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#include <net/busy_poll.h>
29
30#include "i40evf.h"
31#include "i40e_prototype.h"
32
33static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
34				u32 td_tag)
35{
36	return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
37			   ((u64)td_cmd  << I40E_TXD_QW1_CMD_SHIFT) |
38			   ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
39			   ((u64)size  << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
40			   ((u64)td_tag  << I40E_TXD_QW1_L2TAG1_SHIFT));
41}
42
43#define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
44
45/**
46 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
47 * @ring:      the ring that owns the buffer
48 * @tx_buffer: the buffer to free
49 **/
50static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
51					    struct i40e_tx_buffer *tx_buffer)
52{
53	if (tx_buffer->skb) {
54		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
55			kfree(tx_buffer->raw_buf);
56		else
57			dev_kfree_skb_any(tx_buffer->skb);
58
59		if (dma_unmap_len(tx_buffer, len))
60			dma_unmap_single(ring->dev,
61					 dma_unmap_addr(tx_buffer, dma),
62					 dma_unmap_len(tx_buffer, len),
63					 DMA_TO_DEVICE);
64	} else if (dma_unmap_len(tx_buffer, len)) {
65		dma_unmap_page(ring->dev,
66			       dma_unmap_addr(tx_buffer, dma),
67			       dma_unmap_len(tx_buffer, len),
68			       DMA_TO_DEVICE);
69	}
70	tx_buffer->next_to_watch = NULL;
71	tx_buffer->skb = NULL;
72	dma_unmap_len_set(tx_buffer, len, 0);
73	/* tx_buffer must be completely set up in the transmit path */
74}
75
76/**
77 * i40evf_clean_tx_ring - Free any empty Tx buffers
78 * @tx_ring: ring to be cleaned
79 **/
80void i40evf_clean_tx_ring(struct i40e_ring *tx_ring)
81{
82	unsigned long bi_size;
83	u16 i;
84
85	/* ring already cleared, nothing to do */
86	if (!tx_ring->tx_bi)
87		return;
88
89	/* Free all the Tx ring sk_buffs */
90	for (i = 0; i < tx_ring->count; i++)
91		i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
92
93	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
94	memset(tx_ring->tx_bi, 0, bi_size);
95
96	/* Zero out the descriptor ring */
97	memset(tx_ring->desc, 0, tx_ring->size);
98
99	tx_ring->next_to_use = 0;
100	tx_ring->next_to_clean = 0;
101
102	if (!tx_ring->netdev)
103		return;
104
105	/* cleanup Tx queue statistics */
106	netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
107						  tx_ring->queue_index));
108}
109
110/**
111 * i40evf_free_tx_resources - Free Tx resources per queue
112 * @tx_ring: Tx descriptor ring for a specific queue
113 *
114 * Free all transmit software resources
115 **/
116void i40evf_free_tx_resources(struct i40e_ring *tx_ring)
117{
118	i40evf_clean_tx_ring(tx_ring);
119	kfree(tx_ring->tx_bi);
120	tx_ring->tx_bi = NULL;
121
122	if (tx_ring->desc) {
123		dma_free_coherent(tx_ring->dev, tx_ring->size,
124				  tx_ring->desc, tx_ring->dma);
125		tx_ring->desc = NULL;
126	}
127}
128
129/**
130 * i40e_get_head - Retrieve head from head writeback
131 * @tx_ring:  tx ring to fetch head of
132 *
133 * Returns value of Tx ring head based on value stored
134 * in head write-back location
135 **/
136static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
137{
138	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
139
140	return le32_to_cpu(*(volatile __le32 *)head);
141}
142
143/**
144 * i40e_get_tx_pending - how many tx descriptors not processed
145 * @tx_ring: the ring of descriptors
146 *
147 * Since there is no access to the ring head register
148 * in XL710, we need to use our local copies
149 **/
150static u32 i40e_get_tx_pending(struct i40e_ring *ring)
151{
152	u32 head, tail;
153
154	head = i40e_get_head(ring);
155	tail = readl(ring->tail);
156
157	if (head != tail)
158		return (head < tail) ?
159			tail - head : (tail + ring->count - head);
160
161	return 0;
162}
163
164/**
165 * i40e_check_tx_hang - Is there a hang in the Tx queue
166 * @tx_ring: the ring of descriptors
167 **/
168static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
169{
170	u32 tx_done = tx_ring->stats.packets;
171	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
172	u32 tx_pending = i40e_get_tx_pending(tx_ring);
173	bool ret = false;
174
175	clear_check_for_tx_hang(tx_ring);
176
177	/* Check for a hung queue, but be thorough. This verifies
178	 * that a transmit has been completed since the previous
179	 * check AND there is at least one packet pending. The
180	 * ARMED bit is set to indicate a potential hang. The
181	 * bit is cleared if a pause frame is received to remove
182	 * false hang detection due to PFC or 802.3x frames. By
183	 * requiring this to fail twice we avoid races with
184	 * PFC clearing the ARMED bit and conditions where we
185	 * run the check_tx_hang logic with a transmit completion
186	 * pending but without time to complete it yet.
187	 */
188	if ((tx_done_old == tx_done) && tx_pending) {
189		/* make sure it is true for two checks in a row */
190		ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED,
191				       &tx_ring->state);
192	} else if (tx_done_old == tx_done &&
193		   (tx_pending < I40E_MIN_DESC_PENDING) && (tx_pending > 0)) {
194		/* update completed stats and disarm the hang check */
195		tx_ring->tx_stats.tx_done_old = tx_done;
196		clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state);
197	}
198
199	return ret;
200}
201
202#define WB_STRIDE 0x3
203
204/**
205 * i40e_clean_tx_irq - Reclaim resources after transmit completes
206 * @tx_ring:  tx ring to clean
207 * @budget:   how many cleans we're allowed
208 *
209 * Returns true if there's any budget left (e.g. the clean is finished)
210 **/
211static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
212{
213	u16 i = tx_ring->next_to_clean;
214	struct i40e_tx_buffer *tx_buf;
215	struct i40e_tx_desc *tx_head;
216	struct i40e_tx_desc *tx_desc;
217	unsigned int total_packets = 0;
218	unsigned int total_bytes = 0;
219
220	tx_buf = &tx_ring->tx_bi[i];
221	tx_desc = I40E_TX_DESC(tx_ring, i);
222	i -= tx_ring->count;
223
224	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
225
226	do {
227		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
228
229		/* if next_to_watch is not set then there is no work pending */
230		if (!eop_desc)
231			break;
232
233		/* prevent any other reads prior to eop_desc */
234		read_barrier_depends();
235
236		/* we have caught up to head, no work left to do */
237		if (tx_head == tx_desc)
238			break;
239
240		/* clear next_to_watch to prevent false hangs */
241		tx_buf->next_to_watch = NULL;
242
243		/* update the statistics for this packet */
244		total_bytes += tx_buf->bytecount;
245		total_packets += tx_buf->gso_segs;
246
247		/* free the skb */
248		dev_kfree_skb_any(tx_buf->skb);
249
250		/* unmap skb header data */
251		dma_unmap_single(tx_ring->dev,
252				 dma_unmap_addr(tx_buf, dma),
253				 dma_unmap_len(tx_buf, len),
254				 DMA_TO_DEVICE);
255
256		/* clear tx_buffer data */
257		tx_buf->skb = NULL;
258		dma_unmap_len_set(tx_buf, len, 0);
259
260		/* unmap remaining buffers */
261		while (tx_desc != eop_desc) {
