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#define WB_STRIDE 0x3
144
145/**
146 * i40e_clean_tx_irq - Reclaim resources after transmit completes
147 * @tx_ring:  tx ring to clean
148 * @budget:   how many cleans we're allowed
149 *
150 * Returns true if there's any budget left (e.g. the clean is finished)
151 **/
152static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
153{
154	u16 i = tx_ring->next_to_clean;
155	struct i40e_tx_buffer *tx_buf;
156	struct i40e_tx_desc *tx_head;
157	struct i40e_tx_desc *tx_desc;
158	unsigned int total_packets = 0;
159	unsigned int total_bytes = 0;
160
161	tx_buf = &tx_ring->tx_bi[i];
162	tx_desc = I40E_TX_DESC(tx_ring, i);
163	i -= tx_ring->count;
164
165	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
166
167	do {
168		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
169
170		/* if next_to_watch is not set then there is no work pending */
171		if (!eop_desc)
172			break;
173
174		/* prevent any other reads prior to eop_desc */
175		read_barrier_depends();
176
177		/* we have caught up to head, no work left to do */
178		if (tx_head == tx_desc)
179			break;
180
181		/* clear next_to_watch to prevent false hangs */
182		tx_buf->next_to_watch = NULL;
183
184		/* update the statistics for this packet */
185		total_bytes += tx_buf->bytecount;
186		total_packets += tx_buf->gso_segs;
187
188		/* free the skb */
189		dev_kfree_skb_any(tx_buf->skb);
190
191		/* unmap skb header data */
192		dma_unmap_single(tx_ring->dev,
193				 dma_unmap_addr(tx_buf, dma),
194				 dma_unmap_len(tx_buf, len),
195				 DMA_TO_DEVICE);
196
197		/* clear tx_buffer data */
198		tx_buf->skb = NULL;
199		dma_unmap_len_set(tx_buf, len, 0);
200
201		/* unmap remaining buffers */
202		while (tx_desc != eop_desc) {
203
204			tx_buf++;
205			tx_desc++;
206			i++;
207			if (unlikely(!i)) {
208				i -= tx_ring->count;
209				tx_buf = tx_ring->tx_bi;
210				tx_desc = I40E_TX_DESC(tx_ring, 0);
211			}
212
213			/* unmap any remaining paged data */
214			if (dma_unmap_len(tx_buf, len)) {
215				dma_unmap_page(tx_ring->dev,
216					       dma_unmap_addr(tx_buf, dma),
217					       dma_unmap_len(tx_buf, len),
218					       DMA_TO_DEVICE);
219				dma_unmap_len_set(tx_buf, len, 0);
220			}
221		}
222
223		/* move us one more past the eop_desc for start of next pkt */
224		tx_buf++;
225		tx_desc++;
226		i++;
227		if (unlikely(!i)) {
228			i -= tx_ring->count;
229			tx_buf = tx_ring->tx_bi;
230			tx_desc = I40E_TX_DESC(tx_ring, 0);
231		}
232
233		prefetch(tx_desc);
234
235		/* update budget accounting */
236		budget--;
237	} while (likely(budget));
238
239	i += tx_ring->count;
240	tx_ring->next_to_clean = i;
241	u64_stats_update_begin(&tx_ring->syncp);
242	tx_ring->stats.bytes += total_bytes;
243	tx_ring->stats.packets += total_packets;
244	u64_stats_update_end(&tx_ring->syncp);
245	tx_ring->q_vector->tx.total_bytes += total_bytes;
246	tx_ring->q_vector->tx.total_packets += total_packets;
247
248	/* check to see if there are any non-cache aligned descriptors
249	 * waiting to be written back, and kick the hardware to force
250	 * them to be written back in case of napi polling
251	 */
252	if (budget &&
253	    !((i & WB_STRIDE) == WB_STRIDE) &&
254	    !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
255	    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
256		tx_ring->arm_wb = true;
257
258	netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
259						      tx_ring->queue_index),
260				  total_packets, total_bytes);
261
262#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
263	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
264		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
265		/* Make sure that anybody stopping the queue after this
266		 * sees the new next_to_clean.
267		 */
268		smp_mb();
269		if (__netif_subqueue_stopped(tx_ring->netdev,
270					     tx_ring->queue_index) &&
271		   !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
272			netif_wake_subqueue(tx_ring->netdev,
273					    tx_ring->queue_index);
274			++tx_ring->tx_stats.restart_queue;
275		}
276	}
277
278	return !!budget;
279}
280
281/**
282 * i40evf_force_wb -Arm hardware to do a wb on noncache aligned descriptors
283 * @vsi: the VSI we care about
284 * @q_vector: the vector  on which to force writeback
285 *
286 **/
287static void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
288{
289	u16 flags = q_vector->tx.ring[0].flags;
290
291	if (flags & I40E_TXR_FLAGS_WB_ON_ITR) {
292		u32 val;
293
294		if (q_vector->arm_wb_state)
295			return;
296
297		val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK;
298
299		wr32(&vsi->back->hw,
300		     I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
301					  vsi->base_vector - 1),
302		     val);
303		q_vector->arm_wb_state = true;
304	} else {
305		u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
306			  I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
307			  I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
308			  I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK;
309			  /* allow 00 to be written to the index */
310
311		wr32(&vsi->back->hw,
312		     I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
313					  vsi->base_vector - 1), val);
314	}
315}
316
317/**
318 * i40e_set_new_dynamic_itr - Find new ITR level
319 * @rc: structure containing ring performance data
320 *
321 * Returns true if ITR changed, false if not
322 *
323 * Stores a new ITR value based on packets and byte counts during
324 * the last interrupt.  The advantage of per interrupt computation
325 * is faster updates and more accurate ITR for the current traffic
326 * pattern.  Constants in this function were computed based on
327 * theoretical maximum wire speed and thresholds were set based on
328 * testing data as well as attempting to minimize response time
329 * while increasing bulk throughput.
330 **/
331static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
332{
333	enum i40e_latency_range new_latency_range = rc->latency_range;
334	struct i40e_q_vector *qv = rc->ring->q_vector;
335	u32 new_itr = rc->itr;
336	int bytes_per_int;
337	int usecs;
338
339	if (rc->total_packets == 0 || !rc->itr)
340		return false;
341
342	/* simple throttlerate management
343	 *   0-10MB/s   lowest (50000 ints/s)
344	 *  10-20MB/s   low    (20000 ints/s)
345	 *  20-1249MB/s bulk   (18000 ints/s)
346	 *  > 40000 Rx packets per second (8000 ints/s)
347	 *
348	 * The math works out because the divisor is in 10^(-6) which
349	 * turns the bytes/us input value into MB/s values, but
350	 * make sure to use usecs, as the register values written
351	 * are in 2 usec increments in the ITR registers, and make sure
352	 * to use the smoothed values that the countdown timer gives us.
