info,
46 	[board_82599_vf_hv]	= &ixgbevf_82599_vf_hv_info,
47 	[board_X540_vf]		= &ixgbevf_X540_vf_info,
48 	[board_X540_vf_hv]	= &ixgbevf_X540_vf_hv_info,
49 	[board_X550_vf]		= &ixgbevf_X550_vf_info,
50 	[board_X550_vf_hv]	= &ixgbevf_X550_vf_hv_info,
51 	[board_X550EM_x_vf]	= &ixgbevf_X550EM_x_vf_info,
52 	[board_X550EM_x_vf_hv]	= &ixgbevf_X550EM_x_vf_hv_info,
53 	[board_x550em_a_vf]	= &ixgbevf_x550em_a_vf_info,
54 };
55 
56 /* ixgbevf_pci_tbl - PCI Device ID Table
57  *
58  * Wildcard entries (PCI_ANY_ID) should come last
59  * Last entry must be all 0s
60  *
61  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
62  *   Class, Class Mask, private data (not used) }
63  */
64 static const struct pci_device_id ixgbevf_pci_tbl[] = {
65 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
66 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
67 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
68 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
69 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
70 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
71 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
72 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
73 	{PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
74 	/* required last entry */
75 	{0, }
76 };
77 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
78 
79 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
80 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
81 MODULE_LICENSE("GPL v2");
82 
83 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
84 static int debug = -1;
85 module_param(debug, int, 0);
86 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
87 
88 static struct workqueue_struct *ixgbevf_wq;
89 
90 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
91 {
92 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
93 	    !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
94 	    !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
95 		queue_work(ixgbevf_wq, &adapter->service_task);
96 }
97 
98 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
99 {
100 	BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
101 
102 	/* flush memory to make sure state is correct before next watchdog */
103 	smp_mb__before_atomic();
104 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
105 }
106 
107 /* forward decls */
108 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
109 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
110 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
111 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
112 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
113 				  struct ixgbevf_rx_buffer *old_buff);
114 
115 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
116 {
117 	struct ixgbevf_adapter *adapter = hw->back;
118 
119 	if (!hw->hw_addr)
120 		return;
121 	hw->hw_addr = NULL;
122 	dev_err(&adapter->pdev->dev, "Adapter removed\n");
123 	if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
124 		ixgbevf_service_event_schedule(adapter);
125 }
126 
127 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
128 {
129 	u32 value;
130 
131 	/* The following check not only optimizes a bit by not
132 	 * performing a read on the status register when the
133 	 * register just read was a status register read that
134 	 * returned IXGBE_FAILED_READ_REG. It also blocks any
135 	 * potential recursion.
136 	 */
137 	if (reg == IXGBE_VFSTATUS) {
138 		ixgbevf_remove_adapter(hw);
139 		return;
140 	}
141 	value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
142 	if (value == IXGBE_FAILED_READ_REG)
143 		ixgbevf_remove_adapter(hw);
144 }
145 
146 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
147 {
148 	u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
149 	u32 value;
150 
151 	if (IXGBE_REMOVED(reg_addr))
152 		return IXGBE_FAILED_READ_REG;
153 	value = readl(reg_addr + reg);
154 	if (unlikely(value == IXGBE_FAILED_READ_REG))
155 		ixgbevf_check_remove(hw, reg);
156 	return value;
157 }
158 
159 /**
160  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
161  * @adapter: pointer to adapter struct
162  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
163  * @queue: queue to map the corresponding interrupt to
164  * @msix_vector: the vector to map to the corresponding queue
165  **/
166 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
167 			     u8 queue, u8 msix_vector)
168 {
169 	u32 ivar, index;
170 	struct ixgbe_hw *hw = &adapter->hw;
171 
172 	if (direction == -1) {
173 		/* other causes */
174 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
175 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
176 		ivar &= ~0xFF;
177 		ivar |= msix_vector;
178 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
179 	} else {
180 		/* Tx or Rx causes */
181 		msix_vector |= IXGBE_IVAR_ALLOC_VAL;
182 		index = ((16 * (queue & 1)) + (8 * direction));
183 		ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
184 		ivar &= ~(0xFF << index);
185 		ivar |= (msix_vector << index);
186 		IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
187 	}
188 }
189 
190 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
191 {
192 	return ring->stats.packets;
193 }
194 
195 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
196 {
197 	struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
198 	struct ixgbe_hw *hw = &adapter->hw;
199 
200 	u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
201 	u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
202 
203 	if (head != tail)
204 		return (head < tail) ?
205 			tail - head : (tail + ring->count - head);
206 
207 	return 0;
208 }
209 
210 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
211 {
212 	u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
213 	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
214 	u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
215 
216 	clear_check_for_tx_hang(tx_ring);
217 
218 	/* Check for a hung queue, but be thorough. This verifies
219 	 * that a transmit has been completed since the previous
220 	 * check AND there is at least one packet pending. The
221 	 * ARMED bit is set to indicate a potential hang.
222 	 */
223 	if ((tx_done_old == tx_done) && tx_pending) {
224 		/* make sure it is true for two checks in a row */
225 		return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
226 					&tx_ring->state);
227 	}
228 	/* reset the countdown */
229 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
230 
231 	/* update completed stats and continue */
232 	tx_ring->tx_stats.tx_done_old = tx_done;
233 
234 	return false;
235 }
236 
237 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
238 {
239 	/* Do the reset outside of interrupt context */
240 	if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
241 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
242 		ixgbevf_service_event_schedule(adapter);
243 	}
244 }
245 
246 /**
247  * ixgbevf_tx_timeout - Respond to a Tx Hang
248  * @netdev: network interface device structure
249  * @txqueue: transmit queue hanging (unused)
250  **/
251 static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
252 {
253 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
254 
255 	ixgbevf_tx_timeout_reset(adapter);
256 }
257 
258 /**
259  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
260  * @q_vector: board private structure
261  * @tx_ring: tx ring to clean
262  * @napi_budget: Used to determine if we are in netpoll
263  **/
264 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
265 				 struct ixgbevf_ring *tx_ring, int napi_budget)
266 {
267 	struct ixgbevf_adapter *adapter = q_vector->adapter;
268 	struct ixgbevf_tx_buffer *tx_buffer;
269 	union ixgbe_adv_tx_desc *tx_desc;
270 	unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
271 	unsigned int budget = tx_ring->count / 2;
272 	unsigned int i = tx_ring->next_to_clean;
273 
274 	if (test_bit(__IXGBEVF_DOWN, &adapter->state))
275 		return true;
276 
277 	tx_buffer = &tx_ring->tx_buffer_info[i];
278 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
279 	i -= tx_ring->count;
280 
281 	do {
282 		union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
283 
284 		/* if next_to_watch is not set then there is no work pending */
285 		if (!eop_desc)
286 			break;
287 
288 		/* prevent any other reads prior to eop_desc */
289 		smp_rmb();
290 
291 		/* if DD is not set pending work has not been completed */
292 		if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
293 			break;
294 
295 		/* clear next_to_watch to prevent false hangs */
296 		tx_buffer->next_to_watch = NULL;
297 
298 		/* update the statistics for this packet */
299 		total_bytes += tx_buffer->bytecount;
300 		total_packets += tx_buffer->gso_segs;
301 		if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
302 			total_ipsec++;
303 
304 		/* free the skb */
305 		if (ring_is_xdp(tx_ring))
306 			page_frag_free(tx_buffer->data);
307 		else
308 			napi_consume_skb(tx_buffer->skb, napi_budget);
309 
310 		/* unmap skb header data */
311 		dma_unmap_single(tx_ring->dev,
312 				 dma_unmap_addr(tx_buffer, dma),
313 				 dma_unmap_len(tx_buffer, len),
314 				 DMA_TO_DEVICE);
315 
316 		/* clear tx_buffer data */
317 		dma_unmap_len_set(tx_buffer, len, 0);
318 
319 		/* unmap remaining buffers */
320 		while (tx_desc != eop_desc) {
321 			tx_buffer++;
322 			tx_desc++;
323 			i++;
324 			if (unlikely(!i)) {
325 				i -= tx_ring->count;
326 				tx_buffer = tx_ring->tx_buffer_info;
327 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
328 			}
329 
330 			/* unmap any remaining paged data */
331 			if (dma_unmap_len(tx_buffer, len)) {
332 				dma_unmap_page(tx_ring->dev,
333 					       dma_unmap_addr(tx_buffer, dma),
334 					       dma_unmap_len(tx_buffer, len),
335 					       DMA_TO_DEVICE);
336 				dma_unmap_len_set(tx_buffer, len, 0);
337 			}
338 		}
339 
340 		/* move us one more past the eop_desc for start of next pkt */
341 		tx_buffer++;
342 		tx_desc++;
343 		i++;
344 		if (unlikely(!i)) {
345 			i -= tx_ring->count;
346 			tx_buffer = tx_ring->tx_buffer_info;
347 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
348 		}
349 
350 		/* issue prefetch for next Tx descriptor */
351 		prefetch(tx_desc);
352 
353 		/* update budget accounting */
354 		budget--;
355 	} while (likely(budget));
356 
357 	i += tx_ring->count;
358 	tx_ring->next_to_clean = i;
359 	u64_stats_update_begin(&tx_ring->syncp);
360 	tx_ring->stats.bytes += total_bytes;
361 	tx_ring->stats.packets += total_packets;
362 	u64_stats_update_end(&tx_ring->syncp);
363 	q_vector->tx.total_bytes += total_bytes;
364 	q_vector->tx.total_packets += total_packets;
365 	adapter->tx_ipsec += total_ipsec;
366 
367 	if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
368 		struct ixgbe_hw *hw = &adapter->hw;
369 		union ixgbe_adv_tx_desc *eop_desc;
370 
371 		eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
372 
373 		pr_err("Detected Tx Unit Hang%s\n"
374 		       "  Tx Queue             <%d>\n"
375 		       "  TDH, TDT             <%x>, <%x>\n"
376 		       "  next_to_use          <%x>\n"
377 		       "  next_to_clean        <%x>\n"
378 		       "tx_buffer_info[next_to_clean]\n"
379 		       "  next_to_watch        <%p>\n"
380 		       "  eop_desc->wb.status  <%x>\n"
381 		       "  time_stamp           <%lx>\n"
382 		       "  jiffies              <%lx>\n",
383 		       ring_is_xdp(tx_ring) ? " XDP" : "",
384 		       tx_ring->queue_index,
385 		       IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
386 		       IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
387 		       tx_ring->next_to_use, i,
388 		       eop_desc, (eop_desc ? eop_desc->wb.status : 0),
389 		       tx_ring->tx_buffer_info[i].time_stamp, jiffies);
390 
391 		if (!ring_is_xdp(tx_ring))
392 			netif_stop_subqueue(tx_ring->netdev,
393 					    tx_ring->queue_index);
394 
395 		/* schedule immediate reset if we believe we hung */
396 		ixgbevf_tx_timeout_reset(adapter);
397 
398 		return true;
399 	}
400 
401 	if (ring_is_xdp(tx_ring))
402 		return !!budget;
403 
404 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
405 	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
406 		     (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
407 		/* Make sure that anybody stopping the queue after this
408 		 * sees the new next_to_clean.
409 		 */
410 		smp_mb();
411 
412 		if (__netif_subqueue_stopped(tx_ring->netdev,
413 					     tx_ring->queue_index) &&
414 		    !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
415 			netif_wake_subqueue(tx_ring->netdev,
416 					    tx_ring->queue_index);
417 			++tx_ring->tx_stats.restart_queue;
418 		}
419 	}
420 
421 	return !!budget;
422 }
423 
424 /**
425  * ixgbevf_rx_skb - Helper function to determine proper Rx method
426  * @q_vector: structure containing interrupt and ring information
427  * @skb: packet to send up
428  **/
429 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
430 			   struct sk_buff *skb)
431 {
432 	napi_gro_receive(&q_vector->napi, skb);
433 }
434 
435 #define IXGBE_RSS_L4_TYPES_MASK \
436 	((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
437 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
438 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
439 	 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
440 
441 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
442 				   union ixgbe_adv_rx_desc *rx_desc,
443 				   struct sk_buff *skb)
444 {
445 	u16 rss_type;
446 
447 	if (!(ring->netdev->features & NETIF_F_RXHASH))
448 		return;
449 
450 	rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
451 		   IXGBE_RXDADV_RSSTYPE_MASK;
452 
453 	if (!rss_type)
454 		return;
455 
456 	skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
457 		     (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
458 		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
459 }
460 
461 /**
462  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
463  * @ring: structure containig ring specific data
464  * @rx_desc: current Rx descriptor being processed
465  * @skb: skb currently being received and modified
466  **/
467 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
468 				       union ixgbe_adv_rx_desc *rx_desc,
469 				       struct sk_buff *skb)
470 {
471 	skb_checksum_none_assert(skb);
472 
473 	/* Rx csum disabled */
474 	if (!(ring->netdev->features & NETIF_F_RXCSUM))
475 		return;
476 
477 	/* if IP and error */
478 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
479 	    ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
480 		ring->rx_stats.csum_err++;
481 		return;
482 	}
483 
484 	if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
485 		return;
486 
487 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
488 		ring->rx_stats.csum_err++;
489 		return;
490 	}
491 
492 	/* It must be a TCP or UDP packet with a valid checksum */
493 	skb->ip_summed = CHECKSUM_UNNECESSARY;
494 }
495 
496 /**
497  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
498  * @rx_ring: rx descriptor ring packet is being transacted on
499  * @rx_desc: pointer to the EOP Rx descriptor
500  * @skb: pointer to current skb being populated
501  *
502  * This function checks the ring, descriptor, and packet information in
503  * order to populate the checksum, VLAN, protocol, and other fields within
504  * the skb.
505  **/
506 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
507 				       union ixgbe_adv_rx_desc *rx_desc,
508 				       struct sk_buff *skb)
509 {
510 	ixgbevf_rx_hash(rx_ring, rx_desc, skb);
511 	ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
512 
513 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
514 		u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
515 		unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
516 
517 		if (test_bit(vid & VLAN_VID_MASK, active_vlans))
518 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
519 	}
520 
521 	if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
522 		ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
523 
524 	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
525 }
526 
527 static
528 struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
529 						const unsigned int size)
530 {
531 	struct ixgbevf_rx_buffer *rx_buffer;
532 
533 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
534 	prefetchw(rx_buffer->page);
535 
536 	/* we are reusing so sync this buffer for CPU use */
537 	dma_sync_single_range_for_cpu(rx_ring->dev,
538 				      rx_buffer->dma,
539 				      rx_buffer->page_offset,
540 				      size,
541 				      DMA_FROM_DEVICE);
542 
543 	rx_buffer->pagecnt_bias--;
544 
545 	return rx_buffer;
546 }
547 
548 static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
549 				  struct ixgbevf_rx_buffer *rx_buffer,
550 				  struct sk_buff *skb)
551 {
552 	if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
553 		/* hand second half of page back to the ring */
554 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
555 	} else {
556 		if (IS_ERR(skb))
557 			/* We are not reusing the buffer so unmap it and free
558 			 * any references we are holding to it
559 			 */
560 			dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
561 					     ixgbevf_rx_pg_size(rx_ring),
562 					     DMA_FROM_DEVICE,
563 					     IXGBEVF_RX_DMA_ATTR);
564 		__page_frag_cache_drain(rx_buffer->page,
565 					rx_buffer->pagecnt_bias);
566 	}
567 
568 	/* clear contents of rx_buffer */
569 	rx_buffer->page = NULL;
570 }
571 
572 /**
573  * ixgbevf_is_non_eop - process handling of non-EOP buffers
574  * @rx_ring: Rx ring being processed
575  * @rx_desc: Rx descriptor for current buffer
576  *
577  * This function updates next to clean.  If the buffer is an EOP buffer
578  * this function exits returning false, otherwise it will place the
579  * sk_buff in the next buffer to be chained and return true indicating
580  * that this is in fact a non-EOP buffer.
581  **/
582 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
583 			       union ixgbe_adv_rx_desc *rx_desc)
584 {
585 	u32 ntc = rx_ring->next_to_clean + 1;
586 
587 	/* fetch, update, and store next to clean */
588 	ntc = (ntc < rx_ring->count) ? ntc : 0;
589 	rx_ring->next_to_clean = ntc;
590 
591 	prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
592 
593 	if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
594 		return false;
595 
596 	return true;
597 }
598 
599 static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
600 {
601 	return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
602 }
603 
604 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
605 				      struct ixgbevf_rx_buffer *bi)
606 {
607 	struct page *page = bi->page;
608 	dma_addr_t dma;
609 
610 	/* since we are recycling buffers we should seldom need to alloc */
611 	if (likely(page))
612 		return true;
613 
614 	/* alloc new page for storage */
615 	page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
616 	if (unlikely(!page)) {
617 		rx_ring->rx_stats.alloc_rx_page_failed++;
618 		return false;
619 	}
620 
621 	/* map page for use */
622 	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
623 				 ixgbevf_rx_pg_size(rx_ring),
624 				 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
625 
626 	/* if mapping failed free memory back to system since
627 	 * there isn't much point in holding memory we can't use
628 	 */
629 	if (dma_mapping_error(rx_ring->dev, dma)) {
630 		__free_pages(page, ixgbevf_rx_pg_order(rx_ring));
631 
632 		rx_ring->rx_stats.alloc_rx_page_failed++;
633 		return false;
634 	}
635 
636 	bi->dma = dma;
637 	bi->page = page;
638 	bi->page_offset = ixgbevf_rx_offset(rx_ring);
639 	bi->pagecnt_bias = 1;
640 	rx_ring->rx_stats.alloc_rx_page++;
641 
642 	return true;
643 }
644 
645 /**
646  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
647  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
648  * @cleaned_count: number of buffers to replace
649  **/
650 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
651 				     u16 cleaned_count)
652 {
653 	union ixgbe_adv_rx_desc *rx_desc;
654 	struct ixgbevf_rx_buffer *bi;
655 	unsigned int i = rx_ring->next_to_use;
656 
657 	/* nothing to do or no valid netdev defined */
658 	if (!cleaned_count || !rx_ring->netdev)
659 		return;
660 
661 	rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
662 	bi = &rx_ring->rx_buffer_info[i];
663 	i -= rx_ring->count;
664 
665 	do {
666 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
667 			break;
668 
669 		/* sync the buffer for use by the device */
670 		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
671 						 bi->page_offset,
672 						 ixgbevf_rx_bufsz(rx_ring),
673 						 DMA_FROM_DEVICE);
674 
675 		/* Refresh the desc even if pkt_addr didn't change
676 		 * because each write-back erases this info.