262
263			tx_buf++;
264			tx_desc++;
265			i++;
266			if (unlikely(!i)) {
267				i -= tx_ring->count;
268				tx_buf = tx_ring->tx_bi;
269				tx_desc = I40E_TX_DESC(tx_ring, 0);
270			}
271
272			/* unmap any remaining paged data */
273			if (dma_unmap_len(tx_buf, len)) {
274				dma_unmap_page(tx_ring->dev,
275					       dma_unmap_addr(tx_buf, dma),
276					       dma_unmap_len(tx_buf, len),
277					       DMA_TO_DEVICE);
278				dma_unmap_len_set(tx_buf, len, 0);
279			}
280		}
281
282		/* move us one more past the eop_desc for start of next pkt */
283		tx_buf++;
284		tx_desc++;
285		i++;
286		if (unlikely(!i)) {
287			i -= tx_ring->count;
288			tx_buf = tx_ring->tx_bi;
289			tx_desc = I40E_TX_DESC(tx_ring, 0);
290		}
291
292		prefetch(tx_desc);
293
294		/* update budget accounting */
295		budget--;
296	} while (likely(budget));
297
298	i += tx_ring->count;
299	tx_ring->next_to_clean = i;
300	u64_stats_update_begin(&tx_ring->syncp);
301	tx_ring->stats.bytes += total_bytes;
302	tx_ring->stats.packets += total_packets;
303	u64_stats_update_end(&tx_ring->syncp);
304	tx_ring->q_vector->tx.total_bytes += total_bytes;
305	tx_ring->q_vector->tx.total_packets += total_packets;
306
307	if (budget &&
308	    !((i & WB_STRIDE) == WB_STRIDE) &&
309	    !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
310	    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
311		tx_ring->arm_wb = true;
312	else
313		tx_ring->arm_wb = false;
314
315	if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) {
316		/* schedule immediate reset if we believe we hung */
317		dev_info(tx_ring->dev, "Detected Tx Unit Hang\n"
318			 "  VSI                  <%d>\n"
319			 "  Tx Queue             <%d>\n"
320			 "  next_to_use          <%x>\n"
321			 "  next_to_clean        <%x>\n",
322			 tx_ring->vsi->seid,
323			 tx_ring->queue_index,
324			 tx_ring->next_to_use, i);
325		dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n"
326			 "  time_stamp           <%lx>\n"
327			 "  jiffies              <%lx>\n",
328			 tx_ring->tx_bi[i].time_stamp, jiffies);
329
330		netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
331
332		dev_info(tx_ring->dev,
333			 "tx hang detected on queue %d, resetting adapter\n",
334			 tx_ring->queue_index);
335
336		tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev);
337
338		/* the adapter is about to reset, no point in enabling stuff */
339		return true;
340	}
341
342	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
343						      tx_ring->queue_index),
344				  total_packets, total_bytes);
345
346#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
347	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
348		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
349		/* Make sure that anybody stopping the queue after this
350		 * sees the new next_to_clean.
351		 */
352		smp_mb();
353		if (__netif_subqueue_stopped(tx_ring->netdev,
354					     tx_ring->queue_index) &&
355		   !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
356			netif_wake_subqueue(tx_ring->netdev,
357					    tx_ring->queue_index);
358			++tx_ring->tx_stats.restart_queue;
359		}
360	}
361
362	return budget > 0;
363}
364
365/**
366 * i40e_force_wb -Arm hardware to do a wb on noncache aligned descriptors
367 * @vsi: the VSI we care about
368 * @q_vector: the vector  on which to force writeback
369 *
370 **/
371static void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
372{
373	u32 val = I40E_VFINT_DYN_CTLN_INTENA_MASK |
374		  I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
375		  I40E_VFINT_DYN_CTLN_SWINT_TRIG_MASK |
376		  I40E_VFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
377		  /* allow 00 to be written to the index */
378
379	wr32(&vsi->back->hw,
380	     I40E_VFINT_DYN_CTLN1(q_vector->v_idx + vsi->base_vector - 1),
381	     val);
382}
383
384/**
385 * i40e_set_new_dynamic_itr - Find new ITR level
386 * @rc: structure containing ring performance data
387 *
388 * Stores a new ITR value based on packets and byte counts during
389 * the last interrupt.  The advantage of per interrupt computation
390 * is faster updates and more accurate ITR for the current traffic
391 * pattern.  Constants in this function were computed based on
392 * theoretical maximum wire speed and thresholds were set based on
393 * testing data as well as attempting to minimize response time
394 * while increasing bulk throughput.
395 **/
396static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
397{
398	enum i40e_latency_range new_latency_range = rc->latency_range;
399	u32 new_itr = rc->itr;
400	int bytes_per_int;
401
402	if (rc->total_packets == 0 || !rc->itr)
403		return;
404
405	/* simple throttlerate management
406	 *   0-10MB/s   lowest (100000 ints/s)
407	 *  10-20MB/s   low    (20000 ints/s)
408	 *  20-1249MB/s bulk   (8000 ints/s)
409	 */
410	bytes_per_int = rc->total_bytes / rc->itr;
411	switch (rc->itr) {
412	case I40E_LOWEST_LATENCY:
413		if (bytes_per_int > 10)
414			new_latency_range = I40E_LOW_LATENCY;
415		break;
416	case I40E_LOW_LATENCY:
417		if (bytes_per_int > 20)
418			new_latency_range = I40E_BULK_LATENCY;
419		else if (bytes_per_int <= 10)
420			new_latency_range = I40E_LOWEST_LATENCY;
421		break;
422	case I40E_BULK_LATENCY:
423		if (bytes_per_int <= 20)
424			rc->latency_range = I40E_LOW_LATENCY;
425		break;
426	}
427
428	switch (new_latency_range) {
429	case I40E_LOWEST_LATENCY:
430		new_itr = I40E_ITR_100K;
431		break;
432	case I40E_LOW_LATENCY:
433		new_itr = I40E_ITR_20K;
434		break;
435	case I40E_BULK_LATENCY:
436		new_itr = I40E_ITR_8K;
437		break;
438	default:
439		break;
440	}
441
442	if (new_itr != rc->itr) {
443		/* do an exponential smoothing */
444		new_itr = (10 * new_itr * rc->itr) /
445			  ((9 * new_itr) + rc->itr);
446		rc->itr = new_itr & I40E_MAX_ITR;
447	}
448
449	rc->total_bytes = 0;
450	rc->total_packets = 0;
451}
452
453/**
454 * i40e_update_dynamic_itr - Adjust ITR based on bytes per int
455 * @q_vector: the vector to adjust
456 **/
457static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector)
458{
459	u16 vector = q_vector->vsi->base_vector + q_vector->v_idx;
460	struct i40e_hw *hw = &q_vector->vsi->back->hw;
461	u32 reg_addr;
462	u16 old_itr;
463
464	reg_addr = I40E_VFINT_ITRN1(I40E_RX_ITR, vector - 1);
465	old_itr = q_vector->rx.itr;
466	i40e_set_new_dynamic_itr(&q_vector->rx);
467	if (old_itr != q_vector->rx.itr)
468		wr32(hw, reg_addr, q_vector->rx.itr);
469
470	reg_addr = I40E_VFINT_ITRN1(I40E_TX_ITR, vector - 1);
471	old_itr = q_vector->tx.itr;
472	i40e_set_new_dynamic_itr(&q_vector->tx);
473	if (old_itr != q_vector->tx.itr)
474		wr32(hw, reg_addr, q_vector->tx.itr);
475}
476
477/**
478 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
479 * @tx_ring: the tx ring to set up
480 *
481 * Return 0 on success, negative on error
482 **/
483int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
484{
485	struct device *dev = tx_ring->dev;
486	int bi_size;
487
488	if (!dev)
489		return -ENOMEM;
490
491	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