353	 */
354	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
355	bytes_per_int = rc->total_bytes / usecs;
356
357	switch (new_latency_range) {
358	case I40E_LOWEST_LATENCY:
359		if (bytes_per_int > 10)
360			new_latency_range = I40E_LOW_LATENCY;
361		break;
362	case I40E_LOW_LATENCY:
363		if (bytes_per_int > 20)
364			new_latency_range = I40E_BULK_LATENCY;
365		else if (bytes_per_int <= 10)
366			new_latency_range = I40E_LOWEST_LATENCY;
367		break;
368	case I40E_BULK_LATENCY:
369	case I40E_ULTRA_LATENCY:
370	default:
371		if (bytes_per_int <= 20)
372			new_latency_range = I40E_LOW_LATENCY;
373		break;
374	}
375
376	/* this is to adjust RX more aggressively when streaming small
377	 * packets.  The value of 40000 was picked as it is just beyond
378	 * what the hardware can receive per second if in low latency
379	 * mode.
380	 */
381#define RX_ULTRA_PACKET_RATE 40000
382
383	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
384	    (&qv->rx == rc))
385		new_latency_range = I40E_ULTRA_LATENCY;
386
387	rc->latency_range = new_latency_range;
388
389	switch (new_latency_range) {
390	case I40E_LOWEST_LATENCY:
391		new_itr = I40E_ITR_50K;
392		break;
393	case I40E_LOW_LATENCY:
394		new_itr = I40E_ITR_20K;
395		break;
396	case I40E_BULK_LATENCY:
397		new_itr = I40E_ITR_18K;
398		break;
399	case I40E_ULTRA_LATENCY:
400		new_itr = I40E_ITR_8K;
401		break;
402	default:
403		break;
404	}
405
406	rc->total_bytes = 0;
407	rc->total_packets = 0;
408
409	if (new_itr != rc->itr) {
410		rc->itr = new_itr;
411		return true;
412	}
413
414	return false;
415}
416
417/*
418 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
419 * @tx_ring: the tx ring to set up
420 *
421 * Return 0 on success, negative on error
422 **/
423int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
424{
425	struct device *dev = tx_ring->dev;
426	int bi_size;
427
428	if (!dev)
429		return -ENOMEM;
430
431	/* warn if we are about to overwrite the pointer */
432	WARN_ON(tx_ring->tx_bi);
433	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
434	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
435	if (!tx_ring->tx_bi)
436		goto err;
437
438	/* round up to nearest 4K */
439	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
440	/* add u32 for head writeback, align after this takes care of
441	 * guaranteeing this is at least one cache line in size
442	 */
443	tx_ring->size += sizeof(u32);
444	tx_ring->size = ALIGN(tx_ring->size, 4096);
445	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
446					   &tx_ring->dma, GFP_KERNEL);
447	if (!tx_ring->desc) {
448		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
449			 tx_ring->size);
450		goto err;
451	}
452
453	tx_ring->next_to_use = 0;
454	tx_ring->next_to_clean = 0;
455	return 0;
456
457err:
458	kfree(tx_ring->tx_bi);
459	tx_ring->tx_bi = NULL;
460	return -ENOMEM;
461}
462
463/**
464 * i40evf_clean_rx_ring - Free Rx buffers
465 * @rx_ring: ring to be cleaned
466 **/
467void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
468{
469	struct device *dev = rx_ring->dev;
470	struct i40e_rx_buffer *rx_bi;
471	unsigned long bi_size;
472	u16 i;
473
474	/* ring already cleared, nothing to do */
475	if (!rx_ring->rx_bi)
476		return;
477
478	if (ring_is_ps_enabled(rx_ring)) {
479		int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
480
481		rx_bi = &rx_ring->rx_bi[0];
482		if (rx_bi->hdr_buf) {
483			dma_free_coherent(dev,
484					  bufsz,
485					  rx_bi->hdr_buf,
486					  rx_bi->dma);
487			for (i = 0; i < rx_ring->count; i++) {
488				rx_bi = &rx_ring->rx_bi[i];
489				rx_bi->dma = 0;
490				rx_bi->hdr_buf = NULL;
491			}
492		}
493	}
494	/* Free all the Rx ring sk_buffs */
495	for (i = 0; i < rx_ring->count; i++) {
496		rx_bi = &rx_ring->rx_bi[i];
497		if (rx_bi->dma) {
498			dma_unmap_single(dev,
499					 rx_bi->dma,
500					 rx_ring->rx_buf_len,
501					 DMA_FROM_DEVICE);
502			rx_bi->dma = 0;
503		}
504		if (rx_bi->skb) {
505			dev_kfree_skb(rx_bi->skb);
506			rx_bi->skb = NULL;
507		}
508		if (rx_bi->page) {
509			if (rx_bi->page_dma) {
510				dma_unmap_page(dev,
511					       rx_bi->page_dma,
512					       PAGE_SIZE / 2,
513					       DMA_FROM_DEVICE);
514				rx_bi->page_dma = 0;
515			}
516			__free_page(rx_bi->page);
517			rx_bi->page = NULL;
518			rx_bi->page_offset = 0;
519		}
520	}
521
522	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
523	memset(rx_ring->rx_bi, 0, bi_size);
524
525	/* Zero out the descriptor ring */
526	memset(rx_ring->desc, 0, rx_ring->size);
527
528	rx_ring->next_to_clean = 0;
529	rx_ring->next_to_use = 0;
530}
531
532/**
533 * i40evf_free_rx_resources - Free Rx resources
534 * @rx_ring: ring to clean the resources from
535 *
536 * Free all receive software resources
537 **/
538void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
539{
540	i40evf_clean_rx_ring(rx_ring);
541	kfree(rx_ring->rx_bi);
542	rx_ring->rx_bi = NULL;
543
544	if (rx_ring->desc) {
545		dma_free_coherent(rx_ring->dev, rx_ring->size,
546				  rx_ring->desc, rx_ring->dma);
547		rx_ring->desc = NULL;
548	}
549}
550
551/**
552 * i40evf_alloc_rx_headers - allocate rx header buffers
553 * @rx_ring: ring to alloc buffers
554 *
555 * Allocate rx header buffers for the entire ring. As these are static,
556 * this is only called when setting up a new ring.
557 **/
558void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
559{
560	struct device *dev = rx_ring->dev;
561	struct i40e_rx_buffer *rx_bi;
562	dma_addr_t dma;
563	void *buffer;
564	int buf_size;
565	int i;
566
567	if (rx_ring->rx_bi[0].hdr_buf)
568		return;
569	/* Make sure the buffers don't cross cache line boundaries. */
570	buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
571	buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
572				    &dma, GFP_KERNEL);
573	if (!buffer)
574		return;
575	for (i = 0; i < rx_ring->count; i++) {
576		rx_bi = &rx_ring->rx_bi[i];
577		rx_bi->dma = dma + (i * buf_size);
578		rx_bi->hdr_buf = buffer + (i * buf_size);
579	}
580}
581
582/**
583 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
584 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
585 *
586 * Returns 0 on success, negative on failure
587 **/
588int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
589{
590	struct device *dev = rx_ring->dev;
591	int bi_size;
592
593	/* warn if we are about to overwrite the pointer */
594	WARN_ON(rx_ring->rx_bi);
595	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
596	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
597	if (!rx_ring->rx_bi)
598		goto err;
599
600	u64_stats_init(&rx_ring->syncp);
601
602	/* Round up to nearest 4K */
603	rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
604		? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
605		: rx_ring->count * sizeof(union i40e_32byte_rx_desc);
606	rx_ring->size = ALIGN(rx_ring->size, 4096);
607	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
608					   &rx_ring->dma, GFP_KERNEL);
609
610	if (!rx_ring->desc) {
611		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
612			 rx_ring->size);
613		goto err;
614	}
615
616	rx_ring->next_to_clean = 0;
617	rx_ring->next_to_use = 0;
618
619	return 0;
620err:
621	kfree(rx_ring->rx_bi);
622	rx_ring->rx_bi = NULL;
623	return -ENOMEM;
624}
625
626/**
627 * i40e_release_rx_desc - Store the new tail and head values
628 * @rx_ring: ring to bump
629 * @val: new head index
630 **/
631static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
632{
633	rx_ring->next_to_use = val;
634	/* Force memory writes to complete before letting h/w
635	 * know there are new descriptors to fetch.  (Only
636	 * applicable for weak-ordered memory model archs,
637	 * such as IA-64).