677 		 */
678 		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
679 
680 		rx_desc++;
681 		bi++;
682 		i++;
683 		if (unlikely(!i)) {
684 			rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
685 			bi = rx_ring->rx_buffer_info;
686 			i -= rx_ring->count;
687 		}
688 
689 		/* clear the length for the next_to_use descriptor */
690 		rx_desc->wb.upper.length = 0;
691 
692 		cleaned_count--;
693 	} while (cleaned_count);
694 
695 	i += rx_ring->count;
696 
697 	if (rx_ring->next_to_use != i) {
698 		/* record the next descriptor to use */
699 		rx_ring->next_to_use = i;
700 
701 		/* update next to alloc since we have filled the ring */
702 		rx_ring->next_to_alloc = i;
703 
704 		/* Force memory writes to complete before letting h/w
705 		 * know there are new descriptors to fetch.  (Only
706 		 * applicable for weak-ordered memory model archs,
707 		 * such as IA-64).
708 		 */
709 		wmb();
710 		ixgbevf_write_tail(rx_ring, i);
711 	}
712 }
713 
714 /**
715  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
716  * @rx_ring: rx descriptor ring packet is being transacted on
717  * @rx_desc: pointer to the EOP Rx descriptor
718  * @skb: pointer to current skb being fixed
719  *
720  * Check for corrupted packet headers caused by senders on the local L2
721  * embedded NIC switch not setting up their Tx Descriptors right.  These
722  * should be very rare.
723  *
724  * Also address the case where we are pulling data in on pages only
725  * and as such no data is present in the skb header.
726  *
727  * In addition if skb is not at least 60 bytes we need to pad it so that
728  * it is large enough to qualify as a valid Ethernet frame.
729  *
730  * Returns true if an error was encountered and skb was freed.
731  **/
732 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
733 				    union ixgbe_adv_rx_desc *rx_desc,
734 				    struct sk_buff *skb)
735 {
736 	/* XDP packets use error pointer so abort at this point */
737 	if (IS_ERR(skb))
738 		return true;
739 
740 	/* verify that the packet does not have any known errors */
741 	if (unlikely(ixgbevf_test_staterr(rx_desc,
742 					  IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
743 		struct net_device *netdev = rx_ring->netdev;
744 
745 		if (!(netdev->features & NETIF_F_RXALL)) {
746 			dev_kfree_skb_any(skb);
747 			return true;
748 		}
749 	}
750 
751 	/* if eth_skb_pad returns an error the skb was freed */
752 	if (eth_skb_pad(skb))
753 		return true;
754 
755 	return false;
756 }
757 
758 /**
759  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
760  * @rx_ring: rx descriptor ring to store buffers on
761  * @old_buff: donor buffer to have page reused
762  *
763  * Synchronizes page for reuse by the adapter
764  **/
765 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
766 				  struct ixgbevf_rx_buffer *old_buff)
767 {
768 	struct ixgbevf_rx_buffer *new_buff;
769 	u16 nta = rx_ring->next_to_alloc;
770 
771 	new_buff = &rx_ring->rx_buffer_info[nta];
772 
773 	/* update, and store next to alloc */
774 	nta++;
775 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
776 
777 	/* transfer page from old buffer to new buffer */
778 	new_buff->page = old_buff->page;
779 	new_buff->dma = old_buff->dma;
780 	new_buff->page_offset = old_buff->page_offset;
781 	new_buff->pagecnt_bias = old_buff->pagecnt_bias;
782 }
783 
784 static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
785 {
786 	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
787 	struct page *page = rx_buffer->page;
788 
789 	/* avoid re-using remote and pfmemalloc pages */
790 	if (!dev_page_is_reusable(page))
791 		return false;
792 
793 #if (PAGE_SIZE < 8192)
794 	/* if we are only owner of page we can reuse it */
795 	if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
796 		return false;
797 #else
798 #define IXGBEVF_LAST_OFFSET \
799 	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
800 
801 	if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
802 		return false;
803 
804 #endif
805 
806 	/* If we have drained the page fragment pool we need to update
807 	 * the pagecnt_bias and page count so that we fully restock the
808 	 * number of references the driver holds.
809 	 */
810 	if (unlikely(!pagecnt_bias)) {
811 		page_ref_add(page, USHRT_MAX);
812 		rx_buffer->pagecnt_bias = USHRT_MAX;
813 	}
814 
815 	return true;
816 }
817 
818 /**
819  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
820  * @rx_ring: rx descriptor ring to transact packets on
821  * @rx_buffer: buffer containing page to add
822  * @skb: sk_buff to place the data into
823  * @size: size of buffer to be added
824  *
825  * This function will add the data contained in rx_buffer->page to the skb.
826  **/
827 static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
828 				struct ixgbevf_rx_buffer *rx_buffer,
829 				struct sk_buff *skb,
830 				unsigned int size)
831 {
832 #if (PAGE_SIZE < 8192)
833 	unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
834 #else
835 	unsigned int truesize = ring_uses_build_skb(rx_ring) ?
836 				SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
837 				SKB_DATA_ALIGN(size);
838 #endif
839 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
840 			rx_buffer->page_offset, size, truesize);
841 #if (PAGE_SIZE < 8192)
842 	rx_buffer->page_offset ^= truesize;
843 #else
844 	rx_buffer->page_offset += truesize;
845 #endif
846 }
847 
848 static
849 struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
850 				      struct ixgbevf_rx_buffer *rx_buffer,
851 				      struct xdp_buff *xdp,
852 				      union ixgbe_adv_rx_desc *rx_desc)
853 {
854 	unsigned int size = xdp->data_end - xdp->data;
855 #if (PAGE_SIZE < 8192)
856 	unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
857 #else
858 	unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
859 					       xdp->data_hard_start);
860 #endif
861 	unsigned int headlen;
862 	struct sk_buff *skb;
863 
864 	/* prefetch first cache line of first page */
865 	net_prefetch(xdp->data);
866 
867 	/* Note, we get here by enabling legacy-rx via:
868 	 *
869 	 *    ethtool --set-priv-flags <dev> legacy-rx on
870 	 *
871 	 * In this mode, we currently get 0 extra XDP headroom as
872 	 * opposed to having legacy-rx off, where we process XDP
873 	 * packets going to stack via ixgbevf_build_skb().
874 	 *
875 	 * For ixgbevf_construct_skb() mode it means that the
876 	 * xdp->data_meta will always point to xdp->data, since
877 	 * the helper cannot expand the head. Should this ever
878 	 * changed in future for legacy-rx mode on, then lets also
879 	 * add xdp->data_meta handling here.
880 	 */
881 
882 	/* allocate a skb to store the frags */
883 	skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
884 	if (unlikely(!skb))
885 		return NULL;
886 
887 	/* Determine available headroom for copy */
888 	headlen = size;
889 	if (headlen > IXGBEVF_RX_HDR_SIZE)
890 		headlen = eth_get_headlen(skb->dev, xdp->data,
891 					  IXGBEVF_RX_HDR_SIZE);
892 
893 	/* align pull length to size of long to optimize memcpy performance */
894 	memcpy(__skb_put(skb, headlen), xdp->data,
895 	       ALIGN(headlen, sizeof(long)));
896 
897 	/* update all of the pointers */
898 	size -= headlen;
899 	if (size) {
900 		skb_add_rx_frag(skb, 0, rx_buffer->page,
901 				(xdp->data + headlen) -
902 					page_address(rx_buffer->page),
903 				size, truesize);
904 #if (PAGE_SIZE < 8192)
905 		rx_buffer->page_offset ^= truesize;
906 #else
907 		rx_buffer->page_offset += truesize;
908 #endif
909 	} else {
910 		rx_buffer->pagecnt_bias++;
911 	}
912 
913 	return skb;
914 }
915 
916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
917 					     u32 qmask)
918 {
919 	struct ixgbe_hw *hw = &adapter->hw;
920 
921 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
922 }
923 
924 static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
925 					 struct ixgbevf_rx_buffer *rx_buffer,
926 					 struct xdp_buff *xdp,
927 					 union ixgbe_adv_rx_desc *rx_desc)
928 {
929 	unsigned int metasize = xdp->data - xdp->data_meta;
930 #if (PAGE_SIZE < 8192)
931 	unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
932 #else
933 	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
934 				SKB_DATA_ALIGN(xdp->data_end -
935 					       xdp->data_hard_start);
936 #endif
937 	struct sk_buff *skb;
938 
939 	/* Prefetch first cache line of first page. If xdp->data_meta
940 	 * is unused, this points to xdp->data, otherwise, we likely
941 	 * have a consumer accessing first few bytes of meta data,
942 	 * and then actual data.
943 	 */
944 	net_prefetch(xdp->data_meta);
945 
946 	/* build an skb around the page buffer */
947 	skb = build_skb(xdp->data_hard_start, truesize);
948 	if (unlikely(!skb))
949 		return NULL;
950 
951 	/* update pointers within the skb to store the data */
952 	skb_reserve(skb, xdp->data - xdp->data_hard_start);
953 	__skb_put(skb, xdp->data_end - xdp->data);
954 	if (metasize)
955 		skb_metadata_set(skb, metasize);
956 
957 	/* update buffer offset */
958 #if (PAGE_SIZE < 8192)
959 	rx_buffer->page_offset ^= truesize;
960 #else
961 	rx_buffer->page_offset += truesize;
962 #endif
963 
964 	return skb;
965 }
966 
967 #define IXGBEVF_XDP_PASS 0
968 #define IXGBEVF_XDP_CONSUMED 1
969 #define IXGBEVF_XDP_TX 2
970 
971 static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
972 				 struct xdp_buff *xdp)
973 {
974 	struct ixgbevf_tx_buffer *tx_buffer;
975 	union ixgbe_adv_tx_desc *tx_desc;
976 	u32 len, cmd_type;
977 	dma_addr_t dma;
978 	u16 i;
979 
980 	len = xdp->data_end - xdp->data;
981 
982 	if (unlikely(!ixgbevf_desc_unused(ring)))
983 		return IXGBEVF_XDP_CONSUMED;
984 
985 	dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
986 	if (dma_mapping_error(ring->dev, dma))
987 		return IXGBEVF_XDP_CONSUMED;
988 
989 	/* record the location of the first descriptor for this packet */
990 	i = ring->next_to_use;
991 	tx_buffer = &ring->tx_buffer_info[i];
992 
993 	dma_unmap_len_set(tx_buffer, len, len);
994 	dma_unmap_addr_set(tx_buffer, dma, dma);
995 	tx_buffer->data = xdp->data;
996 	tx_buffer->bytecount = len;
997 	tx_buffer->gso_segs = 1;
998 	tx_buffer->protocol = 0;
999 
1000 	/* Populate minimal context descriptor that will provide for the
1001 	 * fact that we are expected to process Ethernet frames.
1002 	 */
1003 	if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1004 		struct ixgbe_adv_tx_context_desc *context_desc;
1005 
1006 		set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1007 
1008 		context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1009 		context_desc->vlan_macip_lens	=
1010 			cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1011 		context_desc->fceof_saidx	= 0;
1012 		context_desc->type_tucmd_mlhl	=
1013 			cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1014 				    IXGBE_ADVTXD_DTYP_CTXT);
1015 		context_desc->mss_l4len_idx	= 0;
1016 
1017 		i = 1;
1018 	}
1019 
1020 	/* put descriptor type bits */
1021 	cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1022 		   IXGBE_ADVTXD_DCMD_DEXT |
1023 		   IXGBE_ADVTXD_DCMD_IFCS;
1024 	cmd_type |= len | IXGBE_TXD_CMD;
1025 
1026 	tx_desc = IXGBEVF_TX_DESC(ring, i);
1027 	tx_desc->read.buffer_addr = cpu_to_le64(dma);
1028 
1029 	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1030 	tx_desc->read.olinfo_status =
1031 			cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1032 				    IXGBE_ADVTXD_CC);
1033 
1034 	/* Avoid any potential race with cleanup */
1035 	smp_wmb();
1036 
1037 	/* set next_to_watch value indicating a packet is present */
1038 	i++;
1039 	if (i == ring->count)
1040 		i = 0;
1041 
1042 	tx_buffer->next_to_watch = tx_desc;
1043 	ring->next_to_use = i;
1044 
1045 	return IXGBEVF_XDP_TX;
1046 }
1047 
1048 static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1049 				       struct ixgbevf_ring  *rx_ring,
1050 				       struct xdp_buff *xdp)
1051 {
1052 	int result = IXGBEVF_XDP_PASS;
1053 	struct ixgbevf_ring *xdp_ring;
1054 	struct bpf_prog *xdp_prog;
1055 	u32 act;
1056 
1057 	rcu_read_lock();
1058 	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1059 
1060 	if (!xdp_prog)
1061 		goto xdp_out;
1062 
1063 	act = bpf_prog_run_xdp(xdp_prog, xdp);
1064 	switch (act) {
1065 	case XDP_PASS:
1066 		break;
1067 	case XDP_TX:
1068 		xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1069 		result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1070 		break;
1071 	default:
1072 		bpf_warn_invalid_xdp_action(act);
1073 		fallthrough;
1074 	case XDP_ABORTED:
1075 		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1076 		fallthrough; /* handle aborts by dropping packet */
1077 	case XDP_DROP:
1078 		result = IXGBEVF_XDP_CONSUMED;
1079 		break;
1080 	}
1081 xdp_out:
1082 	rcu_read_unlock();
1083 	return ERR_PTR(-result);
1084 }
1085 
1086 static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1087 					      unsigned int size)
1088 {
1089 	unsigned int truesize;
1090 
1091 #if (PAGE_SIZE < 8192)
1092 	truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1093 #else
1094 	truesize = ring_uses_build_skb(rx_ring) ?
1095 		SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1096 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1097 		SKB_DATA_ALIGN(size);
1098 #endif
1099 	return truesize;
1100 }
1101 
1102 static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1103 				   struct ixgbevf_rx_buffer *rx_buffer,
1104 				   unsigned int size)
1105 {
1106 	unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1107 
1108 #if (PAGE_SIZE < 8192)
1109 	rx_buffer->page_offset ^= truesize;
1110 #else
1111 	rx_buffer->page_offset += truesize;
1112 #endif
1113 }
1114 
1115 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1116 				struct ixgbevf_ring *rx_ring,
1117 				int budget)
1118 {
1119 	unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1120 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1121 	u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1122 	struct sk_buff *skb = rx_ring->skb;
1123 	bool xdp_xmit = false;
1124 	struct xdp_buff xdp;
1125 
1126 	/* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1127 #if (PAGE_SIZE < 8192)
1128 	frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1129 #endif
1130 	xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1131 
1132 	while (likely(total_rx_packets < budget)) {
1133 		struct ixgbevf_rx_buffer *rx_buffer;
1134 		union ixgbe_adv_rx_desc *rx_desc;
1135 		unsigned int size;
1136 
1137 		/* return some buffers to hardware, one at a time is too slow */
1138 		if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1139 			ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1140 			cleaned_count = 0;
1141 		}
1142 
1143 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1144 		size = le16_to_cpu(rx_desc->wb.upper.length);
1145 		if (!size)
1146 			break;
1147 
1148 		/* This memory barrier is needed to keep us from reading
1149 		 * any other fields out of the rx_desc until we know the
1150 		 * RXD_STAT_DD bit is set
1151 		 */
1152 		rmb();
1153 
1154 		rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1155 
1156 		/* retrieve a buffer from the ring */
1157 		if (!skb) {
1158 			unsigned int offset = ixgbevf_rx_offset(rx_ring);
1159 			unsigned char *hard_start;
1160 
1161 			hard_start = page_address(rx_buffer->page) +
1162 				     rx_buffer->page_offset - offset;
1163 			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1164 #if (PAGE_SIZE > 4096)
1165 			/* At larger PAGE_SIZE, frame_sz depend on len size */
1166 			xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1167 #endif
1168 			skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1169 		}
1170 
1171 		if (IS_ERR(skb)) {
1172 			if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1173 				xdp_xmit = true;
1174 				ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1175 						       size);
1176 			} else {
1177 				rx_buffer->pagecnt_bias++;
1178 			}
1179 			total_rx_packets++;
1180 			total_rx_bytes += size;
1181 		} else if (skb) {
1182 			ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1183 		} else if (ring_uses_build_skb(rx_ring)) {
1184 			skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1185 						&xdp, rx_desc);
1186 		} else {
1187 			skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1188 						    &xdp, rx_desc);
1189 		}
1190 
1191 		/* exit if we failed to retrieve a buffer */
1192 		if (!skb) {
1193 			rx_ring->rx_stats.alloc_rx_buff_failed++;
1194 			rx_buffer->pagecnt_bias++;
1195 			break;
1196 		}
1197 
1198 		ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1199 		cleaned_count++;
1200 
1201 		/* fetch next buffer in frame if non-eop */
1202 		if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1203 			continue;
1204 
1205 		/* verify the packet layout is correct */
1206 		if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1207 			skb = NULL;
1208 			continue;
1209 		}
1210 
1211 		/* probably a little skewed due to removing CRC */
1212 		total_rx_bytes += skb->len;
1213 
1214 		/* Workaround hardware that can't do proper VEPA multicast
1215 		 * source pruning.
1216 		 */
1217 		if ((skb->pkt_type == PACKET_BROADCAST ||
1218 		     skb->pkt_type == PACKET_MULTICAST) &&
1219 		    ether_addr_equal(rx_ring->netdev->dev_addr,
1220 				     eth_hdr(skb)->h_source)) {
1221 			dev_kfree_skb_irq(skb);
1222 			continue;
1223 		}
1224 
1225 		/* populate checksum, VLAN, and protocol */
1226 		ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1227 
1228 		ixgbevf_rx_skb(q_vector, skb);
1229 
1230 		/* reset skb pointer */
1231 		skb = NULL;
1232 
1233 		/* update budget accounting */
1234 		total_rx_packets++;
1235 	}
1236 
1237 	/* place incomplete frames back on ring for completion */
1238 	rx_ring->skb = skb;
1239 
1240 	if (xdp_xmit) {
1241 		struct ixgbevf_ring *xdp_ring =
1242 			adapter->xdp_ring[rx_ring->queue_index];
1243 
1244 		/* Force memory writes to complete before letting h/w
1245 		 * know there are new descriptors to fetch.
1246 		 */
1247 		wmb();
1248 		ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1249 	}
1250 
1251 	u64_stats_update_begin(&rx_ring->syncp);
1252 	rx_ring->stats.packets += total_rx_packets;
1253 	rx_ring->stats.bytes += total_rx_bytes;
1254 	u64_stats_update_end(&rx_ring->syncp);
1255 	q_vector->rx.total_packets += total_rx_packets;
1256 	q_vector->rx.total_bytes += total_rx_bytes;
1257 
1258 	return total_rx_packets;
1259 }
1260 
1261 /**
1262  * ixgbevf_poll - NAPI polling calback
1263  * @napi: napi struct with our devices info in it
1264  * @budget: amount of work driver is allowed to do this pass, in packets
1265  *
1266  * This function will clean more than one or more rings associated with a
1267  * q_vector.
