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 /*$FreeBSD$*/ 34 35 #ifndef IXGBE_STANDALONE_BUILD 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_rss.h" 39 #endif 40 41 #include "ixgbe.h" 42 43 /************************************************************************ 44 * Local Function prototypes 45 ************************************************************************/ 46 static int ixgbe_isc_txd_encap(void *, if_pkt_info_t); 47 static void ixgbe_isc_txd_flush(void *, uint16_t, qidx_t); 48 static int ixgbe_isc_txd_credits_update(void *, uint16_t, bool); 49 50 static void ixgbe_isc_rxd_refill(void *, if_rxd_update_t); 51 static void ixgbe_isc_rxd_flush(void *, uint16_t, uint8_t, qidx_t); 52 static int ixgbe_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t); 53 static int ixgbe_isc_rxd_pkt_get(void *, if_rxd_info_t); 54 55 static void ixgbe_rx_checksum(uint32_t, if_rxd_info_t, uint32_t); 56 static int ixgbe_tx_ctx_setup(struct ixgbe_adv_tx_context_desc *, 57 if_pkt_info_t); 58 59 extern void ixgbe_if_enable_intr(if_ctx_t ctx); 60 static int ixgbe_determine_rsstype(uint16_t pkt_info); 61 62 struct if_txrx ixgbe_txrx = { 63 .ift_txd_encap = ixgbe_isc_txd_encap, 64 .ift_txd_flush = ixgbe_isc_txd_flush, 65 .ift_txd_credits_update = ixgbe_isc_txd_credits_update, 66 .ift_rxd_available = ixgbe_isc_rxd_available, 67 .ift_rxd_pkt_get = ixgbe_isc_rxd_pkt_get, 68 .ift_rxd_refill = ixgbe_isc_rxd_refill, 69 .ift_rxd_flush = ixgbe_isc_rxd_flush, 70 .ift_legacy_intr = NULL 71 }; 72 73 /************************************************************************ 74 * ixgbe_tx_ctx_setup 75 * 76 * Advanced Context Descriptor setup for VLAN, CSUM or TSO 77 * 78 ************************************************************************/ 79 static int 80 ixgbe_tx_ctx_setup(struct ixgbe_adv_tx_context_desc *TXD, if_pkt_info_t pi) 81 { 82 uint32_t vlan_macip_lens, type_tucmd_mlhl; 83 uint32_t olinfo_status, mss_l4len_idx, pktlen, offload; 84 u8 ehdrlen; 85 86 offload = true; 87 olinfo_status = mss_l4len_idx = vlan_macip_lens = type_tucmd_mlhl = 0; 88 /* VLAN MACLEN IPLEN */ 89 vlan_macip_lens |= (htole16(pi->ipi_vtag) << IXGBE_ADVTXD_VLAN_SHIFT); 90 91 /* 92 * Some of our VF devices need a context descriptor for every 93 * packet. That means the ehdrlen needs to be non-zero in order 94 * for the host driver not to flag a malicious event. The stack 95 * will most likely populate this for all other reasons of why 96 * this function was called. 97 */ 98 if (pi->ipi_ehdrlen == 0) { 99 ehdrlen = ETHER_HDR_LEN; 100 ehdrlen += (pi->ipi_vtag != 0) ? ETHER_VLAN_ENCAP_LEN : 0; 101 } else 102 ehdrlen = pi->ipi_ehdrlen; 103 vlan_macip_lens |= ehdrlen << IXGBE_ADVTXD_MACLEN_SHIFT; 104 105 pktlen = pi->ipi_len; 106 /* First check if TSO is to be used */ 107 if (pi->ipi_csum_flags & CSUM_TSO) { 108 /* This is used in the transmit desc in encap */ 109 pktlen = pi->ipi_len - ehdrlen - pi->ipi_ip_hlen - pi->ipi_tcp_hlen; 110 mss_l4len_idx |= (pi->ipi_tso_segsz << IXGBE_ADVTXD_MSS_SHIFT); 111 mss_l4len_idx |= (pi->ipi_tcp_hlen << IXGBE_ADVTXD_L4LEN_SHIFT); 112 } 113 114 olinfo_status |= pktlen << IXGBE_ADVTXD_PAYLEN_SHIFT; 115 116 if (pi->ipi_flags & IPI_TX_IPV4) { 117 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4; 118 /* Tell transmit desc to also do IPv4 checksum. */ 119 if (pi->ipi_csum_flags & (CSUM_IP|CSUM_TSO)) 120 olinfo_status |= IXGBE_TXD_POPTS_IXSM << 8; 121 } else if (pi->ipi_flags & IPI_TX_IPV6) 122 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV6; 123 else 124 offload = false; 125 126 vlan_macip_lens |= pi->ipi_ip_hlen; 127 128 switch (pi->ipi_ipproto) { 129 case IPPROTO_TCP: 130 if (pi->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP6_TCP | CSUM_TSO)) 131 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP; 132 else 133 offload = false; 134 break; 135 case IPPROTO_UDP: 136 if (pi->ipi_csum_flags & (CSUM_IP_UDP | CSUM_IP6_UDP)) 137 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP; 138 else 139 offload = false; 140 break; 141 case IPPROTO_SCTP: 142 if (pi->ipi_csum_flags & (CSUM_IP_SCTP | CSUM_IP6_SCTP)) 143 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_SCTP; 144 else 145 offload = false; 146 break; 147 default: 148 offload = false; 149 break; 150 } 151 /* Insert L4 checksum into data descriptors */ 152 if (offload) 153 olinfo_status |= IXGBE_TXD_POPTS_TXSM << 8; 154 155 type_tucmd_mlhl |= IXGBE_ADVTXD_DCMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT; 156 157 /* Now copy bits into descriptor */ 158 TXD->vlan_macip_lens = htole32(vlan_macip_lens); 159 TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl); 160 TXD->seqnum_seed = htole32(0); 161 TXD->mss_l4len_idx = htole32(mss_l4len_idx); 162 163 return (olinfo_status); 164 } /* ixgbe_tx_ctx_setup */ 165 166 /************************************************************************ 167 * ixgbe_isc_txd_encap 168 ************************************************************************/ 169 static int 170 ixgbe_isc_txd_encap(void *arg, if_pkt_info_t pi) 171 { 172 struct ixgbe_softc *sc = arg; 173 if_softc_ctx_t scctx = sc->shared; 174 struct ix_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx]; 175 struct tx_ring *txr = &que->txr; 176 int nsegs = pi->ipi_nsegs; 177 bus_dma_segment_t *segs = pi->ipi_segs; 178 union ixgbe_adv_tx_desc *txd = NULL; 179 struct ixgbe_adv_tx_context_desc *TXD; 180 int i, j, first, pidx_last; 181 uint32_t olinfo_status, cmd, flags; 182 qidx_t ntxd; 183 184 cmd = (IXGBE_ADVTXD_DTYP_DATA | 185 IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT); 186 187 if (pi->ipi_mflags & M_VLANTAG) 188 cmd |= IXGBE_ADVTXD_DCMD_VLE; 189 190 i = first = pi->ipi_pidx; 191 flags = (pi->ipi_flags & IPI_TX_INTR) ? IXGBE_TXD_CMD_RS : 0; 192 ntxd = scctx->isc_ntxd[0]; 193 194 TXD = (struct ixgbe_adv_tx_context_desc *) &txr->tx_base[first]; 195 if ((pi->ipi_csum_flags & CSUM_OFFLOAD) || 196 (sc->feat_en & IXGBE_FEATURE_NEEDS_CTXD) || 197 pi->ipi_vtag) { 198 /********************************************* 199 * Set up the appropriate offload context 200 * this will consume the first descriptor 201 *********************************************/ 202 olinfo_status = ixgbe_tx_ctx_setup(TXD, pi); 203 if (pi->ipi_csum_flags & CSUM_TSO) { 204 cmd |= IXGBE_ADVTXD_DCMD_TSE; 205 ++txr->tso_tx; 206 } 207 208 if (++i == scctx->isc_ntxd[0]) 209 i = 0; 210 } else { 211 /* Indicate the whole packet as payload when not doing TSO */ 212 olinfo_status = pi->ipi_len << IXGBE_ADVTXD_PAYLEN_SHIFT; 213 } 214 215 olinfo_status |= IXGBE_ADVTXD_CC; 216 pidx_last = 0; 217 for (j = 0; j < nsegs; j++) { 218 bus_size_t seglen; 219 220 txd = &txr->tx_base[i]; 