638	 */
639	wmb();
640	writel(val, rx_ring->tail);
641}
642
643/**
644 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
645 * @rx_ring: ring to place buffers on
646 * @cleaned_count: number of buffers to replace
647 **/
648void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
649{
650	u16 i = rx_ring->next_to_use;
651	union i40e_rx_desc *rx_desc;
652	struct i40e_rx_buffer *bi;
653
654	/* do nothing if no valid netdev defined */
655	if (!rx_ring->netdev || !cleaned_count)
656		return;
657
658	while (cleaned_count--) {
659		rx_desc = I40E_RX_DESC(rx_ring, i);
660		bi = &rx_ring->rx_bi[i];
661
662		if (bi->skb) /* desc is in use */
663			goto no_buffers;
664		if (!bi->page) {
665			bi->page = alloc_page(GFP_ATOMIC);
666			if (!bi->page) {
667				rx_ring->rx_stats.alloc_page_failed++;
668				goto no_buffers;
669			}
670		}
671
672		if (!bi->page_dma) {
673			/* use a half page if we're re-using */
674			bi->page_offset ^= PAGE_SIZE / 2;
675			bi->page_dma = dma_map_page(rx_ring->dev,
676						    bi->page,
677						    bi->page_offset,
678						    PAGE_SIZE / 2,
679						    DMA_FROM_DEVICE);
680			if (dma_mapping_error(rx_ring->dev,
681					      bi->page_dma)) {
682				rx_ring->rx_stats.alloc_page_failed++;
683				bi->page_dma = 0;
684				goto no_buffers;
685			}
686		}
687
688		dma_sync_single_range_for_device(rx_ring->dev,
689						 bi->dma,
690						 0,
691						 rx_ring->rx_hdr_len,
692						 DMA_FROM_DEVICE);
693		/* Refresh the desc even if buffer_addrs didn't change
694		 * because each write-back erases this info.
695		 */
696		rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
697		rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
698		i++;
699		if (i == rx_ring->count)
700			i = 0;
701	}
702
703no_buffers:
704	if (rx_ring->next_to_use != i)
705		i40e_release_rx_desc(rx_ring, i);
706}
707
708/**
709 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
710 * @rx_ring: ring to place buffers on
711 * @cleaned_count: number of buffers to replace
712 **/
713void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
714{
715	u16 i = rx_ring->next_to_use;
716	union i40e_rx_desc *rx_desc;
717	struct i40e_rx_buffer *bi;
718	struct sk_buff *skb;
719
720	/* do nothing if no valid netdev defined */
721	if (!rx_ring->netdev || !cleaned_count)
722		return;
723
724	while (cleaned_count--) {
725		rx_desc = I40E_RX_DESC(rx_ring, i);
726		bi = &rx_ring->rx_bi[i];
727		skb = bi->skb;
728
729		if (!skb) {
730			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
731							rx_ring->rx_buf_len);
732			if (!skb) {
733				rx_ring->rx_stats.alloc_buff_failed++;
734				goto no_buffers;
735			}
736			/* initialize queue mapping */
737			skb_record_rx_queue(skb, rx_ring->queue_index);
738			bi->skb = skb;
739		}
740
741		if (!bi->dma) {
742			bi->dma = dma_map_single(rx_ring->dev,
743						 skb->data,
744						 rx_ring->rx_buf_len,
745						 DMA_FROM_DEVICE);
746			if (dma_mapping_error(rx_ring->dev, bi->dma)) {
747				rx_ring->rx_stats.alloc_buff_failed++;
748				bi->dma = 0;
749				goto no_buffers;
750			}
751		}
752
753		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
754		rx_desc->read.hdr_addr = 0;
755		i++;
756		if (i == rx_ring->count)
757			i = 0;
758	}
759
760no_buffers:
761	if (rx_ring->next_to_use != i)
762		i40e_release_rx_desc(rx_ring, i);
763}
764
765/**
766 * i40e_receive_skb - Send a completed packet up the stack
767 * @rx_ring:  rx ring in play
768 * @skb: packet to send up
769 * @vlan_tag: vlan tag for packet
770 **/
771static void i40e_receive_skb(struct i40e_ring *rx_ring,
772			     struct sk_buff *skb, u16 vlan_tag)
773{
774	struct i40e_q_vector *q_vector = rx_ring->q_vector;
775
776	if (vlan_tag & VLAN_VID_MASK)
777		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
778
779	napi_gro_receive(&q_vector->napi, skb);
780}
781
782/**
783 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
784 * @vsi: the VSI we care about
785 * @skb: skb currently being received and modified
786 * @rx_status: status value of last descriptor in packet
787 * @rx_error: error value of last descriptor in packet
788 * @rx_ptype: ptype value of last descriptor in packet
789 **/
790static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
791				    struct sk_buff *skb,
792				    u32 rx_status,
793				    u32 rx_error,
794				    u16 rx_ptype)
795{
796	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
797	bool ipv4 = false, ipv6 = false;
798	bool ipv4_tunnel, ipv6_tunnel;
799	__wsum rx_udp_csum;
800	struct iphdr *iph;
801	__sum16 csum;
802
803	ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
804		     (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
805	ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
806		     (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
807
808	skb->ip_summed = CHECKSUM_NONE;
809
810	/* Rx csum enabled and ip headers found? */
811	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
812		return;
813
814	/* did the hardware decode the packet and checksum? */
815	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
816		return;
817
818	/* both known and outer_ip must be set for the below code to work */
819	if (!(decoded.known && decoded.outer_ip))
820		return;
821
822	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
823	    decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
824		ipv4 = true;
825	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
826		 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
827		ipv6 = true;
828
829	if (ipv4 &&
830	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
831			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
832		goto checksum_fail;
833
834	/* likely incorrect csum if alternate IP extension headers found */
835	if (ipv6 &&
836	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
837		/* don't increment checksum err here, non-fatal err */
838		return;
839
840	/* there was some L4 error, count error and punt packet to the stack */
841	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
842		goto checksum_fail;
843
844	/* handle packets that were not able to be checksummed due
845	 * to arrival speed, in this case the stack can compute
846	 * the csum.
847	 */
848	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
849		return;
850
851	/* If VXLAN traffic has an outer UDPv4 checksum we need to check
852	 * it in the driver, hardware does not do it for us.