1268  **/
1269 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1270 {
1271 	struct ixgbevf_q_vector *q_vector =
1272 		container_of(napi, struct ixgbevf_q_vector, napi);
1273 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1274 	struct ixgbevf_ring *ring;
1275 	int per_ring_budget, work_done = 0;
1276 	bool clean_complete = true;
1277 
1278 	ixgbevf_for_each_ring(ring, q_vector->tx) {
1279 		if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1280 			clean_complete = false;
1281 	}
1282 
1283 	if (budget <= 0)
1284 		return budget;
1285 
1286 	/* attempt to distribute budget to each queue fairly, but don't allow
1287 	 * the budget to go below 1 because we'll exit polling
1288 	 */
1289 	if (q_vector->rx.count > 1)
1290 		per_ring_budget = max(budget/q_vector->rx.count, 1);
1291 	else
1292 		per_ring_budget = budget;
1293 
1294 	ixgbevf_for_each_ring(ring, q_vector->rx) {
1295 		int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1296 						   per_ring_budget);
1297 		work_done += cleaned;
1298 		if (cleaned >= per_ring_budget)
1299 			clean_complete = false;
1300 	}
1301 
1302 	/* If all work not completed, return budget and keep polling */
1303 	if (!clean_complete)
1304 		return budget;
1305 
1306 	/* Exit the polling mode, but don't re-enable interrupts if stack might
1307 	 * poll us due to busy-polling
1308 	 */
1309 	if (likely(napi_complete_done(napi, work_done))) {
1310 		if (adapter->rx_itr_setting == 1)
1311 			ixgbevf_set_itr(q_vector);
1312 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1313 		    !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1314 			ixgbevf_irq_enable_queues(adapter,
1315 						  BIT(q_vector->v_idx));
1316 	}
1317 
1318 	return min(work_done, budget - 1);
1319 }
1320 
1321 /**
1322  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1323  * @q_vector: structure containing interrupt and ring information
1324  **/
1325 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1326 {
1327 	struct ixgbevf_adapter *adapter = q_vector->adapter;
1328 	struct ixgbe_hw *hw = &adapter->hw;
1329 	int v_idx = q_vector->v_idx;
1330 	u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1331 
1332 	/* set the WDIS bit to not clear the timer bits and cause an
1333 	 * immediate assertion of the interrupt
1334 	 */
1335 	itr_reg |= IXGBE_EITR_CNT_WDIS;
1336 
1337 	IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1338 }
1339 
1340 /**
1341  * ixgbevf_configure_msix - Configure MSI-X hardware
1342  * @adapter: board private structure
1343  *
1344  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1345  * interrupts.
1346  **/
1347 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1348 {
1349 	struct ixgbevf_q_vector *q_vector;
1350 	int q_vectors, v_idx;
1351 
1352 	q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1353 	adapter->eims_enable_mask = 0;
1354 
1355 	/* Populate the IVAR table and set the ITR values to the
1356 	 * corresponding register.
1357 	 */
1358 	for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1359 		struct ixgbevf_ring *ring;
1360 
1361 		q_vector = adapter->q_vector[v_idx];
1362 
1363 		ixgbevf_for_each_ring(ring, q_vector->rx)
1364 			ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1365 
1366 		ixgbevf_for_each_ring(ring, q_vector->tx)
1367 			ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1368 
1369 		if (q_vector->tx.ring && !q_vector->rx.ring) {
1370 			/* Tx only vector */
1371 			if (adapter->tx_itr_setting == 1)
1372 				q_vector->itr = IXGBE_12K_ITR;
1373 			else
1374 				q_vector->itr = adapter->tx_itr_setting;
1375 		} else {
1376 			/* Rx or Rx/Tx vector */
1377 			if (adapter->rx_itr_setting == 1)
1378 				q_vector->itr = IXGBE_20K_ITR;
1379 			else
1380 				q_vector->itr = adapter->rx_itr_setting;
1381 		}
1382 
1383 		/* add q_vector eims value to global eims_enable_mask */
1384 		adapter->eims_enable_mask |= BIT(v_idx);
1385 
1386 		ixgbevf_write_eitr(q_vector);
1387 	}
1388 
1389 	ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1390 	/* setup eims_other and add value to global eims_enable_mask */
1391 	adapter->eims_other = BIT(v_idx);
1392 	adapter->eims_enable_mask |= adapter->eims_other;
1393 }
1394 
1395 enum latency_range {
1396 	lowest_latency = 0,
1397 	low_latency = 1,
1398 	bulk_latency = 2,
1399 	latency_invalid = 255
1400 };
1401 
1402 /**
1403  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1404  * @q_vector: structure containing interrupt and ring information
1405  * @ring_container: structure containing ring performance data
1406  *
1407  * Stores a new ITR value based on packets and byte
1408  * counts during the last interrupt.  The advantage of per interrupt
1409  * computation is faster updates and more accurate ITR for the current
1410  * traffic pattern.  Constants in this function were computed
1411  * based on theoretical maximum wire speed and thresholds were set based
1412  * on testing data as well as attempting to minimize response time
1413  * while increasing bulk throughput.
1414  **/
1415 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1416 			       struct ixgbevf_ring_container *ring_container)
1417 {
1418 	int bytes = ring_container->total_bytes;
1419 	int packets = ring_container->total_packets;
1420 	u32 timepassed_us;
1421 	u64 bytes_perint;
1422 	u8 itr_setting = ring_container->itr;
1423 
1424 	if (packets == 0)
1425 		return;
1426 
1427 	/* simple throttle rate management
1428 	 *    0-20MB/s lowest (100000 ints/s)
1429 	 *   20-100MB/s low   (20000 ints/s)
1430 	 *  100-1249MB/s bulk (12000 ints/s)
1431 	 */
1432 	/* what was last interrupt timeslice? */
1433 	timepassed_us = q_vector->itr >> 2;
1434 	if (timepassed_us == 0)
1435 		return;
1436 
1437 	bytes_perint = bytes / timepassed_us; /* bytes/usec */
1438 
1439 	switch (itr_setting) {
1440 	case lowest_latency:
1441 		if (bytes_perint > 10)
1442 			itr_setting = low_latency;
1443 		break;
1444 	case low_latency:
1445 		if (bytes_perint > 20)
1446 			itr_setting = bulk_latency;
1447 		else if (bytes_perint <= 10)
1448 			itr_setting = lowest_latency;
1449 		break;
1450 	case bulk_latency:
1451 		if (bytes_perint <= 20)
1452 			itr_setting = low_latency;
1453 		break;
1454 	}
1455 
1456 	/* clear work counters since we have the values we need */
1457 	ring_container->total_bytes = 0;
1458 	ring_container->total_packets = 0;
1459 
1460 	/* write updated itr to ring container */
1461 	ring_container->itr = itr_setting;
1462 }
1463 
1464 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1465 {
1466 	u32 new_itr = q_vector->itr;
1467 	u8 current_itr;
1468 
1469 	ixgbevf_update_itr(q_vector, &q_vector->tx);
1470 	ixgbevf_update_itr(q_vector, &q_vector->rx);
1471 
1472 	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1473 
1474 	switch (current_itr) {
1475 	/* counts and packets in update_itr are dependent on these numbers */
1476 	case lowest_latency:
1477 		new_itr = IXGBE_100K_ITR;
1478 		break;
1479 	case low_latency:
1480 		new_itr = IXGBE_20K_ITR;
1481 		break;
1482 	case bulk_latency:
1483 		new_itr = IXGBE_12K_ITR;
1484 		break;
1485 	default:
1486 		break;
1487 	}
1488 
1489 	if (new_itr != q_vector->itr) {
1490 		/* do an exponential smoothing */
1491 		new_itr = (10 * new_itr * q_vector->itr) /
1492 			  ((9 * new_itr) + q_vector->itr);
1493 
1494 		/* save the algorithm value here */
1495 		q_vector->itr = new_itr;
1496 
1497 		ixgbevf_write_eitr(q_vector);
1498 	}
1499 }
1500 
1501 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1502 {
1503 	struct ixgbevf_adapter *adapter = data;
1504 	struct ixgbe_hw *hw = &adapter->hw;
1505 
1506 	hw->mac.get_link_status = 1;
1507 
1508 	ixgbevf_service_event_schedule(adapter);
1509 
1510 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1511 
1512 	return IRQ_HANDLED;
1513 }
1514 
1515 /**
1516  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1517  * @irq: unused
1518  * @data: pointer to our q_vector struct for this interrupt vector
1519  **/
1520 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1521 {
1522 	struct ixgbevf_q_vector *q_vector = data;
1523 
1524 	/* EIAM disabled interrupts (on this vector) for us */
1525 	if (q_vector->rx.ring || q_vector->tx.ring)
1526 		napi_schedule_irqoff(&q_vector->napi);
1527 
1528 	return IRQ_HANDLED;
1529 }
1530 
1531 /**
1532  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1533  * @adapter: board private structure
1534  *
1535  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1536  * interrupts from the kernel.
1537  **/
1538 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1539 {
1540 	struct net_device *netdev = adapter->netdev;
1541 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1542 	unsigned int ri = 0, ti = 0;
1543 	int vector, err;
1544 
1545 	for (vector = 0; vector < q_vectors; vector++) {
1546 		struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1547 		struct msix_entry *entry = &adapter->msix_entries[vector];
1548 
1549 		if (q_vector->tx.ring && q_vector->rx.ring) {
1550 			snprintf(q_vector->name, sizeof(q_vector->name),
1551 				 "%s-TxRx-%u", netdev->name, ri++);
1552 			ti++;
1553 		} else if (q_vector->rx.ring) {
1554 			snprintf(q_vector->name, sizeof(q_vector->name),
1555 				 "%s-rx-%u", netdev->name, ri++);
1556 		} else if (q_vector->tx.ring) {
1557 			snprintf(q_vector->name, sizeof(q_vector->name),
1558 				 "%s-tx-%u", netdev->name, ti++);
1559 		} else {
1560 			/* skip this unused q_vector */
1561 			continue;
1562 		}
1563 		err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1564 				  q_vector->name, q_vector);
1565 		if (err) {
1566 			hw_dbg(&adapter->hw,
1567 			       "request_irq failed for MSIX interrupt Error: %d\n",
1568 			       err);
1569 			goto free_queue_irqs;
1570 		}
1571 	}
1572 
1573 	err = request_irq(adapter->msix_entries[vector].vector,
1574 			  &ixgbevf_msix_other, 0, netdev->name, adapter);
1575 	if (err) {
1576 		hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1577 		       err);
1578 		goto free_queue_irqs;
1579 	}
1580 
1581 	return 0;
1582 
1583 free_queue_irqs:
1584 	while (vector) {
1585 		vector--;
1586 		free_irq(adapter->msix_entries[vector].vector,
1587 			 adapter->q_vector[vector]);
1588 	}
1589 	/* This failure is non-recoverable - it indicates the system is
1590 	 * out of MSIX vector resources and the VF driver cannot run
1591 	 * without them.  Set the number of msix vectors to zero
1592 	 * indicating that not enough can be allocated.  The error
1593 	 * will be returned to the user indicating device open failed.
1594 	 * Any further attempts to force the driver to open will also
1595 	 * fail.  The only way to recover is to unload the driver and
1596 	 * reload it again.  If the system has recovered some MSIX
1597 	 * vectors then it may succeed.
1598 	 */
1599 	adapter->num_msix_vectors = 0;
1600 	return err;
1601 }
1602 
1603 /**
1604  * ixgbevf_request_irq - initialize interrupts
1605  * @adapter: board private structure
1606  *
1607  * Attempts to configure interrupts using the best available
1608  * capabilities of the hardware and kernel.
1609  **/
1610 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1611 {
1612 	int err = ixgbevf_request_msix_irqs(adapter);
1613 
1614 	if (err)
1615 		hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1616 
1617 	return err;
1618 }
1619 
1620 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1621 {
1622 	int i, q_vectors;
1623 
1624 	if (!adapter->msix_entries)
1625 		return;
1626 
1627 	q_vectors = adapter->num_msix_vectors;
1628 	i = q_vectors - 1;
1629 
1630 	free_irq(adapter->msix_entries[i].vector, adapter);
1631 	i--;
1632 
1633 	for (; i >= 0; i--) {
1634 		/* free only the irqs that were actually requested */
1635 		if (!adapter->q_vector[i]->rx.ring &&
1636 		    !adapter->q_vector[i]->tx.ring)
1637 			continue;
1638 
1639 		free_irq(adapter->msix_entries[i].vector,
1640 			 adapter->q_vector[i]);
1641 	}
1642 }
1643 
1644 /**
1645  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1646  * @adapter: board private structure
1647  **/
1648 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1649 {
1650 	struct ixgbe_hw *hw = &adapter->hw;
1651 	int i;
1652 
1653 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1654 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1655 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1656 
1657 	IXGBE_WRITE_FLUSH(hw);
1658 
1659 	for (i = 0; i < adapter->num_msix_vectors; i++)
1660 		synchronize_irq(adapter->msix_entries[i].vector);
1661 }
1662 
1663 /**
1664  * ixgbevf_irq_enable - Enable default interrupt generation settings
1665  * @adapter: board private structure
1666  **/
1667 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1668 {
1669 	struct ixgbe_hw *hw = &adapter->hw;
1670 
1671 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1672 	IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1673 	IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1674 }
1675 
1676 /**
1677  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1678  * @adapter: board private structure
1679  * @ring: structure containing ring specific data
1680  *
1681  * Configure the Tx descriptor ring after a reset.
1682  **/
1683 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1684 				      struct ixgbevf_ring *ring)
1685 {
1686 	struct ixgbe_hw *hw = &adapter->hw;
1687 	u64 tdba = ring->dma;
1688 	int wait_loop = 10;
1689 	u32 txdctl = IXGBE_TXDCTL_ENABLE;
1690 	u8 reg_idx = ring->reg_idx;
1691 
1692 	/* disable queue to avoid issues while updating state */
1693 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1694 	IXGBE_WRITE_FLUSH(hw);
1695 
1696 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1697 	IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1698 	IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1699 			ring->count * sizeof(union ixgbe_adv_tx_desc));
1700 
1701 	/* disable head writeback */
1702 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1703 	IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1704 
1705 	/* enable relaxed ordering */
1706 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1707 			(IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1708 			 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1709 
1710 	/* reset head and tail pointers */
1711 	IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1712 	IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1713 	ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1714 
1715 	/* reset ntu and ntc to place SW in sync with hardwdare */
1716 	ring->next_to_clean = 0;
1717 	ring->next_to_use = 0;
1718 
1719 	/* In order to avoid issues WTHRESH + PTHRESH should always be equal
1720 	 * to or less than the number of on chip descriptors, which is
1721 	 * currently 40.
1722 	 */
1723 	txdctl |= (8 << 16);    /* WTHRESH = 8 */
1724 
1725 	/* Setting PTHRESH to 32 both improves performance */
1726 	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
1727 		   32;           /* PTHRESH = 32 */
1728 
1729 	/* reinitialize tx_buffer_info */
1730 	memset(ring->tx_buffer_info, 0,
1731 	       sizeof(struct ixgbevf_tx_buffer) * ring->count);
1732 
1733 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1734 	clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1735 
1736 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1737 
1738 	/* poll to verify queue is enabled */
1739 	do {
1740 		usleep_range(1000, 2000);
1741 		txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1742 	}  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1743 	if (!wait_loop)
1744 		hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1745 }
1746 
1747 /**
1748  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1749  * @adapter: board private structure
1750  *
1751  * Configure the Tx unit of the MAC after a reset.
1752  **/
1753 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1754 {
1755 	u32 i;
1756 
1757 	/* Setup the HW Tx Head and Tail descriptor pointers */
1758 	for (i = 0; i < adapter->num_tx_queues; i++)
1759 		ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1760 	for (i = 0; i < adapter->num_xdp_queues; i++)
1761 		ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1762 }
1763 
1764 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT	2
1765 
1766 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1767 				     struct ixgbevf_ring *ring, int index)
1768 {
1769 	struct ixgbe_hw *hw = &adapter->hw;
1770 	u32 srrctl;
1771 
1772 	srrctl = IXGBE_SRRCTL_DROP_EN;
1773 
1774 	srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1775 	if (ring_uses_large_buffer(ring))
1776 		srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1777 	else
1778 		srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1779 	srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1780 
1781 	IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1782 }
1783 
1784 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1785 {
1786 	struct ixgbe_hw *hw = &adapter->hw;
1787 
1788 	/* PSRTYPE must be initialized in 82599 */
1789 	u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1790 		      IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1791 		      IXGBE_PSRTYPE_L2HDR;
1792 
1793 	if (adapter->num_rx_queues > 1)
1794 		psrtype |= BIT(29);
1795 
1796 	IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1797 }
1798 
1799 #define IXGBEVF_MAX_RX_DESC_POLL 10
1800 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1801 				     struct ixgbevf_ring *ring)
1802 {
1803 	struct ixgbe_hw *hw = &adapter->hw;
1804 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1805 	u32 rxdctl;
1806 	u8 reg_idx = ring->reg_idx;
1807 
1808 	if (IXGBE_REMOVED(hw->hw_addr))
1809 		return;
1810 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1811 	rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1812 
1813 	/* write value back with RXDCTL.ENABLE bit cleared */
1814 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1815 
1816 	/* the hardware may take up to 100us to really disable the Rx queue */
1817 	do {
1818 		udelay(10);
1819 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1820 	} while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1821 
1822 	if (!wait_loop)
1823 		pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1824 		       reg_idx);
1825 }
1826 
1827 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1828 					 struct ixgbevf_ring *ring)
1829 {
1830 	struct ixgbe_hw *hw = &adapter->hw;
1831 	int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1832 	u32 rxdctl;
1833 	u8 reg_idx = ring->reg_idx;
1834 
1835 	if (IXGBE_REMOVED(hw->hw_addr))
1836 		return;
1837 	do {
1838 		usleep_range(1000, 2000);
1839 		rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1840 	} while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1841 
1842 	if (!wait_loop)
1843 		pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1844 		       reg_idx);
1845 }
1846 
1847 /**
1848  * ixgbevf_init_rss_key - Initialize adapter RSS key
1849  * @adapter: device handle
1850  *
1851  * Allocates and initializes the RSS key if it is not allocated.