221 seglen = segs[j].ds_len; 222 223 txd->read.buffer_addr = htole64(segs[j].ds_addr); 224 txd->read.cmd_type_len = htole32(cmd | seglen); 225 txd->read.olinfo_status = htole32(olinfo_status); 226 227 pidx_last = i; 228 if (++i == scctx->isc_ntxd[0]) { 229 i = 0; 230 } 231 } 232 233 if (flags) { 234 txr->tx_rsq[txr->tx_rs_pidx] = pidx_last; 235 txr->tx_rs_pidx = (txr->tx_rs_pidx + 1) & (ntxd - 1); 236 } 237 txd->read.cmd_type_len |= htole32(IXGBE_TXD_CMD_EOP | flags); 238 239 txr->bytes += pi->ipi_len; 240 pi->ipi_new_pidx = i; 241 242 ++txr->total_packets; 243 244 return (0); 245 } /* ixgbe_isc_txd_encap */ 246 247 /************************************************************************ 248 * ixgbe_isc_txd_flush 249 ************************************************************************/ 250 static void 251 ixgbe_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx) 252 { 253 struct ixgbe_softc *sc = arg; 254 struct ix_tx_queue *que = &sc->tx_queues[txqid]; 255 struct tx_ring *txr = &que->txr; 256 257 IXGBE_WRITE_REG(&sc->hw, txr->tail, pidx); 258 } /* ixgbe_isc_txd_flush */ 259 260 /************************************************************************ 261 * ixgbe_isc_txd_credits_update 262 ************************************************************************/ 263 static int 264 ixgbe_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear) 265 { 266 struct ixgbe_softc *sc = arg; 267 if_softc_ctx_t scctx = sc->shared; 268 struct ix_tx_queue *que = &sc->tx_queues[txqid]; 269 struct tx_ring *txr = &que->txr; 270 qidx_t processed = 0; 271 int updated; 272 qidx_t cur, prev, ntxd, rs_cidx; 273 int32_t delta; 274 uint8_t status; 275 276 rs_cidx = txr->tx_rs_cidx; 277 if (rs_cidx == txr->tx_rs_pidx) 278 return (0); 279 280 cur = txr->tx_rsq[rs_cidx]; 281 status = txr->tx_base[cur].wb.status; 282 updated = !!(status & IXGBE_TXD_STAT_DD); 283 284 if (!updated) 285 return (0); 286 287 /* If clear is false just let caller know that there 288 * are descriptors to reclaim */ 289 if (!clear) 290 return (1); 291 292 prev = txr->tx_cidx_processed; 293 ntxd = scctx->isc_ntxd[0]; 294 do { 295 MPASS(prev != cur); 296 delta = (int32_t)cur - (int32_t)prev; 297 if (delta < 0) 298 delta += ntxd; 299 MPASS(delta > 0); 300 301 processed += delta; 302 prev = cur; 303 rs_cidx = (rs_cidx + 1) & (ntxd - 1); 304 if (rs_cidx == txr->tx_rs_pidx) 305 break; 306 307 cur = txr->tx_rsq[rs_cidx]; 308 status = txr->tx_base[cur].wb.status; 309 } while ((status & IXGBE_TXD_STAT_DD)); 310 311 txr->tx_rs_cidx = rs_cidx; 312 txr->tx_cidx_processed = prev; 313 314 return (processed); 315 } /* ixgbe_isc_txd_credits_update */ 316 317 /************************************************************************ 318 * ixgbe_isc_rxd_refill 319 ************************************************************************/ 320 static void 321 ixgbe_isc_rxd_refill(void *arg, if_rxd_update_t iru) 322 { 323 struct ixgbe_softc *sc = arg; 324 struct ix_rx_queue *que = &sc->rx_queues[iru->iru_qsidx]; 325 struct rx_ring *rxr = &que->rxr; 326 uint64_t *paddrs; 327 int i; 328 uint32_t next_pidx, pidx; 329 uint16_t count; 330 331 paddrs = iru->iru_paddrs; 332 pidx = iru->iru_pidx; 333 count = iru->iru_count; 334 335 for (i = 0, next_pidx = pidx; i < count; i++) { 336 rxr->rx_base[next_pidx].read.pkt_addr = htole64(paddrs[i]); 337 if (++next_pidx == sc->shared->isc_nrxd[0]) 338 next_pidx = 0; 339 } 340 } /* ixgbe_isc_rxd_refill */ 341 342 /************************************************************************ 343 * ixgbe_isc_rxd_flush 344 ************************************************************************/ 345 static void 346 ixgbe_isc_rxd_flush(void *arg, uint16_t qsidx, uint8_t flidx __unused, qidx_t pidx) 347 { 348 struct ixgbe_softc *sc = arg; 349 struct ix_rx_queue *que = &sc->rx_queues[qsidx]; 350 struct rx_ring *rxr = &que->rxr; 351 352 IXGBE_WRITE_REG(&sc->hw, rxr->tail, pidx); 353 } /* ixgbe_isc_rxd_flush */ 354 355 /************************************************************************ 356 * ixgbe_isc_rxd_available 357 ************************************************************************/ 358 static int 359 ixgbe_isc_rxd_available(void *arg, uint16_t qsidx, qidx_t pidx, qidx_t budget) 360 { 361 struct ixgbe_softc *sc = arg; 362 struct ix_rx_queue *que = &sc->rx_queues[qsidx]; 363 struct rx_ring *rxr = &que->rxr; 364 union ixgbe_adv_rx_desc *rxd; 365 uint32_t staterr; 366 int cnt, i, nrxd; 367 368 nrxd = sc->shared->isc_nrxd[0]; 369 for (cnt = 0, i = pidx; cnt < nrxd && cnt <= budget;) { 370 rxd = &rxr->rx_base[i]; 371 staterr = le32toh(rxd->wb.upper.status_error); 372 373 if ((staterr & IXGBE_RXD_STAT_DD) == 0) 374 break; 375 if (++i == nrxd) 376 i = 0; 377 if (staterr & IXGBE_RXD_STAT_EOP) 378 cnt++; 379 } 380 return (cnt); 381 } /* ixgbe_isc_rxd_available */ 382 383 /************************************************************************ 384 * ixgbe_isc_rxd_pkt_get 385 * 386 * Routine sends data which has been dma'ed into host memory 387 * to upper layer. Initialize ri structure. 388 * 389 * Returns 0 upon success, errno on failure 390 ************************************************************************/ 391 392 static int 393 ixgbe_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri) 394 { 395 struct ixgbe_softc *sc = arg; 396 struct ix_rx_queue *que = &sc->rx_queues[ri->iri_qsidx]; 397 struct rx_ring *rxr = &que->rxr; 398 struct ifnet *ifp = iflib_get_ifp(sc->ctx); 399 union ixgbe_adv_rx_desc *rxd; 400 401 uint16_t pkt_info, len, cidx, i; 402 uint16_t vtag = 0; 403 uint32_t ptype; 404 uint32_t staterr = 0; 405 bool eop; 406 407 i = 0; 408 cidx = ri->iri_cidx; 409 do { 410 rxd = &rxr->rx_base[cidx]; 411 staterr = le32toh(rxd->wb.upper.status_error); 412 pkt_info = le16toh(rxd->wb.lower.lo_dword.hs_rss.pkt_info); 413 414 /* Error Checking then decrement count */ 415 MPASS ((staterr & IXGBE_RXD_STAT_DD) != 0); 416 417 len = le16toh(rxd->wb.upper.length); 418 ptype = le32toh(rxd->wb.lower.lo_dword.data) & 419 IXGBE_RXDADV_PKTTYPE_MASK; 420 421 ri->iri_len += len; 422 rxr->bytes += len; 423 424 rxd->wb.upper.status_error = 0; 425 eop = ((staterr & IXGBE_RXD_STAT_EOP) != 0); 426 427 if ( (rxr->vtag_strip) && (staterr & IXGBE_RXD_STAT_VP) ) { 428 vtag = le16toh(rxd->wb.upper.vlan); 429 } else { 430 vtag = 0; 431 } 432 433 /* Make sure bad packets are discarded */ 434 if (eop && (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) != 0) { 435 if (sc->feat_en & IXGBE_FEATURE_VF) 436 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); 437 438 rxr->rx_discarded++; 439 return (EBADMSG); 440 } 441 ri->iri_frags[i].irf_flid = 0; 442 