Linux Audio

Check our new training course

Loading...
v6.8
   1// SPDX-License-Identifier: GPL-2.0
   2/* Copyright(c) 2013 - 2018 Intel Corporation. */
   3
   4#include <linux/bpf_trace.h>
   5#include <linux/prefetch.h>
   6#include <linux/sctp.h>
   7#include <net/mpls.h>
   8#include <net/xdp.h>
   9#include "i40e_txrx_common.h"
  10#include "i40e_trace.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 = FIELD_PREP(I40E_TXD_FLTR_QW0_QINDEX_MASK, fdata->q_index);
 
  37
  38	flex_ptype |= FIELD_PREP(I40E_TXD_FLTR_QW0_FLEXOFF_MASK,
  39				 fdata->flex_off);
  40
  41	flex_ptype |= FIELD_PREP(I40E_TXD_FLTR_QW0_PCTYPE_MASK, fdata->pctype);
 
  42
  43	/* Use LAN VSI Id if not programmed by user */
  44	flex_ptype |= FIELD_PREP(I40E_TXD_FLTR_QW0_DEST_VSI_MASK,
  45				 fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id);
 
  46
  47	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
  48
  49	dtype_cmd |= add ?
  50		     I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
  51		     I40E_TXD_FLTR_QW1_PCMD_SHIFT :
  52		     I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
  53		     I40E_TXD_FLTR_QW1_PCMD_SHIFT;
  54
  55	dtype_cmd |= FIELD_PREP(I40E_TXD_FLTR_QW1_DEST_MASK, fdata->dest_ctl);
 
  56
  57	dtype_cmd |= FIELD_PREP(I40E_TXD_FLTR_QW1_FD_STATUS_MASK,
  58				fdata->fd_status);
  59
  60	if (fdata->cnt_index) {
  61		dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
  62		dtype_cmd |= FIELD_PREP(I40E_TXD_FLTR_QW1_CNTINDEX_MASK,
  63					fdata->cnt_index);
 
  64	}
  65
  66	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
  67	fdir_desc->rsvd = cpu_to_le32(0);
  68	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
  69	fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
  70}
  71
  72#define I40E_FD_CLEAN_DELAY 10
  73/**
  74 * i40e_program_fdir_filter - Program a Flow Director filter
  75 * @fdir_data: Packet data that will be filter parameters
  76 * @raw_packet: the pre-allocated packet buffer for FDir
  77 * @pf: The PF pointer
  78 * @add: True for add/update, False for remove
  79 **/
  80static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
  81				    u8 *raw_packet, struct i40e_pf *pf,
  82				    bool add)
  83{
  84	struct i40e_tx_buffer *tx_buf, *first;
  85	struct i40e_tx_desc *tx_desc;
  86	struct i40e_ring *tx_ring;
  87	struct i40e_vsi *vsi;
  88	struct device *dev;
  89	dma_addr_t dma;
  90	u32 td_cmd = 0;
  91	u16 i;
  92
  93	/* find existing FDIR VSI */
  94	vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
  95	if (!vsi)
  96		return -ENOENT;
  97
  98	tx_ring = vsi->tx_rings[0];
  99	dev = tx_ring->dev;
 100
 101	/* we need two descriptors to add/del a filter and we can wait */
 102	for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
 103		if (!i)
 104			return -EAGAIN;
 105		msleep_interruptible(1);
 106	}
 107
 108	dma = dma_map_single(dev, raw_packet,
 109			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
 110	if (dma_mapping_error(dev, dma))
 111		goto dma_fail;
 112
 113	/* grab the next descriptor */
 114	i = tx_ring->next_to_use;
 115	first = &tx_ring->tx_bi[i];
 116	i40e_fdir(tx_ring, fdir_data, add);
 117
 118	/* Now program a dummy descriptor */
 119	i = tx_ring->next_to_use;
 120	tx_desc = I40E_TX_DESC(tx_ring, i);
 121	tx_buf = &tx_ring->tx_bi[i];
 122
 123	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
 124
 125	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
 126
 127	/* record length, and DMA address */
 128	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
 129	dma_unmap_addr_set(tx_buf, dma, dma);
 130
 131	tx_desc->buffer_addr = cpu_to_le64(dma);
 132	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
 133
 134	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
 135	tx_buf->raw_buf = (void *)raw_packet;
 136
 137	tx_desc->cmd_type_offset_bsz =
 138		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
 139
 140	/* Force memory writes to complete before letting h/w
 141	 * know there are new descriptors to fetch.
 142	 */
 143	wmb();
 144
 145	/* Mark the data descriptor to be watched */
 146	first->next_to_watch = tx_desc;
 147
 148	writel(tx_ring->next_to_use, tx_ring->tail);
 149	return 0;
 150
 151dma_fail:
 152	return -1;
 153}
 154
 155/**
 156 * i40e_create_dummy_packet - Constructs dummy packet for HW
 157 * @dummy_packet: preallocated space for dummy packet
 158 * @ipv4: is layer 3 packet of version 4 or 6
 159 * @l4proto: next level protocol used in data portion of l3
 160 * @data: filter data
 161 *
 162 * Returns address of layer 4 protocol dummy packet.
 163 **/
 164static char *i40e_create_dummy_packet(u8 *dummy_packet, bool ipv4, u8 l4proto,
 165				      struct i40e_fdir_filter *data)
 166{
 167	bool is_vlan = !!data->vlan_tag;
 168	struct vlan_hdr vlan = {};
 169	struct ipv6hdr ipv6 = {};
 170	struct ethhdr eth = {};
 171	struct iphdr ip = {};
 172	u8 *tmp;
 173
 174	if (ipv4) {
 175		eth.h_proto = cpu_to_be16(ETH_P_IP);
 176		ip.protocol = l4proto;
 177		ip.version = 0x4;
 178		ip.ihl = 0x5;
 179
 180		ip.daddr = data->dst_ip;
 181		ip.saddr = data->src_ip;
 182	} else {
 183		eth.h_proto = cpu_to_be16(ETH_P_IPV6);
 184		ipv6.nexthdr = l4proto;
 185		ipv6.version = 0x6;
 186
 187		memcpy(&ipv6.saddr.in6_u.u6_addr32, data->src_ip6,
 188		       sizeof(__be32) * 4);
 189		memcpy(&ipv6.daddr.in6_u.u6_addr32, data->dst_ip6,
 190		       sizeof(__be32) * 4);
 191	}
 192
 193	if (is_vlan) {
 194		vlan.h_vlan_TCI = data->vlan_tag;
 195		vlan.h_vlan_encapsulated_proto = eth.h_proto;
 196		eth.h_proto = data->vlan_etype;
 197	}
 198
 199	tmp = dummy_packet;
 200	memcpy(tmp, &eth, sizeof(eth));
 201	tmp += sizeof(eth);
 202
 203	if (is_vlan) {
 204		memcpy(tmp, &vlan, sizeof(vlan));
 205		tmp += sizeof(vlan);
 206	}
 207
 208	if (ipv4) {
 209		memcpy(tmp, &ip, sizeof(ip));
 210		tmp += sizeof(ip);
 211	} else {
 212		memcpy(tmp, &ipv6, sizeof(ipv6));
 213		tmp += sizeof(ipv6);
 214	}
 215
 216	return tmp;
 217}
 218
 219/**
 220 * i40e_create_dummy_udp_packet - helper function to create UDP packet
 221 * @raw_packet: preallocated space for dummy packet
 222 * @ipv4: is layer 3 packet of version 4 or 6
 223 * @l4proto: next level protocol used in data portion of l3
 224 * @data: filter data
 225 *
 226 * Helper function to populate udp fields.
 227 **/
 228static void i40e_create_dummy_udp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
 229					 struct i40e_fdir_filter *data)
 230{
 231	struct udphdr *udp;
 232	u8 *tmp;
 233
 234	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_UDP, data);
 235	udp = (struct udphdr *)(tmp);
 236	udp->dest = data->dst_port;
 237	udp->source = data->src_port;
 238}
 239
 240/**
 241 * i40e_create_dummy_tcp_packet - helper function to create TCP packet
 242 * @raw_packet: preallocated space for dummy packet
 243 * @ipv4: is layer 3 packet of version 4 or 6
 244 * @l4proto: next level protocol used in data portion of l3
 245 * @data: filter data
 246 *
 247 * Helper function to populate tcp fields.
 248 **/
 249static void i40e_create_dummy_tcp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
 250					 struct i40e_fdir_filter *data)
 251{
 252	struct tcphdr *tcp;
 253	u8 *tmp;
 254	/* Dummy tcp packet */
 255	static const char tcp_packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 256		0x50, 0x11, 0x0, 0x72, 0, 0, 0, 0};
 257
 258	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_TCP, data);
 259
 260	tcp = (struct tcphdr *)tmp;
 261	memcpy(tcp, tcp_packet, sizeof(tcp_packet));
 262	tcp->dest = data->dst_port;
 263	tcp->source = data->src_port;
 264}
 265
 266/**
 267 * i40e_create_dummy_sctp_packet - helper function to create SCTP packet
 268 * @raw_packet: preallocated space for dummy packet
 269 * @ipv4: is layer 3 packet of version 4 or 6
 270 * @l4proto: next level protocol used in data portion of l3
 271 * @data: filter data
 272 *
 273 * Helper function to populate sctp fields.
 274 **/
 275static void i40e_create_dummy_sctp_packet(u8 *raw_packet, bool ipv4,
 276					  u8 l4proto,
 277					  struct i40e_fdir_filter *data)
 278{
 279	struct sctphdr *sctp;
 280	u8 *tmp;
 281
 282	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_SCTP, data);
 283
 284	sctp = (struct sctphdr *)tmp;
 285	sctp->dest = data->dst_port;
 286	sctp->source = data->src_port;
 287}
 288
 289/**
 290 * i40e_prepare_fdir_filter - Prepare and program fdir filter
 291 * @pf: physical function to attach filter to
 292 * @fd_data: filter data
 293 * @add: add or delete filter
 294 * @packet_addr: address of dummy packet, used in filtering
 295 * @payload_offset: offset from dummy packet address to user defined data
 296 * @pctype: Packet type for which filter is used
 297 *
 298 * Helper function to offset data of dummy packet, program it and
 299 * handle errors.
 300 **/
 301static int i40e_prepare_fdir_filter(struct i40e_pf *pf,
 302				    struct i40e_fdir_filter *fd_data,
 303				    bool add, char *packet_addr,
 304				    int payload_offset, u8 pctype)
 305{
 306	int ret;
 307
 308	if (fd_data->flex_filter) {
 309		u8 *payload;
 310		__be16 pattern = fd_data->flex_word;
 311		u16 off = fd_data->flex_offset;
 312
 313		payload = packet_addr + payload_offset;
 314
 315		/* If user provided vlan, offset payload by vlan header length */
 316		if (!!fd_data->vlan_tag)
 317			payload += VLAN_HLEN;
 318
 319		*((__force __be16 *)(payload + off)) = pattern;
 320	}
 321
 322	fd_data->pctype = pctype;
 323	ret = i40e_program_fdir_filter(fd_data, packet_addr, pf, add);
 324	if (ret) {
 325		dev_info(&pf->pdev->dev,
 326			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 327			 fd_data->pctype, fd_data->fd_id, ret);
 328		/* Free the packet buffer since it wasn't added to the ring */
 329		return -EOPNOTSUPP;
 330	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 331		if (add)
 332			dev_info(&pf->pdev->dev,
 333				 "Filter OK for PCTYPE %d loc = %d\n",
 334				 fd_data->pctype, fd_data->fd_id);
 335		else
 336			dev_info(&pf->pdev->dev,
 337				 "Filter deleted for PCTYPE %d loc = %d\n",
 338				 fd_data->pctype, fd_data->fd_id);
 339	}
 340
 341	return ret;
 342}
 343
 344/**
 345 * i40e_change_filter_num - Prepare and program fdir filter
 346 * @ipv4: is layer 3 packet of version 4 or 6
 347 * @add: add or delete filter
 348 * @ipv4_filter_num: field to update
 349 * @ipv6_filter_num: field to update
 350 *
 351 * Update filter number field for pf.
 352 **/
 353static void i40e_change_filter_num(bool ipv4, bool add, u16 *ipv4_filter_num,
 354				   u16 *ipv6_filter_num)
 355{
 356	if (add) {
 357		if (ipv4)
 358			(*ipv4_filter_num)++;
 359		else
 360			(*ipv6_filter_num)++;
 361	} else {
 362		if (ipv4)
 363			(*ipv4_filter_num)--;
 364		else
 365			(*ipv6_filter_num)--;
 366	}
 367}
 368
 369#define I40E_UDPIP_DUMMY_PACKET_LEN	42
 370#define I40E_UDPIP6_DUMMY_PACKET_LEN	62
 371/**
 372 * i40e_add_del_fdir_udp - Add/Remove UDP filters
 373 * @vsi: pointer to the targeted VSI
 374 * @fd_data: the flow director data required for the FDir descriptor
 375 * @add: true adds a filter, false removes it
 376 * @ipv4: true is v4, false is v6
 377 *
 378 * Returns 0 if the filters were successfully added or removed
 379 **/
 380static int i40e_add_del_fdir_udp(struct i40e_vsi *vsi,
 381				 struct i40e_fdir_filter *fd_data,
 382				 bool add,
 383				 bool ipv4)
 384{
 385	struct i40e_pf *pf = vsi->back;
 386	u8 *raw_packet;
 387	int ret;
 388
 389	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 390	if (!raw_packet)
 391		return -ENOMEM;
 392
 393	i40e_create_dummy_udp_packet(raw_packet, ipv4, IPPROTO_UDP, fd_data);
 394
 395	if (ipv4)
 396		ret = i40e_prepare_fdir_filter
 397			(pf, fd_data, add, raw_packet,
 398			 I40E_UDPIP_DUMMY_PACKET_LEN,
 399			 I40E_FILTER_PCTYPE_NONF_IPV4_UDP);
 400	else
 401		ret = i40e_prepare_fdir_filter
 402			(pf, fd_data, add, raw_packet,
 403			 I40E_UDPIP6_DUMMY_PACKET_LEN,
 404			 I40E_FILTER_PCTYPE_NONF_IPV6_UDP);
 405
 406	if (ret) {
 407		kfree(raw_packet);
 408		return ret;
 409	}
 410
 411	i40e_change_filter_num(ipv4, add, &pf->fd_udp4_filter_cnt,
 412			       &pf->fd_udp6_filter_cnt);
 413
 414	return 0;
 415}
 416
 417#define I40E_TCPIP_DUMMY_PACKET_LEN	54
 418#define I40E_TCPIP6_DUMMY_PACKET_LEN	74
 419/**
 420 * i40e_add_del_fdir_tcp - Add/Remove TCPv4 filters
 421 * @vsi: pointer to the targeted VSI
 422 * @fd_data: the flow director data required for the FDir descriptor
 423 * @add: true adds a filter, false removes it
 424 * @ipv4: true is v4, false is v6
 425 *
 426 * Returns 0 if the filters were successfully added or removed
 427 **/
 428static int i40e_add_del_fdir_tcp(struct i40e_vsi *vsi,
 429				 struct i40e_fdir_filter *fd_data,
 430				 bool add,
 431				 bool ipv4)
 432{
 433	struct i40e_pf *pf = vsi->back;
 434	u8 *raw_packet;
 435	int ret;
 436
 437	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 438	if (!raw_packet)
 439		return -ENOMEM;
 440
 441	i40e_create_dummy_tcp_packet(raw_packet, ipv4, IPPROTO_TCP, fd_data);
 442	if (ipv4)
 443		ret = i40e_prepare_fdir_filter
 444			(pf, fd_data, add, raw_packet,
 445			 I40E_TCPIP_DUMMY_PACKET_LEN,
 446			 I40E_FILTER_PCTYPE_NONF_IPV4_TCP);
 447	else
 448		ret = i40e_prepare_fdir_filter
 449			(pf, fd_data, add, raw_packet,
 450			 I40E_TCPIP6_DUMMY_PACKET_LEN,
 451			 I40E_FILTER_PCTYPE_NONF_IPV6_TCP);
 452
 453	if (ret) {
 454		kfree(raw_packet);
 455		return ret;
 456	}
 457
 458	i40e_change_filter_num(ipv4, add, &pf->fd_tcp4_filter_cnt,
 459			       &pf->fd_tcp6_filter_cnt);
 460
 461	if (add) {
 462		if (test_bit(I40E_FLAG_FD_ATR_ENA, pf->flags) &&
 463		    I40E_DEBUG_FD & pf->hw.debug_mask)
 464			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
 465		set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 466	}
 467	return 0;
 468}
 469
 470#define I40E_SCTPIP_DUMMY_PACKET_LEN	46
 471#define I40E_SCTPIP6_DUMMY_PACKET_LEN	66
 472/**
 473 * i40e_add_del_fdir_sctp - Add/Remove SCTPv4 Flow Director filters for
 474 * a specific flow spec
 475 * @vsi: pointer to the targeted VSI
 476 * @fd_data: the flow director data required for the FDir descriptor
 477 * @add: true adds a filter, false removes it
 478 * @ipv4: true is v4, false is v6
 479 *
 480 * Returns 0 if the filters were successfully added or removed
 481 **/
 482static int i40e_add_del_fdir_sctp(struct i40e_vsi *vsi,
 483				  struct i40e_fdir_filter *fd_data,
 484				  bool add,
 485				  bool ipv4)
 486{
 487	struct i40e_pf *pf = vsi->back;
 488	u8 *raw_packet;
 489	int ret;
 490
 491	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 492	if (!raw_packet)
 493		return -ENOMEM;
 494
 495	i40e_create_dummy_sctp_packet(raw_packet, ipv4, IPPROTO_SCTP, fd_data);
 496
 497	if (ipv4)
 498		ret = i40e_prepare_fdir_filter
 499			(pf, fd_data, add, raw_packet,
 500			 I40E_SCTPIP_DUMMY_PACKET_LEN,
 501			 I40E_FILTER_PCTYPE_NONF_IPV4_SCTP);
 502	else
 503		ret = i40e_prepare_fdir_filter
 504			(pf, fd_data, add, raw_packet,
 505			 I40E_SCTPIP6_DUMMY_PACKET_LEN,
 506			 I40E_FILTER_PCTYPE_NONF_IPV6_SCTP);
 507
 508	if (ret) {
 509		kfree(raw_packet);
 510		return ret;
 511	}
 512
 513	i40e_change_filter_num(ipv4, add, &pf->fd_sctp4_filter_cnt,
 514			       &pf->fd_sctp6_filter_cnt);
 515
 516	return 0;
 517}
 518
 519#define I40E_IP_DUMMY_PACKET_LEN	34
 520#define I40E_IP6_DUMMY_PACKET_LEN	54
 521/**
 522 * i40e_add_del_fdir_ip - Add/Remove IPv4 Flow Director filters for
 523 * a specific flow spec
 524 * @vsi: pointer to the targeted VSI
 525 * @fd_data: the flow director data required for the FDir descriptor
 526 * @add: true adds a filter, false removes it
 527 * @ipv4: true is v4, false is v6
 528 *
 529 * Returns 0 if the filters were successfully added or removed
 530 **/
 531static int i40e_add_del_fdir_ip(struct i40e_vsi *vsi,
 532				struct i40e_fdir_filter *fd_data,
 533				bool add,
 534				bool ipv4)
 535{
 536	struct i40e_pf *pf = vsi->back;
 537	int payload_offset;
 538	u8 *raw_packet;
 539	int iter_start;
 540	int iter_end;
 541	int ret;
 542	int i;
 543
 544	if (ipv4) {
 545		iter_start = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
 546		iter_end = I40E_FILTER_PCTYPE_FRAG_IPV4;
 547	} else {
 548		iter_start = I40E_FILTER_PCTYPE_NONF_IPV6_OTHER;
 549		iter_end = I40E_FILTER_PCTYPE_FRAG_IPV6;
 550	}
 551
 552	for (i = iter_start; i <= iter_end; i++) {
 553		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 554		if (!raw_packet)
 555			return -ENOMEM;
 556
 557		/* IPv6 no header option differs from IPv4 */
 558		(void)i40e_create_dummy_packet
 559			(raw_packet, ipv4, (ipv4) ? IPPROTO_IP : IPPROTO_NONE,
 560			 fd_data);
 561
 562		payload_offset = (ipv4) ? I40E_IP_DUMMY_PACKET_LEN :
 563			I40E_IP6_DUMMY_PACKET_LEN;
 564		ret = i40e_prepare_fdir_filter(pf, fd_data, add, raw_packet,
 565					       payload_offset, i);
 566		if (ret)
 567			goto err;
 568	}
 569
 570	i40e_change_filter_num(ipv4, add, &pf->fd_ip4_filter_cnt,
 571			       &pf->fd_ip6_filter_cnt);
 572
 573	return 0;
 574err:
 575	kfree(raw_packet);
 576	return ret;
 577}
 578
 579/**
 580 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
 581 * @vsi: pointer to the targeted VSI
 582 * @input: filter to add or delete
 583 * @add: true adds a filter, false removes it
 584 *
 585 **/
 586int i40e_add_del_fdir(struct i40e_vsi *vsi,
 587		      struct i40e_fdir_filter *input, bool add)
 588{
 589	enum ip_ver { ipv6 = 0, ipv4 = 1 };
 590	struct i40e_pf *pf = vsi->back;
 591	int ret;
 592
 593	switch (input->flow_type & ~FLOW_EXT) {
 594	case TCP_V4_FLOW:
 595		ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
 596		break;
 597	case UDP_V4_FLOW:
 598		ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
 599		break;
 600	case SCTP_V4_FLOW:
 601		ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
 602		break;
 603	case TCP_V6_FLOW:
 604		ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
 605		break;
 606	case UDP_V6_FLOW:
 607		ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
 608		break;
 609	case SCTP_V6_FLOW:
 610		ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
 611		break;
 612	case IP_USER_FLOW:
 613		switch (input->ipl4_proto) {
 614		case IPPROTO_TCP:
 615			ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
 616			break;
 617		case IPPROTO_UDP:
 618			ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
 619			break;
 620		case IPPROTO_SCTP:
 621			ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
 622			break;
 623		case IPPROTO_IP:
 624			ret = i40e_add_del_fdir_ip(vsi, input, add, ipv4);
 625			break;
 626		default:
 627			/* We cannot support masking based on protocol */
 628			dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
 629				 input->ipl4_proto);
 630			return -EINVAL;
 631		}
 632		break;
 633	case IPV6_USER_FLOW:
 634		switch (input->ipl4_proto) {
 635		case IPPROTO_TCP:
 636			ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
 637			break;
 638		case IPPROTO_UDP:
 639			ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
 640			break;
 641		case IPPROTO_SCTP:
 642			ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
 643			break;
 644		case IPPROTO_IP:
 645			ret = i40e_add_del_fdir_ip(vsi, input, add, ipv6);
 646			break;
 647		default:
 648			/* We cannot support masking based on protocol */
 649			dev_info(&pf->pdev->dev, "Unsupported IPv6 protocol 0x%02x\n",
 650				 input->ipl4_proto);
 651			return -EINVAL;
 652		}
 653		break;
 654	default:
 655		dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
 656			 input->flow_type);
 657		return -EINVAL;
 658	}
 659
 660	/* The buffer allocated here will be normally be freed by
 661	 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
 662	 * completion. In the event of an error adding the buffer to the FDIR
 663	 * ring, it will immediately be freed. It may also be freed by
 664	 * i40e_clean_tx_ring() when closing the VSI.
 665	 */
 666	return ret;
 667}
 668
 669/**
 670 * i40e_fd_handle_status - check the Programming Status for FD
 671 * @rx_ring: the Rx ring for this descriptor
 672 * @qword0_raw: qword0
 673 * @qword1: qword1 after le_to_cpu
 674 * @prog_id: the id originally used for programming
 675 *
 676 * This is used to verify if the FD programming or invalidation
 677 * requested by SW to the HW is successful or not and take actions accordingly.
 678 **/
 679static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u64 qword0_raw,
 680				  u64 qword1, u8 prog_id)
 681{
 682	struct i40e_pf *pf = rx_ring->vsi->back;
 683	struct pci_dev *pdev = pf->pdev;
 684	struct i40e_16b_rx_wb_qw0 *qw0;
 685	u32 fcnt_prog, fcnt_avail;
 686	u32 error;
 687
 688	qw0 = (struct i40e_16b_rx_wb_qw0 *)&qword0_raw;
 689	error = FIELD_GET(I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK, qword1);
 
