1 /******************************************************************************* 2 * 3 * Intel Ethernet Controller XL710 Family Linux Driver 4 * Copyright(c) 2013 - 2014 Intel Corporation. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms and conditions of the GNU General Public License, 8 * version 2, as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License along 16 * with this program. If not, see <http://www.gnu.org/licenses/>. 17 * 18 * The full GNU General Public License is included in this distribution in 19 * the file called "COPYING". 20 * 21 * Contact Information: 22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> 23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 24 * 25 ******************************************************************************/ 26 27 #include <linux/prefetch.h> 28 #include <net/busy_poll.h> 29 #include "i40e.h" 30 #include "i40e_prototype.h" 31 32 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size, 33 u32 td_tag) 34 { 35 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA | 36 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) | 37 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) | 38 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) | 39 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT)); 40 } 41 42 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS) 43 #define I40E_FD_CLEAN_DELAY 10 44 /** 45 * i40e_program_fdir_filter - Program a Flow Director filter 46 * @fdir_data: Packet data that will be filter parameters 47 * @raw_packet: the pre-allocated packet buffer for FDir 48 * @pf: The PF pointer 49 * @add: True for add/update, False for remove 50 **/ 51 int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet, 52 struct i40e_pf *pf, bool add) 53 { 54 struct i40e_filter_program_desc *fdir_desc; 55 struct i40e_tx_buffer *tx_buf, *first; 56 struct i40e_tx_desc *tx_desc; 57 struct i40e_ring *tx_ring; 58 unsigned int fpt, dcc; 59 struct i40e_vsi *vsi; 60 struct device *dev; 61 dma_addr_t dma; 62 u32 td_cmd = 0; 63 u16 delay = 0; 64 u16 i; 65 66 /* find existing FDIR VSI */ 67 vsi = NULL; 68 for (i = 0; i < pf->num_alloc_vsi; i++) 69 if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR) 70 vsi = pf->vsi[i]; 71 if (!vsi) 72 return -ENOENT; 73 74 tx_ring = vsi->tx_rings[0]; 75 dev = tx_ring->dev; 76 77 /* we need two descriptors to add/del a filter and we can wait */ 78 do { 79 if (I40E_DESC_UNUSED(tx_ring) > 1) 80 break; 81 msleep_interruptible(1); 82 delay++; 83 } while (delay < I40E_FD_CLEAN_DELAY); 84 85 if (!(I40E_DESC_UNUSED(tx_ring) > 1)) 86 return -EAGAIN; 87 88 dma = dma_map_single(dev, raw_packet, 89 I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE); 90 if (dma_mapping_error(dev, dma)) 91 goto dma_fail; 92 93 /* grab the next descriptor */ 94 i = tx_ring->next_to_use; 95 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); 96 first = &tx_ring->tx_bi[i]; 97 memset(first, 0, sizeof(struct i40e_tx_buffer)); 98 99 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0; 100 101 fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) & 102 I40E_TXD_FLTR_QW0_QINDEX_MASK; 103 104 fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) & 105 I40E_TXD_FLTR_QW0_FLEXOFF_MASK; 106 107 fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) & 108 I40E_TXD_FLTR_QW0_PCTYPE_MASK; 109 110 /* Use LAN VSI Id if not programmed by user */ 111 if (fdir_data->dest_vsi == 0) 112 fpt |= (pf->vsi[pf->lan_vsi]->id) << 113 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT; 114 else 115 fpt |= ((u32)fdir_data->dest_vsi << 116 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) & 117 I40E_TXD_FLTR_QW0_DEST_VSI_MASK; 118 119 dcc = I40E_TX_DESC_DTYPE_FILTER_PROG; 120 121 if (add) 122 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE << 123 I40E_TXD_FLTR_QW1_PCMD_SHIFT; 124 else 125 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE << 126 I40E_TXD_FLTR_QW1_PCMD_SHIFT; 127 128 dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) & 129 I40E_TXD_FLTR_QW1_DEST_MASK; 130 131 dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) & 132 I40E_TXD_FLTR_QW1_FD_STATUS_MASK; 133 134 if (fdir_data->cnt_index != 0) { 135 dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK; 136 dcc |= ((u32)fdir_data->cnt_index << 137 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) & 138 I40E_TXD_FLTR_QW1_CNTINDEX_MASK; 139 } 140 141 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt); 142 fdir_desc->rsvd = cpu_to_le32(0); 143 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc); 144 fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id); 145 146 /* Now program a dummy descriptor */ 147 i = tx_ring->next_to_use; 148 tx_desc = I40E_TX_DESC(tx_ring, i); 149 tx_buf = &tx_ring->tx_bi[i]; 150 151 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0; 152 153 memset(tx_buf, 0, sizeof(struct i40e_tx_buffer)); 154 155 /* record length, and DMA address */ 156 dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE); 157 dma_unmap_addr_set(tx_buf, dma, dma); 158 159 tx_desc->buffer_addr = cpu_to_le64(dma); 160 td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY; 161 162 tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB; 163 tx_buf->raw_buf = (void *)raw_packet; 164 165 tx_desc->cmd_type_offset_bsz = 166 build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0); 167 168 /* set the timestamp */ 169 tx_buf->time_stamp = jiffies; 170 171 /* Force memory writes to complete before letting h/w 172 * know there are new descriptors to fetch. 173 */ 174 wmb(); 175 176 /* Mark the data descriptor to be watched */ 177 first->next_to_watch = tx_desc; 178 179 writel(tx_ring->next_to_use, tx_ring->tail); 180 return 0; 181 182 dma_fail: 183 return -1; 184 } 185 186 #define IP_HEADER_OFFSET 14 187 #define I40E_UDPIP_DUMMY_PACKET_LEN 42 188 /** 189 * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters 190 * @vsi: pointer to the targeted VSI 191 * @fd_data: the flow director data required for the FDir descriptor 192 * @add: true adds a filter, false removes it 193 * 194 * Returns 0 if the filters were successfully added or removed 195 **/ 196 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi, 197 struct i40e_fdir_filter *fd_data, 198 bool add) 199 { 200 struct i40e_pf *pf = vsi->back; 201 struct udphdr *udp; 202 struct iphdr *ip; 203 bool err = false; 204 u8 *raw_packet; 205 int ret; 206 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, 207 0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0, 208 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; 209 210 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL); 211 if (!raw_packet) 212 return -ENOMEM; 213 memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN); 214 215 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); 216 udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET 217 + sizeof(struct iphdr)); 218 219 ip->daddr = fd_data->dst_ip[0]; 220 udp->dest = fd_data->dst_port; 221 ip->saddr = fd_data->src_ip[0]; 222 udp->source = fd_data->src_port; 223 224 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP; 225 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); 226 if (ret) { 227 dev_info(&pf->pdev->dev, 228 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n", 229 fd_data->pctype, fd_data->fd_id, ret); 230 err = true; 231 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) { 232 if (add) 233 dev_info(&pf->pdev->dev, 234 "Filter OK for PCTYPE %d loc = %d\n", 235 fd_data->pctype, fd_data->fd_id); 236 else 237 dev_info(&pf->pdev->dev, 238 "Filter deleted for PCTYPE %d loc = %d\n", 239 fd_data->pctype, fd_data->fd_id); 240 } 241 return err ? -EOPNOTSUPP : 0; 242 } 243 244 #define I40E_TCPIP_DUMMY_PACKET_LEN 54 245 /** 246 * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters 247 * @vsi: pointer to the targeted VSI 248 * @fd_data: the flow director data required for the FDir descriptor 249 * @add: true adds a filter, false removes it 250 * 251 * Returns 0 if the filters were successfully added or removed 252 **/ 253 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi, 254 struct i40e_fdir_filter *fd_data, 255 bool add) 256 { 257 struct i40e_pf *pf = vsi->back; 258 struct tcphdr *tcp; 259 struct iphdr *ip; 260 bool err = false; 261 u8 *raw_packet; 262 int ret; 263 /* Dummy packet */ 264 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, 265 0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0, 266 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11, 267 0x0, 0x72, 0, 0, 0, 0}; 268 269 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL); 270 if (!raw_packet) 271 return -ENOMEM; 272 memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN); 273 274 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); 275 tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET 276 + sizeof(struct iphdr)); 277 278 ip->daddr = fd_data->dst_ip[0]; 279 tcp->dest = fd_data->dst_port; 280 ip->saddr = fd_data->src_ip[0]; 281 tcp->source = fd_data->src_port; 282 283 if (add) { 284 pf->fd_tcp_rule++; 285 if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) { 286 dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n"); 287 pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED; 288 } 289 } else { 290 pf->fd_tcp_rule = (pf->fd_tcp_rule > 0) ? 