1852  **/
1853 static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1854 {
1855 	u32 *rss_key;
1856 
1857 	if (!adapter->rss_key) {
1858 		rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1859 		if (unlikely(!rss_key))
1860 			return -ENOMEM;
1861 
1862 		netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1863 		adapter->rss_key = rss_key;
1864 	}
1865 
1866 	return 0;
1867 }
1868 
1869 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1870 {
1871 	struct ixgbe_hw *hw = &adapter->hw;
1872 	u32 vfmrqc = 0, vfreta = 0;
1873 	u16 rss_i = adapter->num_rx_queues;
1874 	u8 i, j;
1875 
1876 	/* Fill out hash function seeds */
1877 	for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1878 		IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1879 
1880 	for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1881 		if (j == rss_i)
1882 			j = 0;
1883 
1884 		adapter->rss_indir_tbl[i] = j;
1885 
1886 		vfreta |= j << (i & 0x3) * 8;
1887 		if ((i & 3) == 3) {
1888 			IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1889 			vfreta = 0;
1890 		}
1891 	}
1892 
1893 	/* Perform hash on these packet types */
1894 	vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1895 		IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1896 		IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1897 		IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1898 
1899 	vfmrqc |= IXGBE_VFMRQC_RSSEN;
1900 
1901 	IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1902 }
1903 
1904 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1905 				      struct ixgbevf_ring *ring)
1906 {
1907 	struct ixgbe_hw *hw = &adapter->hw;
1908 	union ixgbe_adv_rx_desc *rx_desc;
1909 	u64 rdba = ring->dma;
1910 	u32 rxdctl;
1911 	u8 reg_idx = ring->reg_idx;
1912 
1913 	/* disable queue to avoid issues while updating state */
1914 	rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1915 	ixgbevf_disable_rx_queue(adapter, ring);
1916 
1917 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1918 	IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1919 	IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1920 			ring->count * sizeof(union ixgbe_adv_rx_desc));
1921 
1922 #ifndef CONFIG_SPARC
1923 	/* enable relaxed ordering */
1924 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1925 			IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1926 #else
1927 	IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1928 			IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1929 			IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1930 #endif
1931 
1932 	/* reset head and tail pointers */
1933 	IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1934 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1935 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1936 
1937 	/* initialize rx_buffer_info */
1938 	memset(ring->rx_buffer_info, 0,
1939 	       sizeof(struct ixgbevf_rx_buffer) * ring->count);
1940 
1941 	/* initialize Rx descriptor 0 */
1942 	rx_desc = IXGBEVF_RX_DESC(ring, 0);
1943 	rx_desc->wb.upper.length = 0;
1944 
1945 	/* reset ntu and ntc to place SW in sync with hardwdare */
1946 	ring->next_to_clean = 0;
1947 	ring->next_to_use = 0;
1948 	ring->next_to_alloc = 0;
1949 
1950 	ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1951 
1952 	/* RXDCTL.RLPML does not work on 82599 */
1953 	if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1954 		rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1955 			    IXGBE_RXDCTL_RLPML_EN);
1956 
1957 #if (PAGE_SIZE < 8192)
1958 		/* Limit the maximum frame size so we don't overrun the skb */
1959 		if (ring_uses_build_skb(ring) &&
1960 		    !ring_uses_large_buffer(ring))
1961 			rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1962 				  IXGBE_RXDCTL_RLPML_EN;
1963 #endif
1964 	}
1965 
1966 	rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1967 	IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1968 
1969 	ixgbevf_rx_desc_queue_enable(adapter, ring);
1970 	ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1971 }
1972 
1973 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1974 				      struct ixgbevf_ring *rx_ring)
1975 {
1976 	struct net_device *netdev = adapter->netdev;
1977 	unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1978 
1979 	/* set build_skb and buffer size flags */
1980 	clear_ring_build_skb_enabled(rx_ring);
1981 	clear_ring_uses_large_buffer(rx_ring);
1982 
1983 	if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1984 		return;
1985 
1986 	set_ring_build_skb_enabled(rx_ring);
1987 
1988 	if (PAGE_SIZE < 8192) {
1989 		if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1990 			return;
1991 
1992 		set_ring_uses_large_buffer(rx_ring);
1993 	}
1994 }
1995 
1996 /**
1997  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1998  * @adapter: board private structure
1999  *
2000  * Configure the Rx unit of the MAC after a reset.
2001  **/
2002 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2003 {
2004 	struct ixgbe_hw *hw = &adapter->hw;
2005 	struct net_device *netdev = adapter->netdev;
2006 	int i, ret;
2007 
2008 	ixgbevf_setup_psrtype(adapter);
2009 	if (hw->mac.type >= ixgbe_mac_X550_vf)
2010 		ixgbevf_setup_vfmrqc(adapter);
2011 
2012 	spin_lock_bh(&adapter->mbx_lock);
2013 	/* notify the PF of our intent to use this size of frame */
2014 	ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2015 	spin_unlock_bh(&adapter->mbx_lock);
2016 	if (ret)
2017 		dev_err(&adapter->pdev->dev,
2018 			"Failed to set MTU at %d\n", netdev->mtu);
2019 
2020 	/* Setup the HW Rx Head and Tail Descriptor Pointers and
2021 	 * the Base and Length of the Rx Descriptor Ring
2022 	 */
2023 	for (i = 0; i < adapter->num_rx_queues; i++) {
2024 		struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2025 
2026 		ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2027 		ixgbevf_configure_rx_ring(adapter, rx_ring);
2028 	}
2029 }
2030 
2031 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2032 				   __be16 proto, u16 vid)
2033 {
2034 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2035 	struct ixgbe_hw *hw = &adapter->hw;
2036 	int err;
2037 
2038 	spin_lock_bh(&adapter->mbx_lock);
2039 
2040 	/* add VID to filter table */
2041 	err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2042 
2043 	spin_unlock_bh(&adapter->mbx_lock);
2044 
2045 	/* translate error return types so error makes sense */
2046 	if (err == IXGBE_ERR_MBX)
2047 		return -EIO;
2048 
2049 	if (err == IXGBE_ERR_INVALID_ARGUMENT)
2050 		return -EACCES;
2051 
2052 	set_bit(vid, adapter->active_vlans);
2053 
2054 	return err;
2055 }
2056 
2057 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2058 				    __be16 proto, u16 vid)
2059 {
2060 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2061 	struct ixgbe_hw *hw = &adapter->hw;
2062 	int err;
2063 
2064 	spin_lock_bh(&adapter->mbx_lock);
2065 
2066 	/* remove VID from filter table */
2067 	err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2068 
2069 	spin_unlock_bh(&adapter->mbx_lock);
2070 
2071 	clear_bit(vid, adapter->active_vlans);
2072 
2073 	return err;
2074 }
2075 
2076 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2077 {
2078 	u16 vid;
2079 
2080 	for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2081 		ixgbevf_vlan_rx_add_vid(adapter->netdev,
2082 					htons(ETH_P_8021Q), vid);
2083 }
2084 
2085 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2086 {
2087 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2088 	struct ixgbe_hw *hw = &adapter->hw;
2089 	int count = 0;
2090 
2091 	if (!netdev_uc_empty(netdev)) {
2092 		struct netdev_hw_addr *ha;
2093 
2094 		netdev_for_each_uc_addr(ha, netdev) {
2095 			hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2096 			udelay(200);
2097 		}
2098 	} else {
2099 		/* If the list is empty then send message to PF driver to
2100 		 * clear all MAC VLANs on this VF.
2101 		 */
2102 		hw->mac.ops.set_uc_addr(hw, 0, NULL);
2103 	}
2104 
2105 	return count;
2106 }
2107 
2108 /**
2109  * ixgbevf_set_rx_mode - Multicast and unicast set
2110  * @netdev: network interface device structure
2111  *
2112  * The set_rx_method entry point is called whenever the multicast address
2113  * list, unicast address list or the network interface flags are updated.
2114  * This routine is responsible for configuring the hardware for proper
2115  * multicast mode and configuring requested unicast filters.
2116  **/
2117 static void ixgbevf_set_rx_mode(struct net_device *netdev)
2118 {
2119 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2120 	struct ixgbe_hw *hw = &adapter->hw;
2121 	unsigned int flags = netdev->flags;
2122 	int xcast_mode;
2123 
2124 	/* request the most inclusive mode we need */
2125 	if (flags & IFF_PROMISC)
2126 		xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2127 	else if (flags & IFF_ALLMULTI)
2128 		xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2129 	else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2130 		xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2131 	else
2132 		xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2133 
2134 	spin_lock_bh(&adapter->mbx_lock);
2135 
2136 	hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2137 
2138 	/* reprogram multicast list */
2139 	hw->mac.ops.update_mc_addr_list(hw, netdev);
2140 
2141 	ixgbevf_write_uc_addr_list(netdev);
2142 
2143 	spin_unlock_bh(&adapter->mbx_lock);
2144 }
2145 
2146 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2147 {
2148 	int q_idx;
2149 	struct ixgbevf_q_vector *q_vector;
2150 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2151 
2152 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2153 		q_vector = adapter->q_vector[q_idx];
2154 		napi_enable(&q_vector->napi);
2155 	}
2156 }
2157 
2158 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2159 {
2160 	int q_idx;
2161 	struct ixgbevf_q_vector *q_vector;
2162 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2163 
2164 	for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2165 		q_vector = adapter->q_vector[q_idx];
2166 		napi_disable(&q_vector->napi);
2167 	}
2168 }
2169 
2170 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2171 {
2172 	struct ixgbe_hw *hw = &adapter->hw;
2173 	unsigned int def_q = 0;
2174 	unsigned int num_tcs = 0;
2175 	unsigned int num_rx_queues = adapter->num_rx_queues;
2176 	unsigned int num_tx_queues = adapter->num_tx_queues;
2177 	int err;
2178 
2179 	spin_lock_bh(&adapter->mbx_lock);
2180 
2181 	/* fetch queue configuration from the PF */
2182 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2183 
2184 	spin_unlock_bh(&adapter->mbx_lock);
2185 
2186 	if (err)
2187 		return err;
2188 
2189 	if (num_tcs > 1) {
2190 		/* we need only one Tx queue */
2191 		num_tx_queues = 1;
2192 
2193 		/* update default Tx ring register index */
2194 		adapter->tx_ring[0]->reg_idx = def_q;
2195 
2196 		/* we need as many queues as traffic classes */
2197 		num_rx_queues = num_tcs;
2198 	}
2199 
2200 	/* if we have a bad config abort request queue reset */
2201 	if ((adapter->num_rx_queues != num_rx_queues) ||
2202 	    (adapter->num_tx_queues != num_tx_queues)) {
2203 		/* force mailbox timeout to prevent further messages */
2204 		hw->mbx.timeout = 0;
2205 
2206 		/* wait for watchdog to come around and bail us out */
2207 		set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2208 	}
2209 
2210 	return 0;
2211 }
2212 
2213 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2214 {
2215 	ixgbevf_configure_dcb(adapter);
2216 
2217 	ixgbevf_set_rx_mode(adapter->netdev);
2218 
2219 	ixgbevf_restore_vlan(adapter);
2220 	ixgbevf_ipsec_restore(adapter);
2221 
2222 	ixgbevf_configure_tx(adapter);
2223 	ixgbevf_configure_rx(adapter);
2224 }
2225 
2226 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2227 {
2228 	/* Only save pre-reset stats if there are some */
2229 	if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2230 		adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2231 			adapter->stats.base_vfgprc;
2232 		adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2233 			adapter->stats.base_vfgptc;
2234 		adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2235 			adapter->stats.base_vfgorc;
2236 		adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2237 			adapter->stats.base_vfgotc;
2238 		adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2239 			adapter->stats.base_vfmprc;
2240 	}
2241 }
2242 
2243 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2244 {
2245 	struct ixgbe_hw *hw = &adapter->hw;
2246 
2247 	adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2248 	adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2249 	adapter->stats.last_vfgorc |=
2250 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2251 	adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2252 	adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2253 	adapter->stats.last_vfgotc |=
2254 		(((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2255 	adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2256 
2257 	adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2258 	adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2259 	adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2260 	adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2261 	adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2262 }
2263 
2264 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2265 {
2266 	struct ixgbe_hw *hw = &adapter->hw;
2267 	static const int api[] = {
2268 		ixgbe_mbox_api_14,
2269 		ixgbe_mbox_api_13,
2270 		ixgbe_mbox_api_12,
2271 		ixgbe_mbox_api_11,
2272 		ixgbe_mbox_api_10,
2273 		ixgbe_mbox_api_unknown
2274 	};
2275 	int err, idx = 0;
2276 
2277 	spin_lock_bh(&adapter->mbx_lock);
2278 
2279 	while (api[idx] != ixgbe_mbox_api_unknown) {
2280 		err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2281 		if (!err)
2282 			break;
2283 		idx++;
2284 	}
2285 
2286 	spin_unlock_bh(&adapter->mbx_lock);
2287 }
2288 
2289 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2290 {
2291 	struct net_device *netdev = adapter->netdev;
2292 	struct ixgbe_hw *hw = &adapter->hw;
2293 
2294 	ixgbevf_configure_msix(adapter);
2295 
2296 	spin_lock_bh(&adapter->mbx_lock);
2297 
2298 	if (is_valid_ether_addr(hw->mac.addr))
2299 		hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2300 	else
2301 		hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2302 
2303 	spin_unlock_bh(&adapter->mbx_lock);
2304 
2305 	smp_mb__before_atomic();
2306 	clear_bit(__IXGBEVF_DOWN, &adapter->state);
2307 	ixgbevf_napi_enable_all(adapter);
2308 
2309 	/* clear any pending interrupts, may auto mask */
2310 	IXGBE_READ_REG(hw, IXGBE_VTEICR);
2311 	ixgbevf_irq_enable(adapter);
2312 
2313 	/* enable transmits */
2314 	netif_tx_start_all_queues(netdev);
2315 
2316 	ixgbevf_save_reset_stats(adapter);
2317 	ixgbevf_init_last_counter_stats(adapter);
2318 
2319 	hw->mac.get_link_status = 1;
2320 	mod_timer(&adapter->service_timer, jiffies);
2321 }
2322 
2323 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2324 {
2325 	ixgbevf_configure(adapter);
2326 
2327 	ixgbevf_up_complete(adapter);
2328 }
2329 
2330 /**
2331  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2332  * @rx_ring: ring to free buffers from
2333  **/
2334 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2335 {
2336 	u16 i = rx_ring->next_to_clean;
2337 
2338 	/* Free Rx ring sk_buff */
2339 	if (rx_ring->skb) {
2340 		dev_kfree_skb(rx_ring->skb);
2341 		rx_ring->skb = NULL;
2342 	}
2343 
2344 	/* Free all the Rx ring pages */
2345 	while (i != rx_ring->next_to_alloc) {
2346 		struct ixgbevf_rx_buffer *rx_buffer;
2347 
2348 		rx_buffer = &rx_ring->rx_buffer_info[i];
2349 
2350 		/* Invalidate cache lines that may have been written to by
2351 		 * device so that we avoid corrupting memory.
2352 		 */
2353 		dma_sync_single_range_for_cpu(rx_ring->dev,
2354 					      rx_buffer->dma,
2355 					      rx_buffer->page_offset,
2356 					      ixgbevf_rx_bufsz(rx_ring),
2357 					      DMA_FROM_DEVICE);
2358 
2359 		/* free resources associated with mapping */
2360 		dma_unmap_page_attrs(rx_ring->dev,
2361 				     rx_buffer->dma,
2362 				     ixgbevf_rx_pg_size(rx_ring),
2363 				     DMA_FROM_DEVICE,
2364 				     IXGBEVF_RX_DMA_ATTR);
2365 
2366 		__page_frag_cache_drain(rx_buffer->page,
2367 					rx_buffer->pagecnt_bias);
2368 
2369 		i++;
2370 		if (i == rx_ring->count)
2371 			i = 0;
2372 	}
2373 
2374 	rx_ring->next_to_alloc = 0;
2375 	rx_ring->next_to_clean = 0;
2376 	rx_ring->next_to_use = 0;
2377 }
2378 
2379 /**
2380  * ixgbevf_clean_tx_ring - Free Tx Buffers
2381  * @tx_ring: ring to be cleaned
2382  **/
2383 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2384 {
2385 	u16 i = tx_ring->next_to_clean;
2386 	struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2387 
2388 	while (i != tx_ring->next_to_use) {
2389 		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2390 
2391 		/* Free all the Tx ring sk_buffs */
2392 		if (ring_is_xdp(tx_ring))
2393 			page_frag_free(tx_buffer->data);
2394 		else
2395 			dev_kfree_skb_any(tx_buffer->skb);
2396 
2397 		/* unmap skb header data */
2398 		dma_unmap_single(tx_ring->dev,
2399 				 dma_unmap_addr(tx_buffer, dma),
2400 				 dma_unmap_len(tx_buffer, len),
2401 				 DMA_TO_DEVICE);
2402 
2403 		/* check for eop_desc to determine the end of the packet */
2404 		eop_desc = tx_buffer->next_to_watch;
2405 		tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2406 
2407 		/* unmap remaining buffers */
2408 		while (tx_desc != eop_desc) {
2409 			tx_buffer++;
2410 			tx_desc++;
2411 			i++;
2412 			if (unlikely(i == tx_ring->count)) {
2413 				i = 0;
2414 				tx_buffer = tx_ring->tx_buffer_info;
2415 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2416 			}
2417 
2418 			/* unmap any remaining paged data */
2419 			if (dma_unmap_len(tx_buffer, len))
2420 				dma_unmap_page(tx_ring->dev,
2421 					       dma_unmap_addr(tx_buffer, dma),
2422 					       dma_unmap_len(tx_buffer, len),
2423 					       DMA_TO_DEVICE);
2424 		}
2425 
2426 		/* move us one more past the eop_desc for start of next pkt */
2427 		tx_buffer++;
2428 		i++;
2429 		if (unlikely(i == tx_ring->count)) {
2430 			i = 0;
2431 			tx_buffer = tx_ring->tx_buffer_info;
2432 		}
2433 	}
2434 
2435 	/* reset next_to_use and next_to_clean */
2436 	tx_ring->next_to_use = 0;
2437 	tx_ring->next_to_clean = 0;
2438 
2439 }
2440 
2441 /**
2442  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2443  * @adapter: board private structure
2444  **/
2445 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2446 {
2447 	int i;
2448 
2449 	for (i = 0; i < adapter->num_rx_queues; i++)
2450 		ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2451 }
2452 
2453 /**
2454  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2455  * @adapter: board private structure
2456  **/
2457 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2458 {
2459 	int i;
2460 
2461 	for (i = 0; i < adapter->num_tx_queues; i++)
2462 		ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2463 	for (i = 0; i < adapter->num_xdp_queues; i++)
2464 		ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2465 }
2466 
2467 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2468 {
2469 	struct net_device *netdev = adapter->netdev;
2470 	struct ixgbe_hw *hw = &adapter->hw;
2471 	int i;
2472 
2473 	/* signal that we are down to the interrupt handler */
2474 	if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2475 		return; /* do nothing if already down */
2476 
2477 	/* disable all enabled Rx queues */
2478 	for (i = 0; i < adapter->num_rx_queues; i++)
2479 		ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2480 
2481 	usleep_range(10000, 20000);
2482 
2483 	netif_tx_stop_all_queues(netdev);
2484 
2485 	/* call carrier off first to avoid false dev_watchdog timeouts */
2486 	netif_carrier_off(netdev);
2487 	netif_tx_disable(netdev);
2488 
2489 	ixgbevf_irq_disable(adapter);
2490 
2491 	ixgbevf_napi_disable_all(adapter);
2492 
2493 	del_timer_sync(&adapter->service_timer);
2494 
2495 	/* disable transmits in the hardware now that interrupts are off */
2496 	for (i = 0; i < adapter->num_tx_queues; i++) {
2497 		u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2498 
2499 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2500 				IXGBE_TXDCTL_SWFLSH);
2501 	}
2502 
2503 	for (i = 0; i < adapter->num_xdp_queues; i++) {
2504 		u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2505 
2506 		IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2507 				IXGBE_TXDCTL_SWFLSH);
2508 	}
2509 
2510 	if (!pci_channel_offline(adapter->pdev))
2511 		ixgbevf_reset(adapter);
2512 
2513 	ixgbevf_clean_all_tx_rings(adapter);
2514 	ixgbevf_clean_all_rx_rings(adapter);
2515 }
2516 
2517 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2518 {
2519 	while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2520 		msleep(1);
2521 
2522 	ixgbevf_down(adapter);
2523 	pci_set_master(adapter->pdev);
2524 	ixgbevf_up(adapter);
2525 
2526 	clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2527 }
2528 
2529 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2530 {
2531 	struct ixgbe_hw *hw = &adapter->hw;
2532 	struct net_device *netdev = adapter->netdev;
2533 
2534 	if (hw->mac.ops.reset_hw(hw)) {
2535 		hw_dbg(hw, "PF still resetting\n");
2536 	} else {
2537 		hw->mac.ops.init_hw(hw);
2538 		ixgbevf_negotiate_api(adapter);
2539 	}
2540 
2541 	if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2542 		ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2543 		ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2544 	}
2545 
2546 	adapter->last_reset = jiffies;
2547 }
2548 
2549 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2550 					int vectors)
2551 {
2552 	int vector_threshold;
2553 
2554 	/* We'll want at least 2 (vector_threshold):
2555 	 * 1) TxQ[0] + RxQ[0] handler
2556 	 * 2) Other (Link Status Change, etc.)