492	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
493	if (!tx_ring->tx_bi)
494		goto err;
495
496	/* round up to nearest 4K */
497	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
498	/* add u32 for head writeback, align after this takes care of
499	 * guaranteeing this is at least one cache line in size
500	 */
501	tx_ring->size += sizeof(u32);
502	tx_ring->size = ALIGN(tx_ring->size, 4096);
503	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
504					   &tx_ring->dma, GFP_KERNEL);
505	if (!tx_ring->desc) {
506		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
507			 tx_ring->size);
508		goto err;
509	}
510
511	tx_ring->next_to_use = 0;
512	tx_ring->next_to_clean = 0;
513	return 0;
514
515err:
516	kfree(tx_ring->tx_bi);
517	tx_ring->tx_bi = NULL;
518	return -ENOMEM;
519}
520
521/**
522 * i40evf_clean_rx_ring - Free Rx buffers
523 * @rx_ring: ring to be cleaned
524 **/
525void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
526{
527	struct device *dev = rx_ring->dev;
528	struct i40e_rx_buffer *rx_bi;
529	unsigned long bi_size;
530	u16 i;
531
532	/* ring already cleared, nothing to do */
533	if (!rx_ring->rx_bi)
534		return;
535
536	if (ring_is_ps_enabled(rx_ring)) {
537		int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
538
539		rx_bi = &rx_ring->rx_bi[0];
540		if (rx_bi->hdr_buf) {
541			dma_free_coherent(dev,
542					  bufsz,
543					  rx_bi->hdr_buf,
544					  rx_bi->dma);
545			for (i = 0; i < rx_ring->count; i++) {
546				rx_bi = &rx_ring->rx_bi[i];
547				rx_bi->dma = 0;
548				rx_bi->hdr_buf = NULL;
549			}
550		}
551	}
552	/* Free all the Rx ring sk_buffs */
553	for (i = 0; i < rx_ring->count; i++) {
554		rx_bi = &rx_ring->rx_bi[i];
555		if (rx_bi->dma) {
556			dma_unmap_single(dev,
557					 rx_bi->dma,
558					 rx_ring->rx_buf_len,
559					 DMA_FROM_DEVICE);
560			rx_bi->dma = 0;
561		}
562		if (rx_bi->skb) {
563			dev_kfree_skb(rx_bi->skb);
564			rx_bi->skb = NULL;
565		}
566		if (rx_bi->page) {
567			if (rx_bi->page_dma) {
568				dma_unmap_page(dev,
569					       rx_bi->page_dma,
570					       PAGE_SIZE / 2,
571					       DMA_FROM_DEVICE);
572				rx_bi->page_dma = 0;
573			}
574			__free_page(rx_bi->page);
575			rx_bi->page = NULL;
576			rx_bi->page_offset = 0;
577		}
578	}
579
580	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
581	memset(rx_ring->rx_bi, 0, bi_size);
582
583	/* Zero out the descriptor ring */
584	memset(rx_ring->desc, 0, rx_ring->size);
585
586	rx_ring->next_to_clean = 0;
587	rx_ring->next_to_use = 0;
588}
589
590/**
591 * i40evf_free_rx_resources - Free Rx resources
592 * @rx_ring: ring to clean the resources from
593 *
594 * Free all receive software resources
595 **/
596void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
597{
598	i40evf_clean_rx_ring(rx_ring);
599	kfree(rx_ring->rx_bi);
600	rx_ring->rx_bi = NULL;
601
602	if (rx_ring->desc) {
603		dma_free_coherent(rx_ring->dev, rx_ring->size,
604				  rx_ring->desc, rx_ring->dma);
605		rx_ring->desc = NULL;
606	}
607}
608
609/**
610 * i40evf_alloc_rx_headers - allocate rx header buffers
611 * @rx_ring: ring to alloc buffers
612 *
613 * Allocate rx header buffers for the entire ring. As these are static,
614 * this is only called when setting up a new ring.
615 **/
616void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
617{
618	struct device *dev = rx_ring->dev;
619	struct i40e_rx_buffer *rx_bi;
620	dma_addr_t dma;
621	void *buffer;
622	int buf_size;
623	int i;
624
625	if (rx_ring->rx_bi[0].hdr_buf)
626		return;
627	/* Make sure the buffers don't cross cache line boundaries. */
628	buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
629	buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
630				    &dma, GFP_KERNEL);
631	if (!buffer)
632		return;
633	for (i = 0; i < rx_ring->count; i++) {
634		rx_bi = &rx_ring->rx_bi[i];
635		rx_bi->dma = dma + (i * buf_size);
636		rx_bi->hdr_buf = buffer + (i * buf_size);
637	}
638}
639
640/**
641 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
642 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
643 *
644 * Returns 0 on success, negative on failure
645 **/
646int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
647{
648	struct device *dev = rx_ring->dev;
649	int bi_size;
650
651	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
652	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
653	if (!rx_ring->rx_bi)
654		goto err;
655
656	u64_stats_init(&rx_ring->syncp);
657
658	/* Round up to nearest 4K */
659	rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
660		? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
661		: rx_ring->count * sizeof(union i40e_32byte_rx_desc);
662	rx_ring->size = ALIGN(rx_ring->size, 4096);
663	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
664					   &rx_ring->dma, GFP_KERNEL);
665
666	if (!rx_ring->desc) {
667		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
668			 rx_ring->size);
669		goto err;
670	}
671
672	rx_ring->next_to_clean = 0;
673	rx_ring->next_to_use = 0;
674
675	return 0;
676err:
677	kfree(rx_ring->rx_bi);
678	rx_ring->rx_bi = NULL;
679	return -ENOMEM;
680}
681
682/**
683 * i40e_release_rx_desc - Store the new tail and head values
684 * @rx_ring: ring to bump
685 * @val: new head index
686 **/
687static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
688{
689	rx_ring->next_to_use = val;
690	/* Force memory writes to complete before letting h/w
691	 * know there are new descriptors to fetch.  (Only
692	 * applicable for weak-ordered memory model archs,
693	 * such as IA-64).
694	 */
695	wmb();
696	writel(val, rx_ring->tail);
697}
698
699/**
700 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
701 * @rx_ring: ring to place buffers on
702 * @cleaned_count: number of buffers to replace
703 **/
704void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
705{
706	u16 i = rx_ring->next_to_use;
707	union i40e_rx_desc *rx_desc;
708	struct i40e_rx_buffer *bi;
709
710	/* do nothing if no valid netdev defined */
711	if (!rx_ring->netdev || !cleaned_count)
712		return;
713
714	while (cleaned_count--) {
715		rx_desc = I40E_RX_DESC(rx_ring, i);
716		bi = &rx_ring->rx_bi[i];
717
718		if (bi->skb) /* desc is in use */
719			goto no_buffers;
720		if (!bi->page) {
721			bi->page = alloc_page(GFP_ATOMIC);
722			if (!bi->page) {
723				rx_ring->rx_stats.alloc_page_failed++;
724				goto no_buffers;
725			}
726		}
727
728		if (!bi->page_dma) {
729			/* use a half page if we're re-using */
730			bi->page_offset ^= PAGE_SIZE / 2;
731			bi->page_dma = dma_map_page(rx_ring->dev,
732						    bi->page,
733						    bi->page_offset,
734						    PAGE_SIZE / 2,
735						    DMA_FROM_DEVICE);
736			if (dma_mapping_error(rx_ring->dev,
737					      bi->page_dma)) {
738				rx_ring->rx_stats.alloc_page_failed++;
739				bi->page_dma = 0;
740				goto no_buffers;
741			}
742		}
743
744		dma_sync_single_range_for_device(rx_ring->dev,
745						 bi->dma,
746						 0,
747						 rx_ring->rx_hdr_len,
748						 DMA_FROM_DEVICE);
749		/* Refresh the desc even if buffer_addrs didn't change
750		 * because each write-back erases this info.
