[go: nahoru, domu]

1/*******************************************************************************
2 *
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
4 * Copyright(c) 2013 - 2014 Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program.  If not, see <http://www.gnu.org/licenses/>.
17 *
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
20 *
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
24 *
25 ******************************************************************************/
26
27#include <linux/prefetch.h>
28
29#include "i40evf.h"
30#include "i40e_prototype.h"
31
32static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
33				u32 td_tag)
34{
35	return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
36			   ((u64)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
37			   ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
38			   ((u64)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
39			   ((u64)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
40}
41
42#define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
43
44/**
45 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
46 * @ring:      the ring that owns the buffer
47 * @tx_buffer: the buffer to free
48 **/
49static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
50					    struct i40e_tx_buffer *tx_buffer)
51{
52	if (tx_buffer->skb) {
53		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
54			kfree(tx_buffer->raw_buf);
55		else
56			dev_kfree_skb_any(tx_buffer->skb);
57
58		if (dma_unmap_len(tx_buffer, len))
59			dma_unmap_single(ring->dev,
60					 dma_unmap_addr(tx_buffer, dma),
61					 dma_unmap_len(tx_buffer, len),
62					 DMA_TO_DEVICE);
63	} else if (dma_unmap_len(tx_buffer, len)) {
64		dma_unmap_page(ring->dev,
65			       dma_unmap_addr(tx_buffer, dma),
66			       dma_unmap_len(tx_buffer, len),
67			       DMA_TO_DEVICE);
68	}
69	tx_buffer->next_to_watch = NULL;
70	tx_buffer->skb = NULL;
71	dma_unmap_len_set(tx_buffer, len, 0);
72	/* tx_buffer must be completely set up in the transmit path */
73}
74
75/**
76 * i40evf_clean_tx_ring - Free any empty Tx buffers
77 * @tx_ring: ring to be cleaned
78 **/
79void i40evf_clean_tx_ring(struct i40e_ring *tx_ring)
80{
81	unsigned long bi_size;
82	u16 i;
83
84	/* ring already cleared, nothing to do */
85	if (!tx_ring->tx_bi)
86		return;
87
88	/* Free all the Tx ring sk_buffs */
89	for (i = 0; i < tx_ring->count; i++)
90		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
91
92	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
93	memset(tx_ring->tx_bi, 0, bi_size);
94
95	/* Zero out the descriptor ring */
96	memset(tx_ring->desc, 0, tx_ring->size);
97
98	tx_ring->next_to_use = 0;
99	tx_ring->next_to_clean = 0;
100
101	if (!tx_ring->netdev)
102		return;
103
104	/* cleanup Tx queue statistics */
105	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
106						  tx_ring->queue_index));
107}
108
109/**
110 * i40evf_free_tx_resources - Free Tx resources per queue
111 * @tx_ring: Tx descriptor ring for a specific queue
112 *
113 * Free all transmit software resources
114 **/
115void i40evf_free_tx_resources(struct i40e_ring *tx_ring)
116{
117	i40evf_clean_tx_ring(tx_ring);
118	kfree(tx_ring->tx_bi);
119	tx_ring->tx_bi = NULL;
120
121	if (tx_ring->desc) {
122		dma_free_coherent(tx_ring->dev, tx_ring->size,
123				  tx_ring->desc, tx_ring->dma);
124		tx_ring->desc = NULL;
125	}
126}
127
128/**
129 * i40e_get_tx_pending - how many tx descriptors not processed
130 * @tx_ring: the ring of descriptors
131 *
132 * Since there is no access to the ring head register
133 * in XL710, we need to use our local copies
134 **/
135static u32 i40e_get_tx_pending(struct i40e_ring *ring)
136{
137	u32 ntu = ((ring->next_to_clean <= ring->next_to_use)
138			? ring->next_to_use
139			: ring->next_to_use + ring->count);
140	return ntu - ring->next_to_clean;
141}
142
143/**
144 * i40e_check_tx_hang - Is there a hang in the Tx queue
145 * @tx_ring: the ring of descriptors
146 **/
147static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
148{
149	u32 tx_pending = i40e_get_tx_pending(tx_ring);
150	bool ret = false;
151
152	clear_check_for_tx_hang(tx_ring);
153
154	/* Check for a hung queue, but be thorough. This verifies
155	 * that a transmit has been completed since the previous
156	 * check AND there is at least one packet pending. The
157	 * ARMED bit is set to indicate a potential hang. The
158	 * bit is cleared if a pause frame is received to remove
159	 * false hang detection due to PFC or 802.3x frames. By
160	 * requiring this to fail twice we avoid races with
161	 * PFC clearing the ARMED bit and conditions where we
162	 * run the check_tx_hang logic with a transmit completion
163	 * pending but without time to complete it yet.