ri->iri_frags[i].irf_idx = cidx; 443 ri->iri_frags[i].irf_len = len; 444 if (++cidx == sc->shared->isc_nrxd[0]) 445 cidx = 0; 446 i++; 447 /* even a 16K packet shouldn't consume more than 8 clusters */ 448 MPASS(i < 9); 449 } while (!eop); 450 451 rxr->rx_packets++; 452 rxr->packets++; 453 rxr->rx_bytes += ri->iri_len; 454 455 if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) 456 ixgbe_rx_checksum(staterr, ri, ptype); 457 458 ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss); 459 ri->iri_rsstype = ixgbe_determine_rsstype(pkt_info); 460 if ((sc->feat_en & IXGBE_FEATURE_RSS) == 0) { 461 if (ri->iri_rsstype == M_HASHTYPE_OPAQUE) 462 ri->iri_rsstype = M_HASHTYPE_NONE; 463 else 464 ri->iri_rsstype = M_HASHTYPE_OPAQUE_HASH; 465 } 466 ri->iri_vtag = vtag; 467 ri->iri_nfrags = i; 468 if (vtag) 469 ri->iri_flags |= M_VLANTAG; 470 return (0); 471 } /* ixgbe_isc_rxd_pkt_get */ 472 473 /************************************************************************ 474 * ixgbe_rx_checksum 475 * 476 * Verify that the hardware indicated that the checksum is valid. 477 * Inform the stack about the status of checksum so that stack 478 * doesn't spend time verifying the checksum. 479 ************************************************************************/ 480 static void 481 ixgbe_rx_checksum(uint32_t staterr, if_rxd_info_t ri, uint32_t ptype) 482 { 483 uint16_t status = (uint16_t)staterr; 484 uint8_t errors = (uint8_t)(staterr >> 24); 485 486 /* If there is a layer 3 or 4 error we are done */ 487 if (__predict_false(errors & (IXGBE_RXD_ERR_IPE | IXGBE_RXD_ERR_TCPE))) 488 return; 489 490 /* IP Checksum Good */ 491 if (status & IXGBE_RXD_STAT_IPCS) 492 ri->iri_csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID); 493 494 /* Valid L4E checksum */ 495 if (__predict_true(status & IXGBE_RXD_STAT_L4CS)) { 496 /* SCTP header present. */ 497 if (__predict_false((ptype & IXGBE_RXDADV_PKTTYPE_ETQF) == 0 && 498 (ptype & IXGBE_RXDADV_PKTTYPE_SCTP) != 0)) { 499 ri->iri_csum_flags |= CSUM_SCTP_VALID; 500 } else { 501 ri->iri_csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 502 ri->iri_csum_data = htons(0xffff); 503 } 504 } 505 } /* ixgbe_rx_checksum */ 506 507 /************************************************************************ 508 * ixgbe_determine_rsstype 509 * 510 * Parse the packet type to determine the appropriate hash 511 ************************************************************************/ 512 static int 513 ixgbe_determine_rsstype(uint16_t pkt_info) 514 { 515 switch (pkt_info & IXGBE_RXDADV_RSSTYPE_MASK) { 516 case IXGBE_RXDADV_RSSTYPE_IPV4_TCP: 517 return M_HASHTYPE_RSS_TCP_IPV4; 518 case IXGBE_RXDADV_RSSTYPE_IPV4: 519 return M_HASHTYPE_RSS_IPV4; 520 case IXGBE_RXDADV_RSSTYPE_IPV6_TCP: 521 return M_HASHTYPE_RSS_TCP_IPV6; 522 case IXGBE_RXDADV_RSSTYPE_IPV6_EX: 523 return M_HASHTYPE_RSS_IPV6_EX; 524 case IXGBE_RXDADV_RSSTYPE_IPV6: 525 return M_HASHTYPE_RSS_IPV6; 526 case IXGBE_RXDADV_RSSTYPE_IPV6_TCP_EX: 527 return M_HASHTYPE_RSS_TCP_IPV6_EX; 528 case IXGBE_RXDADV_RSSTYPE_IPV4_UDP: 529 return M_HASHTYPE_RSS_UDP_IPV4; 530 case IXGBE_RXDADV_RSSTYPE_IPV6_UDP: 531 return M_HASHTYPE_RSS_UDP_IPV6; 532 case IXGBE_RXDADV_RSSTYPE_IPV6_UDP_EX: 533 return M_HASHTYPE_RSS_UDP_IPV6_EX; 534 default: 535 return M_HASHTYPE_OPAQUE; 536 } 537 } /* ixgbe_determine_rsstype */ 538