751		 */
752		rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
753		rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
754		i++;
755		if (i == rx_ring->count)
756			i = 0;
757	}
758
759no_buffers:
760	if (rx_ring->next_to_use != i)
761		i40e_release_rx_desc(rx_ring, i);
762}
763
764/**
765 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
766 * @rx_ring: ring to place buffers on
767 * @cleaned_count: number of buffers to replace
768 **/
769void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
770{
771	u16 i = rx_ring->next_to_use;
772	union i40e_rx_desc *rx_desc;
773	struct i40e_rx_buffer *bi;
774	struct sk_buff *skb;
775
776	/* do nothing if no valid netdev defined */
777	if (!rx_ring->netdev || !cleaned_count)
778		return;
779
780	while (cleaned_count--) {
781		rx_desc = I40E_RX_DESC(rx_ring, i);
782		bi = &rx_ring->rx_bi[i];
783		skb = bi->skb;
784
785		if (!skb) {
786			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
787							rx_ring->rx_buf_len);
788			if (!skb) {
789				rx_ring->rx_stats.alloc_buff_failed++;
790				goto no_buffers;
791			}
792			/* initialize queue mapping */
793			skb_record_rx_queue(skb, rx_ring->queue_index);
794			bi->skb = skb;
795		}
796
797		if (!bi->dma) {
798			bi->dma = dma_map_single(rx_ring->dev,
799						 skb->data,
800						 rx_ring->rx_buf_len,
801						 DMA_FROM_DEVICE);
802			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
803				rx_ring->rx_stats.alloc_buff_failed++;
804				bi->dma = 0;
805				goto no_buffers;
806			}
807		}
808
809		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
810		rx_desc->read.hdr_addr = 0;
811		i++;
812		if (i == rx_ring->count)
813			i = 0;
814	}
815
816no_buffers:
817	if (rx_ring->next_to_use != i)
818		i40e_release_rx_desc(rx_ring, i);
819}
820
821/**
822 * i40e_receive_skb - Send a completed packet up the stack
823 * @rx_ring:  rx ring in play
824 * @skb: packet to send up
825 * @vlan_tag: vlan tag for packet
826 **/
827static void i40e_receive_skb(struct i40e_ring *rx_ring,
828			     struct sk_buff *skb, u16 vlan_tag)
829{
830	struct i40e_q_vector *q_vector = rx_ring->q_vector;
831	struct i40e_vsi *vsi = rx_ring->vsi;
832	u64 flags = vsi->back->flags;
833
834	if (vlan_tag & VLAN_VID_MASK)
835		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
836
837	if (flags & I40E_FLAG_IN_NETPOLL)
838		netif_rx(skb);
839	else
840		napi_gro_receive(&q_vector->napi, skb);
841}
842
843/**
844 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
845 * @vsi: the VSI we care about
846 * @skb: skb currently being received and modified
847 * @rx_status: status value of last descriptor in packet
848 * @rx_error: error value of last descriptor in packet
849 * @rx_ptype: ptype value of last descriptor in packet
850 **/
851static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
852				    struct sk_buff *skb,
853				    u32 rx_status,
854				    u32 rx_error,
855				    u16 rx_ptype)
856{
857	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
858	bool ipv4 = false, ipv6 = false;
859	bool ipv4_tunnel, ipv6_tunnel;
860	__wsum rx_udp_csum;
861	struct iphdr *iph;
862	__sum16 csum;
863
864	ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
865		     (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
866	ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
867		     (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
868
869	skb->ip_summed = CHECKSUM_NONE;
870
871	/* Rx csum enabled and ip headers found? */
872	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
873		return;
874
875	/* did the hardware decode the packet and checksum? */
876	if (!(rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
877		return;
878
879	/* both known and outer_ip must be set for the below code to work */
880	if (!(decoded.known && decoded.outer_ip))
881		return;
882
883	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
884	    decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
885		ipv4 = true;
886	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
887		 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
888		ipv6 = true;
889
890	if (ipv4 &&
891	    (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) |
892			 (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT))))
893		goto checksum_fail;
894
895	/* likely incorrect csum if alternate IP extension headers found */
896	if (ipv6 &&
897	    rx_status & (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
898		/* don't increment checksum err here, non-fatal err */
899		return;
900
901	/* there was some L4 error, count error and punt packet to the stack */
902	if (rx_error & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT))
903		goto checksum_fail;
904
905	/* handle packets that were not able to be checksummed due
906	 * to arrival speed, in this case the stack can compute
907	 * the csum.
908	 */
909	if (rx_error & (1 << I40E_RX_DESC_ERROR_PPRS_SHIFT))
910		return;
911
912	/* If VXLAN traffic has an outer UDPv4 checksum we need to check
913	 * it in the driver, hardware does not do it for us.
914	 * Since L3L4P bit was set we assume a valid IHL value (>=5)
915	 * so the total length of IPv4 header is IHL*4 bytes
916	 * The UDP_0 bit *may* bet set if the *inner* header is UDP
917	 */
918	if (ipv4_tunnel) {
919		skb->transport_header = skb->mac_header +
920					sizeof(struct ethhdr) +
921					(ip_hdr(skb)->ihl * 4);
922
923		/* Add 4 bytes for VLAN tagged packets */
924		skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
925					  skb->protocol == htons(ETH_P_8021AD))
926					  ? VLAN_HLEN : 0;
927
928		if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
929		    (udp_hdr(skb)->check != 0)) {
930			rx_udp_csum = udp_csum(skb);
931			iph = ip_hdr(skb);
932			csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
933						 (skb->len -
934						  skb_transport_offset(skb)),
935						 IPPROTO_UDP, rx_udp_csum);
936
937			if (udp_hdr(skb)->check != csum)
938				goto checksum_fail;
939
940		} /* else its GRE and so no outer UDP header */
941	}
942
943	skb->ip_summed = CHECKSUM_UNNECESSARY;
944	skb->csum_level = ipv4_tunnel || ipv6_tunnel;
945
946	return;
947
948checksum_fail:
949	vsi->back->hw_csum_rx_error++;
950}
951
952/**
953 * i40e_rx_hash - returns the hash value from the Rx descriptor
954 * @ring: descriptor ring
955 * @rx_desc: specific descriptor
956 **/
957static inline u32 i40e_rx_hash(struct i40e_ring *ring,
958			       union i40e_rx_desc *rx_desc)
959{
960	const __le64 rss_mask =
961		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
962			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
963
964	if ((ring->netdev->features & NETIF_F_RXHASH) &&
965	    (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
966		return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
967	else
968		return 0;
969}
970
971/**
972 * i40e_ptype_to_hash - get a hash type
973 * @ptype: the ptype value from the descriptor
974 *
975 * Returns a hash type to be used by skb_set_hash
976 **/
977static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
978{
979	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
980
981	if (!decoded.known)
982		return PKT_HASH_TYPE_NONE;
983
984	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
985	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
986		return PKT_HASH_TYPE_L4;
987	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
988		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
989		return PKT_HASH_TYPE_L3;
990	else
991		return PKT_HASH_TYPE_L2;
992}
993
994/**
995 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
996 * @rx_ring:  rx ring to clean
997 * @budget:   how many cleans we're allowed
998 *
999 * Returns true if there's any budget left (e.g. the clean is finished)
1000 **/
1001static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
1002{
1003	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1004	u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
1005	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1006	const int current_node = numa_node_id();
1007	struct i40e_vsi *vsi = rx_ring->vsi;
1008	u16 i = rx_ring->next_to_clean;
1009	union i40e_rx_desc *rx_desc;
1010	u32 rx_error, rx_status;
1011	u8 rx_ptype;
1012	u64 qword;
1013
1014	do {
1015		struct i40e_rx_buffer *rx_bi;
1016		struct sk_buff *skb;
1017		u16 vlan_tag;
1018		/* return some buffers to hardware, one at a time is too slow */
1019		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1020			i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
1021			cleaned_count = 0;
1022		}
1023
1024		i = rx_ring->next_to_clean;
1025		rx_desc = I40E_RX_DESC(rx_ring, i);
1026		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1027		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1028			I40E_RXD_QW1_STATUS_SHIFT;
1029
1030		if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
1031			break;
1032
1033		/* This memory barrier is needed to keep us from reading
1034		 * any other fields out of the rx_desc until we know the
1035		 * DD bit is set.
