Linux Audio

Check our new training course

Loading...
Note: File does not exist in v6.13.7.
   1/*******************************************************************************
   2 *
   3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
   4 * Copyright(c) 2013 - 2016 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		if (dma_unmap_len(tx_buffer, len))
  59			dma_unmap_single(ring->dev,
  60					 dma_unmap_addr(tx_buffer, dma),
  61					 dma_unmap_len(tx_buffer, len),
  62					 DMA_TO_DEVICE);
  63	} else if (dma_unmap_len(tx_buffer, len)) {
  64		dma_unmap_page(ring->dev,
  65			       dma_unmap_addr(tx_buffer, dma),
  66			       dma_unmap_len(tx_buffer, len),
  67			       DMA_TO_DEVICE);
  68	}
  69
  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(txring_txq(tx_ring));
 107}
 108
 109/**
 110 * i40evf_free_tx_resources - Free Tx resources per queue
 111 * @tx_ring: Tx descriptor ring for a specific queue
 112 *
 113 * Free all transmit software resources
 114 **/
 115void i40evf_free_tx_resources(struct i40e_ring *tx_ring)
 116{
 117	i40evf_clean_tx_ring(tx_ring);
 118	kfree(tx_ring->tx_bi);
 119	tx_ring->tx_bi = NULL;
 120
 121	if (tx_ring->desc) {
 122		dma_free_coherent(tx_ring->dev, tx_ring->size,
 123				  tx_ring->desc, tx_ring->dma);
 124		tx_ring->desc = NULL;
 125	}
 126}
 127
 128/**
 129 * i40evf_get_tx_pending - how many Tx descriptors not processed
 130 * @tx_ring: the ring of descriptors
 131 * @in_sw: is tx_pending being checked in SW or HW
 132 *
 133 * Since there is no access to the ring head register
 134 * in XL710, we need to use our local copies
 135 **/
 136u32 i40evf_get_tx_pending(struct i40e_ring *ring, bool in_sw)
 137{
 138	u32 head, tail;
 139
 140	if (!in_sw)
 141		head = i40e_get_head(ring);
 142	else
 143		head = ring->next_to_clean;
 144	tail = readl(ring->tail);
 145
 146	if (head != tail)
 147		return (head < tail) ?
 148			tail - head : (tail + ring->count - head);
 149
 150	return 0;
 151}
 152
 153#define WB_STRIDE 4
 154
 155/**
 156 * i40e_clean_tx_irq - Reclaim resources after transmit completes
 157 * @vsi: the VSI we care about
 158 * @tx_ring: Tx ring to clean
 159 * @napi_budget: Used to determine if we are in netpoll
 160 *
 161 * Returns true if there's any budget left (e.g. the clean is finished)
 162 **/
 163static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
 164			      struct i40e_ring *tx_ring, int napi_budget)
 165{
 166	u16 i = tx_ring->next_to_clean;
 167	struct i40e_tx_buffer *tx_buf;
 168	struct i40e_tx_desc *tx_head;
 169	struct i40e_tx_desc *tx_desc;
 170	unsigned int total_bytes = 0, total_packets = 0;
 171	unsigned int budget = vsi->work_limit;
 172
 173	tx_buf = &tx_ring->tx_bi[i];
 174	tx_desc = I40E_TX_DESC(tx_ring, i);
 175	i -= tx_ring->count;
 176
 177	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
 178
 179	do {
 180		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 181
 182		/* if next_to_watch is not set then there is no work pending */
 183		if (!eop_desc)
 184			break;
 185
 186		/* prevent any other reads prior to eop_desc */
 187		read_barrier_depends();
 188
 189		/* we have caught up to head, no work left to do */
 190		if (tx_head == tx_desc)
 191			break;
 192
 193		/* clear next_to_watch to prevent false hangs */
 194		tx_buf->next_to_watch = NULL;
 195
 196		/* update the statistics for this packet */
 197		total_bytes += tx_buf->bytecount;
 198		total_packets += tx_buf->gso_segs;
 199
 200		/* free the skb */
 201		napi_consume_skb(tx_buf->skb, napi_budget);
 202
 203		/* unmap skb header data */
 204		dma_unmap_single(tx_ring->dev,
 205				 dma_unmap_addr(tx_buf, dma),
 206				 dma_unmap_len(tx_buf, len),
 207				 DMA_TO_DEVICE);
 208
 209		/* clear tx_buffer data */
 210		tx_buf->skb = NULL;
 211		dma_unmap_len_set(tx_buf, len, 0);
 212
 213		/* unmap remaining buffers */
 214		while (tx_desc != eop_desc) {
 215
 216			tx_buf++;
 217			tx_desc++;
 218			i++;
 219			if (unlikely(!i)) {
 220				i -= tx_ring->count;
 221				tx_buf = tx_ring->tx_bi;
 222				tx_desc = I40E_TX_DESC(tx_ring, 0);
 223			}
 224
 225			/* unmap any remaining paged data */
 226			if (dma_unmap_len(tx_buf, len)) {
 227				dma_unmap_page(tx_ring->dev,
 228					       dma_unmap_addr(tx_buf, dma),
 229					       dma_unmap_len(tx_buf, len),
 230					       DMA_TO_DEVICE);
 231				dma_unmap_len_set(tx_buf, len, 0);
 232			}
 233		}
 234
 235		/* move us one more past the eop_desc for start of next pkt */
 236		tx_buf++;
 237		tx_desc++;
 238		i++;
 239		if (unlikely(!i)) {
 240			i -= tx_ring->count;
 241			tx_buf = tx_ring->tx_bi;
 242			tx_desc = I40E_TX_DESC(tx_ring, 0);
 243		}
 244
 245		prefetch(tx_desc);
 246
 247		/* update budget accounting */
 248		budget--;
 249	} while (likely(budget));
 250
 251	i += tx_ring->count;
 252	tx_ring->next_to_clean = i;
 253	u64_stats_update_begin(&tx_ring->syncp);
 254	tx_ring->stats.bytes += total_bytes;
 255	tx_ring->stats.packets += total_packets;
 256	u64_stats_update_end(&tx_ring->syncp);
 257	tx_ring->q_vector->tx.total_bytes += total_bytes;
 258	tx_ring->q_vector->tx.total_packets += total_packets;
 259
 260	if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
 261		/* check to see if there are < 4 descriptors
 262		 * waiting to be written back, then kick the hardware to force
 263		 * them to be written back in case we stay in NAPI.
 264		 * In this mode on X722 we do not enable Interrupt.
 265		 */
 266		unsigned int j = i40evf_get_tx_pending(tx_ring, false);
 267
 268		if (budget &&
 269		    ((j / WB_STRIDE) == 0) && (j > 0) &&
 270		    !test_bit(__I40E_DOWN, &vsi->state) &&
 271		    (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
 272			tx_ring->arm_wb = true;
 273	}
 274
 275	/* notify netdev of completed buffers */
 276	netdev_tx_completed_queue(txring_txq(tx_ring),
 277				  total_packets, total_bytes);
 278
 279#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 280	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 281		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 282		/* Make sure that anybody stopping the queue after this
 283		 * sees the new next_to_clean.
 284		 */
 285		smp_mb();
 286		if (__netif_subqueue_stopped(tx_ring->netdev,
 287					     tx_ring->queue_index) &&
 288		   !test_bit(__I40E_DOWN, &vsi->state)) {
 289			netif_wake_subqueue(tx_ring->netdev,
 290					    tx_ring->queue_index);
 291			++tx_ring->tx_stats.restart_queue;
 292		}
 293	}
 294
 295	return !!budget;
 296}
 297
 298/**
 299 * i40evf_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
 300 * @vsi: the VSI we care about
 301 * @q_vector: the vector on which to enable writeback
 302 *
 303 **/
 304static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
 305				  struct i40e_q_vector *q_vector)
 306{
 307	u16 flags = q_vector->tx.ring[0].flags;
 308	u32 val;
 309
 310	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
 311		return;
 312
 313	if (q_vector->arm_wb_state)
 314		return;
 315
 316	val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
 317	      I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK; /* set noitr */
 318
 319	wr32(&vsi->back->hw,
 320	     I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
 321				  vsi->base_vector - 1), val);
 322	q_vector->arm_wb_state = true;
 323}
 324
 325/**
 326 * i40evf_force_wb - Issue SW Interrupt so HW does a wb
 327 * @vsi: the VSI we care about
 328 * @q_vector: the vector  on which to force writeback
 329 *
 330 **/
 331void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
 332{
 333	u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
 334		  I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
 335		  I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
 336		  I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK
 337		  /* allow 00 to be written to the index */;
 338
 339	wr32(&vsi->back->hw,
 340	     I40E_VFINT_DYN_CTLN1(q_vector->v_idx + vsi->base_vector - 1),
 341	     val);
 342}
 343
 344/**
 345 * i40e_set_new_dynamic_itr - Find new ITR level
 346 * @rc: structure containing ring performance data
 347 *
 348 * Returns true if ITR changed, false if not
 349 *
 350 * Stores a new ITR value based on packets and byte counts during
 351 * the last interrupt.  The advantage of per interrupt computation
 352 * is faster updates and more accurate ITR for the current traffic
 353 * pattern.  Constants in this function were computed based on
 354 * theoretical maximum wire speed and thresholds were set based on
 355 * testing data as well as attempting to minimize response time
 356 * while increasing bulk throughput.