853	 * Since L3L4P bit was set we assume a valid IHL value (>=5)
854	 * so the total length of IPv4 header is IHL*4 bytes
855	 * The UDP_0 bit *may* bet set if the *inner* header is UDP
856	 */
857	if (ipv4_tunnel) {
858		skb->transport_header = skb->mac_header +
859					sizeof(struct ethhdr) +
860					(ip_hdr(skb)->ihl * 4);
861
862		/* Add 4 bytes for VLAN tagged packets */
863		skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
864					  skb->protocol == htons(ETH_P_8021AD))
865					  ? VLAN_HLEN : 0;
866
867		if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
868		    (udp_hdr(skb)->check != 0)) {
869			rx_udp_csum = udp_csum(skb);
870			iph = ip_hdr(skb);
871			csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
872						 (skb->len -
873						  skb_transport_offset(skb)),
874						 IPPROTO_UDP, rx_udp_csum);
875
876			if (udp_hdr(skb)->check != csum)
877				goto checksum_fail;
878
879		} /* else its GRE and so no outer UDP header */
880	}
881
882	skb->ip_summed = CHECKSUM_UNNECESSARY;
883	skb->csum_level = ipv4_tunnel || ipv6_tunnel;
884
885	return;
886
887checksum_fail:
888	vsi->back->hw_csum_rx_error++;
889}
890
891/**
892 * i40e_rx_hash - returns the hash value from the Rx descriptor
893 * @ring: descriptor ring
894 * @rx_desc: specific descriptor
895 **/
896static inline u32 i40e_rx_hash(struct i40e_ring *ring,
897			       union i40e_rx_desc *rx_desc)
898{
899	const __le64 rss_mask =
900		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
901			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
902
903	if ((ring->netdev->features & NETIF_F_RXHASH) &&
904	    (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
905		return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
906	else
907		return 0;
908}
909
910/**
911 * i40e_ptype_to_hash - get a hash type
912 * @ptype: the ptype value from the descriptor
913 *
914 * Returns a hash type to be used by skb_set_hash
915 **/
916static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
917{
918	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
919
920	if (!decoded.known)
921		return PKT_HASH_TYPE_NONE;
922
923	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
924	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
925		return PKT_HASH_TYPE_L4;
926	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
927		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
928		return PKT_HASH_TYPE_L3;
929	else
930		return PKT_HASH_TYPE_L2;
931}
932
933/**
934 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
935 * @rx_ring:  rx ring to clean
936 * @budget:   how many cleans we're allowed
937 *
938 * Returns true if there's any budget left (e.g. the clean is finished)
939 **/
940static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
941{
942	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
943	u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
944	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
945	const int current_node = numa_mem_id();
946	struct i40e_vsi *vsi = rx_ring->vsi;
947	u16 i = rx_ring->next_to_clean;
948	union i40e_rx_desc *rx_desc;
949	u32 rx_error, rx_status;
950	u8 rx_ptype;
951	u64 qword;
952
953	do {
954		struct i40e_rx_buffer *rx_bi;
955		struct sk_buff *skb;
956		u16 vlan_tag;
957		/* return some buffers to hardware, one at a time is too slow */
958		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
959			i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
960			cleaned_count = 0;
961		}
962
963		i = rx_ring->next_to_clean;
964		rx_desc = I40E_RX_DESC(rx_ring, i);
965		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
966		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
967			I40E_RXD_QW1_STATUS_SHIFT;
968
969		if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
970			break;
971
972		/* This memory barrier is needed to keep us from reading
973		 * any other fields out of the rx_desc until we know the
974		 * DD bit is set.
975		 */
976		dma_rmb();
977		rx_bi = &rx_ring->rx_bi[i];
978		skb = rx_bi->skb;
979		if (likely(!skb)) {
980			skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
981							rx_ring->rx_hdr_len);
982			if (!skb) {
983				rx_ring->rx_stats.alloc_buff_failed++;
984				break;
985			}
986
987			/* initialize queue mapping */
988			skb_record_rx_queue(skb, rx_ring->queue_index);
989			/* we are reusing so sync this buffer for CPU use */
990			dma_sync_single_range_for_cpu(rx_ring->dev,
991						      rx_bi->dma,
992						      0,
993						      rx_ring->rx_hdr_len,
994						      DMA_FROM_DEVICE);
995		}
996		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
997				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
998		rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
999				I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1000		rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1001			 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1002
1003		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1004			   I40E_RXD_QW1_ERROR_SHIFT;
1005		rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1006		rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1007
1008		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1009			   I40E_RXD_QW1_PTYPE_SHIFT;
1010		prefetch(rx_bi->page);
1011		rx_bi->skb = NULL;
1012		cleaned_count++;
1013		if (rx_hbo || rx_sph) {
1014			int len;
1015
1016			if (rx_hbo)
1017				len = I40E_RX_HDR_SIZE;
1018			else
1019				len = rx_header_len;
1020			memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1021		} else if (skb->len == 0) {
1022			int len;
1023
1024			len = (rx_packet_len > skb_headlen(skb) ?
1025				skb_headlen(skb) : rx_packet_len);
1026			memcpy(__skb_put(skb, len),
1027			       rx_bi->page + rx_bi->page_offset,
1028			       len);
1029			rx_bi->page_offset += len;
1030			rx_packet_len -= len;
1031		}
1032
1033		/* Get the rest of the data if this was a header split */
1034		if (rx_packet_len) {
1035			skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1036					   rx_bi->page,
1037					   rx_bi->page_offset,
1038					   rx_packet_len);
1039
1040			skb->len += rx_packet_len;
1041			skb->data_len += rx_packet_len;
1042			skb->truesize += rx_packet_len;
1043
1044			if ((page_count(rx_bi->page) == 1) &&
1045			    (page_to_nid(rx_bi->page) == current_node))
1046				get_page(rx_bi->page);
1047			else
1048				rx_bi->page = NULL;
1049
1050			dma_unmap_page(rx_ring->dev,
1051				       rx_bi->page_dma,
1052				       PAGE_SIZE / 2,
1053				       DMA_FROM_DEVICE);
1054			rx_bi->page_dma = 0;
1055		}
1056		I40E_RX_INCREMENT(rx_ring, i);
1057
1058		if (unlikely(
1059		    !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1060			struct i40e_rx_buffer *next_buffer;
1061
1062			next_buffer = &rx_ring->rx_bi[i];
1063			next_buffer->skb = skb;
1064			rx_ring->rx_stats.non_eop_descs++;
1065			continue;
1066		}
1067
1068		/* ERR_MASK will only have valid bits if EOP set */
1069		if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1070			dev_kfree_skb_any(skb);
1071			continue;
1072		}
1073
1074		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1075			     i40e_ptype_to_hash(rx_ptype));
1076		/* probably a little skewed due to removing CRC */
1077		total_rx_bytes += skb->len;
1078		total_rx_packets++;
1079
1080		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1081
1082		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1083
1084		vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1085			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1086			 : 0;
1087#ifdef I40E_FCOE
1088		if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1089			dev_kfree_skb_any(skb);
1090			continue;
1091		}
1092#endif
1093		skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
1094		i40e_receive_skb(rx_ring, skb, vlan_tag);
1095
1096		rx_desc->wb.qword1.status_error_len = 0;
1097
1098	} while (likely(total_rx_packets < budget));
1099
1100	u64_stats_update_begin(&rx_ring->syncp);
1101	rx_ring->stats.packets += total_rx_packets;
1102	rx_ring->stats.bytes += total_rx_bytes;
1103	u64_stats_update_end(&rx_ring->syncp);
1104	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1105	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1106
1107	return total_rx_packets;
1108}
1109
1110/**
1111 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1112 * @rx_ring:  rx ring to clean
1113 * @budget:   how many cleans we're allowed
1114 *
1115 * Returns number of packets cleaned
1116 **/
1117static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1118{
1119	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1120	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1121	struct i40e_vsi *vsi = rx_ring->vsi;
1122	union i40e_rx_desc *rx_desc;
1123	u32 rx_error, rx_status;
1124	u16 rx_packet_len;
1125	u8 rx_ptype;
1126	u64 qword;
1127	u16 i;
1128
1129	do {
1130		struct i40e_rx_buffer *rx_bi;
1131		struct sk_buff *skb;
1132		u16 vlan_tag;
1133		/* return some buffers to hardware, one at a time is too slow */
1134		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1135			i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1136			cleaned_count = 0;
1137		}
1138
1139		i = rx_ring->next_to_clean;
1140		rx_desc = I40E_RX_DESC(rx_ring, i);
1141		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1142		rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1143			I40E_RXD_QW1_STATUS_SHIFT;
1144
1145		if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1146			break;
1147
1148		/* This memory barrier is needed to keep us from reading
1149		 * any other fields out of the rx_desc until we know the
1150		 * DD bit is set.