 690
 691	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
 692		pf->fd_inv = le32_to_cpu(qw0->hi_dword.fd_id);
 693		if (qw0->hi_dword.fd_id != 0 ||
 694		    (I40E_DEBUG_FD & pf->hw.debug_mask))
 695			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
 696				 pf->fd_inv);
 697
 698		/* Check if the programming error is for ATR.
 699		 * If so, auto disable ATR and set a state for
 700		 * flush in progress. Next time we come here if flush is in
 701		 * progress do nothing, once flush is complete the state will
 702		 * be cleared.
 703		 */
 704		if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
 705			return;
 706
 707		pf->fd_add_err++;
 708		/* store the current atr filter count */
 709		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
 710
 711		if (qw0->hi_dword.fd_id == 0 &&
 712		    test_bit(__I40E_FD_SB_AUTO_DISABLED, pf->state)) {
 713			/* These set_bit() calls aren't atomic with the
 714			 * test_bit() here, but worse case we potentially
 715			 * disable ATR and queue a flush right after SB
 716			 * support is re-enabled. That shouldn't cause an
 717			 * issue in practice
 718			 */
 719			set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 720			set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
 721		}
 722
 723		/* filter programming failed most likely due to table full */
 724		fcnt_prog = i40e_get_global_fd_count(pf);
 725		fcnt_avail = pf->fdir_pf_filter_count;
 726		/* If ATR is running fcnt_prog can quickly change,
 727		 * if we are very close to full, it makes sense to disable
 728		 * FD ATR/SB and then re-enable it when there is room.
 729		 */
 730		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
 731			if (test_bit(I40E_FLAG_FD_SB_ENA, pf->flags) &&
 732			    !test_and_set_bit(__I40E_FD_SB_AUTO_DISABLED,
 733					      pf->state))
 734				if (I40E_DEBUG_FD & pf->hw.debug_mask)
 735					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
 736		}
 737	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
 738		if (I40E_DEBUG_FD & pf->hw.debug_mask)
 739			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
 740				 qw0->hi_dword.fd_id);
 741	}
 742}
 743
 744/**
 745 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
 746 * @ring:      the ring that owns the buffer
 747 * @tx_buffer: the buffer to free
 748 **/
 749static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
 750					    struct i40e_tx_buffer *tx_buffer)
 751{
 752	if (tx_buffer->skb) {
 753		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
 754			kfree(tx_buffer->raw_buf);
 755		else if (ring_is_xdp(ring))
 756			xdp_return_frame(tx_buffer->xdpf);
 757		else
 758			dev_kfree_skb_any(tx_buffer->skb);
 759		if (dma_unmap_len(tx_buffer, len))
 760			dma_unmap_single(ring->dev,
 761					 dma_unmap_addr(tx_buffer, dma),
 762					 dma_unmap_len(tx_buffer, len),
 763					 DMA_TO_DEVICE);
 764	} else if (dma_unmap_len(tx_buffer, len)) {
 765		dma_unmap_page(ring->dev,
 766			       dma_unmap_addr(tx_buffer, dma),
 767			       dma_unmap_len(tx_buffer, len),
 768			       DMA_TO_DEVICE);
 769	}
 770
 771	tx_buffer->next_to_watch = NULL;
 772	tx_buffer->skb = NULL;
 773	dma_unmap_len_set(tx_buffer, len, 0);
 774	/* tx_buffer must be completely set up in the transmit path */
 775}
 776
 777/**
 778 * i40e_clean_tx_ring - Free any empty Tx buffers
 779 * @tx_ring: ring to be cleaned
 780 **/
 781void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
 782{
 783	unsigned long bi_size;
 784	u16 i;
 785
 786	if (ring_is_xdp(tx_ring) && tx_ring->xsk_pool) {
 787		i40e_xsk_clean_tx_ring(tx_ring);
 788	} else {
 789		/* ring already cleared, nothing to do */
 790		if (!tx_ring->tx_bi)
 791			return;
 792
 793		/* Free all the Tx ring sk_buffs */
 794		for (i = 0; i < tx_ring->count; i++)
 795			i40e_unmap_and_free_tx_resource(tx_ring,
 796							&tx_ring->tx_bi[i]);
 797	}
 798
 799	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
 800	memset(tx_ring->tx_bi, 0, bi_size);
 801
 802	/* Zero out the descriptor ring */
 803	memset(tx_ring->desc, 0, tx_ring->size);
 804
 805	tx_ring->next_to_use = 0;
 806	tx_ring->next_to_clean = 0;
 807
 808	if (!tx_ring->netdev)
 809		return;
 810
 811	/* cleanup Tx queue statistics */
 812	netdev_tx_reset_queue(txring_txq(tx_ring));
 813}
 814
 815/**
 816 * i40e_free_tx_resources - Free Tx resources per queue
 817 * @tx_ring: Tx descriptor ring for a specific queue
 818 *
 819 * Free all transmit software resources
 820 **/
 821void i40e_free_tx_resources(struct i40e_ring *tx_ring)
 822{
 823	i40e_clean_tx_ring(tx_ring);
 824	kfree(tx_ring->tx_bi);
 825	tx_ring->tx_bi = NULL;
 826
 827	if (tx_ring->desc) {
 828		dma_free_coherent(tx_ring->dev, tx_ring->size,
 829				  tx_ring->desc, tx_ring->dma);
 830		tx_ring->desc = NULL;
 831	}
 832}
 833
 834/**
 835 * i40e_get_tx_pending - how many tx descriptors not processed
 836 * @ring: the ring of descriptors
 837 * @in_sw: use SW variables
 838 *
 839 * Since there is no access to the ring head register
 840 * in XL710, we need to use our local copies
 841 **/
 842u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
 843{
 844	u32 head, tail;
 845
 846	if (!in_sw) {
 847		head = i40e_get_head(ring);
 848		tail = readl(ring->tail);
 849	} else {
 850		head = ring->next_to_clean;
 851		tail = ring->next_to_use;
 852	}
 853
 854	if (head != tail)
 855		return (head < tail) ?
 856			tail - head : (tail + ring->count - head);
 857
 858	return 0;
 859}
 860
 861/**
 862 * i40e_detect_recover_hung - Function to detect and recover hung_queues
 863 * @vsi:  pointer to vsi struct with tx queues
 864 *
 865 * VSI has netdev and netdev has TX queues. This function is to check each of
 866 * those TX queues if they are hung, trigger recovery by issuing SW interrupt.
 867 **/
 868void i40e_detect_recover_hung(struct i40e_vsi *vsi)
 869{
 870	struct i40e_ring *tx_ring = NULL;
 871	struct net_device *netdev;
 872	unsigned int i;
 873	int packets;
 874
 875	if (!vsi)
 876		return;
 877
 878	if (test_bit(__I40E_VSI_DOWN, vsi->state))
 879		return;
 880
 881	netdev = vsi->netdev;
 882	if (!netdev)
 883		return;
 884
 885	if (!netif_carrier_ok(netdev))
 886		return;
 887
 888	for (i = 0; i < vsi->num_queue_pairs; i++) {
 889		tx_ring = vsi->tx_rings[i];
 890		if (tx_ring && tx_ring->desc) {
 891			/* If packet counter has not changed the queue is
 892			 * likely stalled, so force an interrupt for this
 893			 * queue.
 894			 *
 895			 * prev_pkt_ctr would be negative if there was no
 896			 * pending work.
 897			 */
 898			packets = tx_ring->stats.packets & INT_MAX;
 899			if (tx_ring->tx_stats.prev_pkt_ctr == packets) {
 900				i40e_force_wb(vsi, tx_ring->q_vector);
 901				continue;
 902			}
 903
 904			/* Memory barrier between read of packet count and call
 905			 * to i40e_get_tx_pending()
 906			 */
 907			smp_rmb();
 908			tx_ring->tx_stats.prev_pkt_ctr =
 909			    i40e_get_tx_pending(tx_ring, true) ? packets : -1;
 910		}
 911	}
 912}
 913
 914/**
 915 * i40e_clean_tx_irq - Reclaim resources after transmit completes
 916 * @vsi: the VSI we care about
 917 * @tx_ring: Tx ring to clean
 918 * @napi_budget: Used to determine if we are in netpoll
 919 * @tx_cleaned: Out parameter set to the number of TXes cleaned
 920 *
 921 * Returns true if there's any budget left (e.g. the clean is finished)
 922 **/
 923static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
 924			      struct i40e_ring *tx_ring, int napi_budget,
 925			      unsigned int *tx_cleaned)
 926{
 927	int i = tx_ring->next_to_clean;
 928	struct i40e_tx_buffer *tx_buf;
 929	struct i40e_tx_desc *tx_head;
 930	struct i40e_tx_desc *tx_desc;
 931	unsigned int total_bytes = 0, total_packets = 0;
 932	unsigned int budget = vsi->work_limit;
 933
 934	tx_buf = &tx_ring->tx_bi[i];
 935	tx_desc = I40E_TX_DESC(tx_ring, i);
 936	i -= tx_ring->count;
 937
 938	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
 939
 940	do {
 941		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 942
 943		/* if next_to_watch is not set then there is no work pending */
 944		if (!eop_desc)
 945			break;
 946
 947		/* prevent any other reads prior to eop_desc */
 948		smp_rmb();
 949
 950		i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
 951		/* we have caught up to head, no work left to do */
 952		if (tx_head == tx_desc)
 953			break;
 954
 955		/* clear next_to_watch to prevent false hangs */
 956		tx_buf->next_to_watch = NULL;
 957
 958		/* update the statistics for this packet */
 959		total_bytes += tx_buf->bytecount;
 960		total_packets += tx_buf->gso_segs;
 961
 962		/* free the skb/XDP data */
 963		if (ring_is_xdp(tx_ring))
 964			xdp_return_frame(tx_buf->xdpf);
 965		else
 966			napi_consume_skb(tx_buf->skb, napi_budget);
 967
 968		/* unmap skb header data */
 969		dma_unmap_single(tx_ring->dev,
 970				 dma_unmap_addr(tx_buf, dma),
 971				 dma_unmap_len(tx_buf, len),
 972				 DMA_TO_DEVICE);
 973
 974		/* clear tx_buffer data */
 975		tx_buf->skb = NULL;
 976		dma_unmap_len_set(tx_buf, len, 0);
 977
 978		/* unmap remaining buffers */
 979		while (tx_desc != eop_desc) {
 980			i40e_trace(clean_tx_irq_unmap,
 981				   tx_ring, tx_desc, tx_buf);
 982
 983			tx_buf++;
 984			tx_desc++;
 985			i++;
 986			if (unlikely(!i)) {
 987				i -= tx_ring->count;
 988				tx_buf = tx_ring->tx_bi;
 989				tx_desc = I40E_TX_DESC(tx_ring, 0);
 990			}
 991
 992			/* unmap any remaining paged data */
 993			if (dma_unmap_len(tx_buf, len)) {
 994				dma_unmap_page(tx_ring->dev,
 995					       dma_unmap_addr(tx_buf, dma),
 996					       dma_unmap_len(tx_buf, len),
 997					       DMA_TO_DEVICE);
 998				dma_unmap_len_set(tx_buf, len, 0);
 999			}
1000		}
1001
1002		/* move us one more past the eop_desc for start of next pkt */
1003		tx_buf++;
1004		tx_desc++;
1005		i++;
1006		if (unlikely(!i)) {
1007			i -= tx_ring->count;
1008			tx_buf = tx_ring->tx_bi;
1009			tx_desc = I40E_TX_DESC(tx_ring, 0);
1010		}
1011
1012		prefetch(tx_desc);
1013
1014		/* update budget accounting */
1015		budget--;
1016	} while (likely(budget));
1017
1018	i += tx_ring->count;
1019	tx_ring->next_to_clean = i;
1020	i40e_update_tx_stats(tx_ring, total_packets, total_bytes);
1021	i40e_arm_wb(tx_ring, vsi, budget);
1022
1023	if (ring_is_xdp(tx_ring))
1024		return !!budget;
1025
1026	/* notify netdev of completed buffers */
1027	netdev_tx_completed_queue(txring_txq(tx_ring),
1028				  total_packets, total_bytes);
1029
1030#define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
1031	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1032		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
1033		/* Make sure that anybody stopping the queue after this
1034		 * sees the new next_to_clean.
1035		 */
1036		smp_mb();
1037		if (__netif_subqueue_stopped(tx_ring->netdev,
1038					     tx_ring->queue_index) &&
1039		   !test_bit(__I40E_VSI_DOWN, vsi->state)) {
1040			netif_wake_subqueue(tx_ring->netdev,
1041					    tx_ring->queue_index);
1042			++tx_ring->tx_stats.restart_queue;
1043		}
1044	}
1045
1046	*tx_cleaned = total_packets;
1047	return !!budget;
1048}
1049
1050/**
1051 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
1052 * @vsi: the VSI we care about
1053 * @q_vector: the vector on which to enable writeback
1054 *
1055 **/
1056static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
1057				  struct i40e_q_vector *q_vector)
1058{
1059	u16 flags = q_vector->tx.ring[0].flags;
1060	u32 val;
1061
1062	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
1063		return;
1064
1065	if (q_vector->arm_wb_state)
1066		return;
1067
1068	if (test_bit(I40E_FLAG_MSIX_ENA, vsi->back->flags)) {
1069		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
1070		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
1071
1072		wr32(&vsi->back->hw,
1073		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx),
1074		     val);
1075	} else {
1076		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
1077		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
1078
1079		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1080	}
1081	q_vector->arm_wb_state = true;
1082}
1083
1084/**
1085 * i40e_force_wb - Issue SW Interrupt so HW does a wb
1086 * @vsi: the VSI we care about
1087 * @q_vector: the vector  on which to force writeback
1088 *
1089 **/
1090void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
1091{
1092	if (test_bit(I40E_FLAG_MSIX_ENA, vsi->back->flags)) {
1093		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1094			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
1095			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
1096			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
1097			  /* allow 00 to be written to the index */
1098
1099		wr32(&vsi->back->hw,
1100		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx), val);
1101	} else {
1102		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
1103			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
1104			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
1105			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
1106			/* allow 00 to be written to the index */
1107
1108		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1109	}
1110}
1111
1112static inline bool i40e_container_is_rx(struct i40e_q_vector *q_vector,
1113					struct i40e_ring_container *rc)
1114{
1115	return &q_vector->rx == rc;
1116}
1117
1118static inline unsigned int i40e_itr_divisor(struct i40e_q_vector *q_vector)
1119{
1120	unsigned int divisor;
1121
1122	switch (q_vector->vsi->back->hw.phy.link_info.link_speed) {
1123	case I40E_LINK_SPEED_40GB:
1124		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 1024;
1125		break;
1126	case I40E_LINK_SPEED_25GB:
1127	case I40E_LINK_SPEED_20GB:
1128		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 512;
1129		break;
1130	default:
1131	case I40E_LINK_SPEED_10GB:
1132		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 256;
1133		break;
1134	case I40E_LINK_SPEED_1GB:
1135	case I40E_LINK_SPEED_100MB:
1136		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 32;
1137		break;
1138	}
1139
1140	return divisor;
1141}
1142
1143/**
1144 * i40e_update_itr - update the dynamic ITR value based on statistics
1145 * @q_vector: structure containing interrupt and ring information
1146 * @rc: structure containing ring performance data
1147 *
1148 * Stores a new ITR value based on packets and byte
1149 * counts during the last interrupt.  The advantage of per interrupt
1150 * computation is faster updates and more accurate ITR for the current
1151 * traffic pattern.  Constants in this function were computed
1152 * based on theoretical maximum wire speed and thresholds were set based
1153 * on testing data as well as attempting to minimize response time
1154 * while increasing bulk throughput.
1155 **/
1156static void i40e_update_itr(struct i40e_q_vector *q_vector,
1157			    struct i40e_ring_container *rc)
1158{
1159	unsigned int avg_wire_size, packets, bytes, itr;
1160	unsigned long next_update = jiffies;
1161
1162	/* If we don't have any rings just leave ourselves set for maximum
1163	 * possible latency so we take ourselves out of the equation.
1164	 */
1165	if (!rc->ring || !ITR_IS_DYNAMIC(rc->ring->itr_setting))
1166		return;
1167
1168	/* For Rx we want to push the delay up and default to low latency.
1169	 * for Tx we want to pull the delay down and default to high latency.
1170	 */
1171	itr = i40e_container_is_rx(q_vector, rc) ?
1172	      I40E_ITR_ADAPTIVE_MIN_USECS | I40E_ITR_ADAPTIVE_LATENCY :
1173	      I40E_ITR_ADAPTIVE_MAX_USECS | I40E_ITR_ADAPTIVE_LATENCY;
1174
1175	/* If we didn't update within up to 1 - 2 jiffies we can assume
1176	 * that either packets are coming in so slow there hasn't been
1177	 * any work, or that there is so much work that NAPI is dealing
1178	 * with interrupt moderation and we don't need to do anything.
1179	 */
1180	if (time_after(next_update, rc->next_update))
1181		goto clear_counts;
1182
1183	/* If itr_countdown is set it means we programmed an ITR within
1184	 * the last 4 interrupt cycles. This has a side effect of us
1185	 * potentially firing an early interrupt. In order to work around
1186	 * this we need to throw out any data received for a few
1187	 * interrupts following the update.
1188	 */
1189	if (q_vector->itr_countdown) {
1190		itr = rc->target_itr;
1191		goto clear_counts;
1192	}
1193
1194	packets = rc->total_packets;
1195	bytes = rc->total_bytes;
1196
1197	if (i40e_container_is_rx(q_vector, rc)) {
1198		/* If Rx there are 1 to 4 packets and bytes are less than
1199		 * 9000 assume insufficient data to use bulk rate limiting
1200		 * approach unless Tx is already in bulk rate limiting. We
1201		 * are likely latency driven.
1202		 */
1203		if (packets && packets < 4 && bytes < 9000 &&
1204		    (q_vector->tx.target_itr & I40E_ITR_ADAPTIVE_LATENCY)) {
1205			itr = I40E_ITR_ADAPTIVE_LATENCY;
1206			goto adjust_by_size;
1207		}
1208	} else if (packets < 4) {
1209		/* If we have Tx and Rx ITR maxed and Tx ITR is running in
1210		 * bulk mode and we are receiving 4 or fewer packets just
1211		 * reset the ITR_ADAPTIVE_LATENCY bit for latency mode so
1212		 * that the Rx can relax.
1213		 */
1214		if (rc->target_itr == I40E_ITR_ADAPTIVE_MAX_USECS &&
1215		    (q_vector->rx.target_itr & I40E_ITR_MASK) ==
1216		     I40E_ITR_ADAPTIVE_MAX_USECS)
1217			goto clear_counts;
1218	} else if (packets > 32) {
1219		/* If we have processed over 32 packets in a single interrupt
1220		 * for Tx assume we need to switch over to "bulk" mode.
1221		 */
1222		rc->target_itr &= ~I40E_ITR_ADAPTIVE_LATENCY;
1223	}
1224
1225	/* We have no packets to actually measure against. This means
1226	 * either one of the other queues on this vector is active or
1227	 * we are a Tx queue doing TSO with too high of an interrupt rate.
1228	 *
1229	 * Between 4 and 56 we can assume that our current interrupt delay
1230	 * is only slightly too low. As such we should increase it by a small
1231	 * fixed amount.
1232	 */
1233	if (packets < 56) {
1234		itr = rc->target_itr + I40E_ITR_ADAPTIVE_MIN_INC;
1235		if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1236			itr &= I40E_ITR_ADAPTIVE_LATENCY;
1237			itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1238		}
1239		goto clear_counts;
1240	}
1241
1242	if (packets <= 256) {
1243		itr = min(q_vector->tx.current_itr, q_vector->rx.current_itr);
1244		itr &= I40E_ITR_MASK;
1245
1246		/* Between 56 and 112 is our "goldilocks" zone where we are
1247		 * working out "just right". Just report that our current
1248		 * ITR is good for us.
1249		 */
1250		if (packets <= 112)
1251			goto clear_counts;
1252
1253		/* If packet count is 128 or greater we are likely looking
1254		 * at a slight overrun of the delay we want. Try halving
1255		 * our delay to see if that will cut the number of packets
1256		 * in half per interrupt.
1257		 */
1258		itr /= 2;
1259		itr &= I40E_ITR_MASK;
1260		if (itr < I40E_ITR_ADAPTIVE_MIN_USECS)
1261			itr = I40E_ITR_ADAPTIVE_MIN_USECS;
1262
1263		goto clear_counts;
1264	}
1265
1266	/* The paths below assume we are dealing with a bulk ITR since
1267	 * number of packets is greater than 256. We are just going to have
1268	 * to compute a value and try to bring the count under control,
1269	 * though for smaller packet sizes there isn't much we can do as
1270	 * NAPI polling will likely be kicking in sooner rather than later.
1271	 */
1272	itr = I40E_ITR_ADAPTIVE_BULK;
1273
1274adjust_by_size:
1275	/* If packet counts are 256 or greater we can assume we have a gross
1276	 * overestimation of what the rate should be. Instead of trying to fine
1277	 * tune it just use the formula below to try and dial in an exact value
1278	 * give the current packet size of the frame.
1279	 */
1280	avg_wire_size = bytes / packets;
1281
1282	/* The following is a crude approximation of:
1283	 *  wmem_default / (size + overhead) = desired_pkts_per_int
1284	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1285	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1286	 *
1287	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1288	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1289	 * formula down to
1290	 *
1291	 *  (170 * (size + 24)) / (size + 640) = ITR
1292	 *
1293	 * We first do some math on the packet size and then finally bitshift
1294	 * by 8 after rounding up. We also have to account for PCIe link speed
1295	 * difference as ITR scales based on this.
1296	 */
1297	if (avg_wire_size <= 60) {
1298		/* Start at 250k ints/sec */
1299		avg_wire_size = 4096;
1300	} else if (avg_wire_size <= 380) {
1301		/* 250K ints/sec to 60K ints/sec */
1302		avg_wire_size *= 40;
1303		avg_wire_size += 1696;
1304	} else if (avg_wire_size <= 1084) {
1305		/* 60K ints/sec to 36K ints/sec */
1306		avg_wire_size *= 15;
1307		avg_wire_size += 11452;
1308	} else if (avg_wire_size <= 1980) {
1309		/* 36K ints/sec to 30K ints/sec */
1310		avg_wire_size *= 5;
1311		avg_wire_size += 22420;
1312	} else {
1313		/* plateau at a limit of 30K ints/sec */
1314		avg_wire_size = 32256;
1315	}
1316
1317	/* If we are in low latency mode halve our delay which doubles the
1318	 * rate to somewhere between 100K to 16K ints/sec
1319	 */
1320	if (itr & I40E_ITR_ADAPTIVE_LATENCY)
1321		avg_wire_size /= 2;
1322
1323	/* Resultant value is 256 times larger than it needs to be. This
1324	 * gives us room to adjust the value as needed to either increase
1325	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
1326	 *
1327	 * Use addition as we have already recorded the new latency flag
1328	 * for the ITR value.
1329	 */
1330	itr += DIV_ROUND_UP(avg_wire_size, i40e_itr_divisor(q_vector)) *
1331	       I40E_ITR_ADAPTIVE_MIN_INC;
1332
1333	if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1334		itr &= I40E_ITR_ADAPTIVE_LATENCY;
1335		itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1336	}
1337
1338clear_counts:
1339	/* write back value */
1340	rc->target_itr = itr;
1341
1342	/* next update should occur within next jiffy */
1343	rc->next_update = next_update + 1;
1344
1345	rc->total_bytes = 0;
1346	rc->total_packets = 0;
1347}
1348
1349static struct i40e_rx_buffer *i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
1350{
1351	return &rx_ring->rx_bi[idx];
1352}
1353
1354/**
1355 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1356 * @rx_ring: rx descriptor ring to store buffers on
1357 * @old_buff: donor buffer to have page reused
1358 *
1359 * Synchronizes page for reuse by the adapter
1360 **/
1361static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1362			       struct i40e_rx_buffer *old_buff)
1363{
1364	struct i40e_rx_buffer *new_buff;
1365	u16 nta = rx_ring->next_to_alloc;
1366
1367	new_buff = i40e_rx_bi(rx_ring, nta);
1368
1369	/* update, and store next to alloc */
1370	nta++;
1371	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1372
1373	/* transfer page from old buffer to new buffer */
1374	new_buff->dma		= old_buff->dma;
1375	new_buff->page		= old_buff->page;
1376	new_buff->page_offset	= old_buff->page_offset;
1377	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1378
1379	/* clear contents of buffer_info */
1380	old_buff->page = NULL;
1381}
1382
1383/**
1384 * i40e_clean_programming_status - clean the programming status descriptor
1385 * @rx_ring: the rx ring that has this descriptor
1386 * @qword0_raw: qword0
1387 * @qword1: qword1 representing status_error_len in CPU ordering
1388 *
1389 * Flow director should handle FD_FILTER_STATUS to check its filter programming
1390 * status being successful or not and take actions accordingly. FCoE should
1391 * handle its context/filter programming/invalidation status and take actions.
1392 *
1393 * Returns an i40e_rx_buffer to reuse if the cleanup occurred, otherwise NULL.
1394 **/
1395void i40e_clean_programming_status(struct i40e_ring *rx_ring, u64 qword0_raw,
1396				   u64 qword1)
1397{
1398	u8 id;
1399
1400	id = FIELD_GET(I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK, qword1);
 
1401
1402	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1403		i40e_fd_handle_status(rx_ring, qword0_raw, qword1, id);
1404}
1405
1406/**
1407 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1408 * @tx_ring: the tx ring to set up
1409 *
1410 * Return 0 on success, negative on error
1411 **/
1412int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1413{
1414	struct device *dev = tx_ring->dev;
1415	int bi_size;
1416
1417	if (!dev)
1418		return -ENOMEM;
1419
1420	/* warn if we are about to overwrite the pointer */
1421	WARN_ON(tx_ring->tx_bi);
1422	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1423	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1424	if (!tx_ring->tx_bi)
1425		goto err;
1426
1427	u64_stats_init(&tx_ring->syncp);
1428
1429	/* round up to nearest 4K */
1430	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1431	/* add u32 for head writeback, align after this takes care of
1432	 * guaranteeing this is at least one cache line in size
1433	 */
1434	tx_ring->size += sizeof(u32);
1435	tx_ring->size = ALIGN(tx_ring->size, 4096);
1436	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1437					   &tx_ring->dma, GFP_KERNEL);
1438	if (!tx_ring->desc) {
1439		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1440			 tx_ring->size);
1441		goto err;
1442	}
1443
1444	tx_ring->next_to_use = 0;
1445	tx_ring->next_to_clean = 0;
1446	tx_ring->tx_stats.prev_pkt_ctr = -1;
1447	return 0;
1448
1449err:
1450	kfree(tx_ring->tx_bi);
1451	tx_ring->tx_bi = NULL;
1452	return -ENOMEM;
1453}
1454
1455static void i40e_clear_rx_bi(struct i40e_ring *rx_ring)
1456{
1457	memset(rx_ring->rx_bi, 0, sizeof(*rx_ring->rx_bi) * rx_ring->count);
1458}
1459
1460/**
1461 * i40e_clean_rx_ring - Free Rx buffers
1462 * @rx_ring: ring to be cleaned
1463 **/
1464void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1465{
1466	u16 i;
1467
1468	/* ring already cleared, nothing to do */
1469	if (!rx_ring->rx_bi)
1470		return;
1471
 
 
 
1472	if (rx_ring->xsk_pool) {
1473		i40e_xsk_clean_rx_ring(rx_ring);
1474		goto skip_free;
1475	}
1476
1477	/* Free all the Rx ring sk_buffs */
1478	for (i = 0; i < rx_ring->count; i++) {
1479		struct i40e_rx_buffer *rx_bi = i40e_rx_bi(rx_ring, i);
1480
1481		if (!rx_bi->page)
1482			continue;
1483
1484		/* Invalidate cache lines that may have been written to by
1485		 * device so that we avoid corrupting memory.
1486		 */
1487		dma_sync_single_range_for_cpu(rx_ring->dev,
1488					      rx_bi->dma,
1489					      rx_bi->page_offset,
1490					      rx_ring->rx_buf_len,
1491					      DMA_FROM_DEVICE);
1492
1493		/* free resources associated with mapping */
1494		dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1495				     i40e_rx_pg_size(rx_ring),
1496				     DMA_FROM_DEVICE,
1497				     I40E_RX_DMA_ATTR);
1498
1499		__page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1500
1501		rx_bi->page = NULL;
1502		rx_bi->page_offset = 0;
1503	}
1504
1505skip_free:
1506	if (rx_ring->xsk_pool)
1507		i40e_clear_rx_bi_zc(rx_ring);
1508	else
1509		i40e_clear_rx_bi(rx_ring);
1510
1511	/* Zero out the descriptor ring */
1512	memset(rx_ring->desc, 0, rx_ring->size);
1513
1514	rx_ring->next_to_alloc = 0;
1515	rx_ring->next_to_clean = 0;
1516	rx_ring->next_to_process = 0;
1517	rx_ring->next_to_use = 0;
1518}
1519
1520/**
1521 * i40e_free_rx_resources - Free Rx resources
1522 * @rx_ring: ring to clean the resources from
1523 *
1524 * Free all receive software resources
1525 **/
1526void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1527{
1528	i40e_clean_rx_ring(rx_ring);
1529	if (rx_ring->vsi->type == I40E_VSI_MAIN)
1530		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
1531	rx_ring->xdp_prog = NULL;
1532	kfree(rx_ring->rx_bi);
1533	rx_ring->rx_bi = NULL;
1534
1535	if (rx_ring->desc) {
1536		dma_free_coherent(rx_ring->dev, rx_ring->size,
1537				  rx_ring->desc, rx_ring->dma);
1538		rx_ring->desc = NULL;
1539	}
1540}
1541
1542/**
1543 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1544 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1545 *
1546 * Returns 0 on success, negative on failure
1547 **/
1548int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1549{
1550	struct device *dev = rx_ring->dev;
 
1551
1552	u64_stats_init(&rx_ring->syncp);
1553
1554	/* Round up to nearest 4K */
1555	rx_ring->size = rx_ring->count * sizeof(union i40e_rx_desc);
1556	rx_ring->size = ALIGN(rx_ring->size, 4096);
1557	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1558					   &rx_ring->dma, GFP_KERNEL);
1559
1560	if (!rx_ring->desc) {
1561		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1562			 rx_ring->size);
1563		return -ENOMEM;
1564	}
1565
1566	rx_ring->next_to_alloc = 0;
1567	rx_ring->next_to_clean = 0;
1568	rx_ring->next_to_process = 0;
1569	rx_ring->next_to_use = 0;
1570
 
 
 
 
 
 
 
 
1571	rx_ring->xdp_prog = rx_ring->vsi->xdp_prog;
1572
1573	rx_ring->rx_bi =
1574		kcalloc(rx_ring->count, sizeof(*rx_ring->rx_bi), GFP_KERNEL);
1575	if (!rx_ring->rx_bi)
1576		return -ENOMEM;
1577
1578	return 0;
1579}
1580
1581/**
1582 * i40e_release_rx_desc - Store the new tail and head values
1583 * @rx_ring: ring to bump
1584 * @val: new head index
1585 **/
1586void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1587{
1588	rx_ring->next_to_use = val;
1589
1590	/* update next to alloc since we have filled the ring */
1591	rx_ring->next_to_alloc = val;
1592
1593	/* Force memory writes to complete before letting h/w
1594	 * know there are new descriptors to fetch.  (Only
1595	 * applicable for weak-ordered memory model archs,
1596	 * such as IA-64).
1597	 */
1598	wmb();
1599	writel(val, rx_ring->tail);
1600}
1601
1602#if (PAGE_SIZE >= 8192)
1603static unsigned int i40e_rx_frame_truesize(struct i40e_ring *rx_ring,
1604					   unsigned int size)
1605{
1606	unsigned int truesize;
1607
 
 
 
1608	truesize = rx_ring->rx_offset ?
1609		SKB_DATA_ALIGN(size + rx_ring->rx_offset) +
1610		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1611		SKB_DATA_ALIGN(size);
 
1612	return truesize;
1613}
1614#endif
1615
1616/**
1617 * i40e_alloc_mapped_page - recycle or make a new page
1618 * @rx_ring: ring to use
1619 * @bi: rx_buffer struct to modify
1620 *
1621 * Returns true if the page was successfully allocated or
1622 * reused.
1623 **/
1624static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1625				   struct i40e_rx_buffer *bi)
1626{
1627	struct page *page = bi->page;
1628	dma_addr_t dma;
1629
1630	/* since we are recycling buffers we should seldom need to alloc */
1631	if (likely(page)) {
1632		rx_ring->rx_stats.page_reuse_count++;
1633		return true;
1634	}
1635
1636	/* alloc new page for storage */
1637	page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1638	if (unlikely(!page)) {
1639		rx_ring->rx_stats.alloc_page_failed++;
1640		return false;
1641	}
1642
1643	rx_ring->rx_stats.page_alloc_count++;
1644
1645	/* map page for use */
1646	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1647				 i40e_rx_pg_size(rx_ring),
1648				 DMA_FROM_DEVICE,
1649				 I40E_RX_DMA_ATTR);
1650
1651	/* if mapping failed free memory back to system since
1652	 * there isn't much point in holding memory we can't use
1653	 */
1654	if (dma_mapping_error(rx_ring->dev, dma)) {
1655		__free_pages(page, i40e_rx_pg_order(rx_ring));
1656		rx_ring->rx_stats.alloc_page_failed++;
1657		return false;
1658	}
1659
1660	bi->dma = dma;
1661	bi->page = page;
1662	bi->page_offset = rx_ring->rx_offset;
1663	page_ref_add(page, USHRT_MAX - 1);
1664	bi->pagecnt_bias = USHRT_MAX;
1665
1666	return true;
1667}
1668
1669/**
1670 * i40e_alloc_rx_buffers - Replace used receive buffers
1671 * @rx_ring: ring to place buffers on
1672 * @cleaned_count: number of buffers to replace
1673 *
1674 * Returns false if all allocations were successful, true if any fail
1675 **/
1676bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1677{
1678	u16 ntu = rx_ring->next_to_use;
1679	union i40e_rx_desc *rx_desc;
1680	struct i40e_rx_buffer *bi;
1681
1682	/* do nothing if no valid netdev defined */
1683	if (!rx_ring->netdev || !cleaned_count)
1684		return false;
1685
1686	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1687	bi = i40e_rx_bi(rx_ring, ntu);
1688
1689	do {
1690		if (!i40e_alloc_mapped_page(rx_ring, bi))
1691			goto no_buffers;
1692
1693		/* sync the buffer for use by the device */
1694		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1695						 bi->page_offset,
1696						 rx_ring->rx_buf_len,
1697						 DMA_FROM_DEVICE);
1698
1699		/* Refresh the desc even if buffer_addrs didn't change
1700		 * because each write-back erases this info.
1701		 */
1702		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1703
1704		rx_desc++;
1705		bi++;
1706		ntu++;
1707		if (unlikely(ntu == rx_ring->count)) {
1708			rx_desc = I40E_RX_DESC(rx_ring, 0);
1709			bi = i40e_rx_bi(rx_ring, 0);
1710			ntu = 0;
1711		}
1712
1713		/* clear the status bits for the next_to_use descriptor */
1714		rx_desc->wb.qword1.status_error_len = 0;
1715
1716		cleaned_count--;
1717	} while (cleaned_count);
1718
1719	if (rx_ring->next_to_use != ntu)
1720		i40e_release_rx_desc(rx_ring, ntu);
1721
1722	return false;
1723
1724no_buffers:
1725	if (rx_ring->next_to_use != ntu)
1726		i40e_release_rx_desc(rx_ring, ntu);
1727
1728	/* make sure to come back via polling to try again after
1729	 * allocation failure
1730	 */
1731	return true;
1732}
1733
1734/**
1735 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1736 * @vsi: the VSI we care about
1737 * @skb: skb currently being received and modified
1738 * @rx_desc: the receive descriptor
1739 **/
1740static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1741				    struct sk_buff *skb,
1742				    union i40e_rx_desc *rx_desc)
1743{
1744	struct i40e_rx_ptype_decoded decoded;
1745	u32 rx_error, rx_status;
1746	bool ipv4, ipv6;
1747	u8 ptype;
1748	u64 qword;
1749
1750	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1751	ptype = FIELD_GET(I40E_RXD_QW1_PTYPE_MASK, qword);
1752	rx_error = FIELD_GET(I40E_RXD_QW1_ERROR_MASK, qword);
1753	rx_status = FIELD_GET(I40E_RXD_QW1_STATUS_MASK, qword);
 
 
1754	decoded = decode_rx_desc_ptype(ptype);
1755
1756	skb->ip_summed = CHECKSUM_NONE;
1757
1758	skb_checksum_none_assert(skb);
1759
1760	/* Rx csum enabled and ip headers found? */
1761	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1762		return;
1763
1764	/* did the hardware decode the packet and checksum? */
1765	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1766		return;
1767
1768	/* both known and outer_ip must be set for the below code to work */
1769	if (!(decoded.known && decoded.outer_ip))
1770		return;
1771
1772	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1773	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1774	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1775	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1776
1777	if (ipv4 &&
1778	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1779			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1780		goto checksum_fail;
1781
1782	/* likely incorrect csum if alternate IP extension headers found */
1783	if (ipv6 &&
1784	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1785		/* don't increment checksum err here, non-fatal err */
1786		return;
1787
1788	/* there was some L4 error, count error and punt packet to the stack */
1789	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1790		goto checksum_fail;
1791
1792	/* handle packets that were not able to be checksummed due
1793	 * to arrival speed, in this case the stack can compute
1794	 * the csum.
1795	 */
1796	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1797		return;
1798
1799	/* If there is an outer header present that might contain a checksum
1800	 * we need to bump the checksum level by 1 to reflect the fact that
1801	 * we are indicating we validated the inner checksum.
1802	 */
1803	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1804		skb->csum_level = 1;
1805
1806	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1807	switch (decoded.inner_prot) {
1808	case I40E_RX_PTYPE_INNER_PROT_TCP:
1809	case I40E_RX_PTYPE_INNER_PROT_UDP:
1810	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1811		skb->ip_summed = CHECKSUM_UNNECESSARY;
1812		fallthrough;
1813	default:
1814		break;
1815	}
1816
1817	return;
1818
1819checksum_fail:
1820	vsi->back->hw_csum_rx_error++;
1821}
1822
1823/**
1824 * i40e_ptype_to_htype - get a hash type
1825 * @ptype: the ptype value from the descriptor
1826 *
1827 * Returns a hash type to be used by skb_set_hash
1828 **/
1829static inline int i40e_ptype_to_htype(u8 ptype)
1830{
1831	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1832
1833	if (!decoded.known)
1834		return PKT_HASH_TYPE_NONE;
1835
1836	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1837	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1838		return PKT_HASH_TYPE_L4;
1839	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1840		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1841		return PKT_HASH_TYPE_L3;
1842	else
1843		return PKT_HASH_TYPE_L2;
1844}
1845
1846/**
1847 * i40e_rx_hash - set the hash value in the skb
1848 * @ring: descriptor ring
1849 * @rx_desc: specific descriptor
1850 * @skb: skb currently being received and modified
1851 * @rx_ptype: Rx packet type
1852 **/
1853static inline void i40e_rx_hash(struct i40e_ring *ring,
1854				union i40e_rx_desc *rx_desc,
1855				struct sk_buff *skb,
1856				u8 rx_ptype)
1857{
1858	u32 hash;
1859	const __le64 rss_mask =
1860		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1861			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1862
1863	if (!(ring->netdev->features & NETIF_F_RXHASH))
1864		return;
1865
1866	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1867		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1868		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1869	}
1870}
1871
1872/**
1873 * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1874 * @rx_ring: rx descriptor ring packet is being transacted on
1875 * @rx_desc: pointer to the EOP Rx descriptor
1876 * @skb: pointer to current skb being populated
1877 *
1878 * This function checks the ring, descriptor, and packet information in
1879 * order to populate the hash, checksum, VLAN, protocol, and
1880 * other fields within the skb.
1881 **/
1882void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1883			     union i40e_rx_desc *rx_desc, struct sk_buff *skb)
1884{
1885	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1886	u32 rx_status = FIELD_GET(I40E_RXD_QW1_STATUS_MASK, qword);
 
1887	u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1888	u32 tsyn = FIELD_GET(I40E_RXD_QW1_STATUS_TSYNINDX_MASK, rx_status);
1889	u8 rx_ptype = FIELD_GET(I40E_RXD_QW1_PTYPE_MASK, qword);
 
 
1890
1891	if (unlikely(tsynvalid))
1892		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1893
1894	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1895
1896	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1897
1898	skb_record_rx_queue(skb, rx_ring->queue_index);
1899
1900	if (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
1901		__le16 vlan_tag = rx_desc->wb.qword0.lo_dword.l2tag1;
1902
1903		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1904				       le16_to_cpu(vlan_tag));
1905	}
1906
1907	/* modifies the skb - consumes the enet header */
1908	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1909}
1910
1911/**
1912 * i40e_cleanup_headers - Correct empty headers
1913 * @rx_ring: rx descriptor ring packet is being transacted on
1914 * @skb: pointer to current skb being fixed
1915 * @rx_desc: pointer to the EOP Rx descriptor
1916 *
1917 * In addition if skb is not at least 60 bytes we need to pad it so that
1918 * it is large enough to qualify as a valid Ethernet frame.
1919 *
1920 * Returns true if an error was encountered and skb was freed.
1921 **/
1922static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
1923				 union i40e_rx_desc *rx_desc)
1924
1925{
1926	/* ERR_MASK will only have valid bits if EOP set, and
1927	 * what we are doing here is actually checking
1928	 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1929	 * the error field
1930	 */
1931	if (unlikely(i40e_test_staterr(rx_desc,
1932				       BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1933		dev_kfree_skb_any(skb);
1934		return true;
1935	}
1936
1937	/* if eth_skb_pad returns an error the skb was freed */
1938	if (eth_skb_pad(skb))
1939		return true;
1940
1941	return false;
1942}
1943
1944/**
1945 * i40e_can_reuse_rx_page - Determine if page can be reused for another Rx
1946 * @rx_buffer: buffer containing the page
1947 * @rx_stats: rx stats structure for the rx ring
 
1948 *
1949 * If page is reusable, we have a green light for calling i40e_reuse_rx_page,
1950 * which will assign the current buffer to the buffer that next_to_alloc is
1951 * pointing to; otherwise, the DMA mapping needs to be destroyed and
1952 * page freed.
1953 *
1954 * rx_stats will be updated to indicate whether the page was waived
1955 * or busy if it could not be reused.
1956 */
1957static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
1958				   struct i40e_rx_queue_stats *rx_stats)
 
1959{
1960	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1961	struct page *page = rx_buffer->page;
1962
1963	/* Is any reuse possible? */
1964	if (!dev_page_is_reusable(page)) {
1965		rx_stats->page_waive_count++;
1966		return false;
1967	}
1968
1969#if (PAGE_SIZE < 8192)
1970	/* if we are only owner of page we can reuse it */
1971	if (unlikely((rx_buffer->page_count - pagecnt_bias) > 1)) {
1972		rx_stats->page_busy_count++;
1973		return false;
1974	}
1975#else
1976#define I40E_LAST_OFFSET \
1977	(SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
1978	if (rx_buffer->page_offset > I40E_LAST_OFFSET) {
1979		rx_stats->page_busy_count++;
1980		return false;
1981	}
1982#endif
1983
1984	/* If we have drained the page fragment pool we need to update
1985	 * the pagecnt_bias and page count so that we fully restock the
1986	 * number of references the driver holds.
1987	 */
1988	if (unlikely(pagecnt_bias == 1)) {
1989		page_ref_add(page, USHRT_MAX - 1);
1990		rx_buffer->pagecnt_bias = USHRT_MAX;
1991	}
1992
1993	return true;
1994}
1995
1996/**
1997 * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
1998 * @rx_buffer: Rx buffer to adjust
1999 * @truesize: Size of adjustment
2000 **/
2001static void i40e_rx_buffer_flip(struct i40e_rx_buffer *rx_buffer,
2002				unsigned int truesize)
 