 357 **/
 358static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
 359{
 360	enum i40e_latency_range new_latency_range = rc->latency_range;
 361	struct i40e_q_vector *qv = rc->ring->q_vector;
 362	u32 new_itr = rc->itr;
 363	int bytes_per_int;
 364	int usecs;
 365
 366	if (rc->total_packets == 0 || !rc->itr)
 367		return false;
 368
 369	/* simple throttlerate management
 370	 *   0-10MB/s   lowest (50000 ints/s)
 371	 *  10-20MB/s   low    (20000 ints/s)
 372	 *  20-1249MB/s bulk   (18000 ints/s)
 373	 *  > 40000 Rx packets per second (8000 ints/s)
 374	 *
 375	 * The math works out because the divisor is in 10^(-6) which
 376	 * turns the bytes/us input value into MB/s values, but
 377	 * make sure to use usecs, as the register values written
 378	 * are in 2 usec increments in the ITR registers, and make sure
 379	 * to use the smoothed values that the countdown timer gives us.
 380	 */
 381	usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
 382	bytes_per_int = rc->total_bytes / usecs;
 383
 384	switch (new_latency_range) {
 385	case I40E_LOWEST_LATENCY:
 386		if (bytes_per_int > 10)
 387			new_latency_range = I40E_LOW_LATENCY;
 388		break;
 389	case I40E_LOW_LATENCY:
 390		if (bytes_per_int > 20)
 391			new_latency_range = I40E_BULK_LATENCY;
 392		else if (bytes_per_int <= 10)
 393			new_latency_range = I40E_LOWEST_LATENCY;
 394		break;
 395	case I40E_BULK_LATENCY:
 396	case I40E_ULTRA_LATENCY:
 397	default:
 398		if (bytes_per_int <= 20)
 399			new_latency_range = I40E_LOW_LATENCY;
 400		break;
 401	}
 402
 403	/* this is to adjust RX more aggressively when streaming small
 404	 * packets.  The value of 40000 was picked as it is just beyond
 405	 * what the hardware can receive per second if in low latency
 406	 * mode.
 407	 */
 408#define RX_ULTRA_PACKET_RATE 40000
 409
 410	if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
 411	    (&qv->rx == rc))
 412		new_latency_range = I40E_ULTRA_LATENCY;
 413
 414	rc->latency_range = new_latency_range;
 415
 416	switch (new_latency_range) {
 417	case I40E_LOWEST_LATENCY:
 418		new_itr = I40E_ITR_50K;
 419		break;
 420	case I40E_LOW_LATENCY:
 421		new_itr = I40E_ITR_20K;
 422		break;
 423	case I40E_BULK_LATENCY:
 424		new_itr = I40E_ITR_18K;
 425		break;
 426	case I40E_ULTRA_LATENCY:
 427		new_itr = I40E_ITR_8K;
 428		break;
 429	default:
 430		break;
 431	}
 432
 433	rc->total_bytes = 0;
 434	rc->total_packets = 0;
 435
 436	if (new_itr != rc->itr) {
 437		rc->itr = new_itr;
 438		return true;
 439	}
 440
 441	return false;
 442}
 443
 444/**
 445 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
 446 * @tx_ring: the tx ring to set up
 447 *
 448 * Return 0 on success, negative on error
 449 **/
 450int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
 451{
 452	struct device *dev = tx_ring->dev;
 453	int bi_size;
 454
 455	if (!dev)
 456		return -ENOMEM;
 457
 458	/* warn if we are about to overwrite the pointer */
 459	WARN_ON(tx_ring->tx_bi);
 460	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
 461	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
 462	if (!tx_ring->tx_bi)
 463		goto err;
 464
 465	/* round up to nearest 4K */
 466	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
 467	/* add u32 for head writeback, align after this takes care of
 468	 * guaranteeing this is at least one cache line in size
 469	 */
 470	tx_ring->size += sizeof(u32);
 471	tx_ring->size = ALIGN(tx_ring->size, 4096);
 472	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 473					   &tx_ring->dma, GFP_KERNEL);
 474	if (!tx_ring->desc) {
 475		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
 476			 tx_ring->size);
 477		goto err;
 478	}
 479
 480	tx_ring->next_to_use = 0;
 481	tx_ring->next_to_clean = 0;
 482	return 0;
 483
 484err:
 485	kfree(tx_ring->tx_bi);
 486	tx_ring->tx_bi = NULL;
 487	return -ENOMEM;
 488}
 489
 490/**
 491 * i40evf_clean_rx_ring - Free Rx buffers
 492 * @rx_ring: ring to be cleaned
 493 **/
 494void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
 495{
 496	struct device *dev = rx_ring->dev;
 497	unsigned long bi_size;
 498	u16 i;
 499
 500	/* ring already cleared, nothing to do */
 501	if (!rx_ring->rx_bi)
 502		return;
 503
 504	/* Free all the Rx ring sk_buffs */
 505	for (i = 0; i < rx_ring->count; i++) {
 506		struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
 507
 508		if (rx_bi->skb) {
 509			dev_kfree_skb(rx_bi->skb);
 510			rx_bi->skb = NULL;
 511		}
 512		if (!rx_bi->page)
 513			continue;
 514
 515		dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
 516		__free_pages(rx_bi->page, 0);
 517
 518		rx_bi->page = NULL;
 519		rx_bi->page_offset = 0;
 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_alloc = 0;
 529	rx_ring->next_to_clean = 0;
 530	rx_ring->next_to_use = 0;
 531}
 532
 533/**
 534 * i40evf_free_rx_resources - Free Rx resources
 535 * @rx_ring: ring to clean the resources from
 536 *
 537 * Free all receive software resources
 538 **/
 539void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
 540{
 541	i40evf_clean_rx_ring(rx_ring);
 542	kfree(rx_ring->rx_bi);
 543	rx_ring->rx_bi = NULL;
 544
 545	if (rx_ring->desc) {
 546		dma_free_coherent(rx_ring->dev, rx_ring->size,
 547				  rx_ring->desc, rx_ring->dma);
 548		rx_ring->desc = NULL;
 549	}
 550}
 551
 552/**
 553 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
 554 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
 555 *
 556 * Returns 0 on success, negative on failure
 557 **/
 558int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
 559{
 560	struct device *dev = rx_ring->dev;
 561	int bi_size;
 562
 563	/* warn if we are about to overwrite the pointer */
 564	WARN_ON(rx_ring->rx_bi);
 565	bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
 566	rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
 567	if (!rx_ring->rx_bi)
 568		goto err;
 569
 570	u64_stats_init(&rx_ring->syncp);
 571
 572	/* Round up to nearest 4K */
 573	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
 574	rx_ring->size = ALIGN(rx_ring->size, 4096);
 575	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
 576					   &rx_ring->dma, GFP_KERNEL);
 577
 578	if (!rx_ring->desc) {
 579		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
 580			 rx_ring->size);
 581		goto err;
 582	}
 583
 584	rx_ring->next_to_alloc = 0;
 585	rx_ring->next_to_clean = 0;
 586	rx_ring->next_to_use = 0;
 587
 588	return 0;
 589err:
 590	kfree(rx_ring->rx_bi);
 591	rx_ring->rx_bi = NULL;
 592	return -ENOMEM;
 593}
 594
 595/**
 596 * i40e_release_rx_desc - Store the new tail and head values
 597 * @rx_ring: ring to bump
 598 * @val: new head index
 599 **/
 600static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
 601{
 602	rx_ring->next_to_use = val;
 603
 604	/* update next to alloc since we have filled the ring */
 605	rx_ring->next_to_alloc = val;
 606
 607	/* Force memory writes to complete before letting h/w
 608	 * know there are new descriptors to fetch.  (Only
 609	 * applicable for weak-ordered memory model archs,
 610	 * such as IA-64).
 611	 */
 612	wmb();
 613	writel(val, rx_ring->tail);
 614}
 615
 616/**
 617 * i40e_alloc_mapped_page - recycle or make a new page
 618 * @rx_ring: ring to use
 619 * @bi: rx_buffer struct to modify
 620 *
 621 * Returns true if the page was successfully allocated or
 622 * reused.