1151		 */
1152		dma_rmb();
1153
1154		rx_bi = &rx_ring->rx_bi[i];
1155		skb = rx_bi->skb;
1156		prefetch(skb->data);
1157
1158		rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1159				I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1160
1161		rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1162			   I40E_RXD_QW1_ERROR_SHIFT;
1163		rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1164
1165		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1166			   I40E_RXD_QW1_PTYPE_SHIFT;
1167		rx_bi->skb = NULL;
1168		cleaned_count++;
1169
1170		/* Get the header and possibly the whole packet
1171		 * If this is an skb from previous receive dma will be 0
1172		 */
1173		skb_put(skb, rx_packet_len);
1174		dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1175				 DMA_FROM_DEVICE);
1176		rx_bi->dma = 0;
1177
1178		I40E_RX_INCREMENT(rx_ring, i);
1179
1180		if (unlikely(
1181		    !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1182			rx_ring->rx_stats.non_eop_descs++;
1183			continue;
1184		}
1185
1186		/* ERR_MASK will only have valid bits if EOP set */
1187		if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1188			dev_kfree_skb_any(skb);
1189			continue;
1190		}
1191
1192		skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1193			     i40e_ptype_to_hash(rx_ptype));
1194		/* probably a little skewed due to removing CRC */
1195		total_rx_bytes += skb->len;
1196		total_rx_packets++;
1197
1198		skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1199
1200		i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1201
1202		vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1203			 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1204			 : 0;
1205		i40e_receive_skb(rx_ring, skb, vlan_tag);
1206
1207		rx_desc->wb.qword1.status_error_len = 0;
1208	} while (likely(total_rx_packets < budget));
1209
1210	u64_stats_update_begin(&rx_ring->syncp);
1211	rx_ring->stats.packets += total_rx_packets;
1212	rx_ring->stats.bytes += total_rx_bytes;
1213	u64_stats_update_end(&rx_ring->syncp);
1214	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1215	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1216
1217	return total_rx_packets;
1218}
1219
1220static u32 i40e_buildreg_itr(const int type, const u16 itr)
1221{
1222	u32 val;
1223
1224	val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1225	      I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1226	      (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1227	      (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1228
1229	return val;
1230}
1231
1232/* a small macro to shorten up some long lines */
1233#define INTREG I40E_VFINT_DYN_CTLN1
1234
1235/**
1236 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1237 * @vsi: the VSI we care about
1238 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1239 *
1240 **/
1241static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1242					  struct i40e_q_vector *q_vector)
1243{
1244	struct i40e_hw *hw = &vsi->back->hw;
1245	bool rx = false, tx = false;
1246	u32 rxval, txval;
1247	int vector;
1248
1249	vector = (q_vector->v_idx + vsi->base_vector);
1250
1251	/* avoid dynamic calculation if in countdown mode OR if
1252	 * all dynamic is disabled
1253	 */
1254	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1255
1256	if (q_vector->itr_countdown > 0 ||
1257	    (!ITR_IS_DYNAMIC(vsi->rx_itr_setting) &&
1258	     !ITR_IS_DYNAMIC(vsi->tx_itr_setting))) {
1259		goto enable_int;
1260	}
1261
1262	if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1263		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1264		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1265	}
1266	if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1267		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1268		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1269	}
1270	if (rx || tx) {
1271		/* get the higher of the two ITR adjustments and
1272		 * use the same value for both ITR registers
1273		 * when in adaptive mode (Rx and/or Tx)
1274		 */
1275		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1276
1277		q_vector->tx.itr = q_vector->rx.itr = itr;
1278		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1279		tx = true;
1280		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1281		rx = true;
1282	}
1283
1284	/* only need to enable the interrupt once, but need
1285	 * to possibly update both ITR values
1286	 */
1287	if (rx) {
1288		/* set the INTENA_MSK_MASK so that this first write
1289		 * won't actually enable the interrupt, instead just
1290		 * updating the ITR (it's bit 31 PF and VF)
1291		 */
1292		rxval |= BIT(31);
1293		/* don't check _DOWN because interrupt isn't being enabled */
1294		wr32(hw, INTREG(vector - 1), rxval);
1295	}
1296
1297enable_int:
1298	if (!test_bit(__I40E_DOWN, &vsi->state))
1299		wr32(hw, INTREG(vector - 1), txval);
1300
1301	if (q_vector->itr_countdown)
1302		q_vector->itr_countdown--;
1303	else
1304		q_vector->itr_countdown = ITR_COUNTDOWN_START;
1305
1306}
1307
1308/**
1309 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1310 * @napi: napi struct with our devices info in it
1311 * @budget: amount of work driver is allowed to do this pass, in packets
1312 *
1313 * This function will clean all queues associated with a q_vector.
1314 *
1315 * Returns the amount of work done
1316 **/
1317int i40evf_napi_poll(struct napi_struct *napi, int budget)
1318{
1319	struct i40e_q_vector *q_vector =
1320			       container_of(napi, struct i40e_q_vector, napi);
1321	struct i40e_vsi *vsi = q_vector->vsi;
1322	struct i40e_ring *ring;
1323	bool clean_complete = true;
1324	bool arm_wb = false;
1325	int budget_per_ring;
1326	int work_done = 0;
1327
1328	if (test_bit(__I40E_DOWN, &vsi->state)) {
1329		napi_complete(napi);
1330		return 0;
1331	}
1332
1333	/* Since the actual Tx work is minimal, we can give the Tx a larger
1334	 * budget and be more aggressive about cleaning up the Tx descriptors.
1335	 */
1336	i40e_for_each_ring(ring, q_vector->tx) {
1337		clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1338		arm_wb |= ring->arm_wb;
1339		ring->arm_wb = false;
1340	}
1341
1342	/* Handle case where we are called by netpoll with a budget of 0 */
1343	if (budget <= 0)
1344		goto tx_only;
1345
1346	/* We attempt to distribute budget to each Rx queue fairly, but don't
1347	 * allow the budget to go below 1 because that would exit polling early.