164	 */
165	if ((tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) &&
166	    (tx_pending >= I40E_MIN_DESC_PENDING)) {
167		/* make sure it is true for two checks in a row */
168		ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED,
169				       &tx_ring->state);
170	} else if (!(tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) ||
171		   !(tx_pending < I40E_MIN_DESC_PENDING) ||
172		   !(tx_pending > 0)) {
173		/* update completed stats and disarm the hang check */
174		tx_ring->tx_stats.tx_done_old = tx_ring->stats.packets;
175		clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state);
176	}
177
178	return ret;
179}
180
181/**
182 * i40e_get_head - Retrieve head from head writeback
183 * @tx_ring:  tx ring to fetch head of
184 *
185 * Returns value of Tx ring head based on value stored
186 * in head write-back location
187 **/
188static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
189{
190	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
191
192	return le32_to_cpu(*(volatile __le32 *)head);
193}
194
195/**
196 * i40e_clean_tx_irq - Reclaim resources after transmit completes
197 * @tx_ring:  tx ring to clean
198 * @budget:   how many cleans we're allowed
199 *
200 * Returns true if there's any budget left (e.g. the clean is finished)
201 **/
202static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
203{
204	u16 i = tx_ring->next_to_clean;
205	struct i40e_tx_buffer *tx_buf;
206	struct i40e_tx_desc *tx_head;
207	struct i40e_tx_desc *tx_desc;
208	unsigned int total_packets = 0;
209	unsigned int total_bytes = 0;
210
211	tx_buf = &tx_ring->tx_bi[i];
212	tx_desc = I40E_TX_DESC(tx_ring, i);
213	i -= tx_ring->count;
214
215	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
216
217	do {
218		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
219
220		/* if next_to_watch is not set then there is no work pending */
221		if (!eop_desc)
222			break;
223
224		/* prevent any other reads prior to eop_desc */
225		read_barrier_depends();
226
227		/* we have caught up to head, no work left to do */
228		if (tx_head == tx_desc)
229			break;
230
231		/* clear next_to_watch to prevent false hangs */
232		tx_buf->next_to_watch = NULL;
233
234		/* update the statistics for this packet */
235		total_bytes += tx_buf->bytecount;
236		total_packets += tx_buf->gso_segs;
237
238		/* free the skb */
239		dev_kfree_skb_any(tx_buf->skb);
240
241		/* unmap skb header data */
242		dma_unmap_single(tx_ring->dev,
243				 dma_unmap_addr(tx_buf, dma),
244				 dma_unmap_len(tx_buf, len),
245				 DMA_TO_DEVICE);
246
247		/* clear tx_buffer data */
248		tx_buf->skb = NULL;
249		dma_unmap_len_set(tx_buf, len, 0);
250
251		/* unmap remaining buffers */
252		while (tx_desc != eop_desc) {
253
254			tx_buf++;
255			tx_desc++;
256			i++;
257			if (unlikely(!i)) {
258				i -= tx_ring->count;
259				tx_buf = tx_ring->tx_bi;
260				tx_desc = I40E_TX_DESC(tx_ring, 0);
261			}
262
263			/* unmap any remaining paged data */
264			if (dma_unmap_len(tx_buf, len)) {
265				dma_unmap_page(tx_ring->dev,
266					       dma_unmap_addr(tx_buf, dma),
267					       dma_unmap_len(tx_buf, len),
268					       DMA_TO_DEVICE);
269				dma_unmap_len_set(tx_buf, len, 0);
270			}
271		}
272
273		/* move us one more past the eop_desc for start of next pkt */
274		tx_buf++;
275		tx_desc++;
276		i++;
277		if (unlikely(!i)) {
278			i -= tx_ring->count;
279			tx_buf = tx_ring->tx_bi;
280			tx_desc = I40E_TX_DESC(tx_ring, 0);
281		}
282
283		/* update budget accounting */
284		budget--;
285	} while (likely(budget));
286
287	i += tx_ring->count;
288	tx_ring->next_to_clean = i;
289	u64_stats_update_begin(&tx_ring->syncp);
290	tx_ring->stats.bytes += total_bytes;
291	tx_ring->stats.packets += total_packets;
292	u64_stats_update_end(&tx_ring->syncp);
293	tx_ring->q_vector->tx.total_bytes += total_bytes;
294	tx_ring->q_vector->tx.total_packets += total_packets;
295
296	if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) {
297		/* schedule immediate reset if we believe we hung */
298		dev_info(tx_ring->dev, "Detected Tx Unit Hang\n"
299			 "  VSI                  <%d>\n"
300			 "  Tx Queue             <%d>\n"
301			 "  next_to_use          <%x>\n"
302			 "  next_to_clean        <%x>\n",
303			 tx_ring->vsi->seid,
304			 tx_ring->queue_index,
305			 tx_ring->next_to_use, i);
306		dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n"
307			 "  time_stamp           <%lx>\n"
308			 "  jiffies              <%lx>\n",
309			 tx_ring->tx_bi[i].time_stamp, jiffies);
310
311		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
312
313		dev_info(tx_ring->dev,
314			 "tx hang detected on queue %d, resetting adapter\n",
315			 tx_ring->queue_index);
316
317		tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev);
318
319		/* the adapter is about to reset, no point in enabling stuff */
320		return true;
321	}
322
323	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
324						      tx_ring->queue_index),
325				  total_packets, total_bytes);
326
327#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
328	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
329		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
330		/* Make sure that anybody stopping the queue after this
331		 * sees the new next_to_clean.
332		 */
333		smp_mb();
334		if (__netif_subqueue_stopped(tx_ring->netdev,
335					     tx_ring->queue_index) &&
336		   !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
337			netif_wake_subqueue(tx_ring->netdev,
338					    tx_ring->queue_index);
339			++tx_ring->tx_stats.restart_queue;
340		}
341	}
342
343	return budget > 0;
344}
345
346/**
347 * i40e_set_new_dynamic_itr - Find new ITR level
348 * @rc: structure containing ring performance data
349 *
350 * Stores a new ITR value based on packets and byte counts during
351 * the last interrupt.  The advantage of per interrupt computation
352 * is faster updates and more accurate ITR for the current traffic
353 * pattern.  Constants in this function were computed based on
354 * theoretical maximum wire speed and thresholds were set based on
355 * testing data as well as attempting to minimize response time
356 * while increasing bulk throughput.
357 **/
358static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
359{
360	enum i40e_latency_range new_latency_range = rc->latency_range;
361	u32 new_itr = rc->itr;
362	int bytes_per_int;
363
364	if (rc->total_packets == 0 || !rc->itr)
365		return;
366
367	/* simple throttlerate management