1036		 */
1037		dma_rmb();
1038		rx_bi = &rx_ring->rx_bi[i];
1039		skb = rx_bi->skb;
1040		if (likely(!skb)) {
1041			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1042							rx_ring->rx_hdr_len);
1043			if (!skb) {
1044				rx_ring->rx_stats.alloc_buff_failed++;
1045				break;
1046			}
1047
1048			/* initialize queue mapping */
1049			skb_record_rx_queue(skb, rx_ring->queue_index);
1050			/* we are reusing so sync this buffer for CPU use */
1051			dma_sync_single_range_for_cpu(rx_ring->dev,
1052						      rx_bi->dma,
1053						      0,
1054						      rx_ring->rx_hdr_len,
1055						      DMA_FROM_DEVICE);
1056		}
1057		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1058				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1059		rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
1060				I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1061		rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1062			 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1063
1064		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1065			   I40E_RXD_QW1_ERROR_SHIFT;
1066		rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
1067		rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
1068
1069		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1070			   I40E_RXD_QW1_PTYPE_SHIFT;
1071		prefetch(rx_bi->page);
1072		rx_bi->skb = NULL;
1073		cleaned_count++;
1074		if (rx_hbo || rx_sph) {
1075			int len;
1076			if (rx_hbo)
1077				len = I40E_RX_HDR_SIZE;
1078			else
1079				len = rx_header_len;
1080			memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1081		} else if (skb->len == 0) {
1082			int len;
1083
1084			len = (rx_packet_len > skb_headlen(skb) ?
1085				skb_headlen(skb) : rx_packet_len);
1086			memcpy(__skb_put(skb, len),
1087			       rx_bi->page + rx_bi->page_offset,
1088			       len);
1089			rx_bi->page_offset += len;
1090			rx_packet_len -= len;
1091		}
1092
1093		/* Get the rest of the data if this was a header split */
1094		if (rx_packet_len) {
1095			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1096					   rx_bi->page,
1097					   rx_bi->page_offset,
1098					   rx_packet_len);
1099
1100			skb->len += rx_packet_len;
1101			skb->data_len += rx_packet_len;
1102			skb->truesize += rx_packet_len;
1103
1104			if ((page_count(rx_bi->page) == 1) &&
1105			    (page_to_nid(rx_bi->page) == current_node))
1106				get_page(rx_bi->page);
1107			else
1108				rx_bi->page = NULL;
1109
1110			dma_unmap_page(rx_ring->dev,
1111				       rx_bi->page_dma,
1112				       PAGE_SIZE / 2,
1113				       DMA_FROM_DEVICE);
1114			rx_bi->page_dma = 0;
1115		}
1116		I40E_RX_INCREMENT(rx_ring, i);
1117
1118		if (unlikely(
1119		    !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1120			struct i40e_rx_buffer *next_buffer;
1121
1122			next_buffer = &rx_ring->rx_bi[i];
1123			next_buffer->skb = skb;
1124			rx_ring->rx_stats.non_eop_descs++;
1125			continue;
1126		}
1127
1128		/* ERR_MASK will only have valid bits if EOP set */
1129		if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1130			dev_kfree_skb_any(skb);
1131			/* TODO: shouldn't we increment a counter indicating the
1132			 * drop?
1133			 */
1134			continue;
1135		}
1136
1137		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1138			     i40e_ptype_to_hash(rx_ptype));
1139		/* probably a little skewed due to removing CRC */
1140		total_rx_bytes += skb->len;
1141		total_rx_packets++;
1142
1143		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1144
1145		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1146
1147		vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1148			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1149			 : 0;
1150#ifdef I40E_FCOE
1151		if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1152			dev_kfree_skb_any(skb);
1153			continue;
1154		}
1155#endif
1156		skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
1157		i40e_receive_skb(rx_ring, skb, vlan_tag);
1158
1159		rx_ring->netdev->last_rx = jiffies;
1160		rx_desc->wb.qword1.status_error_len = 0;
1161
1162	} while (likely(total_rx_packets < budget));
1163
1164	u64_stats_update_begin(&rx_ring->syncp);
1165	rx_ring->stats.packets += total_rx_packets;
1166	rx_ring->stats.bytes += total_rx_bytes;
1167	u64_stats_update_end(&rx_ring->syncp);
1168	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1169	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1170
1171	return total_rx_packets;
1172}
1173
1174/**
1175 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1176 * @rx_ring:  rx ring to clean
1177 * @budget:   how many cleans we're allowed
1178 *
1179 * Returns number of packets cleaned
1180 **/
1181static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1182{
1183	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1184	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1185	struct i40e_vsi *vsi = rx_ring->vsi;
1186	union i40e_rx_desc *rx_desc;
1187	u32 rx_error, rx_status;
1188	u16 rx_packet_len;
1189	u8 rx_ptype;
1190	u64 qword;
1191	u16 i;
1192
1193	do {
1194		struct i40e_rx_buffer *rx_bi;
1195		struct sk_buff *skb;
1196		u16 vlan_tag;
1197		/* return some buffers to hardware, one at a time is too slow */
1198		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1199			i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1200			cleaned_count = 0;
1201		}
1202
1203		i = rx_ring->next_to_clean;
1204		rx_desc = I40E_RX_DESC(rx_ring, i);
1205		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1206		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1207			I40E_RXD_QW1_STATUS_SHIFT;
1208
1209		if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)))
1210			break;
1211
1212		/* This memory barrier is needed to keep us from reading
1213		 * any other fields out of the rx_desc until we know the
1214		 * DD bit is set.
1215		 */
1216		dma_rmb();
1217
1218		rx_bi = &rx_ring->rx_bi[i];
1219		skb = rx_bi->skb;
1220		prefetch(skb->data);
1221
1222		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1223				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1224
1225		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1226			   I40E_RXD_QW1_ERROR_SHIFT;
1227		rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
1228
1229		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1230			   I40E_RXD_QW1_PTYPE_SHIFT;
1231		rx_bi->skb = NULL;
1232		cleaned_count++;
1233
1234		/* Get the header and possibly the whole packet
1235		 * If this is an skb from previous receive dma will be 0
1236		 */
1237		skb_put(skb, rx_packet_len);
1238		dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1239				 DMA_FROM_DEVICE);
1240		rx_bi->dma = 0;
1241
1242		I40E_RX_INCREMENT(rx_ring, i);
1243
1244		if (unlikely(
1245		    !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1246			rx_ring->rx_stats.non_eop_descs++;
1247			continue;
1248		}
1249
1250		/* ERR_MASK will only have valid bits if EOP set */
1251		if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1252			dev_kfree_skb_any(skb);
1253			/* TODO: shouldn't we increment a counter indicating the
1254			 * drop?
1255			 */
1256			continue;
1257		}
1258
1259		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1260			     i40e_ptype_to_hash(rx_ptype));
1261		/* probably a little skewed due to removing CRC */
1262		total_rx_bytes += skb->len;
1263		total_rx_packets++;
1264
1265		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1266
1267		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1268
1269		vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1270			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1271			 : 0;
1272		i40e_receive_skb(rx_ring, skb, vlan_tag);
1273
1274		rx_ring->netdev->last_rx = jiffies;
1275		rx_desc->wb.qword1.status_error_len = 0;
1276	} while (likely(total_rx_packets < budget));
1277
1278	u64_stats_update_begin(&rx_ring->syncp);
1279	rx_ring->stats.packets += total_rx_packets;
1280	rx_ring->stats.bytes += total_rx_bytes;
1281	u64_stats_update_end(&rx_ring->syncp);
1282	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1283	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1284
1285	return total_rx_packets;
1286}
1287
1288/**
1289 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1290 * @napi: napi struct with our devices info in it
1291 * @budget: amount of work driver is allowed to do this pass, in packets
1292 *
1293 * This function will clean all queues associated with a q_vector.