 
 
 
 
 
 
 
 
2003{
2004#if (PAGE_SIZE < 8192)
 
 
 
 
 
 
 
 
 
 
2005	rx_buffer->page_offset ^= truesize;
2006#else
2007	rx_buffer->page_offset += truesize;
2008#endif
2009}
2010
2011/**
2012 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
2013 * @rx_ring: rx descriptor ring to transact packets on
2014 * @size: size of buffer to add to skb
 
2015 *
2016 * This function will pull an Rx buffer from the ring and synchronize it
2017 * for use by the CPU.
2018 */
2019static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
2020						 const unsigned int size)
 
2021{
2022	struct i40e_rx_buffer *rx_buffer;
2023
2024	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_process);
2025	rx_buffer->page_count =
2026#if (PAGE_SIZE < 8192)
2027		page_count(rx_buffer->page);
2028#else
2029		0;
2030#endif
2031	prefetch_page_address(rx_buffer->page);
2032
2033	/* we are reusing so sync this buffer for CPU use */
2034	dma_sync_single_range_for_cpu(rx_ring->dev,
2035				      rx_buffer->dma,
2036				      rx_buffer->page_offset,
2037				      size,
2038				      DMA_FROM_DEVICE);
2039
2040	/* We have pulled a buffer for use, so decrement pagecnt_bias */
2041	rx_buffer->pagecnt_bias--;
2042
2043	return rx_buffer;
2044}
2045
2046/**
2047 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
2048 * @rx_ring: rx descriptor ring to transact packets on
2049 * @rx_buffer: rx buffer to pull data from
2050 *
2051 * This function will clean up the contents of the rx_buffer.  It will
2052 * either recycle the buffer or unmap it and free the associated resources.
2053 */
2054static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
2055			       struct i40e_rx_buffer *rx_buffer)
2056{
2057	if (i40e_can_reuse_rx_page(rx_buffer, &rx_ring->rx_stats)) {
2058		/* hand second half of page back to the ring */
2059		i40e_reuse_rx_page(rx_ring, rx_buffer);
2060	} else {
2061		/* we are not reusing the buffer so unmap it */
2062		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2063				     i40e_rx_pg_size(rx_ring),
2064				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
2065		__page_frag_cache_drain(rx_buffer->page,
2066					rx_buffer->pagecnt_bias);
2067		/* clear contents of buffer_info */
2068		rx_buffer->page = NULL;
2069	}
2070}
2071
2072/**
2073 * i40e_process_rx_buffs- Processing of buffers post XDP prog or on error
2074 * @rx_ring: Rx descriptor ring to transact packets on
2075 * @xdp_res: Result of the XDP program
2076 * @xdp: xdp_buff pointing to the data
2077 **/
2078static void i40e_process_rx_buffs(struct i40e_ring *rx_ring, int xdp_res,
2079				  struct xdp_buff *xdp)
2080{
2081	u32 nr_frags = xdp_get_shared_info_from_buff(xdp)->nr_frags;
2082	u32 next = rx_ring->next_to_clean, i = 0;
2083	struct i40e_rx_buffer *rx_buffer;
2084
2085	xdp->flags = 0;
2086
2087	while (1) {
2088		rx_buffer = i40e_rx_bi(rx_ring, next);
2089		if (++next == rx_ring->count)
2090			next = 0;
2091
2092		if (!rx_buffer->page)
2093			continue;
2094
2095		if (xdp_res != I40E_XDP_CONSUMED)
2096			i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
2097		else if (i++ <= nr_frags)
2098			rx_buffer->pagecnt_bias++;
2099
2100		/* EOP buffer will be put in i40e_clean_rx_irq() */
2101		if (next == rx_ring->next_to_process)
2102			return;
2103
2104		i40e_put_rx_buffer(rx_ring, rx_buffer);
2105	}
2106}
2107
2108/**
2109 * i40e_construct_skb - Allocate skb and populate it
2110 * @rx_ring: rx descriptor ring to transact packets on
 
2111 * @xdp: xdp_buff pointing to the data
2112 *
2113 * This function allocates an skb.  It then populates it with the page
2114 * data from the current receive descriptor, taking care to set up the
2115 * skb correctly.
2116 */
2117static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
 
2118					  struct xdp_buff *xdp)
2119{
2120	unsigned int size = xdp->data_end - xdp->data;
2121	struct i40e_rx_buffer *rx_buffer;
2122	struct skb_shared_info *sinfo;
 
 
 
2123	unsigned int headlen;
2124	struct sk_buff *skb;
2125	u32 nr_frags = 0;
2126
2127	/* prefetch first cache line of first page */
2128	net_prefetch(xdp->data);
2129
2130	/* Note, we get here by enabling legacy-rx via:
2131	 *
2132	 *    ethtool --set-priv-flags <dev> legacy-rx on
2133	 *
2134	 * In this mode, we currently get 0 extra XDP headroom as
2135	 * opposed to having legacy-rx off, where we process XDP
2136	 * packets going to stack via i40e_build_skb(). The latter
2137	 * provides us currently with 192 bytes of headroom.
2138	 *
2139	 * For i40e_construct_skb() mode it means that the
2140	 * xdp->data_meta will always point to xdp->data, since
2141	 * the helper cannot expand the head. Should this ever
2142	 * change in future for legacy-rx mode on, then lets also
2143	 * add xdp->data_meta handling here.
2144	 */
2145
2146	/* allocate a skb to store the frags */
2147	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
2148			       I40E_RX_HDR_SIZE,
2149			       GFP_ATOMIC | __GFP_NOWARN);
2150	if (unlikely(!skb))
2151		return NULL;
2152
2153	/* Determine available headroom for copy */
2154	headlen = size;
2155	if (headlen > I40E_RX_HDR_SIZE)
2156		headlen = eth_get_headlen(skb->dev, xdp->data,
2157					  I40E_RX_HDR_SIZE);
2158
2159	/* align pull length to size of long to optimize memcpy performance */
2160	memcpy(__skb_put(skb, headlen), xdp->data,
2161	       ALIGN(headlen, sizeof(long)));
2162
2163	if (unlikely(xdp_buff_has_frags(xdp))) {
2164		sinfo = xdp_get_shared_info_from_buff(xdp);
2165		nr_frags = sinfo->nr_frags;
2166	}
2167	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2168	/* update all of the pointers */
2169	size -= headlen;
2170	if (size) {
2171		if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2172			dev_kfree_skb(skb);
2173			return NULL;
2174		}
2175		skb_add_rx_frag(skb, 0, rx_buffer->page,
2176				rx_buffer->page_offset + headlen,
2177				size, xdp->frame_sz);
 
2178		/* buffer is used by skb, update page_offset */
2179		i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
 
 
 
 
2180	} else {
2181		/* buffer is unused, reset bias back to rx_buffer */
2182		rx_buffer->pagecnt_bias++;
2183	}
2184
2185	if (unlikely(xdp_buff_has_frags(xdp))) {
2186		struct skb_shared_info *skinfo = skb_shinfo(skb);
2187
2188		memcpy(&skinfo->frags[skinfo->nr_frags], &sinfo->frags[0],
2189		       sizeof(skb_frag_t) * nr_frags);
2190
2191		xdp_update_skb_shared_info(skb, skinfo->nr_frags + nr_frags,
2192					   sinfo->xdp_frags_size,
2193					   nr_frags * xdp->frame_sz,
2194					   xdp_buff_is_frag_pfmemalloc(xdp));
2195
2196		/* First buffer has already been processed, so bump ntc */
2197		if (++rx_ring->next_to_clean == rx_ring->count)
2198			rx_ring->next_to_clean = 0;
2199
2200		i40e_process_rx_buffs(rx_ring, I40E_XDP_PASS, xdp);
2201	}
2202
2203	return skb;
2204}
2205
2206/**
2207 * i40e_build_skb - Build skb around an existing buffer
2208 * @rx_ring: Rx descriptor ring to transact packets on
 
2209 * @xdp: xdp_buff pointing to the data
2210 *
2211 * This function builds an skb around an existing Rx buffer, taking care
2212 * to set up the skb correctly and avoid any memcpy overhead.
2213 */
2214static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
 
2215				      struct xdp_buff *xdp)
2216{
2217	unsigned int metasize = xdp->data - xdp->data_meta;
2218	struct skb_shared_info *sinfo;
 
 
 
 
 
 
2219	struct sk_buff *skb;
2220	u32 nr_frags;
2221
2222	/* Prefetch first cache line of first page. If xdp->data_meta
2223	 * is unused, this points exactly as xdp->data, otherwise we
2224	 * likely have a consumer accessing first few bytes of meta
2225	 * data, and then actual data.
2226	 */
2227	net_prefetch(xdp->data_meta);
2228
2229	if (unlikely(xdp_buff_has_frags(xdp))) {
2230		sinfo = xdp_get_shared_info_from_buff(xdp);
2231		nr_frags = sinfo->nr_frags;
2232	}
2233
2234	/* build an skb around the page buffer */
2235	skb = napi_build_skb(xdp->data_hard_start, xdp->frame_sz);
2236	if (unlikely(!skb))
2237		return NULL;
2238
2239	/* update pointers within the skb to store the data */
2240	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2241	__skb_put(skb, xdp->data_end - xdp->data);
2242	if (metasize)
2243		skb_metadata_set(skb, metasize);
2244
2245	if (unlikely(xdp_buff_has_frags(xdp))) {
2246		xdp_update_skb_shared_info(skb, nr_frags,
2247					   sinfo->xdp_frags_size,
2248					   nr_frags * xdp->frame_sz,
2249					   xdp_buff_is_frag_pfmemalloc(xdp));
 
2250
2251		i40e_process_rx_buffs(rx_ring, I40E_XDP_PASS, xdp);
2252	} else {
2253		struct i40e_rx_buffer *rx_buffer;
2254
2255		rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2256		/* buffer is used by skb, update page_offset */
2257		i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2258	}
2259
2260	return skb;
2261}
2262
2263/**
2264 * i40e_is_non_eop - process handling of non-EOP buffers
2265 * @rx_ring: Rx ring being processed
2266 * @rx_desc: Rx descriptor for current buffer
2267 *
2268 * If the buffer is an EOP buffer, this function exits returning false,
2269 * otherwise return true indicating that this is in fact a non-EOP buffer.
2270 */
2271bool i40e_is_non_eop(struct i40e_ring *rx_ring,
2272		     union i40e_rx_desc *rx_desc)
2273{
2274	/* if we are the last buffer then there is nothing else to do */
2275#define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
2276	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
2277		return false;
2278
2279	rx_ring->rx_stats.non_eop_descs++;
2280
2281	return true;
2282}
2283
2284static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
2285			      struct i40e_ring *xdp_ring);
2286
2287int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp, struct i40e_ring *xdp_ring)
2288{
2289	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2290
2291	if (unlikely(!xdpf))
2292		return I40E_XDP_CONSUMED;
2293
2294	return i40e_xmit_xdp_ring(xdpf, xdp_ring);
2295}
2296
2297/**
2298 * i40e_run_xdp - run an XDP program
2299 * @rx_ring: Rx ring being processed
2300 * @xdp: XDP buffer containing the frame
2301 * @xdp_prog: XDP program to run
2302 **/
2303static int i40e_run_xdp(struct i40e_ring *rx_ring, struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
2304{
2305	int err, result = I40E_XDP_PASS;
2306	struct i40e_ring *xdp_ring;
2307	u32 act;
2308
2309	if (!xdp_prog)
2310		goto xdp_out;
2311
2312	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2313
2314	act = bpf_prog_run_xdp(xdp_prog, xdp);
2315	switch (act) {
2316	case XDP_PASS:
2317		break;
2318	case XDP_TX:
2319		xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2320		result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
2321		if (result == I40E_XDP_CONSUMED)
2322			goto out_failure;
2323		break;
2324	case XDP_REDIRECT:
2325		err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
2326		if (err)
2327			goto out_failure;
2328		result = I40E_XDP_REDIR;
2329		break;
2330	default:
2331		bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
2332		fallthrough;
2333	case XDP_ABORTED:
2334out_failure:
2335		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2336		fallthrough; /* handle aborts by dropping packet */
2337	case XDP_DROP:
2338		result = I40E_XDP_CONSUMED;
2339		break;
2340	}
2341xdp_out:
2342	return result;
2343}
2344
2345/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2346 * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
2347 * @xdp_ring: XDP Tx ring
2348 *
2349 * This function updates the XDP Tx ring tail register.
2350 **/
2351void i40e_xdp_ring_update_tail(struct i40e_ring *xdp_ring)
2352{
2353	/* Force memory writes to complete before letting h/w
2354	 * know there are new descriptors to fetch.
2355	 */
2356	wmb();
2357	writel_relaxed(xdp_ring->next_to_use, xdp_ring->tail);
2358}
2359
2360/**
2361 * i40e_update_rx_stats - Update Rx ring statistics
2362 * @rx_ring: rx descriptor ring
2363 * @total_rx_bytes: number of bytes received
2364 * @total_rx_packets: number of packets received
2365 *
2366 * This function updates the Rx ring statistics.
2367 **/
2368void i40e_update_rx_stats(struct i40e_ring *rx_ring,
2369			  unsigned int total_rx_bytes,
2370			  unsigned int total_rx_packets)
2371{
2372	u64_stats_update_begin(&rx_ring->syncp);
2373	rx_ring->stats.packets += total_rx_packets;
2374	rx_ring->stats.bytes += total_rx_bytes;
2375	u64_stats_update_end(&rx_ring->syncp);
2376	rx_ring->q_vector->rx.total_packets += total_rx_packets;
2377	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2378}
2379
2380/**
2381 * i40e_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map
2382 * @rx_ring: Rx ring
2383 * @xdp_res: Result of the receive batch
2384 *
2385 * This function bumps XDP Tx tail and/or flush redirect map, and
2386 * should be called when a batch of packets has been processed in the
2387 * napi loop.
2388 **/
2389void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
2390{
2391	if (xdp_res & I40E_XDP_REDIR)
2392		xdp_do_flush();
2393
2394	if (xdp_res & I40E_XDP_TX) {
2395		struct i40e_ring *xdp_ring =
2396			rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2397
2398		i40e_xdp_ring_update_tail(xdp_ring);
2399	}
2400}
2401
2402/**
2403 * i40e_inc_ntp: Advance the next_to_process index
2404 * @rx_ring: Rx ring
2405 **/
2406static void i40e_inc_ntp(struct i40e_ring *rx_ring)
2407{
2408	u32 ntp = rx_ring->next_to_process + 1;
2409
2410	ntp = (ntp < rx_ring->count) ? ntp : 0;
2411	rx_ring->next_to_process = ntp;
2412	prefetch(I40E_RX_DESC(rx_ring, ntp));
2413}
2414
2415/**
2416 * i40e_add_xdp_frag: Add a frag to xdp_buff
2417 * @xdp: xdp_buff pointing to the data
2418 * @nr_frags: return number of buffers for the packet
2419 * @rx_buffer: rx_buffer holding data of the current frag
2420 * @size: size of data of current frag
2421 */
2422static int i40e_add_xdp_frag(struct xdp_buff *xdp, u32 *nr_frags,
2423			     struct i40e_rx_buffer *rx_buffer, u32 size)
2424{
2425	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
2426
2427	if (!xdp_buff_has_frags(xdp)) {
2428		sinfo->nr_frags = 0;
2429		sinfo->xdp_frags_size = 0;
2430		xdp_buff_set_frags_flag(xdp);
2431	} else if (unlikely(sinfo->nr_frags >= MAX_SKB_FRAGS)) {
2432		/* Overflowing packet: All frags need to be dropped */
2433		return -ENOMEM;
2434	}
2435
2436	__skb_fill_page_desc_noacc(sinfo, sinfo->nr_frags++, rx_buffer->page,
2437				   rx_buffer->page_offset, size);
2438
2439	sinfo->xdp_frags_size += size;
2440
2441	if (page_is_pfmemalloc(rx_buffer->page))
2442		xdp_buff_set_frag_pfmemalloc(xdp);
2443	*nr_frags = sinfo->nr_frags;
2444
2445	return 0;
2446}
2447
2448/**
2449 * i40e_consume_xdp_buff - Consume all the buffers of the packet and update ntc
2450 * @rx_ring: rx descriptor ring to transact packets on
2451 * @xdp: xdp_buff pointing to the data
2452 * @rx_buffer: rx_buffer of eop desc
2453 */
2454static void i40e_consume_xdp_buff(struct i40e_ring *rx_ring,
2455				  struct xdp_buff *xdp,
2456				  struct i40e_rx_buffer *rx_buffer)
2457{
2458	i40e_process_rx_buffs(rx_ring, I40E_XDP_CONSUMED, xdp);
2459	i40e_put_rx_buffer(rx_ring, rx_buffer);
2460	rx_ring->next_to_clean = rx_ring->next_to_process;
2461	xdp->data = NULL;
2462}
2463
2464/**
2465 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2466 * @rx_ring: rx descriptor ring to transact packets on
2467 * @budget: Total limit on number of packets to process
2468 * @rx_cleaned: Out parameter of the number of packets processed
2469 *
2470 * This function provides a "bounce buffer" approach to Rx interrupt
2471 * processing.  The advantage to this is that on systems that have
2472 * expensive overhead for IOMMU access this provides a means of avoiding
2473 * it by maintaining the mapping of the page to the system.
2474 *
2475 * Returns amount of work completed
2476 **/
2477static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
2478			     unsigned int *rx_cleaned)
2479{
2480	unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2481	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
2482	u16 clean_threshold = rx_ring->count / 2;
2483	unsigned int offset = rx_ring->rx_offset;
2484	struct xdp_buff *xdp = &rx_ring->xdp;
2485	unsigned int xdp_xmit = 0;
2486	struct bpf_prog *xdp_prog;
2487	bool failure = false;
 
2488	int xdp_res = 0;
2489
 
 
 
 
 
2490	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2491
2492	while (likely(total_rx_packets < (unsigned int)budget)) {
2493		u16 ntp = rx_ring->next_to_process;
2494		struct i40e_rx_buffer *rx_buffer;
2495		union i40e_rx_desc *rx_desc;
2496		struct sk_buff *skb;
2497		unsigned int size;
2498		u32 nfrags = 0;
2499		bool neop;
2500		u64 qword;
2501
2502		/* return some buffers to hardware, one at a time is too slow */
2503		if (cleaned_count >= clean_threshold) {
2504			failure = failure ||
2505				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
2506			cleaned_count = 0;
2507		}
2508
2509		rx_desc = I40E_RX_DESC(rx_ring, ntp);
2510
2511		/* status_error_len will always be zero for unused descriptors
2512		 * because it's cleared in cleanup, and overlaps with hdr_addr
2513		 * which is always zero because packet split isn't used, if the
2514		 * hardware wrote DD then the length will be non-zero
2515		 */
2516		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2517
2518		/* This memory barrier is needed to keep us from reading
2519		 * any other fields out of the rx_desc until we have
2520		 * verified the descriptor has been written back.
2521		 */
2522		dma_rmb();
2523
2524		if (i40e_rx_is_programming_status(qword)) {
2525			i40e_clean_programming_status(rx_ring,
2526						      rx_desc->raw.qword[0],
2527						      qword);
2528			rx_buffer = i40e_rx_bi(rx_ring, ntp);
2529			i40e_inc_ntp(rx_ring);
2530			i40e_reuse_rx_page(rx_ring, rx_buffer);
2531			/* Update ntc and bump cleaned count if not in the
2532			 * middle of mb packet.
2533			 */
2534			if (rx_ring->next_to_clean == ntp) {
2535				rx_ring->next_to_clean =
2536					rx_ring->next_to_process;
2537				cleaned_count++;
2538			}
2539			continue;
2540		}
2541
2542		size = FIELD_GET(I40E_RXD_QW1_LENGTH_PBUF_MASK, qword);
 
2543		if (!size)
2544			break;
2545
2546		i40e_trace(clean_rx_irq, rx_ring, rx_desc, xdp);
2547		/* retrieve a buffer from the ring */
2548		rx_buffer = i40e_get_rx_buffer(rx_ring, size);
2549
2550		neop = i40e_is_non_eop(rx_ring, rx_desc);
2551		i40e_inc_ntp(rx_ring);
2552
2553		if (!xdp->data) {
 
2554			unsigned char *hard_start;
2555
2556			hard_start = page_address(rx_buffer->page) +
2557				     rx_buffer->page_offset - offset;
2558			xdp_prepare_buff(xdp, hard_start, offset, size, true);
 
2559#if (PAGE_SIZE > 4096)
2560			/* At larger PAGE_SIZE, frame_sz depend on len size */
2561			xdp->frame_sz = i40e_rx_frame_truesize(rx_ring, size);
2562#endif
2563		} else if (i40e_add_xdp_frag(xdp, &nfrags, rx_buffer, size) &&
2564			   !neop) {
2565			/* Overflowing packet: Drop all frags on EOP */
2566			i40e_consume_xdp_buff(rx_ring, xdp, rx_buffer);
2567			break;
2568		}
2569
2570		if (neop)
2571			continue;
2572
2573		xdp_res = i40e_run_xdp(rx_ring, xdp, xdp_prog);
2574
2575		if (xdp_res) {
2576			xdp_xmit |= xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR);
2577
2578			if (unlikely(xdp_buff_has_frags(xdp))) {
2579				i40e_process_rx_buffs(rx_ring, xdp_res, xdp);
2580				size = xdp_get_buff_len(xdp);
2581			} else if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
2582				i40e_rx_buffer_flip(rx_buffer, xdp->frame_sz);
2583			} else {
2584				rx_buffer->pagecnt_bias++;
2585			}
2586			total_rx_bytes += size;
 
 
 
 
 
2587		} else {
2588			if (ring_uses_build_skb(rx_ring))
2589				skb = i40e_build_skb(rx_ring, xdp);
2590			else
2591				skb = i40e_construct_skb(rx_ring, xdp);
2592
2593			/* drop if we failed to retrieve a buffer */
2594			if (!skb) {
2595				rx_ring->rx_stats.alloc_buff_failed++;
2596				i40e_consume_xdp_buff(rx_ring, xdp, rx_buffer);
2597				break;
2598			}
2599
2600			if (i40e_cleanup_headers(rx_ring, skb, rx_desc))
2601				goto process_next;
 
 
 
 
2602
2603			/* probably a little skewed due to removing CRC */
2604			total_rx_bytes += skb->len;
2605
2606			/* populate checksum, VLAN, and protocol */
2607			i40e_process_skb_fields(rx_ring, rx_desc, skb);
 
2608
2609			i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, xdp);
2610			napi_gro_receive(&rx_ring->q_vector->napi, skb);
 
2611		}
2612
 
 
 
 
 
 
 
 
 
 
2613		/* update budget accounting */
2614		total_rx_packets++;
2615process_next:
2616		cleaned_count += nfrags + 1;
2617		i40e_put_rx_buffer(rx_ring, rx_buffer);
2618		rx_ring->next_to_clean = rx_ring->next_to_process;
2619
2620		xdp->data = NULL;
2621	}
2622
2623	i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
 
2624
2625	i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
2626
2627	*rx_cleaned = total_rx_packets;
2628
2629	/* guarantee a trip back through this routine if there was a failure */
2630	return failure ? budget : (int)total_rx_packets;
2631}
2632
2633static inline u32 i40e_buildreg_itr(const int type, u16 itr)
2634{
2635	u32 val;
2636
2637	/* We don't bother with setting the CLEARPBA bit as the data sheet
2638	 * points out doing so is "meaningless since it was already
2639	 * auto-cleared". The auto-clearing happens when the interrupt is
2640	 * asserted.
2641	 *
2642	 * Hardware errata 28 for also indicates that writing to a
2643	 * xxINT_DYN_CTLx CSR with INTENA_MSK (bit 31) set to 0 will clear
2644	 * an event in the PBA anyway so we need to rely on the automask
2645	 * to hold pending events for us until the interrupt is re-enabled
2646	 *
2647	 * The itr value is reported in microseconds, and the register
2648	 * value is recorded in 2 microsecond units. For this reason we
2649	 * only need to shift by the interval shift - 1 instead of the
2650	 * full value.
2651	 */
2652	itr &= I40E_ITR_MASK;
2653
2654	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2655	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2656	      (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1));
2657
2658	return val;
2659}
2660
2661/* a small macro to shorten up some long lines */
2662#define INTREG I40E_PFINT_DYN_CTLN
2663
2664/* The act of updating the ITR will cause it to immediately trigger. In order
2665 * to prevent this from throwing off adaptive update statistics we defer the
2666 * update so that it can only happen so often. So after either Tx or Rx are
2667 * updated we make the adaptive scheme wait until either the ITR completely
2668 * expires via the next_update expiration or we have been through at least
2669 * 3 interrupts.
2670 */
2671#define ITR_COUNTDOWN_START 3
2672
2673/**
2674 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2675 * @vsi: the VSI we care about
2676 * @q_vector: q_vector for which itr is being updated and interrupt enabled
2677 *
2678 **/
2679static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2680					  struct i40e_q_vector *q_vector)
2681{
2682	struct i40e_hw *hw = &vsi->back->hw;
2683	u32 intval;
2684
2685	/* If we don't have MSIX, then we only need to re-enable icr0 */
2686	if (!test_bit(I40E_FLAG_MSIX_ENA, vsi->back->flags)) {
2687		i40e_irq_dynamic_enable_icr0(vsi->back);
2688		return;
2689	}
2690
2691	/* These will do nothing if dynamic updates are not enabled */
2692	i40e_update_itr(q_vector, &q_vector->tx);
2693	i40e_update_itr(q_vector, &q_vector->rx);
2694
2695	/* This block of logic allows us to get away with only updating
2696	 * one ITR value with each interrupt. The idea is to perform a
2697	 * pseudo-lazy update with the following criteria.
2698	 *
2699	 * 1. Rx is given higher priority than Tx if both are in same state
2700	 * 2. If we must reduce an ITR that is given highest priority.
2701	 * 3. We then give priority to increasing ITR based on amount.
2702	 */
2703	if (q_vector->rx.target_itr < q_vector->rx.current_itr) {
2704		/* Rx ITR needs to be reduced, this is highest priority */
2705		intval = i40e_buildreg_itr(I40E_RX_ITR,
2706					   q_vector->rx.target_itr);
2707		q_vector->rx.current_itr = q_vector->rx.target_itr;
2708		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2709	} else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) ||
2710		   ((q_vector->rx.target_itr - q_vector->rx.current_itr) <
2711		    (q_vector->tx.target_itr - q_vector->tx.current_itr))) {
2712		/* Tx ITR needs to be reduced, this is second priority
2713		 * Tx ITR needs to be increased more than Rx, fourth priority
2714		 */
2715		intval = i40e_buildreg_itr(I40E_TX_ITR,
2716					   q_vector->tx.target_itr);
2717		q_vector->tx.current_itr = q_vector->tx.target_itr;
2718		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2719	} else if (q_vector->rx.current_itr != q_vector->rx.target_itr) {
2720		/* Rx ITR needs to be increased, third priority */
2721		intval = i40e_buildreg_itr(I40E_RX_ITR,
2722					   q_vector->rx.target_itr);
2723		q_vector->rx.current_itr = q_vector->rx.target_itr;
2724		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2725	} else {
2726		/* No ITR update, lowest priority */
2727		intval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2728		if (q_vector->itr_countdown)
2729			q_vector->itr_countdown--;
2730	}
2731
2732	if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2733		wr32(hw, INTREG(q_vector->reg_idx), intval);
2734}
2735
2736/**
2737 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2738 * @napi: napi struct with our devices info in it
2739 * @budget: amount of work driver is allowed to do this pass, in packets
2740 *
2741 * This function will clean all queues associated with a q_vector.
2742 *
2743 * Returns the amount of work done
2744 **/
2745int i40e_napi_poll(struct napi_struct *napi, int budget)
2746{
2747	struct i40e_q_vector *q_vector =
2748			       container_of(napi, struct i40e_q_vector, napi);
2749	struct i40e_vsi *vsi = q_vector->vsi;
2750	struct i40e_ring *ring;
2751	bool tx_clean_complete = true;
2752	bool rx_clean_complete = true;
2753	unsigned int tx_cleaned = 0;
2754	unsigned int rx_cleaned = 0;
2755	bool clean_complete = true;
2756	bool arm_wb = false;
2757	int budget_per_ring;
2758	int work_done = 0;
2759
2760	if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2761		napi_complete(napi);
2762		return 0;
2763	}
2764
2765	/* Since the actual Tx work is minimal, we can give the Tx a larger
2766	 * budget and be more aggressive about cleaning up the Tx descriptors.
2767	 */
2768	i40e_for_each_ring(ring, q_vector->tx) {
2769		bool wd = ring->xsk_pool ?
2770			  i40e_clean_xdp_tx_irq(vsi, ring) :
2771			  i40e_clean_tx_irq(vsi, ring, budget, &tx_cleaned);
2772
2773		if (!wd) {
2774			clean_complete = tx_clean_complete = false;
2775			continue;
2776		}
2777		arm_wb |= ring->arm_wb;
2778		ring->arm_wb = false;
2779	}
2780
2781	/* Handle case where we are called by netpoll with a budget of 0 */
2782	if (budget <= 0)
2783		goto tx_only;
2784
2785	/* normally we have 1 Rx ring per q_vector */
2786	if (unlikely(q_vector->num_ringpairs > 1))
2787		/* We attempt to distribute budget to each Rx queue fairly, but
2788		 * don't allow the budget to go below 1 because that would exit
2789		 * polling early.
2790		 */
2791		budget_per_ring = max_t(int, budget / q_vector->num_ringpairs, 1);
2792	else
2793		/* Max of 1 Rx ring in this q_vector so give it the budget */
2794		budget_per_ring = budget;
2795
2796	i40e_for_each_ring(ring, q_vector->rx) {
2797		int cleaned = ring->xsk_pool ?
2798			      i40e_clean_rx_irq_zc(ring, budget_per_ring) :
2799			      i40e_clean_rx_irq(ring, budget_per_ring, &rx_cleaned);
2800
2801		work_done += cleaned;
2802		/* if we clean as many as budgeted, we must not be done */
2803		if (cleaned >= budget_per_ring)
2804			clean_complete = rx_clean_complete = false;
2805	}
2806
2807	if (!i40e_enabled_xdp_vsi(vsi))
2808		trace_i40e_napi_poll(napi, q_vector, budget, budget_per_ring, rx_cleaned,
2809				     tx_cleaned, rx_clean_complete, tx_clean_complete);
2810
2811	/* If work not completed, return budget and polling will return */
2812	if (!clean_complete) {
2813		int cpu_id = smp_processor_id();
2814
2815		/* It is possible that the interrupt affinity has changed but,
2816		 * if the cpu is pegged at 100%, polling will never exit while
2817		 * traffic continues and the interrupt will be stuck on this
2818		 * cpu.  We check to make sure affinity is correct before we
2819		 * continue to poll, otherwise we must stop polling so the
2820		 * interrupt can move to the correct cpu.
2821		 */
2822		if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
2823			/* Tell napi that we are done polling */
2824			napi_complete_done(napi, work_done);
2825
2826			/* Force an interrupt */
2827			i40e_force_wb(vsi, q_vector);
2828
2829			/* Return budget-1 so that polling stops */
2830			return budget - 1;
2831		}
2832tx_only:
2833		if (arm_wb) {
2834			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2835			i40e_enable_wb_on_itr(vsi, q_vector);
2836		}
2837		return budget;
2838	}
2839
2840	if (q_vector->tx.ring[0].flags & I40E_TXR_FLAGS_WB_ON_ITR)
2841		q_vector->arm_wb_state = false;
2842
2843	/* Exit the polling mode, but don't re-enable interrupts if stack might
2844	 * poll us due to busy-polling
2845	 */
2846	if (likely(napi_complete_done(napi, work_done)))
2847		i40e_update_enable_itr(vsi, q_vector);
2848
2849	return min(work_done, budget - 1);
2850}
2851
2852/**
2853 * i40e_atr - Add a Flow Director ATR filter
2854 * @tx_ring:  ring to add programming descriptor to
2855 * @skb:      send buffer
2856 * @tx_flags: send tx flags
2857 **/
2858static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2859		     u32 tx_flags)
2860{
2861	struct i40e_filter_program_desc *fdir_desc;
2862	struct i40e_pf *pf = tx_ring->vsi->back;
2863	union {
2864		unsigned char *network;
2865		struct iphdr *ipv4;
2866		struct ipv6hdr *ipv6;
2867	} hdr;
2868	struct tcphdr *th;
2869	unsigned int hlen;
2870	u32 flex_ptype, dtype_cmd;
2871	int l4_proto;
2872	u16 i;
2873
2874	/* make sure ATR is enabled */
2875	if (!test_bit(I40E_FLAG_FD_ATR_ENA, pf->flags))
2876		return;
2877
2878	if (test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2879		return;
2880
2881	/* if sampling is disabled do nothing */
2882	if (!tx_ring->atr_sample_rate)
2883		return;
2884
2885	/* Currently only IPv4/IPv6 with TCP is supported */
2886	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2887		return;
2888
2889	/* snag network header to get L4 type and address */
2890	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2891		      skb_inner_network_header(skb) : skb_network_header(skb);
2892
2893	/* Note: tx_flags gets modified to reflect inner protocols in
2894	 * tx_enable_csum function if encap is enabled.
2895	 */
2896	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2897		/* access ihl as u8 to avoid unaligned access on ia64 */
2898		hlen = (hdr.network[0] & 0x0F) << 2;
2899		l4_proto = hdr.ipv4->protocol;
2900	} else {
2901		/* find the start of the innermost ipv6 header */
2902		unsigned int inner_hlen = hdr.network - skb->data;
2903		unsigned int h_offset = inner_hlen;
2904
2905		/* this function updates h_offset to the end of the header */
2906		l4_proto =
2907		  ipv6_find_hdr(skb, &h_offset, IPPROTO_TCP, NULL, NULL);
2908		/* hlen will contain our best estimate of the tcp header */
2909		hlen = h_offset - inner_hlen;
2910	}
2911
2912	if (l4_proto != IPPROTO_TCP)
2913		return;
2914
2915	th = (struct tcphdr *)(hdr.network + hlen);
2916
2917	/* Due to lack of space, no more new filters can be programmed */
2918	if (th->syn && test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2919		return;
2920	if (test_bit(I40E_FLAG_HW_ATR_EVICT_ENA, pf->flags)) {
2921		/* HW ATR eviction will take care of removing filters on FIN
2922		 * and RST packets.
2923		 */
2924		if (th->fin || th->rst)
2925			return;
2926	}
2927
2928	tx_ring->atr_count++;
2929
2930	/* sample on all syn/fin/rst packets or once every atr sample rate */
2931	if (!th->fin &&
2932	    !th->syn &&
2933	    !th->rst &&
2934	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2935		return;
2936
2937	tx_ring->atr_count = 0;
2938
2939	/* grab the next descriptor */
2940	i = tx_ring->next_to_use;
2941	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2942
2943	i++;
2944	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2945
2946	flex_ptype = FIELD_PREP(I40E_TXD_FLTR_QW0_QINDEX_MASK,
2947				tx_ring->queue_index);
2948	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2949		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2950		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2951		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2952		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2953
2954	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2955
2956	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2957
2958	dtype_cmd |= (th->fin || th->rst) ?
2959		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2960		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2961		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2962		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2963
2964	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2965		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2966
2967	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2968		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2969
2970	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2971	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2972		dtype_cmd |=
2973			FIELD_PREP(I40E_TXD_FLTR_QW1_CNTINDEX_MASK,
2974				   I40E_FD_ATR_STAT_IDX(pf->hw.pf_id));
 