2557 	 */
2558 	vector_threshold = MIN_MSIX_COUNT;
2559 
2560 	/* The more we get, the more we will assign to Tx/Rx Cleanup
2561 	 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2562 	 * Right now, we simply care about how many we'll get; we'll
2563 	 * set them up later while requesting irq's.
2564 	 */
2565 	vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2566 					vector_threshold, vectors);
2567 
2568 	if (vectors < 0) {
2569 		dev_err(&adapter->pdev->dev,
2570 			"Unable to allocate MSI-X interrupts\n");
2571 		kfree(adapter->msix_entries);
2572 		adapter->msix_entries = NULL;
2573 		return vectors;
2574 	}
2575 
2576 	/* Adjust for only the vectors we'll use, which is minimum
2577 	 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2578 	 * vectors we were allocated.
2579 	 */
2580 	adapter->num_msix_vectors = vectors;
2581 
2582 	return 0;
2583 }
2584 
2585 /**
2586  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2587  * @adapter: board private structure to initialize
2588  *
2589  * This is the top level queue allocation routine.  The order here is very
2590  * important, starting with the "most" number of features turned on at once,
2591  * and ending with the smallest set of features.  This way large combinations
2592  * can be allocated if they're turned on, and smaller combinations are the
2593  * fall through conditions.
2594  *
2595  **/
2596 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2597 {
2598 	struct ixgbe_hw *hw = &adapter->hw;
2599 	unsigned int def_q = 0;
2600 	unsigned int num_tcs = 0;
2601 	int err;
2602 
2603 	/* Start with base case */
2604 	adapter->num_rx_queues = 1;
2605 	adapter->num_tx_queues = 1;
2606 	adapter->num_xdp_queues = 0;
2607 
2608 	spin_lock_bh(&adapter->mbx_lock);
2609 
2610 	/* fetch queue configuration from the PF */
2611 	err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2612 
2613 	spin_unlock_bh(&adapter->mbx_lock);
2614 
2615 	if (err)
2616 		return;
2617 
2618 	/* we need as many queues as traffic classes */
2619 	if (num_tcs > 1) {
2620 		adapter->num_rx_queues = num_tcs;
2621 	} else {
2622 		u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2623 
2624 		switch (hw->api_version) {
2625 		case ixgbe_mbox_api_11:
2626 		case ixgbe_mbox_api_12:
2627 		case ixgbe_mbox_api_13:
2628 		case ixgbe_mbox_api_14:
2629 			if (adapter->xdp_prog &&
2630 			    hw->mac.max_tx_queues == rss)
2631 				rss = rss > 3 ? 2 : 1;
2632 
2633 			adapter->num_rx_queues = rss;
2634 			adapter->num_tx_queues = rss;
2635 			adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2636 			break;
2637 		default:
2638 			break;
2639 		}
2640 	}
2641 }
2642 
2643 /**
2644  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2645  * @adapter: board private structure to initialize
2646  *
2647  * Attempt to configure the interrupts using the best available
2648  * capabilities of the hardware and the kernel.
2649  **/
2650 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2651 {
2652 	int vector, v_budget;
2653 
2654 	/* It's easy to be greedy for MSI-X vectors, but it really
2655 	 * doesn't do us much good if we have a lot more vectors
2656 	 * than CPU's.  So let's be conservative and only ask for
2657 	 * (roughly) the same number of vectors as there are CPU's.
2658 	 * The default is to use pairs of vectors.
2659 	 */
2660 	v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2661 	v_budget = min_t(int, v_budget, num_online_cpus());
2662 	v_budget += NON_Q_VECTORS;
2663 
2664 	adapter->msix_entries = kcalloc(v_budget,
2665 					sizeof(struct msix_entry), GFP_KERNEL);
2666 	if (!adapter->msix_entries)
2667 		return -ENOMEM;
2668 
2669 	for (vector = 0; vector < v_budget; vector++)
2670 		adapter->msix_entries[vector].entry = vector;
2671 
2672 	/* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2673 	 * does not support any other modes, so we will simply fail here. Note
2674 	 * that we clean up the msix_entries pointer else-where.
2675 	 */
2676 	return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2677 }
2678 
2679 static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2680 			     struct ixgbevf_ring_container *head)
2681 {
2682 	ring->next = head->ring;
2683 	head->ring = ring;
2684 	head->count++;
2685 }
2686 
2687 /**
2688  * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2689  * @adapter: board private structure to initialize
2690  * @v_idx: index of vector in adapter struct
2691  * @txr_count: number of Tx rings for q vector
2692  * @txr_idx: index of first Tx ring to assign
2693  * @xdp_count: total number of XDP rings to allocate
2694  * @xdp_idx: index of first XDP ring to allocate
2695  * @rxr_count: number of Rx rings for q vector
2696  * @rxr_idx: index of first Rx ring to assign
2697  *
2698  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
2699  **/
2700 static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2701 				  int txr_count, int txr_idx,
2702 				  int xdp_count, int xdp_idx,
2703 				  int rxr_count, int rxr_idx)
2704 {
2705 	struct ixgbevf_q_vector *q_vector;
2706 	int reg_idx = txr_idx + xdp_idx;
2707 	struct ixgbevf_ring *ring;
2708 	int ring_count, size;
2709 
2710 	ring_count = txr_count + xdp_count + rxr_count;
2711 	size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2712 
2713 	/* allocate q_vector and rings */
2714 	q_vector = kzalloc(size, GFP_KERNEL);
2715 	if (!q_vector)
2716 		return -ENOMEM;
2717 
2718 	/* initialize NAPI */
2719 	netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2720 
2721 	/* tie q_vector and adapter together */
2722 	adapter->q_vector[v_idx] = q_vector;
2723 	q_vector->adapter = adapter;
2724 	q_vector->v_idx = v_idx;
2725 
2726 	/* initialize pointer to rings */
2727 	ring = q_vector->ring;
2728 
2729 	while (txr_count) {
2730 		/* assign generic ring traits */
2731 		ring->dev = &adapter->pdev->dev;
2732 		ring->netdev = adapter->netdev;
2733 
2734 		/* configure backlink on ring */
2735 		ring->q_vector = q_vector;
2736 
2737 		/* update q_vector Tx values */
2738 		ixgbevf_add_ring(ring, &q_vector->tx);
2739 
2740 		/* apply Tx specific ring traits */
2741 		ring->count = adapter->tx_ring_count;
2742 		ring->queue_index = txr_idx;
2743 		ring->reg_idx = reg_idx;
2744 
2745 		/* assign ring to adapter */
2746 		 adapter->tx_ring[txr_idx] = ring;
2747 
2748 		/* update count and index */
2749 		txr_count--;
2750 		txr_idx++;
2751 		reg_idx++;
2752 
2753 		/* push pointer to next ring */
2754 		ring++;
2755 	}
2756 
2757 	while (xdp_count) {
2758 		/* assign generic ring traits */
2759 		ring->dev = &adapter->pdev->dev;
2760 		ring->netdev = adapter->netdev;
2761 
2762 		/* configure backlink on ring */
2763 		ring->q_vector = q_vector;
2764 
2765 		/* update q_vector Tx values */
2766 		ixgbevf_add_ring(ring, &q_vector->tx);
2767 
2768 		/* apply Tx specific ring traits */
2769 		ring->count = adapter->tx_ring_count;
2770 		ring->queue_index = xdp_idx;
2771 		ring->reg_idx = reg_idx;
2772 		set_ring_xdp(ring);
2773 
2774 		/* assign ring to adapter */
2775 		adapter->xdp_ring[xdp_idx] = ring;
2776 
2777 		/* update count and index */
2778 		xdp_count--;
2779 		xdp_idx++;
2780 		reg_idx++;
2781 
2782 		/* push pointer to next ring */
2783 		ring++;
2784 	}
2785 
2786 	while (rxr_count) {
2787 		/* assign generic ring traits */
2788 		ring->dev = &adapter->pdev->dev;
2789 		ring->netdev = adapter->netdev;
2790 
2791 		/* configure backlink on ring */
2792 		ring->q_vector = q_vector;
2793 
2794 		/* update q_vector Rx values */
2795 		ixgbevf_add_ring(ring, &q_vector->rx);
2796 
2797 		/* apply Rx specific ring traits */
2798 		ring->count = adapter->rx_ring_count;
2799 		ring->queue_index = rxr_idx;
2800 		ring->reg_idx = rxr_idx;
2801 
2802 		/* assign ring to adapter */
2803 		adapter->rx_ring[rxr_idx] = ring;
2804 
2805 		/* update count and index */
2806 		rxr_count--;
2807 		rxr_idx++;
2808 
2809 		/* push pointer to next ring */
2810 		ring++;
2811 	}
2812 
2813 	return 0;
2814 }
2815 
2816 /**
2817  * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2818  * @adapter: board private structure to initialize
2819  * @v_idx: index of vector in adapter struct
2820  *
2821  * This function frees the memory allocated to the q_vector.  In addition if
2822  * NAPI is enabled it will delete any references to the NAPI struct prior
2823  * to freeing the q_vector.
2824  **/
2825 static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2826 {
2827 	struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2828 	struct ixgbevf_ring *ring;
2829 
2830 	ixgbevf_for_each_ring(ring, q_vector->tx) {
2831 		if (ring_is_xdp(ring))
2832 			adapter->xdp_ring[ring->queue_index] = NULL;
2833 		else
2834 			adapter->tx_ring[ring->queue_index] = NULL;
2835 	}
2836 
2837 	ixgbevf_for_each_ring(ring, q_vector->rx)
2838 		adapter->rx_ring[ring->queue_index] = NULL;
2839 
2840 	adapter->q_vector[v_idx] = NULL;
2841 	netif_napi_del(&q_vector->napi);
2842 
2843 	/* ixgbevf_get_stats() might access the rings on this vector,
2844 	 * we must wait a grace period before freeing it.
2845 	 */
2846 	kfree_rcu(q_vector, rcu);
2847 }
2848 
2849 /**
2850  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2851  * @adapter: board private structure to initialize
2852  *
2853  * We allocate one q_vector per queue interrupt.  If allocation fails we
2854  * return -ENOMEM.
2855  **/
2856 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2857 {
2858 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2859 	int rxr_remaining = adapter->num_rx_queues;
2860 	int txr_remaining = adapter->num_tx_queues;
2861 	int xdp_remaining = adapter->num_xdp_queues;
2862 	int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2863 	int err;
2864 
2865 	if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2866 		for (; rxr_remaining; v_idx++, q_vectors--) {
2867 			int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2868 
2869 			err = ixgbevf_alloc_q_vector(adapter, v_idx,
2870 						     0, 0, 0, 0, rqpv, rxr_idx);
2871 			if (err)
2872 				goto err_out;
2873 
2874 			/* update counts and index */
2875 			rxr_remaining -= rqpv;
2876 			rxr_idx += rqpv;
2877 		}
2878 	}
2879 
2880 	for (; q_vectors; v_idx++, q_vectors--) {
2881 		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2882 		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2883 		int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2884 
2885 		err = ixgbevf_alloc_q_vector(adapter, v_idx,
2886 					     tqpv, txr_idx,
2887 					     xqpv, xdp_idx,
2888 					     rqpv, rxr_idx);
2889 
2890 		if (err)
2891 			goto err_out;
2892 
2893 		/* update counts and index */
2894 		rxr_remaining -= rqpv;
2895 		rxr_idx += rqpv;
2896 		txr_remaining -= tqpv;
2897 		txr_idx += tqpv;
2898 		xdp_remaining -= xqpv;
2899 		xdp_idx += xqpv;
2900 	}
2901 
2902 	return 0;
2903 
2904 err_out:
2905 	while (v_idx) {
2906 		v_idx--;
2907 		ixgbevf_free_q_vector(adapter, v_idx);
2908 	}
2909 
2910 	return -ENOMEM;
2911 }
2912 
2913 /**
2914  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2915  * @adapter: board private structure to initialize
2916  *
2917  * This function frees the memory allocated to the q_vectors.  In addition if
2918  * NAPI is enabled it will delete any references to the NAPI struct prior
2919  * to freeing the q_vector.
2920  **/
2921 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2922 {
2923 	int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2924 
2925 	while (q_vectors) {
2926 		q_vectors--;
2927 		ixgbevf_free_q_vector(adapter, q_vectors);
2928 	}
2929 }
2930 
2931 /**
2932  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2933  * @adapter: board private structure
2934  *
2935  **/
2936 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2937 {
2938 	if (!adapter->msix_entries)
2939 		return;
2940 
2941 	pci_disable_msix(adapter->pdev);
2942 	kfree(adapter->msix_entries);
2943 	adapter->msix_entries = NULL;
2944 }
2945 
2946 /**
2947  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2948  * @adapter: board private structure to initialize
2949  *
2950  **/
2951 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2952 {
2953 	int err;
2954 
2955 	/* Number of supported queues */
2956 	ixgbevf_set_num_queues(adapter);
2957 
2958 	err = ixgbevf_set_interrupt_capability(adapter);
2959 	if (err) {
2960 		hw_dbg(&adapter->hw,
2961 		       "Unable to setup interrupt capabilities\n");
2962 		goto err_set_interrupt;
2963 	}
2964 
2965 	err = ixgbevf_alloc_q_vectors(adapter);
2966 	if (err) {
2967 		hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2968 		goto err_alloc_q_vectors;
2969 	}
2970 
2971 	hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2972 	       (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2973 	       adapter->num_rx_queues, adapter->num_tx_queues,
2974 	       adapter->num_xdp_queues);
2975 
2976 	set_bit(__IXGBEVF_DOWN, &adapter->state);
2977 
2978 	return 0;
2979 err_alloc_q_vectors:
2980 	ixgbevf_reset_interrupt_capability(adapter);
2981 err_set_interrupt:
2982 	return err;
2983 }
2984 
2985 /**
2986  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2987  * @adapter: board private structure to clear interrupt scheme on
2988  *
2989  * We go through and clear interrupt specific resources and reset the structure
2990  * to pre-load conditions
2991  **/
2992 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2993 {
2994 	adapter->num_tx_queues = 0;
2995 	adapter->num_xdp_queues = 0;
2996 	adapter->num_rx_queues = 0;
2997 
2998 	ixgbevf_free_q_vectors(adapter);
2999 	ixgbevf_reset_interrupt_capability(adapter);
3000 }
3001 
3002 /**
3003  * ixgbevf_sw_init - Initialize general software structures
3004  * @adapter: board private structure to initialize
3005  *
3006  * ixgbevf_sw_init initializes the Adapter private data structure.
3007  * Fields are initialized based on PCI device information and
3008  * OS network device settings (MTU size).
3009  **/
3010 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3011 {
3012 	struct ixgbe_hw *hw = &adapter->hw;
3013 	struct pci_dev *pdev = adapter->pdev;
3014 	struct net_device *netdev = adapter->netdev;
3015 	int err;
3016 
3017 	/* PCI config space info */
3018 	hw->vendor_id = pdev->vendor;
3019 	hw->device_id = pdev->device;
3020 	hw->revision_id = pdev->revision;
3021 	hw->subsystem_vendor_id = pdev->subsystem_vendor;
3022 	hw->subsystem_device_id = pdev->subsystem_device;
3023 
3024 	hw->mbx.ops.init_params(hw);
3025 
3026 	if (hw->mac.type >= ixgbe_mac_X550_vf) {
3027 		err = ixgbevf_init_rss_key(adapter);
3028 		if (err)
3029 			goto out;
3030 	}
3031 
3032 	/* assume legacy case in which PF would only give VF 2 queues */
3033 	hw->mac.max_tx_queues = 2;
3034 	hw->mac.max_rx_queues = 2;
3035 
3036 	/* lock to protect mailbox accesses */
3037 	spin_lock_init(&adapter->mbx_lock);
3038 
3039 	err = hw->mac.ops.reset_hw(hw);
3040 	if (err) {
3041 		dev_info(&pdev->dev,
3042 			 "PF still in reset state.  Is the PF interface up?\n");
3043 	} else {
3044 		err = hw->mac.ops.init_hw(hw);
3045 		if (err) {
3046 			pr_err("init_shared_code failed: %d\n", err);
3047 			goto out;
3048 		}
3049 		ixgbevf_negotiate_api(adapter);
3050 		err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3051 		if (err)
3052 			dev_info(&pdev->dev, "Error reading MAC address\n");
3053 		else if (is_zero_ether_addr(adapter->hw.mac.addr))
3054 			dev_info(&pdev->dev,
3055 				 "MAC address not assigned by administrator.\n");
3056 		ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3057 	}
3058 
3059 	if (!is_valid_ether_addr(netdev->dev_addr)) {
3060 		dev_info(&pdev->dev, "Assigning random MAC address\n");
3061 		eth_hw_addr_random(netdev);
3062 		ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3063 		ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3064 	}
3065 
3066 	/* Enable dynamic interrupt throttling rates */
3067 	adapter->rx_itr_setting = 1;
3068 	adapter->tx_itr_setting = 1;
3069 
3070 	/* set default ring sizes */
3071 	adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3072 	adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3073 
3074 	set_bit(__IXGBEVF_DOWN, &adapter->state);
3075 	return 0;
3076 
3077 out:
3078 	return err;
3079 }
3080 
3081 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)	\
3082 	{							\
3083 		u32 current_counter = IXGBE_READ_REG(hw, reg);	\
3084 		if (current_counter < last_counter)		\
3085 			counter += 0x100000000LL;		\
3086 		last_counter = current_counter;			\
3087 		counter &= 0xFFFFFFFF00000000LL;		\
3088 		counter |= current_counter;			\
3089 	}
3090 
3091 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3092 	{								 \
3093 		u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);	 \
3094 		u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);	 \
3095 		u64 current_counter = (current_counter_msb << 32) |	 \
3096 			current_counter_lsb;				 \
3097 		if (current_counter < last_counter)			 \
3098 			counter += 0x1000000000LL;			 \
3099 		last_counter = current_counter;				 \
3100 		counter &= 0xFFFFFFF000000000LL;			 \
3101 		counter |= current_counter;				 \
3102 	}
3103 /**
3104  * ixgbevf_update_stats - Update the board statistics counters.