291 (pf->fd_tcp_rule - 1) : 0; 292 if (pf->fd_tcp_rule == 0) { 293 pf->flags |= I40E_FLAG_FD_ATR_ENABLED; 294 dev_info(&pf->pdev->dev, "ATR re-enabled due to no sideband TCP/IPv4 rules\n"); 295 } 296 } 297 298 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP; 299 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); 300 301 if (ret) { 302 dev_info(&pf->pdev->dev, 303 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n", 304 fd_data->pctype, fd_data->fd_id, ret); 305 err = true; 306 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) { 307 if (add) 308 dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n", 309 fd_data->pctype, fd_data->fd_id); 310 else 311 dev_info(&pf->pdev->dev, 312 "Filter deleted for PCTYPE %d loc = %d\n", 313 fd_data->pctype, fd_data->fd_id); 314 } 315 316 return err ? -EOPNOTSUPP : 0; 317 } 318 319 /** 320 * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for 321 * a specific flow spec 322 * @vsi: pointer to the targeted VSI 323 * @fd_data: the flow director data required for the FDir descriptor 324 * @add: true adds a filter, false removes it 325 * 326 * Always returns -EOPNOTSUPP 327 **/ 328 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi, 329 struct i40e_fdir_filter *fd_data, 330 bool add) 331 { 332 return -EOPNOTSUPP; 333 } 334 335 #define I40E_IP_DUMMY_PACKET_LEN 34 336 /** 337 * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for 338 * a specific flow spec 339 * @vsi: pointer to the targeted VSI 340 * @fd_data: the flow director data required for the FDir descriptor 341 * @add: true adds a filter, false removes it 342 * 343 * Returns 0 if the filters were successfully added or removed 344 **/ 345 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi, 346 struct i40e_fdir_filter *fd_data, 347 bool add) 348 { 349 struct i40e_pf *pf = vsi->back; 350 struct iphdr *ip; 351 bool err = false; 352 u8 *raw_packet; 353 int ret; 354 int i; 355 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0, 356 0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0, 357 0, 0, 0, 0}; 358 359 for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER; 360 i <= I40E_FILTER_PCTYPE_FRAG_IPV4; i++) { 361 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL); 362 if (!raw_packet) 363 return -ENOMEM; 364 memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN); 365 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET); 366 367 ip->saddr = fd_data->src_ip[0]; 368 ip->daddr = fd_data->dst_ip[0]; 369 ip->protocol = 0; 370 371 fd_data->pctype = i; 372 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add); 373 374 if (ret) { 375 dev_info(&pf->pdev->dev, 376 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n", 377 fd_data->pctype, fd_data->fd_id, ret); 378 err = true; 379 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) { 380 if (add) 381 dev_info(&pf->pdev->dev, 382 "Filter OK for PCTYPE %d loc = %d\n", 383 fd_data->pctype, fd_data->fd_id); 384 else 385 dev_info(&pf->pdev->dev, 386 "Filter deleted for PCTYPE %d loc = %d\n", 387 fd_data->pctype, fd_data->fd_id); 388 } 389 } 390 391 return err ? -EOPNOTSUPP : 0; 392 } 393 394 /** 395 * i40e_add_del_fdir - Build raw packets to add/del fdir filter 396 * @vsi: pointer to the targeted VSI 397 * @cmd: command to get or set RX flow classification rules 398 * @add: true adds a filter, false removes it 399 * 400 **/ 401 int i40e_add_del_fdir(struct i40e_vsi *vsi, 402 struct i40e_fdir_filter *input, bool add) 403 { 404 struct i40e_pf *pf = vsi->back; 405 int ret; 406 407 switch (input->flow_type & ~FLOW_EXT) { 408 case TCP_V4_FLOW: 409 ret = i40e_add_del_fdir_tcpv4(vsi, input, add); 410 break; 411 case UDP_V4_FLOW: 412 ret = i40e_add_del_fdir_udpv4(vsi, input, add); 413 break; 414 case SCTP_V4_FLOW: 415 ret = i40e_add_del_fdir_sctpv4(vsi, input, add); 416 break; 417 case IPV4_FLOW: 418 ret = i40e_add_del_fdir_ipv4(vsi, input, add); 419 break; 420 case IP_USER_FLOW: 421 switch (input->ip4_proto) { 422 case IPPROTO_TCP: 423 ret = i40e_add_del_fdir_tcpv4(vsi, input, add); 424 break; 425 case IPPROTO_UDP: 426 ret = i40e_add_del_fdir_udpv4(vsi, input, add); 427 break; 428 case IPPROTO_SCTP: 429 ret = i40e_add_del_fdir_sctpv4(vsi, input, add); 430 break; 431 default: 432 ret = i40e_add_del_fdir_ipv4(vsi, input, add); 433 break; 434 } 435 break; 436 default: 437 dev_info(&pf->pdev->dev, "Could not specify spec type %d\n", 438 input->flow_type); 439 ret = -EINVAL; 440 } 441 442 /* The buffer allocated here is freed by the i40e_clean_tx_ring() */ 443 return ret; 444 } 445 446 /** 447 * i40e_fd_handle_status - check the Programming Status for FD 448 * @rx_ring: the Rx ring for this descriptor 449 * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor. 450 * @prog_id: the id originally used for programming 451 * 452 * This is used to verify if the FD programming or invalidation 453 * requested by SW to the HW is successful or not and take actions accordingly. 454 **/ 455 static void i40e_fd_handle_status(struct i40e_ring *rx_ring, 456 union i40e_rx_desc *rx_desc, u8 prog_id) 457 { 458 struct i40e_pf *pf = rx_ring->vsi->back; 459 struct pci_dev *pdev = pf->pdev; 460 u32 fcnt_prog, fcnt_avail; 461 u32 error; 462 u64 qw; 463 464 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len); 465 error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >> 466 I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT; 467 468 if (error == (0x1 << I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) { 469 if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) || 470 (I40E_DEBUG_FD & pf->hw.debug_mask)) 471 dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n", 472 rx_desc->wb.qword0.hi_dword.fd_id); 473 474 /* Check if the programming error is for ATR. 475 * If so, auto disable ATR and set a state for 476 * flush in progress. Next time we come here if flush is in 477 * progress do nothing, once flush is complete the state will 478 * be cleared. 479 */ 480 if (test_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state)) 481 return; 482 483 pf->fd_add_err++; 484 /* store the current atr filter count */ 485 pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf); 486 487 if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) && 488 (pf->auto_disable_flags & I40E_FLAG_FD_SB_ENABLED)) { 489 pf->auto_disable_flags |= I40E_FLAG_FD_ATR_ENABLED; 490 set_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state); 491 } 492 493 /* filter programming failed most likely due to table full */ 494 fcnt_prog = i40e_get_global_fd_count(pf); 495 fcnt_avail = pf->fdir_pf_filter_count; 496 /* If ATR is running fcnt_prog can quickly change, 497 * if we are very close to full, it makes sense to disable 498 * FD ATR/SB and then re-enable it when there is room. 499 */ 500 if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) { 501 if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) && 502 !(pf->auto_disable_flags & 503 I40E_FLAG_FD_SB_ENABLED)) { 504 dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n"); 505 pf->auto_disable_flags |= 506 I40E_FLAG_FD_SB_ENABLED; 507 } 508 } else { 509 dev_info(&pdev->dev, 510 "FD filter programming failed due to incorrect filter parameters\n"); 511 } 512 } else if (error == 513 (0x1 << I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) { 514 if (I40E_DEBUG_FD & pf->hw.debug_mask) 515 dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n", 516 rx_desc->wb.qword0.hi_dword.fd_id); 517 } 518 } 519 520 /** 521 * i40e_unmap_and_free_tx_resource - Release a Tx buffer 522 * @ring: the ring that owns the buffer 523 * @tx_buffer: the buffer to free 524 **/ 525 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring, 526 struct i40e_tx_buffer *tx_buffer) 527 { 528 if (tx_buffer->skb) { 529 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB) 530 kfree(tx_buffer->raw_buf); 531 else 532 dev_kfree_skb_any(tx_buffer->skb); 533 534 if (dma_unmap_len(tx_buffer, len)) 535 dma_unmap_single(ring->dev, 536 dma_unmap_addr(tx_buffer, dma), 537 dma_unmap_len(tx_buffer, len), 538 DMA_TO_DEVICE); 539 } else if (dma_unmap_len(tx_buffer, len)) { 540 dma_unmap_page(ring->dev, 541 dma_unmap_addr(tx_buffer, dma), 542 dma_unmap_len(tx_buffer, len), 543 DMA_TO_DEVICE); 544 } 545 tx_buffer->next_to_watch = NULL; 546 tx_buffer->skb = NULL; 547 dma_unmap_len_set(tx_buffer, len, 0); 548 /* tx_buffer must be completely set up in the transmit path */ 549 } 550 551 /** 552 * i40e_clean_tx_ring - Free any empty Tx buffers 553 * @tx_ring: ring to be cleaned 554 **/ 555 void i40e_clean_tx_ring(struct i40e_ring *tx_ring) 556 { 557 unsigned long bi_size; 558 u16 i; 559 560 /* ring already cleared, nothing to do */ 561 if (!tx_ring->tx_bi) 562 return; 563 564 /* Free all the Tx ring sk_buffs */ 565 for (i = 0; i < tx_ring->count; i++) 566 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]); 567 568 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; 569 memset(tx_ring->tx_bi, 0, bi_size); 570 571 /* Zero out the descriptor ring */ 572 memset(tx_ring->desc, 0, tx_ring->size); 573 574 tx_ring->next_to_use = 0; 575 tx_ring->next_to_clean = 0; 576 577 if (!tx_ring->netdev) 578 return; 579 580 /* cleanup Tx queue statistics */ 581 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev, 582 tx_ring->queue_index)); 583 } 584 585 /** 586 * i40e_free_tx_resources - Free Tx resources per queue 587 * @tx_ring: Tx descriptor ring for a specific queue 588 * 589 * Free all transmit software resources 590 **/ 591 void i40e_free_tx_resources(struct i40e_ring *tx_ring) 592 { 593 i40e_clean_tx_ring(tx_ring); 594 kfree(tx_ring->tx_bi); 595 tx_ring->tx_bi = NULL; 596 597 if (tx_ring->desc) { 598 dma_free_coherent(tx_ring->dev, tx_ring->size, 599 tx_ring->desc, tx_ring->dma); 600 tx_ring->desc = NULL; 601 } 602 } 603 604 /** 605 * i40e_get_head - Retrieve head from head writeback 606 * @tx_ring: tx ring to fetch head of 607 * 608 * Returns value of Tx ring head based on value stored 609 * in head write-back location 610 **/ 611 static inline u32 i40e_get_head(struct i40e_ring *tx_ring) 612 { 613 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count; 614 615 return le32_to_cpu(*(volatile __le32 *)head); 616 } 617 618 /** 619 * i40e_get_tx_pending - how many tx descriptors not processed 620 * @tx_ring: the ring of descriptors 621 * 622 * Since there is no access to the ring head register 623 * in XL710, we need to use our local copies 624 **/ 625 static u32 i40e_get_tx_pending(struct i40e_ring *ring) 626 { 627 u32 head, tail; 628 629 head = i40e_get_head(ring); 630 tail = readl(ring->tail); 631 632 if (head != tail) 633 return (head < tail) ? 634 tail - head : (tail + ring->count - head); 635 636 return 0; 637 } 638 639 /** 640 * i40e_check_tx_hang - Is there a hang in the Tx queue 641 * @tx_ring: the ring of descriptors 642 **/ 643 static bool i40e_check_tx_hang(struct i40e_ring *tx_ring) 644 { 645 u32 tx_done = tx_ring->stats.packets; 646 u32 tx_done_old = tx_ring->tx_stats.tx_done_old; 647 u32 tx_pending = i40e_get_tx_pending(tx_ring); 648 struct i40e_pf *pf = tx_ring->vsi->back; 649 bool ret = false; 650 651 clear_check_for_tx_hang(tx_ring); 652 653 /* Check for a hung queue, but be thorough. This verifies 654 * that a transmit has been completed since the previous 655 * check AND there is at least one packet pending. The 656 * ARMED bit is set to indicate a potential hang. The 657 * bit is cleared if a pause frame is received to remove 658 * false hang detection due to PFC or 802.3x frames. By 659 * requiring this to fail twice we avoid races with 660 * PFC clearing the ARMED bit and conditions where we 661 * run the check_tx_hang logic with a transmit completion 662 * pending but without time to complete it yet. 663 */ 664 if ((tx_done_old == tx_done) && tx_pending) { 665 /* make sure it is true for two checks in a row */ 666 ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED, 667 &tx_ring->state); 668 } else if (tx_done_old == tx_done && 669 (tx_pending < I40E_MIN_DESC_PENDING) && (tx_pending > 0)) { 670 if (I40E_DEBUG_FLOW & pf->hw.debug_mask) 671 dev_info(tx_ring->dev, "HW needs some more descs to do a cacheline flush. tx_pending %d, queue %d", 672 tx_pending, tx_ring->queue_index); 673 pf->tx_sluggish_count++; 674 } else { 675 /* update completed stats and disarm the hang check */ 676 tx_ring->tx_stats.tx_done_old = tx_done; 677 clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state); 678 } 679 680 return ret; 681 } 682 683 #define WB_STRIDE 0x3 684 685 /** 686 * i40e_clean_tx_irq - Reclaim resources after transmit completes 687 * @tx_ring: tx ring to clean 688 * @budget: how many cleans we're allowed 689 * 690 * Returns true if there's any budget left (e.g. the clean is finished) 691 **/ 692 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget) 693 { 694 u16 i = tx_ring->next_to_clean; 695 struct i40e_tx_buffer *tx_buf; 696 struct i40e_tx_desc *tx_head; 697 struct i40e_tx_desc *tx_desc; 698 unsigned int total_packets = 0; 699 unsigned int total_bytes = 0; 700 701 tx_buf = &tx_ring->tx_bi[i]; 702 tx_desc = I40E_TX_DESC(tx_ring, i); 703 i -= tx_ring->count; 704 705 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring)); 706 707 do { 708 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch; 709 710 /* if next_to_watch is not set then there is no work pending */ 711 if (!eop_desc) 712 break; 713 714 /* prevent any other reads prior to eop_desc */ 715 read_barrier_depends(); 716 717 /* we have caught up to head, no work left to do */ 718 if (tx_head == tx_desc) 719 break; 720 721 /* clear next_to_watch to prevent false hangs */ 722 tx_buf->next_to_watch = NULL; 723 724 /* update the statistics for this packet */ 725 total_bytes += tx_buf->bytecount; 726 total_packets += tx_buf->gso_segs; 727 728 /* free the skb */ 729 dev_consume_skb_any(tx_buf->skb); 730 731 /* unmap skb header data */ 732 dma_unmap_single(tx_ring->dev, 733 dma_unmap_addr(tx_buf, dma), 734 dma_unmap_len(tx_buf, len), 735 DMA_TO_DEVICE); 736 737 /* clear tx_buffer data */ 738 tx_buf->skb = NULL; 739 dma_unmap_len_set(tx_buf, len, 0); 740 741 /* unmap remaining buffers */ 742 while (tx_desc != eop_desc) { 743 744 tx_buf++; 745 tx_desc++; 746 i++; 747 if (unlikely(!i)) { 748 i -= tx_ring->count; 749 tx_buf = tx_ring->tx_bi; 750 tx_desc = I40E_TX_DESC(tx_ring, 0); 751 } 752 753 /* unmap any remaining paged data */ 754 if (dma_unmap_len(tx_buf, len)) { 755 dma_unmap_page(tx_ring->dev, 756 dma_unmap_addr(tx_buf, dma), 757 dma_unmap_len(tx_buf, len), 758 DMA_TO_DEVICE); 759 dma_unmap_len_set(tx_buf, len, 0); 760 } 761 } 762 763 /* move us one more past the eop_desc for start of next pkt */ 764 tx_buf++; 765 tx_desc++; 766 i++; 767 if (unlikely(!i)) { 768 i -= tx_ring->count; 769 tx_buf = tx_ring->tx_bi; 770 tx_desc = I40E_TX_DESC(tx_ring, 0); 771 } 772 773 prefetch(tx_desc); 774 775 /* update budget accounting */ 776 budget--; 777 } while (likely(budget)); 778 779 i += tx_ring->count; 780 tx_ring->next_to_clean = i; 781 u64_stats_update_begin(&tx_ring->syncp); 782 tx_ring->stats.bytes += total_bytes; 783 tx_ring->stats.packets += total_packets; 784 u64_stats_update_end(&tx_ring->syncp); 785 tx_ring->q_vector->tx.total_bytes += total_bytes; 786 tx_ring->q_vector->tx.total_packets += total_packets; 787 788 /* check to see if there are any non-cache aligned descriptors 789 * waiting to be written back, and kick the hardware to force 790 * them to be written back in case of napi polling 791 */ 792 if (budget && 793 !((i & WB_STRIDE) == WB_STRIDE) && 794 !test_bit(__I40E_DOWN, &tx_ring->vsi->state) && 795 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count)) 796 tx_ring->arm_wb = true; 797 else 798 tx_ring->arm_wb = false; 799 800 if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) { 801 /* schedule immediate reset if we believe we hung */ 802 dev_info(tx_ring->dev, "Detected Tx Unit Hang\n" 803 " VSI <%d>\n" 804 " Tx Queue <%d>\n" 805 " next_to_use <%x>\n" 806 " next_to_clean <%x>\n", 807 tx_ring->vsi->seid, 808 tx_ring->queue_index, 809 tx_ring->next_to_use, i); 810 dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n" 811 " time_stamp <%lx>\n" 812 " jiffies <%lx>\n", 813 tx_ring->tx_bi[i].time_stamp, jiffies); 814 815 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); 816 817 dev_info(tx_ring->dev, 818 "tx hang detected on queue %d, reset requested\n", 819 tx_ring->queue_index); 820 821 /* do not fire the reset immediately, wait for the stack to 822 * decide we are truly stuck, also prevents every queue from 823 * simultaneously requesting a reset 824 */ 825 826 /* the adapter is about to reset, no point in enabling polling */ 827 budget = 1; 828 } 829 830 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev, 831 tx_ring->queue_index), 832 total_packets, total_bytes); 833 834 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2) 835 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) && 836 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) { 837 /* Make sure that anybody stopping the queue after this 838 * sees the new next_to_clean. 839 */ 840 smp_mb(); 841 if (__netif_subqueue_stopped(tx_ring->netdev, 842 tx_ring->queue_index) && 843 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) { 844 netif_wake_subqueue(tx_ring->netdev, 845 tx_ring->queue_index); 846 ++tx_ring->tx_stats.restart_queue; 847 } 848 } 849 850 return !!budget; 851 } 852 853 /** 854 * i40e_force_wb - Arm hardware to do a wb on noncache aligned descriptors 855 * @vsi: the VSI we care about 856 * @q_vector: the vector on which to force writeback 857 * 858 **/ 859 static void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector) 860 { 861 u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK | 862 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */ 863 I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK | 864 I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK; 865 /* allow 00 to be written to the index */ 866 867 wr32(&vsi->back->hw, 868 I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1), 869 val); 870 } 871 872 /** 873 * i40e_set_new_dynamic_itr - Find new ITR level 874 * @rc: structure containing ring performance data 875 * 876 * Stores a new ITR value based on packets and byte counts during 877 * the last interrupt. The advantage of per interrupt computation 878 * is faster updates and more accurate ITR for the current traffic 879 * pattern. Constants in this function were computed based on 880 * theoretical maximum wire speed and thresholds were set based on 881 * testing data as well as attempting to minimize response time 882 * while increasing bulk throughput. 883 **/ 884 static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc) 885 { 886 enum i40e_latency_range new_latency_range = rc->latency_range; 887 u32 new_itr = rc->itr; 888 int bytes_per_int; 889 890 if (rc->total_packets == 0 || !rc->itr) 891 return; 892 893 /* simple throttlerate management 894 * 0-10MB/s lowest (100000 ints/s) 895 * 10-20MB/s low (20000 ints/s) 896 * 20-1249MB/s bulk (8000 ints/s) 897 */ 898 bytes_per_int = rc->total_bytes / rc->itr; 899 switch (rc->itr) { 900 case I40E_LOWEST_LATENCY: 901 if (bytes_per_int > 10) 902 new_latency_range = I40E_LOW_LATENCY; 903 break; 904 case I40E_LOW_LATENCY: 905 if (bytes_per_int > 20) 906 new_latency_range = I40E_BULK_LATENCY; 907 else if (bytes_per_int <= 10) 908 new_latency_range = I40E_LOWEST_LATENCY; 909 break; 910 case I40E_BULK_LATENCY: 911 if (bytes_per_int <= 20) 912 rc->latency_range = I40E_LOW_LATENCY; 913 break; 914 } 915 916 switch (new_latency_range) { 917 case I40E_LOWEST_LATENCY: 918 new_itr = I40E_ITR_100K; 919 break; 920 case I40E_LOW_LATENCY: 921 new_itr = I40E_ITR_20K; 922 break; 923 case I40E_BULK_LATENCY: 924 new_itr = I40E_ITR_8K; 925 break; 926 default: 927 break; 928 } 929 930 if (new_itr != rc->itr) { 931 /* do an exponential smoothing */ 932 new_itr = (10 * new_itr * rc->itr) / 933 ((9 * new_itr) + rc->itr); 934 rc->itr = new_itr & I40E_MAX_ITR; 935 } 936 937 rc->total_bytes = 0; 938 rc->total_packets = 0; 939 } 940 941 /** 942 * i40e_update_dynamic_itr - Adjust ITR based on bytes per int 943 * @q_vector: the vector to adjust 944 **/ 945 static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector) 946 { 947 u16 vector = q_vector->vsi->base_vector + q_vector->v_idx; 948 struct i40e_hw *hw = &q_vector->vsi->back->hw; 949 u32 reg_addr; 950 u16 old_itr; 951 952 reg_addr = I40E_PFINT_ITRN(I40E_RX_ITR, vector - 1); 953 old_itr = q_vector->rx.itr; 954 i40e_set_new_dynamic_itr(&q_vector->rx); 955 if (old_itr != q_vector->rx.itr) 956 wr32(hw, reg_addr, q_vector->rx.itr); 957 958 reg_addr = I40E_PFINT_ITRN(I40E_TX_ITR, vector - 1); 959 old_itr = q_vector->tx.itr; 960 i40e_set_new_dynamic_itr(&q_vector->tx); 961 if (old_itr != q_vector->tx.itr) 962 wr32(hw, reg_addr, q_vector->tx.itr); 963 } 964 965 /** 966 * i40e_clean_programming_status - clean the programming status descriptor 967 * @rx_ring: the rx ring that has this descriptor 968 * @rx_desc: the rx descriptor written back by HW 969 * 970 * Flow director should handle FD_FILTER_STATUS to check its filter programming 971 * status being successful or not and take actions accordingly. FCoE should 972 * handle its context/filter programming/invalidation status and take actions. 973 * 974 **/ 975 static void i40e_clean_programming_status(struct i40e_ring *rx_ring, 976 union i40e_rx_desc *rx_desc) 977 { 978 u64 qw; 979 u8 id; 980 981 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len); 982 id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >> 983 I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT; 984 985 if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS) 986 i40e_fd_handle_status(rx_ring, rx_desc, id); 987 #ifdef I40E_FCOE 988 else if ((id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_PROG_STATUS) || 989 (id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_INVL_STATUS)) 990 i40e_fcoe_handle_status(rx_ring, rx_desc, id); 991 #endif 992 } 993 994 /** 995 * i40e_setup_tx_descriptors - Allocate the Tx descriptors 996 * @tx_ring: the tx ring to set up 997 * 998 * Return 0 on success, negative on error 999 **/ 1000 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring) 1001 { 1002 struct device *dev = tx_ring->dev; 1003 int bi_size; 1004 1005 if (!dev) 1006 return -ENOMEM; 1007 1008 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count; 1009 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL); 1010 if (!tx_ring->tx_bi) 1011 goto err; 1012 1013 /* round up to nearest 4K */ 1014 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc); 1015 /* add u32 for head writeback, align after this takes care of 1016 * guaranteeing this is at least one cache line in size 1017 */ 1018 tx_ring->size += sizeof(u32); 1019 tx_ring->size = ALIGN(tx_ring->size, 4096); 1020 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size, 1021 &tx_ring->dma, GFP_KERNEL); 1022 if (!tx_ring->desc) { 1023 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n", 1024 tx_ring->size); 