1294 *
1295 * Returns the amount of work done
1296 **/
1297int i40evf_napi_poll(struct napi_struct *napi, int budget)
1298{
1299	struct i40e_q_vector *q_vector =
1300			       container_of(napi, struct i40e_q_vector, napi);
1301	struct i40e_vsi *vsi = q_vector->vsi;
1302	struct i40e_ring *ring;
1303	bool clean_complete = true;
1304	bool arm_wb = false;
1305	int budget_per_ring;
1306	int cleaned;
1307
1308	if (test_bit(__I40E_DOWN, &vsi->state)) {
1309		napi_complete(napi);
1310		return 0;
1311	}
1312
1313	/* Since the actual Tx work is minimal, we can give the Tx a larger
1314	 * budget and be more aggressive about cleaning up the Tx descriptors.
1315	 */
1316	i40e_for_each_ring(ring, q_vector->tx) {
1317		clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1318		arm_wb |= ring->arm_wb;
1319	}
1320
1321	/* We attempt to distribute budget to each Rx queue fairly, but don't
1322	 * allow the budget to go below 1 because that would exit polling early.
1323	 */
1324	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1325
1326	i40e_for_each_ring(ring, q_vector->rx) {
1327		if (ring_is_ps_enabled(ring))
1328			cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1329		else
1330			cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1331		/* if we didn't clean as many as budgeted, we must be done */
1332		clean_complete &= (budget_per_ring != cleaned);
1333	}
1334
1335	/* If work not completed, return budget and polling will return */
1336	if (!clean_complete) {
1337		if (arm_wb)
1338			i40e_force_wb(vsi, q_vector);
1339		return budget;
1340	}
1341
1342	/* Work is done so exit the polling mode and re-enable the interrupt */
1343	napi_complete(napi);
1344	if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) ||
1345	    ITR_IS_DYNAMIC(vsi->tx_itr_setting))
1346		i40e_update_dynamic_itr(q_vector);
1347
1348	if (!test_bit(__I40E_DOWN, &vsi->state))
1349		i40evf_irq_enable_queues(vsi->back, 1 << q_vector->v_idx);
1350
1351	return 0;
1352}
1353
1354/**
1355 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1356 * @skb:     send buffer
1357 * @tx_ring: ring to send buffer on
1358 * @flags:   the tx flags to be set
1359 *
1360 * Checks the skb and set up correspondingly several generic transmit flags
1361 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1362 *
1363 * Returns error code indicate the frame should be dropped upon error and the
1364 * otherwise  returns 0 to indicate the flags has been set properly.
1365 **/
1366static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
1367				      struct i40e_ring *tx_ring,
1368				      u32 *flags)
1369{
1370	__be16 protocol = skb->protocol;
1371	u32  tx_flags = 0;
1372
1373	if (protocol == htons(ETH_P_8021Q) &&
1374	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1375		/* When HW VLAN acceleration is turned off by the user the
1376		 * stack sets the protocol to 8021q so that the driver
1377		 * can take any steps required to support the SW only
1378		 * VLAN handling.  In our case the driver doesn't need
1379		 * to take any further steps so just set the protocol
1380		 * to the encapsulated ethertype.
1381		 */
1382		skb->protocol = vlan_get_protocol(skb);
1383		goto out;
1384	}
1385
1386	/* if we have a HW VLAN tag being added, default to the HW one */
1387	if (skb_vlan_tag_present(skb)) {
1388		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1389		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1390	/* else if it is a SW VLAN, check the next protocol and store the tag */
1391	} else if (protocol == htons(ETH_P_8021Q)) {
1392		struct vlan_hdr *vhdr, _vhdr;
1393		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1394		if (!vhdr)
1395			return -EINVAL;
1396
1397		protocol = vhdr->h_vlan_encapsulated_proto;
1398		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1399		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1400	}
1401
1402out:
1403	*flags = tx_flags;
1404	return 0;
1405}
1406
1407/**
1408 * i40e_tso - set up the tso context descriptor
1409 * @tx_ring:  ptr to the ring to send
1410 * @skb:      ptr to the skb we're sending
1411 * @tx_flags: the collected send information
1412 * @protocol: the send protocol
1413 * @hdr_len:  ptr to the size of the packet header
1414 * @cd_tunneling: ptr to context descriptor bits
1415 *
1416 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1417 **/
1418static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1419		    u32 tx_flags, __be16 protocol, u8 *hdr_len,
1420		    u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling)
1421{
1422	u32 cd_cmd, cd_tso_len, cd_mss;
1423	struct ipv6hdr *ipv6h;
1424	struct tcphdr *tcph;
1425	struct iphdr *iph;
1426	u32 l4len;
1427	int err;
1428
1429	if (!skb_is_gso(skb))
1430		return 0;
1431
1432	err = skb_cow_head(skb, 0);
1433	if (err < 0)
1434		return err;
1435
1436	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1437	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1438
1439	if (iph->version == 4) {
1440		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1441		iph->tot_len = 0;
1442		iph->check = 0;
1443		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1444						 0, IPPROTO_TCP, 0);
1445	} else if (ipv6h->version == 6) {
1446		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1447		ipv6h->payload_len = 0;
1448		tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1449					       0, IPPROTO_TCP, 0);
1450	}
1451
1452	l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1453	*hdr_len = (skb->encapsulation
1454		    ? (skb_inner_transport_header(skb) - skb->data)
1455		    : skb_transport_offset(skb)) + l4len;
1456
1457	/* find the field values */
1458	cd_cmd = I40E_TX_CTX_DESC_TSO;
1459	cd_tso_len = skb->len - *hdr_len;
1460	cd_mss = skb_shinfo(skb)->gso_size;
1461	*cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1462				((u64)cd_tso_len <<
1463				 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1464				((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1465	return 1;
1466}
1467
1468/**
1469 * i40e_tx_enable_csum - Enable Tx checksum offloads
1470 * @skb: send buffer
1471 * @tx_flags: Tx flags currently set
1472 * @td_cmd: Tx descriptor command bits to set
1473 * @td_offset: Tx descriptor header offsets to set
1474 * @cd_tunneling: ptr to context desc bits
1475 **/
1476static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags,
1477				u32 *td_cmd, u32 *td_offset,
1478				struct i40e_ring *tx_ring,
1479				u32 *cd_tunneling)
1480{
1481	struct ipv6hdr *this_ipv6_hdr;
1482	unsigned int this_tcp_hdrlen;
1483	struct iphdr *this_ip_hdr;
1484	u32 network_hdr_len;
1485	u8 l4_hdr = 0;
1486	u32 l4_tunnel = 0;
1487
1488	if (skb->encapsulation) {
1489		switch (ip_hdr(skb)->protocol) {
1490		case IPPROTO_UDP:
1491			l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
1492			break;
1493		default:
1494			return;
1495		}
1496		network_hdr_len = skb_inner_network_header_len(skb);
1497		this_ip_hdr = inner_ip_hdr(skb);
1498		this_ipv6_hdr = inner_ipv6_hdr(skb);
1499		this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1500
1501		if (tx_flags & I40E_TX_FLAGS_IPV4) {
1502
1503			if (tx_flags & I40E_TX_FLAGS_TSO) {
1504				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1505				ip_hdr(skb)->check = 0;
1506			} else {
1507				*cd_tunneling |=
1508					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1509			}
1510		} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1511			*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1512			if (tx_flags & I40E_TX_FLAGS_TSO)
1513				ip_hdr(skb)->check = 0;
1514		}
1515
1516		/* Now set the ctx descriptor fields */
1517		*cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1518				   I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT      |
1519				   l4_tunnel                             |
1520				   ((skb_inner_network_offset(skb) -
1521					skb_transport_offset(skb)) >> 1) <<
1522				   I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1523		if (this_ip_hdr->version == 6) {
1524			tx_flags &= ~I40E_TX_FLAGS_IPV4;
1525			tx_flags |= I40E_TX_FLAGS_IPV6;
1526		}
1527
1528
1529	} else {
1530		network_hdr_len = skb_network_header_len(skb);
1531		this_ip_hdr = ip_hdr(skb);
1532		this_ipv6_hdr = ipv6_hdr(skb);
1533		this_tcp_hdrlen = tcp_hdrlen(skb);
1534	}
1535
1536	/* Enable IP checksum offloads */
1537	if (tx_flags & I40E_TX_FLAGS_IPV4) {
1538		l4_hdr = this_ip_hdr->protocol;
1539		/* the stack computes the IP header already, the only time we
1540		 * need the hardware to recompute it is in the case of TSO.