368	 *   0-10MB/s   lowest (100000 ints/s)
369	 *  10-20MB/s   low    (20000 ints/s)
370	 *  20-1249MB/s bulk   (8000 ints/s)
371	 */
372	bytes_per_int = rc->total_bytes / rc->itr;
373	switch (rc->itr) {
374	case I40E_LOWEST_LATENCY:
375		if (bytes_per_int > 10)
376			new_latency_range = I40E_LOW_LATENCY;
377		break;
378	case I40E_LOW_LATENCY:
379		if (bytes_per_int > 20)
380			new_latency_range = I40E_BULK_LATENCY;
381		else if (bytes_per_int <= 10)
382			new_latency_range = I40E_LOWEST_LATENCY;
383		break;
384	case I40E_BULK_LATENCY:
385		if (bytes_per_int <= 20)
386			rc->latency_range = I40E_LOW_LATENCY;
387		break;
388	}
389
390	switch (new_latency_range) {
391	case I40E_LOWEST_LATENCY:
392		new_itr = I40E_ITR_100K;
393		break;
394	case I40E_LOW_LATENCY:
395		new_itr = I40E_ITR_20K;
396		break;
397	case I40E_BULK_LATENCY:
398		new_itr = I40E_ITR_8K;
399		break;
400	default:
401		break;
402	}
403
404	if (new_itr != rc->itr) {
405		/* do an exponential smoothing */
406		new_itr = (10 * new_itr * rc->itr) /
407			  ((9 * new_itr) + rc->itr);
408		rc->itr = new_itr & I40E_MAX_ITR;
409	}
410
411	rc->total_bytes = 0;
412	rc->total_packets = 0;
413}
414
415/**
416 * i40e_update_dynamic_itr - Adjust ITR based on bytes per int
417 * @q_vector: the vector to adjust
418 **/
419static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector)
420{
421	u16 vector = q_vector->vsi->base_vector + q_vector->v_idx;
422	struct i40e_hw *hw = &q_vector->vsi->back->hw;
423	u32 reg_addr;
424	u16 old_itr;
425
426	reg_addr = I40E_VFINT_ITRN1(I40E_RX_ITR, vector - 1);
427	old_itr = q_vector->rx.itr;
428	i40e_set_new_dynamic_itr(&q_vector->rx);
429	if (old_itr != q_vector->rx.itr)
430		wr32(hw, reg_addr, q_vector->rx.itr);
431
432	reg_addr = I40E_VFINT_ITRN1(I40E_TX_ITR, vector - 1);
433	old_itr = q_vector->tx.itr;
434	i40e_set_new_dynamic_itr(&q_vector->tx);
435	if (old_itr != q_vector->tx.itr)
436		wr32(hw, reg_addr, q_vector->tx.itr);
437}
438
439/**
440 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
441 * @tx_ring: the tx ring to set up
442 *
443 * Return 0 on success, negative on error
444 **/
445int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
446{
447	struct device *dev = tx_ring->dev;
448	int bi_size;
449
450	if (!dev)
451		return -ENOMEM;
452
453	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
454	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
455	if (!tx_ring->tx_bi)
456		goto err;
457
458	/* round up to nearest 4K */
459	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
460	/* add u32 for head writeback, align after this takes care of
461	 * guaranteeing this is at least one cache line in size
462	 */
463	tx_ring->size += sizeof(u32);
464	tx_ring->size = ALIGN(tx_ring->size, 4096);
465	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
466					   &tx_ring->dma, GFP_KERNEL);
467	if (!tx_ring->desc) {
468		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
469			 tx_ring->size);
470		goto err;
471	}
472
473	tx_ring->next_to_use = 0;
474	tx_ring->next_to_clean = 0;
475	return 0;
476
477err:
478	kfree(tx_ring->tx_bi);
479	tx_ring->tx_bi = NULL;
480	return -ENOMEM;
481}
482
483/**
484 * i40evf_clean_rx_ring - Free Rx buffers
485 * @rx_ring: ring to be cleaned
486 **/
487void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
488{
489	struct device *dev = rx_ring->dev;
490	struct i40e_rx_buffer *rx_bi;
491	unsigned long bi_size;
492	u16 i;
493
494	/* ring already cleared, nothing to do */
495	if (!rx_ring->rx_bi)
496		return;
497
498	/* Free all the Rx ring sk_buffs */
499	for (i = 0; i < rx_ring->count; i++) {
500		rx_bi = &rx_ring->rx_bi[i];
501		if (rx_bi->dma) {
502			dma_unmap_single(dev,
503					 rx_bi->dma,
504					 rx_ring->rx_buf_len,
505					 DMA_FROM_DEVICE);
506			rx_bi->dma = 0;
507		}
508		if (rx_bi->skb) {
509			dev_kfree_skb(rx_bi->skb);
510			rx_bi->skb = NULL;
511		}
512		if (rx_bi->page) {
513			if (rx_bi->page_dma) {
514				dma_unmap_page(dev,
515					       rx_bi->page_dma,
516					       PAGE_SIZE / 2,
517					       DMA_FROM_DEVICE);
518				rx_bi->page_dma = 0;
519			}
520			__free_page(rx_bi->page);
521			rx_bi->page = NULL;
522			rx_bi->page_offset = 0;
523		}
524	}
525
526	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
527	memset(rx_ring->rx_bi, 0, bi_size);
528
529	/* Zero out the descriptor ring */
530	memset(rx_ring->desc, 0, rx_ring->size);
531
532	rx_ring->next_to_clean = 0;
533	rx_ring->next_to_use = 0;
534}
535
536/**
537 * i40evf_free_rx_resources - Free Rx resources
538 * @rx_ring: ring to clean the resources from
539 *
540 * Free all receive software resources
541 **/
542void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
543{
544	i40evf_clean_rx_ring(rx_ring);
545	kfree(rx_ring->rx_bi);
546	rx_ring->rx_bi = NULL;
547
548	if (rx_ring->desc) {
549		dma_free_coherent(rx_ring->dev, rx_ring->size,
550				  rx_ring->desc, rx_ring->dma);
551		rx_ring->desc = NULL;
552	}
553}
554
555/**
556 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
557 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
558 *
559 * Returns 0 on success, negative on failure
560 **/
561int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
562{
563	struct device *dev = rx_ring->dev;
564	int bi_size;
565
566	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
567	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
568	if (!rx_ring->rx_bi)
569		goto err;
570
571	/* Round up to nearest 4K */
572	rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
573		? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
574		: rx_ring->count * sizeof(union i40e_32byte_rx_desc);
575	rx_ring->size = ALIGN(rx_ring->size, 4096);
576	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
577					   &rx_ring->dma, GFP_KERNEL);
578
579	if (!rx_ring->desc) {
580		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
581			 rx_ring->size);
582		goto err;
583	}
584
585	rx_ring->next_to_clean = 0;
586	rx_ring->next_to_use = 0;
587
588	return 0;
589err:
590	kfree(rx_ring->rx_bi);
591	rx_ring->rx_bi = NULL;
592	return -ENOMEM;
593}
594
595/**
596 * i40e_release_rx_desc - Store the new tail and head values
597 * @rx_ring: ring to bump
598 * @val: new head index
599 **/
600static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
601{
602	rx_ring->next_to_use = val;
603	/* Force memory writes to complete before letting h/w
604	 * know there are new descriptors to fetch.  (Only
605	 * applicable for weak-ordered memory model archs,
606	 * such as IA-64).
