Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.5.6.
   1// SPDX-License-Identifier: GPL-2.0
   2/* Copyright(c) 2013 - 2018 Intel Corporation. */
   3
   4#include <linux/prefetch.h>
   5#include <linux/bpf_trace.h>
   6#include <net/xdp.h>
   7#include "i40e.h"
   8#include "i40e_trace.h"
   9#include "i40e_prototype.h"
  10#include "i40e_txrx_common.h"
  11#include "i40e_xsk.h"
  12
  13#define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
  14/**
  15 * i40e_fdir - Generate a Flow Director descriptor based on fdata
  16 * @tx_ring: Tx ring to send buffer on
  17 * @fdata: Flow director filter data
  18 * @add: Indicate if we are adding a rule or deleting one
  19 *
  20 **/
  21static void i40e_fdir(struct i40e_ring *tx_ring,
  22		      struct i40e_fdir_filter *fdata, bool add)
  23{
  24	struct i40e_filter_program_desc *fdir_desc;
  25	struct i40e_pf *pf = tx_ring->vsi->back;
  26	u32 flex_ptype, dtype_cmd;
  27	u16 i;
  28
  29	/* grab the next descriptor */
  30	i = tx_ring->next_to_use;
  31	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
  32
  33	i++;
  34	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
  35
  36	flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
  37		     (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
  38
  39	flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
  40		      (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
  41
  42	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
  43		      (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
  44
  45	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
  46		      (fdata->flex_offset << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
  47
  48	/* Use LAN VSI Id if not programmed by user */
  49	flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
  50		      ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
  51		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
  52
  53	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
  54
  55	dtype_cmd |= add ?
  56		     I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
  57		     I40E_TXD_FLTR_QW1_PCMD_SHIFT :
  58		     I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
  59		     I40E_TXD_FLTR_QW1_PCMD_SHIFT;
  60
  61	dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
  62		     (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
  63
  64	dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
  65		     (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
  66
  67	if (fdata->cnt_index) {
  68		dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
  69		dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
  70			     ((u32)fdata->cnt_index <<
  71			      I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
  72	}
  73
  74	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
  75	fdir_desc->rsvd = cpu_to_le32(0);
  76	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
  77	fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
  78}
  79
  80#define I40E_FD_CLEAN_DELAY 10
  81/**
  82 * i40e_program_fdir_filter - Program a Flow Director filter
  83 * @fdir_data: Packet data that will be filter parameters
  84 * @raw_packet: the pre-allocated packet buffer for FDir
  85 * @pf: The PF pointer
  86 * @add: True for add/update, False for remove
  87 **/
  88static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
  89				    u8 *raw_packet, struct i40e_pf *pf,
  90				    bool add)
  91{
  92	struct i40e_tx_buffer *tx_buf, *first;
  93	struct i40e_tx_desc *tx_desc;
  94	struct i40e_ring *tx_ring;
  95	struct i40e_vsi *vsi;
  96	struct device *dev;
  97	dma_addr_t dma;
  98	u32 td_cmd = 0;
  99	u16 i;
 100
 101	/* find existing FDIR VSI */
 102	vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
 103	if (!vsi)
 104		return -ENOENT;
 105
 106	tx_ring = vsi->tx_rings[0];
 107	dev = tx_ring->dev;
 108
 109	/* we need two descriptors to add/del a filter and we can wait */
 110	for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
 111		if (!i)
 112			return -EAGAIN;
 113		msleep_interruptible(1);
 114	}
 115
 116	dma = dma_map_single(dev, raw_packet,
 117			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
 118	if (dma_mapping_error(dev, dma))
 119		goto dma_fail;
 120
 121	/* grab the next descriptor */
 122	i = tx_ring->next_to_use;
 123	first = &tx_ring->tx_bi[i];
 124	i40e_fdir(tx_ring, fdir_data, add);
 125
 126	/* Now program a dummy descriptor */
 127	i = tx_ring->next_to_use;
 128	tx_desc = I40E_TX_DESC(tx_ring, i);
 129	tx_buf = &tx_ring->tx_bi[i];
 130
 131	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
 132
 133	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
 134
 135	/* record length, and DMA address */
 136	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
 137	dma_unmap_addr_set(tx_buf, dma, dma);
 138
 139	tx_desc->buffer_addr = cpu_to_le64(dma);
 140	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
 141
 142	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
 143	tx_buf->raw_buf = (void *)raw_packet;
 144
 145	tx_desc->cmd_type_offset_bsz =
 146		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
 147
 148	/* Force memory writes to complete before letting h/w
 149	 * know there are new descriptors to fetch.
 150	 */
 151	wmb();
 152
 153	/* Mark the data descriptor to be watched */
 154	first->next_to_watch = tx_desc;
 155
 156	writel(tx_ring->next_to_use, tx_ring->tail);
 157	return 0;
 158
 159dma_fail:
 160	return -1;
 161}
 162
 163#define IP_HEADER_OFFSET 14
 164#define I40E_UDPIP_DUMMY_PACKET_LEN 42
 165/**
 166 * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
 167 * @vsi: pointer to the targeted VSI
 168 * @fd_data: the flow director data required for the FDir descriptor
 169 * @add: true adds a filter, false removes it
 170 *
 171 * Returns 0 if the filters were successfully added or removed
 172 **/
 173static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
 174				   struct i40e_fdir_filter *fd_data,
 175				   bool add)
 176{
 177	struct i40e_pf *pf = vsi->back;
 178	struct udphdr *udp;
 179	struct iphdr *ip;
 180	u8 *raw_packet;
 181	int ret;
 182	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
 183		0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
 184		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
 185
 186	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 187	if (!raw_packet)
 188		return -ENOMEM;
 189	memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
 190
 191	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
 192	udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
 193	      + sizeof(struct iphdr));
 194
 195	ip->daddr = fd_data->dst_ip;
 196	udp->dest = fd_data->dst_port;
 197	ip->saddr = fd_data->src_ip;
 198	udp->source = fd_data->src_port;
 199
 200	if (fd_data->flex_filter) {
 201		u8 *payload = raw_packet + I40E_UDPIP_DUMMY_PACKET_LEN;
 202		__be16 pattern = fd_data->flex_word;
 203		u16 off = fd_data->flex_offset;
 204
 205		*((__force __be16 *)(payload + off)) = pattern;
 206	}
 207
 208	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
 209	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
 210	if (ret) {
 211		dev_info(&pf->pdev->dev,
 212			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 213			 fd_data->pctype, fd_data->fd_id, ret);
 214		/* Free the packet buffer since it wasn't added to the ring */
 215		kfree(raw_packet);
 216		return -EOPNOTSUPP;
 217	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 218		if (add)
 219			dev_info(&pf->pdev->dev,
 220				 "Filter OK for PCTYPE %d loc = %d\n",
 221				 fd_data->pctype, fd_data->fd_id);
 222		else
 223			dev_info(&pf->pdev->dev,
 224				 "Filter deleted for PCTYPE %d loc = %d\n",
 225				 fd_data->pctype, fd_data->fd_id);
 226	}
 227
 228	if (add)
 229		pf->fd_udp4_filter_cnt++;
 230	else
 231		pf->fd_udp4_filter_cnt--;
 232
 233	return 0;
 234}
 235
 236#define I40E_TCPIP_DUMMY_PACKET_LEN 54
 237/**
 238 * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
 239 * @vsi: pointer to the targeted VSI
 240 * @fd_data: the flow director data required for the FDir descriptor
 241 * @add: true adds a filter, false removes it
 242 *
 243 * Returns 0 if the filters were successfully added or removed
 244 **/
 245static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
 246				   struct i40e_fdir_filter *fd_data,
 247				   bool add)
 248{
 249	struct i40e_pf *pf = vsi->back;
 250	struct tcphdr *tcp;
 251	struct iphdr *ip;
 252	u8 *raw_packet;
 253	int ret;
 254	/* Dummy packet */
 255	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
 256		0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
 257		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
 258		0x0, 0x72, 0, 0, 0, 0};
 259
 260	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 261	if (!raw_packet)
 262		return -ENOMEM;
 263	memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
 264
 265	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
 266	tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
 267	      + sizeof(struct iphdr));
 268
 269	ip->daddr = fd_data->dst_ip;
 270	tcp->dest = fd_data->dst_port;
 271	ip->saddr = fd_data->src_ip;
 272	tcp->source = fd_data->src_port;
 273
 274	if (fd_data->flex_filter) {
 275		u8 *payload = raw_packet + I40E_TCPIP_DUMMY_PACKET_LEN;
 276		__be16 pattern = fd_data->flex_word;
 277		u16 off = fd_data->flex_offset;
 278
 279		*((__force __be16 *)(payload + off)) = pattern;
 280	}
 281
 282	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
 283	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
 284	if (ret) {
 285		dev_info(&pf->pdev->dev,
 286			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 287			 fd_data->pctype, fd_data->fd_id, ret);
 288		/* Free the packet buffer since it wasn't added to the ring */
 289		kfree(raw_packet);
 290		return -EOPNOTSUPP;
 291	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 292		if (add)
 293			dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
 294				 fd_data->pctype, fd_data->fd_id);
 295		else
 296			dev_info(&pf->pdev->dev,
 297				 "Filter deleted for PCTYPE %d loc = %d\n",
 298				 fd_data->pctype, fd_data->fd_id);
 299	}
 300
 301	if (add) {
 302		pf->fd_tcp4_filter_cnt++;
 303		if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
 304		    I40E_DEBUG_FD & pf->hw.debug_mask)
 305			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
 306		set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 307	} else {
 308		pf->fd_tcp4_filter_cnt--;
 309	}
 310
 311	return 0;
 312}
 313
 314#define I40E_SCTPIP_DUMMY_PACKET_LEN 46
 315/**
 316 * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
 317 * a specific flow spec
 318 * @vsi: pointer to the targeted VSI
 319 * @fd_data: the flow director data required for the FDir descriptor
 320 * @add: true adds a filter, false removes it
 321 *
 322 * Returns 0 if the filters were successfully added or removed
 323 **/
 324static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
 325				    struct i40e_fdir_filter *fd_data,
 326				    bool add)
 327{
 328	struct i40e_pf *pf = vsi->back;
 329	struct sctphdr *sctp;
 330	struct iphdr *ip;
 331	u8 *raw_packet;
 332	int ret;
 333	/* Dummy packet */
 334	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
 335		0x45, 0, 0, 0x20, 0, 0, 0x40, 0, 0x40, 0x84, 0, 0, 0, 0, 0, 0,
 336		0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
 337
 338	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 339	if (!raw_packet)
 340		return -ENOMEM;
 341	memcpy(raw_packet, packet, I40E_SCTPIP_DUMMY_PACKET_LEN);
 342
 343	ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
 344	sctp = (struct sctphdr *)(raw_packet + IP_HEADER_OFFSET
 345	      + sizeof(struct iphdr));
 346
 347	ip->daddr = fd_data->dst_ip;
 348	sctp->dest = fd_data->dst_port;
 349	ip->saddr = fd_data->src_ip;
 350	sctp->source = fd_data->src_port;
 351
 352	if (fd_data->flex_filter) {
 353		u8 *payload = raw_packet + I40E_SCTPIP_DUMMY_PACKET_LEN;
 354		__be16 pattern = fd_data->flex_word;
 355		u16 off = fd_data->flex_offset;
 356
 357		*((__force __be16 *)(payload + off)) = pattern;
 358	}
 359
 360	fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_SCTP;
 361	ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
 362	if (ret) {
 363		dev_info(&pf->pdev->dev,
 364			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 365			 fd_data->pctype, fd_data->fd_id, ret);
 366		/* Free the packet buffer since it wasn't added to the ring */
 367		kfree(raw_packet);
 368		return -EOPNOTSUPP;
 369	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 370		if (add)
 371			dev_info(&pf->pdev->dev,
 372				 "Filter OK for PCTYPE %d loc = %d\n",
 373				 fd_data->pctype, fd_data->fd_id);
 374		else
 375			dev_info(&pf->pdev->dev,
 376				 "Filter deleted for PCTYPE %d loc = %d\n",
 377				 fd_data->pctype, fd_data->fd_id);
 378	}
 379
 380	if (add)
 381		pf->fd_sctp4_filter_cnt++;
 382	else
 383		pf->fd_sctp4_filter_cnt--;
 384
 385	return 0;
 386}
 387
 388#define I40E_IP_DUMMY_PACKET_LEN 34
 389/**
 390 * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
 391 * a specific flow spec
 392 * @vsi: pointer to the targeted VSI
 393 * @fd_data: the flow director data required for the FDir descriptor
 394 * @add: true adds a filter, false removes it
 395 *
 396 * Returns 0 if the filters were successfully added or removed
 397 **/
 398static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
 399				  struct i40e_fdir_filter *fd_data,
 400				  bool add)
 401{
 402	struct i40e_pf *pf = vsi->back;
 403	struct iphdr *ip;
 404	u8 *raw_packet;
 405	int ret;
 406	int i;
 407	static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
 408		0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
 409		0, 0, 0, 0};
 410
 411	for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
 412	     i <= I40E_FILTER_PCTYPE_FRAG_IPV4;	i++) {
 413		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 414		if (!raw_packet)
 415			return -ENOMEM;
 416		memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
 417		ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
 418
 419		ip->saddr = fd_data->src_ip;
 420		ip->daddr = fd_data->dst_ip;
 421		ip->protocol = 0;
 422
 423		if (fd_data->flex_filter) {
 424			u8 *payload = raw_packet + I40E_IP_DUMMY_PACKET_LEN;
 425			__be16 pattern = fd_data->flex_word;
 426			u16 off = fd_data->flex_offset;
 427
 428			*((__force __be16 *)(payload + off)) = pattern;
 429		}
 430
 431		fd_data->pctype = i;
 432		ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
 433		if (ret) {
 434			dev_info(&pf->pdev->dev,
 435				 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 436				 fd_data->pctype, fd_data->fd_id, ret);
 437			/* The packet buffer wasn't added to the ring so we
 438			 * need to free it now.
 439			 */
 440			kfree(raw_packet);
 441			return -EOPNOTSUPP;
 442		} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 443			if (add)
 444				dev_info(&pf->pdev->dev,
 445					 "Filter OK for PCTYPE %d loc = %d\n",
 446					 fd_data->pctype, fd_data->fd_id);
 447			else
 448				dev_info(&pf->pdev->dev,
 449					 "Filter deleted for PCTYPE %d loc = %d\n",
 450					 fd_data->pctype, fd_data->fd_id);
 451		}
 452	}
 453
 454	if (add)
 455		pf->fd_ip4_filter_cnt++;
 456	else
 457		pf->fd_ip4_filter_cnt--;
 458
 459	return 0;
 460}
 461
 462/**
 463 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
 464 * @vsi: pointer to the targeted VSI
 465 * @input: filter to add or delete
 466 * @add: true adds a filter, false removes it
 467 *
 468 **/
 469int i40e_add_del_fdir(struct i40e_vsi *vsi,
 470		      struct i40e_fdir_filter *input, bool add)
 471{
 472	struct i40e_pf *pf = vsi->back;
 473	int ret;
 474
 475	switch (input->flow_type & ~FLOW_EXT) {
 476	case TCP_V4_FLOW:
 477		ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
 478		break;
 479	case UDP_V4_FLOW:
 480		ret = i40e_add_del_fdir_udpv4(vsi, input, add);
 481		break;
 482	case SCTP_V4_FLOW:
 483		ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
 484		break;
 485	case IP_USER_FLOW:
 486		switch (input->ip4_proto) {
 487		case IPPROTO_TCP:
 488			ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
 489			break;
 490		case IPPROTO_UDP:
 491			ret = i40e_add_del_fdir_udpv4(vsi, input, add);
 492			break;
 493		case IPPROTO_SCTP:
 494			ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
 495			break;
 496		case IPPROTO_IP:
 497			ret = i40e_add_del_fdir_ipv4(vsi, input, add);
 498			break;
 499		default:
 500			/* We cannot support masking based on protocol */
 501			dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
 502				 input->ip4_proto);
 503			return -EINVAL;
 504		}
 505		break;
 506	default:
 507		dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
 508			 input->flow_type);
 509		return -EINVAL;
 510	}
 511
 512	/* The buffer allocated here will be normally be freed by
 513	 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
 514	 * completion. In the event of an error adding the buffer to the FDIR
 515	 * ring, it will immediately be freed. It may also be freed by
 516	 * i40e_clean_tx_ring() when closing the VSI.