3105  * @adapter: board private structure
3106  **/
3107 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3108 {
3109 	struct ixgbe_hw *hw = &adapter->hw;
3110 	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3111 	u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3112 	int i;
3113 
3114 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3115 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3116 		return;
3117 
3118 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3119 				adapter->stats.vfgprc);
3120 	UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3121 				adapter->stats.vfgptc);
3122 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3123 				adapter->stats.last_vfgorc,
3124 				adapter->stats.vfgorc);
3125 	UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3126 				adapter->stats.last_vfgotc,
3127 				adapter->stats.vfgotc);
3128 	UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3129 				adapter->stats.vfmprc);
3130 
3131 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
3132 		struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3133 
3134 		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3135 		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3136 		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3137 		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3138 	}
3139 
3140 	adapter->hw_csum_rx_error = hw_csum_rx_error;
3141 	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3142 	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3143 	adapter->alloc_rx_page = alloc_rx_page;
3144 }
3145 
3146 /**
3147  * ixgbevf_service_timer - Timer Call-back
3148  * @t: pointer to timer_list struct
3149  **/
3150 static void ixgbevf_service_timer(struct timer_list *t)
3151 {
3152 	struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3153 						     service_timer);
3154 
3155 	/* Reset the timer */
3156 	mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3157 
3158 	ixgbevf_service_event_schedule(adapter);
3159 }
3160 
3161 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3162 {
3163 	if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3164 		return;
3165 
3166 	rtnl_lock();
3167 	/* If we're already down or resetting, just bail */
3168 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3169 	    test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3170 	    test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3171 		rtnl_unlock();
3172 		return;
3173 	}
3174 
3175 	adapter->tx_timeout_count++;
3176 
3177 	ixgbevf_reinit_locked(adapter);
3178 	rtnl_unlock();
3179 }
3180 
3181 /**
3182  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3183  * @adapter: pointer to the device adapter structure
3184  *
3185  * This function serves two purposes.  First it strobes the interrupt lines
3186  * in order to make certain interrupts are occurring.  Secondly it sets the
3187  * bits needed to check for TX hangs.  As a result we should immediately
3188  * determine if a hang has occurred.
3189  **/
3190 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3191 {
3192 	struct ixgbe_hw *hw = &adapter->hw;
3193 	u32 eics = 0;
3194 	int i;
3195 
3196 	/* If we're down or resetting, just bail */
3197 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3198 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3199 		return;
3200 
3201 	/* Force detection of hung controller */
3202 	if (netif_carrier_ok(adapter->netdev)) {
3203 		for (i = 0; i < adapter->num_tx_queues; i++)
3204 			set_check_for_tx_hang(adapter->tx_ring[i]);
3205 		for (i = 0; i < adapter->num_xdp_queues; i++)
3206 			set_check_for_tx_hang(adapter->xdp_ring[i]);
3207 	}
3208 
3209 	/* get one bit for every active Tx/Rx interrupt vector */
3210 	for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3211 		struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3212 
3213 		if (qv->rx.ring || qv->tx.ring)
3214 			eics |= BIT(i);
3215 	}
3216 
3217 	/* Cause software interrupt to ensure rings are cleaned */
3218 	IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3219 }
3220 
3221 /**
3222  * ixgbevf_watchdog_update_link - update the link status
3223  * @adapter: pointer to the device adapter structure
3224  **/
3225 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3226 {
3227 	struct ixgbe_hw *hw = &adapter->hw;
3228 	u32 link_speed = adapter->link_speed;
3229 	bool link_up = adapter->link_up;
3230 	s32 err;
3231 
3232 	spin_lock_bh(&adapter->mbx_lock);
3233 
3234 	err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3235 
3236 	spin_unlock_bh(&adapter->mbx_lock);
3237 
3238 	/* if check for link returns error we will need to reset */
3239 	if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3240 		set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3241 		link_up = false;
3242 	}
3243 
3244 	adapter->link_up = link_up;
3245 	adapter->link_speed = link_speed;
3246 }
3247 
3248 /**
3249  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3250  *				 print link up message
3251  * @adapter: pointer to the device adapter structure
3252  **/
3253 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3254 {
3255 	struct net_device *netdev = adapter->netdev;
3256 
3257 	/* only continue if link was previously down */
3258 	if (netif_carrier_ok(netdev))
3259 		return;
3260 
3261 	dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3262 		 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3263 		 "10 Gbps" :
3264 		 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3265 		 "1 Gbps" :
3266 		 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3267 		 "100 Mbps" :
3268 		 "unknown speed");
3269 
3270 	netif_carrier_on(netdev);
3271 }
3272 
3273 /**
3274  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3275  *				   print link down message
3276  * @adapter: pointer to the adapter structure
3277  **/
3278 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3279 {
3280 	struct net_device *netdev = adapter->netdev;
3281 
3282 	adapter->link_speed = 0;
3283 
3284 	/* only continue if link was up previously */
3285 	if (!netif_carrier_ok(netdev))
3286 		return;
3287 
3288 	dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3289 
3290 	netif_carrier_off(netdev);
3291 }
3292 
3293 /**
3294  * ixgbevf_watchdog_subtask - worker thread to bring link up
3295  * @adapter: board private structure
3296  **/
3297 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3298 {
3299 	/* if interface is down do nothing */
3300 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3301 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3302 		return;
3303 
3304 	ixgbevf_watchdog_update_link(adapter);
3305 
3306 	if (adapter->link_up)
3307 		ixgbevf_watchdog_link_is_up(adapter);
3308 	else
3309 		ixgbevf_watchdog_link_is_down(adapter);
3310 
3311 	ixgbevf_update_stats(adapter);
3312 }
3313 
3314 /**
3315  * ixgbevf_service_task - manages and runs subtasks
3316  * @work: pointer to work_struct containing our data
3317  **/
3318 static void ixgbevf_service_task(struct work_struct *work)
3319 {
3320 	struct ixgbevf_adapter *adapter = container_of(work,
3321 						       struct ixgbevf_adapter,
3322 						       service_task);
3323 	struct ixgbe_hw *hw = &adapter->hw;
3324 
3325 	if (IXGBE_REMOVED(hw->hw_addr)) {
3326 		if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3327 			rtnl_lock();
3328 			ixgbevf_down(adapter);
3329 			rtnl_unlock();
3330 		}
3331 		return;
3332 	}
3333 
3334 	ixgbevf_queue_reset_subtask(adapter);
3335 	ixgbevf_reset_subtask(adapter);
3336 	ixgbevf_watchdog_subtask(adapter);
3337 	ixgbevf_check_hang_subtask(adapter);
3338 
3339 	ixgbevf_service_event_complete(adapter);
3340 }
3341 
3342 /**
3343  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3344  * @tx_ring: Tx descriptor ring for a specific queue
3345  *
3346  * Free all transmit software resources
3347  **/
3348 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3349 {
3350 	ixgbevf_clean_tx_ring(tx_ring);
3351 
3352 	vfree(tx_ring->tx_buffer_info);
3353 	tx_ring->tx_buffer_info = NULL;
3354 
3355 	/* if not set, then don't free */
3356 	if (!tx_ring->desc)
3357 		return;
3358 
3359 	dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3360 			  tx_ring->dma);
3361 
3362 	tx_ring->desc = NULL;
3363 }
3364 
3365 /**
3366  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3367  * @adapter: board private structure
3368  *
3369  * Free all transmit software resources
3370  **/
3371 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3372 {
3373 	int i;
3374 
3375 	for (i = 0; i < adapter->num_tx_queues; i++)
3376 		if (adapter->tx_ring[i]->desc)
3377 			ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3378 	for (i = 0; i < adapter->num_xdp_queues; i++)
3379 		if (adapter->xdp_ring[i]->desc)
3380 			ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3381 }
3382 
3383 /**
3384  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3385  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3386  *
3387  * Return 0 on success, negative on failure
3388  **/
3389 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3390 {
3391 	struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3392 	int size;
3393 
3394 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3395 	tx_ring->tx_buffer_info = vmalloc(size);
3396 	if (!tx_ring->tx_buffer_info)
3397 		goto err;
3398 
3399 	u64_stats_init(&tx_ring->syncp);
3400 
3401 	/* round up to nearest 4K */
3402 	tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3403 	tx_ring->size = ALIGN(tx_ring->size, 4096);
3404 
3405 	tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3406 					   &tx_ring->dma, GFP_KERNEL);
3407 	if (!tx_ring->desc)
3408 		goto err;
3409 
3410 	return 0;
3411 
3412 err:
3413 	vfree(tx_ring->tx_buffer_info);
3414 	tx_ring->tx_buffer_info = NULL;
3415 	hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3416 	return -ENOMEM;
3417 }
3418 
3419 /**
3420  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3421  * @adapter: board private structure
3422  *
3423  * If this function returns with an error, then it's possible one or
3424  * more of the rings is populated (while the rest are not).  It is the
3425  * callers duty to clean those orphaned rings.
3426  *
3427  * Return 0 on success, negative on failure
3428  **/
3429 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3430 {
3431 	int i, j = 0, err = 0;
3432 
3433 	for (i = 0; i < adapter->num_tx_queues; i++) {
3434 		err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3435 		if (!err)
3436 			continue;
3437 		hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3438 		goto err_setup_tx;
3439 	}
3440 
3441 	for (j = 0; j < adapter->num_xdp_queues; j++) {
3442 		err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3443 		if (!err)
3444 			continue;
3445 		hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3446 		goto err_setup_tx;
3447 	}
3448 
3449 	return 0;
3450 err_setup_tx:
3451 	/* rewind the index freeing the rings as we go */
3452 	while (j--)
3453 		ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3454 	while (i--)
3455 		ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3456 
3457 	return err;
3458 }
3459 
3460 /**
3461  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3462  * @adapter: board private structure
3463  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3464  *
3465  * Returns 0 on success, negative on failure
3466  **/
3467 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3468 			       struct ixgbevf_ring *rx_ring)
3469 {
3470 	int size;
3471 
3472 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3473 	rx_ring->rx_buffer_info = vmalloc(size);
3474 	if (!rx_ring->rx_buffer_info)
3475 		goto err;
3476 
3477 	u64_stats_init(&rx_ring->syncp);
3478 
3479 	/* Round up to nearest 4K */
3480 	rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3481 	rx_ring->size = ALIGN(rx_ring->size, 4096);
3482 
3483 	rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3484 					   &rx_ring->dma, GFP_KERNEL);
3485 
3486 	if (!rx_ring->desc)
3487 		goto err;
3488 
3489 	/* XDP RX-queue info */
3490 	if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3491 			     rx_ring->queue_index, 0) < 0)
3492 		goto err;
3493 
3494 	rx_ring->xdp_prog = adapter->xdp_prog;
3495 
3496 	return 0;
3497 err:
3498 	vfree(rx_ring->rx_buffer_info);
3499 	rx_ring->rx_buffer_info = NULL;
3500 	dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3501 	return -ENOMEM;
3502 }
3503 
3504 /**
3505  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3506  * @adapter: board private structure
3507  *
3508  * If this function returns with an error, then it's possible one or
3509  * more of the rings is populated (while the rest are not).  It is the
3510  * callers duty to clean those orphaned rings.
3511  *
3512  * Return 0 on success, negative on failure
3513  **/
3514 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3515 {
3516 	int i, err = 0;
3517 
3518 	for (i = 0; i < adapter->num_rx_queues; i++) {
3519 		err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3520 		if (!err)
3521 			continue;
3522 		hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3523 		goto err_setup_rx;
3524 	}
3525 
3526 	return 0;
3527 err_setup_rx:
3528 	/* rewind the index freeing the rings as we go */
3529 	while (i--)
3530 		ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3531 	return err;
3532 }
3533 
3534 /**
3535  * ixgbevf_free_rx_resources - Free Rx Resources
3536  * @rx_ring: ring to clean the resources from
3537  *
3538  * Free all receive software resources
3539  **/
3540 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3541 {
3542 	ixgbevf_clean_rx_ring(rx_ring);
3543 
3544 	rx_ring->xdp_prog = NULL;
3545 	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3546 	vfree(rx_ring->rx_buffer_info);
3547 	rx_ring->rx_buffer_info = NULL;
3548 
3549 	dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3550 			  rx_ring->dma);
3551 
3552 	rx_ring->desc = NULL;
3553 }
3554 
3555 /**
3556  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3557  * @adapter: board private structure
3558  *
3559  * Free all receive software resources
3560  **/
3561 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3562 {
3563 	int i;
3564 
3565 	for (i = 0; i < adapter->num_rx_queues; i++)
3566 		if (adapter->rx_ring[i]->desc)
3567 			ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3568 }
3569 
3570 /**
3571  * ixgbevf_open - Called when a network interface is made active
3572  * @netdev: network interface device structure
3573  *
3574  * Returns 0 on success, negative value on failure
3575  *
3576  * The open entry point is called when a network interface is made
3577  * active by the system (IFF_UP).  At this point all resources needed
3578  * for transmit and receive operations are allocated, the interrupt
3579  * handler is registered with the OS, the watchdog timer is started,
3580  * and the stack is notified that the interface is ready.
3581  **/
3582 int ixgbevf_open(struct net_device *netdev)
3583 {
3584 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3585 	struct ixgbe_hw *hw = &adapter->hw;
3586 	int err;
3587 
3588 	/* A previous failure to open the device because of a lack of
3589 	 * available MSIX vector resources may have reset the number
3590 	 * of msix vectors variable to zero.  The only way to recover
3591 	 * is to unload/reload the driver and hope that the system has
3592 	 * been able to recover some MSIX vector resources.
3593 	 */
3594 	if (!adapter->num_msix_vectors)
3595 		return -ENOMEM;
3596 
3597 	if (hw->adapter_stopped) {
3598 		ixgbevf_reset(adapter);
3599 		/* if adapter is still stopped then PF isn't up and
3600 		 * the VF can't start.
3601 		 */
3602 		if (hw->adapter_stopped) {
3603 			err = IXGBE_ERR_MBX;
3604 			pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3605 			goto err_setup_reset;
3606 		}
3607 	}
3608 
3609 	/* disallow open during test */
3610 	if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3611 		return -EBUSY;
3612 
3613 	netif_carrier_off(netdev);
3614 
3615 	/* allocate transmit descriptors */
3616 	err = ixgbevf_setup_all_tx_resources(adapter);
3617 	if (err)
3618 		goto err_setup_tx;
3619 
3620 	/* allocate receive descriptors */
3621 	err = ixgbevf_setup_all_rx_resources(adapter);
3622 	if (err)
3623 		goto err_setup_rx;
3624 
3625 	ixgbevf_configure(adapter);
3626 
3627 	err = ixgbevf_request_irq(adapter);
3628 	if (err)
3629 		goto err_req_irq;
3630 
3631 	/* Notify the stack of the actual queue counts. */
3632 	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3633 	if (err)
3634 		goto err_set_queues;
3635 
3636 	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3637 	if (err)
3638 		goto err_set_queues;
3639 
3640 	ixgbevf_up_complete(adapter);
3641 
3642 	return 0;
3643 
3644 err_set_queues:
3645 	ixgbevf_free_irq(adapter);
3646 err_req_irq:
3647 	ixgbevf_free_all_rx_resources(adapter);
3648 err_setup_rx:
3649 	ixgbevf_free_all_tx_resources(adapter);
3650 err_setup_tx:
3651 	ixgbevf_reset(adapter);
3652 err_setup_reset:
3653 
3654 	return err;
3655 }
3656 
3657 /**
3658  * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3659  * @adapter: the private adapter struct
3660  *
3661  * This function should contain the necessary work common to both suspending
3662  * and closing of the device.
3663  */
3664 static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3665 {
3666 	ixgbevf_down(adapter);
3667 	ixgbevf_free_irq(adapter);
3668 	ixgbevf_free_all_tx_resources(adapter);
3669 	ixgbevf_free_all_rx_resources(adapter);
3670 }
3671 
3672 /**
3673  * ixgbevf_close - Disables a network interface
3674  * @netdev: network interface device structure
3675  *
3676  * Returns 0, this is not allowed to fail
3677  *
3678  * The close entry point is called when an interface is de-activated
3679  * by the OS.  The hardware is still under the drivers control, but
3680  * needs to be disabled.  A global MAC reset is issued to stop the
3681  * hardware, and all transmit and receive resources are freed.
3682  **/
3683 int ixgbevf_close(struct net_device *netdev)
3684 {
3685 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3686 
3687 	if (netif_device_present(netdev))
3688 		ixgbevf_close_suspend(adapter);
3689 
3690 	return 0;
3691 }
3692 
3693 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3694 {
3695 	struct net_device *dev = adapter->netdev;
3696 
3697 	if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3698 				&adapter->state))
3699 		return;
3700 
3701 	/* if interface is down do nothing */
3702 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3703 	    test_bit(__IXGBEVF_RESETTING, &adapter->state))
3704 		return;
3705 
3706 	/* Hardware has to reinitialize queues and interrupts to
3707 	 * match packet buffer alignment. Unfortunately, the
3708 	 * hardware is not flexible enough to do this dynamically.