1348	 */
1349	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1350
1351	i40e_for_each_ring(ring, q_vector->rx) {
1352		int cleaned;
1353
1354		if (ring_is_ps_enabled(ring))
1355			cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1356		else
1357			cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1358
1359		work_done += cleaned;
1360		/* if we didn't clean as many as budgeted, we must be done */
1361		clean_complete &= (budget_per_ring != cleaned);
1362	}
1363
1364	/* If work not completed, return budget and polling will return */
1365	if (!clean_complete) {
1366tx_only:
1367		if (arm_wb)
1368			i40evf_force_wb(vsi, q_vector);
1369		return budget;
1370	}
1371
1372	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1373		q_vector->arm_wb_state = false;
1374
1375	/* Work is done so exit the polling mode and re-enable the interrupt */
1376	napi_complete_done(napi, work_done);
1377	i40e_update_enable_itr(vsi, q_vector);
1378	return 0;
1379}
1380
1381/**
1382 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1383 * @skb:     send buffer
1384 * @tx_ring: ring to send buffer on
1385 * @flags:   the tx flags to be set
1386 *
1387 * Checks the skb and set up correspondingly several generic transmit flags
1388 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1389 *
1390 * Returns error code indicate the frame should be dropped upon error and the
1391 * otherwise  returns 0 to indicate the flags has been set properly.
1392 **/
1393static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1394					       struct i40e_ring *tx_ring,
1395					       u32 *flags)
1396{
1397	__be16 protocol = skb->protocol;
1398	u32  tx_flags = 0;
1399
1400	if (protocol == htons(ETH_P_8021Q) &&
1401	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1402		/* When HW VLAN acceleration is turned off by the user the
1403		 * stack sets the protocol to 8021q so that the driver
1404		 * can take any steps required to support the SW only
1405		 * VLAN handling.  In our case the driver doesn't need
1406		 * to take any further steps so just set the protocol
1407		 * to the encapsulated ethertype.
1408		 */
1409		skb->protocol = vlan_get_protocol(skb);
1410		goto out;
1411	}
1412
1413	/* if we have a HW VLAN tag being added, default to the HW one */
1414	if (skb_vlan_tag_present(skb)) {
1415		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1416		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1417	/* else if it is a SW VLAN, check the next protocol and store the tag */
1418	} else if (protocol == htons(ETH_P_8021Q)) {
1419		struct vlan_hdr *vhdr, _vhdr;
1420
1421		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1422		if (!vhdr)
1423			return -EINVAL;
1424
1425		protocol = vhdr->h_vlan_encapsulated_proto;
1426		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1427		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1428	}
1429
1430out:
1431	*flags = tx_flags;
1432	return 0;
1433}
1434
1435/**
1436 * i40e_tso - set up the tso context descriptor
1437 * @tx_ring:  ptr to the ring to send
1438 * @skb:      ptr to the skb we're sending
1439 * @hdr_len:  ptr to the size of the packet header
1440 * @cd_tunneling: ptr to context descriptor bits
1441 *
1442 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1443 **/
1444static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1445		    u8 *hdr_len, u64 *cd_type_cmd_tso_mss,
1446		    u32 *cd_tunneling)
1447{
1448	u32 cd_cmd, cd_tso_len, cd_mss;
1449	struct ipv6hdr *ipv6h;
1450	struct tcphdr *tcph;
1451	struct iphdr *iph;
1452	u32 l4len;
1453	int err;
1454
1455	if (!skb_is_gso(skb))
1456		return 0;
1457
1458	err = skb_cow_head(skb, 0);
1459	if (err < 0)
1460		return err;
1461
1462	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1463	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1464
1465	if (iph->version == 4) {
1466		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1467		iph->tot_len = 0;
1468		iph->check = 0;
1469		tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1470						 0, IPPROTO_TCP, 0);
1471	} else if (ipv6h->version == 6) {
1472		tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1473		ipv6h->payload_len = 0;
1474		tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1475					       0, IPPROTO_TCP, 0);
1476	}
1477
1478	l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1479	*hdr_len = (skb->encapsulation
1480		    ? (skb_inner_transport_header(skb) - skb->data)
1481		    : skb_transport_offset(skb)) + l4len;
1482
1483	/* find the field values */
1484	cd_cmd = I40E_TX_CTX_DESC_TSO;
1485	cd_tso_len = skb->len - *hdr_len;
1486	cd_mss = skb_shinfo(skb)->gso_size;
1487	*cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1488				((u64)cd_tso_len <<
1489				 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1490				((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1491	return 1;
1492}
1493
1494/**
1495 * i40e_tx_enable_csum - Enable Tx checksum offloads
1496 * @skb: send buffer
1497 * @tx_flags: pointer to Tx flags currently set
1498 * @td_cmd: Tx descriptor command bits to set
1499 * @td_offset: Tx descriptor header offsets to set
1500 * @cd_tunneling: ptr to context desc bits
1501 **/
1502static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1503				u32 *td_cmd, u32 *td_offset,
1504				struct i40e_ring *tx_ring,
1505				u32 *cd_tunneling)
1506{
1507	struct ipv6hdr *this_ipv6_hdr;
1508	unsigned int this_tcp_hdrlen;
1509	struct iphdr *this_ip_hdr;
1510	u32 network_hdr_len;
1511	u8 l4_hdr = 0;
1512	struct udphdr *oudph;
1513	struct iphdr *oiph;
1514	u32 l4_tunnel = 0;
1515
1516	if (skb->encapsulation) {
1517		switch (ip_hdr(skb)->protocol) {
1518		case IPPROTO_UDP:
1519			oudph = udp_hdr(skb);
1520			oiph = ip_hdr(skb);
1521			l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
1522			*tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1523			break;
1524		default:
1525			return;
1526		}
1527		network_hdr_len = skb_inner_network_header_len(skb);
1528		this_ip_hdr = inner_ip_hdr(skb);
1529		this_ipv6_hdr = inner_ipv6_hdr(skb);
1530		this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1531
1532		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1533			if (*tx_flags & I40E_TX_FLAGS_TSO) {
1534				*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1535				ip_hdr(skb)->check = 0;
1536			} else {
1537				*cd_tunneling |=
1538					 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1539			}
1540		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1541			*cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1542			if (*tx_flags & I40E_TX_FLAGS_TSO)
1543				ip_hdr(skb)->check = 0;
1544		}
1545
1546		/* Now set the ctx descriptor fields */
1547		*cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1548				   I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT      |
1549				   l4_tunnel                             |
1550				   ((skb_inner_network_offset(skb) -
1551					skb_transport_offset(skb)) >> 1) <<
1552				   I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1553		if (this_ip_hdr->version == 6) {
1554			*tx_flags &= ~I40E_TX_FLAGS_IPV4;
1555			*tx_flags |= I40E_TX_FLAGS_IPV6;
1556		}
1557
1558
1559		if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
1560		    (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING)        &&
1561		    (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
1562			oudph->check = ~csum_tcpudp_magic(oiph->saddr,
1563					oiph->daddr,
1564					(skb->len - skb_transport_offset(skb)),
1565					IPPROTO_UDP, 0);
1566			*cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1567		}
1568	} else {
1569		network_hdr_len = skb_network_header_len(skb);
1570		this_ip_hdr = ip_hdr(skb);
1571		this_ipv6_hdr = ipv6_hdr(skb);
1572		this_tcp_hdrlen = tcp_hdrlen(skb);
1573	}
1574
1575	/* Enable IP checksum offloads */
1576	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1577		l4_hdr = this_ip_hdr->protocol;
1578		/* the stack computes the IP header already, the only time we
1579		 * need the hardware to recompute it is in the case of TSO.