 517	 */
 518	return ret;
 519}
 520
 521/**
 522 * i40e_fd_handle_status - check the Programming Status for FD
 523 * @rx_ring: the Rx ring for this descriptor
 524 * @qword0_raw: qword0
 525 * @qword1: qword1 after le_to_cpu
 526 * @prog_id: the id originally used for programming
 527 *
 528 * This is used to verify if the FD programming or invalidation
 529 * requested by SW to the HW is successful or not and take actions accordingly.
 530 **/
 531static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u64 qword0_raw,
 532				  u64 qword1, u8 prog_id)
 533{
 534	struct i40e_pf *pf = rx_ring->vsi->back;
 535	struct pci_dev *pdev = pf->pdev;
 536	struct i40e_32b_rx_wb_qw0 *qw0;
 537	u32 fcnt_prog, fcnt_avail;
 538	u32 error;
 539
 540	qw0 = (struct i40e_32b_rx_wb_qw0 *)&qword0_raw;
 541	error = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
 542		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
 543
 544	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
 545		pf->fd_inv = le32_to_cpu(qw0->hi_dword.fd_id);
 546		if (qw0->hi_dword.fd_id != 0 ||
 547		    (I40E_DEBUG_FD & pf->hw.debug_mask))
 548			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
 549				 pf->fd_inv);
 550
 551		/* Check if the programming error is for ATR.
 552		 * If so, auto disable ATR and set a state for
 553		 * flush in progress. Next time we come here if flush is in
 554		 * progress do nothing, once flush is complete the state will
 555		 * be cleared.
 556		 */
 557		if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
 558			return;
 559
 560		pf->fd_add_err++;
 561		/* store the current atr filter count */
 562		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
 563
 564		if (qw0->hi_dword.fd_id == 0 &&
 565		    test_bit(__I40E_FD_SB_AUTO_DISABLED, pf->state)) {
 566			/* These set_bit() calls aren't atomic with the
 567			 * test_bit() here, but worse case we potentially
 568			 * disable ATR and queue a flush right after SB
 569			 * support is re-enabled. That shouldn't cause an
 570			 * issue in practice
 571			 */
 572			set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 573			set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
 574		}
 575
 576		/* filter programming failed most likely due to table full */
 577		fcnt_prog = i40e_get_global_fd_count(pf);
 578		fcnt_avail = pf->fdir_pf_filter_count;
 579		/* If ATR is running fcnt_prog can quickly change,
 580		 * if we are very close to full, it makes sense to disable
 581		 * FD ATR/SB and then re-enable it when there is room.
 582		 */
 583		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
 584			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
 585			    !test_and_set_bit(__I40E_FD_SB_AUTO_DISABLED,
 586					      pf->state))
 587				if (I40E_DEBUG_FD & pf->hw.debug_mask)
 588					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
 589		}
 590	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
 591		if (I40E_DEBUG_FD & pf->hw.debug_mask)
 592			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
 593				 qw0->hi_dword.fd_id);
 594	}
 595}
 596
 597/**
 598 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
 599 * @ring:      the ring that owns the buffer
 600 * @tx_buffer: the buffer to free
 601 **/
 602static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
 603					    struct i40e_tx_buffer *tx_buffer)
 604{
 605	if (tx_buffer->skb) {
 606		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
 607			kfree(tx_buffer->raw_buf);
 608		else if (ring_is_xdp(ring))
 609			xdp_return_frame(tx_buffer->xdpf);
 610		else
 611			dev_kfree_skb_any(tx_buffer->skb);
 612		if (dma_unmap_len(tx_buffer, len))
 613			dma_unmap_single(ring->dev,
 614					 dma_unmap_addr(tx_buffer, dma),
 615					 dma_unmap_len(tx_buffer, len),
 616					 DMA_TO_DEVICE);
 617	} else if (dma_unmap_len(tx_buffer, len)) {
 618		dma_unmap_page(ring->dev,
 619			       dma_unmap_addr(tx_buffer, dma),
 620			       dma_unmap_len(tx_buffer, len),
 621			       DMA_TO_DEVICE);
 622	}
 623
 624	tx_buffer->next_to_watch = NULL;
 625	tx_buffer->skb = NULL;
 626	dma_unmap_len_set(tx_buffer, len, 0);
 627	/* tx_buffer must be completely set up in the transmit path */
 628}
 629
 630/**
 631 * i40e_clean_tx_ring - Free any empty Tx buffers
 632 * @tx_ring: ring to be cleaned
 633 **/
 634void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
 635{
 636	unsigned long bi_size;
 637	u16 i;
 638
 639	if (ring_is_xdp(tx_ring) && tx_ring->xsk_umem) {
 640		i40e_xsk_clean_tx_ring(tx_ring);
 641	} else {
 642		/* ring already cleared, nothing to do */
 643		if (!tx_ring->tx_bi)
 644			return;
 645
 646		/* Free all the Tx ring sk_buffs */
 647		for (i = 0; i < tx_ring->count; i++)
 648			i40e_unmap_and_free_tx_resource(tx_ring,
 649							&tx_ring->tx_bi[i]);
 650	}
 651
 652	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
 653	memset(tx_ring->tx_bi, 0, bi_size);
 654
 655	/* Zero out the descriptor ring */
 656	memset(tx_ring->desc, 0, tx_ring->size);
 657
 658	tx_ring->next_to_use = 0;
 659	tx_ring->next_to_clean = 0;
 660
 661	if (!tx_ring->netdev)
 662		return;
 663
 664	/* cleanup Tx queue statistics */
 665	netdev_tx_reset_queue(txring_txq(tx_ring));
 666}
 667
 668/**
 669 * i40e_free_tx_resources - Free Tx resources per queue
 670 * @tx_ring: Tx descriptor ring for a specific queue
 671 *
 672 * Free all transmit software resources
 673 **/
 674void i40e_free_tx_resources(struct i40e_ring *tx_ring)
 675{
 676	i40e_clean_tx_ring(tx_ring);
 677	kfree(tx_ring->tx_bi);
 678	tx_ring->tx_bi = NULL;
 679
 680	if (tx_ring->desc) {
 681		dma_free_coherent(tx_ring->dev, tx_ring->size,
 682				  tx_ring->desc, tx_ring->dma);
 683		tx_ring->desc = NULL;
 684	}
 685}
 686
 687/**
 688 * i40e_get_tx_pending - how many tx descriptors not processed
 689 * @ring: the ring of descriptors
 690 * @in_sw: use SW variables
 691 *
 692 * Since there is no access to the ring head register
 693 * in XL710, we need to use our local copies
 694 **/
 695u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
 696{
 697	u32 head, tail;
 698
 699	if (!in_sw) {
 700		head = i40e_get_head(ring);
 701		tail = readl(ring->tail);
 702	} else {
 703		head = ring->next_to_clean;
 704		tail = ring->next_to_use;
 705	}
 706
 707	if (head != tail)
 708		return (head < tail) ?
 709			tail - head : (tail + ring->count - head);
 710
 711	return 0;
 712}
 713
 714/**
 715 * i40e_detect_recover_hung - Function to detect and recover hung_queues
 716 * @vsi:  pointer to vsi struct with tx queues
 717 *
 718 * VSI has netdev and netdev has TX queues. This function is to check each of
 719 * those TX queues if they are hung, trigger recovery by issuing SW interrupt.
 720 **/
 721void i40e_detect_recover_hung(struct i40e_vsi *vsi)
 722{
 723	struct i40e_ring *tx_ring = NULL;
 724	struct net_device *netdev;
 725	unsigned int i;
 726	int packets;
 727
 728	if (!vsi)
 729		return;
 730
 731	if (test_bit(__I40E_VSI_DOWN, vsi->state))
 732		return;
 733
 734	netdev = vsi->netdev;
 735	if (!netdev)
 736		return;
 737
 738	if (!netif_carrier_ok(netdev))
 739		return;
 740
 741	for (i = 0; i < vsi->num_queue_pairs; i++) {
 742		tx_ring = vsi->tx_rings[i];
 743		if (tx_ring && tx_ring->desc) {
 744			/* If packet counter has not changed the queue is
 745			 * likely stalled, so force an interrupt for this
 746			 * queue.
 747			 *
 748			 * prev_pkt_ctr would be negative if there was no
 749			 * pending work.
 750			 */
 751			packets = tx_ring->stats.packets & INT_MAX;
 752			if (tx_ring->tx_stats.prev_pkt_ctr == packets) {
 753				i40e_force_wb(vsi, tx_ring->q_vector);
 754				continue;
 755			}
 756
 757			/* Memory barrier between read of packet count and call
 758			 * to i40e_get_tx_pending()
 759			 */
 760			smp_rmb();
 761			tx_ring->tx_stats.prev_pkt_ctr =
 762			    i40e_get_tx_pending(tx_ring, true) ? packets : -1;
 763		}
 764	}
 765}
 766
 767/**
 768 * i40e_clean_tx_irq - Reclaim resources after transmit completes
 769 * @vsi: the VSI we care about
 770 * @tx_ring: Tx ring to clean
 771 * @napi_budget: Used to determine if we are in netpoll
 772 *
 773 * Returns true if there's any budget left (e.g. the clean is finished)
 774 **/
 775static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
 776			      struct i40e_ring *tx_ring, int napi_budget)
 777{
 778	int i = tx_ring->next_to_clean;
 779	struct i40e_tx_buffer *tx_buf;
 780	struct i40e_tx_desc *tx_head;
 781	struct i40e_tx_desc *tx_desc;
 782	unsigned int total_bytes = 0, total_packets = 0;
 783	unsigned int budget = vsi->work_limit;
 784
 785	tx_buf = &tx_ring->tx_bi[i];
 786	tx_desc = I40E_TX_DESC(tx_ring, i);
 787	i -= tx_ring->count;
 788
 789	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
 790
 791	do {
 792		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 793
 794		/* if next_to_watch is not set then there is no work pending */
 795		if (!eop_desc)
 796			break;
 797
 798		/* prevent any other reads prior to eop_desc */
 799		smp_rmb();
 800
 801		i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
 802		/* we have caught up to head, no work left to do */
 803		if (tx_head == tx_desc)
 804			break;
 805
 806		/* clear next_to_watch to prevent false hangs */
 807		tx_buf->next_to_watch = NULL;
 808
 809		/* update the statistics for this packet */
 810		total_bytes += tx_buf->bytecount;
 811		total_packets += tx_buf->gso_segs;
 812
 813		/* free the skb/XDP data */
 814		if (ring_is_xdp(tx_ring))
 815			xdp_return_frame(tx_buf->xdpf);
 816		else
 817			napi_consume_skb(tx_buf->skb, napi_budget);
 818
 819		/* unmap skb header data */
 820		dma_unmap_single(tx_ring->dev,
 821				 dma_unmap_addr(tx_buf, dma),
 822				 dma_unmap_len(tx_buf, len),
 823				 DMA_TO_DEVICE);
 824
 825		/* clear tx_buffer data */
 826		tx_buf->skb = NULL;
 827		dma_unmap_len_set(tx_buf, len, 0);
 828
 829		/* unmap remaining buffers */
 830		while (tx_desc != eop_desc) {
 831			i40e_trace(clean_tx_irq_unmap,
 832				   tx_ring, tx_desc, tx_buf);
 833
 834			tx_buf++;
 835			tx_desc++;
 836			i++;
 837			if (unlikely(!i)) {
 838				i -= tx_ring->count;
 839				tx_buf = tx_ring->tx_bi;
 840				tx_desc = I40E_TX_DESC(tx_ring, 0);
 841			}
 842
 843			/* unmap any remaining paged data */
 844			if (dma_unmap_len(tx_buf, len)) {
 845				dma_unmap_page(tx_ring->dev,
 846					       dma_unmap_addr(tx_buf, dma),
 847					       dma_unmap_len(tx_buf, len),
 848					       DMA_TO_DEVICE);
 849				dma_unmap_len_set(tx_buf, len, 0);
 850			}
 851		}
 852
 853		/* move us one more past the eop_desc for start of next pkt */
 854		tx_buf++;
 855		tx_desc++;
 856		i++;
 857		if (unlikely(!i)) {
 858			i -= tx_ring->count;
 859			tx_buf = tx_ring->tx_bi;
 860			tx_desc = I40E_TX_DESC(tx_ring, 0);
 861		}
 862
 863		prefetch(tx_desc);
 864
 865		/* update budget accounting */
 866		budget--;
 867	} while (likely(budget));
 868
 869	i += tx_ring->count;
 870	tx_ring->next_to_clean = i;
 871	i40e_update_tx_stats(tx_ring, total_packets, total_bytes);
 872	i40e_arm_wb(tx_ring, vsi, budget);
 873
 874	if (ring_is_xdp(tx_ring))
 875		return !!budget;
 876
 877	/* notify netdev of completed buffers */
 878	netdev_tx_completed_queue(txring_txq(tx_ring),
 879				  total_packets, total_bytes);
 880
 881#define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
 882	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 883		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 884		/* Make sure that anybody stopping the queue after this
 885		 * sees the new next_to_clean.