607	 */
608	wmb();
609	writel(val, rx_ring->tail);
610}
611
612/**
613 * i40evf_alloc_rx_buffers - Replace used receive buffers; packet split
614 * @rx_ring: ring to place buffers on
615 * @cleaned_count: number of buffers to replace
616 **/
617void i40evf_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
618{
619	u16 i = rx_ring->next_to_use;
620	union i40e_rx_desc *rx_desc;
621	struct i40e_rx_buffer *bi;
622	struct sk_buff *skb;
623
624	/* do nothing if no valid netdev defined */
625	if (!rx_ring->netdev || !cleaned_count)
626		return;
627
628	while (cleaned_count--) {
629		rx_desc = I40E_RX_DESC(rx_ring, i);
630		bi = &rx_ring->rx_bi[i];
631		skb = bi->skb;
632
633		if (!skb) {
634			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
635							rx_ring->rx_buf_len);
636			if (!skb) {
637				rx_ring->rx_stats.alloc_buff_failed++;
638				goto no_buffers;
639			}
640			/* initialize queue mapping */
641			skb_record_rx_queue(skb, rx_ring->queue_index);
642			bi->skb = skb;
643		}
644
645		if (!bi->dma) {
646			bi->dma = dma_map_single(rx_ring->dev,
647						 skb->data,
648						 rx_ring->rx_buf_len,
649						 DMA_FROM_DEVICE);
650			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
651				rx_ring->rx_stats.alloc_buff_failed++;
652				bi->dma = 0;
653				goto no_buffers;
654			}
655		}
656
657		if (ring_is_ps_enabled(rx_ring)) {
658			if (!bi->page) {
659				bi->page = alloc_page(GFP_ATOMIC);
660				if (!bi->page) {
661					rx_ring->rx_stats.alloc_page_failed++;
662					goto no_buffers;
663				}
664			}
665
666			if (!bi->page_dma) {
667				/* use a half page if we're re-using */
668				bi->page_offset ^= PAGE_SIZE / 2;
669				bi->page_dma = dma_map_page(rx_ring->dev,
670							    bi->page,
671							    bi->page_offset,
672							    PAGE_SIZE / 2,
673							    DMA_FROM_DEVICE);
674				if (dma_mapping_error(rx_ring->dev,
675						      bi->page_dma)) {
676					rx_ring->rx_stats.alloc_page_failed++;
677					bi->page_dma = 0;
678					goto no_buffers;
679				}
680			}
681
682			/* Refresh the desc even if buffer_addrs didn't change
683			 * because each write-back erases this info.
684			 */
685			rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
686			rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
687		} else {
688			rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
689			rx_desc->read.hdr_addr = 0;
690		}
691		i++;
692		if (i == rx_ring->count)
693			i = 0;
694	}
695
696no_buffers:
697	if (rx_ring->next_to_use != i)
698		i40e_release_rx_desc(rx_ring, i);
699}
700
701/**
702 * i40e_receive_skb - Send a completed packet up the stack
703 * @rx_ring:  rx ring in play
704 * @skb: packet to send up
705 * @vlan_tag: vlan tag for packet
706 **/
707static void i40e_receive_skb(struct i40e_ring *rx_ring,
708			     struct sk_buff *skb, u16 vlan_tag)
709{
710	struct i40e_q_vector *q_vector = rx_ring->q_vector;
711	struct i40e_vsi *vsi = rx_ring->vsi;
712	u64 flags = vsi->back->flags;
713
714	if (vlan_tag & VLAN_VID_MASK)
715		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
716
717	if (flags & I40E_FLAG_IN_NETPOLL)
718		netif_rx(skb);
719	else
720		napi_gro_receive(&q_vector->napi, skb);
721}
722
723/**
724 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
725 * @vsi: the VSI we care about
726 * @skb: skb currently being received and modified
727 * @rx_status: status value of last descriptor in packet
728 * @rx_error: error value of last descriptor in packet
729 * @rx_ptype: ptype value of last descriptor in packet
730 **/
731static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
732				    struct sk_buff *skb,
733				    u32 rx_status,
734				    u32 rx_error,
735				    u16 rx_ptype)
736{
737	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
738	bool ipv4 = false, ipv6 = false;
739	bool ipv4_tunnel, ipv6_tunnel;
740	__wsum rx_udp_csum;
741	struct iphdr *iph;
742	__sum16 csum;
743
744	ipv4_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
745		      (rx_ptype < I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
746	ipv6_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
747		      (rx_ptype < I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
748
749	skb->ip_summed = CHECKSUM_NONE;
750
751	/* Rx csum enabled and ip headers found? */
752	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
753		return;
754
755	/* did the hardware decode the packet and checksum? */
756	if (!(rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
757		return;
758
759	/* both known and outer_ip must be set for the below code to work */
760	if (!(decoded.known && decoded.outer_ip))
761		return;
762
763	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
764	    decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
765		ipv4 = true;
766	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
767		 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
768		ipv6 = true;
769
770	if (ipv4 &&
771	    (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) |
772			 (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT))))
773		goto checksum_fail;
774
775	/* likely incorrect csum if alternate IP extension headers found */
776	if (ipv6 &&
777	    rx_status & (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
778		/* don't increment checksum err here, non-fatal err */
779		return;
780
781	/* there was some L4 error, count error and punt packet to the stack */
782	if (rx_error & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT))
783		goto checksum_fail;
784
785	/* handle packets that were not able to be checksummed due
786	 * to arrival speed, in this case the stack can compute
787	 * the csum.
788	 */
789	if (rx_error & (1 << I40E_RX_DESC_ERROR_PPRS_SHIFT))
790		return;
791
792	/* If VXLAN traffic has an outer UDPv4 checksum we need to check
793	 * it in the driver, hardware does not do it for us.
