1 /****************************************************************************** 2 3 Copyright (c) 2001-2017, Intel Corporation 4 All rights reserved. 5 6 Redistribution and use in source and binary forms, with or without 7 modification, are permitted provided that the following conditions are met: 8 9 1. Redistributions of source code must retain the above copyright notice, 10 this list of conditions and the following disclaimer. 11 12 2. Redistributions in binary form must reproduce the above copyright 13 notice, this list of conditions and the following disclaimer in the 14 documentation and/or other materials provided with the distribution. 15 16 3. Neither the name of the Intel Corporation nor the names of its 17 contributors may be used to endorse or promote products derived from 18 this software without specific prior written permission. 19 20 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 21 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 24 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 POSSIBILITY OF SUCH DAMAGE. 31 32 ******************************************************************************/ 33 34 #ifndef IXGBE_STANDALONE_BUILD 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 #include "opt_rss.h" 38 #endif 39 40 #include "ixgbe.h" 41 42 /************************************************************************ 43 * Local Function prototypes 44 ************************************************************************/ 45 static int ixgbe_isc_txd_encap(void *, if_pkt_info_t); 46 static void ixgbe_isc_txd_flush(void *, uint16_t, qidx_t); 47 static int ixgbe_isc_txd_credits_update(void *, uint16_t, bool); 48 49 static void ixgbe_isc_rxd_refill(void *, if_rxd_update_t); 50 static void ixgbe_isc_rxd_flush(void *, uint16_t, uint8_t, qidx_t); 51 static int ixgbe_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t); 52 static int ixgbe_isc_rxd_pkt_get(void *, if_rxd_info_t); 53 54 static void ixgbe_rx_checksum(uint32_t, if_rxd_info_t, uint32_t); 55 static int ixgbe_tx_ctx_setup(struct ixgbe_adv_tx_context_desc *, 56 if_pkt_info_t); 57 58 extern void ixgbe_if_enable_intr(if_ctx_t ctx); 59 static int ixgbe_determine_rsstype(uint16_t pkt_info); 60 61 struct if_txrx ixgbe_txrx = { 62 .ift_txd_encap = ixgbe_isc_txd_encap, 63 .ift_txd_flush = ixgbe_isc_txd_flush, 64 .ift_txd_credits_update = ixgbe_isc_txd_credits_update, 65 .ift_rxd_available = ixgbe_isc_rxd_available, 66 .ift_rxd_pkt_get = ixgbe_isc_rxd_pkt_get, 67 .ift_rxd_refill = ixgbe_isc_rxd_refill, 68 .ift_rxd_flush = ixgbe_isc_rxd_flush, 69 .ift_legacy_intr = NULL 70 }; 71 72 /************************************************************************ 73 * ixgbe_tx_ctx_setup 74 * 75 * Advanced Context Descriptor setup for VLAN, CSUM or TSO 76 * 77 ************************************************************************/ 78 static int 79 ixgbe_tx_ctx_setup(struct ixgbe_adv_tx_context_desc *TXD, if_pkt_info_t pi) 80 { 81 uint32_t vlan_macip_lens, type_tucmd_mlhl; 82 uint32_t olinfo_status, mss_l4len_idx, pktlen, offload; 83 u8 ehdrlen; 84 85 offload = true; 86 olinfo_status = mss_l4len_idx = vlan_macip_lens = type_tucmd_mlhl = 0; 87 /* VLAN MACLEN IPLEN */ 88 vlan_macip_lens |= (htole16(pi->ipi_vtag) << IXGBE_ADVTXD_VLAN_SHIFT); 89 90 /* 91 * Some of our VF devices need a context descriptor for every 92 * packet. That means the ehdrlen needs to be non-zero in order 93 * for the host driver not to flag a malicious event. The stack 94 * will most likely populate this for all other reasons of why 95 * this function was called. 