 886		 */
 887		smp_mb();
 888		if (__netif_subqueue_stopped(tx_ring->netdev,
 889					     tx_ring->queue_index) &&
 890		   !test_bit(__I40E_VSI_DOWN, vsi->state)) {
 891			netif_wake_subqueue(tx_ring->netdev,
 892					    tx_ring->queue_index);
 893			++tx_ring->tx_stats.restart_queue;
 894		}
 895	}
 896
 897	return !!budget;
 898}
 899
 900/**
 901 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
 902 * @vsi: the VSI we care about
 903 * @q_vector: the vector on which to enable writeback
 904 *
 905 **/
 906static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
 907				  struct i40e_q_vector *q_vector)
 908{
 909	u16 flags = q_vector->tx.ring[0].flags;
 910	u32 val;
 911
 912	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
 913		return;
 914
 915	if (q_vector->arm_wb_state)
 916		return;
 917
 918	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
 919		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
 920		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
 921
 922		wr32(&vsi->back->hw,
 923		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx),
 924		     val);
 925	} else {
 926		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
 927		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
 928
 929		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
 930	}
 931	q_vector->arm_wb_state = true;
 932}
 933
 934/**
 935 * i40e_force_wb - Issue SW Interrupt so HW does a wb
 936 * @vsi: the VSI we care about
 937 * @q_vector: the vector  on which to force writeback
 938 *
 939 **/
 940void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
 941{
 942	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
 943		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
 944			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
 945			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
 946			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
 947			  /* allow 00 to be written to the index */
 948
 949		wr32(&vsi->back->hw,
 950		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx), val);
 951	} else {
 952		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
 953			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
 954			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
 955			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
 956			/* allow 00 to be written to the index */
 957
 958		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
 959	}
 960}
 961
 962static inline bool i40e_container_is_rx(struct i40e_q_vector *q_vector,
 963					struct i40e_ring_container *rc)
 964{
 965	return &q_vector->rx == rc;
 966}
 967
 968static inline unsigned int i40e_itr_divisor(struct i40e_q_vector *q_vector)
 969{
 970	unsigned int divisor;
 971
 972	switch (q_vector->vsi->back->hw.phy.link_info.link_speed) {
 973	case I40E_LINK_SPEED_40GB:
 974		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 1024;
 975		break;
 976	case I40E_LINK_SPEED_25GB:
 977	case I40E_LINK_SPEED_20GB:
 978		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 512;
 979		break;
 980	default:
 981	case I40E_LINK_SPEED_10GB:
 982		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 256;
 983		break;
 984	case I40E_LINK_SPEED_1GB:
 985	case I40E_LINK_SPEED_100MB:
 986		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 32;
 987		break;
 988	}
 989
 990	return divisor;
 991}
 992
 993/**
 994 * i40e_update_itr - update the dynamic ITR value based on statistics
 995 * @q_vector: structure containing interrupt and ring information
 996 * @rc: structure containing ring performance data
 997 *
 998 * Stores a new ITR value based on packets and byte
 999 * counts during the last interrupt.  The advantage of per interrupt
1000 * computation is faster updates and more accurate ITR for the current
1001 * traffic pattern.  Constants in this function were computed
1002 * based on theoretical maximum wire speed and thresholds were set based
1003 * on testing data as well as attempting to minimize response time
1004 * while increasing bulk throughput.
1005 **/
1006static void i40e_update_itr(struct i40e_q_vector *q_vector,
1007			    struct i40e_ring_container *rc)
1008{
1009	unsigned int avg_wire_size, packets, bytes, itr;
1010	unsigned long next_update = jiffies;
1011
1012	/* If we don't have any rings just leave ourselves set for maximum
1013	 * possible latency so we take ourselves out of the equation.
1014	 */
1015	if (!rc->ring || !ITR_IS_DYNAMIC(rc->ring->itr_setting))
1016		return;
1017
1018	/* For Rx we want to push the delay up and default to low latency.
1019	 * for Tx we want to pull the delay down and default to high latency.
1020	 */
1021	itr = i40e_container_is_rx(q_vector, rc) ?
1022	      I40E_ITR_ADAPTIVE_MIN_USECS | I40E_ITR_ADAPTIVE_LATENCY :
1023	      I40E_ITR_ADAPTIVE_MAX_USECS | I40E_ITR_ADAPTIVE_LATENCY;
1024
1025	/* If we didn't update within up to 1 - 2 jiffies we can assume
1026	 * that either packets are coming in so slow there hasn't been
1027	 * any work, or that there is so much work that NAPI is dealing
1028	 * with interrupt moderation and we don't need to do anything.
1029	 */
1030	if (time_after(next_update, rc->next_update))
1031		goto clear_counts;
1032
1033	/* If itr_countdown is set it means we programmed an ITR within
1034	 * the last 4 interrupt cycles. This has a side effect of us
1035	 * potentially firing an early interrupt. In order to work around
1036	 * this we need to throw out any data received for a few
1037	 * interrupts following the update.
1038	 */
1039	if (q_vector->itr_countdown) {
1040		itr = rc->target_itr;
1041		goto clear_counts;
1042	}
1043
1044	packets = rc->total_packets;
1045	bytes = rc->total_bytes;
1046
1047	if (i40e_container_is_rx(q_vector, rc)) {
1048		/* If Rx there are 1 to 4 packets and bytes are less than
1049		 * 9000 assume insufficient data to use bulk rate limiting
1050		 * approach unless Tx is already in bulk rate limiting. We
1051		 * are likely latency driven.
1052		 */
1053		if (packets && packets < 4 && bytes < 9000 &&
1054		    (q_vector->tx.target_itr & I40E_ITR_ADAPTIVE_LATENCY)) {
1055			itr = I40E_ITR_ADAPTIVE_LATENCY;
1056			goto adjust_by_size;
1057		}
1058	} else if (packets < 4) {
1059		/* If we have Tx and Rx ITR maxed and Tx ITR is running in
1060		 * bulk mode and we are receiving 4 or fewer packets just
1061		 * reset the ITR_ADAPTIVE_LATENCY bit for latency mode so
1062		 * that the Rx can relax.
1063		 */
1064		if (rc->target_itr == I40E_ITR_ADAPTIVE_MAX_USECS &&
1065		    (q_vector->rx.target_itr & I40E_ITR_MASK) ==
1066		     I40E_ITR_ADAPTIVE_MAX_USECS)
1067			goto clear_counts;
1068	} else if (packets > 32) {
1069		/* If we have processed over 32 packets in a single interrupt
1070		 * for Tx assume we need to switch over to "bulk" mode.
1071		 */
1072		rc->target_itr &= ~I40E_ITR_ADAPTIVE_LATENCY;
1073	}
1074
1075	/* We have no packets to actually measure against. This means
1076	 * either one of the other queues on this vector is active or
1077	 * we are a Tx queue doing TSO with too high of an interrupt rate.
1078	 *
1079	 * Between 4 and 56 we can assume that our current interrupt delay
1080	 * is only slightly too low. As such we should increase it by a small
1081	 * fixed amount.
1082	 */
1083	if (packets < 56) {
1084		itr = rc->target_itr + I40E_ITR_ADAPTIVE_MIN_INC;
1085		if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1086			itr &= I40E_ITR_ADAPTIVE_LATENCY;
1087			itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1088		}
1089		goto clear_counts;
1090	}
1091
1092	if (packets <= 256) {
1093		itr = min(q_vector->tx.current_itr, q_vector->rx.current_itr);
1094		itr &= I40E_ITR_MASK;
1095
1096		/* Between 56 and 112 is our "goldilocks" zone where we are
1097		 * working out "just right". Just report that our current
1098		 * ITR is good for us.
1099		 */
1100		if (packets <= 112)
1101			goto clear_counts;
1102
1103		/* If packet count is 128 or greater we are likely looking
1104		 * at a slight overrun of the delay we want. Try halving
1105		 * our delay to see if that will cut the number of packets
1106		 * in half per interrupt.
1107		 */
1108		itr /= 2;
1109		itr &= I40E_ITR_MASK;
1110		if (itr < I40E_ITR_ADAPTIVE_MIN_USECS)
1111			itr = I40E_ITR_ADAPTIVE_MIN_USECS;
1112
1113		goto clear_counts;
1114	}
1115
1116	/* The paths below assume we are dealing with a bulk ITR since
1117	 * number of packets is greater than 256. We are just going to have
1118	 * to compute a value and try to bring the count under control,
1119	 * though for smaller packet sizes there isn't much we can do as
1120	 * NAPI polling will likely be kicking in sooner rather than later.
1121	 */
1122	itr = I40E_ITR_ADAPTIVE_BULK;
1123
1124adjust_by_size:
1125	/* If packet counts are 256 or greater we can assume we have a gross
1126	 * overestimation of what the rate should be. Instead of trying to fine
1127	 * tune it just use the formula below to try and dial in an exact value
1128	 * give the current packet size of the frame.
1129	 */
1130	avg_wire_size = bytes / packets;
1131
1132	/* The following is a crude approximation of:
1133	 *  wmem_default / (size + overhead) = desired_pkts_per_int
1134	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1135	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1136	 *
1137	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1138	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1139	 * formula down to
1140	 *
1141	 *  (170 * (size + 24)) / (size + 640) = ITR
1142	 *
1143	 * We first do some math on the packet size and then finally bitshift
1144	 * by 8 after rounding up. We also have to account for PCIe link speed
1145	 * difference as ITR scales based on this.
1146	 */
1147	if (avg_wire_size <= 60) {
1148		/* Start at 250k ints/sec */
1149		avg_wire_size = 4096;
1150	} else if (avg_wire_size <= 380) {
1151		/* 250K ints/sec to 60K ints/sec */
1152		avg_wire_size *= 40;
1153		avg_wire_size += 1696;
1154	} else if (avg_wire_size <= 1084) {
1155		/* 60K ints/sec to 36K ints/sec */
1156		avg_wire_size *= 15;
1157		avg_wire_size += 11452;
1158	} else if (avg_wire_size <= 1980) {
1159		/* 36K ints/sec to 30K ints/sec */
1160		avg_wire_size *= 5;
1161		avg_wire_size += 22420;
1162	} else {
1163		/* plateau at a limit of 30K ints/sec */
1164		avg_wire_size = 32256;
1165	}
1166
1167	/* If we are in low latency mode halve our delay which doubles the
1168	 * rate to somewhere between 100K to 16K ints/sec
1169	 */
1170	if (itr & I40E_ITR_ADAPTIVE_LATENCY)
1171		avg_wire_size /= 2;
1172
1173	/* Resultant value is 256 times larger than it needs to be. This
1174	 * gives us room to adjust the value as needed to either increase
1175	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
1176	 *
1177	 * Use addition as we have already recorded the new latency flag
1178	 * for the ITR value.
1179	 */
1180	itr += DIV_ROUND_UP(avg_wire_size, i40e_itr_divisor(q_vector)) *
1181	       I40E_ITR_ADAPTIVE_MIN_INC;
1182
1183	if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1184		itr &= I40E_ITR_ADAPTIVE_LATENCY;
1185		itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1186	}
1187
1188clear_counts:
1189	/* write back value */
1190	rc->target_itr = itr;
1191
1192	/* next update should occur within next jiffy */
1193	rc->next_update = next_update + 1;
1194
1195	rc->total_bytes = 0;
1196	rc->total_packets = 0;
1197}
1198
1199static struct i40e_rx_buffer *i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
1200{
1201	return &rx_ring->rx_bi[idx];
1202}
1203
1204/**
1205 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1206 * @rx_ring: rx descriptor ring to store buffers on
1207 * @old_buff: donor buffer to have page reused
1208 *
1209 * Synchronizes page for reuse by the adapter
1210 **/
1211static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1212			       struct i40e_rx_buffer *old_buff)
1213{
1214	struct i40e_rx_buffer *new_buff;
1215	u16 nta = rx_ring->next_to_alloc;
1216
1217	new_buff = i40e_rx_bi(rx_ring, nta);
1218
1219	/* update, and store next to alloc */
1220	nta++;
1221	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1222
1223	/* transfer page from old buffer to new buffer */
1224	new_buff->dma		= old_buff->dma;
1225	new_buff->page		= old_buff->page;
1226	new_buff->page_offset	= old_buff->page_offset;
1227	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1228
1229	rx_ring->rx_stats.page_reuse_count++;
1230
1231	/* clear contents of buffer_info */
1232	old_buff->page = NULL;
1233}
1234
1235/**
1236 * i40e_clean_programming_status - clean the programming status descriptor
1237 * @rx_ring: the rx ring that has this descriptor
1238 * @qword0_raw: qword0
1239 * @qword1: qword1 representing status_error_len in CPU ordering
1240 *
1241 * Flow director should handle FD_FILTER_STATUS to check its filter programming
1242 * status being successful or not and take actions accordingly. FCoE should
1243 * handle its context/filter programming/invalidation status and take actions.
1244 *
1245 * Returns an i40e_rx_buffer to reuse if the cleanup occurred, otherwise NULL.
1246 **/
1247void i40e_clean_programming_status(struct i40e_ring *rx_ring, u64 qword0_raw,
1248				   u64 qword1)
1249{
1250	u8 id;
1251
1252	id = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1253		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1254
1255	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1256		i40e_fd_handle_status(rx_ring, qword0_raw, qword1, id);
1257}
1258
1259/**
1260 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1261 * @tx_ring: the tx ring to set up
1262 *
1263 * Return 0 on success, negative on error
1264 **/
1265int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1266{
1267	struct device *dev = tx_ring->dev;
1268	int bi_size;
1269
1270	if (!dev)
1271		return -ENOMEM;
1272
1273	/* warn if we are about to overwrite the pointer */
1274	WARN_ON(tx_ring->tx_bi);
1275	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1276	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1277	if (!tx_ring->tx_bi)
1278		goto err;
1279
1280	u64_stats_init(&tx_ring->syncp);
1281
1282	/* round up to nearest 4K */
1283	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1284	/* add u32 for head writeback, align after this takes care of
1285	 * guaranteeing this is at least one cache line in size
1286	 */
1287	tx_ring->size += sizeof(u32);
1288	tx_ring->size = ALIGN(tx_ring->size, 4096);
1289	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1290					   &tx_ring->dma, GFP_KERNEL);
1291	if (!tx_ring->desc) {
1292		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1293			 tx_ring->size);
1294		goto err;
1295	}
1296
1297	tx_ring->next_to_use = 0;
1298	tx_ring->next_to_clean = 0;
1299	tx_ring->tx_stats.prev_pkt_ctr = -1;
1300	return 0;
1301
1302err:
1303	kfree(tx_ring->tx_bi);
1304	tx_ring->tx_bi = NULL;
1305	return -ENOMEM;
1306}
1307
1308int i40e_alloc_rx_bi(struct i40e_ring *rx_ring)
1309{
1310	unsigned long sz = sizeof(*rx_ring->rx_bi) * rx_ring->count;
1311
1312	rx_ring->rx_bi = kzalloc(sz, GFP_KERNEL);
1313	return rx_ring->rx_bi ? 0 : -ENOMEM;
1314}
1315
1316static void i40e_clear_rx_bi(struct i40e_ring *rx_ring)
1317{
1318	memset(rx_ring->rx_bi, 0, sizeof(*rx_ring->rx_bi) * rx_ring->count);
1319}
1320
1321/**
1322 * i40e_clean_rx_ring - Free Rx buffers
1323 * @rx_ring: ring to be cleaned
1324 **/
1325void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1326{
1327	u16 i;
1328
1329	/* ring already cleared, nothing to do */
1330	if (!rx_ring->rx_bi)
1331		return;
1332
1333	if (rx_ring->skb) {
1334		dev_kfree_skb(rx_ring->skb);
1335		rx_ring->skb = NULL;
1336	}
1337
1338	if (rx_ring->xsk_umem) {
1339		i40e_xsk_clean_rx_ring(rx_ring);
1340		goto skip_free;
1341	}
1342
1343	/* Free all the Rx ring sk_buffs */
1344	for (i = 0; i < rx_ring->count; i++) {
1345		struct i40e_rx_buffer *rx_bi = i40e_rx_bi(rx_ring, i);
1346
1347		if (!rx_bi->page)
1348			continue;
1349
1350		/* Invalidate cache lines that may have been written to by
1351		 * device so that we avoid corrupting memory.