794	 * Since L3L4P bit was set we assume a valid IHL value (>=5)
795	 * so the total length of IPv4 header is IHL*4 bytes
796	 * The UDP_0 bit *may* bet set if the *inner* header is UDP
797	 */
798	if (ipv4_tunnel &&
799	    (decoded.inner_prot != I40E_RX_PTYPE_INNER_PROT_UDP) &&
800	    !(rx_status & (1 << I40E_RX_DESC_STATUS_UDP_0_SHIFT))) {
801		skb->transport_header = skb->mac_header +
802					sizeof(struct ethhdr) +
803					(ip_hdr(skb)->ihl * 4);
804
805		/* Add 4 bytes for VLAN tagged packets */
806		skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
807					  skb->protocol == htons(ETH_P_8021AD))
808					  ? VLAN_HLEN : 0;
809
810		rx_udp_csum = udp_csum(skb);
811		iph = ip_hdr(skb);
812		csum = csum_tcpudp_magic(
813				iph->saddr, iph->daddr,
814				(skb->len - skb_transport_offset(skb)),
815				IPPROTO_UDP, rx_udp_csum);
816
817		if (udp_hdr(skb)->check != csum)
818			goto checksum_fail;
819	}
820
821	skb->ip_summed = CHECKSUM_UNNECESSARY;
822	skb->csum_level = ipv4_tunnel || ipv6_tunnel;
823
824	return;
825
826checksum_fail:
827	vsi->back->hw_csum_rx_error++;
828}
829
830/**
831 * i40e_rx_hash - returns the hash value from the Rx descriptor
832 * @ring: descriptor ring
833 * @rx_desc: specific descriptor
834 **/
835static inline u32 i40e_rx_hash(struct i40e_ring *ring,
836			       union i40e_rx_desc *rx_desc)
837{
838	const __le64 rss_mask =
839		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
840			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
841
842	if ((ring->netdev->features & NETIF_F_RXHASH) &&
843	    (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
844		return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
845	else
846		return 0;
847}
848
849/**
850 * i40e_ptype_to_hash - get a hash type
851 * @ptype: the ptype value from the descriptor
852 *
853 * Returns a hash type to be used by skb_set_hash
854 **/
855static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
856{
857	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
858
859	if (!decoded.known)
860		return PKT_HASH_TYPE_NONE;
861
862	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
863	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
864		return PKT_HASH_TYPE_L4;
865	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
866		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
867		return PKT_HASH_TYPE_L3;
868	else
869		return PKT_HASH_TYPE_L2;
870}
871
872/**
873 * i40e_clean_rx_irq - Reclaim resources after receive completes
874 * @rx_ring:  rx ring to clean
875 * @budget:   how many cleans we're allowed
876 *
877 * Returns true if there's any budget left (e.g. the clean is finished)
878 **/
879static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
880{
881	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
882	u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
883	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
884	const int current_node = numa_node_id();
885	struct i40e_vsi *vsi = rx_ring->vsi;
886	u16 i = rx_ring->next_to_clean;
887	union i40e_rx_desc *rx_desc;
888	u32 rx_error, rx_status;
889	u8 rx_ptype;
890	u64 qword;
891
892	rx_desc = I40E_RX_DESC(rx_ring, i);
893	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
894	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
895		    I40E_RXD_QW1_STATUS_SHIFT;
896
897	while (rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
898		union i40e_rx_desc *next_rxd;
899		struct i40e_rx_buffer *rx_bi;
900		struct sk_buff *skb;
901		u16 vlan_tag;
902		rx_bi = &rx_ring->rx_bi[i];
903		skb = rx_bi->skb;
904		prefetch(skb->data);
905
906		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
907				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
908		rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
909				I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
910		rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
911			 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
912
913		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
914			   I40E_RXD_QW1_ERROR_SHIFT;
915		rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
916		rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
917
918		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
919			   I40E_RXD_QW1_PTYPE_SHIFT;
920		rx_bi->skb = NULL;
921
922		/* This memory barrier is needed to keep us from reading
923		 * any other fields out of the rx_desc until we know the
924		 * STATUS_DD bit is set
925		 */
926		rmb();
927
928		/* Get the header and possibly the whole packet
929		 * If this is an skb from previous receive dma will be 0
930		 */
931		if (rx_bi->dma) {
932			u16 len;
933
934			if (rx_hbo)
935				len = I40E_RX_HDR_SIZE;
936			else if (rx_sph)
937				len = rx_header_len;
938			else if (rx_packet_len)
939				len = rx_packet_len;   /* 1buf/no split found */
940			else
941				len = rx_header_len;   /* split always mode */
942
943			skb_put(skb, len);
944			dma_unmap_single(rx_ring->dev,
945					 rx_bi->dma,
946					 rx_ring->rx_buf_len,
947					 DMA_FROM_DEVICE);
948			rx_bi->dma = 0;
949		}
950
951		/* Get the rest of the data if this was a header split */
952		if (ring_is_ps_enabled(rx_ring) && rx_packet_len) {
953
954			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
955					   rx_bi->page,
956					   rx_bi->page_offset,
957					   rx_packet_len);
958
959			skb->len += rx_packet_len;
960			skb->data_len += rx_packet_len;
961			skb->truesize += rx_packet_len;
962
963			if ((page_count(rx_bi->page) == 1) &&
964			    (page_to_nid(rx_bi->page) == current_node))
965				get_page(rx_bi->page);
966			else
967				rx_bi->page = NULL;
968
969			dma_unmap_page(rx_ring->dev,
970				       rx_bi->page_dma,
971				       PAGE_SIZE / 2,
972				       DMA_FROM_DEVICE);
973			rx_bi->page_dma = 0;
974		}
975		I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd);
976
977		if (unlikely(
978		    !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
979			struct i40e_rx_buffer *next_buffer;
980
981			next_buffer = &rx_ring->rx_bi[i];
982
983			if (ring_is_ps_enabled(rx_ring)) {
984				rx_bi->skb = next_buffer->skb;
985				rx_bi->dma = next_buffer->dma;
986				next_buffer->skb = skb;
987				next_buffer->dma = 0;
988			}
989			rx_ring->rx_stats.non_eop_descs++;
990			goto next_desc;
991		}
992
993		/* ERR_MASK will only have valid bits if EOP set */
994		if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
995			dev_kfree_skb_any(skb);
996			/* TODO: shouldn't we increment a counter indicating the
997			 * drop?
998			 */
999			goto next_desc;
1000		}
1001
1002		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1003			     i40e_ptype_to_hash(rx_ptype));
1004		/* probably a little skewed due to removing CRC */
1005		total_rx_bytes += skb->len;
1006		total_rx_packets++;
1007
1008		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1009
1010		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1011
1012		vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1013			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1014			 : 0;
1015		i40e_receive_skb(rx_ring, skb, vlan_tag);
1016
1017		rx_ring->netdev->last_rx = jiffies;
1018		budget--;
1019next_desc:
1020		rx_desc->wb.qword1.status_error_len = 0;
1021		if (!budget)
1022			break;
1023
1024		cleaned_count++;
1025		/* return some buffers to hardware, one at a time is too slow */
1026		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1027			i40evf_alloc_rx_buffers(rx_ring, cleaned_count);
1028			cleaned_count = 0;
1029		}
1030
1031		/* use prefetched values */
1032		rx_desc = next_rxd;
1033		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1034		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1035			    I40E_RXD_QW1_STATUS_SHIFT;
1036	}
1037
1038	rx_ring->next_to_clean = i;
1039	u64_stats_update_begin(&rx_ring->syncp);
1040	rx_ring->stats.packets += total_rx_packets;
1041	rx_ring->stats.bytes += total_rx_bytes;
1042	u64_stats_update_end(&rx_ring->syncp);
1043	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1044	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1045
1046	if (cleaned_count)
1047		i40evf_alloc_rx_buffers(rx_ring, cleaned_count);
1048
1049	return budget > 0;
1050}
1051
1052/**
1053 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1054 * @napi: napi struct with our devices info in it
1055 * @budget: amount of work driver is allowed to do this pass, in packets
1056 *
1057 * This function will clean all queues associated with a q_vector.