2975	else
2976		dtype_cmd |=
2977			FIELD_PREP(I40E_TXD_FLTR_QW1_CNTINDEX_MASK,
2978				   I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id));
 
2979
2980	if (test_bit(I40E_FLAG_HW_ATR_EVICT_ENA, pf->flags))
2981		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2982
2983	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2984	fdir_desc->rsvd = cpu_to_le32(0);
2985	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2986	fdir_desc->fd_id = cpu_to_le32(0);
2987}
2988
2989/**
2990 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2991 * @skb:     send buffer
2992 * @tx_ring: ring to send buffer on
2993 * @flags:   the tx flags to be set
2994 *
2995 * Checks the skb and set up correspondingly several generic transmit flags
2996 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2997 *
2998 * Returns error code indicate the frame should be dropped upon error and the
2999 * otherwise  returns 0 to indicate the flags has been set properly.
3000 **/
3001static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
3002					     struct i40e_ring *tx_ring,
3003					     u32 *flags)
3004{
3005	__be16 protocol = skb->protocol;
3006	u32  tx_flags = 0;
3007
3008	if (protocol == htons(ETH_P_8021Q) &&
3009	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
3010		/* When HW VLAN acceleration is turned off by the user the
3011		 * stack sets the protocol to 8021q so that the driver
3012		 * can take any steps required to support the SW only
3013		 * VLAN handling.  In our case the driver doesn't need
3014		 * to take any further steps so just set the protocol
3015		 * to the encapsulated ethertype.
3016		 */
3017		skb->protocol = vlan_get_protocol(skb);
3018		goto out;
3019	}
3020
3021	/* if we have a HW VLAN tag being added, default to the HW one */
3022	if (skb_vlan_tag_present(skb)) {
3023		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
3024		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
3025	/* else if it is a SW VLAN, check the next protocol and store the tag */
3026	} else if (protocol == htons(ETH_P_8021Q)) {
3027		struct vlan_hdr *vhdr, _vhdr;
3028
3029		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
3030		if (!vhdr)
3031			return -EINVAL;
3032
3033		protocol = vhdr->h_vlan_encapsulated_proto;
3034		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
3035		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
3036	}
3037
3038	if (!test_bit(I40E_FLAG_DCB_ENA, tx_ring->vsi->back->flags))
3039		goto out;
3040
3041	/* Insert 802.1p priority into VLAN header */
3042	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
3043	    (skb->priority != TC_PRIO_CONTROL)) {
3044		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
3045		tx_flags |= (skb->priority & 0x7) <<
3046				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
3047		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
3048			struct vlan_ethhdr *vhdr;
3049			int rc;
3050
3051			rc = skb_cow_head(skb, 0);
3052			if (rc < 0)
3053				return rc;
3054			vhdr = skb_vlan_eth_hdr(skb);
3055			vhdr->h_vlan_TCI = htons(tx_flags >>
3056						 I40E_TX_FLAGS_VLAN_SHIFT);
3057		} else {
3058			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
3059		}
3060	}
3061
3062out:
3063	*flags = tx_flags;
3064	return 0;
3065}
3066
3067/**
3068 * i40e_tso - set up the tso context descriptor
3069 * @first:    pointer to first Tx buffer for xmit
3070 * @hdr_len:  ptr to the size of the packet header
3071 * @cd_type_cmd_tso_mss: Quad Word 1
3072 *
3073 * Returns 0 if no TSO can happen, 1 if tso is going, or error
3074 **/
3075static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
3076		    u64 *cd_type_cmd_tso_mss)
3077{
3078	struct sk_buff *skb = first->skb;
3079	u64 cd_cmd, cd_tso_len, cd_mss;
3080	__be16 protocol;
3081	union {
3082		struct iphdr *v4;
3083		struct ipv6hdr *v6;
3084		unsigned char *hdr;
3085	} ip;
3086	union {
3087		struct tcphdr *tcp;
3088		struct udphdr *udp;
3089		unsigned char *hdr;
3090	} l4;
3091	u32 paylen, l4_offset;
3092	u16 gso_size;
3093	int err;
3094
3095	if (skb->ip_summed != CHECKSUM_PARTIAL)
3096		return 0;
3097
3098	if (!skb_is_gso(skb))
3099		return 0;
3100
3101	err = skb_cow_head(skb, 0);
3102	if (err < 0)
3103		return err;
3104
3105	protocol = vlan_get_protocol(skb);
3106
3107	if (eth_p_mpls(protocol))
3108		ip.hdr = skb_inner_network_header(skb);
3109	else
3110		ip.hdr = skb_network_header(skb);
3111	l4.hdr = skb_checksum_start(skb);
3112
3113	/* initialize outer IP header fields */
3114	if (ip.v4->version == 4) {
3115		ip.v4->tot_len = 0;
3116		ip.v4->check = 0;
3117
3118		first->tx_flags |= I40E_TX_FLAGS_TSO;
3119	} else {
3120		ip.v6->payload_len = 0;
3121		first->tx_flags |= I40E_TX_FLAGS_TSO;
3122	}
3123
3124	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
3125					 SKB_GSO_GRE_CSUM |
3126					 SKB_GSO_IPXIP4 |
3127					 SKB_GSO_IPXIP6 |
3128					 SKB_GSO_UDP_TUNNEL |
3129					 SKB_GSO_UDP_TUNNEL_CSUM)) {
3130		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3131		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
3132			l4.udp->len = 0;
3133
3134			/* determine offset of outer transport header */
3135			l4_offset = l4.hdr - skb->data;
3136
3137			/* remove payload length from outer checksum */
3138			paylen = skb->len - l4_offset;
3139			csum_replace_by_diff(&l4.udp->check,
3140					     (__force __wsum)htonl(paylen));
3141		}
3142
3143		/* reset pointers to inner headers */
3144		ip.hdr = skb_inner_network_header(skb);
3145		l4.hdr = skb_inner_transport_header(skb);
3146
3147		/* initialize inner IP header fields */
3148		if (ip.v4->version == 4) {
3149			ip.v4->tot_len = 0;
3150			ip.v4->check = 0;
3151		} else {
3152			ip.v6->payload_len = 0;
3153		}
3154	}
3155
3156	/* determine offset of inner transport header */
3157	l4_offset = l4.hdr - skb->data;
3158
3159	/* remove payload length from inner checksum */
3160	paylen = skb->len - l4_offset;
3161
3162	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
3163		csum_replace_by_diff(&l4.udp->check, (__force __wsum)htonl(paylen));
3164		/* compute length of segmentation header */
3165		*hdr_len = sizeof(*l4.udp) + l4_offset;
3166	} else {
3167		csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3168		/* compute length of segmentation header */
3169		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
3170	}
3171
3172	/* pull values out of skb_shinfo */
3173	gso_size = skb_shinfo(skb)->gso_size;
3174
3175	/* update GSO size and bytecount with header size */
3176	first->gso_segs = skb_shinfo(skb)->gso_segs;
3177	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3178
3179	/* find the field values */
3180	cd_cmd = I40E_TX_CTX_DESC_TSO;
3181	cd_tso_len = skb->len - *hdr_len;
3182	cd_mss = gso_size;
3183	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
3184				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
3185				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
3186	return 1;
3187}
3188
3189/**
3190 * i40e_tsyn - set up the tsyn context descriptor
3191 * @tx_ring:  ptr to the ring to send
3192 * @skb:      ptr to the skb we're sending
3193 * @tx_flags: the collected send information
3194 * @cd_type_cmd_tso_mss: Quad Word 1
3195 *
3196 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
3197 **/
3198static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
3199		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
3200{
3201	struct i40e_pf *pf;
3202
3203	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
3204		return 0;
3205
3206	/* Tx timestamps cannot be sampled when doing TSO */
3207	if (tx_flags & I40E_TX_FLAGS_TSO)
3208		return 0;
3209
3210	/* only timestamp the outbound packet if the user has requested it and
3211	 * we are not already transmitting a packet to be timestamped
3212	 */
3213	pf = i40e_netdev_to_pf(tx_ring->netdev);
3214	if (!test_bit(I40E_FLAG_PTP_ENA, pf->flags))
3215		return 0;
3216
3217	if (pf->ptp_tx &&
3218	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
3219		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3220		pf->ptp_tx_start = jiffies;
3221		pf->ptp_tx_skb = skb_get(skb);
3222	} else {
3223		pf->tx_hwtstamp_skipped++;
3224		return 0;
3225	}
3226
3227	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
3228				I40E_TXD_CTX_QW1_CMD_SHIFT;
3229
3230	return 1;
3231}
3232
3233/**
3234 * i40e_tx_enable_csum - Enable Tx checksum offloads
3235 * @skb: send buffer
3236 * @tx_flags: pointer to Tx flags currently set
3237 * @td_cmd: Tx descriptor command bits to set
3238 * @td_offset: Tx descriptor header offsets to set
3239 * @tx_ring: Tx descriptor ring
3240 * @cd_tunneling: ptr to context desc bits
3241 **/
3242static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
3243			       u32 *td_cmd, u32 *td_offset,
3244			       struct i40e_ring *tx_ring,
3245			       u32 *cd_tunneling)
3246{
3247	union {
3248		struct iphdr *v4;
3249		struct ipv6hdr *v6;
3250		unsigned char *hdr;
3251	} ip;
3252	union {
3253		struct tcphdr *tcp;
3254		struct udphdr *udp;
3255		unsigned char *hdr;
3256	} l4;
3257	unsigned char *exthdr;
3258	u32 offset, cmd = 0;
3259	__be16 frag_off;
3260	__be16 protocol;
3261	u8 l4_proto = 0;
3262
3263	if (skb->ip_summed != CHECKSUM_PARTIAL)
3264		return 0;
3265
3266	protocol = vlan_get_protocol(skb);
3267
3268	if (eth_p_mpls(protocol)) {
3269		ip.hdr = skb_inner_network_header(skb);
3270		l4.hdr = skb_checksum_start(skb);
3271	} else {
3272		ip.hdr = skb_network_header(skb);
3273		l4.hdr = skb_transport_header(skb);
3274	}
3275
3276	/* set the tx_flags to indicate the IP protocol type. this is
3277	 * required so that checksum header computation below is accurate.
3278	 */
3279	if (ip.v4->version == 4)
3280		*tx_flags |= I40E_TX_FLAGS_IPV4;
3281	else
3282		*tx_flags |= I40E_TX_FLAGS_IPV6;
3283
3284	/* compute outer L2 header size */
3285	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
3286
3287	if (skb->encapsulation) {
3288		u32 tunnel = 0;
3289		/* define outer network header type */
3290		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3291			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3292				  I40E_TX_CTX_EXT_IP_IPV4 :
3293				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
3294
3295			l4_proto = ip.v4->protocol;
3296		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3297			int ret;
3298
3299			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
3300
3301			exthdr = ip.hdr + sizeof(*ip.v6);
3302			l4_proto = ip.v6->nexthdr;
3303			ret = ipv6_skip_exthdr(skb, exthdr - skb->data,
3304					       &l4_proto, &frag_off);
3305			if (ret < 0)
3306				return -1;
3307		}
3308
3309		/* define outer transport */
3310		switch (l4_proto) {
3311		case IPPROTO_UDP:
3312			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
3313			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3314			break;
3315		case IPPROTO_GRE:
3316			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
3317			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3318			break;
3319		case IPPROTO_IPIP:
3320		case IPPROTO_IPV6:
3321			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3322			l4.hdr = skb_inner_network_header(skb);
3323			break;
3324		default:
3325			if (*tx_flags & I40E_TX_FLAGS_TSO)
3326				return -1;
3327
3328			skb_checksum_help(skb);
3329			return 0;
3330		}
3331
3332		/* compute outer L3 header size */
3333		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
3334			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
3335
3336		/* switch IP header pointer from outer to inner header */
3337		ip.hdr = skb_inner_network_header(skb);
3338
3339		/* compute tunnel header size */
3340		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
3341			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
3342
3343		/* indicate if we need to offload outer UDP header */
3344		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
3345		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3346		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
3347			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
3348
3349		/* record tunnel offload values */
3350		*cd_tunneling |= tunnel;
3351
3352		/* switch L4 header pointer from outer to inner */
3353		l4.hdr = skb_inner_transport_header(skb);
3354		l4_proto = 0;
3355
3356		/* reset type as we transition from outer to inner headers */
3357		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
3358		if (ip.v4->version == 4)
3359			*tx_flags |= I40E_TX_FLAGS_IPV4;
3360		if (ip.v6->version == 6)
3361			*tx_flags |= I40E_TX_FLAGS_IPV6;
3362	}
3363
3364	/* Enable IP checksum offloads */
3365	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3366		l4_proto = ip.v4->protocol;
3367		/* the stack computes the IP header already, the only time we
3368		 * need the hardware to recompute it is in the case of TSO.
3369		 */
3370		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3371		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
3372		       I40E_TX_DESC_CMD_IIPT_IPV4;
3373	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3374		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
3375
3376		exthdr = ip.hdr + sizeof(*ip.v6);
3377		l4_proto = ip.v6->nexthdr;
3378		if (l4.hdr != exthdr)
3379			ipv6_skip_exthdr(skb, exthdr - skb->data,
3380					 &l4_proto, &frag_off);
3381	}
3382
3383	/* compute inner L3 header size */
3384	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
3385
3386	/* Enable L4 checksum offloads */
3387	switch (l4_proto) {
3388	case IPPROTO_TCP:
3389		/* enable checksum offloads */
3390		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
3391		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3392		break;
3393	case IPPROTO_SCTP:
3394		/* enable SCTP checksum offload */
3395		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
3396		offset |= (sizeof(struct sctphdr) >> 2) <<
3397			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3398		break;
3399	case IPPROTO_UDP:
3400		/* enable UDP checksum offload */
3401		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
3402		offset |= (sizeof(struct udphdr) >> 2) <<
3403			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3404		break;
3405	default:
3406		if (*tx_flags & I40E_TX_FLAGS_TSO)
3407			return -1;
3408		skb_checksum_help(skb);
3409		return 0;
3410	}
3411
3412	*td_cmd |= cmd;
3413	*td_offset |= offset;
3414
3415	return 1;
3416}
3417
3418/**
3419 * i40e_create_tx_ctx - Build the Tx context descriptor
3420 * @tx_ring:  ring to create the descriptor on
3421 * @cd_type_cmd_tso_mss: Quad Word 1
3422 * @cd_tunneling: Quad Word 0 - bits 0-31
3423 * @cd_l2tag2: Quad Word 0 - bits 32-63
3424 **/
3425static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
3426			       const u64 cd_type_cmd_tso_mss,
3427			       const u32 cd_tunneling, const u32 cd_l2tag2)
3428{
3429	struct i40e_tx_context_desc *context_desc;
3430	int i = tx_ring->next_to_use;
3431
3432	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
3433	    !cd_tunneling && !cd_l2tag2)
3434		return;
3435
3436	/* grab the next descriptor */
3437	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
3438
3439	i++;
3440	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3441
3442	/* cpu_to_le32 and assign to struct fields */
3443	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
3444	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
3445	context_desc->rsvd = cpu_to_le16(0);
3446	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
3447}
3448
3449/**
3450 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
3451 * @tx_ring: the ring to be checked
3452 * @size:    the size buffer we want to assure is available
3453 *
3454 * Returns -EBUSY if a stop is needed, else 0
3455 **/
3456int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
3457{
3458	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3459	/* Memory barrier before checking head and tail */
3460	smp_mb();
3461
3462	++tx_ring->tx_stats.tx_stopped;
3463
3464	/* Check again in a case another CPU has just made room available. */
3465	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
3466		return -EBUSY;
3467
3468	/* A reprieve! - use start_queue because it doesn't call schedule */
3469	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3470	++tx_ring->tx_stats.restart_queue;
3471	return 0;
3472}
3473
3474/**
3475 * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
3476 * @skb:      send buffer
3477 *
3478 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
3479 * and so we need to figure out the cases where we need to linearize the skb.
3480 *
3481 * For TSO we need to count the TSO header and segment payload separately.
3482 * As such we need to check cases where we have 7 fragments or more as we
3483 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
3484 * the segment payload in the first descriptor, and another 7 for the
3485 * fragments.
3486 **/
3487bool __i40e_chk_linearize(struct sk_buff *skb)
3488{
3489	const skb_frag_t *frag, *stale;
3490	int nr_frags, sum;
3491
3492	/* no need to check if number of frags is less than 7 */
3493	nr_frags = skb_shinfo(skb)->nr_frags;
3494	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
3495		return false;
3496
3497	/* We need to walk through the list and validate that each group
3498	 * of 6 fragments totals at least gso_size.
3499	 */
3500	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
3501	frag = &skb_shinfo(skb)->frags[0];
3502
3503	/* Initialize size to the negative value of gso_size minus 1.  We
3504	 * use this as the worst case scenerio in which the frag ahead
3505	 * of us only provides one byte which is why we are limited to 6
3506	 * descriptors for a single transmit as the header and previous
3507	 * fragment are already consuming 2 descriptors.
3508	 */
3509	sum = 1 - skb_shinfo(skb)->gso_size;
3510
3511	/* Add size of frags 0 through 4 to create our initial sum */
3512	sum += skb_frag_size(frag++);
3513	sum += skb_frag_size(frag++);
3514	sum += skb_frag_size(frag++);
3515	sum += skb_frag_size(frag++);
3516	sum += skb_frag_size(frag++);
3517
3518	/* Walk through fragments adding latest fragment, testing it, and
3519	 * then removing stale fragments from the sum.
3520	 */
3521	for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
3522		int stale_size = skb_frag_size(stale);
3523
3524		sum += skb_frag_size(frag++);
3525
3526		/* The stale fragment may present us with a smaller
3527		 * descriptor than the actual fragment size. To account
3528		 * for that we need to remove all the data on the front and
3529		 * figure out what the remainder would be in the last
3530		 * descriptor associated with the fragment.
3531		 */
3532		if (stale_size > I40E_MAX_DATA_PER_TXD) {
3533			int align_pad = -(skb_frag_off(stale)) &
3534					(I40E_MAX_READ_REQ_SIZE - 1);
3535
3536			sum -= align_pad;
3537			stale_size -= align_pad;
3538
3539			do {
3540				sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3541				stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3542			} while (stale_size > I40E_MAX_DATA_PER_TXD);
3543		}
3544
3545		/* if sum is negative we failed to make sufficient progress */
3546		if (sum < 0)
3547			return true;
3548
3549		if (!nr_frags--)
3550			break;
3551
3552		sum -= stale_size;
3553	}
3554
3555	return false;
3556}
3557
3558/**
3559 * i40e_tx_map - Build the Tx descriptor
3560 * @tx_ring:  ring to send buffer on
3561 * @skb:      send buffer
3562 * @first:    first buffer info buffer to use
3563 * @tx_flags: collected send information
3564 * @hdr_len:  size of the packet header
3565 * @td_cmd:   the command field in the descriptor
3566 * @td_offset: offset for checksum or crc
3567 *
3568 * Returns 0 on success, -1 on failure to DMA
3569 **/
3570static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
3571			      struct i40e_tx_buffer *first, u32 tx_flags,
3572			      const u8 hdr_len, u32 td_cmd, u32 td_offset)
3573{
3574	unsigned int data_len = skb->data_len;
3575	unsigned int size = skb_headlen(skb);
3576	skb_frag_t *frag;
3577	struct i40e_tx_buffer *tx_bi;
3578	struct i40e_tx_desc *tx_desc;
3579	u16 i = tx_ring->next_to_use;
3580	u32 td_tag = 0;
3581	dma_addr_t dma;
3582	u16 desc_count = 1;
3583
3584	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
3585		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
3586		td_tag = FIELD_GET(I40E_TX_FLAGS_VLAN_MASK, tx_flags);
 