1352		 */
1353		dma_sync_single_range_for_cpu(rx_ring->dev,
1354					      rx_bi->dma,
1355					      rx_bi->page_offset,
1356					      rx_ring->rx_buf_len,
1357					      DMA_FROM_DEVICE);
1358
1359		/* free resources associated with mapping */
1360		dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1361				     i40e_rx_pg_size(rx_ring),
1362				     DMA_FROM_DEVICE,
1363				     I40E_RX_DMA_ATTR);
1364
1365		__page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1366
1367		rx_bi->page = NULL;
1368		rx_bi->page_offset = 0;
1369	}
1370
1371skip_free:
1372	if (rx_ring->xsk_umem)
1373		i40e_clear_rx_bi_zc(rx_ring);
1374	else
1375		i40e_clear_rx_bi(rx_ring);
1376
1377	/* Zero out the descriptor ring */
1378	memset(rx_ring->desc, 0, rx_ring->size);
1379
1380	rx_ring->next_to_alloc = 0;
1381	rx_ring->next_to_clean = 0;
1382	rx_ring->next_to_use = 0;
1383}
1384
1385/**
1386 * i40e_free_rx_resources - Free Rx resources
1387 * @rx_ring: ring to clean the resources from
1388 *
1389 * Free all receive software resources
1390 **/
1391void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1392{
1393	i40e_clean_rx_ring(rx_ring);
1394	if (rx_ring->vsi->type == I40E_VSI_MAIN)
1395		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
1396	rx_ring->xdp_prog = NULL;
1397	kfree(rx_ring->rx_bi);
1398	rx_ring->rx_bi = NULL;
1399
1400	if (rx_ring->desc) {
1401		dma_free_coherent(rx_ring->dev, rx_ring->size,
1402				  rx_ring->desc, rx_ring->dma);
1403		rx_ring->desc = NULL;
1404	}
1405}
1406
1407/**
1408 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1409 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1410 *
1411 * Returns 0 on success, negative on failure
1412 **/
1413int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1414{
1415	struct device *dev = rx_ring->dev;
1416	int err;
1417
1418	u64_stats_init(&rx_ring->syncp);
1419
1420	/* Round up to nearest 4K */
1421	rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1422	rx_ring->size = ALIGN(rx_ring->size, 4096);
1423	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1424					   &rx_ring->dma, GFP_KERNEL);
1425
1426	if (!rx_ring->desc) {
1427		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1428			 rx_ring->size);
1429		return -ENOMEM;
1430	}
1431
1432	rx_ring->next_to_alloc = 0;
1433	rx_ring->next_to_clean = 0;
1434	rx_ring->next_to_use = 0;
1435
1436	/* XDP RX-queue info only needed for RX rings exposed to XDP */
1437	if (rx_ring->vsi->type == I40E_VSI_MAIN) {
1438		err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
1439				       rx_ring->queue_index);
1440		if (err < 0)
1441			return err;
1442	}
1443
1444	rx_ring->xdp_prog = rx_ring->vsi->xdp_prog;
1445
1446	return 0;
1447}
1448
1449/**
1450 * i40e_release_rx_desc - Store the new tail and head values
1451 * @rx_ring: ring to bump
1452 * @val: new head index
1453 **/
1454void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1455{
1456	rx_ring->next_to_use = val;
1457
1458	/* update next to alloc since we have filled the ring */
1459	rx_ring->next_to_alloc = val;
1460
1461	/* Force memory writes to complete before letting h/w
1462	 * know there are new descriptors to fetch.  (Only
1463	 * applicable for weak-ordered memory model archs,
1464	 * such as IA-64).
1465	 */
1466	wmb();
1467	writel(val, rx_ring->tail);
1468}
1469
1470/**
1471 * i40e_rx_offset - Return expected offset into page to access data
1472 * @rx_ring: Ring we are requesting offset of
1473 *
1474 * Returns the offset value for ring into the data buffer.
1475 */
1476static inline unsigned int i40e_rx_offset(struct i40e_ring *rx_ring)
1477{
1478	return ring_uses_build_skb(rx_ring) ? I40E_SKB_PAD : 0;
1479}
1480
1481static unsigned int i40e_rx_frame_truesize(struct i40e_ring *rx_ring,
1482					   unsigned int size)
1483{
1484	unsigned int truesize;
1485
1486#if (PAGE_SIZE < 8192)
1487	truesize = i40e_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1488#else
1489	truesize = i40e_rx_offset(rx_ring) ?
1490		SKB_DATA_ALIGN(size + i40e_rx_offset(rx_ring)) +
1491		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1492		SKB_DATA_ALIGN(size);
1493#endif
1494	return truesize;
1495}
1496
1497/**
1498 * i40e_alloc_mapped_page - recycle or make a new page
1499 * @rx_ring: ring to use
1500 * @bi: rx_buffer struct to modify
1501 *
1502 * Returns true if the page was successfully allocated or
1503 * reused.
1504 **/
1505static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1506				   struct i40e_rx_buffer *bi)
1507{
1508	struct page *page = bi->page;
1509	dma_addr_t dma;
1510
1511	/* since we are recycling buffers we should seldom need to alloc */
1512	if (likely(page)) {
1513		rx_ring->rx_stats.page_reuse_count++;
1514		return true;
1515	}
1516
1517	/* alloc new page for storage */
1518	page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1519	if (unlikely(!page)) {
1520		rx_ring->rx_stats.alloc_page_failed++;
1521		return false;
1522	}
1523
1524	/* map page for use */
1525	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1526				 i40e_rx_pg_size(rx_ring),
1527				 DMA_FROM_DEVICE,
1528				 I40E_RX_DMA_ATTR);
1529
1530	/* if mapping failed free memory back to system since
1531	 * there isn't much point in holding memory we can't use
1532	 */
1533	if (dma_mapping_error(rx_ring->dev, dma)) {
1534		__free_pages(page, i40e_rx_pg_order(rx_ring));
1535		rx_ring->rx_stats.alloc_page_failed++;
1536		return false;
1537	}
1538
1539	bi->dma = dma;
1540	bi->page = page;
1541	bi->page_offset = i40e_rx_offset(rx_ring);
1542	page_ref_add(page, USHRT_MAX - 1);
1543	bi->pagecnt_bias = USHRT_MAX;
1544
1545	return true;
1546}
1547
1548/**
1549 * i40e_alloc_rx_buffers - Replace used receive buffers
1550 * @rx_ring: ring to place buffers on
1551 * @cleaned_count: number of buffers to replace
1552 *
1553 * Returns false if all allocations were successful, true if any fail
1554 **/
1555bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1556{
1557	u16 ntu = rx_ring->next_to_use;
1558	union i40e_rx_desc *rx_desc;
1559	struct i40e_rx_buffer *bi;
1560
1561	/* do nothing if no valid netdev defined */
1562	if (!rx_ring->netdev || !cleaned_count)
1563		return false;
1564
1565	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1566	bi = i40e_rx_bi(rx_ring, ntu);
1567
1568	do {
1569		if (!i40e_alloc_mapped_page(rx_ring, bi))
1570			goto no_buffers;
1571
1572		/* sync the buffer for use by the device */
1573		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1574						 bi->page_offset,
1575						 rx_ring->rx_buf_len,
1576						 DMA_FROM_DEVICE);
1577
1578		/* Refresh the desc even if buffer_addrs didn't change
1579		 * because each write-back erases this info.
1580		 */
1581		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1582
1583		rx_desc++;
1584		bi++;
1585		ntu++;
1586		if (unlikely(ntu == rx_ring->count)) {
1587			rx_desc = I40E_RX_DESC(rx_ring, 0);
1588			bi = i40e_rx_bi(rx_ring, 0);
1589			ntu = 0;
1590		}
1591
1592		/* clear the status bits for the next_to_use descriptor */
1593		rx_desc->wb.qword1.status_error_len = 0;
1594
1595		cleaned_count--;
1596	} while (cleaned_count);
1597
1598	if (rx_ring->next_to_use != ntu)
1599		i40e_release_rx_desc(rx_ring, ntu);
1600
1601	return false;
1602
1603no_buffers:
1604	if (rx_ring->next_to_use != ntu)
1605		i40e_release_rx_desc(rx_ring, ntu);
1606
1607	/* make sure to come back via polling to try again after
1608	 * allocation failure
1609	 */
1610	return true;
1611}
1612
1613/**
1614 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1615 * @vsi: the VSI we care about
1616 * @skb: skb currently being received and modified
1617 * @rx_desc: the receive descriptor
1618 **/
1619static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1620				    struct sk_buff *skb,
1621				    union i40e_rx_desc *rx_desc)
1622{
1623	struct i40e_rx_ptype_decoded decoded;
1624	u32 rx_error, rx_status;
1625	bool ipv4, ipv6;
1626	u8 ptype;
1627	u64 qword;
1628
1629	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1630	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1631	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1632		   I40E_RXD_QW1_ERROR_SHIFT;
1633	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1634		    I40E_RXD_QW1_STATUS_SHIFT;
1635	decoded = decode_rx_desc_ptype(ptype);
1636
1637	skb->ip_summed = CHECKSUM_NONE;
1638
1639	skb_checksum_none_assert(skb);
1640
1641	/* Rx csum enabled and ip headers found? */
1642	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1643		return;
1644
1645	/* did the hardware decode the packet and checksum? */
1646	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1647		return;
1648
1649	/* both known and outer_ip must be set for the below code to work */
1650	if (!(decoded.known && decoded.outer_ip))
1651		return;
1652
1653	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1654	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1655	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1656	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1657
1658	if (ipv4 &&
1659	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1660			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1661		goto checksum_fail;
1662
1663	/* likely incorrect csum if alternate IP extension headers found */
1664	if (ipv6 &&
1665	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1666		/* don't increment checksum err here, non-fatal err */
1667		return;
1668
1669	/* there was some L4 error, count error and punt packet to the stack */
1670	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1671		goto checksum_fail;
1672
1673	/* handle packets that were not able to be checksummed due
1674	 * to arrival speed, in this case the stack can compute
1675	 * the csum.
1676	 */
1677	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1678		return;
1679
1680	/* If there is an outer header present that might contain a checksum
1681	 * we need to bump the checksum level by 1 to reflect the fact that
1682	 * we are indicating we validated the inner checksum.
1683	 */
1684	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1685		skb->csum_level = 1;
1686
1687	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1688	switch (decoded.inner_prot) {
1689	case I40E_RX_PTYPE_INNER_PROT_TCP:
1690	case I40E_RX_PTYPE_INNER_PROT_UDP:
1691	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1692		skb->ip_summed = CHECKSUM_UNNECESSARY;
1693		fallthrough;
1694	default:
1695		break;
1696	}
1697
1698	return;
1699
1700checksum_fail:
1701	vsi->back->hw_csum_rx_error++;
1702}
1703
1704/**
1705 * i40e_ptype_to_htype - get a hash type
1706 * @ptype: the ptype value from the descriptor
1707 *
1708 * Returns a hash type to be used by skb_set_hash
1709 **/
1710static inline int i40e_ptype_to_htype(u8 ptype)
1711{
1712	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1713
1714	if (!decoded.known)
1715		return PKT_HASH_TYPE_NONE;
1716
1717	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1718	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1719		return PKT_HASH_TYPE_L4;
1720	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1721		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1722		return PKT_HASH_TYPE_L3;
1723	else
1724		return PKT_HASH_TYPE_L2;
1725}
1726
1727/**
1728 * i40e_rx_hash - set the hash value in the skb
1729 * @ring: descriptor ring
1730 * @rx_desc: specific descriptor
1731 * @skb: skb currently being received and modified
1732 * @rx_ptype: Rx packet type
1733 **/
1734static inline void i40e_rx_hash(struct i40e_ring *ring,
1735				union i40e_rx_desc *rx_desc,
1736				struct sk_buff *skb,
1737				u8 rx_ptype)
1738{
1739	u32 hash;
1740	const __le64 rss_mask =
1741		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1742			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1743
1744	if (!(ring->netdev->features & NETIF_F_RXHASH))
1745		return;
1746
1747	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1748		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1749		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1750	}
1751}
1752
1753/**
1754 * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1755 * @rx_ring: rx descriptor ring packet is being transacted on
1756 * @rx_desc: pointer to the EOP Rx descriptor
1757 * @skb: pointer to current skb being populated
1758 * @rx_ptype: the packet type decoded by hardware
1759 *
1760 * This function checks the ring, descriptor, and packet information in
1761 * order to populate the hash, checksum, VLAN, protocol, and
1762 * other fields within the skb.
1763 **/
1764void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1765			     union i40e_rx_desc *rx_desc, struct sk_buff *skb)
1766{
1767	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1768	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1769			I40E_RXD_QW1_STATUS_SHIFT;
1770	u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1771	u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1772		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1773	u8 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1774		      I40E_RXD_QW1_PTYPE_SHIFT;
1775
1776	if (unlikely(tsynvalid))
1777		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1778
1779	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1780
1781	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1782
1783	skb_record_rx_queue(skb, rx_ring->queue_index);
1784
1785	if (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
1786		u16 vlan_tag = rx_desc->wb.qword0.lo_dword.l2tag1;
1787
1788		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1789				       le16_to_cpu(vlan_tag));
1790	}
1791
1792	/* modifies the skb - consumes the enet header */
1793	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1794}
1795
1796/**
1797 * i40e_cleanup_headers - Correct empty headers
1798 * @rx_ring: rx descriptor ring packet is being transacted on
1799 * @skb: pointer to current skb being fixed
1800 * @rx_desc: pointer to the EOP Rx descriptor
1801 *
1802 * Also address the case where we are pulling data in on pages only
1803 * and as such no data is present in the skb header.
1804 *
1805 * In addition if skb is not at least 60 bytes we need to pad it so that
1806 * it is large enough to qualify as a valid Ethernet frame.
1807 *
1808 * Returns true if an error was encountered and skb was freed.
1809 **/
1810static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
1811				 union i40e_rx_desc *rx_desc)
1812
1813{
1814	/* XDP packets use error pointer so abort at this point */
1815	if (IS_ERR(skb))
1816		return true;
1817
1818	/* ERR_MASK will only have valid bits if EOP set, and
1819	 * what we are doing here is actually checking
1820	 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1821	 * the error field
1822	 */
1823	if (unlikely(i40e_test_staterr(rx_desc,
1824				       BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1825		dev_kfree_skb_any(skb);
1826		return true;
1827	}
1828
1829	/* if eth_skb_pad returns an error the skb was freed */
1830	if (eth_skb_pad(skb))
1831		return true;
1832
1833	return false;
1834}
1835
1836/**
1837 * i40e_page_is_reusable - check if any reuse is possible
1838 * @page: page struct to check
1839 *
1840 * A page is not reusable if it was allocated under low memory
1841 * conditions, or it's not in the same NUMA node as this CPU.
1842 */
1843static inline bool i40e_page_is_reusable(struct page *page)
1844{
1845	return (page_to_nid(page) == numa_mem_id()) &&
1846		!page_is_pfmemalloc(page);
1847}
1848
1849/**
1850 * i40e_can_reuse_rx_page - Determine if this page can be reused by
1851 * the adapter for another receive
1852 *
1853 * @rx_buffer: buffer containing the page
1854 *
1855 * If page is reusable, rx_buffer->page_offset is adjusted to point to
1856 * an unused region in the page.
1857 *
1858 * For small pages, @truesize will be a constant value, half the size
1859 * of the memory at page.  We'll attempt to alternate between high and
1860 * low halves of the page, with one half ready for use by the hardware
1861 * and the other half being consumed by the stack.  We use the page
1862 * ref count to determine whether the stack has finished consuming the
1863 * portion of this page that was passed up with a previous packet.  If
1864 * the page ref count is >1, we'll assume the "other" half page is
1865 * still busy, and this page cannot be reused.
1866 *
1867 * For larger pages, @truesize will be the actual space used by the
1868 * received packet (adjusted upward to an even multiple of the cache
1869 * line size).  This will advance through the page by the amount
1870 * actually consumed by the received packets while there is still
1871 * space for a buffer.  Each region of larger pages will be used at
1872 * most once, after which the page will not be reused.