1058 *
1059 * Returns the amount of work done
1060 **/
1061int i40evf_napi_poll(struct napi_struct *napi, int budget)
1062{
1063	struct i40e_q_vector *q_vector =
1064			       container_of(napi, struct i40e_q_vector, napi);
1065	struct i40e_vsi *vsi = q_vector->vsi;
1066	struct i40e_ring *ring;
1067	bool clean_complete = true;
1068	int budget_per_ring;
1069
1070	if (test_bit(__I40E_DOWN, &vsi->state)) {
1071		napi_complete(napi);
1072		return 0;
1073	}
1074
1075	/* Since the actual Tx work is minimal, we can give the Tx a larger
1076	 * budget and be more aggressive about cleaning up the Tx descriptors.
1077	 */
1078	i40e_for_each_ring(ring, q_vector->tx)
1079		clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1080
1081	/* We attempt to distribute budget to each Rx queue fairly, but don't
1082	 * allow the budget to go below 1 because that would exit polling early.
1083	 */
1084	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1085
1086	i40e_for_each_ring(ring, q_vector->rx)
1087		clean_complete &= i40e_clean_rx_irq(ring, budget_per_ring);
1088
1089	/* If work not completed, return budget and polling will return */
1090	if (!clean_complete)
1091		return budget;
1092
1093	/* Work is done so exit the polling mode and re-enable the interrupt */
1094	napi_complete(napi);
1095	if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) ||
1096	    ITR_IS_DYNAMIC(vsi->tx_itr_setting))
1097		i40e_update_dynamic_itr(q_vector);
1098
1099	if (!test_bit(__I40E_DOWN, &vsi->state))
1100		i40evf_irq_enable_queues(vsi->back, 1 << q_vector->v_idx);
1101
1102	return 0;
1103}
1104
1105/**
1106 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1107 * @skb:     send buffer
1108 * @tx_ring: ring to send buffer on
1109 * @flags:   the tx flags to be set
1110 *
1111 * Checks the skb and set up correspondingly several generic transmit flags
1112 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1113 *
1114 * Returns error code indicate the frame should be dropped upon error and the
1115 * otherwise  returns 0 to indicate the flags has been set properly.
1116 **/
1117static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
1118				      struct i40e_ring *tx_ring,
1119				      u32 *flags)
1120{
1121	__be16 protocol = skb->protocol;
1122	u32  tx_flags = 0;
1123
1124	/* if we have a HW VLAN tag being added, default to the HW one */
1125	if (vlan_tx_tag_present(skb)) {
1126		tx_flags |= vlan_tx_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1127		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1128	/* else if it is a SW VLAN, check the next protocol and store the tag */
1129	} else if (protocol == htons(ETH_P_8021Q)) {
1130		struct vlan_hdr *vhdr, _vhdr;
1131		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1132		if (!vhdr)
1133			return -EINVAL;
1134
1135		protocol = vhdr->h_vlan_encapsulated_proto;
1136		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1137		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1138	}
1139
1140	*flags = tx_flags;
1141	return 0;
1142}
1143
1144/**
1145 * i40e_tso - set up the tso context descriptor
1146 * @tx_ring:  ptr to the ring to send
1147 * @skb:      ptr to the skb we're sending
1148 * @tx_flags: the collected send information
1149 * @protocol: the send protocol
1150 * @hdr_len:  ptr to the size of the packet header
1151 * @cd_tunneling: ptr to context descriptor bits
1152 *
1153 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1154 **/
1155static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1156		    u32 tx_flags, __be16 protocol, u8 *hdr_len,
1157		    u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling)
1158{
1159	u32 cd_cmd, cd_tso_len, cd_mss;
1160	struct ipv6hdr *ipv6h;
1161	struct tcphdr *tcph;
1162	struct iphdr *iph;
1163	u32 l4len;
1164	int err;
1165
1166	if (!skb_is_gso(skb))
1167		return 0;
1168
1169	err = skb_cow_head(skb, 0);
1170	if (err < 0)
1171		return err;
1172
1173	if (protocol == htons(ETH_P_IP)) {
1174		iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1175		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1176		iph->tot_len = 0;
1177		iph->check = 0;
1178		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1179						 0, IPPROTO_TCP, 0);
1180	} else if (skb_is_gso_v6(skb)) {
1181
1182		ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb)
1183					   : ipv6_hdr(skb);
1184		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1185		ipv6h->payload_len = 0;
1186		tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1187					       0, IPPROTO_TCP, 0);
1188	}
1189
1190	l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1191	*hdr_len = (skb->encapsulation
1192		    ? (skb_inner_transport_header(skb) - skb->data)
1193		    : skb_transport_offset(skb)) + l4len;
1194
1195	/* find the field values */
1196	cd_cmd = I40E_TX_CTX_DESC_TSO;
1197	cd_tso_len = skb->len - *hdr_len;
1198	cd_mss = skb_shinfo(skb)->gso_size;
1199	*cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1200				((u64)cd_tso_len <<
1201				 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1202				((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1203	return 1;
1204}
1205
1206/**
1207 * i40e_tx_enable_csum - Enable Tx checksum offloads
1208 * @skb: send buffer
1209 * @tx_flags: Tx flags currently set
1210 * @td_cmd: Tx descriptor command bits to set
1211 * @td_offset: Tx descriptor header offsets to set
1212 * @cd_tunneling: ptr to context desc bits
1213 **/
1214static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags,
1215				u32 *td_cmd, u32 *td_offset,
1216				struct i40e_ring *tx_ring,
1217				u32 *cd_tunneling)
1218{
1219	struct ipv6hdr *this_ipv6_hdr;
1220	unsigned int this_tcp_hdrlen;
1221	struct iphdr *this_ip_hdr;
1222	u32 network_hdr_len;
1223	u8 l4_hdr = 0;
1224
1225	if (skb->encapsulation) {
1226		network_hdr_len = skb_inner_network_header_len(skb);
1227		this_ip_hdr = inner_ip_hdr(skb);
1228		this_ipv6_hdr = inner_ipv6_hdr(skb);
1229		this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1230
1231		if (tx_flags & I40E_TX_FLAGS_IPV4) {
1232
1233			if (tx_flags & I40E_TX_FLAGS_TSO) {
1234				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1235				ip_hdr(skb)->check = 0;
1236			} else {
1237				*cd_tunneling |=
1238					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1239			}
1240		} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1241			if (tx_flags & I40E_TX_FLAGS_TSO) {
1242				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1243				ip_hdr(skb)->check = 0;
1244			} else {
1245				*cd_tunneling |=
1246					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1247			}
1248		}
1249
1250		/* Now set the ctx descriptor fields */
1251		*cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1252					I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1253				   I40E_TXD_CTX_UDP_TUNNELING            |
1254				   ((skb_inner_network_offset(skb) -
1255					skb_transport_offset(skb)) >> 1) <<
1256				   I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1257
1258	} else {
1259		network_hdr_len = skb_network_header_len(skb);
1260		this_ip_hdr = ip_hdr(skb);
1261		this_ipv6_hdr = ipv6_hdr(skb);
1262		this_tcp_hdrlen = tcp_hdrlen(skb);
1263	}
1264
1265	/* Enable IP checksum offloads */
1266	if (tx_flags & I40E_TX_FLAGS_IPV4) {
1267		l4_hdr = this_ip_hdr->protocol;
1268		/* the stack computes the IP header already, the only time we
1269		 * need the hardware to recompute it is in the case of TSO.