3587	}
3588
3589	first->tx_flags = tx_flags;
3590
3591	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3592
3593	tx_desc = I40E_TX_DESC(tx_ring, i);
3594	tx_bi = first;
3595
3596	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3597		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3598
3599		if (dma_mapping_error(tx_ring->dev, dma))
3600			goto dma_error;
3601
3602		/* record length, and DMA address */
3603		dma_unmap_len_set(tx_bi, len, size);
3604		dma_unmap_addr_set(tx_bi, dma, dma);
3605
3606		/* align size to end of page */
3607		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
3608		tx_desc->buffer_addr = cpu_to_le64(dma);
3609
3610		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
3611			tx_desc->cmd_type_offset_bsz =
3612				build_ctob(td_cmd, td_offset,
3613					   max_data, td_tag);
3614
3615			tx_desc++;
3616			i++;
3617			desc_count++;
3618
3619			if (i == tx_ring->count) {
3620				tx_desc = I40E_TX_DESC(tx_ring, 0);
3621				i = 0;
3622			}
3623
3624			dma += max_data;
3625			size -= max_data;
3626
3627			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3628			tx_desc->buffer_addr = cpu_to_le64(dma);
3629		}
3630
3631		if (likely(!data_len))
3632			break;
3633
3634		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3635							  size, td_tag);
3636
3637		tx_desc++;
3638		i++;
3639		desc_count++;
3640
3641		if (i == tx_ring->count) {
3642			tx_desc = I40E_TX_DESC(tx_ring, 0);
3643			i = 0;
3644		}
3645
3646		size = skb_frag_size(frag);
3647		data_len -= size;
3648
3649		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3650				       DMA_TO_DEVICE);
3651
3652		tx_bi = &tx_ring->tx_bi[i];
3653	}
3654
3655	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3656
3657	i++;
3658	if (i == tx_ring->count)
3659		i = 0;
3660
3661	tx_ring->next_to_use = i;
3662
3663	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3664
3665	/* write last descriptor with EOP bit */
3666	td_cmd |= I40E_TX_DESC_CMD_EOP;
3667
3668	/* We OR these values together to check both against 4 (WB_STRIDE)
3669	 * below. This is safe since we don't re-use desc_count afterwards.
3670	 */
3671	desc_count |= ++tx_ring->packet_stride;
3672
3673	if (desc_count >= WB_STRIDE) {
3674		/* write last descriptor with RS bit set */
3675		td_cmd |= I40E_TX_DESC_CMD_RS;
3676		tx_ring->packet_stride = 0;
3677	}
3678
3679	tx_desc->cmd_type_offset_bsz =
3680			build_ctob(td_cmd, td_offset, size, td_tag);
3681
3682	skb_tx_timestamp(skb);
3683
3684	/* Force memory writes to complete before letting h/w know there
3685	 * are new descriptors to fetch.
3686	 *
3687	 * We also use this memory barrier to make certain all of the
3688	 * status bits have been updated before next_to_watch is written.
3689	 */
3690	wmb();
3691
3692	/* set next_to_watch value indicating a packet is present */
3693	first->next_to_watch = tx_desc;
3694
3695	/* notify HW of packet */
3696	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
3697		writel(i, tx_ring->tail);
3698	}
3699
3700	return 0;
3701
3702dma_error:
3703	dev_info(tx_ring->dev, "TX DMA map failed\n");
3704
3705	/* clear dma mappings for failed tx_bi map */
3706	for (;;) {
3707		tx_bi = &tx_ring->tx_bi[i];
3708		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3709		if (tx_bi == first)
3710			break;
3711		if (i == 0)
3712			i = tx_ring->count;
3713		i--;
3714	}
3715
3716	tx_ring->next_to_use = i;
3717
3718	return -1;
3719}
3720
3721static u16 i40e_swdcb_skb_tx_hash(struct net_device *dev,
3722				  const struct sk_buff *skb,
3723				  u16 num_tx_queues)
3724{
3725	u32 jhash_initval_salt = 0xd631614b;
3726	u32 hash;
3727
3728	if (skb->sk && skb->sk->sk_hash)
3729		hash = skb->sk->sk_hash;
3730	else
3731		hash = (__force u16)skb->protocol ^ skb->hash;
3732
3733	hash = jhash_1word(hash, jhash_initval_salt);
3734
3735	return (u16)(((u64)hash * num_tx_queues) >> 32);
3736}
3737
3738u16 i40e_lan_select_queue(struct net_device *netdev,
3739			  struct sk_buff *skb,
3740			  struct net_device __always_unused *sb_dev)
3741{
3742	struct i40e_netdev_priv *np = netdev_priv(netdev);
3743	struct i40e_vsi *vsi = np->vsi;
3744	struct i40e_hw *hw;
3745	u16 qoffset;
3746	u16 qcount;
3747	u8 tclass;
3748	u16 hash;
3749	u8 prio;
3750
3751	/* is DCB enabled at all? */
3752	if (vsi->tc_config.numtc == 1 ||
3753	    i40e_is_tc_mqprio_enabled(vsi->back))
3754		return netdev_pick_tx(netdev, skb, sb_dev);
3755
3756	prio = skb->priority;
3757	hw = &vsi->back->hw;
3758	tclass = hw->local_dcbx_config.etscfg.prioritytable[prio];
3759	/* sanity check */
3760	if (unlikely(!(vsi->tc_config.enabled_tc & BIT(tclass))))
3761		tclass = 0;
3762
3763	/* select a queue assigned for the given TC */
3764	qcount = vsi->tc_config.tc_info[tclass].qcount;
3765	hash = i40e_swdcb_skb_tx_hash(netdev, skb, qcount);
3766
3767	qoffset = vsi->tc_config.tc_info[tclass].qoffset;
3768	return qoffset + hash;
3769}
3770
3771/**
3772 * i40e_xmit_xdp_ring - transmits an XDP buffer to an XDP Tx ring
3773 * @xdpf: data to transmit
3774 * @xdp_ring: XDP Tx ring
3775 **/
3776static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
3777			      struct i40e_ring *xdp_ring)
3778{
3779	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
3780	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
3781	u16 i = 0, index = xdp_ring->next_to_use;
3782	struct i40e_tx_buffer *tx_head = &xdp_ring->tx_bi[index];
3783	struct i40e_tx_buffer *tx_bi = tx_head;
3784	struct i40e_tx_desc *tx_desc = I40E_TX_DESC(xdp_ring, index);
3785	void *data = xdpf->data;
3786	u32 size = xdpf->len;
3787
3788	if (unlikely(I40E_DESC_UNUSED(xdp_ring) < 1 + nr_frags)) {
3789		xdp_ring->tx_stats.tx_busy++;
3790		return I40E_XDP_CONSUMED;
3791	}
3792
3793	tx_head->bytecount = xdp_get_frame_len(xdpf);
3794	tx_head->gso_segs = 1;
3795	tx_head->xdpf = xdpf;
3796
3797	for (;;) {
3798		dma_addr_t dma;
3799
3800		dma = dma_map_single(xdp_ring->dev, data, size, DMA_TO_DEVICE);
3801		if (dma_mapping_error(xdp_ring->dev, dma))
3802			goto unmap;
3803
3804		/* record length, and DMA address */
3805		dma_unmap_len_set(tx_bi, len, size);
3806		dma_unmap_addr_set(tx_bi, dma, dma);
3807
3808		tx_desc->buffer_addr = cpu_to_le64(dma);
3809		tx_desc->cmd_type_offset_bsz =
3810			build_ctob(I40E_TX_DESC_CMD_ICRC, 0, size, 0);
3811
3812		if (++index == xdp_ring->count)
3813			index = 0;
3814
3815		if (i == nr_frags)
3816			break;
3817
3818		tx_bi = &xdp_ring->tx_bi[index];
3819		tx_desc = I40E_TX_DESC(xdp_ring, index);
3820
3821		data = skb_frag_address(&sinfo->frags[i]);
3822		size = skb_frag_size(&sinfo->frags[i]);
3823		i++;
3824	}
3825
3826	tx_desc->cmd_type_offset_bsz |=
3827		cpu_to_le64(I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT);
3828
3829	/* Make certain all of the status bits have been updated
3830	 * before next_to_watch is written.
3831	 */
3832	smp_wmb();
3833
3834	xdp_ring->xdp_tx_active++;
3835
3836	tx_head->next_to_watch = tx_desc;
3837	xdp_ring->next_to_use = index;
3838
3839	return I40E_XDP_TX;
3840
3841unmap:
3842	for (;;) {
3843		tx_bi = &xdp_ring->tx_bi[index];
3844		if (dma_unmap_len(tx_bi, len))
3845			dma_unmap_page(xdp_ring->dev,
3846				       dma_unmap_addr(tx_bi, dma),
3847				       dma_unmap_len(tx_bi, len),
3848				       DMA_TO_DEVICE);
3849		dma_unmap_len_set(tx_bi, len, 0);
3850		if (tx_bi == tx_head)
3851			break;
3852
3853		if (!index)
3854			index += xdp_ring->count;
3855		index--;
3856	}
3857
3858	return I40E_XDP_CONSUMED;
3859}
3860
3861/**
3862 * i40e_xmit_frame_ring - Sends buffer on Tx ring
3863 * @skb:     send buffer
3864 * @tx_ring: ring to send buffer on
3865 *
3866 * Returns NETDEV_TX_OK if sent, else an error code
3867 **/
3868static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3869					struct i40e_ring *tx_ring)
3870{
3871	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3872	u32 cd_tunneling = 0, cd_l2tag2 = 0;
3873	struct i40e_tx_buffer *first;
3874	u32 td_offset = 0;
3875	u32 tx_flags = 0;
3876	u32 td_cmd = 0;
3877	u8 hdr_len = 0;
3878	int tso, count;
3879	int tsyn;
3880
3881	/* prefetch the data, we'll need it later */
3882	prefetch(skb->data);
3883
3884	i40e_trace(xmit_frame_ring, skb, tx_ring);
3885
3886	count = i40e_xmit_descriptor_count(skb);
3887	if (i40e_chk_linearize(skb, count)) {
3888		if (__skb_linearize(skb)) {
3889			dev_kfree_skb_any(skb);
3890			return NETDEV_TX_OK;
3891		}
3892		count = i40e_txd_use_count(skb->len);
3893		tx_ring->tx_stats.tx_linearize++;
3894	}
3895
3896	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3897	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3898	 *       + 4 desc gap to avoid the cache line where head is,
3899	 *       + 1 desc for context descriptor,
3900	 * otherwise try next time
3901	 */
3902	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3903		tx_ring->tx_stats.tx_busy++;
3904		return NETDEV_TX_BUSY;
3905	}
3906
3907	/* record the location of the first descriptor for this packet */
3908	first = &tx_ring->tx_bi[tx_ring->next_to_use];
3909	first->skb = skb;
3910	first->bytecount = skb->len;
3911	first->gso_segs = 1;
3912
3913	/* prepare the xmit flags */
3914	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3915		goto out_drop;
3916
3917	tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3918
3919	if (tso < 0)
3920		goto out_drop;
3921	else if (tso)
3922		tx_flags |= I40E_TX_FLAGS_TSO;
3923
3924	/* Always offload the checksum, since it's in the data descriptor */
3925	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3926				  tx_ring, &cd_tunneling);
3927	if (tso < 0)
3928		goto out_drop;
3929
3930	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3931
3932	if (tsyn)
3933		tx_flags |= I40E_TX_FLAGS_TSYN;
3934
3935	/* always enable CRC insertion offload */
3936	td_cmd |= I40E_TX_DESC_CMD_ICRC;
3937
3938	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3939			   cd_tunneling, cd_l2tag2);
3940
3941	/* Add Flow Director ATR if it's enabled.
3942	 *
3943	 * NOTE: this must always be directly before the data descriptor.
3944	 */
3945	i40e_atr(tx_ring, skb, tx_flags);
3946
3947	if (i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3948			td_cmd, td_offset))
3949		goto cleanup_tx_tstamp;
3950
3951	return NETDEV_TX_OK;
3952
3953out_drop:
3954	i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3955	dev_kfree_skb_any(first->skb);
3956	first->skb = NULL;
3957cleanup_tx_tstamp:
3958	if (unlikely(tx_flags & I40E_TX_FLAGS_TSYN)) {
3959		struct i40e_pf *pf = i40e_netdev_to_pf(tx_ring->netdev);
3960
3961		dev_kfree_skb_any(pf->ptp_tx_skb);
3962		pf->ptp_tx_skb = NULL;
3963		clear_bit_unlock(__I40E_PTP_TX_IN_PROGRESS, pf->state);
3964	}
3965
3966	return NETDEV_TX_OK;
3967}
3968
3969/**
3970 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3971 * @skb:    send buffer
3972 * @netdev: network interface device structure
3973 *
3974 * Returns NETDEV_TX_OK if sent, else an error code
3975 **/
3976netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3977{
3978	struct i40e_netdev_priv *np = netdev_priv(netdev);
3979	struct i40e_vsi *vsi = np->vsi;
3980	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3981
3982	/* hardware can't handle really short frames, hardware padding works
3983	 * beyond this point
3984	 */
3985	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3986		return NETDEV_TX_OK;
3987
3988	return i40e_xmit_frame_ring(skb, tx_ring);
3989}
3990
3991/**
3992 * i40e_xdp_xmit - Implements ndo_xdp_xmit
3993 * @dev: netdev
3994 * @n: number of frames
3995 * @frames: array of XDP buffer pointers
3996 * @flags: XDP extra info
3997 *
3998 * Returns number of frames successfully sent. Failed frames
3999 * will be free'ed by XDP core.
4000 *
4001 * For error cases, a negative errno code is returned and no-frames
4002 * are transmitted (caller must handle freeing frames).
4003 **/
4004int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
4005		  u32 flags)
4006{
4007	struct i40e_netdev_priv *np = netdev_priv(dev);
4008	unsigned int queue_index = smp_processor_id();
4009	struct i40e_vsi *vsi = np->vsi;
4010	struct i40e_pf *pf = vsi->back;
4011	struct i40e_ring *xdp_ring;
4012	int nxmit = 0;
4013	int i;
4014
4015	if (test_bit(__I40E_VSI_DOWN, vsi->state))
4016		return -ENETDOWN;
4017
4018	if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs ||
4019	    test_bit(__I40E_CONFIG_BUSY, pf->state))
4020		return -ENXIO;
4021
4022	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
4023		return -EINVAL;
4024
4025	xdp_ring = vsi->xdp_rings[queue_index];
4026
4027	for (i = 0; i < n; i++) {
4028		struct xdp_frame *xdpf = frames[i];
4029		int err;
4030
4031		err = i40e_xmit_xdp_ring(xdpf, xdp_ring);
4032		if (err != I40E_XDP_TX)
4033			break;
4034		nxmit++;
4035	}
4036
4037	if (unlikely(flags & XDP_XMIT_FLUSH))
4038		i40e_xdp_ring_update_tail(xdp_ring);
4039
4040	return nxmit;
4041}
v6.2
   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/mpls.h>
   7#include <net/xdp.h>
   8#include "i40e.h"
   9#include "i40e_trace.h"
  10#include "i40e_prototype.h"
  11#include "i40e_txrx_common.h"
  12#include "i40e_xsk.h"
  13
  14#define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
  15/**
  16 * i40e_fdir - Generate a Flow Director descriptor based on fdata
  17 * @tx_ring: Tx ring to send buffer on
  18 * @fdata: Flow director filter data
  19 * @add: Indicate if we are adding a rule or deleting one
  20 *
  21 **/
  22static void i40e_fdir(struct i40e_ring *tx_ring,
  23		      struct i40e_fdir_filter *fdata, bool add)
  24{
  25	struct i40e_filter_program_desc *fdir_desc;
  26	struct i40e_pf *pf = tx_ring->vsi->back;
  27	u32 flex_ptype, dtype_cmd;
  28	u16 i;
  29
  30	/* grab the next descriptor */
  31	i = tx_ring->next_to_use;
  32	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
  33
  34	i++;
  35	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
  36
  37	flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
  38		     (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
  39
  40	flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
  41		      (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
  42
  43	flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
  44		      (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
  45
  46	/* Use LAN VSI Id if not programmed by user */
  47	flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
  48		      ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
  49		       I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
  50
  51	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
  52
  53	dtype_cmd |= add ?
  54		     I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
  55		     I40E_TXD_FLTR_QW1_PCMD_SHIFT :
  56		     I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
  57		     I40E_TXD_FLTR_QW1_PCMD_SHIFT;
  58
  59	dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
  60		     (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
  61
  62	dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
  63		     (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
  64
  65	if (fdata->cnt_index) {
  66		dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
  67		dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
  68			     ((u32)fdata->cnt_index <<
  69			      I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
  70	}
  71
  72	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
  73	fdir_desc->rsvd = cpu_to_le32(0);
  74	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
  75	fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
  76}
  77
  78#define I40E_FD_CLEAN_DELAY 10
  79/**
  80 * i40e_program_fdir_filter - Program a Flow Director filter
  81 * @fdir_data: Packet data that will be filter parameters
  82 * @raw_packet: the pre-allocated packet buffer for FDir
  83 * @pf: The PF pointer
  84 * @add: True for add/update, False for remove
  85 **/
  86static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
  87				    u8 *raw_packet, struct i40e_pf *pf,
  88				    bool add)
  89{
  90	struct i40e_tx_buffer *tx_buf, *first;
  91	struct i40e_tx_desc *tx_desc;
  92	struct i40e_ring *tx_ring;
  93	struct i40e_vsi *vsi;
  94	struct device *dev;
  95	dma_addr_t dma;
  96	u32 td_cmd = 0;
  97	u16 i;
  98
  99	/* find existing FDIR VSI */
 100	vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
 101	if (!vsi)
 102		return -ENOENT;
 103
 104	tx_ring = vsi->tx_rings[0];
 105	dev = tx_ring->dev;
 106
 107	/* we need two descriptors to add/del a filter and we can wait */
 108	for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
 109		if (!i)
 110			return -EAGAIN;
 111		msleep_interruptible(1);
 112	}
 113
 114	dma = dma_map_single(dev, raw_packet,
 115			     I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
 116	if (dma_mapping_error(dev, dma))
 117		goto dma_fail;
 118
 119	/* grab the next descriptor */
 120	i = tx_ring->next_to_use;
 121	first = &tx_ring->tx_bi[i];
 122	i40e_fdir(tx_ring, fdir_data, add);
 123
 124	/* Now program a dummy descriptor */
 125	i = tx_ring->next_to_use;
 126	tx_desc = I40E_TX_DESC(tx_ring, i);
 127	tx_buf = &tx_ring->tx_bi[i];
 128
 129	tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
 130
 131	memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
 132
 133	/* record length, and DMA address */
 134	dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
 135	dma_unmap_addr_set(tx_buf, dma, dma);
 136
 137	tx_desc->buffer_addr = cpu_to_le64(dma);
 138	td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
 139
 140	tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
 141	tx_buf->raw_buf = (void *)raw_packet;
 142
 143	tx_desc->cmd_type_offset_bsz =
 144		build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
 145
 146	/* Force memory writes to complete before letting h/w
 147	 * know there are new descriptors to fetch.
 148	 */
 149	wmb();
 150
 151	/* Mark the data descriptor to be watched */
 152	first->next_to_watch = tx_desc;
 153
 154	writel(tx_ring->next_to_use, tx_ring->tail);
 155	return 0;
 156
 157dma_fail:
 158	return -1;
 159}
 160
 161/**
 162 * i40e_create_dummy_packet - Constructs dummy packet for HW
 163 * @dummy_packet: preallocated space for dummy packet
 164 * @ipv4: is layer 3 packet of version 4 or 6
 165 * @l4proto: next level protocol used in data portion of l3
 166 * @data: filter data
 167 *
 168 * Returns address of layer 4 protocol dummy packet.
 169 **/
 170static char *i40e_create_dummy_packet(u8 *dummy_packet, bool ipv4, u8 l4proto,
 171				      struct i40e_fdir_filter *data)
 172{
 173	bool is_vlan = !!data->vlan_tag;
 174	struct vlan_hdr vlan;
 175	struct ipv6hdr ipv6;
 176	struct ethhdr eth;
 177	struct iphdr ip;
 178	u8 *tmp;
 179
 180	if (ipv4) {
 181		eth.h_proto = cpu_to_be16(ETH_P_IP);
 182		ip.protocol = l4proto;
 183		ip.version = 0x4;
 184		ip.ihl = 0x5;
 185
 186		ip.daddr = data->dst_ip;
 187		ip.saddr = data->src_ip;
 188	} else {
 189		eth.h_proto = cpu_to_be16(ETH_P_IPV6);
 190		ipv6.nexthdr = l4proto;
 191		ipv6.version = 0x6;
 192
 193		memcpy(&ipv6.saddr.in6_u.u6_addr32, data->src_ip6,
 194		       sizeof(__be32) * 4);
 195		memcpy(&ipv6.daddr.in6_u.u6_addr32, data->dst_ip6,
 196		       sizeof(__be32) * 4);
 197	}
 198
 199	if (is_vlan) {
 200		vlan.h_vlan_TCI = data->vlan_tag;
 201		vlan.h_vlan_encapsulated_proto = eth.h_proto;
 202		eth.h_proto = data->vlan_etype;
 203	}
 204
 205	tmp = dummy_packet;
 206	memcpy(tmp, &eth, sizeof(eth));
 207	tmp += sizeof(eth);
 208
 209	if (is_vlan) {
 210		memcpy(tmp, &vlan, sizeof(vlan));
 211		tmp += sizeof(vlan);
 212	}
 213
 214	if (ipv4) {
 215		memcpy(tmp, &ip, sizeof(ip));
 216		tmp += sizeof(ip);
 217	} else {
 218		memcpy(tmp, &ipv6, sizeof(ipv6));
 219		tmp += sizeof(ipv6);
 220	}
 221
 222	return tmp;
 223}
 224
 225/**
 226 * i40e_create_dummy_udp_packet - helper function to create UDP packet
 227 * @raw_packet: preallocated space for dummy packet
 228 * @ipv4: is layer 3 packet of version 4 or 6
 229 * @l4proto: next level protocol used in data portion of l3
 230 * @data: filter data
 231 *
 232 * Helper function to populate udp fields.
 233 **/
 234static void i40e_create_dummy_udp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
 235					 struct i40e_fdir_filter *data)
 236{
 237	struct udphdr *udp;
 238	u8 *tmp;
 239
 240	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_UDP, data);
 241	udp = (struct udphdr *)(tmp);
 242	udp->dest = data->dst_port;
 243	udp->source = data->src_port;
 244}
 245
 246/**
 247 * i40e_create_dummy_tcp_packet - helper function to create TCP packet
 248 * @raw_packet: preallocated space for dummy packet
 249 * @ipv4: is layer 3 packet of version 4 or 6
 250 * @l4proto: next level protocol used in data portion of l3
 251 * @data: filter data
 252 *
 253 * Helper function to populate tcp fields.
 254 **/
 255static void i40e_create_dummy_tcp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
 256					 struct i40e_fdir_filter *data)
 257{
 258	struct tcphdr *tcp;
 259	u8 *tmp;
 260	/* Dummy tcp packet */
 261	static const char tcp_packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 262		0x50, 0x11, 0x0, 0x72, 0, 0, 0, 0};
 263
 264	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_TCP, data);
 265
 266	tcp = (struct tcphdr *)tmp;
 267	memcpy(tcp, tcp_packet, sizeof(tcp_packet));
 268	tcp->dest = data->dst_port;
 269	tcp->source = data->src_port;
 270}
 271
 272/**
 273 * i40e_create_dummy_sctp_packet - helper function to create SCTP packet
 274 * @raw_packet: preallocated space for dummy packet
 275 * @ipv4: is layer 3 packet of version 4 or 6
 276 * @l4proto: next level protocol used in data portion of l3
 277 * @data: filter data
 278 *
 279 * Helper function to populate sctp fields.
 280 **/
 281static void i40e_create_dummy_sctp_packet(u8 *raw_packet, bool ipv4,
 282					  u8 l4proto,
 283					  struct i40e_fdir_filter *data)
 284{
 285	struct sctphdr *sctp;
 286	u8 *tmp;
 287
 288	tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_SCTP, data);
 289
 290	sctp = (struct sctphdr *)tmp;
 291	sctp->dest = data->dst_port;
 292	sctp->source = data->src_port;
 293}
 294
 295/**
 296 * i40e_prepare_fdir_filter - Prepare and program fdir filter
 297 * @pf: physical function to attach filter to
 298 * @fd_data: filter data
 299 * @add: add or delete filter
 300 * @packet_addr: address of dummy packet, used in filtering
 301 * @payload_offset: offset from dummy packet address to user defined data
 302 * @pctype: Packet type for which filter is used
 303 *
 304 * Helper function to offset data of dummy packet, program it and
 305 * handle errors.
 306 **/
 307static int i40e_prepare_fdir_filter(struct i40e_pf *pf,
 308				    struct i40e_fdir_filter *fd_data,
 309				    bool add, char *packet_addr,
 310				    int payload_offset, u8 pctype)
 311{
 312	int ret;
 313
 314	if (fd_data->flex_filter) {
 315		u8 *payload;
 316		__be16 pattern = fd_data->flex_word;
 317		u16 off = fd_data->flex_offset;
 318
 319		payload = packet_addr + payload_offset;
 320
 321		/* If user provided vlan, offset payload by vlan header length */
 322		if (!!fd_data->vlan_tag)
 323			payload += VLAN_HLEN;
 324
 325		*((__force __be16 *)(payload + off)) = pattern;
 326	}
 327
 328	fd_data->pctype = pctype;
 329	ret = i40e_program_fdir_filter(fd_data, packet_addr, pf, add);
 330	if (ret) {
 331		dev_info(&pf->pdev->dev,
 332			 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
 333			 fd_data->pctype, fd_data->fd_id, ret);
 334		/* Free the packet buffer since it wasn't added to the ring */
 335		return -EOPNOTSUPP;
 336	} else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
 337		if (add)
 338			dev_info(&pf->pdev->dev,
 339				 "Filter OK for PCTYPE %d loc = %d\n",
 340				 fd_data->pctype, fd_data->fd_id);
 341		else
 342			dev_info(&pf->pdev->dev,
 343				 "Filter deleted for PCTYPE %d loc = %d\n",
 344				 fd_data->pctype, fd_data->fd_id);
 345	}
 346
 347	return ret;
 348}
 349
 350/**
 351 * i40e_change_filter_num - Prepare and program fdir filter
 352 * @ipv4: is layer 3 packet of version 4 or 6
 353 * @add: add or delete filter
 354 * @ipv4_filter_num: field to update
 355 * @ipv6_filter_num: field to update
 356 *
 357 * Update filter number field for pf.
 358 **/
 359static void i40e_change_filter_num(bool ipv4, bool add, u16 *ipv4_filter_num,
 360				   u16 *ipv6_filter_num)
 361{
 362	if (add) {
 363		if (ipv4)
 364			(*ipv4_filter_num)++;
 365		else
 366			(*ipv6_filter_num)++;
 367	} else {
 368		if (ipv4)
 369			(*ipv4_filter_num)--;
 370		else
 371			(*ipv6_filter_num)--;
 372	}
 373}
 374
 375#define I40E_UDPIP_DUMMY_PACKET_LEN	42
 376#define I40E_UDPIP6_DUMMY_PACKET_LEN	62
 377/**
 378 * i40e_add_del_fdir_udp - Add/Remove UDP filters
 379 * @vsi: pointer to the targeted VSI
 380 * @fd_data: the flow director data required for the FDir descriptor
 381 * @add: true adds a filter, false removes it
 382 * @ipv4: true is v4, false is v6
 383 *
 384 * Returns 0 if the filters were successfully added or removed
 385 **/
 386static int i40e_add_del_fdir_udp(struct i40e_vsi *vsi,
 387				 struct i40e_fdir_filter *fd_data,
 388				 bool add,
 389				 bool ipv4)
 390{
 391	struct i40e_pf *pf = vsi->back;
 392	u8 *raw_packet;
 393	int ret;
 394
 395	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 396	if (!raw_packet)
 397		return -ENOMEM;
 398
 399	i40e_create_dummy_udp_packet(raw_packet, ipv4, IPPROTO_UDP, fd_data);
 400
 401	if (ipv4)
 402		ret = i40e_prepare_fdir_filter
 403			(pf, fd_data, add, raw_packet,
 404			 I40E_UDPIP_DUMMY_PACKET_LEN,
 405			 I40E_FILTER_PCTYPE_NONF_IPV4_UDP);
 406	else
 407		ret = i40e_prepare_fdir_filter
 408			(pf, fd_data, add, raw_packet,
 409			 I40E_UDPIP6_DUMMY_PACKET_LEN,
 410			 I40E_FILTER_PCTYPE_NONF_IPV6_UDP);
 411
 412	if (ret) {
 413		kfree(raw_packet);
 414		return ret;
 415	}
 416
 417	i40e_change_filter_num(ipv4, add, &pf->fd_udp4_filter_cnt,
 418			       &pf->fd_udp6_filter_cnt);
 419
 420	return 0;
 421}
 422
 423#define I40E_TCPIP_DUMMY_PACKET_LEN	54
 424#define I40E_TCPIP6_DUMMY_PACKET_LEN	74
 425/**
 426 * i40e_add_del_fdir_tcp - Add/Remove TCPv4 filters
 427 * @vsi: pointer to the targeted VSI
 428 * @fd_data: the flow director data required for the FDir descriptor
 429 * @add: true adds a filter, false removes it
 430 * @ipv4: true is v4, false is v6
 431 *
 432 * Returns 0 if the filters were successfully added or removed
 433 **/
 434static int i40e_add_del_fdir_tcp(struct i40e_vsi *vsi,
 435				 struct i40e_fdir_filter *fd_data,
 436				 bool add,
 437				 bool ipv4)
 438{
 439	struct i40e_pf *pf = vsi->back;
 440	u8 *raw_packet;
 441	int ret;
 442
 443	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 444	if (!raw_packet)
 445		return -ENOMEM;
 446
 447	i40e_create_dummy_tcp_packet(raw_packet, ipv4, IPPROTO_TCP, fd_data);
 448	if (ipv4)
 449		ret = i40e_prepare_fdir_filter
 450			(pf, fd_data, add, raw_packet,
 451			 I40E_TCPIP_DUMMY_PACKET_LEN,
 452			 I40E_FILTER_PCTYPE_NONF_IPV4_TCP);
 453	else
 454		ret = i40e_prepare_fdir_filter
 455			(pf, fd_data, add, raw_packet,
 456			 I40E_TCPIP6_DUMMY_PACKET_LEN,
 457			 I40E_FILTER_PCTYPE_NONF_IPV6_TCP);
 458
 459	if (ret) {
 460		kfree(raw_packet);
 461		return ret;
 462	}
 463
 464	i40e_change_filter_num(ipv4, add, &pf->fd_tcp4_filter_cnt,
 465			       &pf->fd_tcp6_filter_cnt);
 466
 467	if (add) {
 468		if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
 469		    I40E_DEBUG_FD & pf->hw.debug_mask)
 470			dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
 471		set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 472	}
 473	return 0;
 474}
 475
 476#define I40E_SCTPIP_DUMMY_PACKET_LEN	46
 477#define I40E_SCTPIP6_DUMMY_PACKET_LEN	66
 478/**
 479 * i40e_add_del_fdir_sctp - Add/Remove SCTPv4 Flow Director filters for
 480 * a specific flow spec
 481 * @vsi: pointer to the targeted VSI
 482 * @fd_data: the flow director data required for the FDir descriptor
 483 * @add: true adds a filter, false removes it
 484 * @ipv4: true is v4, false is v6
 485 *
 486 * Returns 0 if the filters were successfully added or removed
 487 **/
 488static int i40e_add_del_fdir_sctp(struct i40e_vsi *vsi,
 489				  struct i40e_fdir_filter *fd_data,
 490				  bool add,
 491				  bool ipv4)
 492{
 493	struct i40e_pf *pf = vsi->back;
 494	u8 *raw_packet;
 495	int ret;
 496
 497	raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 498	if (!raw_packet)
 499		return -ENOMEM;
 500
 501	i40e_create_dummy_sctp_packet(raw_packet, ipv4, IPPROTO_SCTP, fd_data);
 502
 503	if (ipv4)
 504		ret = i40e_prepare_fdir_filter
 505			(pf, fd_data, add, raw_packet,
 506			 I40E_SCTPIP_DUMMY_PACKET_LEN,
 507			 I40E_FILTER_PCTYPE_NONF_IPV4_SCTP);
 508	else
 509		ret = i40e_prepare_fdir_filter
 510			(pf, fd_data, add, raw_packet,
 511			 I40E_SCTPIP6_DUMMY_PACKET_LEN,
 512			 I40E_FILTER_PCTYPE_NONF_IPV6_SCTP);
 513
 514	if (ret) {
 515		kfree(raw_packet);
 516		return ret;
 517	}
 518
 519	i40e_change_filter_num(ipv4, add, &pf->fd_sctp4_filter_cnt,
 520			       &pf->fd_sctp6_filter_cnt);
 521
 522	return 0;
 523}
 524
 525#define I40E_IP_DUMMY_PACKET_LEN	34
 526#define I40E_IP6_DUMMY_PACKET_LEN	54
 527/**
 528 * i40e_add_del_fdir_ip - Add/Remove IPv4 Flow Director filters for
 529 * a specific flow spec
 530 * @vsi: pointer to the targeted VSI
 531 * @fd_data: the flow director data required for the FDir descriptor
 532 * @add: true adds a filter, false removes it
 533 * @ipv4: true is v4, false is v6
 534 *
 535 * Returns 0 if the filters were successfully added or removed
 536 **/
 537static int i40e_add_del_fdir_ip(struct i40e_vsi *vsi,
 538				struct i40e_fdir_filter *fd_data,
 539				bool add,
 540				bool ipv4)
 541{
 542	struct i40e_pf *pf = vsi->back;
 543	int payload_offset;
 544	u8 *raw_packet;
 545	int iter_start;
 546	int iter_end;
 547	int ret;
 548	int i;
 549
 550	if (ipv4) {
 551		iter_start = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
 552		iter_end = I40E_FILTER_PCTYPE_FRAG_IPV4;
 553	} else {
 554		iter_start = I40E_FILTER_PCTYPE_NONF_IPV6_OTHER;
 555		iter_end = I40E_FILTER_PCTYPE_FRAG_IPV6;
 556	}
 557
 558	for (i = iter_start; i <= iter_end; i++) {
 559		raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
 560		if (!raw_packet)
 561			return -ENOMEM;
 562
 563		/* IPv6 no header option differs from IPv4 */
 564		(void)i40e_create_dummy_packet
 565			(raw_packet, ipv4, (ipv4) ? IPPROTO_IP : IPPROTO_NONE,
 566			 fd_data);
 567
 568		payload_offset = (ipv4) ? I40E_IP_DUMMY_PACKET_LEN :
 569			I40E_IP6_DUMMY_PACKET_LEN;
 570		ret = i40e_prepare_fdir_filter(pf, fd_data, add, raw_packet,
 571					       payload_offset, i);
 572		if (ret)
 573			goto err;
 574	}
 575
 576	i40e_change_filter_num(ipv4, add, &pf->fd_ip4_filter_cnt,
 577			       &pf->fd_ip6_filter_cnt);
 578
 579	return 0;
 580err:
 581	kfree(raw_packet);
 582	return ret;
 583}
 584
 585/**
 586 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
 587 * @vsi: pointer to the targeted VSI
 588 * @input: filter to add or delete
 589 * @add: true adds a filter, false removes it
 590 *
 591 **/
 592int i40e_add_del_fdir(struct i40e_vsi *vsi,
 593		      struct i40e_fdir_filter *input, bool add)
 594{
 595	enum ip_ver { ipv6 = 0, ipv4 = 1 };
 596	struct i40e_pf *pf = vsi->back;
 597	int ret;
 598
 599	switch (input->flow_type & ~FLOW_EXT) {
 600	case TCP_V4_FLOW:
 601		ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
 602		break;
 603	case UDP_V4_FLOW:
 604		ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
 605		break;
 606	case SCTP_V4_FLOW:
 607		ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
 608		break;
 609	case TCP_V6_FLOW:
 610		ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
 611		break;
 612	case UDP_V6_FLOW:
 613		ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
 614		break;
 615	case SCTP_V6_FLOW:
 616		ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
 617		break;
 618	case IP_USER_FLOW:
 619		switch (input->ipl4_proto) {
 620		case IPPROTO_TCP:
 621			ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
 622			break;
 623		case IPPROTO_UDP:
 624			ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
 625			break;
 626		case IPPROTO_SCTP:
 627			ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
 628			break;
 629		case IPPROTO_IP:
 630			ret = i40e_add_del_fdir_ip(vsi, input, add, ipv4);
 631			break;
 632		default:
 633			/* We cannot support masking based on protocol */
 634			dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
 635				 input->ipl4_proto);
 636			return -EINVAL;
 637		}
 638		break;
 639	case IPV6_USER_FLOW:
 640		switch (input->ipl4_proto) {
 641		case IPPROTO_TCP:
 642			ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
 643			break;
 644		case IPPROTO_UDP:
 645			ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
 646			break;
 647		case IPPROTO_SCTP:
 648			ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
 649			break;
 650		case IPPROTO_IP:
 651			ret = i40e_add_del_fdir_ip(vsi, input, add, ipv6);
 652			break;
 653		default:
 654			/* We cannot support masking based on protocol */
 655			dev_info(&pf->pdev->dev, "Unsupported IPv6 protocol 0x%02x\n",
 656				 input->ipl4_proto);
 657			return -EINVAL;
 658		}
 659		break;
 660	default:
 661		dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
 662			 input->flow_type);
 663		return -EINVAL;
 664	}
 665
 666	/* The buffer allocated here will be normally be freed by
 667	 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
 668	 * completion. In the event of an error adding the buffer to the FDIR
 669	 * ring, it will immediately be freed. It may also be freed by
 670	 * i40e_clean_tx_ring() when closing the VSI.
 671	 */
 672	return ret;
 673}
 674
 675/**
 676 * i40e_fd_handle_status - check the Programming Status for FD
 677 * @rx_ring: the Rx ring for this descriptor
 678 * @qword0_raw: qword0
 679 * @qword1: qword1 after le_to_cpu
 680 * @prog_id: the id originally used for programming
 681 *
 682 * This is used to verify if the FD programming or invalidation
 683 * requested by SW to the HW is successful or not and take actions accordingly.
 684 **/
 685static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u64 qword0_raw,
 686				  u64 qword1, u8 prog_id)
 687{
 688	struct i40e_pf *pf = rx_ring->vsi->back;
 689	struct pci_dev *pdev = pf->pdev;
 690	struct i40e_16b_rx_wb_qw0 *qw0;
 691	u32 fcnt_prog, fcnt_avail;
 692	u32 error;
 693
 694	qw0 = (struct i40e_16b_rx_wb_qw0 *)&qword0_raw;
 695	error = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
 696		I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
 697
 698	if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
 699		pf->fd_inv = le32_to_cpu(qw0->hi_dword.fd_id);
 700		if (qw0->hi_dword.fd_id != 0 ||
 701		    (I40E_DEBUG_FD & pf->hw.debug_mask))
 702			dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
 703				 pf->fd_inv);
 704
 705		/* Check if the programming error is for ATR.
 706		 * If so, auto disable ATR and set a state for
 707		 * flush in progress. Next time we come here if flush is in
 708		 * progress do nothing, once flush is complete the state will
 709		 * be cleared.
 710		 */
 711		if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
 712			return;
 713
 714		pf->fd_add_err++;
 715		/* store the current atr filter count */
 716		pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
 717
 718		if (qw0->hi_dword.fd_id == 0 &&
 719		    test_bit(__I40E_FD_SB_AUTO_DISABLED, pf->state)) {
 720			/* These set_bit() calls aren't atomic with the
 721			 * test_bit() here, but worse case we potentially
 722			 * disable ATR and queue a flush right after SB
 723			 * support is re-enabled. That shouldn't cause an
 724			 * issue in practice
 725			 */
 726			set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
 727			set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
 728		}
 729
 730		/* filter programming failed most likely due to table full */
 731		fcnt_prog = i40e_get_global_fd_count(pf);
 732		fcnt_avail = pf->fdir_pf_filter_count;
 733		/* If ATR is running fcnt_prog can quickly change,
 734		 * if we are very close to full, it makes sense to disable
 735		 * FD ATR/SB and then re-enable it when there is room.
 736		 */
 737		if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
 738			if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
 739			    !test_and_set_bit(__I40E_FD_SB_AUTO_DISABLED,
 740					      pf->state))
 741				if (I40E_DEBUG_FD & pf->hw.debug_mask)
 742					dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
 743		}
 744	} else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
 745		if (I40E_DEBUG_FD & pf->hw.debug_mask)
 746			dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
 747				 qw0->hi_dword.fd_id);
 748	}
 749}
 750
 751/**
 752 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
 753 * @ring:      the ring that owns the buffer
 754 * @tx_buffer: the buffer to free
 755 **/
 756static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
 757					    struct i40e_tx_buffer *tx_buffer)
 758{
 759	if (tx_buffer->skb) {
 760		if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
 761			kfree(tx_buffer->raw_buf);
 762		else if (ring_is_xdp(ring))
 763			xdp_return_frame(tx_buffer->xdpf);
 764		else
 765			dev_kfree_skb_any(tx_buffer->skb);
 766		if (dma_unmap_len(tx_buffer, len))
 767			dma_unmap_single(ring->dev,
 768					 dma_unmap_addr(tx_buffer, dma),
 769					 dma_unmap_len(tx_buffer, len),
 770					 DMA_TO_DEVICE);
 771	} else if (dma_unmap_len(tx_buffer, len)) {
 772		dma_unmap_page(ring->dev,
 773			       dma_unmap_addr(tx_buffer, dma),
 774			       dma_unmap_len(tx_buffer, len),
 775			       DMA_TO_DEVICE);
 776	}
 777
 778	tx_buffer->next_to_watch = NULL;
 779	tx_buffer->skb = NULL;
 780	dma_unmap_len_set(tx_buffer, len, 0);
 781	/* tx_buffer must be completely set up in the transmit path */
 782}
 783
 784/**
 785 * i40e_clean_tx_ring - Free any empty Tx buffers
 786 * @tx_ring: ring to be cleaned
 787 **/
 788void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
 789{
 790	unsigned long bi_size;
 791	u16 i;
 792
 793	if (ring_is_xdp(tx_ring) && tx_ring->xsk_pool) {
 794		i40e_xsk_clean_tx_ring(tx_ring);
 795	} else {
 796		/* ring already cleared, nothing to do */
 797		if (!tx_ring->tx_bi)
 798			return;
 799
 800		/* Free all the Tx ring sk_buffs */
 801		for (i = 0; i < tx_ring->count; i++)
 802			i40e_unmap_and_free_tx_resource(tx_ring,
 803							&tx_ring->tx_bi[i]);
 804	}
 805
 806	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
 807	memset(tx_ring->tx_bi, 0, bi_size);
 808
 809	/* Zero out the descriptor ring */
 810	memset(tx_ring->desc, 0, tx_ring->size);
 811
 812	tx_ring->next_to_use = 0;
 813	tx_ring->next_to_clean = 0;
 814
 815	if (!tx_ring->netdev)
 816		return;
 817
 818	/* cleanup Tx queue statistics */
 819	netdev_tx_reset_queue(txring_txq(tx_ring));
 820}
 821
 822/**
 823 * i40e_free_tx_resources - Free Tx resources per queue
 824 * @tx_ring: Tx descriptor ring for a specific queue
 825 *
 826 * Free all transmit software resources
 827 **/
 828void i40e_free_tx_resources(struct i40e_ring *tx_ring)
 829{
 830	i40e_clean_tx_ring(tx_ring);
 831	kfree(tx_ring->tx_bi);
 832	tx_ring->tx_bi = NULL;
 833
 834	if (tx_ring->desc) {
 835		dma_free_coherent(tx_ring->dev, tx_ring->size,
 836				  tx_ring->desc, tx_ring->dma);
 837		tx_ring->desc = NULL;
 838	}
 839}
 840
 841/**
 842 * i40e_get_tx_pending - how many tx descriptors not processed
 843 * @ring: the ring of descriptors
 844 * @in_sw: use SW variables
 845 *
 846 * Since there is no access to the ring head register
 847 * in XL710, we need to use our local copies
 848 **/
 849u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
 850{
 851	u32 head, tail;
 852
 853	if (!in_sw) {
 854		head = i40e_get_head(ring);
 855		tail = readl(ring->tail);
 856	} else {
 857		head = ring->next_to_clean;
 858		tail = ring->next_to_use;
 859	}
 860
 861	if (head != tail)
 862		return (head < tail) ?
 863			tail - head : (tail + ring->count - head);
 864
 865	return 0;
 866}
 867
 868/**
 869 * i40e_detect_recover_hung - Function to detect and recover hung_queues
 870 * @vsi:  pointer to vsi struct with tx queues
 871 *
 872 * VSI has netdev and netdev has TX queues. This function is to check each of
 873 * those TX queues if they are hung, trigger recovery by issuing SW interrupt.
 874 **/
 875void i40e_detect_recover_hung(struct i40e_vsi *vsi)
 876{
 877	struct i40e_ring *tx_ring = NULL;
 878	struct net_device *netdev;
 879	unsigned int i;
 880	int packets;
 881
 882	if (!vsi)
 883		return;
 884
 885	if (test_bit(__I40E_VSI_DOWN, vsi->state))
 886		return;
 887
 888	netdev = vsi->netdev;
 889	if (!netdev)
 890		return;
 891
 892	if (!netif_carrier_ok(netdev))
 893		return;
 894
 895	for (i = 0; i < vsi->num_queue_pairs; i++) {
 896		tx_ring = vsi->tx_rings[i];
 897		if (tx_ring && tx_ring->desc) {
 898			/* If packet counter has not changed the queue is
 899			 * likely stalled, so force an interrupt for this
 900			 * queue.
 901			 *
 902			 * prev_pkt_ctr would be negative if there was no
 903			 * pending work.
 904			 */
 905			packets = tx_ring->stats.packets & INT_MAX;
 906			if (tx_ring->tx_stats.prev_pkt_ctr == packets) {
 907				i40e_force_wb(vsi, tx_ring->q_vector);
 908				continue;
 909			}
 910
 911			/* Memory barrier between read of packet count and call
 912			 * to i40e_get_tx_pending()
 913			 */
 914			smp_rmb();
 915			tx_ring->tx_stats.prev_pkt_ctr =
 916			    i40e_get_tx_pending(tx_ring, true) ? packets : -1;
 917		}
 918	}
 919}
 920
 921/**
 922 * i40e_clean_tx_irq - Reclaim resources after transmit completes
 923 * @vsi: the VSI we care about
 924 * @tx_ring: Tx ring to clean
 925 * @napi_budget: Used to determine if we are in netpoll
 926 * @tx_cleaned: Out parameter set to the number of TXes cleaned
 927 *
 928 * Returns true if there's any budget left (e.g. the clean is finished)
 929 **/
 930static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
 931			      struct i40e_ring *tx_ring, int napi_budget,
 932			      unsigned int *tx_cleaned)
 933{
 934	int i = tx_ring->next_to_clean;
 935	struct i40e_tx_buffer *tx_buf;
 936	struct i40e_tx_desc *tx_head;
 937	struct i40e_tx_desc *tx_desc;
 938	unsigned int total_bytes = 0, total_packets = 0;
 939	unsigned int budget = vsi->work_limit;
 940
 941	tx_buf = &tx_ring->tx_bi[i];
 942	tx_desc = I40E_TX_DESC(tx_ring, i);
 943	i -= tx_ring->count;
 944
 945	tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
 946
 947	do {
 948		struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
 949
 950		/* if next_to_watch is not set then there is no work pending */
 951		if (!eop_desc)
 952			break;
 953
 954		/* prevent any other reads prior to eop_desc */
 955		smp_rmb();
 956
 957		i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
 958		/* we have caught up to head, no work left to do */
 959		if (tx_head == tx_desc)
 960			break;
 961
 962		/* clear next_to_watch to prevent false hangs */
 963		tx_buf->next_to_watch = NULL;
 964
 965		/* update the statistics for this packet */
 966		total_bytes += tx_buf->bytecount;
 967		total_packets += tx_buf->gso_segs;
 968
 969		/* free the skb/XDP data */
 970		if (ring_is_xdp(tx_ring))
 971			xdp_return_frame(tx_buf->xdpf);
 972		else
 973			napi_consume_skb(tx_buf->skb, napi_budget);
 974
 975		/* unmap skb header data */
 976		dma_unmap_single(tx_ring->dev,
 977				 dma_unmap_addr(tx_buf, dma),
 978				 dma_unmap_len(tx_buf, len),
 979				 DMA_TO_DEVICE);
 980
 981		/* clear tx_buffer data */
 982		tx_buf->skb = NULL;
 983		dma_unmap_len_set(tx_buf, len, 0);
 984
 985		/* unmap remaining buffers */
 986		while (tx_desc != eop_desc) {
 987			i40e_trace(clean_tx_irq_unmap,
 988				   tx_ring, tx_desc, tx_buf);
 989
 990			tx_buf++;
 991			tx_desc++;
 992			i++;
 993			if (unlikely(!i)) {
 994				i -= tx_ring->count;
 995				tx_buf = tx_ring->tx_bi;
 996				tx_desc = I40E_TX_DESC(tx_ring, 0);
 997			}
 998
 999			/* unmap any remaining paged data */
1000			if (dma_unmap_len(tx_buf, len)) {
1001				dma_unmap_page(tx_ring->dev,
1002					       dma_unmap_addr(tx_buf, dma),
1003					       dma_unmap_len(tx_buf, len),
1004					       DMA_TO_DEVICE);
1005				dma_unmap_len_set(tx_buf, len, 0);
1006			}
1007		}
1008
1009		/* move us one more past the eop_desc for start of next pkt */
1010		tx_buf++;
1011		tx_desc++;
1012		i++;
1013		if (unlikely(!i)) {
1014			i -= tx_ring->count;
1015			tx_buf = tx_ring->tx_bi;
1016			tx_desc = I40E_TX_DESC(tx_ring, 0);
1017		}
1018
1019		prefetch(tx_desc);
1020
1021		/* update budget accounting */
1022		budget--;
1023	} while (likely(budget));
1024
1025	i += tx_ring->count;
1026	tx_ring->next_to_clean = i;
1027	i40e_update_tx_stats(tx_ring, total_packets, total_bytes);
1028	i40e_arm_wb(tx_ring, vsi, budget);
1029
1030	if (ring_is_xdp(tx_ring))
1031		return !!budget;
1032
1033	/* notify netdev of completed buffers */
1034	netdev_tx_completed_queue(txring_txq(tx_ring),
1035				  total_packets, total_bytes);
1036
1037#define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
1038	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1039		     (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
1040		/* Make sure that anybody stopping the queue after this
1041		 * sees the new next_to_clean.
1042		 */
1043		smp_mb();
1044		if (__netif_subqueue_stopped(tx_ring->netdev,
1045					     tx_ring->queue_index) &&
1046		   !test_bit(__I40E_VSI_DOWN, vsi->state)) {
1047			netif_wake_subqueue(tx_ring->netdev,
1048					    tx_ring->queue_index);
1049			++tx_ring->tx_stats.restart_queue;
1050		}
1051	}
1052
1053	*tx_cleaned = total_packets;
1054	return !!budget;
1055}
1056
1057/**
1058 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
1059 * @vsi: the VSI we care about
1060 * @q_vector: the vector on which to enable writeback
1061 *
1062 **/
1063static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
1064				  struct i40e_q_vector *q_vector)
1065{
1066	u16 flags = q_vector->tx.ring[0].flags;
1067	u32 val;
1068
1069	if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
1070		return;
1071
1072	if (q_vector->arm_wb_state)
1073		return;
1074
1075	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1076		val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
1077		      I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
1078
1079		wr32(&vsi->back->hw,
1080		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx),
1081		     val);
1082	} else {
1083		val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
1084		      I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
1085
1086		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1087	}
1088	q_vector->arm_wb_state = true;
1089}
1090
1091/**
1092 * i40e_force_wb - Issue SW Interrupt so HW does a wb
1093 * @vsi: the VSI we care about
1094 * @q_vector: the vector  on which to force writeback
1095 *
1096 **/
1097void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
1098{
1099	if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1100		u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1101			  I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
1102			  I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
1103			  I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
1104			  /* allow 00 to be written to the index */
1105
1106		wr32(&vsi->back->hw,
1107		     I40E_PFINT_DYN_CTLN(q_vector->reg_idx), val);
1108	} else {
1109		u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
1110			  I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
1111			  I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
1112			  I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
1113			/* allow 00 to be written to the index */
1114
1115		wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1116	}
1117}
1118
1119static inline bool i40e_container_is_rx(struct i40e_q_vector *q_vector,
1120					struct i40e_ring_container *rc)
1121{
1122	return &q_vector->rx == rc;
1123}
1124
1125static inline unsigned int i40e_itr_divisor(struct i40e_q_vector *q_vector)
1126{
1127	unsigned int divisor;
1128
1129	switch (q_vector->vsi->back->hw.phy.link_info.link_speed) {
1130	case I40E_LINK_SPEED_40GB:
1131		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 1024;
1132		break;
1133	case I40E_LINK_SPEED_25GB:
1134	case I40E_LINK_SPEED_20GB:
1135		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 512;
1136		break;
1137	default:
1138	case I40E_LINK_SPEED_10GB:
1139		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 256;
1140		break;
1141	case I40E_LINK_SPEED_1GB:
1142	case I40E_LINK_SPEED_100MB:
1143		divisor = I40E_ITR_ADAPTIVE_MIN_INC * 32;
1144		break;
1145	}
1146
1147	return divisor;
1148}
1149
1150/**
1151 * i40e_update_itr - update the dynamic ITR value based on statistics
1152 * @q_vector: structure containing interrupt and ring information
1153 * @rc: structure containing ring performance data
1154 *
1155 * Stores a new ITR value based on packets and byte
1156 * counts during the last interrupt.  The advantage of per interrupt
1157 * computation is faster updates and more accurate ITR for the current
1158 * traffic pattern.  Constants in this function were computed
1159 * based on theoretical maximum wire speed and thresholds were set based
1160 * on testing data as well as attempting to minimize response time
1161 * while increasing bulk throughput.
1162 **/
1163static void i40e_update_itr(struct i40e_q_vector *q_vector,
1164			    struct i40e_ring_container *rc)
1165{
1166	unsigned int avg_wire_size, packets, bytes, itr;
1167	unsigned long next_update = jiffies;
1168
1169	/* If we don't have any rings just leave ourselves set for maximum
1170	 * possible latency so we take ourselves out of the equation.
1171	 */
1172	if (!rc->ring || !ITR_IS_DYNAMIC(rc->ring->itr_setting))
1173		return;
1174
1175	/* For Rx we want to push the delay up and default to low latency.
1176	 * for Tx we want to pull the delay down and default to high latency.
1177	 */
1178	itr = i40e_container_is_rx(q_vector, rc) ?
1179	      I40E_ITR_ADAPTIVE_MIN_USECS | I40E_ITR_ADAPTIVE_LATENCY :
1180	      I40E_ITR_ADAPTIVE_MAX_USECS | I40E_ITR_ADAPTIVE_LATENCY;
1181
1182	/* If we didn't update within up to 1 - 2 jiffies we can assume
1183	 * that either packets are coming in so slow there hasn't been
1184	 * any work, or that there is so much work that NAPI is dealing
1185	 * with interrupt moderation and we don't need to do anything.
1186	 */
1187	if (time_after(next_update, rc->next_update))
1188		goto clear_counts;
1189
1190	/* If itr_countdown is set it means we programmed an ITR within
1191	 * the last 4 interrupt cycles. This has a side effect of us
1192	 * potentially firing an early interrupt. In order to work around
1193	 * this we need to throw out any data received for a few
1194	 * interrupts following the update.
1195	 */
1196	if (q_vector->itr_countdown) {
1197		itr = rc->target_itr;
1198		goto clear_counts;
1199	}
1200
1201	packets = rc->total_packets;
1202	bytes = rc->total_bytes;
1203
1204	if (i40e_container_is_rx(q_vector, rc)) {
1205		/* If Rx there are 1 to 4 packets and bytes are less than
1206		 * 9000 assume insufficient data to use bulk rate limiting
1207		 * approach unless Tx is already in bulk rate limiting. We
1208		 * are likely latency driven.
1209		 */
1210		if (packets && packets < 4 && bytes < 9000 &&
1211		    (q_vector->tx.target_itr & I40E_ITR_ADAPTIVE_LATENCY)) {
1212			itr = I40E_ITR_ADAPTIVE_LATENCY;
1213			goto adjust_by_size;
1214		}
1215	} else if (packets < 4) {
1216		/* If we have Tx and Rx ITR maxed and Tx ITR is running in
1217		 * bulk mode and we are receiving 4 or fewer packets just
1218		 * reset the ITR_ADAPTIVE_LATENCY bit for latency mode so
1219		 * that the Rx can relax.
1220		 */
1221		if (rc->target_itr == I40E_ITR_ADAPTIVE_MAX_USECS &&
1222		    (q_vector->rx.target_itr & I40E_ITR_MASK) ==
1223		     I40E_ITR_ADAPTIVE_MAX_USECS)
1224			goto clear_counts;
1225	} else if (packets > 32) {
1226		/* If we have processed over 32 packets in a single interrupt
1227		 * for Tx assume we need to switch over to "bulk" mode.
1228		 */
1229		rc->target_itr &= ~I40E_ITR_ADAPTIVE_LATENCY;
1230	}
1231
1232	/* We have no packets to actually measure against. This means
1233	 * either one of the other queues on this vector is active or
1234	 * we are a Tx queue doing TSO with too high of an interrupt rate.
1235	 *
1236	 * Between 4 and 56 we can assume that our current interrupt delay
1237	 * is only slightly too low. As such we should increase it by a small
1238	 * fixed amount.
1239	 */
1240	if (packets < 56) {
1241		itr = rc->target_itr + I40E_ITR_ADAPTIVE_MIN_INC;
1242		if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1243			itr &= I40E_ITR_ADAPTIVE_LATENCY;
1244			itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1245		}
1246		goto clear_counts;
1247	}
1248
1249	if (packets <= 256) {
1250		itr = min(q_vector->tx.current_itr, q_vector->rx.current_itr);
1251		itr &= I40E_ITR_MASK;
1252
1253		/* Between 56 and 112 is our "goldilocks" zone where we are
1254		 * working out "just right". Just report that our current
1255		 * ITR is good for us.
1256		 */
1257		if (packets <= 112)
1258			goto clear_counts;
1259
1260		/* If packet count is 128 or greater we are likely looking
1261		 * at a slight overrun of the delay we want. Try halving
1262		 * our delay to see if that will cut the number of packets
1263		 * in half per interrupt.
1264		 */
1265		itr /= 2;
1266		itr &= I40E_ITR_MASK;
1267		if (itr < I40E_ITR_ADAPTIVE_MIN_USECS)
1268			itr = I40E_ITR_ADAPTIVE_MIN_USECS;
1269
1270		goto clear_counts;
1271	}
1272
1273	/* The paths below assume we are dealing with a bulk ITR since
1274	 * number of packets is greater than 256. We are just going to have
1275	 * to compute a value and try to bring the count under control,
1276	 * though for smaller packet sizes there isn't much we can do as
1277	 * NAPI polling will likely be kicking in sooner rather than later.
1278	 */
1279	itr = I40E_ITR_ADAPTIVE_BULK;
1280
1281adjust_by_size:
1282	/* If packet counts are 256 or greater we can assume we have a gross
1283	 * overestimation of what the rate should be. Instead of trying to fine
1284	 * tune it just use the formula below to try and dial in an exact value
1285	 * give the current packet size of the frame.
1286	 */
1287	avg_wire_size = bytes / packets;
1288
1289	/* The following is a crude approximation of:
1290	 *  wmem_default / (size + overhead) = desired_pkts_per_int
1291	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1292	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1293	 *
1294	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1295	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1296	 * formula down to
1297	 *
1298	 *  (170 * (size + 24)) / (size + 640) = ITR
1299	 *
1300	 * We first do some math on the packet size and then finally bitshift
1301	 * by 8 after rounding up. We also have to account for PCIe link speed
1302	 * difference as ITR scales based on this.
1303	 */
1304	if (avg_wire_size <= 60) {
1305		/* Start at 250k ints/sec */
1306		avg_wire_size = 4096;
1307	} else if (avg_wire_size <= 380) {
1308		/* 250K ints/sec to 60K ints/sec */
1309		avg_wire_size *= 40;
1310		avg_wire_size += 1696;
1311	} else if (avg_wire_size <= 1084) {
1312		/* 60K ints/sec to 36K ints/sec */
1313		avg_wire_size *= 15;
1314		avg_wire_size += 11452;
1315	} else if (avg_wire_size <= 1980) {
1316		/* 36K ints/sec to 30K ints/sec */
1317		avg_wire_size *= 5;
1318		avg_wire_size += 22420;
1319	} else {
1320		/* plateau at a limit of 30K ints/sec */
1321		avg_wire_size = 32256;
1322	}
1323
1324	/* If we are in low latency mode halve our delay which doubles the
1325	 * rate to somewhere between 100K to 16K ints/sec
1326	 */
1327	if (itr & I40E_ITR_ADAPTIVE_LATENCY)
1328		avg_wire_size /= 2;
1329
1330	/* Resultant value is 256 times larger than it needs to be. This
1331	 * gives us room to adjust the value as needed to either increase
1332	 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
1333	 *
1334	 * Use addition as we have already recorded the new latency flag
1335	 * for the ITR value.
1336	 */
1337	itr += DIV_ROUND_UP(avg_wire_size, i40e_itr_divisor(q_vector)) *
1338	       I40E_ITR_ADAPTIVE_MIN_INC;
1339
1340	if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1341		itr &= I40E_ITR_ADAPTIVE_LATENCY;
1342		itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1343	}
1344
1345clear_counts:
1346	/* write back value */
1347	rc->target_itr = itr;
1348
1349	/* next update should occur within next jiffy */
1350	rc->next_update = next_update + 1;
1351
1352	rc->total_bytes = 0;
1353	rc->total_packets = 0;
1354}
1355
1356static struct i40e_rx_buffer *i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
1357{
1358	return &rx_ring->rx_bi[idx];
1359}
1360
1361/**
1362 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1363 * @rx_ring: rx descriptor ring to store buffers on
1364 * @old_buff: donor buffer to have page reused
1365 *
1366 * Synchronizes page for reuse by the adapter
1367 **/
1368static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1369			       struct i40e_rx_buffer *old_buff)
1370{
1371	struct i40e_rx_buffer *new_buff;
1372	u16 nta = rx_ring->next_to_alloc;
1373
1374	new_buff = i40e_rx_bi(rx_ring, nta);
1375
1376	/* update, and store next to alloc */
1377	nta++;
1378	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1379
1380	/* transfer page from old buffer to new buffer */
1381	new_buff->dma		= old_buff->dma;
1382	new_buff->page		= old_buff->page;
1383	new_buff->page_offset	= old_buff->page_offset;
1384	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
1385
1386	/* clear contents of buffer_info */
1387	old_buff->page = NULL;
1388}
1389
1390/**
1391 * i40e_clean_programming_status - clean the programming status descriptor
1392 * @rx_ring: the rx ring that has this descriptor
1393 * @qword0_raw: qword0
1394 * @qword1: qword1 representing status_error_len in CPU ordering
1395 *
1396 * Flow director should handle FD_FILTER_STATUS to check its filter programming
1397 * status being successful or not and take actions accordingly. FCoE should
1398 * handle its context/filter programming/invalidation status and take actions.
1399 *
1400 * Returns an i40e_rx_buffer to reuse if the cleanup occurred, otherwise NULL.
1401 **/
1402void i40e_clean_programming_status(struct i40e_ring *rx_ring, u64 qword0_raw,
1403				   u64 qword1)
1404{
1405	u8 id;
1406
1407	id = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1408		  I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1409
1410	if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1411		i40e_fd_handle_status(rx_ring, qword0_raw, qword1, id);
1412}
1413
1414/**
1415 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1416 * @tx_ring: the tx ring to set up
1417 *
1418 * Return 0 on success, negative on error
1419 **/
1420int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1421{
1422	struct device *dev = tx_ring->dev;
1423	int bi_size;
1424
1425	if (!dev)
1426		return -ENOMEM;
1427
1428	/* warn if we are about to overwrite the pointer */
1429	WARN_ON(tx_ring->tx_bi);
1430	bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1431	tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1432	if (!tx_ring->tx_bi)
1433		goto err;
1434
1435	u64_stats_init(&tx_ring->syncp);
1436
1437	/* round up to nearest 4K */
1438	tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1439	/* add u32 for head writeback, align after this takes care of
1440	 * guaranteeing this is at least one cache line in size
1441	 */
1442	tx_ring->size += sizeof(u32);
1443	tx_ring->size = ALIGN(tx_ring->size, 4096);
1444	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1445					   &tx_ring->dma, GFP_KERNEL);
1446	if (!tx_ring->desc) {
1447		dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1448			 tx_ring->size);
1449		goto err;
1450	}
1451
1452	tx_ring->next_to_use = 0;
1453	tx_ring->next_to_clean = 0;
1454	tx_ring->tx_stats.prev_pkt_ctr = -1;
1455	return 0;
1456
1457err:
1458	kfree(tx_ring->tx_bi);
1459	tx_ring->tx_bi = NULL;
1460	return -ENOMEM;
1461}
1462
1463static void i40e_clear_rx_bi(struct i40e_ring *rx_ring)
1464{
1465	memset(rx_ring->rx_bi, 0, sizeof(*rx_ring->rx_bi) * rx_ring->count);
1466}
1467
1468/**
1469 * i40e_clean_rx_ring - Free Rx buffers
1470 * @rx_ring: ring to be cleaned
1471 **/
1472void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1473{
1474	u16 i;
1475
1476	/* ring already cleared, nothing to do */
1477	if (!rx_ring->rx_bi)
1478		return;
1479
1480	dev_kfree_skb(rx_ring->skb);
1481	rx_ring->skb = NULL;
1482
1483	if (rx_ring->xsk_pool) {
1484		i40e_xsk_clean_rx_ring(rx_ring);
1485		goto skip_free;
1486	}
1487
1488	/* Free all the Rx ring sk_buffs */
1489	for (i = 0; i < rx_ring->count; i++) {
1490		struct i40e_rx_buffer *rx_bi = i40e_rx_bi(rx_ring, i);
1491
1492		if (!rx_bi->page)
1493			continue;
1494
1495		/* Invalidate cache lines that may have been written to by
1496		 * device so that we avoid corrupting memory.
1497		 */
1498		dma_sync_single_range_for_cpu(rx_ring->dev,
1499					      rx_bi->dma,
1500					      rx_bi->page_offset,
1501					      rx_ring->rx_buf_len,
1502					      DMA_FROM_DEVICE);
1503
1504		/* free resources associated with mapping */
1505		dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1506				     i40e_rx_pg_size(rx_ring),
1507				     DMA_FROM_DEVICE,
1508				     I40E_RX_DMA_ATTR);
1509
1510		__page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1511
1512		rx_bi->page = NULL;
1513		rx_bi->page_offset = 0;
1514	}
1515
1516skip_free:
1517	if (rx_ring->xsk_pool)
1518		i40e_clear_rx_bi_zc(rx_ring);
1519	else
1520		i40e_clear_rx_bi(rx_ring);
1521
1522	/* Zero out the descriptor ring */
1523	memset(rx_ring->desc, 0, rx_ring->size);
1524
1525	rx_ring->next_to_alloc = 0;
1526	rx_ring->next_to_clean = 0;
 