1873 *
1874 * In either case, if the page is reusable its refcount is increased.
1875 **/
1876static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer)
1877{
1878	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1879	struct page *page = rx_buffer->page;
1880
1881	/* Is any reuse possible? */
1882	if (unlikely(!i40e_page_is_reusable(page)))
1883		return false;
1884
1885#if (PAGE_SIZE < 8192)
1886	/* if we are only owner of page we can reuse it */
1887	if (unlikely((page_count(page) - pagecnt_bias) > 1))
1888		return false;
1889#else
1890#define I40E_LAST_OFFSET \
1891	(SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
1892	if (rx_buffer->page_offset > I40E_LAST_OFFSET)
1893		return false;
1894#endif
1895
1896	/* If we have drained the page fragment pool we need to update
1897	 * the pagecnt_bias and page count so that we fully restock the
1898	 * number of references the driver holds.
1899	 */
1900	if (unlikely(pagecnt_bias == 1)) {
1901		page_ref_add(page, USHRT_MAX - 1);
1902		rx_buffer->pagecnt_bias = USHRT_MAX;
1903	}
1904
1905	return true;
1906}
1907
1908/**
1909 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1910 * @rx_ring: rx descriptor ring to transact packets on
1911 * @rx_buffer: buffer containing page to add
1912 * @skb: sk_buff to place the data into
1913 * @size: packet length from rx_desc
1914 *
1915 * This function will add the data contained in rx_buffer->page to the skb.
1916 * It will just attach the page as a frag to the skb.
1917 *
1918 * The function will then update the page offset.
1919 **/
1920static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
1921			     struct i40e_rx_buffer *rx_buffer,
1922			     struct sk_buff *skb,
1923			     unsigned int size)
1924{
1925#if (PAGE_SIZE < 8192)
1926	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1927#else
1928	unsigned int truesize = SKB_DATA_ALIGN(size + i40e_rx_offset(rx_ring));
1929#endif
1930
1931	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1932			rx_buffer->page_offset, size, truesize);
1933
1934	/* page is being used so we must update the page offset */
1935#if (PAGE_SIZE < 8192)
1936	rx_buffer->page_offset ^= truesize;
1937#else
1938	rx_buffer->page_offset += truesize;
1939#endif
1940}
1941
1942/**
1943 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
1944 * @rx_ring: rx descriptor ring to transact packets on
1945 * @size: size of buffer to add to skb
1946 *
1947 * This function will pull an Rx buffer from the ring and synchronize it
1948 * for use by the CPU.
1949 */
1950static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
1951						 const unsigned int size)
1952{
1953	struct i40e_rx_buffer *rx_buffer;
1954
1955	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
1956	prefetchw(rx_buffer->page);
1957
1958	/* we are reusing so sync this buffer for CPU use */
1959	dma_sync_single_range_for_cpu(rx_ring->dev,
1960				      rx_buffer->dma,
1961				      rx_buffer->page_offset,
1962				      size,
1963				      DMA_FROM_DEVICE);
1964
1965	/* We have pulled a buffer for use, so decrement pagecnt_bias */
1966	rx_buffer->pagecnt_bias--;
1967
1968	return rx_buffer;
1969}
1970
1971/**
1972 * i40e_construct_skb - Allocate skb and populate it
1973 * @rx_ring: rx descriptor ring to transact packets on
1974 * @rx_buffer: rx buffer to pull data from
1975 * @xdp: xdp_buff pointing to the data
1976 *
1977 * This function allocates an skb.  It then populates it with the page
1978 * data from the current receive descriptor, taking care to set up the
1979 * skb correctly.
1980 */
1981static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
1982					  struct i40e_rx_buffer *rx_buffer,
1983					  struct xdp_buff *xdp)
1984{
1985	unsigned int size = xdp->data_end - xdp->data;
1986#if (PAGE_SIZE < 8192)
1987	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1988#else
1989	unsigned int truesize = SKB_DATA_ALIGN(size);
1990#endif
1991	unsigned int headlen;
1992	struct sk_buff *skb;
1993
1994	/* prefetch first cache line of first page */
1995	prefetch(xdp->data);
1996#if L1_CACHE_BYTES < 128
1997	prefetch(xdp->data + L1_CACHE_BYTES);
1998#endif
1999	/* Note, we get here by enabling legacy-rx via:
2000	 *
2001	 *    ethtool --set-priv-flags <dev> legacy-rx on
2002	 *
2003	 * In this mode, we currently get 0 extra XDP headroom as
2004	 * opposed to having legacy-rx off, where we process XDP
2005	 * packets going to stack via i40e_build_skb(). The latter
2006	 * provides us currently with 192 bytes of headroom.
2007	 *
2008	 * For i40e_construct_skb() mode it means that the
2009	 * xdp->data_meta will always point to xdp->data, since
2010	 * the helper cannot expand the head. Should this ever
2011	 * change in future for legacy-rx mode on, then lets also
2012	 * add xdp->data_meta handling here.
2013	 */
2014
2015	/* allocate a skb to store the frags */
2016	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
2017			       I40E_RX_HDR_SIZE,
2018			       GFP_ATOMIC | __GFP_NOWARN);
2019	if (unlikely(!skb))
2020		return NULL;
2021
2022	/* Determine available headroom for copy */
2023	headlen = size;
2024	if (headlen > I40E_RX_HDR_SIZE)
2025		headlen = eth_get_headlen(skb->dev, xdp->data,
2026					  I40E_RX_HDR_SIZE);
2027
2028	/* align pull length to size of long to optimize memcpy performance */
2029	memcpy(__skb_put(skb, headlen), xdp->data,
2030	       ALIGN(headlen, sizeof(long)));
2031
2032	/* update all of the pointers */
2033	size -= headlen;
2034	if (size) {
2035		skb_add_rx_frag(skb, 0, rx_buffer->page,
2036				rx_buffer->page_offset + headlen,
2037				size, truesize);
2038
2039		/* buffer is used by skb, update page_offset */
2040#if (PAGE_SIZE < 8192)
2041		rx_buffer->page_offset ^= truesize;
2042#else
2043		rx_buffer->page_offset += truesize;
2044#endif
2045	} else {
2046		/* buffer is unused, reset bias back to rx_buffer */
2047		rx_buffer->pagecnt_bias++;
2048	}
2049
2050	return skb;
2051}
2052
2053/**
2054 * i40e_build_skb - Build skb around an existing buffer
2055 * @rx_ring: Rx descriptor ring to transact packets on
2056 * @rx_buffer: Rx buffer to pull data from
2057 * @xdp: xdp_buff pointing to the data
2058 *
2059 * This function builds an skb around an existing Rx buffer, taking care
2060 * to set up the skb correctly and avoid any memcpy overhead.
2061 */
2062static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
2063				      struct i40e_rx_buffer *rx_buffer,
2064				      struct xdp_buff *xdp)
2065{
2066	unsigned int metasize = xdp->data - xdp->data_meta;
2067#if (PAGE_SIZE < 8192)
2068	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2069#else
2070	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2071				SKB_DATA_ALIGN(xdp->data_end -
2072					       xdp->data_hard_start);
2073#endif
2074	struct sk_buff *skb;
2075
2076	/* Prefetch first cache line of first page. If xdp->data_meta
2077	 * is unused, this points exactly as xdp->data, otherwise we
2078	 * likely have a consumer accessing first few bytes of meta
2079	 * data, and then actual data.
2080	 */
2081	prefetch(xdp->data_meta);
2082#if L1_CACHE_BYTES < 128
2083	prefetch(xdp->data_meta + L1_CACHE_BYTES);
2084#endif
2085	/* build an skb around the page buffer */
2086	skb = build_skb(xdp->data_hard_start, truesize);
2087	if (unlikely(!skb))
2088		return NULL;
2089
2090	/* update pointers within the skb to store the data */
2091	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2092	__skb_put(skb, xdp->data_end - xdp->data);
2093	if (metasize)
2094		skb_metadata_set(skb, metasize);
2095
2096	/* buffer is used by skb, update page_offset */
2097#if (PAGE_SIZE < 8192)
2098	rx_buffer->page_offset ^= truesize;
2099#else
2100	rx_buffer->page_offset += truesize;
2101#endif
2102
2103	return skb;
2104}
2105
2106/**
2107 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
2108 * @rx_ring: rx descriptor ring to transact packets on
2109 * @rx_buffer: rx buffer to pull data from
2110 *
2111 * This function will clean up the contents of the rx_buffer.  It will
2112 * either recycle the buffer or unmap it and free the associated resources.
2113 */
2114static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
2115			       struct i40e_rx_buffer *rx_buffer)
2116{
2117	if (i40e_can_reuse_rx_page(rx_buffer)) {
2118		/* hand second half of page back to the ring */
2119		i40e_reuse_rx_page(rx_ring, rx_buffer);
2120	} else {
2121		/* we are not reusing the buffer so unmap it */
2122		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2123				     i40e_rx_pg_size(rx_ring),
2124				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
2125		__page_frag_cache_drain(rx_buffer->page,
2126					rx_buffer->pagecnt_bias);
2127		/* clear contents of buffer_info */
2128		rx_buffer->page = NULL;
2129	}
2130}
2131
2132/**
2133 * i40e_is_non_eop - process handling of non-EOP buffers
2134 * @rx_ring: Rx ring being processed
2135 * @rx_desc: Rx descriptor for current buffer
2136 * @skb: Current socket buffer containing buffer in progress
2137 *
2138 * This function updates next to clean.  If the buffer is an EOP buffer
2139 * this function exits returning false, otherwise it will place the
2140 * sk_buff in the next buffer to be chained and return true indicating
2141 * that this is in fact a non-EOP buffer.
2142 **/
2143static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
2144			    union i40e_rx_desc *rx_desc,
2145			    struct sk_buff *skb)
2146{
2147	u32 ntc = rx_ring->next_to_clean + 1;
2148
2149	/* fetch, update, and store next to clean */
2150	ntc = (ntc < rx_ring->count) ? ntc : 0;
2151	rx_ring->next_to_clean = ntc;
2152
2153	prefetch(I40E_RX_DESC(rx_ring, ntc));
2154
2155	/* if we are the last buffer then there is nothing else to do */
2156#define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
2157	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
2158		return false;
2159
2160	rx_ring->rx_stats.non_eop_descs++;
2161
2162	return true;
2163}
2164
2165static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
2166			      struct i40e_ring *xdp_ring);
2167
2168int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp, struct i40e_ring *xdp_ring)
2169{
2170	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2171
2172	if (unlikely(!xdpf))
2173		return I40E_XDP_CONSUMED;
2174
2175	return i40e_xmit_xdp_ring(xdpf, xdp_ring);
2176}
2177
2178/**
2179 * i40e_run_xdp - run an XDP program
2180 * @rx_ring: Rx ring being processed
2181 * @xdp: XDP buffer containing the frame
2182 **/
2183static struct sk_buff *i40e_run_xdp(struct i40e_ring *rx_ring,
2184				    struct xdp_buff *xdp)
2185{
2186	int err, result = I40E_XDP_PASS;
2187	struct i40e_ring *xdp_ring;
2188	struct bpf_prog *xdp_prog;
2189	u32 act;
2190
2191	rcu_read_lock();
2192	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2193
2194	if (!xdp_prog)
2195		goto xdp_out;
2196
2197	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2198
2199	act = bpf_prog_run_xdp(xdp_prog, xdp);
2200	switch (act) {
2201	case XDP_PASS:
2202		break;
2203	case XDP_TX:
2204		xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2205		result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
2206		break;
2207	case XDP_REDIRECT:
2208		err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
2209		result = !err ? I40E_XDP_REDIR : I40E_XDP_CONSUMED;
2210		break;
2211	default:
2212		bpf_warn_invalid_xdp_action(act);
2213		fallthrough;
2214	case XDP_ABORTED:
2215		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2216		fallthrough; /* handle aborts by dropping packet */
2217	case XDP_DROP:
2218		result = I40E_XDP_CONSUMED;
2219		break;
2220	}
2221xdp_out:
2222	rcu_read_unlock();
2223	return ERR_PTR(-result);
2224}
2225
2226/**
2227 * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
2228 * @rx_ring: Rx ring
2229 * @rx_buffer: Rx buffer to adjust
2230 * @size: Size of adjustment
2231 **/
2232static void i40e_rx_buffer_flip(struct i40e_ring *rx_ring,
2233				struct i40e_rx_buffer *rx_buffer,
2234				unsigned int size)
2235{
2236	unsigned int truesize = i40e_rx_frame_truesize(rx_ring, size);
2237
2238#if (PAGE_SIZE < 8192)
2239	rx_buffer->page_offset ^= truesize;
2240#else
2241	rx_buffer->page_offset += truesize;
2242#endif
2243}
2244
2245/**
2246 * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
2247 * @xdp_ring: XDP Tx ring
2248 *
2249 * This function updates the XDP Tx ring tail register.
2250 **/
2251void i40e_xdp_ring_update_tail(struct i40e_ring *xdp_ring)
2252{
2253	/* Force memory writes to complete before letting h/w
2254	 * know there are new descriptors to fetch.
2255	 */
2256	wmb();
2257	writel_relaxed(xdp_ring->next_to_use, xdp_ring->tail);
2258}
2259
2260/**
2261 * i40e_update_rx_stats - Update Rx ring statistics
2262 * @rx_ring: rx descriptor ring
2263 * @total_rx_bytes: number of bytes received
2264 * @total_rx_packets: number of packets received
2265 *
2266 * This function updates the Rx ring statistics.
2267 **/
2268void i40e_update_rx_stats(struct i40e_ring *rx_ring,
2269			  unsigned int total_rx_bytes,
2270			  unsigned int total_rx_packets)
2271{
2272	u64_stats_update_begin(&rx_ring->syncp);
2273	rx_ring->stats.packets += total_rx_packets;
2274	rx_ring->stats.bytes += total_rx_bytes;
2275	u64_stats_update_end(&rx_ring->syncp);
2276	rx_ring->q_vector->rx.total_packets += total_rx_packets;
2277	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2278}
2279
2280/**
2281 * i40e_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map
2282 * @rx_ring: Rx ring
2283 * @xdp_res: Result of the receive batch
2284 *
2285 * This function bumps XDP Tx tail and/or flush redirect map, and
2286 * should be called when a batch of packets has been processed in the
2287 * napi loop.
2288 **/
2289void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
2290{
2291	if (xdp_res & I40E_XDP_REDIR)
2292		xdp_do_flush_map();
2293
2294	if (xdp_res & I40E_XDP_TX) {
2295		struct i40e_ring *xdp_ring =
2296			rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2297
2298		i40e_xdp_ring_update_tail(xdp_ring);
2299	}
2300}
2301
2302/**
2303 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2304 * @rx_ring: rx descriptor ring to transact packets on
2305 * @budget: Total limit on number of packets to process
2306 *
2307 * This function provides a "bounce buffer" approach to Rx interrupt
2308 * processing.  The advantage to this is that on systems that have
2309 * expensive overhead for IOMMU access this provides a means of avoiding
2310 * it by maintaining the mapping of the page to the system.