1025 goto err; 1026 } 1027 1028 tx_ring->next_to_use = 0; 1029 tx_ring->next_to_clean = 0; 1030 return 0; 1031 1032 err: 1033 kfree(tx_ring->tx_bi); 1034 tx_ring->tx_bi = NULL; 1035 return -ENOMEM; 1036 } 1037 1038 /** 1039 * i40e_clean_rx_ring - Free Rx buffers 1040 * @rx_ring: ring to be cleaned 1041 **/ 1042 void i40e_clean_rx_ring(struct i40e_ring *rx_ring) 1043 { 1044 struct device *dev = rx_ring->dev; 1045 struct i40e_rx_buffer *rx_bi; 1046 unsigned long bi_size; 1047 u16 i; 1048 1049 /* ring already cleared, nothing to do */ 1050 if (!rx_ring->rx_bi) 1051 return; 1052 1053 if (ring_is_ps_enabled(rx_ring)) { 1054 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count; 1055 1056 rx_bi = &rx_ring->rx_bi[0]; 1057 if (rx_bi->hdr_buf) { 1058 dma_free_coherent(dev, 1059 bufsz, 1060 rx_bi->hdr_buf, 1061 rx_bi->dma); 1062 for (i = 0; i < rx_ring->count; i++) { 1063 rx_bi = &rx_ring->rx_bi[i]; 1064 rx_bi->dma = 0; 1065 rx_bi->hdr_buf = NULL; 1066 } 1067 } 1068 } 1069 /* Free all the Rx ring sk_buffs */ 1070 for (i = 0; i < rx_ring->count; i++) { 1071 rx_bi = &rx_ring->rx_bi[i]; 1072 if (rx_bi->dma) { 1073 dma_unmap_single(dev, 1074 rx_bi->dma, 1075 rx_ring->rx_buf_len, 1076 DMA_FROM_DEVICE); 1077 rx_bi->dma = 0; 1078 } 1079 if (rx_bi->skb) { 1080 dev_kfree_skb(rx_bi->skb); 1081 rx_bi->skb = NULL; 1082 } 1083 if (rx_bi->page) { 1084 if (rx_bi->page_dma) { 1085 dma_unmap_page(dev, 1086 rx_bi->page_dma, 1087 PAGE_SIZE / 2, 1088 DMA_FROM_DEVICE); 1089 rx_bi->page_dma = 0; 1090 } 1091 __free_page(rx_bi->page); 1092 rx_bi->page = NULL; 1093 rx_bi->page_offset = 0; 1094 } 1095 } 1096 1097 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; 1098 memset(rx_ring->rx_bi, 0, bi_size); 1099 1100 /* Zero out the descriptor ring */ 1101 memset(rx_ring->desc, 0, rx_ring->size); 1102 1103 rx_ring->next_to_clean = 0; 1104 rx_ring->next_to_use = 0; 1105 } 1106 1107 /** 1108 * i40e_free_rx_resources - Free Rx resources 1109 * @rx_ring: ring to clean the resources from 1110 * 1111 * Free all receive software resources 1112 **/ 1113 void i40e_free_rx_resources(struct i40e_ring *rx_ring) 1114 { 1115 i40e_clean_rx_ring(rx_ring); 1116 kfree(rx_ring->rx_bi); 1117 rx_ring->rx_bi = NULL; 1118 1119 if (rx_ring->desc) { 1120 dma_free_coherent(rx_ring->dev, rx_ring->size, 1121 rx_ring->desc, rx_ring->dma); 1122 rx_ring->desc = NULL; 1123 } 1124 } 1125 1126 /** 1127 * i40e_alloc_rx_headers - allocate rx header buffers 1128 * @rx_ring: ring to alloc buffers 1129 * 1130 * Allocate rx header buffers for the entire ring. As these are static, 1131 * this is only called when setting up a new ring. 1132 **/ 1133 void i40e_alloc_rx_headers(struct i40e_ring *rx_ring) 1134 { 1135 struct device *dev = rx_ring->dev; 1136 struct i40e_rx_buffer *rx_bi; 1137 dma_addr_t dma; 1138 void *buffer; 1139 int buf_size; 1140 int i; 1141 1142 if (rx_ring->rx_bi[0].hdr_buf) 1143 return; 1144 /* Make sure the buffers don't cross cache line boundaries. */ 1145 buf_size = ALIGN(rx_ring->rx_hdr_len, 256); 1146 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count, 1147 &dma, GFP_KERNEL); 1148 if (!buffer) 1149 return; 1150 for (i = 0; i < rx_ring->count; i++) { 1151 rx_bi = &rx_ring->rx_bi[i]; 1152 rx_bi->dma = dma + (i * buf_size); 1153 rx_bi->hdr_buf = buffer + (i * buf_size); 1154 } 1155 } 1156 1157 /** 1158 * i40e_setup_rx_descriptors - Allocate Rx descriptors 1159 * @rx_ring: Rx descriptor ring (for a specific queue) to setup 1160 * 1161 * Returns 0 on success, negative on failure 1162 **/ 1163 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring) 1164 { 1165 struct device *dev = rx_ring->dev; 1166 int bi_size; 1167 1168 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count; 1169 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL); 1170 if (!rx_ring->rx_bi) 1171 goto err; 1172 1173 u64_stats_init(&rx_ring->syncp); 1174 1175 /* Round up to nearest 4K */ 1176 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring) 1177 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc) 1178 : rx_ring->count * sizeof(union i40e_32byte_rx_desc); 1179 rx_ring->size = ALIGN(rx_ring->size, 4096); 1180 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size, 1181 &rx_ring->dma, GFP_KERNEL); 1182 1183 if (!rx_ring->desc) { 1184 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n", 1185 rx_ring->size); 1186 goto err; 1187 } 1188 1189 rx_ring->next_to_clean = 0; 1190 rx_ring->next_to_use = 0; 1191 1192 return 0; 1193 err: 1194 kfree(rx_ring->rx_bi); 1195 rx_ring->rx_bi = NULL; 1196 return -ENOMEM; 1197 } 1198 1199 /** 1200 * i40e_release_rx_desc - Store the new tail and head values 1201 * @rx_ring: ring to bump 1202 * @val: new head index 1203 **/ 1204 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val) 1205 { 1206 rx_ring->next_to_use = val; 1207 /* Force memory writes to complete before letting h/w 1208 * know there are new descriptors to fetch. (Only 1209 * applicable for weak-ordered memory model archs, 1210 * such as IA-64). 1211 */ 1212 wmb(); 1213 writel(val, rx_ring->tail); 1214 } 1215 1216 /** 1217 * i40e_alloc_rx_buffers_ps - Replace used receive buffers; packet split 1218 * @rx_ring: ring to place buffers on 1219 * @cleaned_count: number of buffers to replace 1220 **/ 1221 void i40e_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count) 1222 { 1223 u16 i = rx_ring->next_to_use; 1224 union i40e_rx_desc *rx_desc; 1225 struct i40e_rx_buffer *bi; 1226 1227 /* do nothing if no valid netdev defined */ 1228 if (!rx_ring->netdev || !cleaned_count) 1229 return; 1230 1231 while (cleaned_count--) { 1232 rx_desc = I40E_RX_DESC(rx_ring, i); 1233 bi = &rx_ring->rx_bi[i]; 1234 1235 if (bi->skb) /* desc is in use */ 1236 goto no_buffers; 1237 if (!bi->page) { 1238 bi->page = alloc_page(GFP_ATOMIC); 1239 if (!bi->page) { 1240 rx_ring->rx_stats.alloc_page_failed++; 1241 goto no_buffers; 1242 } 1243 } 1244 1245 if (!bi->page_dma) { 1246 /* use a half page if we're re-using */ 1247 bi->page_offset ^= PAGE_SIZE / 2; 1248 bi->page_dma = dma_map_page(rx_ring->dev, 1249 bi->page, 1250 bi->page_offset, 1251 PAGE_SIZE / 2, 1252 DMA_FROM_DEVICE); 1253 if (dma_mapping_error(rx_ring->dev, 1254 bi->page_dma)) { 1255 rx_ring->rx_stats.alloc_page_failed++; 1256 bi->page_dma = 0; 1257 goto no_buffers; 1258 } 1259 } 1260 1261 dma_sync_single_range_for_device(rx_ring->dev, 1262 bi->dma, 1263 0, 1264 rx_ring->rx_hdr_len, 1265 DMA_FROM_DEVICE); 1266 /* Refresh the desc even if buffer_addrs didn't change 1267 * because each write-back erases this info. 1268 */ 1269 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma); 1270 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma); 1271 i++; 1272 if (i == rx_ring->count) 1273 i = 0; 1274 } 1275 1276 no_buffers: 1277 if (rx_ring->next_to_use != i) 1278 i40e_release_rx_desc(rx_ring, i); 1279 } 1280 1281 /** 1282 * i40e_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer 1283 * @rx_ring: ring to place buffers on 1284 * @cleaned_count: number of buffers to replace 1285 **/ 1286 void i40e_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count) 1287 { 1288 u16 i = rx_ring->next_to_use; 1289 union i40e_rx_desc *rx_desc; 1290 struct i40e_rx_buffer *bi; 1291 struct sk_buff *skb; 1292 1293 /* do nothing if no valid netdev defined */ 1294 if (!rx_ring->netdev || !cleaned_count) 1295 return; 1296 1297 while (cleaned_count--) { 1298 rx_desc = I40E_RX_DESC(rx_ring, i); 1299 bi = &rx_ring->rx_bi[i]; 1300 skb = bi->skb; 1301 1302 if (!skb) { 1303 skb = netdev_alloc_skb_ip_align(rx_ring->netdev, 1304 rx_ring->rx_buf_len); 1305 if (!skb) { 1306 rx_ring->rx_stats.alloc_buff_failed++; 1307 goto no_buffers; 1308 } 1309 /* initialize queue mapping */ 1310 skb_record_rx_queue(skb, rx_ring->queue_index); 1311 bi->skb = skb; 1312 } 1313 1314 if (!bi->dma) { 1315 bi->dma = dma_map_single(rx_ring->dev, 1316 skb->data, 1317 rx_ring->rx_buf_len, 1318 DMA_FROM_DEVICE); 1319 if (dma_mapping_error(rx_ring->dev, bi->dma)) { 1320 rx_ring->rx_stats.alloc_buff_failed++; 1321 bi->dma = 0; 1322 goto no_buffers; 1323 } 1324 } 1325 1326 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma); 1327 rx_desc->read.hdr_addr = 0; 1328 i++; 1329 if (i == rx_ring->count) 1330 i = 0; 1331 } 1332 1333 no_buffers: 1334 if (rx_ring->next_to_use != i) 1335 i40e_release_rx_desc(rx_ring, i); 1336 } 1337 1338 /** 1339 * i40e_receive_skb - Send a completed packet up the stack 1340 * @rx_ring: rx ring in play 1341 * @skb: packet to send up 1342 * @vlan_tag: vlan tag for packet 1343 **/ 1344 static void i40e_receive_skb(struct i40e_ring *rx_ring, 1345 struct sk_buff *skb, u16 vlan_tag) 1346 { 1347 struct i40e_q_vector *q_vector = rx_ring->q_vector; 1348 struct i40e_vsi *vsi = rx_ring->vsi; 1349 u64 flags = vsi->back->flags; 1350 1351 if (vlan_tag & VLAN_VID_MASK) 1352 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag); 1353 1354 if (flags & I40E_FLAG_IN_NETPOLL) 1355 netif_rx(skb); 1356 else 1357 napi_gro_receive(&q_vector->napi, skb); 1358 } 1359 1360 /** 1361 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum 1362 * @vsi: the VSI we care about 1363 * @skb: skb currently being received and modified 1364 * @rx_status: status value of last descriptor in packet 1365 * @rx_error: error value of last descriptor in packet 1366 * @rx_ptype: ptype value of last descriptor in packet 1367 **/ 1368 static inline void i40e_rx_checksum(struct i40e_vsi *vsi, 1369 struct sk_buff *skb, 1370 u32 rx_status, 1371 u32 rx_error, 1372 u16 rx_ptype) 1373 { 1374 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype); 1375 bool ipv4 = false, ipv6 = false; 1376 bool ipv4_tunnel, ipv6_tunnel; 1377 __wsum rx_udp_csum; 1378 struct iphdr *iph; 1379 __sum16 csum; 1380 1381 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) && 1382 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4); 1383 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) && 1384 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4); 1385 1386 skb->ip_summed = CHECKSUM_NONE; 1387 1388 /* Rx csum enabled and ip headers found? */ 1389 if (!(vsi->netdev->features & NETIF_F_RXCSUM)) 1390 return; 1391 1392 /* did the hardware decode the packet and checksum? */ 1393 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT))) 1394 return; 1395 1396 /* both known and outer_ip must be set for the below code to work */ 1397 if (!