1527	rx_ring->next_to_use = 0;
1528}
1529
1530/**
1531 * i40e_free_rx_resources - Free Rx resources
1532 * @rx_ring: ring to clean the resources from
1533 *
1534 * Free all receive software resources
1535 **/
1536void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1537{
1538	i40e_clean_rx_ring(rx_ring);
1539	if (rx_ring->vsi->type == I40E_VSI_MAIN)
1540		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
1541	rx_ring->xdp_prog = NULL;
1542	kfree(rx_ring->rx_bi);
1543	rx_ring->rx_bi = NULL;
1544
1545	if (rx_ring->desc) {
1546		dma_free_coherent(rx_ring->dev, rx_ring->size,
1547				  rx_ring->desc, rx_ring->dma);
1548		rx_ring->desc = NULL;
1549	}
1550}
1551
1552/**
1553 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1554 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1555 *
1556 * Returns 0 on success, negative on failure
1557 **/
1558int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1559{
1560	struct device *dev = rx_ring->dev;
1561	int err;
1562
1563	u64_stats_init(&rx_ring->syncp);
1564
1565	/* Round up to nearest 4K */
1566	rx_ring->size = rx_ring->count * sizeof(union i40e_rx_desc);
1567	rx_ring->size = ALIGN(rx_ring->size, 4096);
1568	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1569					   &rx_ring->dma, GFP_KERNEL);
1570
1571	if (!rx_ring->desc) {
1572		dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1573			 rx_ring->size);
1574		return -ENOMEM;
1575	}
1576
1577	rx_ring->next_to_alloc = 0;
1578	rx_ring->next_to_clean = 0;
 
1579	rx_ring->next_to_use = 0;
1580
1581	/* XDP RX-queue info only needed for RX rings exposed to XDP */
1582	if (rx_ring->vsi->type == I40E_VSI_MAIN) {
1583		err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
1584				       rx_ring->queue_index, rx_ring->q_vector->napi.napi_id);
1585		if (err < 0)
1586			return err;
1587	}
1588
1589	rx_ring->xdp_prog = rx_ring->vsi->xdp_prog;
1590
1591	rx_ring->rx_bi =
1592		kcalloc(rx_ring->count, sizeof(*rx_ring->rx_bi), GFP_KERNEL);
1593	if (!rx_ring->rx_bi)
1594		return -ENOMEM;
1595
1596	return 0;
1597}
1598
1599/**
1600 * i40e_release_rx_desc - Store the new tail and head values
1601 * @rx_ring: ring to bump
1602 * @val: new head index
1603 **/
1604void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1605{
1606	rx_ring->next_to_use = val;
1607
1608	/* update next to alloc since we have filled the ring */
1609	rx_ring->next_to_alloc = val;
1610
1611	/* Force memory writes to complete before letting h/w
1612	 * know there are new descriptors to fetch.  (Only
1613	 * applicable for weak-ordered memory model archs,
1614	 * such as IA-64).
1615	 */
1616	wmb();
1617	writel(val, rx_ring->tail);
1618}
1619
 
1620static unsigned int i40e_rx_frame_truesize(struct i40e_ring *rx_ring,
1621					   unsigned int size)
1622{
1623	unsigned int truesize;
1624
1625#if (PAGE_SIZE < 8192)
1626	truesize = i40e_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1627#else
1628	truesize = rx_ring->rx_offset ?
1629		SKB_DATA_ALIGN(size + rx_ring->rx_offset) +
1630		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1631		SKB_DATA_ALIGN(size);
1632#endif
1633	return truesize;
1634}
 
1635
1636/**
1637 * i40e_alloc_mapped_page - recycle or make a new page
1638 * @rx_ring: ring to use
1639 * @bi: rx_buffer struct to modify
1640 *
1641 * Returns true if the page was successfully allocated or
1642 * reused.
1643 **/
1644static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1645				   struct i40e_rx_buffer *bi)
1646{
1647	struct page *page = bi->page;
1648	dma_addr_t dma;
1649
1650	/* since we are recycling buffers we should seldom need to alloc */
1651	if (likely(page)) {
1652		rx_ring->rx_stats.page_reuse_count++;
1653		return true;
1654	}
1655
1656	/* alloc new page for storage */
1657	page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1658	if (unlikely(!page)) {
1659		rx_ring->rx_stats.alloc_page_failed++;
1660		return false;
1661	}
1662
1663	rx_ring->rx_stats.page_alloc_count++;
1664
1665	/* map page for use */
1666	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1667				 i40e_rx_pg_size(rx_ring),
1668				 DMA_FROM_DEVICE,
1669				 I40E_RX_DMA_ATTR);
1670
1671	/* if mapping failed free memory back to system since
1672	 * there isn't much point in holding memory we can't use
1673	 */
1674	if (dma_mapping_error(rx_ring->dev, dma)) {
1675		__free_pages(page, i40e_rx_pg_order(rx_ring));
1676		rx_ring->rx_stats.alloc_page_failed++;
1677		return false;
1678	}
1679
1680	bi->dma = dma;
1681	bi->page = page;
1682	bi->page_offset = rx_ring->rx_offset;
1683	page_ref_add(page, USHRT_MAX - 1);
1684	bi->pagecnt_bias = USHRT_MAX;
1685
1686	return true;
1687}
1688
1689/**
1690 * i40e_alloc_rx_buffers - Replace used receive buffers
1691 * @rx_ring: ring to place buffers on
1692 * @cleaned_count: number of buffers to replace
1693 *
1694 * Returns false if all allocations were successful, true if any fail
1695 **/
1696bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1697{
1698	u16 ntu = rx_ring->next_to_use;
1699	union i40e_rx_desc *rx_desc;
1700	struct i40e_rx_buffer *bi;
1701
1702	/* do nothing if no valid netdev defined */
1703	if (!rx_ring->netdev || !cleaned_count)
1704		return false;
1705
1706	rx_desc = I40E_RX_DESC(rx_ring, ntu);
1707	bi = i40e_rx_bi(rx_ring, ntu);
1708
1709	do {
1710		if (!i40e_alloc_mapped_page(rx_ring, bi))
1711			goto no_buffers;
1712
1713		/* sync the buffer for use by the device */
1714		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1715						 bi->page_offset,
1716						 rx_ring->rx_buf_len,
1717						 DMA_FROM_DEVICE);
1718
1719		/* Refresh the desc even if buffer_addrs didn't change
1720		 * because each write-back erases this info.
1721		 */
1722		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1723
1724		rx_desc++;
1725		bi++;
1726		ntu++;
1727		if (unlikely(ntu == rx_ring->count)) {
1728			rx_desc = I40E_RX_DESC(rx_ring, 0);
1729			bi = i40e_rx_bi(rx_ring, 0);
1730			ntu = 0;
1731		}
1732
1733		/* clear the status bits for the next_to_use descriptor */
1734		rx_desc->wb.qword1.status_error_len = 0;
1735
1736		cleaned_count--;
1737	} while (cleaned_count);
1738
1739	if (rx_ring->next_to_use != ntu)
1740		i40e_release_rx_desc(rx_ring, ntu);
1741
1742	return false;
1743
1744no_buffers:
1745	if (rx_ring->next_to_use != ntu)
1746		i40e_release_rx_desc(rx_ring, ntu);
1747
1748	/* make sure to come back via polling to try again after
1749	 * allocation failure
1750	 */
1751	return true;
1752}
1753
1754/**
1755 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1756 * @vsi: the VSI we care about
1757 * @skb: skb currently being received and modified
1758 * @rx_desc: the receive descriptor
1759 **/
1760static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1761				    struct sk_buff *skb,
1762				    union i40e_rx_desc *rx_desc)
1763{
1764	struct i40e_rx_ptype_decoded decoded;
1765	u32 rx_error, rx_status;
1766	bool ipv4, ipv6;
1767	u8 ptype;
1768	u64 qword;
1769
1770	qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1771	ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1772	rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1773		   I40E_RXD_QW1_ERROR_SHIFT;
1774	rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1775		    I40E_RXD_QW1_STATUS_SHIFT;
1776	decoded = decode_rx_desc_ptype(ptype);
1777
1778	skb->ip_summed = CHECKSUM_NONE;
1779
1780	skb_checksum_none_assert(skb);
1781
1782	/* Rx csum enabled and ip headers found? */
1783	if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1784		return;
1785
1786	/* did the hardware decode the packet and checksum? */
1787	if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1788		return;
1789
1790	/* both known and outer_ip must be set for the below code to work */
1791	if (!(decoded.known && decoded.outer_ip))
1792		return;
1793
1794	ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1795	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1796	ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1797	       (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1798
1799	if (ipv4 &&
1800	    (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1801			 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1802		goto checksum_fail;
1803
1804	/* likely incorrect csum if alternate IP extension headers found */
1805	if (ipv6 &&
1806	    rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1807		/* don't increment checksum err here, non-fatal err */
1808		return;
1809
1810	/* there was some L4 error, count error and punt packet to the stack */
1811	if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1812		goto checksum_fail;
1813
1814	/* handle packets that were not able to be checksummed due
1815	 * to arrival speed, in this case the stack can compute
1816	 * the csum.
1817	 */
1818	if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1819		return;
1820
1821	/* If there is an outer header present that might contain a checksum
1822	 * we need to bump the checksum level by 1 to reflect the fact that
1823	 * we are indicating we validated the inner checksum.
1824	 */
1825	if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1826		skb->csum_level = 1;
1827
1828	/* Only report checksum unnecessary for TCP, UDP, or SCTP */
1829	switch (decoded.inner_prot) {
1830	case I40E_RX_PTYPE_INNER_PROT_TCP:
1831	case I40E_RX_PTYPE_INNER_PROT_UDP:
1832	case I40E_RX_PTYPE_INNER_PROT_SCTP:
1833		skb->ip_summed = CHECKSUM_UNNECESSARY;
1834		fallthrough;
1835	default:
1836		break;
1837	}
1838
1839	return;
1840
1841checksum_fail:
1842	vsi->back->hw_csum_rx_error++;
1843}
1844
1845/**
1846 * i40e_ptype_to_htype - get a hash type
1847 * @ptype: the ptype value from the descriptor
1848 *
1849 * Returns a hash type to be used by skb_set_hash
1850 **/
1851static inline int i40e_ptype_to_htype(u8 ptype)
1852{
1853	struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1854
1855	if (!decoded.known)
1856		return PKT_HASH_TYPE_NONE;
1857
1858	if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1859	    decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1860		return PKT_HASH_TYPE_L4;
1861	else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1862		 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1863		return PKT_HASH_TYPE_L3;
1864	else
1865		return PKT_HASH_TYPE_L2;
1866}
1867
1868/**
1869 * i40e_rx_hash - set the hash value in the skb
1870 * @ring: descriptor ring
1871 * @rx_desc: specific descriptor
1872 * @skb: skb currently being received and modified
1873 * @rx_ptype: Rx packet type
1874 **/
1875static inline void i40e_rx_hash(struct i40e_ring *ring,
1876				union i40e_rx_desc *rx_desc,
1877				struct sk_buff *skb,
1878				u8 rx_ptype)
1879{
1880	u32 hash;
1881	const __le64 rss_mask =
1882		cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1883			    I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1884
1885	if (!(ring->netdev->features & NETIF_F_RXHASH))
1886		return;
1887
1888	if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1889		hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1890		skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1891	}
1892}
1893
1894/**
1895 * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1896 * @rx_ring: rx descriptor ring packet is being transacted on
1897 * @rx_desc: pointer to the EOP Rx descriptor
1898 * @skb: pointer to current skb being populated
1899 *
1900 * This function checks the ring, descriptor, and packet information in
1901 * order to populate the hash, checksum, VLAN, protocol, and
1902 * other fields within the skb.
1903 **/
1904void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1905			     union i40e_rx_desc *rx_desc, struct sk_buff *skb)
1906{
1907	u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1908	u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1909			I40E_RXD_QW1_STATUS_SHIFT;
1910	u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1911	u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1912		   I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1913	u8 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1914		      I40E_RXD_QW1_PTYPE_SHIFT;
1915
1916	if (unlikely(tsynvalid))
1917		i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1918
1919	i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1920
1921	i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1922
1923	skb_record_rx_queue(skb, rx_ring->queue_index);
1924
1925	if (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
1926		__le16 vlan_tag = rx_desc->wb.qword0.lo_dword.l2tag1;
1927
1928		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1929				       le16_to_cpu(vlan_tag));
1930	}
1931
1932	/* modifies the skb - consumes the enet header */
1933	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1934}
1935
1936/**
1937 * i40e_cleanup_headers - Correct empty headers
1938 * @rx_ring: rx descriptor ring packet is being transacted on
1939 * @skb: pointer to current skb being fixed
1940 * @rx_desc: pointer to the EOP Rx descriptor
1941 *
1942 * In addition if skb is not at least 60 bytes we need to pad it so that
1943 * it is large enough to qualify as a valid Ethernet frame.
1944 *
1945 * Returns true if an error was encountered and skb was freed.
1946 **/
1947static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
1948				 union i40e_rx_desc *rx_desc)
1949
1950{
1951	/* ERR_MASK will only have valid bits if EOP set, and
1952	 * what we are doing here is actually checking
1953	 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1954	 * the error field
1955	 */
1956	if (unlikely(i40e_test_staterr(rx_desc,
1957				       BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1958		dev_kfree_skb_any(skb);
1959		return true;
1960	}
1961
1962	/* if eth_skb_pad returns an error the skb was freed */
1963	if (eth_skb_pad(skb))
1964		return true;
1965
1966	return false;
1967}
1968
1969/**
1970 * i40e_can_reuse_rx_page - Determine if page can be reused for another Rx
1971 * @rx_buffer: buffer containing the page
1972 * @rx_stats: rx stats structure for the rx ring
1973 * @rx_buffer_pgcnt: buffer page refcount pre xdp_do_redirect() call
1974 *
1975 * If page is reusable, we have a green light for calling i40e_reuse_rx_page,
1976 * which will assign the current buffer to the buffer that next_to_alloc is
1977 * pointing to; otherwise, the DMA mapping needs to be destroyed and
1978 * page freed.
1979 *
1980 * rx_stats will be updated to indicate whether the page was waived
1981 * or busy if it could not be reused.
1982 */
1983static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
1984				   struct i40e_rx_queue_stats *rx_stats,
1985				   int rx_buffer_pgcnt)
1986{
1987	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1988	struct page *page = rx_buffer->page;
1989
1990	/* Is any reuse possible? */
1991	if (!dev_page_is_reusable(page)) {
1992		rx_stats->page_waive_count++;
1993		return false;
1994	}
1995
1996#if (PAGE_SIZE < 8192)
1997	/* if we are only owner of page we can reuse it */
1998	if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1)) {
1999		rx_stats->page_busy_count++;
2000		return false;
2001	}
2002#else
2003#define I40E_LAST_OFFSET \
2004	(SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
2005	if (rx_buffer->page_offset > I40E_LAST_OFFSET) {
2006		rx_stats->page_busy_count++;
2007		return false;
2008	}
2009#endif
2010
2011	/* If we have drained the page fragment pool we need to update
2012	 * the pagecnt_bias and page count so that we fully restock the
2013	 * number of references the driver holds.
2014	 */
2015	if (unlikely(pagecnt_bias == 1)) {
2016		page_ref_add(page, USHRT_MAX - 1);
2017		rx_buffer->pagecnt_bias = USHRT_MAX;
2018	}
2019
2020	return true;
2021}
2022
2023/**
2024 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
2025 * @rx_ring: rx descriptor ring to transact packets on
2026 * @rx_buffer: buffer containing page to add
2027 * @skb: sk_buff to place the data into
2028 * @size: packet length from rx_desc
2029 *
2030 * This function will add the data contained in rx_buffer->page to the skb.
2031 * It will just attach the page as a frag to the skb.
2032 *
2033 * The function will then update the page offset.
2034 **/
2035static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
2036			     struct i40e_rx_buffer *rx_buffer,
2037			     struct sk_buff *skb,
2038			     unsigned int size)
2039{
2040#if (PAGE_SIZE < 8192)
2041	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2042#else
2043	unsigned int truesize = SKB_DATA_ALIGN(size + rx_ring->rx_offset);
2044#endif
2045
2046	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2047			rx_buffer->page_offset, size, truesize);
2048
2049	/* page is being used so we must update the page offset */
2050#if (PAGE_SIZE < 8192)
2051	rx_buffer->page_offset ^= truesize;
2052#else
2053	rx_buffer->page_offset += truesize;
2054#endif
2055}
2056
2057/**
2058 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
2059 * @rx_ring: rx descriptor ring to transact packets on
2060 * @size: size of buffer to add to skb
2061 * @rx_buffer_pgcnt: buffer page refcount
2062 *
2063 * This function will pull an Rx buffer from the ring and synchronize it
2064 * for use by the CPU.
2065 */
2066static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
2067						 const unsigned int size,
2068						 int *rx_buffer_pgcnt)
2069{
2070	struct i40e_rx_buffer *rx_buffer;
2071
2072	rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2073	*rx_buffer_pgcnt =
2074#if (PAGE_SIZE < 8192)
2075		page_count(rx_buffer->page);
2076#else
2077		0;
2078#endif
2079	prefetch_page_address(rx_buffer->page);
2080
2081	/* we are reusing so sync this buffer for CPU use */
2082	dma_sync_single_range_for_cpu(rx_ring->dev,
2083				      rx_buffer->dma,
2084				      rx_buffer->page_offset,
2085				      size,
2086				      DMA_FROM_DEVICE);
2087
2088	/* We have pulled a buffer for use, so decrement pagecnt_bias */
2089	rx_buffer->pagecnt_bias--;
2090
2091	return rx_buffer;
2092}
2093
2094/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2095 * i40e_construct_skb - Allocate skb and populate it
2096 * @rx_ring: rx descriptor ring to transact packets on
2097 * @rx_buffer: rx buffer to pull data from
2098 * @xdp: xdp_buff pointing to the data
2099 *
2100 * This function allocates an skb.  It then populates it with the page
2101 * data from the current receive descriptor, taking care to set up the
2102 * skb correctly.
2103 */
2104static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
2105					  struct i40e_rx_buffer *rx_buffer,
2106					  struct xdp_buff *xdp)
2107{
2108	unsigned int size = xdp->data_end - xdp->data;
2109#if (PAGE_SIZE < 8192)
2110	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2111#else
2112	unsigned int truesize = SKB_DATA_ALIGN(size);
2113#endif
2114	unsigned int headlen;
2115	struct sk_buff *skb;
 