 623 **/
 624static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
 625				   struct i40e_rx_buffer *bi)
 626{
 627	struct page *page = bi->page;
 628	dma_addr_t dma;
 629
 630	/* since we are recycling buffers we should seldom need to alloc */
 631	if (likely(page)) {
 632		rx_ring->rx_stats.page_reuse_count++;
 633		return true;
 634	}
 635
 636	/* alloc new page for storage */
 637	page = dev_alloc_page();
 638	if (unlikely(!page)) {
 639		rx_ring->rx_stats.alloc_page_failed++;
 640		return false;
 641	}
 642
 643	/* map page for use */
 644	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
 645
 646	/* if mapping failed free memory back to system since
 647	 * there isn't much point in holding memory we can't use
 648	 */
 649	if (dma_mapping_error(rx_ring->dev, dma)) {
 650		__free_pages(page, 0);
 651		rx_ring->rx_stats.alloc_page_failed++;
 652		return false;
 653	}
 654
 655	bi->dma = dma;
 656	bi->page = page;
 657	bi->page_offset = 0;
 658
 659	return true;
 660}
 661
 662/**
 663 * i40e_receive_skb - Send a completed packet up the stack
 664 * @rx_ring:  rx ring in play
 665 * @skb: packet to send up
 666 * @vlan_tag: vlan tag for packet
 667 **/
 668static void i40e_receive_skb(struct i40e_ring *rx_ring,
 669			     struct sk_buff *skb, u16 vlan_tag)
 670{
 671	struct i40e_q_vector *q_vector = rx_ring->q_vector;
 672
 673	if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
 674	    (vlan_tag & VLAN_VID_MASK))
 675		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
 676
 677	napi_gro_receive(&q_vector->napi, skb);
 678}
 679
 680/**
 681 * i40evf_alloc_rx_buffers - Replace used receive buffers
 682 * @rx_ring: ring to place buffers on
 683 * @cleaned_count: number of buffers to replace
 684 *
 685 * Returns false if all allocations were successful, true if any fail
 686 **/
 687bool i40evf_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
 688{
 689	u16 ntu = rx_ring->next_to_use;
 690	union i40e_rx_desc *rx_desc;
 691	struct i40e_rx_buffer *bi;
 692
 693	/* do nothing if no valid netdev defined */
 694	if (!rx_ring->netdev || !cleaned_count)
 695		return false;
 696
 697	rx_desc = I40E_RX_DESC(rx_ring, ntu);
 698	bi = &rx_ring->rx_bi[ntu];
 699
 700	do {
 701		if (!i40e_alloc_mapped_page(rx_ring, bi))
 702			goto no_buffers;
 703
 704		/* Refresh the desc even if buffer_addrs didn't change
 705		 * because each write-back erases this info.
 706		 */
 707		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 708
 709		rx_desc++;
 710		bi++;
 711		ntu++;
 712		if (unlikely(ntu == rx_ring->count)) {
 713			rx_desc = I40E_RX_DESC(rx_ring, 0);
 714			bi = rx_ring->rx_bi;
 715			ntu = 0;
 716		}
 717
 718		/* clear the status bits for the next_to_use descriptor */
 719		rx_desc->wb.qword1.status_error_len = 0;
 720
 721		cleaned_count--;
 722	} while (cleaned_count);
 723
 724	if (rx_ring->next_to_use != ntu)
 725		i40e_release_rx_desc(rx_ring, ntu);
 726
 727	return false;
 728
 729no_buffers:
 730	if (rx_ring->next_to_use != ntu)
 731		i40e_release_rx_desc(rx_ring, ntu);
 732
 733	/* make sure to come back via polling to try again after
 734	 * allocation failure
 735	 */
 736	return true;
 737}
 738
 739/**
 740 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
 741 * @vsi: the VSI we care about
 742 * @skb: skb currently being received and modified
 743 * @rx_desc: the receive descriptor
 744 *
 745 * skb->protocol must be set before this function is called
 746 **/
 747static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
 748				    struct sk_buff *skb,
 749				    union i40e_rx_desc *rx_desc)
 750{
 751	struct i40e_rx_ptype_decoded decoded;
 752	u32 rx_error, rx_status;
 753	bool ipv4, ipv6;
 754	u8 ptype;
 755	u64 qword;
 756
 757	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
 758	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
 759	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
 760		   I40E_RXD_QW1_ERROR_SHIFT;
 761	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
 762		    I40E_RXD_QW1_STATUS_SHIFT;
 763	decoded = decode_rx_desc_ptype(ptype);
 764
 765	skb->ip_summed = CHECKSUM_NONE;
 766
 767	skb_checksum_none_assert(skb);
 768
 769	/* Rx csum enabled and ip headers found? */
 770	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
 771		return;
 772
 773	/* did the hardware decode the packet and checksum? */
 774	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
 775		return;
 776
 777	/* both known and outer_ip must be set for the below code to work */
 778	if (!(decoded.known && decoded.outer_ip))
 779		return;
 780
 781	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
 782	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
 783	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
 784	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
 785
 786	if (ipv4 &&
 787	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
 788			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
 789		goto checksum_fail;
 790
 791	/* likely incorrect csum if alternate IP extension headers found */
 792	if (ipv6 &&
 793	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
 794		/* don't increment checksum err here, non-fatal err */
 795		return;
 796
 797	/* there was some L4 error, count error and punt packet to the stack */
 798	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
 799		goto checksum_fail;
 800
 801	/* handle packets that were not able to be checksummed due
 802	 * to arrival speed, in this case the stack can compute
 803	 * the csum.
 804	 */
 805	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
 806		return;
 807
 808	/* If there is an outer header present that might contain a checksum
 809	 * we need to bump the checksum level by 1 to reflect the fact that
 810	 * we are indicating we validated the inner checksum.
 811	 */
 812	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
 813		skb->csum_level = 1;
 814
 815	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
 816	switch (decoded.inner_prot) {
 817	case I40E_RX_PTYPE_INNER_PROT_TCP:
 818	case I40E_RX_PTYPE_INNER_PROT_UDP:
 819	case I40E_RX_PTYPE_INNER_PROT_SCTP:
 820		skb->ip_summed = CHECKSUM_UNNECESSARY;
 821		/* fall though */
 822	default:
 823		break;
 824	}
 825
 826	return;
 827
 828checksum_fail:
 829	vsi->back->hw_csum_rx_error++;
 830}
 831
 832/**
 833 * i40e_ptype_to_htype - get a hash type
 834 * @ptype: the ptype value from the descriptor
 835 *
 836 * Returns a hash type to be used by skb_set_hash
 837 **/
 838static inline int i40e_ptype_to_htype(u8 ptype)
 839{
 840	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
 841
 842	if (!decoded.known)
 843		return PKT_HASH_TYPE_NONE;
 844
 845	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
 846	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
 847		return PKT_HASH_TYPE_L4;
 848	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
 849		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
 850		return PKT_HASH_TYPE_L3;
 851	else
 852		return PKT_HASH_TYPE_L2;
 853}
 854
 855/**
 856 * i40e_rx_hash - set the hash value in the skb
 857 * @ring: descriptor ring
 858 * @rx_desc: specific descriptor
 859 **/
 860static inline void i40e_rx_hash(struct i40e_ring *ring,
 861				union i40e_rx_desc *rx_desc,
 862				struct sk_buff *skb,
 863				u8 rx_ptype)
 864{
 865	u32 hash;
 866	const __le64 rss_mask =
 867		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
 868			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
 869
 870	if (ring->netdev->features & NETIF_F_RXHASH)
 871		return;
 872
 873	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
 874		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
 875		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
 876	}
 877}
 878
 879/**
 880 * i40evf_process_skb_fields - Populate skb header fields from Rx descriptor
 881 * @rx_ring: rx descriptor ring packet is being transacted on
 882 * @rx_desc: pointer to the EOP Rx descriptor
 883 * @skb: pointer to current skb being populated
 884 * @rx_ptype: the packet type decoded by hardware
 885 *
 886 * This function checks the ring, descriptor, and packet information in
 887 * order to populate the hash, checksum, VLAN, protocol, and
 888 * other fields within the skb.
 889 **/
 890static inline
 891void i40evf_process_skb_fields(struct i40e_ring *rx_ring,
 892			       union i40e_rx_desc *rx_desc, struct sk_buff *skb,
 893			       u8 rx_ptype)
 894{
 895	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
 896
 897	/* modifies the skb - consumes the enet header */
 898	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 899
 900	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
 901
 902	skb_record_rx_queue(skb, rx_ring->queue_index);
 903}
 904
 905/**
 906 * i40e_pull_tail - i40e specific version of skb_pull_tail
 907 * @rx_ring: rx descriptor ring packet is being transacted on
 908 * @skb: pointer to current skb being adjusted
 909 *
 910 * This function is an i40e specific version of __pskb_pull_tail.  The
 911 * main difference between this version and the original function is that
 912 * this function can make several assumptions about the state of things
 913 * that allow for significant optimizations versus the standard function.