1580		 */
1581		if (*tx_flags & I40E_TX_FLAGS_TSO) {
1582			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1583			this_ip_hdr->check = 0;
1584		} else {
1585			*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1586		}
1587		/* Now set the td_offset for IP header length */
1588		*td_offset = (network_hdr_len >> 2) <<
1589			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1590	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1591		l4_hdr = this_ipv6_hdr->nexthdr;
1592		*td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1593		/* Now set the td_offset for IP header length */
1594		*td_offset = (network_hdr_len >> 2) <<
1595			      I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1596	}
1597	/* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1598	*td_offset |= (skb_network_offset(skb) >> 1) <<
1599		       I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1600
1601	/* Enable L4 checksum offloads */
1602	switch (l4_hdr) {
1603	case IPPROTO_TCP:
1604		/* enable checksum offloads */
1605		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1606		*td_offset |= (this_tcp_hdrlen >> 2) <<
1607			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1608		break;
1609	case IPPROTO_SCTP:
1610		/* enable SCTP checksum offload */
1611		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1612		*td_offset |= (sizeof(struct sctphdr) >> 2) <<
1613			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1614		break;
1615	case IPPROTO_UDP:
1616		/* enable UDP checksum offload */
1617		*td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1618		*td_offset |= (sizeof(struct udphdr) >> 2) <<
1619			       I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1620		break;
1621	default:
1622		break;
1623	}
1624}
1625
1626/**
1627 * i40e_create_tx_ctx Build the Tx context descriptor
1628 * @tx_ring:  ring to create the descriptor on
1629 * @cd_type_cmd_tso_mss: Quad Word 1
1630 * @cd_tunneling: Quad Word 0 - bits 0-31
1631 * @cd_l2tag2: Quad Word 0 - bits 32-63
1632 **/
1633static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1634			       const u64 cd_type_cmd_tso_mss,
1635			       const u32 cd_tunneling, const u32 cd_l2tag2)
1636{
1637	struct i40e_tx_context_desc *context_desc;
1638	int i = tx_ring->next_to_use;
1639
1640	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1641	    !cd_tunneling && !cd_l2tag2)
1642		return;
1643
1644	/* grab the next descriptor */
1645	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1646
1647	i++;
1648	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1649
1650	/* cpu_to_le32 and assign to struct fields */
1651	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1652	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1653	context_desc->rsvd = cpu_to_le16(0);
1654	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1655}
1656
1657 /**
1658 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1659 * @skb:      send buffer
1660 * @tx_flags: collected send information
1661 *
1662 * Note: Our HW can't scatter-gather more than 8 fragments to build
1663 * a packet on the wire and so we need to figure out the cases where we
1664 * need to linearize the skb.
1665 **/
1666static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1667{
1668	struct skb_frag_struct *frag;
1669	bool linearize = false;
1670	unsigned int size = 0;
1671	u16 num_frags;
1672	u16 gso_segs;
1673
1674	num_frags = skb_shinfo(skb)->nr_frags;
1675	gso_segs = skb_shinfo(skb)->gso_segs;
1676
1677	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1678		u16 j = 0;
1679
1680		if (num_frags < (I40E_MAX_BUFFER_TXD))
1681			goto linearize_chk_done;
1682		/* try the simple math, if we have too many frags per segment */
1683		if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1684		    I40E_MAX_BUFFER_TXD) {
1685			linearize = true;
1686			goto linearize_chk_done;
1687		}
1688		frag = &skb_shinfo(skb)->frags[0];
1689		/* we might still have more fragments per segment */
1690		do {
1691			size += skb_frag_size(frag);
1692			frag++; j++;
1693			if ((size >= skb_shinfo(skb)->gso_size) &&
1694			    (j < I40E_MAX_BUFFER_TXD)) {
1695				size = (size % skb_shinfo(skb)->gso_size);
1696				j = (size) ? 1 : 0;
1697			}
1698			if (j == I40E_MAX_BUFFER_TXD) {
1699				linearize = true;
1700				break;
1701			}
1702			num_frags--;
1703		} while (num_frags);
1704	} else {
1705		if (num_frags >= I40E_MAX_BUFFER_TXD)
1706			linearize = true;
1707	}
1708
1709linearize_chk_done:
1710	return linearize;
1711}
1712
1713/**
1714 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1715 * @tx_ring: the ring to be checked
1716 * @size:    the size buffer we want to assure is available
1717 *
1718 * Returns -EBUSY if a stop is needed, else 0
1719 **/
1720static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1721{
1722	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1723	/* Memory barrier before checking head and tail */
1724	smp_mb();
1725
1726	/* Check again in a case another CPU has just made room available. */
1727	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1728		return -EBUSY;
1729
1730	/* A reprieve! - use start_queue because it doesn't call schedule */
1731	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1732	++tx_ring->tx_stats.restart_queue;
1733	return 0;
1734}
1735
1736/**
1737 * i40evf_maybe_stop_tx - 1st level check for tx stop conditions
1738 * @tx_ring: the ring to be checked
1739 * @size:    the size buffer we want to assure is available
1740 *
1741 * Returns 0 if stop is not needed
1742 **/
1743static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1744{
1745	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1746		return 0;
1747	return __i40evf_maybe_stop_tx(tx_ring, size);
1748}
1749
1750/**
1751 * i40evf_tx_map - Build the Tx descriptor
1752 * @tx_ring:  ring to send buffer on
1753 * @skb:      send buffer
1754 * @first:    first buffer info buffer to use
1755 * @tx_flags: collected send information
1756 * @hdr_len:  size of the packet header
1757 * @td_cmd:   the command field in the descriptor
1758 * @td_offset: offset for checksum or crc
1759 **/
1760static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1761				 struct i40e_tx_buffer *first, u32 tx_flags,
1762				 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1763{
1764	unsigned int data_len = skb->data_len;
1765	unsigned int size = skb_headlen(skb);
1766	struct skb_frag_struct *frag;
1767	struct i40e_tx_buffer *tx_bi;
1768	struct i40e_tx_desc *tx_desc;
1769	u16 i = tx_ring->next_to_use;
1770	u32 td_tag = 0;
1771	dma_addr_t dma;
1772	u16 gso_segs;
1773
1774	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1775		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1776		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1777			 I40E_TX_FLAGS_VLAN_SHIFT;
1778	}
1779
1780	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1781		gso_segs = skb_shinfo(skb)->gso_segs;
1782	else
1783		gso_segs = 1;
1784
1785	/* multiply data chunks by size of headers */
1786	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1787	first->gso_segs = gso_segs;
1788	first->skb = skb;
1789	first->tx_flags = tx_flags;
1790
1791	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1792
1793	tx_desc = I40E_TX_DESC(tx_ring, i);
1794	tx_bi = first;
1795
1796	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1797		if (dma_mapping_error(tx_ring->dev, dma))