2116
2117	/* prefetch first cache line of first page */
2118	net_prefetch(xdp->data);
2119
2120	/* Note, we get here by enabling legacy-rx via:
2121	 *
2122	 *    ethtool --set-priv-flags <dev> legacy-rx on
2123	 *
2124	 * In this mode, we currently get 0 extra XDP headroom as
2125	 * opposed to having legacy-rx off, where we process XDP
2126	 * packets going to stack via i40e_build_skb(). The latter
2127	 * provides us currently with 192 bytes of headroom.
2128	 *
2129	 * For i40e_construct_skb() mode it means that the
2130	 * xdp->data_meta will always point to xdp->data, since
2131	 * the helper cannot expand the head. Should this ever
2132	 * change in future for legacy-rx mode on, then lets also
2133	 * add xdp->data_meta handling here.
2134	 */
2135
2136	/* allocate a skb to store the frags */
2137	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
2138			       I40E_RX_HDR_SIZE,
2139			       GFP_ATOMIC | __GFP_NOWARN);
2140	if (unlikely(!skb))
2141		return NULL;
2142
2143	/* Determine available headroom for copy */
2144	headlen = size;
2145	if (headlen > I40E_RX_HDR_SIZE)
2146		headlen = eth_get_headlen(skb->dev, xdp->data,
2147					  I40E_RX_HDR_SIZE);
2148
2149	/* align pull length to size of long to optimize memcpy performance */
2150	memcpy(__skb_put(skb, headlen), xdp->data,
2151	       ALIGN(headlen, sizeof(long)));
2152
 
 
 
 
 
2153	/* update all of the pointers */
2154	size -= headlen;
2155	if (size) {
 
 
 
 
2156		skb_add_rx_frag(skb, 0, rx_buffer->page,
2157				rx_buffer->page_offset + headlen,
2158				size, truesize);
2159
2160		/* buffer is used by skb, update page_offset */
2161#if (PAGE_SIZE < 8192)
2162		rx_buffer->page_offset ^= truesize;
2163#else
2164		rx_buffer->page_offset += truesize;
2165#endif
2166	} else {
2167		/* buffer is unused, reset bias back to rx_buffer */
2168		rx_buffer->pagecnt_bias++;
2169	}
2170
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2171	return skb;
2172}
2173
2174/**
2175 * i40e_build_skb - Build skb around an existing buffer
2176 * @rx_ring: Rx descriptor ring to transact packets on
2177 * @rx_buffer: Rx buffer to pull data from
2178 * @xdp: xdp_buff pointing to the data
2179 *
2180 * This function builds an skb around an existing Rx buffer, taking care
2181 * to set up the skb correctly and avoid any memcpy overhead.
2182 */
2183static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
2184				      struct i40e_rx_buffer *rx_buffer,
2185				      struct xdp_buff *xdp)
2186{
2187	unsigned int metasize = xdp->data - xdp->data_meta;
2188#if (PAGE_SIZE < 8192)
2189	unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2190#else
2191	unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2192				SKB_DATA_ALIGN(xdp->data_end -
2193					       xdp->data_hard_start);
2194#endif
2195	struct sk_buff *skb;
 
2196
2197	/* Prefetch first cache line of first page. If xdp->data_meta
2198	 * is unused, this points exactly as xdp->data, otherwise we
2199	 * likely have a consumer accessing first few bytes of meta
2200	 * data, and then actual data.
2201	 */
2202	net_prefetch(xdp->data_meta);
2203
 
 
 
 
 
2204	/* build an skb around the page buffer */
2205	skb = napi_build_skb(xdp->data_hard_start, truesize);
2206	if (unlikely(!skb))
2207		return NULL;
2208
2209	/* update pointers within the skb to store the data */
2210	skb_reserve(skb, xdp->data - xdp->data_hard_start);
2211	__skb_put(skb, xdp->data_end - xdp->data);
2212	if (metasize)
2213		skb_metadata_set(skb, metasize);
2214
2215	/* buffer is used by skb, update page_offset */
2216#if (PAGE_SIZE < 8192)
2217	rx_buffer->page_offset ^= truesize;
2218#else
2219	rx_buffer->page_offset += truesize;
2220#endif
2221
2222	return skb;
2223}
 
2224
2225/**
2226 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
2227 * @rx_ring: rx descriptor ring to transact packets on
2228 * @rx_buffer: rx buffer to pull data from
2229 * @rx_buffer_pgcnt: rx buffer page refcount pre xdp_do_redirect() call
2230 *
2231 * This function will clean up the contents of the rx_buffer.  It will
2232 * either recycle the buffer or unmap it and free the associated resources.
2233 */
2234static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
2235			       struct i40e_rx_buffer *rx_buffer,
2236			       int rx_buffer_pgcnt)
2237{
2238	if (i40e_can_reuse_rx_page(rx_buffer, &rx_ring->rx_stats, rx_buffer_pgcnt)) {
2239		/* hand second half of page back to the ring */
2240		i40e_reuse_rx_page(rx_ring, rx_buffer);
2241	} else {
2242		/* we are not reusing the buffer so unmap it */
2243		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2244				     i40e_rx_pg_size(rx_ring),
2245				     DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
2246		__page_frag_cache_drain(rx_buffer->page,
2247					rx_buffer->pagecnt_bias);
2248		/* clear contents of buffer_info */
2249		rx_buffer->page = NULL;
2250	}
 
 
2251}
2252
2253/**
2254 * i40e_is_non_eop - process handling of non-EOP buffers
2255 * @rx_ring: Rx ring being processed
2256 * @rx_desc: Rx descriptor for current buffer
2257 *
2258 * If the buffer is an EOP buffer, this function exits returning false,
2259 * otherwise return true indicating that this is in fact a non-EOP buffer.
2260 */
2261static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
2262			    union i40e_rx_desc *rx_desc)
2263{
2264	/* if we are the last buffer then there is nothing else to do */
2265#define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
2266	if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
2267		return false;
2268
2269	rx_ring->rx_stats.non_eop_descs++;
2270
2271	return true;
2272}
2273
2274static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
2275			      struct i40e_ring *xdp_ring);
2276
2277int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp, struct i40e_ring *xdp_ring)
2278{
2279	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2280
2281	if (unlikely(!xdpf))
2282		return I40E_XDP_CONSUMED;
2283
2284	return i40e_xmit_xdp_ring(xdpf, xdp_ring);
2285}
2286
2287/**
2288 * i40e_run_xdp - run an XDP program
2289 * @rx_ring: Rx ring being processed
2290 * @xdp: XDP buffer containing the frame
2291 * @xdp_prog: XDP program to run
2292 **/
2293static int i40e_run_xdp(struct i40e_ring *rx_ring, struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
2294{
2295	int err, result = I40E_XDP_PASS;
2296	struct i40e_ring *xdp_ring;
2297	u32 act;
2298
2299	if (!xdp_prog)
2300		goto xdp_out;
2301
2302	prefetchw(xdp->data_hard_start); /* xdp_frame write */
2303
2304	act = bpf_prog_run_xdp(xdp_prog, xdp);
2305	switch (act) {
2306	case XDP_PASS:
2307		break;
2308	case XDP_TX:
2309		xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2310		result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
2311		if (result == I40E_XDP_CONSUMED)
2312			goto out_failure;
2313		break;
2314	case XDP_REDIRECT:
2315		err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
2316		if (err)
2317			goto out_failure;
2318		result = I40E_XDP_REDIR;
2319		break;
2320	default:
2321		bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
2322		fallthrough;
2323	case XDP_ABORTED:
2324out_failure:
2325		trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2326		fallthrough; /* handle aborts by dropping packet */
2327	case XDP_DROP:
2328		result = I40E_XDP_CONSUMED;
2329		break;
2330	}
2331xdp_out:
2332	return result;
2333}
2334
2335/**
2336 * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
2337 * @rx_ring: Rx ring
2338 * @rx_buffer: Rx buffer to adjust
2339 * @size: Size of adjustment
2340 **/
2341static void i40e_rx_buffer_flip(struct i40e_ring *rx_ring,
2342				struct i40e_rx_buffer *rx_buffer,
2343				unsigned int size)
2344{
2345	unsigned int truesize = i40e_rx_frame_truesize(rx_ring, size);
2346
2347#if (PAGE_SIZE < 8192)
2348	rx_buffer->page_offset ^= truesize;
2349#else
2350	rx_buffer->page_offset += truesize;
2351#endif
2352}
2353
2354/**
2355 * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
2356 * @xdp_ring: XDP Tx ring
2357 *
2358 * This function updates the XDP Tx ring tail register.
2359 **/
2360void i40e_xdp_ring_update_tail(struct i40e_ring *xdp_ring)
2361{
2362	/* Force memory writes to complete before letting h/w
2363	 * know there are new descriptors to fetch.
2364	 */
2365	wmb();
2366	writel_relaxed(xdp_ring->next_to_use, xdp_ring->tail);
2367}
2368
2369/**
2370 * i40e_update_rx_stats - Update Rx ring statistics
2371 * @rx_ring: rx descriptor ring
2372 * @total_rx_bytes: number of bytes received
2373 * @total_rx_packets: number of packets received
2374 *
2375 * This function updates the Rx ring statistics.
2376 **/
2377void i40e_update_rx_stats(struct i40e_ring *rx_ring,
2378			  unsigned int total_rx_bytes,
2379			  unsigned int total_rx_packets)
2380{
2381	u64_stats_update_begin(&rx_ring->syncp);
2382	rx_ring->stats.packets += total_rx_packets;
2383	rx_ring->stats.bytes += total_rx_bytes;
2384	u64_stats_update_end(&rx_ring->syncp);
2385	rx_ring->q_vector->rx.total_packets += total_rx_packets;
2386	rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2387}
2388
2389/**
2390 * i40e_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map
2391 * @rx_ring: Rx ring
2392 * @xdp_res: Result of the receive batch
2393 *
2394 * This function bumps XDP Tx tail and/or flush redirect map, and
2395 * should be called when a batch of packets has been processed in the
2396 * napi loop.
2397 **/
2398void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
2399{
2400	if (xdp_res & I40E_XDP_REDIR)
2401		xdp_do_flush_map();
2402
2403	if (xdp_res & I40E_XDP_TX) {
2404		struct i40e_ring *xdp_ring =
2405			rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2406
2407		i40e_xdp_ring_update_tail(xdp_ring);
2408	}
2409}
2410
2411/**
2412 * i40e_inc_ntc: Advance the next_to_clean index
2413 * @rx_ring: Rx ring
2414 **/
2415static void i40e_inc_ntc(struct i40e_ring *rx_ring)
2416{
2417	u32 ntc = rx_ring->next_to_clean + 1;
2418
2419	ntc = (ntc < rx_ring->count) ? ntc : 0;
2420	rx_ring->next_to_clean = ntc;
2421	prefetch(I40E_RX_DESC(rx_ring, ntc));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2422}
2423
2424/**
2425 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2426 * @rx_ring: rx descriptor ring to transact packets on
2427 * @budget: Total limit on number of packets to process
2428 * @rx_cleaned: Out parameter of the number of packets processed
2429 *
2430 * This function provides a "bounce buffer" approach to Rx interrupt
2431 * processing.  The advantage to this is that on systems that have
2432 * expensive overhead for IOMMU access this provides a means of avoiding
2433 * it by maintaining the mapping of the page to the system.
2434 *
2435 * Returns amount of work completed
2436 **/
2437static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget,
2438			     unsigned int *rx_cleaned)
2439{
2440	unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2441	u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
 
2442	unsigned int offset = rx_ring->rx_offset;
2443	struct sk_buff *skb = rx_ring->skb;
2444	unsigned int xdp_xmit = 0;
2445	struct bpf_prog *xdp_prog;
2446	bool failure = false;
2447	struct xdp_buff xdp;
2448	int xdp_res = 0;
2449
2450#if (PAGE_SIZE < 8192)
2451	frame_sz = i40e_rx_frame_truesize(rx_ring, 0);
2452#endif
2453	xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2454
2455	xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2456
2457	while (likely(total_rx_packets < (unsigned int)budget)) {
 
2458		struct i40e_rx_buffer *rx_buffer;
2459		union i40e_rx_desc *rx_desc;
2460		int rx_buffer_pgcnt;
2461		unsigned int size;
 
 
2462		u64 qword;
2463
2464		/* return some buffers to hardware, one at a time is too slow */
2465		if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
2466			failure = failure ||
2467				  i40e_alloc_rx_buffers(rx_ring, cleaned_count);
2468			cleaned_count = 0;
2469		}
2470
2471		rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
2472
2473		/* status_error_len will always be zero for unused descriptors
2474		 * because it's cleared in cleanup, and overlaps with hdr_addr
2475		 * which is always zero because packet split isn't used, if the
2476		 * hardware wrote DD then the length will be non-zero
2477		 */
2478		qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2479
2480		/* This memory barrier is needed to keep us from reading
2481		 * any other fields out of the rx_desc until we have
2482		 * verified the descriptor has been written back.
2483		 */
2484		dma_rmb();
2485
2486		if (i40e_rx_is_programming_status(qword)) {
2487			i40e_clean_programming_status(rx_ring,
2488						      rx_desc->raw.qword[0],
2489						      qword);
2490			rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2491			i40e_inc_ntc(rx_ring);
2492			i40e_reuse_rx_page(rx_ring, rx_buffer);
2493			cleaned_count++;
 
 
 
 
 
 
 
2494			continue;
2495		}
2496
2497		size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
2498		       I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
2499		if (!size)
2500			break;
2501
2502		i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
2503		rx_buffer = i40e_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
 
 
 
 
2504
2505		/* retrieve a buffer from the ring */
2506		if (!skb) {
2507			unsigned char *hard_start;
2508
2509			hard_start = page_address(rx_buffer->page) +
2510				     rx_buffer->page_offset - offset;
2511			xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2512			xdp_buff_clear_frags_flag(&xdp);
2513#if (PAGE_SIZE > 4096)
2514			/* At larger PAGE_SIZE, frame_sz depend on len size */
2515			xdp.frame_sz = i40e_rx_frame_truesize(rx_ring, size);
2516#endif
2517			xdp_res = i40e_run_xdp(rx_ring, &xdp, xdp_prog);
 
 
 
 
2518		}
2519
 
 
 
 
 
2520		if (xdp_res) {
2521			if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
2522				xdp_xmit |= xdp_res;
2523				i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
 
 
 
 
2524			} else {
2525				rx_buffer->pagecnt_bias++;
2526			}
2527			total_rx_bytes += size;
2528			total_rx_packets++;
2529		} else if (skb) {
2530			i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
2531		} else if (ring_uses_build_skb(rx_ring)) {
2532			skb = i40e_build_skb(rx_ring, rx_buffer, &xdp);
2533		} else {
2534			skb = i40e_construct_skb(rx_ring, rx_buffer, &xdp);
2535		}
 
 
 
 
 
 
 
 
 
2536
2537		/* exit if we failed to retrieve a buffer */
2538		if (!xdp_res && !skb) {
2539			rx_ring->rx_stats.alloc_buff_failed++;
2540			rx_buffer->pagecnt_bias++;
2541			break;
2542		}
2543
2544		i40e_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
2545		cleaned_count++;
2546
2547		i40e_inc_ntc(rx_ring);
2548		if (i40e_is_non_eop(rx_ring, rx_desc))
2549			continue;
2550
2551		if (xdp_res || i40e_cleanup_headers(rx_ring, skb, rx_desc)) {
2552			skb = NULL;
2553			continue;
2554		}
2555
2556		/* probably a little skewed due to removing CRC */
2557		total_rx_bytes += skb->len;
2558
2559		/* populate checksum, VLAN, and protocol */
2560		i40e_process_skb_fields(rx_ring, rx_desc, skb);
2561
2562		i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
2563		napi_gro_receive(&rx_ring->q_vector->napi, skb);
2564		skb = NULL;
2565
2566		/* update budget accounting */
2567		total_rx_packets++;
 
 
 
 
 