2311 *
2312 * Returns amount of work completed
2313 **/
2314static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
2315{
2316	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2317	struct sk_buff *skb = rx_ring->skb;
2318	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
2319	unsigned int xdp_xmit = 0;
2320	bool failure = false;
2321	struct xdp_buff xdp;
2322
2323#if (PAGE_SIZE < 8192)
2324	xdp.frame_sz = i40e_rx_frame_truesize(rx_ring, 0);
2325#endif
2326	xdp.rxq = &rx_ring->xdp_rxq;
2327
2328	while (likely(total_rx_packets < (unsigned int)budget)) {
2329		struct i40e_rx_buffer *rx_buffer;
2330		union i40e_rx_desc *rx_desc;
2331		unsigned int size;
2332		u64 qword;
2333
2334		/* return some buffers to hardware, one at a time is too slow */
2335		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
2336			failure = failure ||
2337				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
2338			cleaned_count = 0;
2339		}
2340
2341		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
2342
2343		/* status_error_len will always be zero for unused descriptors
2344		 * because it's cleared in cleanup, and overlaps with hdr_addr
2345		 * which is always zero because packet split isn't used, if the
2346		 * hardware wrote DD then the length will be non-zero
2347		 */
2348		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2349
2350		/* This memory barrier is needed to keep us from reading
2351		 * any other fields out of the rx_desc until we have
2352		 * verified the descriptor has been written back.
2353		 */
2354		dma_rmb();
2355
2356		if (i40e_rx_is_programming_status(qword)) {
2357			i40e_clean_programming_status(rx_ring,
2358						      rx_desc->raw.qword[0],
2359						      qword);
2360			rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2361			i40e_inc_ntc(rx_ring);
2362			i40e_reuse_rx_page(rx_ring, rx_buffer);
2363			cleaned_count++;
2364			continue;
2365		}
2366
2367		size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
2368		       I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
2369		if (!size)
2370			break;
2371
2372		i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
2373		rx_buffer = i40e_get_rx_buffer(rx_ring, size);
2374
2375		/* retrieve a buffer from the ring */
2376		if (!skb) {
2377			xdp.data = page_address(rx_buffer->page) +
2378				   rx_buffer->page_offset;
2379			xdp.data_meta = xdp.data;
2380			xdp.data_hard_start = xdp.data -
2381					      i40e_rx_offset(rx_ring);
2382			xdp.data_end = xdp.data + size;
2383#if (PAGE_SIZE > 4096)
2384			/* At larger PAGE_SIZE, frame_sz depend on len size */
2385			xdp.frame_sz = i40e_rx_frame_truesize(rx_ring, size);
2386#endif
2387			skb = i40e_run_xdp(rx_ring, &xdp);
2388		}
2389
2390		if (IS_ERR(skb)) {
2391			unsigned int xdp_res = -PTR_ERR(skb);
2392
2393			if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
2394				xdp_xmit |= xdp_res;
2395				i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
2396			} else {
2397				rx_buffer->pagecnt_bias++;
2398			}
2399			total_rx_bytes += size;
2400			total_rx_packets++;
2401		} else if (skb) {
2402			i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
2403		} else if (ring_uses_build_skb(rx_ring)) {
2404			skb = i40e_build_skb(rx_ring, rx_buffer, &xdp);
2405		} else {
2406			skb = i40e_construct_skb(rx_ring, rx_buffer, &xdp);
2407		}
2408
2409		/* exit if we failed to retrieve a buffer */
2410		if (!skb) {
2411			rx_ring->rx_stats.alloc_buff_failed++;
2412			rx_buffer->pagecnt_bias++;
2413			break;
2414		}
2415
2416		i40e_put_rx_buffer(rx_ring, rx_buffer);
2417		cleaned_count++;
2418
2419		if (i40e_is_non_eop(rx_ring, rx_desc, skb))
2420			continue;
2421
2422		if (i40e_cleanup_headers(rx_ring, skb, rx_desc)) {
2423			skb = NULL;
2424			continue;
2425		}
2426
2427		/* probably a little skewed due to removing CRC */
2428		total_rx_bytes += skb->len;
2429
2430		/* populate checksum, VLAN, and protocol */
2431		i40e_process_skb_fields(rx_ring, rx_desc, skb);
2432
2433		i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
2434		napi_gro_receive(&rx_ring->q_vector->napi, skb);
2435		skb = NULL;
2436
2437		/* update budget accounting */
2438		total_rx_packets++;
2439	}
2440
2441	i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
2442	rx_ring->skb = skb;
2443
2444	i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
2445
2446	/* guarantee a trip back through this routine if there was a failure */
2447	return failure ? budget : (int)total_rx_packets;
2448}
2449
2450static inline u32 i40e_buildreg_itr(const int type, u16 itr)
2451{
2452	u32 val;
2453
2454	/* We don't bother with setting the CLEARPBA bit as the data sheet
2455	 * points out doing so is "meaningless since it was already
2456	 * auto-cleared". The auto-clearing happens when the interrupt is
2457	 * asserted.
2458	 *
2459	 * Hardware errata 28 for also indicates that writing to a
2460	 * xxINT_DYN_CTLx CSR with INTENA_MSK (bit 31) set to 0 will clear
2461	 * an event in the PBA anyway so we need to rely on the automask
2462	 * to hold pending events for us until the interrupt is re-enabled
2463	 *
2464	 * The itr value is reported in microseconds, and the register
2465	 * value is recorded in 2 microsecond units. For this reason we
2466	 * only need to shift by the interval shift - 1 instead of the
2467	 * full value.
2468	 */
2469	itr &= I40E_ITR_MASK;
2470
2471	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2472	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2473	      (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1));
2474
2475	return val;
2476}
2477
2478/* a small macro to shorten up some long lines */
2479#define INTREG I40E_PFINT_DYN_CTLN
2480
2481/* The act of updating the ITR will cause it to immediately trigger. In order
2482 * to prevent this from throwing off adaptive update statistics we defer the
2483 * update so that it can only happen so often. So after either Tx or Rx are
2484 * updated we make the adaptive scheme wait until either the ITR completely
2485 * expires via the next_update expiration or we have been through at least
2486 * 3 interrupts.
2487 */
2488#define ITR_COUNTDOWN_START 3
2489
2490/**
2491 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2492 * @vsi: the VSI we care about
2493 * @q_vector: q_vector for which itr is being updated and interrupt enabled
2494 *
2495 **/
2496static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2497					  struct i40e_q_vector *q_vector)
2498{
2499	struct i40e_hw *hw = &vsi->back->hw;
2500	u32 intval;
2501
2502	/* If we don't have MSIX, then we only need to re-enable icr0 */
2503	if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) {
2504		i40e_irq_dynamic_enable_icr0(vsi->back);
2505		return;
2506	}
2507
2508	/* These will do nothing if dynamic updates are not enabled */
2509	i40e_update_itr(q_vector, &q_vector->tx);
2510	i40e_update_itr(q_vector, &q_vector->rx);
2511
2512	/* This block of logic allows us to get away with only updating
2513	 * one ITR value with each interrupt. The idea is to perform a
2514	 * pseudo-lazy update with the following criteria.
2515	 *
2516	 * 1. Rx is given higher priority than Tx if both are in same state
2517	 * 2. If we must reduce an ITR that is given highest priority.
2518	 * 3. We then give priority to increasing ITR based on amount.
2519	 */
2520	if (q_vector->rx.target_itr < q_vector->rx.current_itr) {
2521		/* Rx ITR needs to be reduced, this is highest priority */
2522		intval = i40e_buildreg_itr(I40E_RX_ITR,
2523					   q_vector->rx.target_itr);
2524		q_vector->rx.current_itr = q_vector->rx.target_itr;
2525		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2526	} else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) ||
2527		   ((q_vector->rx.target_itr - q_vector->rx.current_itr) <
2528		    (q_vector->tx.target_itr - q_vector->tx.current_itr))) {
2529		/* Tx ITR needs to be reduced, this is second priority
2530		 * Tx ITR needs to be increased more than Rx, fourth priority
2531		 */
2532		intval = i40e_buildreg_itr(I40E_TX_ITR,
2533					   q_vector->tx.target_itr);
2534		q_vector->tx.current_itr = q_vector->tx.target_itr;
2535		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2536	} else if (q_vector->rx.current_itr != q_vector->rx.target_itr) {
2537		/* Rx ITR needs to be increased, third priority */
2538		intval = i40e_buildreg_itr(I40E_RX_ITR,
2539					   q_vector->rx.target_itr);
2540		q_vector->rx.current_itr = q_vector->rx.target_itr;
2541		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2542	} else {
2543		/* No ITR update, lowest priority */
2544		intval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2545		if (q_vector->itr_countdown)
2546			q_vector->itr_countdown--;
2547	}
2548
2549	if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2550		wr32(hw, INTREG(q_vector->reg_idx), intval);
2551}
2552
2553/**
2554 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2555 * @napi: napi struct with our devices info in it
2556 * @budget: amount of work driver is allowed to do this pass, in packets
2557 *
2558 * This function will clean all queues associated with a q_vector.
2559 *
2560 * Returns the amount of work done
2561 **/
2562int i40e_napi_poll(struct napi_struct *napi, int budget)
2563{
2564	struct i40e_q_vector *q_vector =
2565			       container_of(napi, struct i40e_q_vector, napi);
2566	struct i40e_vsi *vsi = q_vector->vsi;
2567	struct i40e_ring *ring;
2568	bool clean_complete = true;
2569	bool arm_wb = false;
2570	int budget_per_ring;
2571	int work_done = 0;
2572
2573	if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2574		napi_complete(napi);
2575		return 0;
2576	}
2577
2578	/* Since the actual Tx work is minimal, we can give the Tx a larger
2579	 * budget and be more aggressive about cleaning up the Tx descriptors.
2580	 */
2581	i40e_for_each_ring(ring, q_vector->tx) {
2582		bool wd = ring->xsk_umem ?
2583			  i40e_clean_xdp_tx_irq(vsi, ring) :
2584			  i40e_clean_tx_irq(vsi, ring, budget);
2585
2586		if (!wd) {
2587			clean_complete = false;
2588			continue;
2589		}
2590		arm_wb |= ring->arm_wb;
2591		ring->arm_wb = false;
2592	}
2593
2594	/* Handle case where we are called by netpoll with a budget of 0 */
2595	if (budget <= 0)
2596		goto tx_only;
2597
2598	/* normally we have 1 Rx ring per q_vector */
2599	if (unlikely(q_vector->num_ringpairs > 1))
2600		/* We attempt to distribute budget to each Rx queue fairly, but
2601		 * don't allow the budget to go below 1 because that would exit
2602		 * polling early.
2603		 */
2604		budget_per_ring = max_t(int, budget / q_vector->num_ringpairs, 1);
2605	else
2606		/* Max of 1 Rx ring in this q_vector so give it the budget */
2607		budget_per_ring = budget;
2608
2609	i40e_for_each_ring(ring, q_vector->rx) {
2610		int cleaned = ring->xsk_umem ?
2611			      i40e_clean_rx_irq_zc(ring, budget_per_ring) :
2612			      i40e_clean_rx_irq(ring, budget_per_ring);
2613
2614		work_done += cleaned;
2615		/* if we clean as many as budgeted, we must not be done */
2616		if (cleaned >= budget_per_ring)
2617			clean_complete = false;
2618	}
2619
2620	/* If work not completed, return budget and polling will return */
2621	if (!clean_complete) {
2622		int cpu_id = smp_processor_id();
2623
2624		/* It is possible that the interrupt affinity has changed but,
2625		 * if the cpu is pegged at 100%, polling will never exit while
2626		 * traffic continues and the interrupt will be stuck on this
2627		 * cpu.  We check to make sure affinity is correct before we
2628		 * continue to poll, otherwise we must stop polling so the
2629		 * interrupt can move to the correct cpu.
2630		 */
2631		if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
2632			/* Tell napi that we are done polling */
2633			napi_complete_done(napi, work_done);
2634
2635			/* Force an interrupt */
2636			i40e_force_wb(vsi, q_vector);
2637
2638			/* Return budget-1 so that polling stops */
2639			return budget - 1;
2640		}
2641tx_only:
2642		if (arm_wb) {
2643			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2644			i40e_enable_wb_on_itr(vsi, q_vector);
2645		}
2646		return budget;
2647	}
2648
2649	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2650		q_vector->arm_wb_state = false;
2651
2652	/* Exit the polling mode, but don't re-enable interrupts if stack might
2653	 * poll us due to busy-polling
2654	 */
2655	if (likely(napi_complete_done(napi, work_done)))
2656		i40e_update_enable_itr(vsi, q_vector);
2657
2658	return min(work_done, budget - 1);
2659}
2660
2661/**
2662 * i40e_atr - Add a Flow Director ATR filter
2663 * @tx_ring:  ring to add programming descriptor to
2664 * @skb:      send buffer
2665 * @tx_flags: send tx flags
2666 **/
2667static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2668		     u32 tx_flags)
2669{
2670	struct i40e_filter_program_desc *fdir_desc;
2671	struct i40e_pf *pf = tx_ring->vsi->back;
2672	union {
2673		unsigned char *network;
2674		struct iphdr *ipv4;
2675		struct ipv6hdr *ipv6;
2676	} hdr;
2677	struct tcphdr *th;
2678	unsigned int hlen;
2679	u32 flex_ptype, dtype_cmd;
2680	int l4_proto;
2681	u16 i;
2682
2683	/* make sure ATR is enabled */
2684	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2685		return;
2686
2687	if (test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2688		return;
2689
2690	/* if sampling is disabled do nothing */
2691	if (!tx_ring->atr_sample_rate)
2692		return;
2693
2694	/* Currently only IPv4/IPv6 with TCP is supported */
2695	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2696		return;
2697
2698	/* snag network header to get L4 type and address */
2699	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2700		      skb_inner_network_header(skb) : skb_network_header(skb);
2701
2702	/* Note: tx_flags gets modified to reflect inner protocols in
2703	 * tx_enable_csum function if encap is enabled.
2704	 */
2705	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2706		/* access ihl as u8 to avoid unaligned access on ia64 */
2707		hlen = (hdr.network[0] & 0x0F) << 2;
2708		l4_proto = hdr.ipv4->protocol;
2709	} else {
2710		/* find the start of the innermost ipv6 header */
2711		unsigned int inner_hlen = hdr.network - skb->data;
2712		unsigned int h_offset = inner_hlen;
2713
2714		/* this function updates h_offset to the end of the header */
2715		l4_proto =
2716		  ipv6_find_hdr(skb, &h_offset, IPPROTO_TCP, NULL, NULL);
2717		/* hlen will contain our best estimate of the tcp header */
2718		hlen = h_offset - inner_hlen;
2719	}
2720
2721	if (l4_proto != IPPROTO_TCP)
2722		return;
2723
2724	th = (struct tcphdr *)(hdr.network + hlen);
2725
2726	/* Due to lack of space, no more new filters can be programmed */
2727	if (th->syn && test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2728		return;
2729	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED) {
2730		/* HW ATR eviction will take care of removing filters on FIN
2731		 * and RST packets.
2732		 */
2733		if (th->fin || th->rst)
2734			return;
2735	}
2736
2737	tx_ring->atr_count++;
2738
2739	/* sample on all syn/fin/rst packets or once every atr sample rate */
2740	if (!th->fin &&
2741	    !th->syn &&
2742	    !th->rst &&
2743	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2744		return;
2745
2746	tx_ring->atr_count = 0;
2747
2748	/* grab the next descriptor */
2749	i = tx_ring->next_to_use;
2750	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2751
2752	i++;
2753	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2754
2755	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2756		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2757	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2758		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2759		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2760		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2761		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2762
2763	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2764
2765	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2766
2767	dtype_cmd |= (th->fin || th->rst) ?