(decoded.known && decoded.outer_ip)) 1398 return; 1399 1400 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && 1401 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4) 1402 ipv4 = true; 1403 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && 1404 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6) 1405 ipv6 = true; 1406 1407 if (ipv4 && 1408 (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) | 1409 (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT)))) 1410 goto checksum_fail; 1411 1412 /* likely incorrect csum if alternate IP extension headers found */ 1413 if (ipv6 && 1414 rx_status & (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT)) 1415 /* don't increment checksum err here, non-fatal err */ 1416 return; 1417 1418 /* there was some L4 error, count error and punt packet to the stack */ 1419 if (rx_error & (1 << I40E_RX_DESC_ERROR_L4E_SHIFT)) 1420 goto checksum_fail; 1421 1422 /* handle packets that were not able to be checksummed due 1423 * to arrival speed, in this case the stack can compute 1424 * the csum. 1425 */ 1426 if (rx_error & (1 << I40E_RX_DESC_ERROR_PPRS_SHIFT)) 1427 return; 1428 1429 /* If VXLAN traffic has an outer UDPv4 checksum we need to check 1430 * it in the driver, hardware does not do it for us. 1431 * Since L3L4P bit was set we assume a valid IHL value (>=5) 1432 * so the total length of IPv4 header is IHL*4 bytes 1433 * The UDP_0 bit *may* bet set if the *inner* header is UDP 1434 */ 1435 if (ipv4_tunnel) { 1436 skb->transport_header = skb->mac_header + 1437 sizeof(struct ethhdr) + 1438 (ip_hdr(skb)->ihl * 4); 1439 1440 /* Add 4 bytes for VLAN tagged packets */ 1441 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) || 1442 skb->protocol == htons(ETH_P_8021AD)) 1443 ? VLAN_HLEN : 0; 1444 1445 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) && 1446 (udp_hdr(skb)->check != 0)) { 1447 rx_udp_csum = udp_csum(skb); 1448 iph = ip_hdr(skb); 1449 csum = csum_tcpudp_magic( 1450 iph->saddr, iph->daddr, 1451 (skb->len - skb_transport_offset(skb)), 1452 IPPROTO_UDP, rx_udp_csum); 1453 1454 if (udp_hdr(skb)->check != csum) 1455 goto checksum_fail; 1456 1457 } /* else its GRE and so no outer UDP header */ 1458 } 1459 1460 skb->ip_summed = CHECKSUM_UNNECESSARY; 1461 skb->csum_level = ipv4_tunnel || ipv6_tunnel; 1462 1463 return; 1464 1465 checksum_fail: 1466 vsi->back->hw_csum_rx_error++; 1467 } 1468 1469 /** 1470 * i40e_rx_hash - returns the hash value from the Rx descriptor 1471 * @ring: descriptor ring 1472 * @rx_desc: specific descriptor 1473 **/ 1474 static inline u32 i40e_rx_hash(struct i40e_ring *ring, 1475 union i40e_rx_desc *rx_desc) 1476 { 1477 const __le64 rss_mask = 1478 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH << 1479 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT); 1480 1481 if ((ring->netdev->features & NETIF_F_RXHASH) && 1482 (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) 1483 return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss); 1484 else 1485 return 0; 1486 } 1487 1488 /** 1489 * i40e_ptype_to_hash - get a hash type 1490 * @ptype: the ptype value from the descriptor 1491 * 1492 * Returns a hash type to be used by skb_set_hash 1493 **/ 1494 static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype) 1495 { 1496 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype); 1497 1498 if (!decoded.known) 1499 return PKT_HASH_TYPE_NONE; 1500 1501 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && 1502 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4) 1503 return PKT_HASH_TYPE_L4; 1504 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP && 1505 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3) 1506 return PKT_HASH_TYPE_L3; 1507 else 1508 return PKT_HASH_TYPE_L2; 1509 } 1510 1511 /** 1512 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split 1513 * @rx_ring: rx ring to clean 1514 * @budget: how many cleans we're allowed 1515 * 1516 * Returns true if there's any budget left (e.g. the clean is finished) 1517 **/ 1518 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget) 1519 { 1520 unsigned int total_rx_bytes = 0, total_rx_packets = 0; 1521 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo; 1522 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); 1523 const int current_node = numa_node_id(); 1524 struct i40e_vsi *vsi = rx_ring->vsi; 1525 u16 i = rx_ring->next_to_clean; 1526 union i40e_rx_desc *rx_desc; 1527 u32 rx_error, rx_status; 1528 u8 rx_ptype; 1529 u64 qword; 1530 1531 if (budget <= 0) 1532 return 0; 1533 1534 do { 1535 struct i40e_rx_buffer *rx_bi; 1536 struct sk_buff *skb; 1537 u16 vlan_tag; 1538 /* return some buffers to hardware, one at a time is too slow */ 1539 if (cleaned_count >= I40E_RX_BUFFER_WRITE) { 1540 i40e_alloc_rx_buffers_ps(rx_ring, cleaned_count); 1541 cleaned_count = 0; 1542 } 1543 1544 i = rx_ring->next_to_clean; 1545 rx_desc = I40E_RX_DESC(rx_ring, i); 1546 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); 1547 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> 1548 I40E_RXD_QW1_STATUS_SHIFT; 1549 1550 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT))) 1551 break; 1552 1553 /* This memory barrier is needed to keep us from reading 1554 * any other fields out of the rx_desc until we know the 1555 * DD bit is set. 1556 */ 1557 rmb(); 1558 if (i40e_rx_is_programming_status(qword)) { 1559 i40e_clean_programming_status(rx_ring, rx_desc); 1560 I40E_RX_INCREMENT(rx_ring, i); 1561 continue; 1562 } 1563 rx_bi = &rx_ring->rx_bi[i]; 1564 skb = rx_bi->skb; 1565 if (likely(!skb)) { 1566 skb = netdev_alloc_skb_ip_align(rx_ring->netdev, 1567 rx_ring->rx_hdr_len); 1568 if (!skb) 1569 rx_ring->rx_stats.alloc_buff_failed++; 1570 /* initialize queue mapping */ 1571 skb_record_rx_queue(skb, rx_ring->queue_index); 1572 /* we are reusing so sync this buffer for CPU use */ 1573 dma_sync_single_range_for_cpu(rx_ring->dev, 1574 rx_bi->dma, 1575 0, 1576 rx_ring->rx_hdr_len, 1577 DMA_FROM_DEVICE); 1578 } 1579 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >> 1580 I40E_RXD_QW1_LENGTH_PBUF_SHIFT; 1581 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >> 1582 I40E_RXD_QW1_LENGTH_HBUF_SHIFT; 1583 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >> 1584 I40E_RXD_QW1_LENGTH_SPH_SHIFT; 1585 1586 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >> 1587 I40E_RXD_QW1_ERROR_SHIFT; 1588 rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT); 1589 rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT); 1590 1591 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> 1592 I40E_RXD_QW1_PTYPE_SHIFT; 1593 prefetch(rx_bi->page); 1594 rx_bi->skb = NULL; 1595 cleaned_count++; 1596 if (rx_hbo || rx_sph) { 1597 int len; 1598 if (rx_hbo) 1599 len = I40E_RX_HDR_SIZE; 1600 else 1601 len = rx_header_len; 1602 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len); 1603 } else if (skb->len == 0) { 1604 int len; 1605 1606 len = (rx_packet_len > skb_headlen(skb) ? 1607 skb_headlen(skb) : rx_packet_len); 1608 memcpy(__skb_put(skb, len), 1609 rx_bi->page + rx_bi->page_offset, 1610 len); 1611 rx_bi->page_offset += len; 1612 rx_packet_len -= len; 1613 } 1614 1615 /* Get the rest of the data if this was a header split */ 1616 if (rx_packet_len) { 1617 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags, 1618 rx_bi->page, 1619 rx_bi->page_offset, 1620 rx_packet_len); 1621 1622 skb->len += rx_packet_len; 1623 skb->data_len += rx_packet_len; 1624 skb->truesize += rx_packet_len; 1625 1626 if ((page_count(rx_bi->page) == 1) && 1627 (page_to_nid(rx_bi->page) == current_node)) 1628 get_page(rx_bi->page); 1629 else 1630 rx_bi->page = NULL; 1631 1632 dma_unmap_page(rx_ring->dev, 1633 rx_bi->page_dma, 1634 PAGE_SIZE / 2, 1635 DMA_FROM_DEVICE); 1636 rx_bi->page_dma = 0; 1637 } 1638 I40E_RX_INCREMENT(rx_ring, i); 1639 1640 if (unlikely( 1641 !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) { 1642 struct i40e_rx_buffer *next_buffer; 1643 1644 next_buffer = &rx_ring->rx_bi[i]; 1645 next_buffer->skb = skb; 1646 rx_ring->rx_stats.non_eop_descs++; 1647 continue; 1648 } 1649 1650 /* ERR_MASK will only have valid bits if EOP set */ 1651 if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) { 1652 dev_kfree_skb_any(skb); 1653 /* TODO: shouldn't we increment a counter indicating the 1654 * drop? 1655 */ 1656 continue; 1657 } 1658 1659 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc), 1660 i40e_ptype_to_hash(rx_ptype)); 1661 if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) { 1662 i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status & 1663 I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >> 1664 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT); 1665 rx_ring->last_rx_timestamp = jiffies; 1666 } 1667 1668 /* probably a little skewed due to removing CRC */ 1669 total_rx_bytes += skb->len; 1670 total_rx_packets++; 1671 1672 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 1673 1674 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype); 1675 1676 vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) 1677 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) 1678 : 0; 1679 #ifdef I40E_FCOE 1680 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) { 1681 dev_kfree_skb_any(skb); 1682 continue; 1683 } 1684 #endif 1685 skb_mark_napi_id(skb, &rx_ring->q_vector->napi); 1686 i40e_receive_skb(rx_ring, skb, vlan_tag); 1687 1688 rx_ring->netdev->last_rx = jiffies; 1689 rx_desc->wb.qword1.status_error_len = 0; 1690 1691 } while (likely(total_rx_packets < budget)); 1692 1693 u64_stats_update_begin(&rx_ring->syncp); 1694 rx_ring->stats.packets += total_rx_packets; 1695 rx_ring->stats.bytes += total_rx_bytes; 1696 u64_stats_update_end(&rx_ring->syncp); 1697 rx_ring->q_vector->rx.total_packets += total_rx_packets; 1698 rx_ring->q_vector->rx.total_bytes += total_rx_bytes; 1699 1700 return total_rx_packets; 1701 } 1702 1703 /** 1704 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer 1705 * @rx_ring: rx ring to clean 1706 * @budget: how many cleans we're allowed 1707 * 1708 * Returns number of packets cleaned 1709 **/ 1710 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget) 1711 { 1712 unsigned int total_rx_bytes = 0, total_rx_packets = 0; 1713 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring); 1714 struct i40e_vsi *vsi = rx_ring->vsi; 1715 union i40e_rx_desc *rx_desc; 1716 u32 rx_error, rx_status; 1717 u16 rx_packet_len; 1718 u8 rx_ptype; 1719 u64 qword; 1720 u16 i; 1721 1722 do { 1723 struct i40e_rx_buffer *rx_bi; 1724 struct sk_buff *skb; 1725 u16 vlan_tag; 1726 /* return some buffers to hardware, one at a time is too slow */ 1727 if (cleaned_count >= I40E_RX_BUFFER_WRITE) { 1728 i40e_alloc_rx_buffers_1buf(rx_ring, cleaned_count); 1729 cleaned_count = 0; 1730 } 1731 1732 i = rx_ring->next_to_clean; 1733 rx_desc = I40E_RX_DESC(rx_ring, i); 1734 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len); 1735 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >> 1736 I40E_RXD_QW1_STATUS_SHIFT; 1737 1738 if (!(rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT))) 1739 break; 1740 1741 /* This memory barrier is needed to keep us from reading 1742 * any other fields out of the rx_desc until we know the 1743 * DD bit is set. 1744 */ 1745 rmb(); 1746 1747 if (i40e_rx_is_programming_status(qword)) { 1748 i40e_clean_programming_status(rx_ring, rx_desc); 1749 I40E_RX_INCREMENT(rx_ring, i); 1750 continue; 1751 } 1752 rx_bi = &rx_ring->rx_bi[i]; 1753 skb = rx_bi->skb; 1754 prefetch(skb->data); 1755 1756 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >> 1757 I40E_RXD_QW1_LENGTH_PBUF_SHIFT; 1758 1759 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >> 1760 I40E_RXD_QW1_ERROR_SHIFT; 1761 rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT); 1762 1763 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> 1764 I40E_RXD_QW1_PTYPE_SHIFT; 1765 rx_bi->skb = NULL; 1766 cleaned_count++; 1767 1768 /* Get the header and possibly the whole packet 1769 * If this is an skb from previous receive dma will be 0 1770 */ 1771 skb_put(skb, rx_packet_len); 1772 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len, 1773 DMA_FROM_DEVICE); 1774 rx_bi->dma = 0; 1775 1776 I40E_RX_INCREMENT(rx_ring, i); 1777 1778 if (unlikely( 1779 !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) { 1780 rx_ring->rx_stats.non_eop_descs++; 1781 continue; 1782 } 1783 1784 /* ERR_MASK will only have valid bits if EOP set */ 1785 if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) { 1786 dev_kfree_skb_any(skb); 1787 /* TODO: shouldn't we increment a counter indicating the 1788 * drop? 1789 */ 1790 continue; 1791 } 1792 1793 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc), 1794 i40e_ptype_to_hash(rx_ptype)); 1795 if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) { 1796 i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status & 1797 I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >> 1798 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT); 1799 rx_ring->last_rx_timestamp = jiffies; 1800 } 1801 1802 /* probably a little skewed due to removing CRC */ 1803 total_rx_bytes += skb->len; 1804 total_rx_packets++; 1805 1806 skb->protocol = eth_type_trans(skb, rx_ring->netdev); 1807 1808 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype); 1809 1810 vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT) 1811 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) 1812 : 0; 1813 #ifdef I40E_FCOE 1814 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) { 1815 dev_kfree_skb_any(skb); 1816 continue; 1817 } 1818 #endif 1819 i40e_receive_skb(rx_ring, skb, vlan_tag); 1820 1821 rx_ring->netdev->last_rx = jiffies; 1822 rx_desc->wb.qword1.status_error_len = 0; 1823 } while (likely(total_rx_packets < budget)); 1824 1825 u64_stats_update_begin(&rx_ring->syncp); 1826 rx_ring->stats.packets += total_rx_packets; 1827 rx_ring->stats.bytes += total_rx_bytes; 1828 u64_stats_update_end(&rx_ring->syncp); 1829 rx_ring->q_vector->rx.total_packets += total_rx_packets; 1830 rx_ring->q_vector->rx.total_bytes += total_rx_bytes; 1831 1832 return total_rx_packets; 1833 } 1834 1835 /** 1836 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine 1837 * @napi: napi struct with our devices info in it 1838 * @budget: amount of work driver is allowed to do this pass, in packets 1839 * 1840 * This function will clean all queues associated with a q_vector. 1841 * 1842 * Returns the amount of work done 1843 **/ 1844 int i40e_napi_poll(struct napi_struct *napi, int budget) 1845 { 1846 struct i40e_q_vector *q_vector = 1847 container_of(napi, struct i40e_q_vector, napi); 1848 struct i40e_vsi *vsi = q_vector->vsi; 1849 struct i40e_ring *ring; 1850 bool clean_complete = true; 1851 bool arm_wb = false; 1852 int budget_per_ring; 1853 int cleaned; 1854 1855 if (test_bit(__I40E_DOWN, &vsi->state)) { 1856 napi_complete(napi); 1857 return 0; 1858 } 1859 1860 /* Since the actual Tx work is minimal, we can give the Tx a larger 1861 * budget and be more aggressive about cleaning up the Tx descriptors. 1862 */ 1863 i40e_for_each_ring(ring, q_vector->tx) { 1864 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit); 1865 arm_wb |= ring->arm_wb; 1866 } 1867 1868 /* We attempt to distribute budget to each Rx queue fairly, but don't 1869 * allow the budget to go below 1 because that would exit polling early. 1870 */ 1871 budget_per_ring = max(budget/q_vector->num_ringpairs, 1); 1872 1873 i40e_for_each_ring(ring, q_vector->rx) { 1874 if (ring_is_ps_enabled(ring)) 1875 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring); 1876 else 1877 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring); 1878 /* if we didn't clean as many as budgeted, we must be done */ 1879 clean_complete &= (budget_per_ring != cleaned); 1880 } 1881 1882 /* If work not completed, return budget and polling will return */ 1883 if (!clean_complete) { 1884 if (arm_wb) 1885 i40e_force_wb(vsi, q_vector); 1886 return budget; 1887 } 1888 1889 /* Work is done so exit the polling mode and re-enable the interrupt */ 1890 napi_complete(napi); 1891 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) || 1892 ITR_IS_DYNAMIC(vsi->tx_itr_setting)) 1893 i40e_update_dynamic_itr(q_vector); 1894 1895 if (!test_bit(__I40E_DOWN, &vsi->state)) { 1896 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) { 1897 i40e_irq_dynamic_enable(vsi, 1898 q_vector->v_idx + vsi->base_vector); 1899 } else { 1900 struct i40e_hw *hw = &vsi->back->hw; 1901 /* We re-enable the queue 0 cause, but 1902 * don't worry about dynamic_enable 1903 * because we left it on for the other 1904 * possible interrupts during napi 1905 */ 1906 u32 qval = rd32(hw, I40E_QINT_RQCTL(0)); 1907 qval |= I40E_QINT_RQCTL_CAUSE_ENA_MASK; 1908 wr32(hw, I40E_QINT_RQCTL(0), qval); 1909 1910 qval = rd32(hw, I40E_QINT_TQCTL(0)); 1911 qval |= I40E_QINT_TQCTL_CAUSE_ENA_MASK; 1912 wr32(hw, I40E_QINT_TQCTL(0), qval); 1913 1914 i40e_irq_dynamic_enable_icr0(vsi->back); 1915 } 1916 } 1917 1918 return 0; 1919 } 1920 1921 /** 1922 * i40e_atr - Add a Flow Director ATR filter 1923 * @tx_ring: ring to add programming descriptor to 1924 * @skb: send buffer 1925 * @flags: send flags 1926 * @protocol: wire protocol 1927 **/ 1928 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb, 1929 u32 flags, __be16 protocol) 1930 { 1931 struct i40e_filter_program_desc *fdir_desc; 1932 struct i40e_pf *pf = tx_ring->vsi->back; 1933 union { 1934 unsigned char *network; 1935 struct iphdr *ipv4; 1936 struct ipv6hdr *ipv6; 1937 } hdr; 1938 struct tcphdr *th; 1939 unsigned int hlen; 1940 u32 flex_ptype, dtype_cmd; 1941 u16 i; 1942 1943 /* make sure ATR is enabled */ 1944 if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED)) 1945 return; 1946 1947 if ((pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED)) 1948 return; 1949 1950 /* if sampling is disabled do nothing */ 1951 if (!tx_ring->atr_sample_rate) 1952 return; 1953 1954 /* snag network header to get L4 type and address */ 1955 hdr.network = skb_network_header(skb); 1956 1957 /* Currently only IPv4/IPv6 with TCP is supported */ 1958 if (protocol == htons(ETH_P_IP)) { 1959 if (hdr.ipv4->protocol != IPPROTO_TCP) 1960 return; 1961 1962 /* access ihl as a u8 to avoid unaligned access on ia64 */ 1963 hlen = (hdr.network[0] & 0x0F) << 2; 1964 } else if (protocol == htons(ETH_P_IPV6)) { 1965 if (hdr.ipv6->nexthdr != IPPROTO_TCP) 1966 return; 1967 1968 hlen = sizeof(struct ipv6hdr); 1969 } else { 1970 return; 1971 } 1972 1973 th = (struct tcphdr *)(hdr.network + hlen); 1974 1975 /* Due to lack of space, no more new filters can be programmed */ 1976 if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED)) 1977 return; 1978 1979 tx_ring->atr_count++; 1980 1981 /* sample on all syn/fin/rst packets or once every atr sample rate */ 1982 if (!th->fin && 1983 !th->syn && 1984 !th->rst && 1985 (tx_ring->atr_count < tx_ring->atr_sample_rate)) 1986 return; 1987 1988 tx_ring->atr_count = 0; 1989 1990 /* grab the next descriptor */ 1991 i = tx_ring->next_to_use; 1992 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i); 1993 1994 i++; 1995 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 1996 1997 flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) & 1998 I40E_TXD_FLTR_QW0_QINDEX_MASK; 1999 flex_ptype |= (protocol == htons(ETH_P_IP)) ? 2000 (I40E_FILTER_PCTYPE_NONF_IPV4_TCP << 2001 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) : 2002 (I40E_FILTER_PCTYPE_NONF_IPV6_TCP << 2003 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT); 2004 2005 flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT; 2006 2007 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG; 2008 2009 dtype_cmd |= (th->fin || th->rst) ? 2010 (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE << 2011 I40E_TXD_FLTR_QW1_PCMD_SHIFT) : 2012 (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE << 2013 I40E_TXD_FLTR_QW1_PCMD_SHIFT); 2014 2015 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX << 2016 I40E_TXD_FLTR_QW1_DEST_SHIFT; 2017 2018 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID << 2019 I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT; 2020 2021 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK; 2022 dtype_cmd |= 2023 ((u32)pf->fd_atr_cnt_idx << I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) & 2024 I40E_TXD_FLTR_QW1_CNTINDEX_MASK; 2025 2026 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype); 2027 fdir_desc->rsvd = cpu_to_le32(0); 2028 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd); 2029 fdir_desc->fd_id = cpu_to_le32(0); 2030 } 2031 2032 /** 2033 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW 2034 * @skb: send buffer 2035 * @tx_ring: ring to send buffer on 2036 * @flags: the tx flags to be set 2037 * 2038 * Checks the skb and set up correspondingly several generic transmit flags 2039 * related to VLAN tagging for the HW, such as VLAN, DCB, etc. 2040 * 2041 * Returns error code indicate the frame should be dropped upon error and the 2042 * otherwise returns 0 to indicate the flags has been set properly. 