 914 * As a result we can do things like drop a frag and maintain an accurate
 915 * truesize for the skb.
 916 */
 917static void i40e_pull_tail(struct i40e_ring *rx_ring, struct sk_buff *skb)
 918{
 919	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
 920	unsigned char *va;
 921	unsigned int pull_len;
 922
 923	/* it is valid to use page_address instead of kmap since we are
 924	 * working with pages allocated out of the lomem pool per
 925	 * alloc_page(GFP_ATOMIC)
 926	 */
 927	va = skb_frag_address(frag);
 928
 929	/* we need the header to contain the greater of either ETH_HLEN or
 930	 * 60 bytes if the skb->len is less than 60 for skb_pad.
 931	 */
 932	pull_len = eth_get_headlen(va, I40E_RX_HDR_SIZE);
 933
 934	/* align pull length to size of long to optimize memcpy performance */
 935	skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
 936
 937	/* update all of the pointers */
 938	skb_frag_size_sub(frag, pull_len);
 939	frag->page_offset += pull_len;
 940	skb->data_len -= pull_len;
 941	skb->tail += pull_len;
 942}
 943
 944/**
 945 * i40e_cleanup_headers - Correct empty headers
 946 * @rx_ring: rx descriptor ring packet is being transacted on
 947 * @skb: pointer to current skb being fixed
 948 *
 949 * Also address the case where we are pulling data in on pages only
 950 * and as such no data is present in the skb header.
 951 *
 952 * In addition if skb is not at least 60 bytes we need to pad it so that
 953 * it is large enough to qualify as a valid Ethernet frame.
 954 *
 955 * Returns true if an error was encountered and skb was freed.
 956 **/
 957static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
 958{
 959	/* place header in linear portion of buffer */
 960	if (skb_is_nonlinear(skb))
 961		i40e_pull_tail(rx_ring, skb);
 962
 963	/* if eth_skb_pad returns an error the skb was freed */
 964	if (eth_skb_pad(skb))
 965		return true;
 966
 967	return false;
 968}
 969
 970/**
 971 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
 972 * @rx_ring: rx descriptor ring to store buffers on
 973 * @old_buff: donor buffer to have page reused
 974 *
 975 * Synchronizes page for reuse by the adapter
 976 **/
 977static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
 978			       struct i40e_rx_buffer *old_buff)
 979{
 980	struct i40e_rx_buffer *new_buff;
 981	u16 nta = rx_ring->next_to_alloc;
 982
 983	new_buff = &rx_ring->rx_bi[nta];
 984
 985	/* update, and store next to alloc */
 986	nta++;
 987	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 988
 989	/* transfer page from old buffer to new buffer */
 990	*new_buff = *old_buff;
 991}
 992
 993/**
 994 * i40e_page_is_reserved - check if reuse is possible
 995 * @page: page struct to check
 996 */
 997static inline bool i40e_page_is_reserved(struct page *page)
 998{
 999	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1000}
1001
1002/**
1003 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1004 * @rx_ring: rx descriptor ring to transact packets on
1005 * @rx_buffer: buffer containing page to add
1006 * @rx_desc: descriptor containing length of buffer written by hardware
1007 * @skb: sk_buff to place the data into
1008 *
1009 * This function will add the data contained in rx_buffer->page to the skb.
1010 * This is done either through a direct copy if the data in the buffer is
1011 * less than the skb header size, otherwise it will just attach the page as
1012 * a frag to the skb.
1013 *
1014 * The function will then update the page offset if necessary and return
1015 * true if the buffer can be reused by the adapter.
1016 **/
1017static bool i40e_add_rx_frag(struct i40e_ring *rx_ring,
1018			     struct i40e_rx_buffer *rx_buffer,
1019			     union i40e_rx_desc *rx_desc,
1020			     struct sk_buff *skb)
1021{
1022	struct page *page = rx_buffer->page;
1023	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1024	unsigned int size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1025			    I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1026#if (PAGE_SIZE < 8192)
1027	unsigned int truesize = I40E_RXBUFFER_2048;
1028#else
1029	unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1030	unsigned int last_offset = PAGE_SIZE - I40E_RXBUFFER_2048;
1031#endif
1032
1033	/* will the data fit in the skb we allocated? if so, just
1034	 * copy it as it is pretty small anyway
1035	 */
1036	if ((size <= I40E_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1037		unsigned char *va = page_address(page) + rx_buffer->page_offset;
1038
1039		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1040
1041		/* page is not reserved, we can reuse buffer as-is */
1042		if (likely(!i40e_page_is_reserved(page)))
1043			return true;
1044
1045		/* this page cannot be reused so discard it */
1046		__free_pages(page, 0);
1047		return false;
1048	}
1049
1050	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1051			rx_buffer->page_offset, size, truesize);
1052
1053	/* avoid re-using remote pages */
1054	if (unlikely(i40e_page_is_reserved(page)))
1055		return false;
1056
1057#if (PAGE_SIZE < 8192)
1058	/* if we are only owner of page we can reuse it */
1059	if (unlikely(page_count(page) != 1))
1060		return false;
1061
1062	/* flip page offset to other buffer */
1063	rx_buffer->page_offset ^= truesize;
1064#else
1065	/* move offset up to the next cache line */
1066	rx_buffer->page_offset += truesize;
1067
1068	if (rx_buffer->page_offset > last_offset)
1069		return false;
1070#endif
1071
1072	/* Even if we own the page, we are not allowed to use atomic_set()
1073	 * This would break get_page_unless_zero() users.
1074	 */
1075	get_page(rx_buffer->page);
1076
1077	return true;
1078}
1079
1080/**
1081 * i40evf_fetch_rx_buffer - Allocate skb and populate it
1082 * @rx_ring: rx descriptor ring to transact packets on
1083 * @rx_desc: descriptor containing info written by hardware
1084 *
1085 * This function allocates an skb on the fly, and populates it with the page
1086 * data from the current receive descriptor, taking care to set up the skb
1087 * correctly, as well as handling calling the page recycle function if
1088 * necessary.
1089 */
1090static inline
1091struct sk_buff *i40evf_fetch_rx_buffer(struct i40e_ring *rx_ring,
1092				       union i40e_rx_desc *rx_desc)
1093{
1094	struct i40e_rx_buffer *rx_buffer;
1095	struct sk_buff *skb;
1096	struct page *page;
1097
1098	rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1099	page = rx_buffer->page;
1100	prefetchw(page);
1101
1102	skb = rx_buffer->skb;
1103
1104	if (likely(!skb)) {
1105		void *page_addr = page_address(page) + rx_buffer->page_offset;
1106
1107		/* prefetch first cache line of first page */
1108		prefetch(page_addr);
1109#if L1_CACHE_BYTES < 128
1110		prefetch(page_addr + L1_CACHE_BYTES);
1111#endif
1112
1113		/* allocate a skb to store the frags */
1114		skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1115				       I40E_RX_HDR_SIZE,
1116				       GFP_ATOMIC | __GFP_NOWARN);
1117		if (unlikely(!skb)) {
1118			rx_ring->rx_stats.alloc_buff_failed++;
1119			return NULL;
1120		}
1121
1122		/* we will be copying header into skb->data in
1123		 * pskb_may_pull so it is in our interest to prefetch
1124		 * it now to avoid a possible cache miss
1125		 */
1126		prefetchw(skb->data);
1127	} else {
1128		rx_buffer->skb = NULL;
1129	}
1130
1131	/* we are reusing so sync this buffer for CPU use */
1132	dma_sync_single_range_for_cpu(rx_ring->dev,
1133				      rx_buffer->dma,
1134				      rx_buffer->page_offset,
1135				      I40E_RXBUFFER_2048,
1136				      DMA_FROM_DEVICE);
1137
1138	/* pull page into skb */
1139	if (i40e_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
1140		/* hand second half of page back to the ring */
1141		i40e_reuse_rx_page(rx_ring, rx_buffer);
1142		rx_ring->rx_stats.page_reuse_count++;
1143	} else {
1144		/* we are not reusing the buffer so unmap it */
1145		dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
1146			       DMA_FROM_DEVICE);
1147	}
1148
1149	/* clear contents of buffer_info */
1150	rx_buffer->page = NULL;
1151
1152	return skb;
1153}
1154
1155/**
1156 * i40e_is_non_eop - process handling of non-EOP buffers
1157 * @rx_ring: Rx ring being processed
1158 * @rx_desc: Rx descriptor for current buffer
1159 * @skb: Current socket buffer containing buffer in progress
1160 *
1161 * This function updates next to clean.  If the buffer is an EOP buffer
1162 * this function exits returning false, otherwise it will place the
1163 * sk_buff in the next buffer to be chained and return true indicating
1164 * that this is in fact a non-EOP buffer.