3709 	 */
3710 	rtnl_lock();
3711 
3712 	if (netif_running(dev))
3713 		ixgbevf_close(dev);
3714 
3715 	ixgbevf_clear_interrupt_scheme(adapter);
3716 	ixgbevf_init_interrupt_scheme(adapter);
3717 
3718 	if (netif_running(dev))
3719 		ixgbevf_open(dev);
3720 
3721 	rtnl_unlock();
3722 }
3723 
3724 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3725 				u32 vlan_macip_lens, u32 fceof_saidx,
3726 				u32 type_tucmd, u32 mss_l4len_idx)
3727 {
3728 	struct ixgbe_adv_tx_context_desc *context_desc;
3729 	u16 i = tx_ring->next_to_use;
3730 
3731 	context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3732 
3733 	i++;
3734 	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3735 
3736 	/* set bits to identify this as an advanced context descriptor */
3737 	type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3738 
3739 	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
3740 	context_desc->fceof_saidx	= cpu_to_le32(fceof_saidx);
3741 	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
3742 	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
3743 }
3744 
3745 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3746 		       struct ixgbevf_tx_buffer *first,
3747 		       u8 *hdr_len,
3748 		       struct ixgbevf_ipsec_tx_data *itd)
3749 {
3750 	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3751 	struct sk_buff *skb = first->skb;
3752 	union {
3753 		struct iphdr *v4;
3754 		struct ipv6hdr *v6;
3755 		unsigned char *hdr;
3756 	} ip;
3757 	union {
3758 		struct tcphdr *tcp;
3759 		unsigned char *hdr;
3760 	} l4;
3761 	u32 paylen, l4_offset;
3762 	u32 fceof_saidx = 0;
3763 	int err;
3764 
3765 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3766 		return 0;
3767 
3768 	if (!skb_is_gso(skb))
3769 		return 0;
3770 
3771 	err = skb_cow_head(skb, 0);
3772 	if (err < 0)
3773 		return err;
3774 
3775 	if (eth_p_mpls(first->protocol))
3776 		ip.hdr = skb_inner_network_header(skb);
3777 	else
3778 		ip.hdr = skb_network_header(skb);
3779 	l4.hdr = skb_checksum_start(skb);
3780 
3781 	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3782 	type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3783 
3784 	/* initialize outer IP header fields */
3785 	if (ip.v4->version == 4) {
3786 		unsigned char *csum_start = skb_checksum_start(skb);
3787 		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3788 		int len = csum_start - trans_start;
3789 
3790 		/* IP header will have to cancel out any data that
3791 		 * is not a part of the outer IP header, so set to
3792 		 * a reverse csum if needed, else init check to 0.
3793 		 */
3794 		ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3795 					   csum_fold(csum_partial(trans_start,
3796 								  len, 0)) : 0;
3797 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3798 
3799 		ip.v4->tot_len = 0;
3800 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3801 				   IXGBE_TX_FLAGS_CSUM |
3802 				   IXGBE_TX_FLAGS_IPV4;
3803 	} else {
3804 		ip.v6->payload_len = 0;
3805 		first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3806 				   IXGBE_TX_FLAGS_CSUM;
3807 	}
3808 
3809 	/* determine offset of inner transport header */
3810 	l4_offset = l4.hdr - skb->data;
3811 
3812 	/* compute length of segmentation header */
3813 	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
3814 
3815 	/* remove payload length from inner checksum */
3816 	paylen = skb->len - l4_offset;
3817 	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3818 
3819 	/* update gso size and bytecount with header size */
3820 	first->gso_segs = skb_shinfo(skb)->gso_segs;
3821 	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3822 
3823 	/* mss_l4len_id: use 1 as index for TSO */
3824 	mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3825 	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3826 	mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3827 
3828 	fceof_saidx |= itd->pfsa;
3829 	type_tucmd |= itd->flags | itd->trailer_len;
3830 
3831 	/* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3832 	vlan_macip_lens = l4.hdr - ip.hdr;
3833 	vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3834 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3835 
3836 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3837 			    mss_l4len_idx);
3838 
3839 	return 1;
3840 }
3841 
3842 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3843 			    struct ixgbevf_tx_buffer *first,
3844 			    struct ixgbevf_ipsec_tx_data *itd)
3845 {
3846 	struct sk_buff *skb = first->skb;
3847 	u32 vlan_macip_lens = 0;
3848 	u32 fceof_saidx = 0;
3849 	u32 type_tucmd = 0;
3850 
3851 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3852 		goto no_csum;
3853 
3854 	switch (skb->csum_offset) {
3855 	case offsetof(struct tcphdr, check):
3856 		type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3857 		fallthrough;
3858 	case offsetof(struct udphdr, check):
3859 		break;
3860 	case offsetof(struct sctphdr, checksum):
3861 		/* validate that this is actually an SCTP request */
3862 		if (skb_csum_is_sctp(skb)) {
3863 			type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3864 			break;
3865 		}
3866 		fallthrough;
3867 	default:
3868 		skb_checksum_help(skb);
3869 		goto no_csum;
3870 	}
3871 
3872 	if (first->protocol == htons(ETH_P_IP))
3873 		type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3874 
3875 	/* update TX checksum flag */
3876 	first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3877 	vlan_macip_lens = skb_checksum_start_offset(skb) -
3878 			  skb_network_offset(skb);
3879 no_csum:
3880 	/* vlan_macip_lens: MACLEN, VLAN tag */
3881 	vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3882 	vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3883 
3884 	fceof_saidx |= itd->pfsa;
3885 	type_tucmd |= itd->flags | itd->trailer_len;
3886 
3887 	ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3888 			    fceof_saidx, type_tucmd, 0);
3889 }
3890 
3891 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3892 {
3893 	/* set type for advanced descriptor with frame checksum insertion */
3894 	__le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3895 				      IXGBE_ADVTXD_DCMD_IFCS |
3896 				      IXGBE_ADVTXD_DCMD_DEXT);
3897 
3898 	/* set HW VLAN bit if VLAN is present */
3899 	if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3900 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3901 
3902 	/* set segmentation enable bits for TSO/FSO */
3903 	if (tx_flags & IXGBE_TX_FLAGS_TSO)
3904 		cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3905 
3906 	return cmd_type;
3907 }
3908 
3909 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3910 				     u32 tx_flags, unsigned int paylen)
3911 {
3912 	__le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3913 
3914 	/* enable L4 checksum for TSO and TX checksum offload */
3915 	if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3916 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3917 
3918 	/* enble IPv4 checksum for TSO */
3919 	if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3920 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3921 
3922 	/* enable IPsec */
3923 	if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3924 		olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3925 
3926 	/* use index 1 context for TSO/FSO/FCOE/IPSEC */
3927 	if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3928 		olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3929 
3930 	/* Check Context must be set if Tx switch is enabled, which it
3931 	 * always is for case where virtual functions are running
3932 	 */
3933 	olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3934 
3935 	tx_desc->read.olinfo_status = olinfo_status;
3936 }
3937 
3938 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3939 			   struct ixgbevf_tx_buffer *first,
3940 			   const u8 hdr_len)
3941 {
3942 	struct sk_buff *skb = first->skb;
3943 	struct ixgbevf_tx_buffer *tx_buffer;
3944 	union ixgbe_adv_tx_desc *tx_desc;
3945 	skb_frag_t *frag;
3946 	dma_addr_t dma;
3947 	unsigned int data_len, size;
3948 	u32 tx_flags = first->tx_flags;
3949 	__le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3950 	u16 i = tx_ring->next_to_use;
3951 
3952 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3953 
3954 	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3955 
3956 	size = skb_headlen(skb);
3957 	data_len = skb->data_len;
3958 
3959 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3960 
3961 	tx_buffer = first;
3962 
3963 	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3964 		if (dma_mapping_error(tx_ring->dev, dma))
3965 			goto dma_error;
3966 
3967 		/* record length, and DMA address */
3968 		dma_unmap_len_set(tx_buffer, len, size);
3969 		dma_unmap_addr_set(tx_buffer, dma, dma);
3970 
3971 		tx_desc->read.buffer_addr = cpu_to_le64(dma);
3972 
3973 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3974 			tx_desc->read.cmd_type_len =
3975 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3976 
3977 			i++;
3978 			tx_desc++;
3979 			if (i == tx_ring->count) {
3980 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3981 				i = 0;
3982 			}
3983 			tx_desc->read.olinfo_status = 0;
3984 
3985 			dma += IXGBE_MAX_DATA_PER_TXD;
3986 			size -= IXGBE_MAX_DATA_PER_TXD;
3987 
3988 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
3989 		}
3990 
3991 		if (likely(!data_len))
3992 			break;
3993 
3994 		tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3995 
3996 		i++;
3997 		tx_desc++;
3998 		if (i == tx_ring->count) {
3999 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4000 			i = 0;
4001 		}
4002 		tx_desc->read.olinfo_status = 0;
4003 
4004 		size = skb_frag_size(frag);
4005 		data_len -= size;
4006 
4007 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4008 				       DMA_TO_DEVICE);
4009 
4010 		tx_buffer = &tx_ring->tx_buffer_info[i];
4011 	}
4012 
4013 	/* write last descriptor with RS and EOP bits */
4014 	cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4015 	tx_desc->read.cmd_type_len = cmd_type;
4016 
4017 	/* set the timestamp */
4018 	first->time_stamp = jiffies;
4019 
4020 	skb_tx_timestamp(skb);
4021 
4022 	/* Force memory writes to complete before letting h/w know there
4023 	 * are new descriptors to fetch.  (Only applicable for weak-ordered
4024 	 * memory model archs, such as IA-64).
4025 	 *
4026 	 * We also need this memory barrier (wmb) to make certain all of the
4027 	 * status bits have been updated before next_to_watch is written.
4028 	 */
4029 	wmb();
4030 
4031 	/* set next_to_watch value indicating a packet is present */
4032 	first->next_to_watch = tx_desc;
4033 
4034 	i++;
4035 	if (i == tx_ring->count)
4036 		i = 0;
4037 
4038 	tx_ring->next_to_use = i;
4039 
4040 	/* notify HW of packet */
4041 	ixgbevf_write_tail(tx_ring, i);
4042 
4043 	return;
4044 dma_error:
4045 	dev_err(tx_ring->dev, "TX DMA map failed\n");
4046 	tx_buffer = &tx_ring->tx_buffer_info[i];
4047 
4048 	/* clear dma mappings for failed tx_buffer_info map */
4049 	while (tx_buffer != first) {
4050 		if (dma_unmap_len(tx_buffer, len))
4051 			dma_unmap_page(tx_ring->dev,
4052 				       dma_unmap_addr(tx_buffer, dma),
4053 				       dma_unmap_len(tx_buffer, len),
4054 				       DMA_TO_DEVICE);
4055 		dma_unmap_len_set(tx_buffer, len, 0);
4056 
4057 		if (i-- == 0)
4058 			i += tx_ring->count;
4059 		tx_buffer = &tx_ring->tx_buffer_info[i];
4060 	}
4061 
4062 	if (dma_unmap_len(tx_buffer, len))
4063 		dma_unmap_single(tx_ring->dev,
4064 				 dma_unmap_addr(tx_buffer, dma),
4065 				 dma_unmap_len(tx_buffer, len),
4066 				 DMA_TO_DEVICE);
4067 	dma_unmap_len_set(tx_buffer, len, 0);
4068 
4069 	dev_kfree_skb_any(tx_buffer->skb);
4070 	tx_buffer->skb = NULL;
4071 
4072 	tx_ring->next_to_use = i;
4073 }
4074 
4075 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4076 {
4077 	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4078 	/* Herbert's original patch had:
4079 	 *  smp_mb__after_netif_stop_queue();
4080 	 * but since that doesn't exist yet, just open code it.
4081 	 */
4082 	smp_mb();
4083 
4084 	/* We need to check again in a case another CPU has just
4085 	 * made room available.
4086 	 */
4087 	if (likely(ixgbevf_desc_unused(tx_ring) < size))
4088 		return -EBUSY;
4089 
4090 	/* A reprieve! - use start_queue because it doesn't call schedule */
4091 	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4092 	++tx_ring->tx_stats.restart_queue;
4093 
4094 	return 0;
4095 }
4096 
4097 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4098 {
4099 	if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4100 		return 0;
4101 	return __ixgbevf_maybe_stop_tx(tx_ring, size);
4102 }
4103 
4104 static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4105 				   struct ixgbevf_ring *tx_ring)
4106 {
4107 	struct ixgbevf_tx_buffer *first;
4108 	int tso;
4109 	u32 tx_flags = 0;
4110 	u16 count = TXD_USE_COUNT(skb_headlen(skb));
4111 	struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4112 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4113 	unsigned short f;
4114 #endif
4115 	u8 hdr_len = 0;
4116 	u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4117 
4118 	if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4119 		dev_kfree_skb_any(skb);
4120 		return NETDEV_TX_OK;
4121 	}
4122 
4123 	/* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4124 	 *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4125 	 *       + 2 desc gap to keep tail from touching head,
4126 	 *       + 1 desc for context descriptor,
4127 	 * otherwise try next time
4128 	 */
4129 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4130 	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4131 		skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4132 
4133 		count += TXD_USE_COUNT(skb_frag_size(frag));
4134 	}
4135 #else
4136 	count += skb_shinfo(skb)->nr_frags;
4137 #endif
4138 	if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4139 		tx_ring->tx_stats.tx_busy++;
4140 		return NETDEV_TX_BUSY;
4141 	}
4142 
4143 	/* record the location of the first descriptor for this packet */
4144 	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4145 	first->skb = skb;
4146 	first->bytecount = skb->len;
4147 	first->gso_segs = 1;
4148 
4149 	if (skb_vlan_tag_present(skb)) {
4150 		tx_flags |= skb_vlan_tag_get(skb);
4151 		tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4152 		tx_flags |= IXGBE_TX_FLAGS_VLAN;
4153 	}
4154 
4155 	/* record initial flags and protocol */
4156 	first->tx_flags = tx_flags;
4157 	first->protocol = vlan_get_protocol(skb);
4158 
4159 #ifdef CONFIG_IXGBEVF_IPSEC
4160 	if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4161 		goto out_drop;
4162 #endif
4163 	tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4164 	if (tso < 0)
4165 		goto out_drop;
4166 	else if (!tso)
4167 		ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4168 
4169 	ixgbevf_tx_map(tx_ring, first, hdr_len);
4170 
4171 	ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4172 
4173 	return NETDEV_TX_OK;
4174 
4175 out_drop:
4176 	dev_kfree_skb_any(first->skb);
4177 	first->skb = NULL;
4178 
4179 	return NETDEV_TX_OK;
4180 }
4181 
4182 static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4183 {
4184 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4185 	struct ixgbevf_ring *tx_ring;
4186 
4187 	if (skb->len <= 0) {
4188 		dev_kfree_skb_any(skb);
4189 		return NETDEV_TX_OK;
4190 	}
4191 
4192 	/* The minimum packet size for olinfo paylen is 17 so pad the skb
4193 	 * in order to meet this minimum size requirement.
4194 	 */
4195 	if (skb->len < 17) {
4196 		if (skb_padto(skb, 17))
4197 			return NETDEV_TX_OK;
4198 		skb->len = 17;
4199 	}
4200 
4201 	tx_ring = adapter->tx_ring[skb->queue_mapping];
4202 	return ixgbevf_xmit_frame_ring(skb, tx_ring);
4203 }
4204 
4205 /**
4206  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4207  * @netdev: network interface device structure
4208  * @p: pointer to an address structure
4209  *
4210  * Returns 0 on success, negative on failure
4211  **/
4212 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4213 {
4214 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4215 	struct ixgbe_hw *hw = &adapter->hw;
4216 	struct sockaddr *addr = p;
4217 	int err;
4218 
4219 	if (!is_valid_ether_addr(addr->sa_data))
4220 		return -EADDRNOTAVAIL;
4221 
4222 	spin_lock_bh(&adapter->mbx_lock);
4223 
4224 	err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4225 
4226 	spin_unlock_bh(&adapter->mbx_lock);
4227 
4228 	if (err)
4229 		return -EPERM;
4230 
4231 	ether_addr_copy(hw->mac.addr, addr->sa_data);
4232 	ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4233 	ether_addr_copy(netdev->dev_addr, addr->sa_data);
4234 
4235 	return 0;
4236 }
4237 
4238 /**
4239  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4240  * @netdev: network interface device structure
4241  * @new_mtu: new value for maximum frame size
4242  *
4243  * Returns 0 on success, negative on failure
4244  **/
4245 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4246 {
4247 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4248 	struct ixgbe_hw *hw = &adapter->hw;
4249 	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4250 	int ret;
4251 
4252 	/* prevent MTU being changed to a size unsupported by XDP */
4253 	if (adapter->xdp_prog) {
4254 		dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4255 		return -EPERM;
4256 	}
4257 
4258 	spin_lock_bh(&adapter->mbx_lock);
4259 	/* notify the PF of our intent to use this size of frame */
4260 	ret = hw->mac.ops.set_rlpml(hw, max_frame);
4261 	spin_unlock_bh(&adapter->mbx_lock);
4262 	if (ret)
4263 		return -EINVAL;
4264 
4265 	hw_dbg(hw, "changing MTU from %d to %d\n",
4266 	       netdev->mtu, new_mtu);
4267 
4268 	/* must set new MTU before calling down or up */
4269 	netdev->mtu = new_mtu;
4270 
4271 	if (netif_running(netdev))
4272 		ixgbevf_reinit_locked(adapter);
4273 
4274 	return 0;
4275 }
4276 
4277 static int __maybe_unused ixgbevf_suspend(struct device *dev_d)
4278 {
4279 	struct net_device *netdev = dev_get_drvdata(dev_d);
4280 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4281 
4282 	rtnl_lock();
4283 	netif_device_detach(netdev);
4284 
4285 	if (netif_running(netdev))
4286 		ixgbevf_close_suspend(adapter);
4287 
4288 	ixgbevf_clear_interrupt_scheme(adapter);
4289 	rtnl_unlock();
4290 
4291 	return 0;
4292 }
4293 
4294 static int __maybe_unused ixgbevf_resume(struct device *dev_d)
4295 {
4296 	struct pci_dev *pdev = to_pci_dev(dev_d);
4297 	struct net_device *netdev = pci_get_drvdata(pdev);
4298 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4299 	u32 err;
4300 
4301 	adapter->hw.hw_addr = adapter->io_addr;
4302 	smp_mb__before_atomic();
4303 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4304 	pci_set_master(pdev);
4305 
4306 	ixgbevf_reset(adapter);
4307 
4308 	rtnl_lock();
4309 	err = ixgbevf_init_interrupt_scheme(adapter);
4310 	if (!err && netif_running(netdev))
4311 		err = ixgbevf_open(netdev);
4312 	rtnl_unlock();
4313 	if (err)
4314 		return err;
4315 
4316 	netif_device_attach(netdev);
4317 
4318 	return err;
4319 }
4320 
4321 static void ixgbevf_shutdown(struct pci_dev *pdev)
4322 {
4323 	ixgbevf_suspend(&pdev->dev);
4324 }
4325 
4326 static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4327 				      const struct ixgbevf_ring *ring)
4328 {
4329 	u64 bytes, packets;
4330 	unsigned int start;
4331 
4332 	if (ring) {
4333 		do {
4334 			start = u64_stats_fetch_begin_irq(&ring->syncp);
4335 			bytes = ring->stats.bytes;
4336 			packets = ring->stats.packets;
4337 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4338 		stats->tx_bytes += bytes;
4339 		stats->tx_packets += packets;
4340 	}
4341 }
4342 
4343 static void ixgbevf_get_stats(struct net_device *netdev,
4344 			      struct rtnl_link_stats64 *stats)
4345 {
4346 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4347 	unsigned int start;
4348 	u64 bytes, packets;
4349 	const struct ixgbevf_ring *ring;
4350 	int i;
4351 
4352 	ixgbevf_update_stats(adapter);
4353 
4354 	stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4355 
4356 	rcu_read_lock();
4357 	for (i = 0; i < adapter->num_rx_queues; i++) {
4358 		ring = adapter->rx_ring[i];
4359 		do {
4360 			start = u64_stats_fetch_begin_irq(&ring->syncp);
4361 			bytes = ring->stats.bytes;
4362 			packets = ring->stats.packets;
4363 		} while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4364 		stats->rx_bytes += bytes;
4365 		stats->rx_packets += packets;
4366 	}
4367 
4368 	for (i = 0; i < adapter->num_tx_queues; i++) {
4369 		ring = adapter->tx_ring[i];
4370 		ixgbevf_get_tx_ring_stats(stats, ring);
4371 	}
4372 
4373 	for (i = 0; i < adapter->num_xdp_queues; i++) {
4374 		ring = adapter->xdp_ring[i];
4375 		ixgbevf_get_tx_ring_stats(stats, ring);
4376 	}
4377 	rcu_read_unlock();
4378 }
4379 
4380 #define IXGBEVF_MAX_MAC_HDR_LEN		127
4381 #define IXGBEVF_MAX_NETWORK_HDR_LEN	511
4382 
4383 static netdev_features_t
4384 ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4385 		       netdev_features_t features)
4386 {
4387 	unsigned int network_hdr_len, mac_hdr_len;
4388 
4389 	/* Make certain the headers can be described by a context descriptor */
4390 	mac_hdr_len = skb_network_header(skb) - skb->data;
4391 	if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4392 		return features & ~(NETIF_F_HW_CSUM |
4393 				    NETIF_F_SCTP_CRC |
4394 				    NETIF_F_HW_VLAN_CTAG_TX |
4395 				    NETIF_F_TSO |
4396 				    NETIF_F_TSO6);
4397 
4398 	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4399 	if (unlikely(network_hdr_len >  IXGBEVF_MAX_NETWORK_HDR_LEN))
4400 		return features & ~(NETIF_F_HW_CSUM |
4401 				    NETIF_F_SCTP_CRC |
4402 				    NETIF_F_TSO |
4403 				    NETIF_F_TSO6);
4404 
4405 	/* We can only support IPV4 TSO in tunnels if we can mangle the
4406 	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4407 	 */
4408 	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4409 		features &= ~NETIF_F_TSO;
4410 
4411 	return features;
4412 }
4413 
4414 static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4415 {
4416 	int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4417 	struct ixgbevf_adapter *adapter = netdev_priv(dev);
4418 	struct bpf_prog *old_prog;
4419 
4420 	/* verify ixgbevf ring attributes are sufficient for XDP */
4421 	for (i = 0; i < adapter->num_rx_queues; i++) {
4422 		struct ixgbevf_ring *ring = adapter->rx_ring[i];
4423 
4424 		if (frame_size > ixgbevf_rx_bufsz(ring))
4425 			return -EINVAL;
4426 	}
4427 
4428 	old_prog = xchg(&adapter->xdp_prog, prog);
4429 
4430 	/* If transitioning XDP modes reconfigure rings */
4431 	if (!!prog != !!old_prog) {
4432 		/* Hardware has to reinitialize queues and interrupts to
4433 		 * match packet buffer alignment. Unfortunately, the
4434 		 * hardware is not flexible enough to do this dynamically.