96 */ 97 if (pi->ipi_ehdrlen == 0) { 98 ehdrlen = ETHER_HDR_LEN; 99 ehdrlen += (pi->ipi_vtag != 0) ? ETHER_VLAN_ENCAP_LEN : 0; 100 } else 101 ehdrlen = pi->ipi_ehdrlen; 102 vlan_macip_lens |= ehdrlen << IXGBE_ADVTXD_MACLEN_SHIFT; 103 104 pktlen = pi->ipi_len; 105 /* First check if TSO is to be used */ 106 if (pi->ipi_csum_flags & CSUM_TSO) { 107 /* This is used in the transmit desc in encap */ 108 pktlen = pi->ipi_len - ehdrlen - pi->ipi_ip_hlen - pi->ipi_tcp_hlen; 109 mss_l4len_idx |= (pi->ipi_tso_segsz << IXGBE_ADVTXD_MSS_SHIFT); 110 mss_l4len_idx |= (pi->ipi_tcp_hlen << IXGBE_ADVTXD_L4LEN_SHIFT); 111 } 112 113 olinfo_status |= pktlen << IXGBE_ADVTXD_PAYLEN_SHIFT; 114 115 if (pi->ipi_flags & IPI_TX_IPV4) { 116 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4; 117 /* Tell transmit desc to also do IPv4 checksum. */ 118 if (pi->ipi_csum_flags & (CSUM_IP|CSUM_TSO)) 119 olinfo_status |= IXGBE_TXD_POPTS_IXSM << 8; 120 } else if (pi->ipi_flags & IPI_TX_IPV6) 121 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV6; 122 else 123 offload = false; 124 125 vlan_macip_lens |= pi->ipi_ip_hlen; 126 127 switch (pi->ipi_ipproto) { 128 case IPPROTO_TCP: 129 if (pi->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP6_TCP | CSUM_TSO)) 130 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP; 131 else 132 offload = false; 133 break; 134 case IPPROTO_UDP: 135 if (pi->ipi_csum_flags & (CSUM_IP_UDP | CSUM_IP6_UDP)) 136 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP; 137 else 138 offload = false; 139 break; 140 case IPPROTO_SCTP: 141 if (pi->ipi_csum_flags & (CSUM_IP_SCTP | CSUM_IP6_SCTP)) 142 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP; 143 else 144 offload = false; 145 break; 146 default: 147 offload = false; 148 break; 149 } 150 /* Insert L4 checksum into data descriptors */ 151 if (offload) 152 olinfo_status |= IXGBE_TXD_POPTS_TXSM << 8; 153 154 type_tucmd_mlhl |= IXGBE_ADVTXD_DCMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT; 155 156 /* Now copy bits into descriptor */ 157 TXD->vlan_macip_lens = htole32(vlan_macip_lens); 158 TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl); 159 TXD->seqnum_seed = htole32(0); 160 TXD->mss_l4len_idx = htole32(mss_l4len_idx); 161 162 return (olinfo_status); 163 } /* ixgbe_tx_ctx_setup */ 164 165 /************************************************************************ 166 * ixgbe_isc_txd_encap 167 ************************************************************************/ 168 static int 169 ixgbe_isc_txd_encap(void *arg, if_pkt_info_t pi) 170 { 171 struct ixgbe_softc *sc = arg; 172 if_softc_ctx_t scctx = sc->shared; 173 struct ix_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx]; 174 struct tx_ring *txr = &que->txr; 175 int nsegs = pi->ipi_nsegs; 176 bus_dma_segment_t *segs = pi->ipi_segs; 177 union ixgbe_adv_tx_desc *txd = NULL; 178 struct ixgbe_adv_tx_context_desc *TXD; 179 int i, j, first, pidx_last; 180 uint32_t olinfo_status, cmd, flags; 181 qidx_t ntxd; 182 183 cmd = (IXGBE_ADVTXD_DTYP_DATA | 184 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT); 185 186 if (pi->ipi_mflags & M_VLANTAG) 187 cmd |= IXGBE_ADVTXD_DCMD_VLE; 188 189 i = first = pi->ipi_pidx; 190 flags = (pi->ipi_flags & IPI_TX_INTR) ? IXGBE_TXD_CMD_RS : 0; 191 ntxd = scctx->isc_ntxd[0]; 192 193 TXD = (struct ixgbe_adv_tx_context_desc *) &txr->tx_base[first]; 194 if ((pi->ipi_csum_flags & CSUM_OFFLOAD) || 195 (sc->feat_en & IXGBE_FEATURE_NEEDS_CTXD) || 196 