2768		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2769		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2770		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2771		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2772
2773	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2774		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2775
2776	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2777		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2778
2779	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2780	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2781		dtype_cmd |=
2782			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2783			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2784			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2785	else
2786		dtype_cmd |=
2787			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2788			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2789			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2790
2791	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED)
2792		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2793
2794	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2795	fdir_desc->rsvd = cpu_to_le32(0);
2796	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2797	fdir_desc->fd_id = cpu_to_le32(0);
2798}
2799
2800/**
2801 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2802 * @skb:     send buffer
2803 * @tx_ring: ring to send buffer on
2804 * @flags:   the tx flags to be set
2805 *
2806 * Checks the skb and set up correspondingly several generic transmit flags
2807 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2808 *
2809 * Returns error code indicate the frame should be dropped upon error and the
2810 * otherwise  returns 0 to indicate the flags has been set properly.
2811 **/
2812static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2813					     struct i40e_ring *tx_ring,
2814					     u32 *flags)
2815{
2816	__be16 protocol = skb->protocol;
2817	u32  tx_flags = 0;
2818
2819	if (protocol == htons(ETH_P_8021Q) &&
2820	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2821		/* When HW VLAN acceleration is turned off by the user the
2822		 * stack sets the protocol to 8021q so that the driver
2823		 * can take any steps required to support the SW only
2824		 * VLAN handling.  In our case the driver doesn't need
2825		 * to take any further steps so just set the protocol
2826		 * to the encapsulated ethertype.
2827		 */
2828		skb->protocol = vlan_get_protocol(skb);
2829		goto out;
2830	}
2831
2832	/* if we have a HW VLAN tag being added, default to the HW one */
2833	if (skb_vlan_tag_present(skb)) {
2834		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2835		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2836	/* else if it is a SW VLAN, check the next protocol and store the tag */
2837	} else if (protocol == htons(ETH_P_8021Q)) {
2838		struct vlan_hdr *vhdr, _vhdr;
2839
2840		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2841		if (!vhdr)
2842			return -EINVAL;
2843
2844		protocol = vhdr->h_vlan_encapsulated_proto;
2845		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2846		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2847	}
2848
2849	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2850		goto out;
2851
2852	/* Insert 802.1p priority into VLAN header */
2853	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2854	    (skb->priority != TC_PRIO_CONTROL)) {
2855		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2856		tx_flags |= (skb->priority & 0x7) <<
2857				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2858		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2859			struct vlan_ethhdr *vhdr;
2860			int rc;
2861
2862			rc = skb_cow_head(skb, 0);
2863			if (rc < 0)
2864				return rc;
2865			vhdr = (struct vlan_ethhdr *)skb->data;
2866			vhdr->h_vlan_TCI = htons(tx_flags >>
2867						 I40E_TX_FLAGS_VLAN_SHIFT);
2868		} else {
2869			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2870		}
2871	}
2872
2873out:
2874	*flags = tx_flags;
2875	return 0;
2876}
2877
2878/**
2879 * i40e_tso - set up the tso context descriptor
2880 * @first:    pointer to first Tx buffer for xmit
2881 * @hdr_len:  ptr to the size of the packet header
2882 * @cd_type_cmd_tso_mss: Quad Word 1
2883 *
2884 * Returns 0 if no TSO can happen, 1 if tso is going, or error
2885 **/
2886static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
2887		    u64 *cd_type_cmd_tso_mss)
2888{
2889	struct sk_buff *skb = first->skb;
2890	u64 cd_cmd, cd_tso_len, cd_mss;
2891	union {
2892		struct iphdr *v4;
2893		struct ipv6hdr *v6;
2894		unsigned char *hdr;
2895	} ip;
2896	union {
2897		struct tcphdr *tcp;
2898		struct udphdr *udp;
2899		unsigned char *hdr;
2900	} l4;
2901	u32 paylen, l4_offset;
2902	u16 gso_segs, gso_size;
2903	int err;
2904
2905	if (skb->ip_summed != CHECKSUM_PARTIAL)
2906		return 0;
2907
2908	if (!skb_is_gso(skb))
2909		return 0;
2910
2911	err = skb_cow_head(skb, 0);
2912	if (err < 0)
2913		return err;
2914
2915	ip.hdr = skb_network_header(skb);
2916	l4.hdr = skb_transport_header(skb);
2917
2918	/* initialize outer IP header fields */
2919	if (ip.v4->version == 4) {
2920		ip.v4->tot_len = 0;
2921		ip.v4->check = 0;
2922	} else {
2923		ip.v6->payload_len = 0;
2924	}
2925
2926	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2927					 SKB_GSO_GRE_CSUM |
2928					 SKB_GSO_IPXIP4 |
2929					 SKB_GSO_IPXIP6 |
2930					 SKB_GSO_UDP_TUNNEL |
2931					 SKB_GSO_UDP_TUNNEL_CSUM)) {
2932		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2933		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2934			l4.udp->len = 0;
2935
2936			/* determine offset of outer transport header */
2937			l4_offset = l4.hdr - skb->data;
2938
2939			/* remove payload length from outer checksum */
2940			paylen = skb->len - l4_offset;
2941			csum_replace_by_diff(&l4.udp->check,
2942					     (__force __wsum)htonl(paylen));
2943		}
2944
2945		/* reset pointers to inner headers */
2946		ip.hdr = skb_inner_network_header(skb);
2947		l4.hdr = skb_inner_transport_header(skb);
2948
2949		/* initialize inner IP header fields */
2950		if (ip.v4->version == 4) {
2951			ip.v4->tot_len = 0;
2952			ip.v4->check = 0;
2953		} else {
2954			ip.v6->payload_len = 0;
2955		}
2956	}
2957
2958	/* determine offset of inner transport header */
2959	l4_offset = l4.hdr - skb->data;
2960
2961	/* remove payload length from inner checksum */
2962	paylen = skb->len - l4_offset;
2963
2964	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
2965		csum_replace_by_diff(&l4.udp->check, (__force __wsum)htonl(paylen));
2966		/* compute length of segmentation header */
2967		*hdr_len = sizeof(*l4.udp) + l4_offset;
2968	} else {
2969		csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2970		/* compute length of segmentation header */
2971		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
2972	}
2973
2974	/* pull values out of skb_shinfo */
2975	gso_size = skb_shinfo(skb)->gso_size;
2976	gso_segs = skb_shinfo(skb)->gso_segs;
2977
2978	/* update GSO size and bytecount with header size */
2979	first->gso_segs = gso_segs;
2980	first->bytecount += (first->gso_segs - 1) * *hdr_len;
2981
2982	/* find the field values */
2983	cd_cmd = I40E_TX_CTX_DESC_TSO;
2984	cd_tso_len = skb->len - *hdr_len;
2985	cd_mss = gso_size;
2986	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2987				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2988				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2989	return 1;
2990}
2991
2992/**
2993 * i40e_tsyn - set up the tsyn context descriptor
2994 * @tx_ring:  ptr to the ring to send
2995 * @skb:      ptr to the skb we're sending
2996 * @tx_flags: the collected send information
2997 * @cd_type_cmd_tso_mss: Quad Word 1
2998 *
2999 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
3000 **/
3001static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
3002		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
3003{
3004	struct i40e_pf *pf;
3005
3006	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
3007		return 0;
3008
3009	/* Tx timestamps cannot be sampled when doing TSO */
3010	if (tx_flags & I40E_TX_FLAGS_TSO)
3011		return 0;
3012
3013	/* only timestamp the outbound packet if the user has requested it and
3014	 * we are not already transmitting a packet to be timestamped
3015	 */
3016	pf = i40e_netdev_to_pf(tx_ring->netdev);
3017	if (!(pf->flags & I40E_FLAG_PTP))
3018		return 0;
3019
3020	if (pf->ptp_tx &&
3021	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
3022		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3023		pf->ptp_tx_start = jiffies;
3024		pf->ptp_tx_skb = skb_get(skb);
3025	} else {
3026		pf->tx_hwtstamp_skipped++;
3027		return 0;
3028	}
3029
3030	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
3031				I40E_TXD_CTX_QW1_CMD_SHIFT;
3032
3033	return 1;
3034}
3035
3036/**
3037 * i40e_tx_enable_csum - Enable Tx checksum offloads
3038 * @skb: send buffer
3039 * @tx_flags: pointer to Tx flags currently set
3040 * @td_cmd: Tx descriptor command bits to set
3041 * @td_offset: Tx descriptor header offsets to set
3042 * @tx_ring: Tx descriptor ring
3043 * @cd_tunneling: ptr to context desc bits
3044 **/
3045static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
3046			       u32 *td_cmd, u32 *td_offset,
3047			       struct i40e_ring *tx_ring,
3048			       u32 *cd_tunneling)
3049{
3050	union {
3051		struct iphdr *v4;
3052		struct ipv6hdr *v6;
3053		unsigned char *hdr;
3054	} ip;
3055	union {
3056		struct tcphdr *tcp;
3057		struct udphdr *udp;
3058		unsigned char *hdr;
3059	} l4;
3060	unsigned char *exthdr;
3061	u32 offset, cmd = 0;
3062	__be16 frag_off;
3063	u8 l4_proto = 0;
3064
3065	if (skb->ip_summed != CHECKSUM_PARTIAL)
3066		return 0;
3067
3068	ip.hdr = skb_network_header(skb);
3069	l4.hdr = skb_transport_header(skb);
3070
3071	/* compute outer L2 header size */
3072	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
3073
3074	if (skb->encapsulation) {
3075		u32 tunnel = 0;
3076		/* define outer network header type */
3077		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3078			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3079				  I40E_TX_CTX_EXT_IP_IPV4 :
3080				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
3081
3082			l4_proto = ip.v4->protocol;
3083		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3084			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
3085
3086			exthdr = ip.hdr + sizeof(*ip.v6);
3087			l4_proto = ip.v6->nexthdr;
3088			if (l4.hdr != exthdr)
3089				ipv6_skip_exthdr(skb, exthdr - skb->data,
3090						 &l4_proto, &frag_off);
3091		}
3092
3093		/* define outer transport */
3094		switch (l4_proto) {
3095		case IPPROTO_UDP:
3096			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
3097			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3098			break;
3099		case IPPROTO_GRE:
3100			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
3101			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3102			break;
3103		case IPPROTO_IPIP:
3104		case IPPROTO_IPV6:
3105			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3106			l4.hdr = skb_inner_network_header(skb);
3107			break;
3108		default:
3109			if (*tx_flags & I40E_TX_FLAGS_TSO)
3110				return -1;
3111
3112			skb_checksum_help(skb);
3113			return 0;
3114		}
3115
3116		/* compute outer L3 header size */
3117		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
3118			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
3119
3120		/* switch IP header pointer from outer to inner header */
3121		ip.hdr = skb_inner_network_header(skb);
3122
3123		/* compute tunnel header size */
3124		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
3125			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
3126
3127		/* indicate if we need to offload outer UDP header */
3128		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
3129		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3130		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
3131			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
3132
3133		/* record tunnel offload values */
3134		*cd_tunneling |= tunnel;
3135
3136		/* switch L4 header pointer from outer to inner */
3137		l4.hdr = skb_inner_transport_header(skb);
3138		l4_proto = 0;
3139
3140		/* reset type as we transition from outer to inner headers */
3141		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
3142		if (ip.v4->version == 4)
3143			*tx_flags |= I40E_TX_FLAGS_IPV4;
3144		if (ip.v6->version == 6)
3145			*tx_flags |= I40E_TX_FLAGS_IPV6;
3146	}
3147
3148	/* Enable IP checksum offloads */
3149	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3150		l4_proto = ip.v4->protocol;
3151		/* the stack computes the IP header already, the only time we
3152		 * need the hardware to recompute it is in the case of TSO.
3153		 */
3154		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3155		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
3156		       I40E_TX_DESC_CMD_IIPT_IPV4;
3157	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3158		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
3159
3160		exthdr = ip.hdr + sizeof(*ip.v6);
3161		l4_proto = ip.v6->nexthdr;
3162		if (l4.hdr != exthdr)
3163			ipv6_skip_exthdr(skb, exthdr - skb->data,
3164					 &l4_proto, &frag_off);
3165	}
3166
3167	/* compute inner L3 header size */
3168	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
3169
3170	/* Enable L4 checksum offloads */
3171	switch (l4_proto) {
3172	case IPPROTO_TCP:
3173		/* enable checksum offloads */
3174		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
3175		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3176		break;
3177	case IPPROTO_SCTP:
3178		/* enable SCTP checksum offload */
3179		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
3180		offset |= (sizeof(struct sctphdr) >> 2) <<
3181			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3182		break;
3183	case IPPROTO_UDP:
3184		/* enable UDP checksum offload */
3185		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
3186		offset |= (sizeof(struct udphdr) >> 2) <<
3187			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3188		break;
3189	default:
3190		if (*tx_flags & I40E_TX_FLAGS_TSO)
3191			return -1;
3192		skb_checksum_help(skb);
3193		return 0;
3194	}
3195
3196	*td_cmd |= cmd;
3197	*td_offset |= offset;
3198
3199	return 1;
3200}
3201
3202/**
3203 * i40e_create_tx_ctx Build the Tx context descriptor
3204 * @tx_ring:  ring to create the descriptor on
3205 * @cd_type_cmd_tso_mss: Quad Word 1
3206 * @cd_tunneling: Quad Word 0 - bits 0-31
3207 * @cd_l2tag2: Quad Word 0 - bits 32-63
3208 **/
3209static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
3210			       const u64 cd_type_cmd_tso_mss,
3211			       const u32 cd_tunneling, const u32 cd_l2tag2)
3212{
3213	struct i40e_tx_context_desc *context_desc;
3214	int i = tx_ring->next_to_use;
3215
3216	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
3217	    !cd_tunneling && !cd_l2tag2)
3218		return;
3219
3220	/* grab the next descriptor */
3221	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
3222
3223	i++;
3224	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3225
3226	/* cpu_to_le32 and assign to struct fields */
3227	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
3228	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
3229	context_desc->rsvd = cpu_to_le16(0);
3230	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
3231}
3232
3233/**
3234 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
3235 * @tx_ring: the ring to be checked
3236 * @size:    the size buffer we want to assure is available
3237 *
3238 * Returns -EBUSY if a stop is needed, else 0
3239 **/
3240int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
3241{
3242	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3243	/* Memory barrier before checking head and tail */
3244	smp_mb();
3245
3246	/* Check again in a case another CPU has just made room available. */
3247	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
3248		return -EBUSY;
3249
3250	/* A reprieve! - use start_queue because it doesn't call schedule */
3251	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3252	++tx_ring->tx_stats.restart_queue;
3253	return 0;
3254}
3255
3256/**
3257 * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
3258 * @skb:      send buffer
3259 *
3260 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
3261 * and so we need to figure out the cases where we need to linearize the skb.
3262 *
3263 * For TSO we need to count the TSO header and segment payload separately.
3264 * As such we need to check cases where we have 7 fragments or more as we
3265 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
3266 * the segment payload in the first descriptor, and another 7 for the
3267 * fragments.
3268 **/
3269bool __i40e_chk_linearize(struct sk_buff *skb)
3270{
3271	const skb_frag_t *frag, *stale;
3272	int nr_frags, sum;
3273
3274	/* no need to check if number of frags is less than 7 */
3275	nr_frags = skb_shinfo(skb)->nr_frags;
3276	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
3277		return false;
3278
3279	/* We need to walk through the list and validate that each group
3280	 * of 6 fragments totals at least gso_size.