1165 **/
1166static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1167			    union i40e_rx_desc *rx_desc,
1168			    struct sk_buff *skb)
1169{
1170	u32 ntc = rx_ring->next_to_clean + 1;
1171
1172	/* fetch, update, and store next to clean */
1173	ntc = (ntc < rx_ring->count) ? ntc : 0;
1174	rx_ring->next_to_clean = ntc;
1175
1176	prefetch(I40E_RX_DESC(rx_ring, ntc));
1177
1178	/* if we are the last buffer then there is nothing else to do */
1179#define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1180	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1181		return false;
1182
1183	/* place skb in next buffer to be received */
1184	rx_ring->rx_bi[ntc].skb = skb;
1185	rx_ring->rx_stats.non_eop_descs++;
1186
1187	return true;
1188}
1189
1190/**
1191 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1192 * @rx_ring: rx descriptor ring to transact packets on
1193 * @budget: Total limit on number of packets to process
1194 *
1195 * This function provides a "bounce buffer" approach to Rx interrupt
1196 * processing.  The advantage to this is that on systems that have
1197 * expensive overhead for IOMMU access this provides a means of avoiding
1198 * it by maintaining the mapping of the page to the system.
1199 *
1200 * Returns amount of work completed
1201 **/
1202static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1203{
1204	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1205	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1206	bool failure = false;
1207
1208	while (likely(total_rx_packets < budget)) {
1209		union i40e_rx_desc *rx_desc;
1210		struct sk_buff *skb;
1211		u16 vlan_tag;
1212		u8 rx_ptype;
1213		u64 qword;
1214
1215		/* return some buffers to hardware, one at a time is too slow */
1216		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1217			failure = failure ||
1218				  i40evf_alloc_rx_buffers(rx_ring, cleaned_count);
1219			cleaned_count = 0;
1220		}
1221
1222		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1223
1224		/* status_error_len will always be zero for unused descriptors
1225		 * because it's cleared in cleanup, and overlaps with hdr_addr
1226		 * which is always zero because packet split isn't used, if the
1227		 * hardware wrote DD then it will be non-zero
1228		 */
1229		if (!i40e_test_staterr(rx_desc,
1230				       BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1231			break;
1232
1233		/* This memory barrier is needed to keep us from reading
1234		 * any other fields out of the rx_desc until we know the
1235		 * DD bit is set.
1236		 */
1237		dma_rmb();
1238
1239		skb = i40evf_fetch_rx_buffer(rx_ring, rx_desc);
1240		if (!skb)
1241			break;
1242
1243		cleaned_count++;
1244
1245		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1246			continue;
1247
1248		/* ERR_MASK will only have valid bits if EOP set, and
1249		 * what we are doing here is actually checking
1250		 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1251		 * the error field
1252		 */
1253		if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1254			dev_kfree_skb_any(skb);
1255			continue;
1256		}
1257
1258		if (i40e_cleanup_headers(rx_ring, skb))
1259			continue;
1260
1261		/* probably a little skewed due to removing CRC */
1262		total_rx_bytes += skb->len;
1263
1264		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1265		rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1266			   I40E_RXD_QW1_PTYPE_SHIFT;
1267
1268		/* populate checksum, VLAN, and protocol */
1269		i40evf_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1270
1271
1272		vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1273			   le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1274
1275		i40e_receive_skb(rx_ring, skb, vlan_tag);
1276
1277		/* update budget accounting */
1278		total_rx_packets++;
1279	}
1280
1281	u64_stats_update_begin(&rx_ring->syncp);
1282	rx_ring->stats.packets += total_rx_packets;
1283	rx_ring->stats.bytes += total_rx_bytes;
1284	u64_stats_update_end(&rx_ring->syncp);
1285	rx_ring->q_vector->rx.total_packets += total_rx_packets;
1286	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1287
1288	/* guarantee a trip back through this routine if there was a failure */
1289	return failure ? budget : total_rx_packets;
1290}
1291
1292static u32 i40e_buildreg_itr(const int type, const u16 itr)
1293{
1294	u32 val;
1295
1296	val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1297	      /* Don't clear PBA because that can cause lost interrupts that
1298	       * came in while we were cleaning/polling
1299	       */
1300	      (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1301	      (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1302
1303	return val;
1304}
1305
1306/* a small macro to shorten up some long lines */
1307#define INTREG I40E_VFINT_DYN_CTLN1
1308static inline int get_rx_itr_enabled(struct i40e_vsi *vsi, int idx)
1309{
1310	struct i40evf_adapter *adapter = vsi->back;
1311
1312	return !!(adapter->rx_rings[idx].rx_itr_setting);
1313}
1314
1315static inline int get_tx_itr_enabled(struct i40e_vsi *vsi, int idx)
1316{
1317	struct i40evf_adapter *adapter = vsi->back;
1318
1319	return !!(adapter->tx_rings[idx].tx_itr_setting);
1320}
1321
1322/**
1323 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1324 * @vsi: the VSI we care about
1325 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1326 *
1327 **/
1328static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1329					  struct i40e_q_vector *q_vector)
1330{
1331	struct i40e_hw *hw = &vsi->back->hw;
1332	bool rx = false, tx = false;
1333	u32 rxval, txval;
1334	int vector;
1335	int idx = q_vector->v_idx;
1336	int rx_itr_setting, tx_itr_setting;
1337
1338	vector = (q_vector->v_idx + vsi->base_vector);
1339
1340	/* avoid dynamic calculation if in countdown mode OR if
1341	 * all dynamic is disabled
1342	 */
1343	rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1344
1345	rx_itr_setting = get_rx_itr_enabled(vsi, idx);
1346	tx_itr_setting = get_tx_itr_enabled(vsi, idx);
1347
1348	if (q_vector->itr_countdown > 0 ||
1349	    (!ITR_IS_DYNAMIC(rx_itr_setting) &&
1350	     !ITR_IS_DYNAMIC(tx_itr_setting))) {
1351		goto enable_int;
1352	}
1353
1354	if (ITR_IS_DYNAMIC(rx_itr_setting)) {
1355		rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1356		rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1357	}
1358
1359	if (ITR_IS_DYNAMIC(tx_itr_setting)) {
1360		tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1361		txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1362	}
1363
1364	if (rx || tx) {
1365		/* get the higher of the two ITR adjustments and
1366		 * use the same value for both ITR registers
1367		 * when in adaptive mode (Rx and/or Tx)
1368		 */
1369		u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1370
1371		q_vector->tx.itr = q_vector->rx.itr = itr;
1372		txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1373		tx = true;
1374		rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1375		rx = true;
1376	}
1377
1378	/* only need to enable the interrupt once, but need
1379	 * to possibly update both ITR values
1380	 */
1381	if (rx) {
1382		/* set the INTENA_MSK_MASK so that this first write
1383		 * won't actually enable the interrupt, instead just
1384		 * updating the ITR (it's bit 31 PF and VF)
1385		 */
1386		rxval |= BIT(31);
1387		/* don't check _DOWN because interrupt isn't being enabled */
1388		wr32(hw, INTREG(vector - 1), rxval);
1389	}
1390
1391enable_int:
1392	if (!test_bit(__I40E_DOWN, &vsi->state))
1393		wr32(hw, INTREG(vector - 1), txval);
1394
1395	if (q_vector->itr_countdown)
1396		q_vector->itr_countdown--;
1397	else
1398		q_vector->itr_countdown = ITR_COUNTDOWN_START;
1399}
1400
1401/**
1402 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1403 * @napi: napi struct with our devices info in it
1404 * @budget: amount of work driver is allowed to do this pass, in packets
1405 *
1406 * This function will clean all queues associated with a q_vector.
1407 *
1408 * Returns the amount of work done
1409 **/
1410int i40evf_napi_poll(struct napi_struct *napi, int budget)
1411{
1412	struct i40e_q_vector *q_vector =
1413			       container_of(napi, struct i40e_q_vector, napi);
1414	struct i40e_vsi *vsi = q_vector->vsi;
1415	struct i40e_ring *ring;
1416	bool clean_complete = true;
1417	bool arm_wb = false;
1418	int budget_per_ring;
1419	int work_done = 0;
1420
1421	if (test_bit(__I40E_DOWN, &vsi->state)) {
1422		napi_complete(napi);
1423		return 0;
1424	}
1425
1426	/* Since the actual Tx work is minimal, we can give the Tx a larger
1427	 * budget and be more aggressive about cleaning up the Tx descriptors.