4435 		 */
4436 		if (netif_running(dev))
4437 			ixgbevf_close(dev);
4438 
4439 		ixgbevf_clear_interrupt_scheme(adapter);
4440 		ixgbevf_init_interrupt_scheme(adapter);
4441 
4442 		if (netif_running(dev))
4443 			ixgbevf_open(dev);
4444 	} else {
4445 		for (i = 0; i < adapter->num_rx_queues; i++)
4446 			xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4447 	}
4448 
4449 	if (old_prog)
4450 		bpf_prog_put(old_prog);
4451 
4452 	return 0;
4453 }
4454 
4455 static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4456 {
4457 	switch (xdp->command) {
4458 	case XDP_SETUP_PROG:
4459 		return ixgbevf_xdp_setup(dev, xdp->prog);
4460 	default:
4461 		return -EINVAL;
4462 	}
4463 }
4464 
4465 static const struct net_device_ops ixgbevf_netdev_ops = {
4466 	.ndo_open		= ixgbevf_open,
4467 	.ndo_stop		= ixgbevf_close,
4468 	.ndo_start_xmit		= ixgbevf_xmit_frame,
4469 	.ndo_set_rx_mode	= ixgbevf_set_rx_mode,
4470 	.ndo_get_stats64	= ixgbevf_get_stats,
4471 	.ndo_validate_addr	= eth_validate_addr,
4472 	.ndo_set_mac_address	= ixgbevf_set_mac,
4473 	.ndo_change_mtu		= ixgbevf_change_mtu,
4474 	.ndo_tx_timeout		= ixgbevf_tx_timeout,
4475 	.ndo_vlan_rx_add_vid	= ixgbevf_vlan_rx_add_vid,
4476 	.ndo_vlan_rx_kill_vid	= ixgbevf_vlan_rx_kill_vid,
4477 	.ndo_features_check	= ixgbevf_features_check,
4478 	.ndo_bpf		= ixgbevf_xdp,
4479 };
4480 
4481 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4482 {
4483 	dev->netdev_ops = &ixgbevf_netdev_ops;
4484 	ixgbevf_set_ethtool_ops(dev);
4485 	dev->watchdog_timeo = 5 * HZ;
4486 }
4487 
4488 /**
4489  * ixgbevf_probe - Device Initialization Routine
4490  * @pdev: PCI device information struct
4491  * @ent: entry in ixgbevf_pci_tbl
4492  *
4493  * Returns 0 on success, negative on failure
4494  *
4495  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4496  * The OS initialization, configuring of the adapter private structure,
4497  * and a hardware reset occur.
4498  **/
4499 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4500 {
4501 	struct net_device *netdev;
4502 	struct ixgbevf_adapter *adapter = NULL;
4503 	struct ixgbe_hw *hw = NULL;
4504 	const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4505 	int err, pci_using_dac;
4506 	bool disable_dev = false;
4507 
4508 	err = pci_enable_device(pdev);
4509 	if (err)
4510 		return err;
4511 
4512 	if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4513 		pci_using_dac = 1;
4514 	} else {
4515 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4516 		if (err) {
4517 			dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4518 			goto err_dma;
4519 		}
4520 		pci_using_dac = 0;
4521 	}
4522 
4523 	err = pci_request_regions(pdev, ixgbevf_driver_name);
4524 	if (err) {
4525 		dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4526 		goto err_pci_reg;
4527 	}
4528 
4529 	pci_set_master(pdev);
4530 
4531 	netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4532 				   MAX_TX_QUEUES);
4533 	if (!netdev) {
4534 		err = -ENOMEM;
4535 		goto err_alloc_etherdev;
4536 	}
4537 
4538 	SET_NETDEV_DEV(netdev, &pdev->dev);
4539 
4540 	adapter = netdev_priv(netdev);
4541 
4542 	adapter->netdev = netdev;
4543 	adapter->pdev = pdev;
4544 	hw = &adapter->hw;
4545 	hw->back = adapter;
4546 	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4547 
4548 	/* call save state here in standalone driver because it relies on
4549 	 * adapter struct to exist, and needs to call netdev_priv
4550 	 */
4551 	pci_save_state(pdev);
4552 
4553 	hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4554 			      pci_resource_len(pdev, 0));
4555 	adapter->io_addr = hw->hw_addr;
4556 	if (!hw->hw_addr) {
4557 		err = -EIO;
4558 		goto err_ioremap;
4559 	}
4560 
4561 	ixgbevf_assign_netdev_ops(netdev);
4562 
4563 	/* Setup HW API */
4564 	memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4565 	hw->mac.type  = ii->mac;
4566 
4567 	memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4568 	       sizeof(struct ixgbe_mbx_operations));
4569 
4570 	/* setup the private structure */
4571 	err = ixgbevf_sw_init(adapter);
4572 	if (err)
4573 		goto err_sw_init;
4574 
4575 	/* The HW MAC address was set and/or determined in sw_init */
4576 	if (!is_valid_ether_addr(netdev->dev_addr)) {
4577 		pr_err("invalid MAC address\n");
4578 		err = -EIO;
4579 		goto err_sw_init;
4580 	}
4581 
4582 	netdev->hw_features = NETIF_F_SG |
4583 			      NETIF_F_TSO |
4584 			      NETIF_F_TSO6 |
4585 			      NETIF_F_RXCSUM |
4586 			      NETIF_F_HW_CSUM |
4587 			      NETIF_F_SCTP_CRC;
4588 
4589 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4590 				      NETIF_F_GSO_GRE_CSUM | \
4591 				      NETIF_F_GSO_IPXIP4 | \
4592 				      NETIF_F_GSO_IPXIP6 | \
4593 				      NETIF_F_GSO_UDP_TUNNEL | \
4594 				      NETIF_F_GSO_UDP_TUNNEL_CSUM)
4595 
4596 	netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4597 	netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4598 			       IXGBEVF_GSO_PARTIAL_FEATURES;
4599 
4600 	netdev->features = netdev->hw_features;
4601 
4602 	if (pci_using_dac)
4603 		netdev->features |= NETIF_F_HIGHDMA;
4604 
4605 	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4606 	netdev->mpls_features |= NETIF_F_SG |
4607 				 NETIF_F_TSO |
4608 				 NETIF_F_TSO6 |
4609 				 NETIF_F_HW_CSUM;
4610 	netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4611 	netdev->hw_enc_features |= netdev->vlan_features;
4612 
4613 	/* set this bit last since it cannot be part of vlan_features */
4614 	netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4615 			    NETIF_F_HW_VLAN_CTAG_RX |
4616 			    NETIF_F_HW_VLAN_CTAG_TX;
4617 
4618 	netdev->priv_flags |= IFF_UNICAST_FLT;
4619 
4620 	/* MTU range: 68 - 1504 or 9710 */
4621 	netdev->min_mtu = ETH_MIN_MTU;
4622 	switch (adapter->hw.api_version) {
4623 	case ixgbe_mbox_api_11:
4624 	case ixgbe_mbox_api_12:
4625 	case ixgbe_mbox_api_13:
4626 	case ixgbe_mbox_api_14:
4627 		netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4628 				  (ETH_HLEN + ETH_FCS_LEN);
4629 		break;
4630 	default:
4631 		if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4632 			netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4633 					  (ETH_HLEN + ETH_FCS_LEN);
4634 		else
4635 			netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4636 		break;
4637 	}
4638 
4639 	if (IXGBE_REMOVED(hw->hw_addr)) {
4640 		err = -EIO;
4641 		goto err_sw_init;
4642 	}
4643 
4644 	timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4645 
4646 	INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4647 	set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4648 	clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4649 
4650 	err = ixgbevf_init_interrupt_scheme(adapter);
4651 	if (err)
4652 		goto err_sw_init;
4653 
4654 	strcpy(netdev->name, "eth%d");
4655 
4656 	err = register_netdev(netdev);
4657 	if (err)
4658 		goto err_register;
4659 
4660 	pci_set_drvdata(pdev, netdev);
4661 	netif_carrier_off(netdev);
4662 	ixgbevf_init_ipsec_offload(adapter);
4663 
4664 	ixgbevf_init_last_counter_stats(adapter);
4665 
4666 	/* print the VF info */
4667 	dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4668 	dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4669 
4670 	switch (hw->mac.type) {
4671 	case ixgbe_mac_X550_vf:
4672 		dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4673 		break;
4674 	case ixgbe_mac_X540_vf:
4675 		dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4676 		break;
4677 	case ixgbe_mac_82599_vf:
4678 	default:
4679 		dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4680 		break;
4681 	}
4682 
4683 	return 0;
4684 
4685 err_register:
4686 	ixgbevf_clear_interrupt_scheme(adapter);
4687 err_sw_init:
4688 	ixgbevf_reset_interrupt_capability(adapter);
4689 	iounmap(adapter->io_addr);
4690 	kfree(adapter->rss_key);
4691 err_ioremap:
4692 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4693 	free_netdev(netdev);
4694 err_alloc_etherdev:
4695 	pci_release_regions(pdev);
4696 err_pci_reg:
4697 err_dma:
4698 	if (!adapter || disable_dev)
4699 		pci_disable_device(pdev);
4700 	return err;
4701 }
4702 
4703 /**
4704  * ixgbevf_remove - Device Removal Routine
4705  * @pdev: PCI device information struct
4706  *
4707  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4708  * that it should release a PCI device.  The could be caused by a
4709  * Hot-Plug event, or because the driver is going to be removed from
4710  * memory.
4711  **/
4712 static void ixgbevf_remove(struct pci_dev *pdev)
4713 {
4714 	struct net_device *netdev = pci_get_drvdata(pdev);
4715 	struct ixgbevf_adapter *adapter;
4716 	bool disable_dev;
4717 
4718 	if (!netdev)
4719 		return;
4720 
4721 	adapter = netdev_priv(netdev);
4722 
4723 	set_bit(__IXGBEVF_REMOVING, &adapter->state);
4724 	cancel_work_sync(&adapter->service_task);
4725 
4726 	if (netdev->reg_state == NETREG_REGISTERED)
4727 		unregister_netdev(netdev);
4728 
4729 	ixgbevf_stop_ipsec_offload(adapter);
4730 	ixgbevf_clear_interrupt_scheme(adapter);
4731 	ixgbevf_reset_interrupt_capability(adapter);
4732 
4733 	iounmap(adapter->io_addr);
4734 	pci_release_regions(pdev);
4735 
4736 	hw_dbg(&adapter->hw, "Remove complete\n");
4737 
4738 	kfree(adapter->rss_key);
4739 	disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4740 	free_netdev(netdev);
4741 
4742 	if (disable_dev)
4743 		pci_disable_device(pdev);
4744 }
4745 
4746 /**
4747  * ixgbevf_io_error_detected - called when PCI error is detected
4748  * @pdev: Pointer to PCI device
4749  * @state: The current pci connection state
4750  *
4751  * This function is called after a PCI bus error affecting
4752  * this device has been detected.
4753  **/
4754 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4755 						  pci_channel_state_t state)
4756 {
4757 	struct net_device *netdev = pci_get_drvdata(pdev);
4758 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4759 
4760 	if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4761 		return PCI_ERS_RESULT_DISCONNECT;
4762 
4763 	rtnl_lock();
4764 	netif_device_detach(netdev);
4765 
4766 	if (netif_running(netdev))
4767 		ixgbevf_close_suspend(adapter);
4768 
4769 	if (state == pci_channel_io_perm_failure) {
4770 		rtnl_unlock();
4771 		return PCI_ERS_RESULT_DISCONNECT;
4772 	}
4773 
4774 	if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4775 		pci_disable_device(pdev);
4776 	rtnl_unlock();
4777 
4778 	/* Request a slot slot reset. */
4779 	return PCI_ERS_RESULT_NEED_RESET;
4780 }
4781 
4782 /**
4783  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4784  * @pdev: Pointer to PCI device
4785  *
4786  * Restart the card from scratch, as if from a cold-boot. Implementation
4787  * resembles the first-half of the ixgbevf_resume routine.
4788  **/
4789 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4790 {
4791 	struct net_device *netdev = pci_get_drvdata(pdev);
4792 	struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4793 
4794 	if (pci_enable_device_mem(pdev)) {
4795 		dev_err(&pdev->dev,
4796 			"Cannot re-enable PCI device after reset.\n");
4797 		return PCI_ERS_RESULT_DISCONNECT;
4798 	}
4799 
4800 	adapter->hw.hw_addr = adapter->io_addr;
4801 	smp_mb__before_atomic();
4802 	clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4803 	pci_set_master(pdev);
4804 
4805 	ixgbevf_reset(adapter);
4806 
4807 	return PCI_ERS_RESULT_RECOVERED;
4808 }
4809 
4810 /**
4811  * ixgbevf_io_resume - called when traffic can start flowing again.
4812  * @pdev: Pointer to PCI device
4813  *
4814  * This callback is called when the error recovery driver tells us that
4815  * its OK to resume normal operation. Implementation resembles the
4816  * second-half of the ixgbevf_resume routine.
4817  **/
4818 static void ixgbevf_io_resume(struct pci_dev *pdev)
4819 {
4820 	struct net_device *netdev = pci_get_drvdata(pdev);
4821 
4822 	rtnl_lock();
4823 	if (netif_running(netdev))
4824 		ixgbevf_open(netdev);
4825 
4826 	netif_device_attach(netdev);
4827 	rtnl_unlock();
4828 }
4829 
4830 /* PCI Error Recovery (ERS) */
4831 static const struct pci_error_handlers ixgbevf_err_handler = {
4832 	.error_detected = ixgbevf_io_error_detected,
4833 	.slot_reset = ixgbevf_io_slot_reset,
4834 	.resume = ixgbevf_io_resume,
4835 };
4836 
4837 static SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4838 
4839 static struct pci_driver ixgbevf_driver = {
4840 	.name		= ixgbevf_driver_name,
4841 	.id_table	= ixgbevf_pci_tbl,
4842 	.probe		= ixgbevf_probe,
4843 	.remove		= ixgbevf_remove,
4844 
4845 	/* Power Management Hooks */
4846 	.driver.pm	= &ixgbevf_pm_ops,
4847 
4848 	.shutdown	= ixgbevf_shutdown,
4849 	.err_handler	= &ixgbevf_err_handler
4850 };
4851 
4852 /**
4853  * ixgbevf_init_module - Driver Registration Routine
4854  *
4855  * ixgbevf_init_module is the first routine called when the driver is
4856  * loaded. All it does is register with the PCI subsystem.
4857  **/
4858 static int __init ixgbevf_init_module(void)
4859 {
4860 	pr_info("%s\n", ixgbevf_driver_string);
4861 	pr_info("%s\n", ixgbevf_copyright);
4862 	ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4863 	if (!ixgbevf_wq) {
4864 		pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4865 		return -ENOMEM;
4866 	}
4867 
4868 	return pci_register_driver(&ixgbevf_driver);
4869 }
4870 
4871 module_init(ixgbevf_init_module);
4872 
4873 /**
4874  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4875  *
4876  * ixgbevf_exit_module is called just before the driver is removed
4877  * from memory.
4878  **/
4879 static void __exit ixgbevf_exit_module(void)
4880 {
4881 	pci_unregister_driver(&ixgbevf_driver);
4882 	if (ixgbevf_wq) {
4883 		destroy_workqueue(ixgbevf_wq);
4884 		ixgbevf_wq = NULL;
4885 	}
4886 }
4887 
4888 #ifdef DEBUG
4889 /**
4890  * ixgbevf_get_hw_dev_name - return device name string
4891  * used by hardware layer to print debugging information
4892  * @hw: pointer to private hardware struct
4893  **/
4894 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4895 {
4896 	struct ixgbevf_adapter *adapter = hw->back;
4897 
4898 	return adapter->netdev->name;
4899 }
4900 
4901 #endif
4902 module_exit(ixgbevf_exit_module);
4903 
4904 /* ixgbevf_main.c */
4905