pi->ipi_vtag) { 197 /********************************************* 198 * Set up the appropriate offload context 199 * this will consume the first descriptor 200 *********************************************/ 201 olinfo_status = ixgbe_tx_ctx_setup(TXD, pi); 202 if (pi->ipi_csum_flags & CSUM_TSO) { 203 cmd |= IXGBE_ADVTXD_DCMD_TSE; 204 ++txr->tso_tx; 205 } 206 207 if (++i == scctx->isc_ntxd[0]) 208 i = 0; 209 } else { 210 /* Indicate the whole packet as payload when not doing TSO */ 211 olinfo_status = pi->ipi_len << IXGBE_ADVTXD_PAYLEN_SHIFT; 212 } 213 214 olinfo_status |= IXGBE_ADVTXD_CC; 215 pidx_last = 0; 216 for (j = 0; j < nsegs; j++) { 217 bus_size_t seglen; 218 219 txd = &txr->tx_base[i]; 220 seglen = segs[j].ds_len; 221 222 txd->read.buffer_addr = htole64(segs[j].ds_addr); 223 txd->read.cmd_type_len = htole32(cmd | seglen); 224 txd->read.olinfo_status = htole32(olinfo_status); 225 226 pidx_last = i; 227 if (++i == scctx->isc_ntxd[0]) { 228 i = 0; 229 } 230 } 231 232 if (flags) { 233 txr->tx_rsq[txr->tx_rs_pidx] = pidx_last; 234 txr->tx_rs_pidx = (txr->tx_rs_pidx + 1) & (ntxd - 1); 235 } 236 txd->read.cmd_type_len |= htole32(IXGBE_TXD_CMD_EOP | flags); 237 238 txr->bytes += pi->ipi_len; 239 pi->ipi_new_pidx = i; 240 241 ++txr->total_packets; 242 243 return (0); 244 } /* ixgbe_isc_txd_encap */ 245 246 /************************************************************************ 247 * ixgbe_isc_txd_flush 248 ************************************************************************/ 249 static void 250 ixgbe_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx) 251 { 252 struct ixgbe_softc *sc = arg; 253 struct ix_tx_queue *que = &sc->tx_queues[txqid]; 254 struct tx_ring *txr = &que->txr; 255 256 IXGBE_WRITE_REG(&sc->hw, txr->tail, pidx); 257 } /* ixgbe_isc_txd_flush */ 258 259 /************************************************************************ 260 * ixgbe_isc_txd_credits_update 261 ************************************************************************/ 262 static int 263 ixgbe_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear) 264 { 265 struct ixgbe_softc *sc = arg; 266 if_softc_ctx_t scctx = sc->shared; 267 struct ix_tx_queue *que = &sc->tx_queues[txqid]; 268 struct tx_ring *txr = &que->txr; 269 qidx_t processed = 0; 270 int updated; 271 qidx_t cur, prev, ntxd, rs_cidx; 272 int32_t delta; 273 uint8_t status; 274 275 rs_cidx = txr->tx_rs_cidx; 276 if (rs_cidx == txr->tx_rs_pidx) 277 return (0); 278 279 cur = txr->tx_rsq[rs_cidx]; 280 status = txr->tx_base[cur].wb.status; 281 updated = !!(status & IXGBE_TXD_STAT_DD); 282 283 if (!updated) 284 return (0); 285 286 /* If clear is false just let caller know that there 287 * are descriptors to reclaim */ 288 if (!clear) 289 return (1); 290 291 prev = txr->tx_cidx_processed; 292 ntxd = scctx->isc_ntxd[0]; 293 do { 294 MPASS(prev != cur); 295 delta = (int32_t)cur - (int32_t)prev; 296 if (delta < 0) 297 delta += ntxd; 298 MPASS(delta > 0); 299 300 processed += delta; 301 prev = cur; 302 rs_cidx = (rs_cidx + 1) & (ntxd - 1); 303 if (rs_cidx == txr->tx_rs_pidx) 304 break; 305 306 cur = txr->tx_rsq[rs_cidx]; 307 status = txr->tx_base[cur].wb.status; 308 } while ((status & IXGBE_TXD_STAT_DD)); 309 310 txr->tx_rs_cidx = rs_cidx; 311 txr->tx_cidx_processed = prev; 312 313 return (processed); 314 } /* ixgbe_isc_txd_credits_update */ 315 316 /************************************************************************ 317 * ixgbe_isc_rxd_refill 318 ************************************************************************/ 