1270		 */
1271		if (tx_flags & I40E_TX_FLAGS_TSO) {
1272			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1273			this_ip_hdr->check = 0;
1274		} else {
1275			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1276		}
1277		/* Now set the td_offset for IP header length */
1278		*td_offset = (network_hdr_len >> 2) <<
1279			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1280	} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1281		l4_hdr = this_ipv6_hdr->nexthdr;
1282		*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1283		/* Now set the td_offset for IP header length */
1284		*td_offset = (network_hdr_len >> 2) <<
1285			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1286	}
1287	/* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1288	*td_offset |= (skb_network_offset(skb) >> 1) <<
1289		       I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1290
1291	/* Enable L4 checksum offloads */
1292	switch (l4_hdr) {
1293	case IPPROTO_TCP:
1294		/* enable checksum offloads */
1295		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1296		*td_offset |= (this_tcp_hdrlen >> 2) <<
1297			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1298		break;
1299	case IPPROTO_SCTP:
1300		/* enable SCTP checksum offload */
1301		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1302		*td_offset |= (sizeof(struct sctphdr) >> 2) <<
1303			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1304		break;
1305	case IPPROTO_UDP:
1306		/* enable UDP checksum offload */
1307		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1308		*td_offset |= (sizeof(struct udphdr) >> 2) <<
1309			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1310		break;
1311	default:
1312		break;
1313	}
1314}
1315
1316/**
1317 * i40e_create_tx_ctx Build the Tx context descriptor
1318 * @tx_ring:  ring to create the descriptor on
1319 * @cd_type_cmd_tso_mss: Quad Word 1
1320 * @cd_tunneling: Quad Word 0 - bits 0-31
1321 * @cd_l2tag2: Quad Word 0 - bits 32-63
1322 **/
1323static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1324			       const u64 cd_type_cmd_tso_mss,
1325			       const u32 cd_tunneling, const u32 cd_l2tag2)
1326{
1327	struct i40e_tx_context_desc *context_desc;
1328	int i = tx_ring->next_to_use;
1329
1330	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1331	    !cd_tunneling && !cd_l2tag2)
1332		return;
1333
1334	/* grab the next descriptor */
1335	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1336
1337	i++;
1338	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1339
1340	/* cpu_to_le32 and assign to struct fields */
1341	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1342	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1343	context_desc->rsvd = cpu_to_le16(0);
1344	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1345}
1346
1347/**
1348 * i40e_tx_map - Build the Tx descriptor
1349 * @tx_ring:  ring to send buffer on
1350 * @skb:      send buffer
1351 * @first:    first buffer info buffer to use
1352 * @tx_flags: collected send information
1353 * @hdr_len:  size of the packet header
1354 * @td_cmd:   the command field in the descriptor
1355 * @td_offset: offset for checksum or crc
1356 **/
1357static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1358			struct i40e_tx_buffer *first, u32 tx_flags,
1359			const u8 hdr_len, u32 td_cmd, u32 td_offset)
1360{
1361	unsigned int data_len = skb->data_len;
1362	unsigned int size = skb_headlen(skb);
1363	struct skb_frag_struct *frag;
1364	struct i40e_tx_buffer *tx_bi;
1365	struct i40e_tx_desc *tx_desc;
1366	u16 i = tx_ring->next_to_use;
1367	u32 td_tag = 0;
1368	dma_addr_t dma;
1369	u16 gso_segs;
1370
1371	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1372		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1373		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1374			 I40E_TX_FLAGS_VLAN_SHIFT;
1375	}
1376
1377	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1378		gso_segs = skb_shinfo(skb)->gso_segs;
1379	else
1380		gso_segs = 1;
1381
1382	/* multiply data chunks by size of headers */
1383	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1384	first->gso_segs = gso_segs;
1385	first->skb = skb;
1386	first->tx_flags = tx_flags;
1387
1388	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1389
1390	tx_desc = I40E_TX_DESC(tx_ring, i);
1391	tx_bi = first;
1392
1393	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1394		if (dma_mapping_error(tx_ring->dev, dma))
1395			goto dma_error;
1396
1397		/* record length, and DMA address */
1398		dma_unmap_len_set(tx_bi, len, size);
1399		dma_unmap_addr_set(tx_bi, dma, dma);
1400
1401		tx_desc->buffer_addr = cpu_to_le64(dma);
1402
1403		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1404			tx_desc->cmd_type_offset_bsz =
1405				build_ctob(td_cmd, td_offset,
1406					   I40E_MAX_DATA_PER_TXD, td_tag);
1407
1408			tx_desc++;
1409			i++;
1410			if (i == tx_ring->count) {
1411				tx_desc = I40E_TX_DESC(tx_ring, 0);
1412				i = 0;
1413			}
1414
1415			dma += I40E_MAX_DATA_PER_TXD;
1416			size -= I40E_MAX_DATA_PER_TXD;
1417
1418			tx_desc->buffer_addr = cpu_to_le64(dma);
1419		}
1420
1421		if (likely(!data_len))
1422			break;
1423
1424		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1425							  size, td_tag);
1426
1427		tx_desc++;
1428		i++;
1429		if (i == tx_ring->count) {
1430			tx_desc = I40E_TX_DESC(tx_ring, 0);
1431			i = 0;
1432		}
1433
1434		size = skb_frag_size(frag);
1435		data_len -= size;
1436
1437		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1438				       DMA_TO_DEVICE);
1439
1440		tx_bi = &tx_ring->tx_bi[i];
1441	}
1442
1443	/* Place RS bit on last descriptor of any packet that spans across the
1444	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1445	 */
1446#define WB_STRIDE 0x3
1447	if (((i & WB_STRIDE) != WB_STRIDE) &&
1448	    (first <= &tx_ring->tx_bi[i]) &&
1449	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1450		tx_desc->cmd_type_offset_bsz =
1451			build_ctob(td_cmd, td_offset, size, td_tag) |
1452			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1453					 I40E_TXD_QW1_CMD_SHIFT);
1454	} else {
1455		tx_desc->cmd_type_offset_bsz =
1456			build_ctob(td_cmd, td_offset, size, td_tag) |
1457			cpu_to_le64((u64)I40E_TXD_CMD <<
1458					 I40E_TXD_QW1_CMD_SHIFT);
1459	}
1460
1461	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1462						 tx_ring->queue_index),
1463			     first->bytecount);
1464
1465	/* set the timestamp */
1466	first->time_stamp = jiffies;
1467
1468	/* Force memory writes to complete before letting h/w
1469	 * know there are new descriptors to fetch.  (Only
1470	 * applicable for weak-ordered memory model archs,
1471	 * such as IA-64).