1798			goto dma_error;
1799
1800		/* record length, and DMA address */
1801		dma_unmap_len_set(tx_bi, len, size);
1802		dma_unmap_addr_set(tx_bi, dma, dma);
1803
1804		tx_desc->buffer_addr = cpu_to_le64(dma);
1805
1806		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1807			tx_desc->cmd_type_offset_bsz =
1808				build_ctob(td_cmd, td_offset,
1809					   I40E_MAX_DATA_PER_TXD, td_tag);
1810
1811			tx_desc++;
1812			i++;
1813			if (i == tx_ring->count) {
1814				tx_desc = I40E_TX_DESC(tx_ring, 0);
1815				i = 0;
1816			}
1817
1818			dma += I40E_MAX_DATA_PER_TXD;
1819			size -= I40E_MAX_DATA_PER_TXD;
1820
1821			tx_desc->buffer_addr = cpu_to_le64(dma);
1822		}
1823
1824		if (likely(!data_len))
1825			break;
1826
1827		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1828							  size, td_tag);
1829
1830		tx_desc++;
1831		i++;
1832		if (i == tx_ring->count) {
1833			tx_desc = I40E_TX_DESC(tx_ring, 0);
1834			i = 0;
1835		}
1836
1837		size = skb_frag_size(frag);
1838		data_len -= size;
1839
1840		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1841				       DMA_TO_DEVICE);
1842
1843		tx_bi = &tx_ring->tx_bi[i];
1844	}
1845
1846	/* Place RS bit on last descriptor of any packet that spans across the
1847	 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1848	 */
1849#define WB_STRIDE 0x3
1850	if (((i & WB_STRIDE) != WB_STRIDE) &&
1851	    (first <= &tx_ring->tx_bi[i]) &&
1852	    (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1853		tx_desc->cmd_type_offset_bsz =
1854			build_ctob(td_cmd, td_offset, size, td_tag) |
1855			cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1856					 I40E_TXD_QW1_CMD_SHIFT);
1857	} else {
1858		tx_desc->cmd_type_offset_bsz =
1859			build_ctob(td_cmd, td_offset, size, td_tag) |
1860			cpu_to_le64((u64)I40E_TXD_CMD <<
1861					 I40E_TXD_QW1_CMD_SHIFT);
1862	}
1863
1864	netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1865						 tx_ring->queue_index),
1866			     first->bytecount);
1867
1868	/* Force memory writes to complete before letting h/w
1869	 * know there are new descriptors to fetch.  (Only
1870	 * applicable for weak-ordered memory model archs,
1871	 * such as IA-64).
1872	 */
1873	wmb();
1874
1875	/* set next_to_watch value indicating a packet is present */
1876	first->next_to_watch = tx_desc;
1877
1878	i++;
1879	if (i == tx_ring->count)
1880		i = 0;
1881
1882	tx_ring->next_to_use = i;
1883
1884	i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED);
1885	/* notify HW of packet */
1886	if (!skb->xmit_more ||
1887	    netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1888						   tx_ring->queue_index)))
1889		writel(i, tx_ring->tail);
1890	else
1891		prefetchw(tx_desc + 1);
1892
1893	return;
1894
1895dma_error:
1896	dev_info(tx_ring->dev, "TX DMA map failed\n");
1897
1898	/* clear dma mappings for failed tx_bi map */
1899	for (;;) {
1900		tx_bi = &tx_ring->tx_bi[i];
1901		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1902		if (tx_bi == first)
1903			break;
1904		if (i == 0)
1905			i = tx_ring->count;
1906		i--;
1907	}
1908
1909	tx_ring->next_to_use = i;
1910}
1911
1912/**
1913 * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed
1914 * @skb:     send buffer
1915 * @tx_ring: ring to send buffer on
1916 *
1917 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1918 * there is not enough descriptors available in this ring since we need at least
1919 * one descriptor.
1920 **/
1921static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb,
1922					       struct i40e_ring *tx_ring)
1923{
1924	unsigned int f;
1925	int count = 0;
1926
1927	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1928	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1929	 *       + 4 desc gap to avoid the cache line where head is,
1930	 *       + 1 desc for context descriptor,
1931	 * otherwise try next time
1932	 */
1933	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1934		count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1935
1936	count += TXD_USE_COUNT(skb_headlen(skb));
1937	if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1938		tx_ring->tx_stats.tx_busy++;
1939		return 0;
1940	}
1941	return count;
1942}
1943
1944/**
1945 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1946 * @skb:     send buffer
1947 * @tx_ring: ring to send buffer on
1948 *
1949 * Returns NETDEV_TX_OK if sent, else an error code
1950 **/
1951static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1952					struct i40e_ring *tx_ring)
1953{
1954	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1955	u32 cd_tunneling = 0, cd_l2tag2 = 0;
1956	struct i40e_tx_buffer *first;
1957	u32 td_offset = 0;
1958	u32 tx_flags = 0;
1959	__be16 protocol;
1960	u32 td_cmd = 0;
1961	u8 hdr_len = 0;
1962	int tso;
1963
1964	if (0 == i40evf_xmit_descriptor_count(skb, tx_ring))
1965		return NETDEV_TX_BUSY;
1966
1967	/* prepare the xmit flags */
1968	if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1969		goto out_drop;
1970
1971	/* obtain protocol of skb */
1972	protocol = vlan_get_protocol(skb);
1973
1974	/* record the location of the first descriptor for this packet */
1975	first = &tx_ring->tx_bi[tx_ring->next_to_use];
1976
1977	/* setup IPv4/IPv6 offloads */
1978	if (protocol == htons(ETH_P_IP))
1979		tx_flags |= I40E_TX_FLAGS_IPV4;
1980	else if (protocol == htons(ETH_P_IPV6))
1981		tx_flags |= I40E_TX_FLAGS_IPV6;
1982
1983	tso = i40e_tso(tx_ring, skb, &hdr_len,
1984		       &cd_type_cmd_tso_mss, &cd_tunneling);
1985
1986	if (tso < 0)
1987		goto out_drop;
1988	else if (tso)
1989		tx_flags |= I40E_TX_FLAGS_TSO;
1990
1991	if (i40e_chk_linearize(skb, tx_flags)) {
1992		if (skb_linearize(skb))
1993			goto out_drop;
1994		tx_ring->tx_stats.tx_linearize++;
1995	}
1996	skb_tx_timestamp(skb);
1997
1998	/* always enable CRC insertion offload */
1999	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2000
2001	/* Always offload the checksum, since it's in the data descriptor */
2002	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2003		tx_flags |= I40E_TX_FLAGS_CSUM;
2004
2005		i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2006				    tx_ring, &cd_tunneling);
2007	}
2008
2009	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2010			   cd_tunneling, cd_l2tag2);
2011
2012	i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2013		      td_cmd, td_offset);
2014
2015	return NETDEV_TX_OK;
2016
2017out_drop:
2018	dev_kfree_skb_any(skb);
2019	return NETDEV_TX_OK;
2020}
2021
2022/**
2023 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2024 * @skb:    send buffer
2025 * @netdev: network interface device structure
2026 *
2027 * Returns NETDEV_TX_OK if sent, else an error code
2028 **/
2029netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2030{
2031	struct i40evf_adapter *adapter = netdev_priv(netdev);
2032	struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
2033
2034	/* hardware can't handle really short frames, hardware padding works
2035	 * beyond this point
2036	 */
2037	if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2038		if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2039			return NETDEV_TX_OK;
2040		skb->len = I40E_MIN_TX_LEN;
2041		skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2042	}
2043
2044	return i40e_xmit_frame_ring(skb, tx_ring);
2045}
2046