1428	 */
1429	i40e_for_each_ring(ring, q_vector->tx) {
1430		if (!i40e_clean_tx_irq(vsi, ring, budget)) {
1431			clean_complete = false;
1432			continue;
1433		}
1434		arm_wb |= ring->arm_wb;
1435		ring->arm_wb = false;
1436	}
1437
1438	/* Handle case where we are called by netpoll with a budget of 0 */
1439	if (budget <= 0)
1440		goto tx_only;
1441
1442	/* We attempt to distribute budget to each Rx queue fairly, but don't
1443	 * allow the budget to go below 1 because that would exit polling early.
1444	 */
1445	budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1446
1447	i40e_for_each_ring(ring, q_vector->rx) {
1448		int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
1449
1450		work_done += cleaned;
1451		/* if we clean as many as budgeted, we must not be done */
1452		if (cleaned >= budget_per_ring)
1453			clean_complete = false;
1454	}
1455
1456	/* If work not completed, return budget and polling will return */
1457	if (!clean_complete) {
1458		const cpumask_t *aff_mask = &q_vector->affinity_mask;
1459		int cpu_id = smp_processor_id();
1460
1461		/* It is possible that the interrupt affinity has changed but,
1462		 * if the cpu is pegged at 100%, polling will never exit while
1463		 * traffic continues and the interrupt will be stuck on this
1464		 * cpu.  We check to make sure affinity is correct before we
1465		 * continue to poll, otherwise we must stop polling so the
1466		 * interrupt can move to the correct cpu.
1467		 */
1468		if (likely(cpumask_test_cpu(cpu_id, aff_mask))) {
1469tx_only:
1470			if (arm_wb) {
1471				q_vector->tx.ring[0].tx_stats.tx_force_wb++;
1472				i40e_enable_wb_on_itr(vsi, q_vector);
1473			}
1474			return budget;
1475		}
1476	}
1477
1478	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1479		q_vector->arm_wb_state = false;
1480
1481	/* Work is done so exit the polling mode and re-enable the interrupt */
1482	napi_complete_done(napi, work_done);
1483
1484	/* If we're prematurely stopping polling to fix the interrupt
1485	 * affinity we want to make sure polling starts back up so we
1486	 * issue a call to i40evf_force_wb which triggers a SW interrupt.
1487	 */
1488	if (!clean_complete)
1489		i40evf_force_wb(vsi, q_vector);
1490	else
1491		i40e_update_enable_itr(vsi, q_vector);
1492
1493	return min(work_done, budget - 1);
1494}
1495
1496/**
1497 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1498 * @skb:     send buffer
1499 * @tx_ring: ring to send buffer on
1500 * @flags:   the tx flags to be set
1501 *
1502 * Checks the skb and set up correspondingly several generic transmit flags
1503 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1504 *
1505 * Returns error code indicate the frame should be dropped upon error and the
1506 * otherwise  returns 0 to indicate the flags has been set properly.
1507 **/
1508static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1509					       struct i40e_ring *tx_ring,
1510					       u32 *flags)
1511{
1512	__be16 protocol = skb->protocol;
1513	u32  tx_flags = 0;
1514
1515	if (protocol == htons(ETH_P_8021Q) &&
1516	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1517		/* When HW VLAN acceleration is turned off by the user the
1518		 * stack sets the protocol to 8021q so that the driver
1519		 * can take any steps required to support the SW only
1520		 * VLAN handling.  In our case the driver doesn't need
1521		 * to take any further steps so just set the protocol
1522		 * to the encapsulated ethertype.
1523		 */
1524		skb->protocol = vlan_get_protocol(skb);
1525		goto out;
1526	}
1527
1528	/* if we have a HW VLAN tag being added, default to the HW one */
1529	if (skb_vlan_tag_present(skb)) {
1530		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1531		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1532	/* else if it is a SW VLAN, check the next protocol and store the tag */
1533	} else if (protocol == htons(ETH_P_8021Q)) {
1534		struct vlan_hdr *vhdr, _vhdr;
1535
1536		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1537		if (!vhdr)
1538			return -EINVAL;
1539
1540		protocol = vhdr->h_vlan_encapsulated_proto;
1541		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1542		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1543	}
1544
1545out:
1546	*flags = tx_flags;
1547	return 0;
1548}
1549
1550/**
1551 * i40e_tso - set up the tso context descriptor
1552 * @skb:      ptr to the skb we're sending
1553 * @hdr_len:  ptr to the size of the packet header
1554 * @cd_type_cmd_tso_mss: Quad Word 1
1555 *
1556 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1557 **/
1558static int i40e_tso(struct sk_buff *skb, u8 *hdr_len, u64 *cd_type_cmd_tso_mss)
1559{
1560	u64 cd_cmd, cd_tso_len, cd_mss;
1561	union {
1562		struct iphdr *v4;
1563		struct ipv6hdr *v6;
1564		unsigned char *hdr;
1565	} ip;
1566	union {
1567		struct tcphdr *tcp;
1568		struct udphdr *udp;
1569		unsigned char *hdr;
1570	} l4;
1571	u32 paylen, l4_offset;
1572	int err;
1573
1574	if (skb->ip_summed != CHECKSUM_PARTIAL)
1575		return 0;
1576
1577	if (!skb_is_gso(skb))
1578		return 0;
1579
1580	err = skb_cow_head(skb, 0);
1581	if (err < 0)
1582		return err;
1583
1584	ip.hdr = skb_network_header(skb);
1585	l4.hdr = skb_transport_header(skb);
1586
1587	/* initialize outer IP header fields */
1588	if (ip.v4->version == 4) {
1589		ip.v4->tot_len = 0;
1590		ip.v4->check = 0;
1591	} else {
1592		ip.v6->payload_len = 0;
1593	}
1594
1595	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1596					 SKB_GSO_GRE_CSUM |
1597					 SKB_GSO_IPXIP4 |
1598					 SKB_GSO_IPXIP6 |
1599					 SKB_GSO_UDP_TUNNEL |
1600					 SKB_GSO_UDP_TUNNEL_CSUM)) {
1601		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
1602		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
1603			l4.udp->len = 0;
1604
1605			/* determine offset of outer transport header */
1606			l4_offset = l4.hdr - skb->data;
1607
1608			/* remove payload length from outer checksum */
1609			paylen = skb->len - l4_offset;
1610			csum_replace_by_diff(&l4.udp->check, htonl(paylen));
1611		}
1612
1613		/* reset pointers to inner headers */
1614		ip.hdr = skb_inner_network_header(skb);
1615		l4.hdr = skb_inner_transport_header(skb);
1616
1617		/* initialize inner IP header fields */
1618		if (ip.v4->version == 4) {
1619			ip.v4->tot_len = 0;
1620			ip.v4->check = 0;
1621		} else {
1622			ip.v6->payload_len = 0;
1623		}
1624	}
1625
1626	/* determine offset of inner transport header */
1627	l4_offset = l4.hdr - skb->data;
1628
1629	/* remove payload length from inner checksum */
1630	paylen = skb->len - l4_offset;
1631	csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
1632
1633	/* compute length of segmentation header */
1634	*hdr_len = (l4.tcp->doff * 4) + l4_offset;
1635
1636	/* find the field values */
1637	cd_cmd = I40E_TX_CTX_DESC_TSO;
1638	cd_tso_len = skb->len - *hdr_len;
1639	cd_mss = skb_shinfo(skb)->gso_size;
1640	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1641				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1642				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1643	return 1;
1644}
1645
1646/**
1647 * i40e_tx_enable_csum - Enable Tx checksum offloads
1648 * @skb: send buffer
1649 * @tx_flags: pointer to Tx flags currently set
1650 * @td_cmd: Tx descriptor command bits to set
1651 * @td_offset: Tx descriptor header offsets to set
1652 * @tx_ring: Tx descriptor ring
1653 * @cd_tunneling: ptr to context desc bits
1654 **/
1655static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1656			       u32 *td_cmd, u32 *td_offset,
1657			       struct i40e_ring *tx_ring,
1658			       u32 *cd_tunneling)
1659{
1660	union {
1661		struct iphdr *v4;
1662		struct ipv6hdr *v6;
1663		unsigned char *hdr;
1664	} ip;
1665	union {
1666		struct tcphdr *tcp;
1667		struct udphdr *udp;
1668		unsigned char *hdr;
1669	} l4;
1670	unsigned char *exthdr;
1671	u32 offset, cmd = 0;
1672	__be16 frag_off;
1673	u8 l4_proto = 0;
1674
1675	if (skb->ip_summed != CHECKSUM_PARTIAL)
1676		return 0;
1677
1678	ip.hdr = skb_network_header(skb);
1679	l4.hdr = skb_transport_header(skb);
1680
1681	/* compute outer L2 header size */
1682	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1683
1684	if (skb->encapsulation) {
1685		u32 tunnel = 0;
1686		/* define outer network header type */
1687		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1688			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
1689				  I40E_TX_CTX_EXT_IP_IPV4 :
1690				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1691
1692			l4_proto = ip.v4->protocol;
1693		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1694			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
1695
1696			exthdr = ip.hdr + sizeof(*ip.v6);
1697			l4_proto = ip.v6->nexthdr;
1698			if (l4.hdr != exthdr)
1699				ipv6_skip_exthdr(skb, exthdr - skb->data,
1700						 &l4_proto, &frag_off);
1701		}
1702
1703		/* define outer transport */
1704		switch (l4_proto) {
1705		case IPPROTO_UDP:
1706			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
1707			*tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1708			break;
1709		case IPPROTO_GRE:
1710			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
1711			*tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1712			break;
1713		case IPPROTO_IPIP:
1714		case IPPROTO_IPV6:
1715			*tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1716			l4.hdr = skb_inner_network_header(skb);
1717			break;
1718		default:
1719			if (*tx_flags & I40E_TX_FLAGS_TSO)
1720				return -1;
1721
1722			skb_checksum_help(skb);
1723			return 0;
1724		}
1725
1726		/* compute outer L3 header size */
1727		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
1728			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
1729
1730		/* switch IP header pointer from outer to inner header */
1731		ip.hdr = skb_inner_network_header(skb);
1732
1733		/* compute tunnel header size */
1734		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
1735			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1736
1737		/* indicate if we need to offload outer UDP header */
1738		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
1739		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
1740		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
1741			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1742
1743		/* record tunnel offload values */
1744		*cd_tunneling |= tunnel;
1745
1746		/* switch L4 header pointer from outer to inner */
1747		l4.hdr = skb_inner_transport_header(skb);
1748		l4_proto = 0;
1749
1750		/* reset type as we transition from outer to inner headers */
1751		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
1752		if (ip.v4->version == 4)
1753			*tx_flags |= I40E_TX_FLAGS_IPV4;
1754		if (ip.v6->version == 6)
1755			*tx_flags |= I40E_TX_FLAGS_IPV6;
1756	}
1757
1758	/* Enable IP checksum offloads */
1759	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1760		l4_proto = ip.v4->protocol;
1761		/* the stack computes the IP header already, the only time we
1762		 * need the hardware to recompute it is in the case of TSO.