2043 **/ 2044 #ifdef I40E_FCOE 2045 int i40e_tx_prepare_vlan_flags(struct sk_buff *skb, 2046 struct i40e_ring *tx_ring, 2047 u32 *flags) 2048 #else 2049 static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb, 2050 struct i40e_ring *tx_ring, 2051 u32 *flags) 2052 #endif 2053 { 2054 __be16 protocol = skb->protocol; 2055 u32 tx_flags = 0; 2056 2057 /* if we have a HW VLAN tag being added, default to the HW one */ 2058 if (skb_vlan_tag_present(skb)) { 2059 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT; 2060 tx_flags |= I40E_TX_FLAGS_HW_VLAN; 2061 /* else if it is a SW VLAN, check the next protocol and store the tag */ 2062 } else if (protocol == htons(ETH_P_8021Q)) { 2063 struct vlan_hdr *vhdr, _vhdr; 2064 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr); 2065 if (!vhdr) 2066 return -EINVAL; 2067 2068 protocol = vhdr->h_vlan_encapsulated_proto; 2069 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT; 2070 tx_flags |= I40E_TX_FLAGS_SW_VLAN; 2071 } 2072 2073 if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED)) 2074 goto out; 2075 2076 /* Insert 802.1p priority into VLAN header */ 2077 if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) || 2078 (skb->priority != TC_PRIO_CONTROL)) { 2079 tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK; 2080 tx_flags |= (skb->priority & 0x7) << 2081 I40E_TX_FLAGS_VLAN_PRIO_SHIFT; 2082 if (tx_flags & I40E_TX_FLAGS_SW_VLAN) { 2083 struct vlan_ethhdr *vhdr; 2084 int rc; 2085 2086 rc = skb_cow_head(skb, 0); 2087 if (rc < 0) 2088 return rc; 2089 vhdr = (struct vlan_ethhdr *)skb->data; 2090 vhdr->h_vlan_TCI = htons(tx_flags >> 2091 I40E_TX_FLAGS_VLAN_SHIFT); 2092 } else { 2093 tx_flags |= I40E_TX_FLAGS_HW_VLAN; 2094 } 2095 } 2096 2097 out: 2098 *flags = tx_flags; 2099 return 0; 2100 } 2101 2102 /** 2103 * i40e_tso - set up the tso context descriptor 2104 * @tx_ring: ptr to the ring to send 2105 * @skb: ptr to the skb we're sending 2106 * @tx_flags: the collected send information 2107 * @protocol: the send protocol 2108 * @hdr_len: ptr to the size of the packet header 2109 * @cd_tunneling: ptr to context descriptor bits 2110 * 2111 * Returns 0 if no TSO can happen, 1 if tso is going, or error 2112 **/ 2113 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb, 2114 u32 tx_flags, __be16 protocol, u8 *hdr_len, 2115 u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling) 2116 { 2117 u32 cd_cmd, cd_tso_len, cd_mss; 2118 struct ipv6hdr *ipv6h; 2119 struct tcphdr *tcph; 2120 struct iphdr *iph; 2121 u32 l4len; 2122 int err; 2123 2124 if (!skb_is_gso(skb)) 2125 return 0; 2126 2127 err = skb_cow_head(skb, 0); 2128 if (err < 0) 2129 return err; 2130 2131 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb); 2132 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb); 2133 2134 if (iph->version == 4) { 2135 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); 2136 iph->tot_len = 0; 2137 iph->check = 0; 2138 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 2139 0, IPPROTO_TCP, 0); 2140 } else if (ipv6h->version == 6) { 2141 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb); 2142 ipv6h->payload_len = 0; 2143 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, 2144 0, IPPROTO_TCP, 0); 2145 } 2146 2147 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb); 2148 *hdr_len = (skb->encapsulation 2149 ? (skb_inner_transport_header(skb) - skb->data) 2150 : skb_transport_offset(skb)) + l4len; 2151 2152 /* find the field values */ 2153 cd_cmd = I40E_TX_CTX_DESC_TSO; 2154 cd_tso_len = skb->len - *hdr_len; 2155 cd_mss = skb_shinfo(skb)->gso_size; 2156 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) | 2157 ((u64)cd_tso_len << 2158 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) | 2159 ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT); 2160 return 1; 2161 } 2162 2163 /** 2164 * i40e_tsyn - set up the tsyn context descriptor 2165 * @tx_ring: ptr to the ring to send 2166 * @skb: ptr to the skb we're sending 2167 * @tx_flags: the collected send information 2168 * 2169 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen 2170 **/ 2171 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb, 2172 u32 tx_flags, u64 *cd_type_cmd_tso_mss) 2173 { 2174 struct i40e_pf *pf; 2175 2176 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))) 2177 return 0; 2178 2179 /* Tx timestamps cannot be sampled when doing TSO */ 2180 if (tx_flags & I40E_TX_FLAGS_TSO) 2181 return 0; 2182 2183 /* only timestamp the outbound packet if the user has requested it and 2184 * we are not already transmitting a packet to be timestamped 2185 */ 2186 pf = i40e_netdev_to_pf(tx_ring->netdev); 2187 if (!(pf->flags & I40E_FLAG_PTP)) 2188 return 0; 2189 2190 if (pf->ptp_tx && 2191 !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) { 2192 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS; 2193 pf->ptp_tx_skb = skb_get(skb); 2194 } else { 2195 return 0; 2196 } 2197 2198 *cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN << 2199 I40E_TXD_CTX_QW1_CMD_SHIFT; 2200 2201 return 1; 2202 } 2203 2204 /** 2205 * i40e_tx_enable_csum - Enable Tx checksum offloads 2206 * @skb: send buffer 2207 * @tx_flags: Tx flags currently set 2208 * @td_cmd: Tx descriptor command bits to set 2209 * @td_offset: Tx descriptor header offsets to set 2210 * @cd_tunneling: ptr to context desc bits 2211 **/ 2212 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags, 2213 u32 *td_cmd, u32 *td_offset, 2214 struct i40e_ring *tx_ring, 2215 u32 *cd_tunneling) 2216 { 2217 struct ipv6hdr *this_ipv6_hdr; 2218 unsigned int this_tcp_hdrlen; 2219 struct iphdr *this_ip_hdr; 2220 u32 network_hdr_len; 2221 u8 l4_hdr = 0; 2222 u32 l4_tunnel = 0; 2223 2224 if (skb->encapsulation) { 2225 switch (ip_hdr(skb)->protocol) { 2226 case IPPROTO_UDP: 2227 l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING; 2228 break; 2229 default: 2230 return; 2231 } 2232 network_hdr_len = skb_inner_network_header_len(skb); 2233 this_ip_hdr = inner_ip_hdr(skb); 2234 this_ipv6_hdr = inner_ipv6_hdr(skb); 2235 this_tcp_hdrlen = inner_tcp_hdrlen(skb); 2236 2237 if (tx_flags & I40E_TX_FLAGS_IPV4) { 2238 2239 if (tx_flags & I40E_TX_FLAGS_TSO) { 2240 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4; 2241 ip_hdr(skb)->check = 0; 2242 } else { 2243 *cd_tunneling |= 2244 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM; 2245 } 2246 } else if (tx_flags & I40E_TX_FLAGS_IPV6) { 2247 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6; 2248 if (tx_flags & I40E_TX_FLAGS_TSO) 2249 ip_hdr(skb)->check = 0; 2250 } 2251 2252 /* Now set the ctx descriptor fields */ 2253 *cd_tunneling |= (skb_network_header_len(skb) >> 2) << 2254 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT | 2255 l4_tunnel | 2256 ((skb_inner_network_offset(skb) - 2257 skb_transport_offset(skb)) >> 1) << 2258 I40E_TXD_CTX_QW0_NATLEN_SHIFT; 2259 if (this_ip_hdr->version == 6) { 2260 tx_flags &= ~I40E_TX_FLAGS_IPV4; 2261 tx_flags |= I40E_TX_FLAGS_IPV6; 2262 } 2263 } else { 2264 network_hdr_len = skb_network_header_len(skb); 2265 this_ip_hdr = ip_hdr(skb); 2266 this_ipv6_hdr = ipv6_hdr(skb); 2267 this_tcp_hdrlen = tcp_hdrlen(skb); 2268 } 2269 2270 /* Enable IP checksum offloads */ 2271 if (tx_flags & I40E_TX_FLAGS_IPV4) { 2272 l4_hdr = this_ip_hdr->protocol; 2273 /* the stack computes the IP header already, the only time we 2274 * need the hardware to recompute it is in the case of TSO. 2275 */ 2276 if (tx_flags & I40E_TX_FLAGS_TSO) { 2277 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM; 2278 this_ip_hdr->check = 0; 2279 } else { 2280 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4; 2281 } 2282 /* Now set the td_offset for IP header length */ 2283 *td_offset = (network_hdr_len >> 2) << 2284 I40E_TX_DESC_LENGTH_IPLEN_SHIFT; 2285 } else if (tx_flags & I40E_TX_FLAGS_IPV6) { 2286 l4_hdr = this_ipv6_hdr->nexthdr; 2287 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6; 2288 /* Now set the td_offset for IP header length */ 2289 *td_offset = (network_hdr_len >> 2) << 2290 I40E_TX_DESC_LENGTH_IPLEN_SHIFT; 2291 } 2292 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */ 2293 *td_offset |= (skb_network_offset(skb) >> 1) << 2294 I40E_TX_DESC_LENGTH_MACLEN_SHIFT; 2295 2296 /* Enable L4 checksum offloads */ 2297 switch (l4_hdr) { 2298 case IPPROTO_TCP: 2299 /* enable checksum offloads */ 2300 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP; 2301 *td_offset |= (this_tcp_hdrlen >> 2) << 2302 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; 2303 break; 2304 case IPPROTO_SCTP: 2305 /* enable SCTP checksum offload */ 2306 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP; 2307 *td_offset |= (sizeof(struct sctphdr) >> 2) << 2308 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; 2309 break; 2310 case IPPROTO_UDP: 2311 /* enable UDP checksum offload */ 2312 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP; 2313 *td_offset |= (sizeof(struct udphdr) >> 2) << 2314 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT; 2315 break; 2316 default: 2317 break; 2318 } 2319 } 2320 2321 /** 2322 * i40e_create_tx_ctx Build the Tx context descriptor 2323 * @tx_ring: ring to create the descriptor on 2324 * @cd_type_cmd_tso_mss: Quad Word 1 2325 * @cd_tunneling: Quad Word 0 - bits 0-31 2326 * @cd_l2tag2: Quad Word 0 - bits 32-63 2327 **/ 2328 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring, 2329 const u64 cd_type_cmd_tso_mss, 2330 const u32 cd_tunneling, const u32 cd_l2tag2) 2331 { 2332 struct i40e_tx_context_desc *context_desc; 2333 int i = tx_ring->next_to_use; 2334 2335 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) && 2336 !cd_tunneling && !cd_l2tag2) 2337 return; 2338 2339 /* grab the next descriptor */ 2340 context_desc = I40E_TX_CTXTDESC(tx_ring, i); 2341 2342 i++; 2343 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0; 2344 2345 /* cpu_to_le32 and assign to struct fields */ 2346 context_desc->tunneling_params = cpu_to_le32(cd_tunneling); 2347 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2); 2348 context_desc->rsvd = cpu_to_le16(0); 2349 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss); 2350 } 2351 2352 /** 2353 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions 2354 * @tx_ring: the ring to be checked 2355 * @size: the size buffer we want to assure is available 2356 * 2357 * Returns -EBUSY if a stop is needed, else 0 2358 **/ 2359 static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) 2360 { 2361 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index); 2362 /* Memory barrier before checking head and tail */ 2363 smp_mb(); 2364 2365 /* Check again in a case another CPU has just made room available. */ 2366 if (likely(I40E_DESC_UNUSED(tx_ring) < size)) 2367 return -EBUSY; 2368 2369 /* A reprieve! - use start_queue because it doesn't call schedule */ 2370 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index); 2371 ++tx_ring->tx_stats.restart_queue; 2372 return 0; 2373 } 2374 2375 /** 2376 * i40e_maybe_stop_tx - 1st level check for tx stop conditions 2377 * @tx_ring: the ring to be checked 2378 * @size: the size buffer we want to assure is available 2379 * 2380 * Returns 0 if stop is not needed 2381 **/ 2382 #ifdef I40E_FCOE 2383 int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) 2384 #else 2385 static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size) 2386 #endif 2387 { 2388 if (likely(I40E_DESC_UNUSED(tx_ring) >= size)) 2389 return 0; 2390 return __i40e_maybe_stop_tx(tx_ring, size); 2391 } 2392 2393 /** 2394 * i40e_chk_linearize - Check if there are more than 8 fragments per packet 2395 * @skb: send buffer 2396 * @tx_flags: collected send information 2397 * @hdr_len: size of the packet header 2398 * 2399 * Note: Our HW