319 static void 320 ixgbe_isc_rxd_refill(void *arg, if_rxd_update_t iru) 321 { 322 struct ixgbe_softc *sc = arg; 323 struct ix_rx_queue *que = &sc->rx_queues[iru->iru_qsidx]; 324 struct rx_ring *rxr = &que->rxr; 325 uint64_t *paddrs; 326 int i; 327 uint32_t next_pidx, pidx; 328 uint16_t count; 329 330 paddrs = iru->iru_paddrs; 331 pidx = iru->iru_pidx; 332 count = iru->iru_count; 333 334 for (i = 0, next_pidx = pidx; i < count; i++) { 335 rxr->rx_base[next_pidx].read.pkt_addr = htole64(paddrs[i]); 336 if (++next_pidx == sc->shared->isc_nrxd[0]) 337 next_pidx = 0; 338 } 339 } /* ixgbe_isc_rxd_refill */ 340 341 /************************************************************************ 342 * ixgbe_isc_rxd_flush 343 ************************************************************************/ 344 static void 345 ixgbe_isc_rxd_flush(void *arg, uint16_t qsidx, uint8_t flidx __unused, qidx_t pidx) 346 { 347 struct ixgbe_softc *sc = arg; 348 struct ix_rx_queue *que = &sc->rx_queues[qsidx]; 349 struct rx_ring *rxr = &que->rxr; 350 351 IXGBE_WRITE_REG(&sc->hw, rxr->tail, pidx); 352 } /* ixgbe_isc_rxd_flush */ 353 354 /************************************************************************ 355 * ixgbe_isc_rxd_available 356 ************************************************************************/ 357 static int 358 ixgbe_isc_rxd_available(void *arg, uint16_t qsidx, qidx_t pidx, qidx_t budget) 359 { 360 struct ixgbe_softc *sc = arg; 361 struct ix_rx_queue *que = &sc->rx_queues[qsidx]; 362 struct rx_ring *rxr = &que->rxr; 363 union ixgbe_adv_rx_desc *rxd; 364 uint32_t staterr; 365 int cnt, i, nrxd; 366 367 nrxd = sc->shared->isc_nrxd[0]; 368 for (cnt = 0, i = pidx; cnt < nrxd && cnt <= budget;) { 369 rxd = &rxr->rx_base[i]; 370 staterr = le32toh(rxd->wb.upper.status_error); 371 372 if ((staterr & IXGBE_RXD_STAT_DD) == 0) 373 break; 374 if (++i == nrxd) 375 i = 0; 376 if (staterr & IXGBE_RXD_STAT_EOP) 377 cnt++; 378 } 379 return (cnt); 380 } /* ixgbe_isc_rxd_available */ 381 382 /************************************************************************ 383 * ixgbe_isc_rxd_pkt_get 384 * 385 * Routine sends data which has been dma'ed into host memory 386 * to upper layer. Initialize ri structure. 387 * 388 * Returns 0 upon success, errno on failure 389 ************************************************************************/ 390 391 static int 392 ixgbe_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri) 393 { 394 struct ixgbe_softc *sc = arg; 395 if_softc_ctx_t scctx = sc->shared; 396 struct ix_rx_queue *que = &sc->rx_queues[ri->iri_qsidx]; 397 struct rx_ring *rxr = &que->rxr; 398 union ixgbe_adv_rx_desc *rxd; 399 400 uint16_t pkt_info, len, cidx, i; 401 uint32_t ptype; 402 uint32_t staterr = 0; 403 bool eop; 404 405 i = 0; 406 cidx = ri->iri_cidx; 407 do { 408 rxd = &rxr->rx_base[cidx]; 409 staterr = le32toh(rxd->wb.upper.status_error); 410 pkt_info = le16toh(rxd->wb.lower.lo_dword.hs_rss.pkt_info); 411 412 /* Error Checking then decrement count */ 413 MPASS ((staterr & IXGBE_RXD_STAT_DD) != 0); 414 415 len = le16toh(rxd->wb.upper.length); 416 ptype = le32toh(rxd->wb.lower.lo_dword.data) & 417 IXGBE_RXDADV_PKTTYPE_MASK; 418 419 ri->iri_len += len; 420 rxr->bytes += len; 421 422 rxd->wb.upper.status_error = 0; 423 eop = ((staterr & IXGBE_RXD_STAT_EOP) != 0); 424 425 /* Make sure bad packets are discarded */ 426 if (eop && (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) != 0) { 427 if (sc->feat_en & IXGBE_FEATURE_VF) 428 if_inc_counter(ri->iri_ifp, IFCOUNTER_IERRORS, 