3281	 */
3282	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
3283	frag = &skb_shinfo(skb)->frags[0];
3284
3285	/* Initialize size to the negative value of gso_size minus 1.  We
3286	 * use this as the worst case scenerio in which the frag ahead
3287	 * of us only provides one byte which is why we are limited to 6
3288	 * descriptors for a single transmit as the header and previous
3289	 * fragment are already consuming 2 descriptors.
3290	 */
3291	sum = 1 - skb_shinfo(skb)->gso_size;
3292
3293	/* Add size of frags 0 through 4 to create our initial sum */
3294	sum += skb_frag_size(frag++);
3295	sum += skb_frag_size(frag++);
3296	sum += skb_frag_size(frag++);
3297	sum += skb_frag_size(frag++);
3298	sum += skb_frag_size(frag++);
3299
3300	/* Walk through fragments adding latest fragment, testing it, and
3301	 * then removing stale fragments from the sum.
3302	 */
3303	for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
3304		int stale_size = skb_frag_size(stale);
3305
3306		sum += skb_frag_size(frag++);
3307
3308		/* The stale fragment may present us with a smaller
3309		 * descriptor than the actual fragment size. To account
3310		 * for that we need to remove all the data on the front and
3311		 * figure out what the remainder would be in the last
3312		 * descriptor associated with the fragment.
3313		 */
3314		if (stale_size > I40E_MAX_DATA_PER_TXD) {
3315			int align_pad = -(skb_frag_off(stale)) &
3316					(I40E_MAX_READ_REQ_SIZE - 1);
3317
3318			sum -= align_pad;
3319			stale_size -= align_pad;
3320
3321			do {
3322				sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3323				stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3324			} while (stale_size > I40E_MAX_DATA_PER_TXD);
3325		}
3326
3327		/* if sum is negative we failed to make sufficient progress */
3328		if (sum < 0)
3329			return true;
3330
3331		if (!nr_frags--)
3332			break;
3333
3334		sum -= stale_size;
3335	}
3336
3337	return false;
3338}
3339
3340/**
3341 * i40e_tx_map - Build the Tx descriptor
3342 * @tx_ring:  ring to send buffer on
3343 * @skb:      send buffer
3344 * @first:    first buffer info buffer to use
3345 * @tx_flags: collected send information
3346 * @hdr_len:  size of the packet header
3347 * @td_cmd:   the command field in the descriptor
3348 * @td_offset: offset for checksum or crc
3349 *
3350 * Returns 0 on success, -1 on failure to DMA
3351 **/
3352static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
3353			      struct i40e_tx_buffer *first, u32 tx_flags,
3354			      const u8 hdr_len, u32 td_cmd, u32 td_offset)
3355{
3356	unsigned int data_len = skb->data_len;
3357	unsigned int size = skb_headlen(skb);
3358	skb_frag_t *frag;
3359	struct i40e_tx_buffer *tx_bi;
3360	struct i40e_tx_desc *tx_desc;
3361	u16 i = tx_ring->next_to_use;
3362	u32 td_tag = 0;
3363	dma_addr_t dma;
3364	u16 desc_count = 1;
3365
3366	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
3367		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
3368		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
3369			 I40E_TX_FLAGS_VLAN_SHIFT;
3370	}
3371
3372	first->tx_flags = tx_flags;
3373
3374	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3375
3376	tx_desc = I40E_TX_DESC(tx_ring, i);
3377	tx_bi = first;
3378
3379	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3380		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3381
3382		if (dma_mapping_error(tx_ring->dev, dma))
3383			goto dma_error;
3384
3385		/* record length, and DMA address */
3386		dma_unmap_len_set(tx_bi, len, size);
3387		dma_unmap_addr_set(tx_bi, dma, dma);
3388
3389		/* align size to end of page */
3390		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
3391		tx_desc->buffer_addr = cpu_to_le64(dma);
3392
3393		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
3394			tx_desc->cmd_type_offset_bsz =
3395				build_ctob(td_cmd, td_offset,
3396					   max_data, td_tag);
3397
3398			tx_desc++;
3399			i++;
3400			desc_count++;
3401
3402			if (i == tx_ring->count) {
3403				tx_desc = I40E_TX_DESC(tx_ring, 0);
3404				i = 0;
3405			}
3406
3407			dma += max_data;
3408			size -= max_data;
3409
3410			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3411			tx_desc->buffer_addr = cpu_to_le64(dma);
3412		}
3413
3414		if (likely(!data_len))
3415			break;
3416
3417		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3418							  size, td_tag);
3419
3420		tx_desc++;
3421		i++;
3422		desc_count++;
3423
3424		if (i == tx_ring->count) {
3425			tx_desc = I40E_TX_DESC(tx_ring, 0);
3426			i = 0;
3427		}
3428
3429		size = skb_frag_size(frag);
3430		data_len -= size;
3431
3432		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3433				       DMA_TO_DEVICE);
3434
3435		tx_bi = &tx_ring->tx_bi[i];
3436	}
3437
3438	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3439
3440	i++;
3441	if (i == tx_ring->count)
3442		i = 0;
3443
3444	tx_ring->next_to_use = i;
3445
3446	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3447
3448	/* write last descriptor with EOP bit */
3449	td_cmd |= I40E_TX_DESC_CMD_EOP;
3450
3451	/* We OR these values together to check both against 4 (WB_STRIDE)
3452	 * below. This is safe since we don't re-use desc_count afterwards.
3453	 */
3454	desc_count |= ++tx_ring->packet_stride;
3455
3456	if (desc_count >= WB_STRIDE) {
3457		/* write last descriptor with RS bit set */
3458		td_cmd |= I40E_TX_DESC_CMD_RS;
3459		tx_ring->packet_stride = 0;
3460	}
3461
3462	tx_desc->cmd_type_offset_bsz =
3463			build_ctob(td_cmd, td_offset, size, td_tag);
3464
3465	skb_tx_timestamp(skb);
3466
3467	/* Force memory writes to complete before letting h/w know there
3468	 * are new descriptors to fetch.
3469	 *
3470	 * We also use this memory barrier to make certain all of the
3471	 * status bits have been updated before next_to_watch is written.
3472	 */
3473	wmb();
3474
3475	/* set next_to_watch value indicating a packet is present */
3476	first->next_to_watch = tx_desc;
3477
3478	/* notify HW of packet */
3479	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
3480		writel(i, tx_ring->tail);
3481	}
3482
3483	return 0;
3484
3485dma_error:
3486	dev_info(tx_ring->dev, "TX DMA map failed\n");
3487
3488	/* clear dma mappings for failed tx_bi map */
3489	for (;;) {
3490		tx_bi = &tx_ring->tx_bi[i];
3491		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3492		if (tx_bi == first)
3493			break;
3494		if (i == 0)
3495			i = tx_ring->count;
3496		i--;
3497	}
3498
3499	tx_ring->next_to_use = i;
3500
3501	return -1;
3502}
3503
3504/**
3505 * i40e_xmit_xdp_ring - transmits an XDP buffer to an XDP Tx ring
3506 * @xdp: data to transmit
3507 * @xdp_ring: XDP Tx ring
3508 **/
3509static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
3510			      struct i40e_ring *xdp_ring)
3511{
3512	u16 i = xdp_ring->next_to_use;
3513	struct i40e_tx_buffer *tx_bi;
3514	struct i40e_tx_desc *tx_desc;
3515	void *data = xdpf->data;
3516	u32 size = xdpf->len;
3517	dma_addr_t dma;
3518
3519	if (!unlikely(I40E_DESC_UNUSED(xdp_ring))) {
3520		xdp_ring->tx_stats.tx_busy++;
3521		return I40E_XDP_CONSUMED;
3522	}
3523	dma = dma_map_single(xdp_ring->dev, data, size, DMA_TO_DEVICE);
3524	if (dma_mapping_error(xdp_ring->dev, dma))
3525		return I40E_XDP_CONSUMED;
3526
3527	tx_bi = &xdp_ring->tx_bi[i];
3528	tx_bi->bytecount = size;
3529	tx_bi->gso_segs = 1;
3530	tx_bi->xdpf = xdpf;
3531
3532	/* record length, and DMA address */
3533	dma_unmap_len_set(tx_bi, len, size);
3534	dma_unmap_addr_set(tx_bi, dma, dma);
3535
3536	tx_desc = I40E_TX_DESC(xdp_ring, i);
3537	tx_desc->buffer_addr = cpu_to_le64(dma);
3538	tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC
3539						  | I40E_TXD_CMD,
3540						  0, size, 0);
3541
3542	/* Make certain all of the status bits have been updated
3543	 * before next_to_watch is written.
3544	 */
3545	smp_wmb();
3546
3547	xdp_ring->xdp_tx_active++;
3548	i++;
3549	if (i == xdp_ring->count)
3550		i = 0;
3551
3552	tx_bi->next_to_watch = tx_desc;
3553	xdp_ring->next_to_use = i;
3554
3555	return I40E_XDP_TX;
3556}
3557
3558/**
3559 * i40e_xmit_frame_ring - Sends buffer on Tx ring
3560 * @skb:     send buffer
3561 * @tx_ring: ring to send buffer on
3562 *
3563 * Returns NETDEV_TX_OK if sent, else an error code
3564 **/
3565static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3566					struct i40e_ring *tx_ring)
3567{
3568	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3569	u32 cd_tunneling = 0, cd_l2tag2 = 0;
3570	struct i40e_tx_buffer *first;
3571	u32 td_offset = 0;
3572	u32 tx_flags = 0;
3573	__be16 protocol;
3574	u32 td_cmd = 0;
3575	u8 hdr_len = 0;
3576	int tso, count;
3577	int tsyn;
3578
3579	/* prefetch the data, we'll need it later */
3580	prefetch(skb->data);
3581
3582	i40e_trace(xmit_frame_ring, skb, tx_ring);
3583
3584	count = i40e_xmit_descriptor_count(skb);
3585	if (i40e_chk_linearize(skb, count)) {
3586		if (__skb_linearize(skb)) {
3587			dev_kfree_skb_any(skb);
3588			return NETDEV_TX_OK;
3589		}
3590		count = i40e_txd_use_count(skb->len);
3591		tx_ring->tx_stats.tx_linearize++;
3592	}
3593
3594	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3595	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3596	 *       + 4 desc gap to avoid the cache line where head is,
3597	 *       + 1 desc for context descriptor,
3598	 * otherwise try next time
3599	 */
3600	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3601		tx_ring->tx_stats.tx_busy++;
3602		return NETDEV_TX_BUSY;
3603	}
3604
3605	/* record the location of the first descriptor for this packet */
3606	first = &tx_ring->tx_bi[tx_ring->next_to_use];
3607	first->skb = skb;
3608	first->bytecount = skb->len;
3609	first->gso_segs = 1;
3610
3611	/* prepare the xmit flags */
3612	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3613		goto out_drop;
3614
3615	/* obtain protocol of skb */
3616	protocol = vlan_get_protocol(skb);
3617
3618	/* setup IPv4/IPv6 offloads */
3619	if (protocol == htons(ETH_P_IP))
3620		tx_flags |= I40E_TX_FLAGS_IPV4;
3621	else if (protocol == htons(ETH_P_IPV6))
3622		tx_flags |= I40E_TX_FLAGS_IPV6;
3623
3624	tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3625
3626	if (tso < 0)
3627		goto out_drop;
3628	else if (tso)
3629		tx_flags |= I40E_TX_FLAGS_TSO;
3630
3631	/* Always offload the checksum, since it's in the data descriptor */
3632	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3633				  tx_ring, &cd_tunneling);
3634	if (tso < 0)
3635		goto out_drop;
3636
3637	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3638
3639	if (tsyn)
3640		tx_flags |= I40E_TX_FLAGS_TSYN;
3641
3642	/* always enable CRC insertion offload */
3643	td_cmd |= I40E_TX_DESC_CMD_ICRC;
3644
3645	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3646			   cd_tunneling, cd_l2tag2);
3647
3648	/* Add Flow Director ATR if it's enabled.
3649	 *
3650	 * NOTE: this must always be directly before the data descriptor.
3651	 */
3652	i40e_atr(tx_ring, skb, tx_flags);
3653
3654	if (i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3655			td_cmd, td_offset))
3656		goto cleanup_tx_tstamp;
3657
3658	return NETDEV_TX_OK;
3659
3660out_drop:
3661	i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3662	dev_kfree_skb_any(first->skb);
3663	first->skb = NULL;
3664cleanup_tx_tstamp:
3665	if (unlikely(tx_flags & I40E_TX_FLAGS_TSYN)) {
3666		struct i40e_pf *pf = i40e_netdev_to_pf(tx_ring->netdev);
3667
3668		dev_kfree_skb_any(pf->ptp_tx_skb);
3669		pf->ptp_tx_skb = NULL;
3670		clear_bit_unlock(__I40E_PTP_TX_IN_PROGRESS, pf->state);
3671	}
3672
3673	return NETDEV_TX_OK;
3674}
3675
3676/**
3677 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3678 * @skb:    send buffer
3679 * @netdev: network interface device structure
3680 *
3681 * Returns NETDEV_TX_OK if sent, else an error code
3682 **/
3683netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3684{
3685	struct i40e_netdev_priv *np = netdev_priv(netdev);
3686	struct i40e_vsi *vsi = np->vsi;
3687	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3688
3689	/* hardware can't handle really short frames, hardware padding works
3690	 * beyond this point
3691	 */
3692	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3693		return NETDEV_TX_OK;
3694
3695	return i40e_xmit_frame_ring(skb, tx_ring);
3696}
3697
3698/**
3699 * i40e_xdp_xmit - Implements ndo_xdp_xmit
3700 * @dev: netdev
3701 * @xdp: XDP buffer
3702 *
3703 * Returns number of frames successfully sent. Frames that fail are
3704 * free'ed via XDP return API.
3705 *
3706 * For error cases, a negative errno code is returned and no-frames
3707 * are transmitted (caller must handle freeing frames).
3708 **/
3709int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
3710		  u32 flags)
3711{
3712	struct i40e_netdev_priv *np = netdev_priv(dev);
3713	unsigned int queue_index = smp_processor_id();
3714	struct i40e_vsi *vsi = np->vsi;
3715	struct i40e_pf *pf = vsi->back;
3716	struct i40e_ring *xdp_ring;
3717	int drops = 0;
3718	int i;
3719
3720	if (test_bit(__I40E_VSI_DOWN, vsi->state))
3721		return -ENETDOWN;
3722
3723	if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs ||
3724	    test_bit(__I40E_CONFIG_BUSY, pf->state))
3725		return -ENXIO;
3726
3727	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3728		return -EINVAL;
3729
3730	xdp_ring = vsi->xdp_rings[queue_index];
3731
3732	for (i = 0; i < n; i++) {
3733		struct xdp_frame *xdpf = frames[i];
3734		int err;
3735
3736		err = i40e_xmit_xdp_ring(xdpf, xdp_ring);
3737		if (err != I40E_XDP_TX) {
3738			xdp_return_frame_rx_napi(xdpf);
3739			drops++;
3740		}
3741	}
3742
3743	if (unlikely(flags & XDP_XMIT_FLUSH))
3744		i40e_xdp_ring_update_tail(xdp_ring);
3745
3746	return n - drops;
3747}