can't scatter-gather more than 8 fragments to build 2400 * a packet on the wire and so we need to figure out the cases where we 2401 * need to linearize the skb. 2402 **/ 2403 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags, 2404 const u8 hdr_len) 2405 { 2406 struct skb_frag_struct *frag; 2407 bool linearize = false; 2408 unsigned int size = 0; 2409 u16 num_frags; 2410 u16 gso_segs; 2411 2412 num_frags = skb_shinfo(skb)->nr_frags; 2413 gso_segs = skb_shinfo(skb)->gso_segs; 2414 2415 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) { 2416 u16 j = 1; 2417 2418 if (num_frags < (I40E_MAX_BUFFER_TXD)) 2419 goto linearize_chk_done; 2420 /* try the simple math, if we have too many frags per segment */ 2421 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) > 2422 I40E_MAX_BUFFER_TXD) { 2423 linearize = true; 2424 goto linearize_chk_done; 2425 } 2426 frag = &skb_shinfo(skb)->frags[0]; 2427 size = hdr_len; 2428 /* we might still have more fragments per segment */ 2429 do { 2430 size += skb_frag_size(frag); 2431 frag++; j++; 2432 if (j == I40E_MAX_BUFFER_TXD) { 2433 if (size < skb_shinfo(skb)->gso_size) { 2434 linearize = true; 2435 break; 2436 } 2437 j = 1; 2438 size -= skb_shinfo(skb)->gso_size; 2439 if (size) 2440 j++; 2441 size += hdr_len; 2442 } 2443 num_frags--; 2444 } while (num_frags); 2445 } else { 2446 if (num_frags >= I40E_MAX_BUFFER_TXD) 2447 linearize = true; 2448 } 2449 2450 linearize_chk_done: 2451 return linearize; 2452 } 2453 2454 /** 2455 * i40e_tx_map - Build the Tx descriptor 2456 * @tx_ring: ring to send buffer on 2457 * @skb: send buffer 2458 * @first: first buffer info buffer to use 2459 * @tx_flags: collected send information 2460 * @hdr_len: size of the packet header 2461 * @td_cmd: the command field in the descriptor 2462 * @td_offset: offset for checksum or crc 2463 **/ 2464 #ifdef I40E_FCOE 2465 void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb, 2466 struct i40e_tx_buffer *first, u32 tx_flags, 2467 const u8 hdr_len, u32 td_cmd, u32 td_offset) 2468 #else 2469 static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb, 2470 struct i40e_tx_buffer *first, u32 tx_flags, 2471 const u8 hdr_len, u32 td_cmd, u32 td_offset) 2472 #endif 2473 { 2474 unsigned int data_len = skb->data_len; 2475 unsigned int size = skb_headlen(skb); 2476 struct skb_frag_struct *frag; 2477 struct i40e_tx_buffer *tx_bi; 2478 struct i40e_tx_desc *tx_desc; 2479 u16 i = tx_ring->next_to_use; 2480 u32 td_tag = 0; 2481 dma_addr_t dma; 2482 u16 gso_segs; 2483 2484 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) { 2485 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1; 2486 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >> 2487 I40E_TX_FLAGS_VLAN_SHIFT; 2488 } 2489 2490 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) 2491 gso_segs = skb_shinfo(skb)->gso_segs; 2492 else 2493 gso_segs = 1; 2494 2495 /* multiply data chunks by size of headers */ 2496 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len); 2497 first->gso_segs = gso_segs; 2498 first->skb = skb; 2499 first->tx_flags = tx_flags; 2500 2501 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE); 2502 2503 tx_desc = I40E_TX_DESC(tx_ring, i); 2504 tx_bi = first; 2505 2506 for (frag = &skb_shinfo(skb)->frags[0];; frag++) { 2507 if (dma_mapping_error(tx_ring->dev, dma)) 2508 goto dma_error; 2509 2510 /* record length, and DMA address */ 2511 dma_unmap_len_set(tx_bi, len, size); 2512 dma_unmap_addr_set(tx_bi, dma, dma); 2513 2514 tx_desc->buffer_addr = cpu_to_le64(dma); 2515 2516 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) { 2517 tx_desc->cmd_type_offset_bsz = 2518 build_ctob(td_cmd, td_offset, 2519 I40E_MAX_DATA_PER_TXD, td_tag); 2520 2521 tx_desc++; 2522 i++; 2523 if (i == tx_ring->count) { 2524 tx_desc = I40E_TX_DESC(tx_ring, 0); 2525 i = 0; 2526 } 2527 2528 dma += I40E_MAX_DATA_PER_TXD; 2529 size -= I40E_MAX_DATA_PER_TXD; 2530 2531 tx_desc->buffer_addr = cpu_to_le64(dma); 2532 } 2533 2534 if (likely(!data_len)) 2535 break; 2536 2537 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset, 2538 size, td_tag); 2539 2540 tx_desc++; 2541 i++; 2542 if (i == tx_ring->count) { 2543 tx_desc = I40E_TX_DESC(tx_ring, 0); 2544 i = 0; 2545 } 2546 2547 size = skb_frag_size(frag); 2548 data_len -= size; 2549 2550 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size, 2551 DMA_TO_DEVICE); 2552 2553 tx_bi = &tx_ring->tx_bi[i]; 2554 } 2555 2556 /* Place RS bit on last descriptor of any packet that spans across the 2557 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline. 2558 */ 2559 if (((i & WB_STRIDE) != WB_STRIDE) && 2560 (first <= &tx_ring->tx_bi[i]) && 2561 (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) { 2562 tx_desc->cmd_type_offset_bsz = 2563 build_ctob(td_cmd, td_offset, size, td_tag) | 2564 cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP << 2565 I40E_TXD_QW1_CMD_SHIFT); 2566 } else { 2567 tx_desc->cmd_type_offset_bsz = 2568 build_ctob(td_cmd, td_offset, size, td_tag) | 2569 cpu_to_le64((u64)I40E_TXD_CMD << 2570 I40E_TXD_QW1_CMD_SHIFT); 2571 } 2572 2573 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev, 2574 tx_ring->queue_index), 2575 first->bytecount); 2576 2577 /* set the timestamp */ 2578 first->time_stamp = jiffies; 2579 2580 /* Force memory writes to complete before letting h/w 2581 * know there are new descriptors to fetch. (Only 2582 * applicable for weak-ordered memory model archs, 2583 * such as IA-64). 2584 */ 2585 wmb(); 2586 2587 /* set next_to_watch value indicating a packet is present */ 2588 first->next_to_watch = tx_desc; 2589 2590 i++; 2591 if (i == tx_ring->count) 2592 i = 0; 2593 2594 tx_ring->next_to_use = i; 2595 2596 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED); 2597 /* notify HW of packet */ 2598 if (!skb->xmit_more || 2599 netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev, 2600 tx_ring->queue_index))) 2601 writel(i, tx_ring->tail); 2602 2603 return; 2604 2605 dma_error: 2606 dev_info(tx_ring->dev, "TX DMA map failed\n"); 2607 2608 /* clear dma mappings for failed tx_bi map */ 2609 for (;;) { 2610 tx_bi = &tx_ring->tx_bi[i]; 2611 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi); 2612 if (tx_bi == first) 2613 break; 2614 if (i == 0) 2615 i = tx_ring->count; 2616 i--; 2617 } 2618 2619 tx_ring->next_to_use = i; 2620 } 2621 2622 /** 2623 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed 2624 * @skb: send buffer 2625 * @tx_ring: ring to send buffer on 2626 * 2627 * Returns number of data descriptors needed for this skb. Returns 0 to indicate 2628 * there is not enough descriptors available in this ring since we need at least 2629 * one descriptor. 2630 **/ 2631 #ifdef I40E_FCOE 2632 int i40e_xmit_descriptor_count(struct sk_buff *skb, 2633 struct i40e_ring *tx_ring) 2634 #else 2635 static int i40e_xmit_descriptor_count(struct sk_buff *skb, 2636 struct i40e_ring *tx_ring) 2637 #endif 2638 { 2639 unsigned int f; 2640 int count = 0; 2641 2642 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD, 2643 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD, 2644 * + 4 desc gap to avoid the cache line where head is, 2645 * + 1 desc for context descriptor, 2646 * otherwise try next time 2647 */ 2648 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) 2649 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size); 2650 2651 count += TXD_USE_COUNT(skb_headlen(skb)); 2652 if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) { 2653 tx_ring->tx_stats.tx_busy++; 2654 return 0; 2655 } 2656 return count; 2657 } 2658 2659 /** 2660 * i40e_xmit_frame_ring - Sends buffer on Tx ring 2661 * @skb: send buffer 2662 * @tx_ring: ring to send buffer on 2663 * 2664 * Returns NETDEV_TX_OK if sent, else an error code 2665 **/ 2666 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb, 2667 struct i40e_ring *tx_ring) 2668 { 2669 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT; 2670 u32 cd_tunneling = 0, cd_l2tag2 = 0; 2671 struct i40e_tx_buffer *first; 2672 u32 td_offset = 0; 2673 u32 tx_flags = 0; 2674 __be16 protocol; 2675 u32 td_cmd = 0; 2676 u8 hdr_len = 0; 2677 int tsyn; 2678 int tso; 2679 if (0 == i40e_xmit_descriptor_count(skb, tx_ring)) 2680 return NETDEV_TX_BUSY; 2681 2682 /* prepare the xmit flags */ 2683 if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags)) 2684 goto out_drop; 2685 2686 /* obtain protocol of skb */ 2687 protocol = vlan_get_protocol(skb); 2688 2689 /* record the location of the first descriptor for this packet */ 2690 first = &tx_ring->tx_bi[tx_ring->next_to_use]; 2691 2692 /* setup IPv4/IPv6 offloads */ 2693 if (protocol == htons(ETH_P_IP)) 2694 tx_flags |= I40E_TX_FLAGS_IPV4; 2695 else if (protocol == htons(ETH_P_IPV6)) 2696 tx_flags |= I40E_TX_FLAGS_IPV6; 2697 2698 tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len, 2699 &cd_type_cmd_tso_mss, &cd_tunneling); 2700 2701 if (tso < 0) 2702 goto out_drop; 2703 else if (tso) 2704 tx_flags |= I40E_TX_FLAGS_TSO; 2705 2706 tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss); 2707 2708 if (tsyn) 2709 tx_flags |= I40E_TX_FLAGS_TSYN; 2710 2711 if (i40e_chk_linearize(skb, tx_flags, hdr_len)) 2712 if (skb_linearize(skb)) 2713 goto out_drop; 2714 2715 skb_tx_timestamp(skb); 2716 2717 /* always enable CRC insertion offload */ 2718 td_cmd |= I40E_TX_DESC_CMD_ICRC; 2719 2720 /* Always offload the checksum, since it's in the data descriptor */ 2721 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2722 tx_flags |= I40E_TX_FLAGS_CSUM; 2723 2724 i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset, 2725 tx_ring, &cd_tunneling); 2726 } 2727 2728 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss, 2729 cd_tunneling, cd_l2tag2); 2730 2731 /* Add Flow Director ATR if it's enabled. 2732 * 2733 * NOTE: this must always be directly before the data descriptor. 2734 */ 2735 i40e_atr(tx_ring, skb, tx_flags, protocol); 2736 2737 i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len, 2738 td_cmd, td_offset); 2739 2740 return NETDEV_TX_OK; 2741 2742 out_drop: 2743 dev_kfree_skb_any(skb); 2744 return NETDEV_TX_OK; 2745 } 2746 2747 /** 2748 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer 2749 * @skb: send buffer 2750 * @netdev: network interface device structure 2751 * 2752 * Returns NETDEV_TX_OK if sent, else an error code 2753 **/ 2754 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev) 2755 { 2756 struct i40e_netdev_priv *np = netdev_priv(netdev); 2757 struct i40e_vsi *vsi = np->vsi; 2758 struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping]; 2759 2760 /* hardware can't handle really short frames, hardware padding works 2761 * beyond this point 2762 */ 2763 if (skb_put_padto(skb, I40E_MIN_TX_LEN)) 2764 return NETDEV_TX_OK; 2765 2766 return i40e_xmit_frame_ring(skb, tx_ring); 2767 } 2768