1541		 */
1542		if (tx_flags & I40E_TX_FLAGS_TSO) {
1543			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1544			this_ip_hdr->check = 0;
1545		} else {
1546			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1547		}
1548		/* Now set the td_offset for IP header length */
1549		*td_offset = (network_hdr_len >> 2) <<
1550			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1551	} else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1552		l4_hdr = this_ipv6_hdr->nexthdr;
1553		*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1554		/* Now set the td_offset for IP header length */
1555		*td_offset = (network_hdr_len >> 2) <<
1556			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1557	}
1558	/* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1559	*td_offset |= (skb_network_offset(skb) >> 1) <<
1560		       I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1561
1562	/* Enable L4 checksum offloads */
1563	switch (l4_hdr) {
1564	case IPPROTO_TCP:
1565		/* enable checksum offloads */
1566		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1567		*td_offset |= (this_tcp_hdrlen >> 2) <<
1568			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1569		break;
1570	case IPPROTO_SCTP:
1571		/* enable SCTP checksum offload */
1572		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1573		*td_offset |= (sizeof(struct sctphdr) >> 2) <<
1574			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1575		break;
1576	case IPPROTO_UDP:
1577		/* enable UDP checksum offload */
1578		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1579		*td_offset |= (sizeof(struct udphdr) >> 2) <<
1580			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1581		break;
1582	default:
1583		break;
1584	}
1585}
1586
1587/**
1588 * i40e_create_tx_ctx Build the Tx context descriptor
1589 * @tx_ring:  ring to create the descriptor on
1590 * @cd_type_cmd_tso_mss: Quad Word 1
1591 * @cd_tunneling: Quad Word 0 - bits 0-31
1592 * @cd_l2tag2: Quad Word 0 - bits 32-63
1593 **/
1594static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1595			       const u64 cd_type_cmd_tso_mss,
1596			       const u32 cd_tunneling, const u32 cd_l2tag2)
1597{
1598	struct i40e_tx_context_desc *context_desc;
1599	int i = tx_ring->next_to_use;
1600
1601	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1602	    !cd_tunneling && !cd_l2tag2)
1603		return;
1604
1605	/* grab the next descriptor */
1606	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1607
1608	i++;
1609	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1610
1611	/* cpu_to_le32 and assign to struct fields */
1612	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1613	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1614	context_desc->rsvd = cpu_to_le16(0);
1615	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1616}
1617
1618 /**
1619 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1620 * @skb:      send buffer
1621 * @tx_flags: collected send information
1622 *
1623 * Note: Our HW can't scatter-gather more than 8 fragments to build
1624 * a packet on the wire and so we need to figure out the cases where we
1625 * need to linearize the skb.
1626 **/
1627static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1628{
1629	struct skb_frag_struct *frag;
1630	bool linearize = false;
1631	unsigned int size = 0;
1632	u16 num_frags;
1633	u16 gso_segs;
1634
1635	num_frags = skb_shinfo(skb)->nr_frags;
1636	gso_segs = skb_shinfo(skb)->gso_segs;
1637
1638	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1639		u16 j = 0;
1640
1641		if (num_frags < (I40E_MAX_BUFFER_TXD))
1642			goto linearize_chk_done;
1643		/* try the simple math, if we have too many frags per segment */
1644		if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1645		    I40E_MAX_BUFFER_TXD) {
1646			linearize = true;
1647			goto linearize_chk_done;
1648		}
1649		frag = &skb_shinfo(skb)->frags[0];
1650		/* we might still have more fragments per segment */
1651		do {
1652			size += skb_frag_size(frag);
1653			frag++; j++;
1654			if ((size >= skb_shinfo(skb)->gso_size) &&
1655			    (j < I40E_MAX_BUFFER_TXD)) {
1656				size = (size % skb_shinfo(skb)->gso_size);
1657				j = (size) ? 1 : 0;
1658			}
1659			if (j == I40E_MAX_BUFFER_TXD) {
1660				linearize = true;
1661				break;
1662			}
1663			num_frags--;
1664		} while (num_frags);
1665	} else {
1666		if (num_frags >= I40E_MAX_BUFFER_TXD)
1667			linearize = true;
1668	}
1669
1670linearize_chk_done:
1671	return linearize;
1672}
1673
1674/**
1675 * i40e_tx_map - Build the Tx descriptor
1676 * @tx_ring:  ring to send buffer on
1677 * @skb:      send buffer
1678 * @first:    first buffer info buffer to use
1679 * @tx_flags: collected send information
1680 * @hdr_len:  size of the packet header
1681 * @td_cmd:   the command field in the descriptor
1682 * @td_offset: offset for checksum or crc
1683 **/
1684static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1685			struct i40e_tx_buffer *first, u32 tx_flags,
1686			const u8 hdr_len, u32 td_cmd, u32 td_offset)
1687{
1688	unsigned int data_len = skb->data_len;
1689	unsigned int size = skb_headlen(skb);
1690	struct skb_frag_struct *frag;
1691	struct i40e_tx_buffer *tx_bi;
1692	struct i40e_tx_desc *tx_desc;
1693	u16 i = tx_ring->next_to_use;
1694	u32 td_tag = 0;
1695	dma_addr_t dma;
1696	u16 gso_segs;
1697
1698	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1699		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1700		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1701			 I40E_TX_FLAGS_VLAN_SHIFT;
1702	}
1703
1704	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1705		gso_segs = skb_shinfo(skb)->gso_segs;
1706	else
1707		gso_segs = 1;
1708
1709	/* multiply data chunks by size of headers */
1710	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1711	first->gso_segs = gso_segs;
1712	first->skb = skb;
1713	first->tx_flags = tx_flags;
1714
1715	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1716
1717	tx_desc = I40E_TX_DESC(tx_ring, i);
1718	tx_bi = first;
1719
1720	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1721		if (dma_mapping_error(tx_ring->dev, dma))
1722			goto dma_error;
1723
1724		/* record length, and DMA address */
1725		dma_unmap_len_set(tx_bi, len, size);
1726		dma_unmap_addr_set(tx_bi, dma, dma);
1727
1728		tx_desc->buffer_addr = cpu_to_le64(dma);
1729
1730		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1731			tx_desc->cmd_type_offset_bsz =
1732				build_ctob(td_cmd, td_offset,
1733					   I40E_MAX_DATA_PER_TXD, td_tag);
1734
1735			tx_desc++;
1736			i++;
1737			if (i == tx_ring->count) {
1738				tx_desc = I40E_TX_DESC(tx_ring, 0);
1739				i = 0;
1740			}
1741
1742			dma += I40E_MAX_DATA_PER_TXD;
1743			size -= I40E_MAX_DATA_PER_TXD;
1744
1745			tx_desc->buffer_addr = cpu_to_le64(dma);
1746		}
1747
1748		if (likely(!data_len))
1749			break;
1750
1751		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1752							  size, td_tag);
1753
1754		tx_desc++;
1755		i++;
1756		if (i == tx_ring->count) {
1757			tx_desc = I40E_TX_DESC(tx_ring, 0);
1758			i = 0;
1759		}
1760
1761		size = skb_frag_size(frag);
1762		data_len -= size;
1763
1764		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1765				       DMA_TO_DEVICE);
1766
1767		tx_bi = &tx_ring->tx_bi[i];
1768	}
1769
1770	/* Place RS bit on last descriptor of any packet that spans across the
1771	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1772	 */
1773#define WB_STRIDE 0x3
1774	if (((i & WB_STRIDE) != WB_STRIDE) &&
1775	    (first <= &tx_ring->tx_bi[i]) &&
1776	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1777		tx_desc->cmd_type_offset_bsz =
1778			build_ctob(td_cmd, td_offset, size, td_tag) |
1779			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1780					 I40E_TXD_QW1_CMD_SHIFT);
1781	} else {
1782		tx_desc->cmd_type_offset_bsz =
1783			build_ctob(td_cmd, td_offset, size, td_tag) |
1784			cpu_to_le64((u64)I40E_TXD_CMD <<
1785					 I40E_TXD_QW1_CMD_SHIFT);
1786	}
1787
1788	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1789						 tx_ring->queue_index),
1790			     first->bytecount);
1791
1792	/* set the timestamp */
1793	first->time_stamp = jiffies;
1794
1795	/* Force memory writes to complete before letting h/w
1796	 * know there are new descriptors to fetch.  (Only
1797	 * applicable for weak-ordered memory model archs,
1798	 * such as IA-64).