1); 429 430 rxr->rx_discarded++; 431 return (EBADMSG); 432 } 433 ri->iri_frags[i].irf_flid = 0; 434 ri->iri_frags[i].irf_idx = cidx; 435 ri->iri_frags[i].irf_len = len; 436 if (++cidx == sc->shared->isc_nrxd[0]) 437 cidx = 0; 438 i++; 439 /* even a 16K packet shouldn't consume more than 8 clusters */ 440 MPASS(i < 9); 441 } while (!eop); 442 443 rxr->rx_packets++; 444 rxr->packets++; 445 rxr->rx_bytes += ri->iri_len; 446 447 if ((scctx->isc_capenable & IFCAP_RXCSUM) != 0) 448 ixgbe_rx_checksum(staterr, ri, ptype); 449 450 ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss); 451 ri->iri_rsstype = ixgbe_determine_rsstype(pkt_info); 452 if ((sc->feat_en & IXGBE_FEATURE_RSS) == 0) { 453 if (ri->iri_rsstype == M_HASHTYPE_OPAQUE) 454 ri->iri_rsstype = M_HASHTYPE_NONE; 455 else 456 ri->iri_rsstype = M_HASHTYPE_OPAQUE_HASH; 457 } 458 if ((rxr->vtag_strip) && (staterr & IXGBE_RXD_STAT_VP)) { 459 ri->iri_vtag = le16toh(rxd->wb.upper.vlan); 460 ri->iri_flags |= M_VLANTAG; 461 } 462 463 ri->iri_nfrags = i; 464 return (0); 465 } /* ixgbe_isc_rxd_pkt_get */ 466 467 /************************************************************************ 468 * ixgbe_rx_checksum 469 * 470 * Verify that the hardware indicated that the checksum is valid. 471 * Inform the stack about the status of checksum so that stack 472 * doesn't spend time verifying the checksum. 473 ************************************************************************/ 474 static void 475 ixgbe_rx_checksum(uint32_t staterr, if_rxd_info_t ri, uint32_t ptype) 476 { 477 uint16_t status = (uint16_t)staterr; 478 uint8_t errors = (uint8_t)(staterr >> 24); 479 480 /* If there is a layer 3 or 4 error we are done */ 481 if (__predict_false(errors & (IXGBE_RXD_ERR_IPE | IXGBE_RXD_ERR_TCPE))) 482 return; 483 484 /* IP Checksum Good */ 485 if (status & IXGBE_RXD_STAT_IPCS) 486 ri->iri_csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID); 487 488 /* Valid L4E checksum */ 489 if (__predict_true(status & IXGBE_RXD_STAT_L4CS)) { 490 /* SCTP header present. */ 491 if (__predict_false((ptype & IXGBE_RXDADV_PKTTYPE_ETQF) == 0 && 492 (ptype & IXGBE_RXDADV_PKTTYPE_SCTP) != 0)) { 493 ri->iri_csum_flags |= CSUM_SCTP_VALID; 494 } else { 495 ri->iri_csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 496 ri->iri_csum_data = htons(0xffff); 497 } 498 } 499 } /* ixgbe_rx_checksum */ 500 501 /************************************************************************ 502 * ixgbe_determine_rsstype 503 * 504 * Parse the packet type to determine the appropriate hash 505 ************************************************************************/ 506 static int 507 ixgbe_determine_rsstype(uint16_t pkt_info) 508 { 509 switch (pkt_info & IXGBE_RXDADV_RSSTYPE_MASK) { 510 case IXGBE_RXDADV_RSSTYPE_IPV4_TCP: 511 return M_HASHTYPE_RSS_TCP_IPV4; 512 case IXGBE_RXDADV_RSSTYPE_IPV4: 513 return M_HASHTYPE_RSS_IPV4; 514 case IXGBE_RXDADV_RSSTYPE_IPV6_TCP: 515 return M_HASHTYPE_RSS_TCP_IPV6; 516 case IXGBE_RXDADV_RSSTYPE_IPV6_EX: 517 return M_HASHTYPE_RSS_IPV6_EX; 518 case IXGBE_RXDADV_RSSTYPE_IPV6: 519 return M_HASHTYPE_RSS_IPV6; 520 case IXGBE_RXDADV_RSSTYPE_IPV6_TCP_EX: 521 return M_HASHTYPE_RSS_TCP_IPV6_EX; 522 case IXGBE_RXDADV_RSSTYPE_IPV4_UDP: 523 return M_HASHTYPE_RSS_UDP_IPV4; 524 case IXGBE_RXDADV_RSSTYPE_IPV6_UDP: 525 return M_HASHTYPE_RSS_UDP_IPV6; 526 case IXGBE_RXDADV_RSSTYPE_IPV6_UDP_EX: 527 return M_HASHTYPE_RSS_UDP_IPV6_EX; 528 default: 529 return M_HASHTYPE_OPAQUE; 530 } 531 } /* ixgbe_determine_rsstype */ 532