1763		 */
1764		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
1765		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
1766		       I40E_TX_DESC_CMD_IIPT_IPV4;
1767	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1768		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1769
1770		exthdr = ip.hdr + sizeof(*ip.v6);
1771		l4_proto = ip.v6->nexthdr;
1772		if (l4.hdr != exthdr)
1773			ipv6_skip_exthdr(skb, exthdr - skb->data,
1774					 &l4_proto, &frag_off);
1775	}
1776
1777	/* compute inner L3 header size */
1778	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1779
1780	/* Enable L4 checksum offloads */
1781	switch (l4_proto) {
1782	case IPPROTO_TCP:
1783		/* enable checksum offloads */
1784		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1785		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1786		break;
1787	case IPPROTO_SCTP:
1788		/* enable SCTP checksum offload */
1789		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1790		offset |= (sizeof(struct sctphdr) >> 2) <<
1791			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1792		break;
1793	case IPPROTO_UDP:
1794		/* enable UDP checksum offload */
1795		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1796		offset |= (sizeof(struct udphdr) >> 2) <<
1797			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1798		break;
1799	default:
1800		if (*tx_flags & I40E_TX_FLAGS_TSO)
1801			return -1;
1802		skb_checksum_help(skb);
1803		return 0;
1804	}
1805
1806	*td_cmd |= cmd;
1807	*td_offset |= offset;
1808
1809	return 1;
1810}
1811
1812/**
1813 * i40e_create_tx_ctx Build the Tx context descriptor
1814 * @tx_ring:  ring to create the descriptor on
1815 * @cd_type_cmd_tso_mss: Quad Word 1
1816 * @cd_tunneling: Quad Word 0 - bits 0-31
1817 * @cd_l2tag2: Quad Word 0 - bits 32-63
1818 **/
1819static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1820			       const u64 cd_type_cmd_tso_mss,
1821			       const u32 cd_tunneling, const u32 cd_l2tag2)
1822{
1823	struct i40e_tx_context_desc *context_desc;
1824	int i = tx_ring->next_to_use;
1825
1826	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1827	    !cd_tunneling && !cd_l2tag2)
1828		return;
1829
1830	/* grab the next descriptor */
1831	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1832
1833	i++;
1834	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1835
1836	/* cpu_to_le32 and assign to struct fields */
1837	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1838	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1839	context_desc->rsvd = cpu_to_le16(0);
1840	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1841}
1842
1843/**
1844 * __i40evf_chk_linearize - Check if there are more than 8 buffers per packet
1845 * @skb:      send buffer
1846 *
1847 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
1848 * and so we need to figure out the cases where we need to linearize the skb.
1849 *
1850 * For TSO we need to count the TSO header and segment payload separately.
1851 * As such we need to check cases where we have 7 fragments or more as we
1852 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
1853 * the segment payload in the first descriptor, and another 7 for the
1854 * fragments.
1855 **/
1856bool __i40evf_chk_linearize(struct sk_buff *skb)
1857{
1858	const struct skb_frag_struct *frag, *stale;
1859	int nr_frags, sum;
1860
1861	/* no need to check if number of frags is less than 7 */
1862	nr_frags = skb_shinfo(skb)->nr_frags;
1863	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
1864		return false;
1865
1866	/* We need to walk through the list and validate that each group
1867	 * of 6 fragments totals at least gso_size.
1868	 */
1869	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
1870	frag = &skb_shinfo(skb)->frags[0];
1871
1872	/* Initialize size to the negative value of gso_size minus 1.  We
1873	 * use this as the worst case scenerio in which the frag ahead
1874	 * of us only provides one byte which is why we are limited to 6
1875	 * descriptors for a single transmit as the header and previous
1876	 * fragment are already consuming 2 descriptors.
1877	 */
1878	sum = 1 - skb_shinfo(skb)->gso_size;
1879
1880	/* Add size of frags 0 through 4 to create our initial sum */
1881	sum += skb_frag_size(frag++);
1882	sum += skb_frag_size(frag++);
1883	sum += skb_frag_size(frag++);
1884	sum += skb_frag_size(frag++);
1885	sum += skb_frag_size(frag++);
1886
1887	/* Walk through fragments adding latest fragment, testing it, and
1888	 * then removing stale fragments from the sum.