 
2568	}
2569
2570	i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
2571	rx_ring->skb = skb;
2572
2573	i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
2574
2575	*rx_cleaned = total_rx_packets;
2576
2577	/* guarantee a trip back through this routine if there was a failure */
2578	return failure ? budget : (int)total_rx_packets;
2579}
2580
2581static inline u32 i40e_buildreg_itr(const int type, u16 itr)
2582{
2583	u32 val;
2584
2585	/* We don't bother with setting the CLEARPBA bit as the data sheet
2586	 * points out doing so is "meaningless since it was already
2587	 * auto-cleared". The auto-clearing happens when the interrupt is
2588	 * asserted.
2589	 *
2590	 * Hardware errata 28 for also indicates that writing to a
2591	 * xxINT_DYN_CTLx CSR with INTENA_MSK (bit 31) set to 0 will clear
2592	 * an event in the PBA anyway so we need to rely on the automask
2593	 * to hold pending events for us until the interrupt is re-enabled
2594	 *
2595	 * The itr value is reported in microseconds, and the register
2596	 * value is recorded in 2 microsecond units. For this reason we
2597	 * only need to shift by the interval shift - 1 instead of the
2598	 * full value.
2599	 */
2600	itr &= I40E_ITR_MASK;
2601
2602	val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2603	      (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2604	      (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1));
2605
2606	return val;
2607}
2608
2609/* a small macro to shorten up some long lines */
2610#define INTREG I40E_PFINT_DYN_CTLN
2611
2612/* The act of updating the ITR will cause it to immediately trigger. In order
2613 * to prevent this from throwing off adaptive update statistics we defer the
2614 * update so that it can only happen so often. So after either Tx or Rx are
2615 * updated we make the adaptive scheme wait until either the ITR completely
2616 * expires via the next_update expiration or we have been through at least
2617 * 3 interrupts.
2618 */
2619#define ITR_COUNTDOWN_START 3
2620
2621/**
2622 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2623 * @vsi: the VSI we care about
2624 * @q_vector: q_vector for which itr is being updated and interrupt enabled
2625 *
2626 **/
2627static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2628					  struct i40e_q_vector *q_vector)
2629{
2630	struct i40e_hw *hw = &vsi->back->hw;
2631	u32 intval;
2632
2633	/* If we don't have MSIX, then we only need to re-enable icr0 */
2634	if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) {
2635		i40e_irq_dynamic_enable_icr0(vsi->back);
2636		return;
2637	}
2638
2639	/* These will do nothing if dynamic updates are not enabled */
2640	i40e_update_itr(q_vector, &q_vector->tx);
2641	i40e_update_itr(q_vector, &q_vector->rx);
2642
2643	/* This block of logic allows us to get away with only updating
2644	 * one ITR value with each interrupt. The idea is to perform a
2645	 * pseudo-lazy update with the following criteria.
2646	 *
2647	 * 1. Rx is given higher priority than Tx if both are in same state
2648	 * 2. If we must reduce an ITR that is given highest priority.
2649	 * 3. We then give priority to increasing ITR based on amount.
2650	 */
2651	if (q_vector->rx.target_itr < q_vector->rx.current_itr) {
2652		/* Rx ITR needs to be reduced, this is highest priority */
2653		intval = i40e_buildreg_itr(I40E_RX_ITR,
2654					   q_vector->rx.target_itr);
2655		q_vector->rx.current_itr = q_vector->rx.target_itr;
2656		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2657	} else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) ||
2658		   ((q_vector->rx.target_itr - q_vector->rx.current_itr) <
2659		    (q_vector->tx.target_itr - q_vector->tx.current_itr))) {
2660		/* Tx ITR needs to be reduced, this is second priority
2661		 * Tx ITR needs to be increased more than Rx, fourth priority
2662		 */
2663		intval = i40e_buildreg_itr(I40E_TX_ITR,
2664					   q_vector->tx.target_itr);
2665		q_vector->tx.current_itr = q_vector->tx.target_itr;
2666		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2667	} else if (q_vector->rx.current_itr != q_vector->rx.target_itr) {
2668		/* Rx ITR needs to be increased, third priority */
2669		intval = i40e_buildreg_itr(I40E_RX_ITR,
2670					   q_vector->rx.target_itr);
2671		q_vector->rx.current_itr = q_vector->rx.target_itr;
2672		q_vector->itr_countdown = ITR_COUNTDOWN_START;
2673	} else {
2674		/* No ITR update, lowest priority */
2675		intval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2676		if (q_vector->itr_countdown)
2677			q_vector->itr_countdown--;
2678	}
2679
2680	if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2681		wr32(hw, INTREG(q_vector->reg_idx), intval);
2682}
2683
2684/**
2685 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2686 * @napi: napi struct with our devices info in it
2687 * @budget: amount of work driver is allowed to do this pass, in packets
2688 *
2689 * This function will clean all queues associated with a q_vector.
2690 *
2691 * Returns the amount of work done
2692 **/
2693int i40e_napi_poll(struct napi_struct *napi, int budget)
2694{
2695	struct i40e_q_vector *q_vector =
2696			       container_of(napi, struct i40e_q_vector, napi);
2697	struct i40e_vsi *vsi = q_vector->vsi;
2698	struct i40e_ring *ring;
2699	bool tx_clean_complete = true;
2700	bool rx_clean_complete = true;
2701	unsigned int tx_cleaned = 0;
2702	unsigned int rx_cleaned = 0;
2703	bool clean_complete = true;
2704	bool arm_wb = false;
2705	int budget_per_ring;
2706	int work_done = 0;
2707
2708	if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2709		napi_complete(napi);
2710		return 0;
2711	}
2712
2713	/* Since the actual Tx work is minimal, we can give the Tx a larger
2714	 * budget and be more aggressive about cleaning up the Tx descriptors.
2715	 */
2716	i40e_for_each_ring(ring, q_vector->tx) {
2717		bool wd = ring->xsk_pool ?
2718			  i40e_clean_xdp_tx_irq(vsi, ring) :
2719			  i40e_clean_tx_irq(vsi, ring, budget, &tx_cleaned);
2720
2721		if (!wd) {
2722			clean_complete = tx_clean_complete = false;
2723			continue;
2724		}
2725		arm_wb |= ring->arm_wb;
2726		ring->arm_wb = false;
2727	}
2728
2729	/* Handle case where we are called by netpoll with a budget of 0 */
2730	if (budget <= 0)
2731		goto tx_only;
2732
2733	/* normally we have 1 Rx ring per q_vector */
2734	if (unlikely(q_vector->num_ringpairs > 1))
2735		/* We attempt to distribute budget to each Rx queue fairly, but
2736		 * don't allow the budget to go below 1 because that would exit
2737		 * polling early.
2738		 */
2739		budget_per_ring = max_t(int, budget / q_vector->num_ringpairs, 1);
2740	else
2741		/* Max of 1 Rx ring in this q_vector so give it the budget */
2742		budget_per_ring = budget;
2743
2744	i40e_for_each_ring(ring, q_vector->rx) {
2745		int cleaned = ring->xsk_pool ?
2746			      i40e_clean_rx_irq_zc(ring, budget_per_ring) :
2747			      i40e_clean_rx_irq(ring, budget_per_ring, &rx_cleaned);
2748
2749		work_done += cleaned;
2750		/* if we clean as many as budgeted, we must not be done */
2751		if (cleaned >= budget_per_ring)
2752			clean_complete = rx_clean_complete = false;
2753	}
2754
2755	if (!i40e_enabled_xdp_vsi(vsi))
2756		trace_i40e_napi_poll(napi, q_vector, budget, budget_per_ring, rx_cleaned,
2757				     tx_cleaned, rx_clean_complete, tx_clean_complete);
2758
2759	/* If work not completed, return budget and polling will return */
2760	if (!clean_complete) {
2761		int cpu_id = smp_processor_id();
2762
2763		/* It is possible that the interrupt affinity has changed but,
2764		 * if the cpu is pegged at 100%, polling will never exit while
2765		 * traffic continues and the interrupt will be stuck on this
2766		 * cpu.  We check to make sure affinity is correct before we
2767		 * continue to poll, otherwise we must stop polling so the
2768		 * interrupt can move to the correct cpu.
2769		 */
2770		if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
2771			/* Tell napi that we are done polling */
2772			napi_complete_done(napi, work_done);
2773
2774			/* Force an interrupt */
2775			i40e_force_wb(vsi, q_vector);
2776
2777			/* Return budget-1 so that polling stops */
2778			return budget - 1;
2779		}
2780tx_only:
2781		if (arm_wb) {
2782			q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2783			i40e_enable_wb_on_itr(vsi, q_vector);
2784		}
2785		return budget;
2786	}
2787
2788	if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2789		q_vector->arm_wb_state = false;
2790
2791	/* Exit the polling mode, but don't re-enable interrupts if stack might
2792	 * poll us due to busy-polling
2793	 */
2794	if (likely(napi_complete_done(napi, work_done)))
2795		i40e_update_enable_itr(vsi, q_vector);
2796
2797	return min(work_done, budget - 1);
2798}
2799
2800/**
2801 * i40e_atr - Add a Flow Director ATR filter
2802 * @tx_ring:  ring to add programming descriptor to
2803 * @skb:      send buffer
2804 * @tx_flags: send tx flags
2805 **/
2806static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2807		     u32 tx_flags)
2808{
2809	struct i40e_filter_program_desc *fdir_desc;
2810	struct i40e_pf *pf = tx_ring->vsi->back;
2811	union {
2812		unsigned char *network;
2813		struct iphdr *ipv4;
2814		struct ipv6hdr *ipv6;
2815	} hdr;
2816	struct tcphdr *th;
2817	unsigned int hlen;
2818	u32 flex_ptype, dtype_cmd;
2819	int l4_proto;
2820	u16 i;
2821
2822	/* make sure ATR is enabled */
2823	if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2824		return;
2825
2826	if (test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2827		return;
2828
2829	/* if sampling is disabled do nothing */
2830	if (!tx_ring->atr_sample_rate)
2831		return;
2832
2833	/* Currently only IPv4/IPv6 with TCP is supported */
2834	if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2835		return;
2836
2837	/* snag network header to get L4 type and address */
2838	hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2839		      skb_inner_network_header(skb) : skb_network_header(skb);
2840
2841	/* Note: tx_flags gets modified to reflect inner protocols in
2842	 * tx_enable_csum function if encap is enabled.
2843	 */
2844	if (tx_flags & I40E_TX_FLAGS_IPV4) {
2845		/* access ihl as u8 to avoid unaligned access on ia64 */
2846		hlen = (hdr.network[0] & 0x0F) << 2;
2847		l4_proto = hdr.ipv4->protocol;
2848	} else {
2849		/* find the start of the innermost ipv6 header */
2850		unsigned int inner_hlen = hdr.network - skb->data;
2851		unsigned int h_offset = inner_hlen;
2852
2853		/* this function updates h_offset to the end of the header */
2854		l4_proto =
2855		  ipv6_find_hdr(skb, &h_offset, IPPROTO_TCP, NULL, NULL);
2856		/* hlen will contain our best estimate of the tcp header */
2857		hlen = h_offset - inner_hlen;
2858	}
2859
2860	if (l4_proto != IPPROTO_TCP)
2861		return;
2862
2863	th = (struct tcphdr *)(hdr.network + hlen);
2864
2865	/* Due to lack of space, no more new filters can be programmed */
2866	if (th->syn && test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2867		return;
2868	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED) {
2869		/* HW ATR eviction will take care of removing filters on FIN
2870		 * and RST packets.
2871		 */
2872		if (th->fin || th->rst)
2873			return;
2874	}
2875
2876	tx_ring->atr_count++;
2877
2878	/* sample on all syn/fin/rst packets or once every atr sample rate */
2879	if (!th->fin &&
2880	    !th->syn &&
2881	    !th->rst &&
2882	    (tx_ring->atr_count < tx_ring->atr_sample_rate))
2883		return;
2884
2885	tx_ring->atr_count = 0;
2886
2887	/* grab the next descriptor */
2888	i = tx_ring->next_to_use;
2889	fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2890
2891	i++;
2892	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2893
2894	flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2895		      I40E_TXD_FLTR_QW0_QINDEX_MASK;
2896	flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2897		      (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2898		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2899		      (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2900		       I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2901
2902	flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2903
2904	dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2905
2906	dtype_cmd |= (th->fin || th->rst) ?
2907		     (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2908		      I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2909		     (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2910		      I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2911
2912	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2913		     I40E_TXD_FLTR_QW1_DEST_SHIFT;
2914
2915	dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2916		     I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2917
2918	dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2919	if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2920		dtype_cmd |=
2921			((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2922			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2923			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2924	else
2925		dtype_cmd |=
2926			((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2927			I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2928			I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2929
2930	if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED)
2931		dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2932
2933	fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2934	fdir_desc->rsvd = cpu_to_le32(0);
2935	fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2936	fdir_desc->fd_id = cpu_to_le32(0);
2937}
2938
2939/**
2940 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2941 * @skb:     send buffer
2942 * @tx_ring: ring to send buffer on
2943 * @flags:   the tx flags to be set
2944 *
2945 * Checks the skb and set up correspondingly several generic transmit flags
2946 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2947 *
2948 * Returns error code indicate the frame should be dropped upon error and the
2949 * otherwise  returns 0 to indicate the flags has been set properly.
2950 **/
2951static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2952					     struct i40e_ring *tx_ring,
2953					     u32 *flags)
2954{
2955	__be16 protocol = skb->protocol;
2956	u32  tx_flags = 0;
2957
2958	if (protocol == htons(ETH_P_8021Q) &&
2959	    !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2960		/* When HW VLAN acceleration is turned off by the user the
2961		 * stack sets the protocol to 8021q so that the driver
2962		 * can take any steps required to support the SW only
2963		 * VLAN handling.  In our case the driver doesn't need
2964		 * to take any further steps so just set the protocol
2965		 * to the encapsulated ethertype.
2966		 */
2967		skb->protocol = vlan_get_protocol(skb);
2968		goto out;
2969	}
2970
2971	/* if we have a HW VLAN tag being added, default to the HW one */
2972	if (skb_vlan_tag_present(skb)) {
2973		tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2974		tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2975	/* else if it is a SW VLAN, check the next protocol and store the tag */
2976	} else if (protocol == htons(ETH_P_8021Q)) {
2977		struct vlan_hdr *vhdr, _vhdr;
2978
2979		vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2980		if (!vhdr)
2981			return -EINVAL;
2982
2983		protocol = vhdr->h_vlan_encapsulated_proto;
2984		tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2985		tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2986	}
2987
2988	if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2989		goto out;
2990
2991	/* Insert 802.1p priority into VLAN header */
2992	if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2993	    (skb->priority != TC_PRIO_CONTROL)) {
2994		tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2995		tx_flags |= (skb->priority & 0x7) <<
2996				I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2997		if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2998			struct vlan_ethhdr *vhdr;
2999			int rc;
3000
3001			rc = skb_cow_head(skb, 0);
3002			if (rc < 0)
3003				return rc;
3004			vhdr = (struct vlan_ethhdr *)skb->data;
3005			vhdr->h_vlan_TCI = htons(tx_flags >>
3006						 I40E_TX_FLAGS_VLAN_SHIFT);
3007		} else {
3008			tx_flags |= I40E_TX_FLAGS_HW_VLAN;
3009		}
3010	}
3011
3012out:
3013	*flags = tx_flags;
3014	return 0;
3015}
3016
3017/**
3018 * i40e_tso - set up the tso context descriptor
3019 * @first:    pointer to first Tx buffer for xmit
3020 * @hdr_len:  ptr to the size of the packet header
3021 * @cd_type_cmd_tso_mss: Quad Word 1
3022 *
3023 * Returns 0 if no TSO can happen, 1 if tso is going, or error
3024 **/
3025static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
3026		    u64 *cd_type_cmd_tso_mss)
3027{
3028	struct sk_buff *skb = first->skb;
3029	u64 cd_cmd, cd_tso_len, cd_mss;
3030	__be16 protocol;
3031	union {
3032		struct iphdr *v4;
3033		struct ipv6hdr *v6;
3034		unsigned char *hdr;
3035	} ip;
3036	union {
3037		struct tcphdr *tcp;
3038		struct udphdr *udp;
3039		unsigned char *hdr;
3040	} l4;
3041	u32 paylen, l4_offset;
3042	u16 gso_size;
3043	int err;
3044
3045	if (skb->ip_summed != CHECKSUM_PARTIAL)
3046		return 0;
3047
3048	if (!skb_is_gso(skb))
3049		return 0;
3050
3051	err = skb_cow_head(skb, 0);
3052	if (err < 0)
3053		return err;
3054
3055	protocol = vlan_get_protocol(skb);
3056
3057	if (eth_p_mpls(protocol))
3058		ip.hdr = skb_inner_network_header(skb);
3059	else
3060		ip.hdr = skb_network_header(skb);
3061	l4.hdr = skb_checksum_start(skb);
3062
3063	/* initialize outer IP header fields */
3064	if (ip.v4->version == 4) {
3065		ip.v4->tot_len = 0;
3066		ip.v4->check = 0;
3067
3068		first->tx_flags |= I40E_TX_FLAGS_TSO;
3069	} else {
3070		ip.v6->payload_len = 0;
3071		first->tx_flags |= I40E_TX_FLAGS_TSO;
3072	}
3073
3074	if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
3075					 SKB_GSO_GRE_CSUM |
3076					 SKB_GSO_IPXIP4 |
3077					 SKB_GSO_IPXIP6 |
3078					 SKB_GSO_UDP_TUNNEL |
3079					 SKB_GSO_UDP_TUNNEL_CSUM)) {
3080		if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3081		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
3082			l4.udp->len = 0;
3083
3084			/* determine offset of outer transport header */
3085			l4_offset = l4.hdr - skb->data;
3086
3087			/* remove payload length from outer checksum */
3088			paylen = skb->len - l4_offset;
3089			csum_replace_by_diff(&l4.udp->check,
3090					     (__force __wsum)htonl(paylen));
3091		}
3092
3093		/* reset pointers to inner headers */
3094		ip.hdr = skb_inner_network_header(skb);
3095		l4.hdr = skb_inner_transport_header(skb);
3096
3097		/* initialize inner IP header fields */
3098		if (ip.v4->version == 4) {
3099			ip.v4->tot_len = 0;
3100			ip.v4->check = 0;
3101		} else {
3102			ip.v6->payload_len = 0;
3103		}
3104	}
3105
3106	/* determine offset of inner transport header */
3107	l4_offset = l4.hdr - skb->data;
3108
3109	/* remove payload length from inner checksum */
3110	paylen = skb->len - l4_offset;
3111
3112	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
3113		csum_replace_by_diff(&l4.udp->check, (__force __wsum)htonl(paylen));
3114		/* compute length of segmentation header */
3115		*hdr_len = sizeof(*l4.udp) + l4_offset;
3116	} else {
3117		csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3118		/* compute length of segmentation header */
3119		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
3120	}
3121
3122	/* pull values out of skb_shinfo */
3123	gso_size = skb_shinfo(skb)->gso_size;
3124
3125	/* update GSO size and bytecount with header size */
3126	first->gso_segs = skb_shinfo(skb)->gso_segs;
3127	first->bytecount += (first->gso_segs - 1) * *hdr_len;
3128
3129	/* find the field values */
3130	cd_cmd = I40E_TX_CTX_DESC_TSO;
3131	cd_tso_len = skb->len - *hdr_len;
3132	cd_mss = gso_size;
3133	*cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
3134				(cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
3135				(cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
3136	return 1;
3137}
3138
3139/**
3140 * i40e_tsyn - set up the tsyn context descriptor
3141 * @tx_ring:  ptr to the ring to send
3142 * @skb:      ptr to the skb we're sending
3143 * @tx_flags: the collected send information
3144 * @cd_type_cmd_tso_mss: Quad Word 1
3145 *
3146 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
3147 **/
3148static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
3149		     u32 tx_flags, u64 *cd_type_cmd_tso_mss)
3150{
3151	struct i40e_pf *pf;
3152
3153	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
3154		return 0;
3155
3156	/* Tx timestamps cannot be sampled when doing TSO */
3157	if (tx_flags & I40E_TX_FLAGS_TSO)
3158		return 0;
3159
3160	/* only timestamp the outbound packet if the user has requested it and
3161	 * we are not already transmitting a packet to be timestamped
3162	 */
3163	pf = i40e_netdev_to_pf(tx_ring->netdev);
3164	if (!(pf->flags & I40E_FLAG_PTP))
3165		return 0;
3166
3167	if (pf->ptp_tx &&
3168	    !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
3169		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3170		pf->ptp_tx_start = jiffies;
3171		pf->ptp_tx_skb = skb_get(skb);
3172	} else {
3173		pf->tx_hwtstamp_skipped++;
3174		return 0;
3175	}
3176
3177	*cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
3178				I40E_TXD_CTX_QW1_CMD_SHIFT;
3179
3180	return 1;
3181}
3182
3183/**
3184 * i40e_tx_enable_csum - Enable Tx checksum offloads
3185 * @skb: send buffer
3186 * @tx_flags: pointer to Tx flags currently set
3187 * @td_cmd: Tx descriptor command bits to set
3188 * @td_offset: Tx descriptor header offsets to set
3189 * @tx_ring: Tx descriptor ring
3190 * @cd_tunneling: ptr to context desc bits
3191 **/
3192static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
3193			       u32 *td_cmd, u32 *td_offset,
3194			       struct i40e_ring *tx_ring,
3195			       u32 *cd_tunneling)
3196{
3197	union {
3198		struct iphdr *v4;
3199		struct ipv6hdr *v6;
3200		unsigned char *hdr;
3201	} ip;
3202	union {
3203		struct tcphdr *tcp;
3204		struct udphdr *udp;
3205		unsigned char *hdr;
3206	} l4;
3207	unsigned char *exthdr;
3208	u32 offset, cmd = 0;
3209	__be16 frag_off;
3210	__be16 protocol;
3211	u8 l4_proto = 0;
3212
3213	if (skb->ip_summed != CHECKSUM_PARTIAL)
3214		return 0;
3215
3216	protocol = vlan_get_protocol(skb);
3217
3218	if (eth_p_mpls(protocol)) {
3219		ip.hdr = skb_inner_network_header(skb);
3220		l4.hdr = skb_checksum_start(skb);
3221	} else {
3222		ip.hdr = skb_network_header(skb);
3223		l4.hdr = skb_transport_header(skb);
3224	}
3225
3226	/* set the tx_flags to indicate the IP protocol type. this is
3227	 * required so that checksum header computation below is accurate.
3228	 */
3229	if (ip.v4->version == 4)
3230		*tx_flags |= I40E_TX_FLAGS_IPV4;
3231	else
3232		*tx_flags |= I40E_TX_FLAGS_IPV6;
3233
3234	/* compute outer L2 header size */
3235	offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
3236
3237	if (skb->encapsulation) {
3238		u32 tunnel = 0;
3239		/* define outer network header type */
3240		if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3241			tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3242				  I40E_TX_CTX_EXT_IP_IPV4 :
3243				  I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
3244
3245			l4_proto = ip.v4->protocol;
3246		} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3247			int ret;
3248
3249			tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
3250
3251			exthdr = ip.hdr + sizeof(*ip.v6);
3252			l4_proto = ip.v6->nexthdr;
3253			ret = ipv6_skip_exthdr(skb, exthdr - skb->data,
3254					       &l4_proto, &frag_off);
3255			if (ret < 0)
3256				return -1;
3257		}
3258
3259		/* define outer transport */
3260		switch (l4_proto) {
3261		case IPPROTO_UDP:
3262			tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
3263			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3264			break;
3265		case IPPROTO_GRE:
3266			tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
3267			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3268			break;
3269		case IPPROTO_IPIP:
3270		case IPPROTO_IPV6:
3271			*tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3272			l4.hdr = skb_inner_network_header(skb);
3273			break;
3274		default:
3275			if (*tx_flags & I40E_TX_FLAGS_TSO)
3276				return -1;
3277
3278			skb_checksum_help(skb);
3279			return 0;
3280		}
3281
3282		/* compute outer L3 header size */
3283		tunnel |= ((l4.hdr - ip.hdr) / 4) <<
3284			  I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
3285
3286		/* switch IP header pointer from outer to inner header */
3287		ip.hdr = skb_inner_network_header(skb);
3288
3289		/* compute tunnel header size */
3290		tunnel |= ((ip.hdr - l4.hdr) / 2) <<
3291			  I40E_TXD_CTX_QW0_NATLEN_SHIFT;
3292
3293		/* indicate if we need to offload outer UDP header */
3294		if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
3295		    !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3296		    (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
3297			tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
3298
3299		/* record tunnel offload values */
3300		*cd_tunneling |= tunnel;
3301
3302		/* switch L4 header pointer from outer to inner */
3303		l4.hdr = skb_inner_transport_header(skb);
3304		l4_proto = 0;
3305
3306		/* reset type as we transition from outer to inner headers */
3307		*tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
3308		if (ip.v4->version == 4)
3309			*tx_flags |= I40E_TX_FLAGS_IPV4;
3310		if (ip.v6->version == 6)
3311			*tx_flags |= I40E_TX_FLAGS_IPV6;
3312	}
3313
3314	/* Enable IP checksum offloads */
3315	if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3316		l4_proto = ip.v4->protocol;
3317		/* the stack computes the IP header already, the only time we
3318		 * need the hardware to recompute it is in the case of TSO.
3319		 */
3320		cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3321		       I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
3322		       I40E_TX_DESC_CMD_IIPT_IPV4;
3323	} else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3324		cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
3325
3326		exthdr = ip.hdr + sizeof(*ip.v6);
3327		l4_proto = ip.v6->nexthdr;
3328		if (l4.hdr != exthdr)
3329			ipv6_skip_exthdr(skb, exthdr - skb->data,
3330					 &l4_proto, &frag_off);
3331	}
3332
3333	/* compute inner L3 header size */
3334	offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
3335
3336	/* Enable L4 checksum offloads */
3337	switch (l4_proto) {
3338	case IPPROTO_TCP:
3339		/* enable checksum offloads */
3340		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
3341		offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3342		break;
3343	case IPPROTO_SCTP:
3344		/* enable SCTP checksum offload */
3345		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
3346		offset |= (sizeof(struct sctphdr) >> 2) <<
3347			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3348		break;
3349	case IPPROTO_UDP:
3350		/* enable UDP checksum offload */
3351		cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
3352		offset |= (sizeof(struct udphdr) >> 2) <<
3353			  I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3354		break;
3355	default:
3356		if (*tx_flags & I40E_TX_FLAGS_TSO)
3357			return -1;
3358		skb_checksum_help(skb);
3359		return 0;
3360	}
3361
3362	*td_cmd |= cmd;
3363	*td_offset |= offset;
3364
3365	return 1;
3366}
3367
3368/**
3369 * i40e_create_tx_ctx - Build the Tx context descriptor
3370 * @tx_ring:  ring to create the descriptor on
3371 * @cd_type_cmd_tso_mss: Quad Word 1
3372 * @cd_tunneling: Quad Word 0 - bits 0-31
3373 * @cd_l2tag2: Quad Word 0 - bits 32-63
3374 **/
3375static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
3376			       const u64 cd_type_cmd_tso_mss,
3377			       const u32 cd_tunneling, const u32 cd_l2tag2)
3378{
3379	struct i40e_tx_context_desc *context_desc;
3380	int i = tx_ring->next_to_use;
3381
3382	if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
3383	    !cd_tunneling && !cd_l2tag2)
3384		return;
3385
3386	/* grab the next descriptor */
3387	context_desc = I40E_TX_CTXTDESC(tx_ring, i);
3388
3389	i++;
3390	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3391
3392	/* cpu_to_le32 and assign to struct fields */
3393	context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
3394	context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
3395	context_desc->rsvd = cpu_to_le16(0);
3396	context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
3397}
3398
3399/**
3400 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
3401 * @tx_ring: the ring to be checked
3402 * @size:    the size buffer we want to assure is available
3403 *
3404 * Returns -EBUSY if a stop is needed, else 0
3405 **/
3406int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
3407{
3408	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3409	/* Memory barrier before checking head and tail */
3410	smp_mb();
3411
3412	++tx_ring->tx_stats.tx_stopped;
3413
3414	/* Check again in a case another CPU has just made room available. */
3415	if (likely(I40E_DESC_UNUSED(tx_ring) < size))
3416		return -EBUSY;
3417
3418	/* A reprieve! - use start_queue because it doesn't call schedule */
3419	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3420	++tx_ring->tx_stats.restart_queue;
3421	return 0;
3422}
3423
3424/**
3425 * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
3426 * @skb:      send buffer
3427 *
3428 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
3429 * and so we need to figure out the cases where we need to linearize the skb.
3430 *
3431 * For TSO we need to count the TSO header and segment payload separately.
3432 * As such we need to check cases where we have 7 fragments or more as we
3433 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
3434 * the segment payload in the first descriptor, and another 7 for the
3435 * fragments.
3436 **/
3437bool __i40e_chk_linearize(struct sk_buff *skb)
3438{
3439	const skb_frag_t *frag, *stale;
3440	int nr_frags, sum;
3441
3442	/* no need to check if number of frags is less than 7 */
3443	nr_frags = skb_shinfo(skb)->nr_frags;
3444	if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
3445		return false;
3446
3447	/* We need to walk through the list and validate that each group
3448	 * of 6 fragments totals at least gso_size.
3449	 */
3450	nr_frags -= I40E_MAX_BUFFER_TXD - 2;
3451	frag = &skb_shinfo(skb)->frags[0];
3452
3453	/* Initialize size to the negative value of gso_size minus 1.  We
3454	 * use this as the worst case scenerio in which the frag ahead
3455	 * of us only provides one byte which is why we are limited to 6
3456	 * descriptors for a single transmit as the header and previous
3457	 * fragment are already consuming 2 descriptors.
3458	 */
3459	sum = 1 - skb_shinfo(skb)->gso_size;
3460
3461	/* Add size of frags 0 through 4 to create our initial sum */
3462	sum += skb_frag_size(frag++);
3463	sum += skb_frag_size(frag++);
3464	sum += skb_frag_size(frag++);
3465	sum += skb_frag_size(frag++);
3466	sum += skb_frag_size(frag++);
3467
3468	/* Walk through fragments adding latest fragment, testing it, and
3469	 * then removing stale fragments from the sum.
3470	 */
3471	for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
3472		int stale_size = skb_frag_size(stale);
3473
3474		sum += skb_frag_size(frag++);
3475
3476		/* The stale fragment may present us with a smaller
3477		 * descriptor than the actual fragment size. To account
3478		 * for that we need to remove all the data on the front and
3479		 * figure out what the remainder would be in the last
3480		 * descriptor associated with the fragment.
3481		 */
3482		if (stale_size > I40E_MAX_DATA_PER_TXD) {
3483			int align_pad = -(skb_frag_off(stale)) &
3484					(I40E_MAX_READ_REQ_SIZE - 1);
3485
3486			sum -= align_pad;
3487			stale_size -= align_pad;
3488
3489			do {
3490				sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3491				stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3492			} while (stale_size > I40E_MAX_DATA_PER_TXD);
3493		}
3494
3495		/* if sum is negative we failed to make sufficient progress */
3496		if (sum < 0)
3497			return true;
3498
3499		if (!nr_frags--)
3500			break;
3501
3502		sum -= stale_size;
3503	}
3504
3505	return false;
3506}
3507
3508/**
3509 * i40e_tx_map - Build the Tx descriptor
3510 * @tx_ring:  ring to send buffer on
3511 * @skb:      send buffer
3512 * @first:    first buffer info buffer to use
3513 * @tx_flags: collected send information
3514 * @hdr_len:  size of the packet header
3515 * @td_cmd:   the command field in the descriptor
3516 * @td_offset: offset for checksum or crc
3517 *
3518 * Returns 0 on success, -1 on failure to DMA
3519 **/
3520static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
3521			      struct i40e_tx_buffer *first, u32 tx_flags,
3522			      const u8 hdr_len, u32 td_cmd, u32 td_offset)
3523{
3524	unsigned int data_len = skb->data_len;
3525	unsigned int size = skb_headlen(skb);
3526	skb_frag_t *frag;
3527	struct i40e_tx_buffer *tx_bi;
3528	struct i40e_tx_desc *tx_desc;
3529	u16 i = tx_ring->next_to_use;
3530	u32 td_tag = 0;
3531	dma_addr_t dma;
3532	u16 desc_count = 1;
3533
3534	if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
3535		td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
3536		td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
3537			 I40E_TX_FLAGS_VLAN_SHIFT;
3538	}
3539
3540	first->tx_flags = tx_flags;
3541
3542	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3543
3544	tx_desc = I40E_TX_DESC(tx_ring, i);
3545	tx_bi = first;
3546
3547	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3548		unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3549
3550		if (dma_mapping_error(tx_ring->dev, dma))
3551			goto dma_error;
3552
3553		/* record length, and DMA address */
3554		dma_unmap_len_set(tx_bi, len, size);
3555		dma_unmap_addr_set(tx_bi, dma, dma);
3556
3557		/* align size to end of page */
3558		max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
3559		tx_desc->buffer_addr = cpu_to_le64(dma);
3560
3561		while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
3562			tx_desc->cmd_type_offset_bsz =
3563				build_ctob(td_cmd, td_offset,
3564					   max_data, td_tag);
3565
3566			tx_desc++;
3567			i++;
3568			desc_count++;
3569
3570			if (i == tx_ring->count) {
3571				tx_desc = I40E_TX_DESC(tx_ring, 0);
3572				i = 0;
3573			}
3574
3575			dma += max_data;
3576			size -= max_data;
3577
3578			max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3579			tx_desc->buffer_addr = cpu_to_le64(dma);
3580		}
3581
3582		if (likely(!data_len))
3583			break;
3584
3585		tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3586							  size, td_tag);
3587
3588		tx_desc++;
3589		i++;
3590		desc_count++;
3591
3592		if (i == tx_ring->count) {
3593			tx_desc = I40E_TX_DESC(tx_ring, 0);
3594			i = 0;
3595		}
3596
3597		size = skb_frag_size(frag);
3598		data_len -= size;
3599
3600		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3601				       DMA_TO_DEVICE);
3602
3603		tx_bi = &tx_ring->tx_bi[i];
3604	}
3605
3606	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3607
3608	i++;
3609	if (i == tx_ring->count)
3610		i = 0;
3611
3612	tx_ring->next_to_use = i;
3613
3614	i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3615
3616	/* write last descriptor with EOP bit */
3617	td_cmd |= I40E_TX_DESC_CMD_EOP;
3618
3619	/* We OR these values together to check both against 4 (WB_STRIDE)
3620	 * below. This is safe since we don't re-use desc_count afterwards.
3621	 */
3622	desc_count |= ++tx_ring->packet_stride;
3623
3624	if (desc_count >= WB_STRIDE) {
3625		/* write last descriptor with RS bit set */
3626		td_cmd |= I40E_TX_DESC_CMD_RS;
3627		tx_ring->packet_stride = 0;
3628	}
3629
3630	tx_desc->cmd_type_offset_bsz =
3631			build_ctob(td_cmd, td_offset, size, td_tag);
3632
3633	skb_tx_timestamp(skb);
3634
3635	/* Force memory writes to complete before letting h/w know there
3636	 * are new descriptors to fetch.
3637	 *
3638	 * We also use this memory barrier to make certain all of the
3639	 * status bits have been updated before next_to_watch is written.
3640	 */
3641	wmb();
3642
3643	/* set next_to_watch value indicating a packet is present */
3644	first->next_to_watch = tx_desc;
3645
3646	/* notify HW of packet */
3647	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
3648		writel(i, tx_ring->tail);
3649	}
3650
3651	return 0;
3652
3653dma_error:
3654	dev_info(tx_ring->dev, "TX DMA map failed\n");
3655
3656	/* clear dma mappings for failed tx_bi map */
3657	for (;;) {
3658		tx_bi = &tx_ring->tx_bi[i];
3659		i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3660		if (tx_bi == first)
3661			break;
3662		if (i == 0)
3663			i = tx_ring->count;
3664		i--;
3665	}
3666
3667	tx_ring->next_to_use = i;
3668
3669	return -1;
3670}
3671
3672static u16 i40e_swdcb_skb_tx_hash(struct net_device *dev,
3673				  const struct sk_buff *skb,
3674				  u16 num_tx_queues)
3675{
3676	u32 jhash_initval_salt = 0xd631614b;
3677	u32 hash;
3678
3679	if (skb->sk && skb->sk->sk_hash)
3680		hash = skb->sk->sk_hash;
3681	else
3682		hash = (__force u16)skb->protocol ^ skb->hash;
3683
3684	hash = jhash_1word(hash, jhash_initval_salt);
3685
3686	return (u16)(((u64)hash * num_tx_queues) >> 32);
3687}
3688
3689u16 i40e_lan_select_queue(struct net_device *netdev,
3690			  struct sk_buff *skb,
3691			  struct net_device __always_unused *sb_dev)
3692{
3693	struct i40e_netdev_priv *np = netdev_priv(netdev);
3694	struct i40e_vsi *vsi = np->vsi;
3695	struct i40e_hw *hw;
3696	u16 qoffset;
3697	u16 qcount;
3698	u8 tclass;
3699	u16 hash;
3700	u8 prio;
3701
3702	/* is DCB enabled at all? */
3703	if (vsi->tc_config.numtc == 1 ||
3704	    i40e_is_tc_mqprio_enabled(vsi->back))
3705		return netdev_pick_tx(netdev, skb, sb_dev);
3706
3707	prio = skb->priority;
3708	hw = &vsi->back->hw;
3709	tclass = hw->local_dcbx_config.etscfg.prioritytable[prio];
3710	/* sanity check */
3711	if (unlikely(!(vsi->tc_config.enabled_tc & BIT(tclass))))
3712		tclass = 0;
3713
3714	/* select a queue assigned for the given TC */
3715	qcount = vsi->tc_config.tc_info[tclass].qcount;
3716	hash = i40e_swdcb_skb_tx_hash(netdev, skb, qcount);
3717
3718	qoffset = vsi->tc_config.tc_info[tclass].qoffset;
3719	return qoffset + hash;
3720}
3721
3722/**
3723 * i40e_xmit_xdp_ring - transmits an XDP buffer to an XDP Tx ring
3724 * @xdpf: data to transmit
3725 * @xdp_ring: XDP Tx ring
3726 **/
3727static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
3728			      struct i40e_ring *xdp_ring)
3729{
3730	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
3731	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
3732	u16 i = 0, index = xdp_ring->next_to_use;
3733	struct i40e_tx_buffer *tx_head = &xdp_ring->tx_bi[index];
3734	struct i40e_tx_buffer *tx_bi = tx_head;
3735	struct i40e_tx_desc *tx_desc = I40E_TX_DESC(xdp_ring, index);
3736	void *data = xdpf->data;
3737	u32 size = xdpf->len;
3738
3739	if (unlikely(I40E_DESC_UNUSED(xdp_ring) < 1 + nr_frags)) {
3740		xdp_ring->tx_stats.tx_busy++;
3741		return I40E_XDP_CONSUMED;
3742	}
3743
3744	tx_head->bytecount = xdp_get_frame_len(xdpf);
3745	tx_head->gso_segs = 1;
3746	tx_head->xdpf = xdpf;
3747
3748	for (;;) {
3749		dma_addr_t dma;
3750
3751		dma = dma_map_single(xdp_ring->dev, data, size, DMA_TO_DEVICE);
3752		if (dma_mapping_error(xdp_ring->dev, dma))
3753			goto unmap;
3754
3755		/* record length, and DMA address */
3756		dma_unmap_len_set(tx_bi, len, size);
3757		dma_unmap_addr_set(tx_bi, dma, dma);
3758
3759		tx_desc->buffer_addr = cpu_to_le64(dma);
3760		tx_desc->cmd_type_offset_bsz =
3761			build_ctob(I40E_TX_DESC_CMD_ICRC, 0, size, 0);
3762
3763		if (++index == xdp_ring->count)
3764			index = 0;
3765
3766		if (i == nr_frags)
3767			break;
3768
3769		tx_bi = &xdp_ring->tx_bi[index];
3770		tx_desc = I40E_TX_DESC(xdp_ring, index);
3771
3772		data = skb_frag_address(&sinfo->frags[i]);
3773		size = skb_frag_size(&sinfo->frags[i]);
3774		i++;
3775	}
3776
3777	tx_desc->cmd_type_offset_bsz |=
3778		cpu_to_le64(I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT);
3779
3780	/* Make certain all of the status bits have been updated
3781	 * before next_to_watch is written.
3782	 */
3783	smp_wmb();
3784
3785	xdp_ring->xdp_tx_active++;
3786
3787	tx_head->next_to_watch = tx_desc;
3788	xdp_ring->next_to_use = index;
3789
3790	return I40E_XDP_TX;
3791
3792unmap:
3793	for (;;) {
3794		tx_bi = &xdp_ring->tx_bi[index];
3795		if (dma_unmap_len(tx_bi, len))
3796			dma_unmap_page(xdp_ring->dev,
3797				       dma_unmap_addr(tx_bi, dma),
3798				       dma_unmap_len(tx_bi, len),
3799				       DMA_TO_DEVICE);
3800		dma_unmap_len_set(tx_bi, len, 0);
3801		if (tx_bi == tx_head)
3802			break;
3803
3804		if (!index)
3805			index += xdp_ring->count;
3806		index--;
3807	}
3808
3809	return I40E_XDP_CONSUMED;
3810}
3811
3812/**
3813 * i40e_xmit_frame_ring - Sends buffer on Tx ring
3814 * @skb:     send buffer
3815 * @tx_ring: ring to send buffer on
3816 *
3817 * Returns NETDEV_TX_OK if sent, else an error code
3818 **/
3819static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3820					struct i40e_ring *tx_ring)
3821{
3822	u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3823	u32 cd_tunneling = 0, cd_l2tag2 = 0;
3824	struct i40e_tx_buffer *first;
3825	u32 td_offset = 0;
3826	u32 tx_flags = 0;
3827	u32 td_cmd = 0;
3828	u8 hdr_len = 0;
3829	int tso, count;
3830	int tsyn;
3831
3832	/* prefetch the data, we'll need it later */
3833	prefetch(skb->data);
3834
3835	i40e_trace(xmit_frame_ring, skb, tx_ring);
3836
3837	count = i40e_xmit_descriptor_count(skb);
3838	if (i40e_chk_linearize(skb, count)) {
3839		if (__skb_linearize(skb)) {
3840			dev_kfree_skb_any(skb);
3841			return NETDEV_TX_OK;
3842		}
3843		count = i40e_txd_use_count(skb->len);
3844		tx_ring->tx_stats.tx_linearize++;
3845	}
3846
3847	/* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3848	 *       + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3849	 *       + 4 desc gap to avoid the cache line where head is,
3850	 *       + 1 desc for context descriptor,
3851	 * otherwise try next time
3852	 */
3853	if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3854		tx_ring->tx_stats.tx_busy++;
3855		return NETDEV_TX_BUSY;
3856	}
3857
3858	/* record the location of the first descriptor for this packet */
3859	first = &tx_ring->tx_bi[tx_ring->next_to_use];
3860	first->skb = skb;
3861	first->bytecount = skb->len;
3862	first->gso_segs = 1;
3863
3864	/* prepare the xmit flags */
3865	if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3866		goto out_drop;
3867
3868	tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3869
3870	if (tso < 0)
3871		goto out_drop;
3872	else if (tso)
3873		tx_flags |= I40E_TX_FLAGS_TSO;
3874
3875	/* Always offload the checksum, since it's in the data descriptor */
3876	tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3877				  tx_ring, &cd_tunneling);
3878	if (tso < 0)
3879		goto out_drop;
3880
3881	tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3882
3883	if (tsyn)
3884		tx_flags |= I40E_TX_FLAGS_TSYN;
3885
3886	/* always enable CRC insertion offload */
3887	td_cmd |= I40E_TX_DESC_CMD_ICRC;
3888
3889	i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3890			   cd_tunneling, cd_l2tag2);
3891
3892	/* Add Flow Director ATR if it's enabled.
3893	 *
3894	 * NOTE: this must always be directly before the data descriptor.
3895	 */
3896	i40e_atr(tx_ring, skb, tx_flags);
3897
3898	if (i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3899			td_cmd, td_offset))
3900		goto cleanup_tx_tstamp;
3901
3902	return NETDEV_TX_OK;
3903
3904out_drop:
3905	i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3906	dev_kfree_skb_any(first->skb);
3907	first->skb = NULL;
3908cleanup_tx_tstamp:
3909	if (unlikely(tx_flags & I40E_TX_FLAGS_TSYN)) {
3910		struct i40e_pf *pf = i40e_netdev_to_pf(tx_ring->netdev);
3911
3912		dev_kfree_skb_any(pf->ptp_tx_skb);
3913		pf->ptp_tx_skb = NULL;
3914		clear_bit_unlock(__I40E_PTP_TX_IN_PROGRESS, pf->state);
3915	}
3916
3917	return NETDEV_TX_OK;
3918}
3919
3920/**
3921 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3922 * @skb:    send buffer
3923 * @netdev: network interface device structure
3924 *
3925 * Returns NETDEV_TX_OK if sent, else an error code
3926 **/
3927netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3928{
3929	struct i40e_netdev_priv *np = netdev_priv(netdev);
3930	struct i40e_vsi *vsi = np->vsi;
3931	struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3932
3933	/* hardware can't handle really short frames, hardware padding works
3934	 * beyond this point
3935	 */
3936	if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3937		return NETDEV_TX_OK;
3938
3939	return i40e_xmit_frame_ring(skb, tx_ring);
3940}
3941
3942/**
3943 * i40e_xdp_xmit - Implements ndo_xdp_xmit
3944 * @dev: netdev
3945 * @n: number of frames
3946 * @frames: array of XDP buffer pointers
3947 * @flags: XDP extra info
3948 *
3949 * Returns number of frames successfully sent. Failed frames
3950 * will be free'ed by XDP core.
3951 *
3952 * For error cases, a negative errno code is returned and no-frames
3953 * are transmitted (caller must handle freeing frames).
3954 **/
3955int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
3956		  u32 flags)
3957{
3958	struct i40e_netdev_priv *np = netdev_priv(dev);
3959	unsigned int queue_index = smp_processor_id();
3960	struct i40e_vsi *vsi = np->vsi;
3961	struct i40e_pf *pf = vsi->back;
3962	struct i40e_ring *xdp_ring;
3963	int nxmit = 0;
3964	int i;
3965
3966	if (test_bit(__I40E_VSI_DOWN, vsi->state))
3967		return -ENETDOWN;
3968
3969	if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs ||
3970	    test_bit(__I40E_CONFIG_BUSY, pf->state))
3971		return -ENXIO;
3972
3973	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3974		return -EINVAL;
3975
3976	xdp_ring = vsi->xdp_rings[queue_index];
3977
3978	for (i = 0; i < n; i++) {
3979		struct xdp_frame *xdpf = frames[i];
3980		int err;
3981
3982		err = i40e_xmit_xdp_ring(xdpf, xdp_ring);
3983		if (err != I40E_XDP_TX)
3984			break;
3985		nxmit++;
3986	}
3987
3988	if (unlikely(flags & XDP_XMIT_FLUSH))
3989		i40e_xdp_ring_update_tail(xdp_ring);
3990
3991	return nxmit;
3992}