1472	 */
1473	wmb();
1474
1475	/* set next_to_watch value indicating a packet is present */
1476	first->next_to_watch = tx_desc;
1477
1478	i++;
1479	if (i == tx_ring->count)
1480		i = 0;
1481
1482	tx_ring->next_to_use = i;
1483
1484	/* notify HW of packet */
1485	writel(i, tx_ring->tail);
1486
1487	return;
1488
1489dma_error:
1490	dev_info(tx_ring->dev, "TX DMA map failed\n");
1491
1492	/* clear dma mappings for failed tx_bi map */
1493	for (;;) {
1494		tx_bi = &tx_ring->tx_bi[i];
1495		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1496		if (tx_bi == first)
1497			break;
1498		if (i == 0)
1499			i = tx_ring->count;
1500		i--;
1501	}
1502
1503	tx_ring->next_to_use = i;
1504}
1505
1506/**
1507 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
1508 * @tx_ring: the ring to be checked
1509 * @size:    the size buffer we want to assure is available
1510 *
1511 * Returns -EBUSY if a stop is needed, else 0
1512 **/
1513static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1514{
1515	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1516	/* Memory barrier before checking head and tail */
1517	smp_mb();
1518
1519	/* Check again in a case another CPU has just made room available. */
1520	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1521		return -EBUSY;
1522
1523	/* A reprieve! - use start_queue because it doesn't call schedule */
1524	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1525	++tx_ring->tx_stats.restart_queue;
1526	return 0;
1527}
1528
1529/**
1530 * i40e_maybe_stop_tx - 1st level check for tx stop conditions
1531 * @tx_ring: the ring to be checked
1532 * @size:    the size buffer we want to assure is available
1533 *
1534 * Returns 0 if stop is not needed
1535 **/
1536static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1537{
1538	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1539		return 0;
1540	return __i40e_maybe_stop_tx(tx_ring, size);
1541}
1542
1543/**
1544 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
1545 * @skb:     send buffer
1546 * @tx_ring: ring to send buffer on
1547 *
1548 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1549 * there is not enough descriptors available in this ring since we need at least
1550 * one descriptor.
1551 **/
1552static int i40e_xmit_descriptor_count(struct sk_buff *skb,
1553				      struct i40e_ring *tx_ring)
1554{
1555	unsigned int f;
1556	int count = 0;
1557
1558	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1559	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1560	 *       + 4 desc gap to avoid the cache line where head is,
1561	 *       + 1 desc for context descriptor,
1562	 * otherwise try next time
1563	 */
1564	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1565		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1566
1567	count += TXD_USE_COUNT(skb_headlen(skb));
1568	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1569		tx_ring->tx_stats.tx_busy++;
1570		return 0;
1571	}
1572	return count;
1573}
1574
1575/**
1576 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1577 * @skb:     send buffer
1578 * @tx_ring: ring to send buffer on
1579 *
1580 * Returns NETDEV_TX_OK if sent, else an error code
1581 **/
1582static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1583					struct i40e_ring *tx_ring)
1584{
1585	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1586	u32 cd_tunneling = 0, cd_l2tag2 = 0;
1587	struct i40e_tx_buffer *first;
1588	u32 td_offset = 0;
1589	u32 tx_flags = 0;
1590	__be16 protocol;
1591	u32 td_cmd = 0;
1592	u8 hdr_len = 0;
1593	int tso;
1594	if (0 == i40e_xmit_descriptor_count(skb, tx_ring))
1595		return NETDEV_TX_BUSY;
1596
1597	/* prepare the xmit flags */
1598	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1599		goto out_drop;
1600
1601	/* obtain protocol of skb */
1602	protocol = vlan_get_protocol(skb);
1603
1604	/* record the location of the first descriptor for this packet */
1605	first = &tx_ring->tx_bi[tx_ring->next_to_use];
1606
1607	/* setup IPv4/IPv6 offloads */
1608	if (protocol == htons(ETH_P_IP))
1609		tx_flags |= I40E_TX_FLAGS_IPV4;
1610	else if (protocol == htons(ETH_P_IPV6))
1611		tx_flags |= I40E_TX_FLAGS_IPV6;
1612
1613	tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len,
1614		       &cd_type_cmd_tso_mss, &cd_tunneling);
1615
1616	if (tso < 0)
1617		goto out_drop;
1618	else if (tso)
1619		tx_flags |= I40E_TX_FLAGS_TSO;
1620
1621	skb_tx_timestamp(skb);
1622
1623	/* always enable CRC insertion offload */
1624	td_cmd |= I40E_TX_DESC_CMD_ICRC;
1625
1626	/* Always offload the checksum, since it's in the data descriptor */
1627	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1628		tx_flags |= I40E_TX_FLAGS_CSUM;
1629
1630		i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset,
1631				    tx_ring, &cd_tunneling);
1632	}
1633
1634	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
1635			   cd_tunneling, cd_l2tag2);
1636
1637	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
1638		    td_cmd, td_offset);
1639
1640	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
1641
1642	return NETDEV_TX_OK;
1643
1644out_drop:
1645	dev_kfree_skb_any(skb);
1646	return NETDEV_TX_OK;
1647}
1648
1649/**
1650 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1651 * @skb:    send buffer
1652 * @netdev: network interface device structure
1653 *
1654 * Returns NETDEV_TX_OK if sent, else an error code
1655 **/
1656netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1657{
1658	struct i40evf_adapter *adapter = netdev_priv(netdev);
1659	struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
1660
1661	/* hardware can't handle really short frames, hardware padding works
1662	 * beyond this point
1663	 */
1664	if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
1665		if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
1666			return NETDEV_TX_OK;
1667		skb->len = I40E_MIN_TX_LEN;
1668		skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
1669	}
1670
1671	return i40e_xmit_frame_ring(skb, tx_ring);
1672}
1673