1889	 */
1890	stale = &skb_shinfo(skb)->frags[0];
1891	for (;;) {
1892		sum += skb_frag_size(frag++);
1893
1894		/* if sum is negative we failed to make sufficient progress */
1895		if (sum < 0)
1896			return true;
1897
1898		if (!nr_frags--)
1899			break;
1900
1901		sum -= skb_frag_size(stale++);
1902	}
1903
1904	return false;
1905}
1906
1907/**
1908 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1909 * @tx_ring: the ring to be checked
1910 * @size:    the size buffer we want to assure is available
1911 *
1912 * Returns -EBUSY if a stop is needed, else 0
1913 **/
1914int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1915{
1916	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1917	/* Memory barrier before checking head and tail */
1918	smp_mb();
1919
1920	/* Check again in a case another CPU has just made room available. */
1921	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1922		return -EBUSY;
1923
1924	/* A reprieve! - use start_queue because it doesn't call schedule */
1925	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1926	++tx_ring->tx_stats.restart_queue;
1927	return 0;
1928}
1929
1930/**
1931 * i40evf_tx_map - Build the Tx descriptor
1932 * @tx_ring:  ring to send buffer on
1933 * @skb:      send buffer
1934 * @first:    first buffer info buffer to use
1935 * @tx_flags: collected send information
1936 * @hdr_len:  size of the packet header
1937 * @td_cmd:   the command field in the descriptor
1938 * @td_offset: offset for checksum or crc
1939 **/
1940static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1941				 struct i40e_tx_buffer *first, u32 tx_flags,
1942				 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1943{
1944	unsigned int data_len = skb->data_len;
1945	unsigned int size = skb_headlen(skb);
1946	struct skb_frag_struct *frag;
1947	struct i40e_tx_buffer *tx_bi;
1948	struct i40e_tx_desc *tx_desc;
1949	u16 i = tx_ring->next_to_use;
1950	u32 td_tag = 0;
1951	dma_addr_t dma;
1952	u16 gso_segs;
1953	u16 desc_count = 1;
1954
1955	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1956		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1957		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1958			 I40E_TX_FLAGS_VLAN_SHIFT;
1959	}
1960
1961	if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1962		gso_segs = skb_shinfo(skb)->gso_segs;
1963	else
1964		gso_segs = 1;
1965
1966	/* multiply data chunks by size of headers */
1967	first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1968	first->gso_segs = gso_segs;
1969	first->skb = skb;
1970	first->tx_flags = tx_flags;
1971
1972	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1973
1974	tx_desc = I40E_TX_DESC(tx_ring, i);
1975	tx_bi = first;
1976
1977	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1978		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
1979
1980		if (dma_mapping_error(tx_ring->dev, dma))
1981			goto dma_error;
1982
1983		/* record length, and DMA address */
1984		dma_unmap_len_set(tx_bi, len, size);
1985		dma_unmap_addr_set(tx_bi, dma, dma);
1986
1987		/* align size to end of page */
1988		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
1989		tx_desc->buffer_addr = cpu_to_le64(dma);
1990
1991		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1992			tx_desc->cmd_type_offset_bsz =
1993				build_ctob(td_cmd, td_offset,
1994					   max_data, td_tag);
1995
1996			tx_desc++;
1997			i++;
1998			desc_count++;
1999
2000			if (i == tx_ring->count) {
2001				tx_desc = I40E_TX_DESC(tx_ring, 0);
2002				i = 0;
2003			}
2004
2005			dma += max_data;
2006			size -= max_data;
2007
2008			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2009			tx_desc->buffer_addr = cpu_to_le64(dma);
2010		}
2011
2012		if (likely(!data_len))
2013			break;
2014
2015		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2016							  size, td_tag);
2017
2018		tx_desc++;
2019		i++;
2020		desc_count++;
2021
2022		if (i == tx_ring->count) {
2023			tx_desc = I40E_TX_DESC(tx_ring, 0);
2024			i = 0;
2025		}
2026
2027		size = skb_frag_size(frag);
2028		data_len -= size;
2029
2030		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2031				       DMA_TO_DEVICE);
2032
2033		tx_bi = &tx_ring->tx_bi[i];
2034	}
2035
2036	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
2037
2038	i++;
2039	if (i == tx_ring->count)
2040		i = 0;
2041
2042	tx_ring->next_to_use = i;
2043
2044	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2045
2046	/* write last descriptor with EOP bit */
2047	td_cmd |= I40E_TX_DESC_CMD_EOP;
2048
2049	/* We can OR these values together as they both are checked against
2050	 * 4 below and at this point desc_count will be used as a boolean value
2051	 * after this if/else block.
2052	 */
2053	desc_count |= ++tx_ring->packet_stride;
2054
2055	/* Algorithm to optimize tail and RS bit setting:
2056	 * if queue is stopped
2057	 *	mark RS bit
2058	 *	reset packet counter
2059	 * else if xmit_more is supported and is true
2060	 *	advance packet counter to 4
2061	 *	reset desc_count to 0
2062	 *
2063	 * if desc_count >= 4
2064	 *	mark RS bit
2065	 *	reset packet counter
2066	 * if desc_count > 0
2067	 *	update tail
2068	 *
2069	 * Note: If there are less than 4 descriptors
2070	 * pending and interrupts were disabled the service task will
2071	 * trigger a force WB.
2072	 */
2073	if (netif_xmit_stopped(txring_txq(tx_ring))) {
2074		goto do_rs;
2075	} else if (skb->xmit_more) {
2076		/* set stride to arm on next packet and reset desc_count */
2077		tx_ring->packet_stride = WB_STRIDE;
2078		desc_count = 0;
2079	} else if (desc_count >= WB_STRIDE) {
2080do_rs:
2081		/* write last descriptor with RS bit set */
2082		td_cmd |= I40E_TX_DESC_CMD_RS;
2083		tx_ring->packet_stride = 0;
2084	}
2085
2086	tx_desc->cmd_type_offset_bsz =
2087			build_ctob(td_cmd, td_offset, size, td_tag);
2088
2089	/* Force memory writes to complete before letting h/w know there
2090	 * are new descriptors to fetch.
2091	 *
2092	 * We also use this memory barrier to make certain all of the
2093	 * status bits have been updated before next_to_watch is written.
2094	 */
2095	wmb();
2096
2097	/* set next_to_watch value indicating a packet is present */
2098	first->next_to_watch = tx_desc;
2099
2100	/* notify HW of packet */
2101	if (desc_count) {
2102		writel(i, tx_ring->tail);
2103
2104		/* we need this if more than one processor can write to our tail
2105		 * at a time, it synchronizes IO on IA64/Altix systems
2106		 */
2107		mmiowb();
2108	}
2109
2110	return;
2111
2112dma_error:
2113	dev_info(tx_ring->dev, "TX DMA map failed\n");
2114
2115	/* clear dma mappings for failed tx_bi map */
2116	for (;;) {
2117		tx_bi = &tx_ring->tx_bi[i];
2118		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2119		if (tx_bi == first)
2120			break;
2121		if (i == 0)
2122			i = tx_ring->count;
2123		i--;
2124	}
2125
2126	tx_ring->next_to_use = i;
2127}
2128
2129/**
2130 * i40e_xmit_frame_ring - Sends buffer on Tx ring
2131 * @skb:     send buffer
2132 * @tx_ring: ring to send buffer on
2133 *
2134 * Returns NETDEV_TX_OK if sent, else an error code
2135 **/
2136static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2137					struct i40e_ring *tx_ring)
2138{
2139	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2140	u32 cd_tunneling = 0, cd_l2tag2 = 0;
2141	struct i40e_tx_buffer *first;
2142	u32 td_offset = 0;
2143	u32 tx_flags = 0;
2144	__be16 protocol;
2145	u32 td_cmd = 0;
2146	u8 hdr_len = 0;
2147	int tso, count;
2148
2149	/* prefetch the data, we'll need it later */
2150	prefetch(skb->data);
2151
2152	count = i40e_xmit_descriptor_count(skb);
2153	if (i40e_chk_linearize(skb, count)) {
2154		if (__skb_linearize(skb))
2155			goto out_drop;
2156		count = i40e_txd_use_count(skb->len);
2157		tx_ring->tx_stats.tx_linearize++;
2158	}
2159
2160	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2161	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2162	 *       + 4 desc gap to avoid the cache line where head is,
2163	 *       + 1 desc for context descriptor,
2164	 * otherwise try next time
2165	 */
2166	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2167		tx_ring->tx_stats.tx_busy++;
2168		return NETDEV_TX_BUSY;
2169	}
2170
2171	/* prepare the xmit flags */
2172	if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2173		goto out_drop;
2174
2175	/* obtain protocol of skb */
2176	protocol = vlan_get_protocol(skb);
2177
2178	/* record the location of the first descriptor for this packet */
2179	first = &tx_ring->tx_bi[tx_ring->next_to_use];
2180
2181	/* setup IPv4/IPv6 offloads */
2182	if (protocol == htons(ETH_P_IP))
2183		tx_flags |= I40E_TX_FLAGS_IPV4;
2184	else if (protocol == htons(ETH_P_IPV6))
2185		tx_flags |= I40E_TX_FLAGS_IPV6;
2186
2187	tso = i40e_tso(skb, &hdr_len, &cd_type_cmd_tso_mss);
2188
2189	if (tso < 0)
2190		goto out_drop;
2191	else if (tso)
2192		tx_flags |= I40E_TX_FLAGS_TSO;
2193
2194	/* Always offload the checksum, since it's in the data descriptor */
2195	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2196				  tx_ring, &cd_tunneling);
2197	if (tso < 0)
2198		goto out_drop;
2199
2200	skb_tx_timestamp(skb);
2201
2202	/* always enable CRC insertion offload */
2203	td_cmd |= I40E_TX_DESC_CMD_ICRC;
2204
2205	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2206			   cd_tunneling, cd_l2tag2);
2207
2208	i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2209		      td_cmd, td_offset);
2210
2211	return NETDEV_TX_OK;
2212
2213out_drop:
2214	dev_kfree_skb_any(skb);
2215	return NETDEV_TX_OK;
2216}
2217
2218/**
2219 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2220 * @skb:    send buffer
2221 * @netdev: network interface device structure
2222 *
2223 * Returns NETDEV_TX_OK if sent, else an error code
2224 **/
2225netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2226{
2227	struct i40evf_adapter *adapter = netdev_priv(netdev);
2228	struct i40e_ring *tx_ring = &adapter->tx_rings[skb->queue_mapping];
2229
2230	/* hardware can't handle really short frames, hardware padding works
2231	 * beyond this point
2232	 */
2233	if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2234		if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2235			return NETDEV_TX_OK;
2236		skb->len = I40E_MIN_TX_LEN;
2237		skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2238	}
2239
2240	return i40e_xmit_frame_ring(skb, tx_ring);
2241}