1799	 */
1800	wmb();
1801
1802	/* set next_to_watch value indicating a packet is present */
1803	first->next_to_watch = tx_desc;
1804
1805	i++;
1806	if (i == tx_ring->count)
1807		i = 0;
1808
1809	tx_ring->next_to_use = i;
1810
1811	/* notify HW of packet */
1812	writel(i, tx_ring->tail);
1813
1814	return;
1815
1816dma_error:
1817	dev_info(tx_ring->dev, "TX DMA map failed\n");
1818
1819	/* clear dma mappings for failed tx_bi map */
1820	for (;;) {
1821		tx_bi = &tx_ring->tx_bi[i];
1822		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1823		if (tx_bi == first)
1824			break;
1825		if (i == 0)
1826			i = tx_ring->count;
1827		i--;
1828	}
1829
1830	tx_ring->next_to_use = i;
1831}
1832
1833/**
1834 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
1835 * @tx_ring: the ring to be checked
1836 * @size:    the size buffer we want to assure is available
1837 *
1838 * Returns -EBUSY if a stop is needed, else 0
1839 **/
1840static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1841{
1842	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1843	/* Memory barrier before checking head and tail */
1844	smp_mb();
1845
1846	/* Check again in a case another CPU has just made room available. */
1847	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1848		return -EBUSY;
1849
1850	/* A reprieve! - use start_queue because it doesn't call schedule */
1851	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1852	++tx_ring->tx_stats.restart_queue;
1853	return 0;
1854}
1855
1856/**
1857 * i40e_maybe_stop_tx - 1st level check for tx stop conditions
1858 * @tx_ring: the ring to be checked
1859 * @size:    the size buffer we want to assure is available
1860 *
1861 * Returns 0 if stop is not needed
1862 **/
1863static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1864{
1865	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1866		return 0;
1867	return __i40e_maybe_stop_tx(tx_ring, size);
1868}
1869
1870/**
1871 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
1872 * @skb:     send buffer
1873 * @tx_ring: ring to send buffer on
1874 *
1875 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1876 * there is not enough descriptors available in this ring since we need at least
1877 * one descriptor.
1878 **/
1879static int i40e_xmit_descriptor_count(struct sk_buff *skb,
1880				      struct i40e_ring *tx_ring)
1881{
1882	unsigned int f;
1883	int count = 0;
1884
1885	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1886	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1887	 *       + 4 desc gap to avoid the cache line where head is,
1888	 *       + 1 desc for context descriptor,
1889	 * otherwise try next time
1890	 */
1891	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1892		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1893
1894	count += TXD_USE_COUNT(skb_headlen(skb));
1895	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1896		tx_ring->tx_stats.tx_busy++;
1897		return 0;
1898	}
1899	return count;
1900}
1901
1902/**
1903 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1904 * @skb:     send buffer
1905 * @tx_ring: ring to send buffer on
1906 *
1907 * Returns NETDEV_TX_OK if sent, else an error code
1908 **/
1909static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1910					struct i40e_ring *tx_ring)
1911{
1912	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1913	u32 cd_tunneling = 0, cd_l2tag2 = 0;
1914	struct i40e_tx_buffer *first;
1915	u32 td_offset = 0;
1916	u32 tx_flags = 0;
1917	__be16 protocol;
1918	u32 td_cmd = 0;
1919	u8 hdr_len = 0;
1920	int tso;
1921	if (0 == i40e_xmit_descriptor_count(skb, tx_ring))
1922		return NETDEV_TX_BUSY;
1923
1924	/* prepare the xmit flags */
1925	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1926		goto out_drop;
1927
1928	/* obtain protocol of skb */
1929	protocol = vlan_get_protocol(skb);
1930
1931	/* record the location of the first descriptor for this packet */
1932	first = &tx_ring->tx_bi[tx_ring->next_to_use];
1933
1934	/* setup IPv4/IPv6 offloads */
1935	if (protocol == htons(ETH_P_IP))
1936		tx_flags |= I40E_TX_FLAGS_IPV4;
1937	else if (protocol == htons(ETH_P_IPV6))
1938		tx_flags |= I40E_TX_FLAGS_IPV6;
1939
1940	tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len,
1941		       &cd_type_cmd_tso_mss, &cd_tunneling);
1942
1943	if (tso < 0)
1944		goto out_drop;
1945	else if (tso)
1946		tx_flags |= I40E_TX_FLAGS_TSO;
1947
1948	if (i40e_chk_linearize(skb, tx_flags))
1949		if (skb_linearize(skb))
1950			goto out_drop;
1951
1952	skb_tx_timestamp(skb);
1953
1954	/* always enable CRC insertion offload */
1955	td_cmd |= I40E_TX_DESC_CMD_ICRC;
1956
1957	/* Always offload the checksum, since it's in the data descriptor */
1958	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1959		tx_flags |= I40E_TX_FLAGS_CSUM;
1960
1961		i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset,
1962				    tx_ring, &cd_tunneling);
1963	}
1964
1965	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
1966			   cd_tunneling, cd_l2tag2);
1967
1968	i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
1969		    td_cmd, td_offset);
1970
1971	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
1972
1973	return NETDEV_TX_OK;
1974
1975out_drop:
1976	dev_kfree_skb_any(skb);
1977	return NETDEV_TX_OK;
1978}
1979
1980/**
1981 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1982 * @skb:    send buffer
1983 * @netdev: network interface device structure
1984 *
1985 * Returns NETDEV_TX_OK if sent, else an error code
1986 **/
1987netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1988{
1989	struct i40evf_adapter *adapter = netdev_priv(netdev);
1990	struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
1991
1992	/* hardware can't handle really short frames, hardware padding works
1993	 * beyond this point
1994	 */
1995	if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
1996		if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
1997			return NETDEV_TX_OK;
1998		skb->len = I40E_MIN_TX_LEN;
1999		skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2000	}
2001
2002	return i40e_xmit_frame_ring(skb, tx_ring);
2003}
2004