1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (C) 2013 Emulex 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions are met: 9 * 10 * 1. Redistributions of source code must retain the above copyright notice, 11 * this list of conditions and the following disclaimer. 12 * 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * 3. Neither the name of the Emulex Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived from 19 * this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 25 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Contact Information: 34 * freebsd-drivers@emulex.com 35 * 36 * Emulex 37 * 3333 Susan Street 38 * Costa Mesa, CA 92626 39 */ 40 41 /* $FreeBSD$ */ 42 43 #include "opt_inet6.h" 44 #include "opt_inet.h" 45 46 #include "oce_if.h" 47 #include "oce_user.h" 48 49 #define is_tso_pkt(m) (m->m_pkthdr.csum_flags & CSUM_TSO) 50 51 /* UE Status Low CSR */ 52 static char *ue_status_low_desc[] = { 53 "CEV", 54 "CTX", 55 "DBUF", 56 "ERX", 57 "Host", 58 "MPU", 59 "NDMA", 60 "PTC ", 61 "RDMA ", 62 "RXF ", 63 "RXIPS ", 64 "RXULP0 ", 65 "RXULP1 ", 66 "RXULP2 ", 67 "TIM ", 68 "TPOST ", 69 "TPRE ", 70 "TXIPS ", 71 "TXULP0 ", 72 "TXULP1 ", 73 "UC ", 74 "WDMA ", 75 "TXULP2 ", 76 "HOST1 ", 77 "P0_OB_LINK ", 78 "P1_OB_LINK ", 79 "HOST_GPIO ", 80 "MBOX ", 81 "AXGMAC0", 82 "AXGMAC1", 83 "JTAG", 84 "MPU_INTPEND" 85 }; 86 87 /* UE Status High CSR */ 88 static char *ue_status_hi_desc[] = { 89 "LPCMEMHOST", 90 "MGMT_MAC", 91 "PCS0ONLINE", 92 "MPU_IRAM", 93 "PCS1ONLINE", 94 "PCTL0", 95 "PCTL1", 96 "PMEM", 97 "RR", 98 "TXPB", 99 "RXPP", 100 "XAUI", 101 "TXP", 102 "ARM", 103 "IPC", 104 "HOST2", 105 "HOST3", 106 "HOST4", 107 "HOST5", 108 "HOST6", 109 "HOST7", 110 "HOST8", 111 "HOST9", 112 "NETC", 113 "Unknown", 114 "Unknown", 115 "Unknown", 116 "Unknown", 117 "Unknown", 118 "Unknown", 119 "Unknown", 120 "Unknown" 121 }; 122 123 struct oce_common_cqe_info{ 124 uint8_t vtp:1; 125 uint8_t l4_cksum_pass:1; 126 uint8_t ip_cksum_pass:1; 127 uint8_t ipv6_frame:1; 128 uint8_t qnq:1; 129 uint8_t rsvd:3; 130 uint8_t num_frags; 131 uint16_t pkt_size; 132 uint16_t vtag; 133 }; 134 135 /* Driver entry points prototypes */ 136 static int oce_probe(device_t dev); 137 static int oce_attach(device_t dev); 138 static int oce_detach(device_t dev); 139 static int oce_shutdown(device_t dev); 140 static int oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data); 141 static void oce_init(void *xsc); 142 static int oce_multiq_start(struct ifnet *ifp, struct mbuf *m); 143 static void oce_multiq_flush(struct ifnet *ifp); 144 145 /* Driver interrupt routines protypes */ 146 static void oce_intr(void *arg, int pending); 147 static int oce_setup_intr(POCE_SOFTC sc); 148 static int oce_fast_isr(void *arg); 149 static int oce_alloc_intr(POCE_SOFTC sc, int vector, 150 void (*isr) (void *arg, int pending)); 151 152 /* Media callbacks prototypes */ 153 static void oce_media_status(struct ifnet *ifp, struct ifmediareq *req); 154 static int oce_media_change(struct ifnet *ifp); 155 156 /* Transmit routines prototypes */ 157 static int oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index); 158 static void oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq); 159 static void oce_process_tx_completion(struct oce_wq *wq); 160 static int oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, 161 struct oce_wq *wq); 162 163 /* Receive routines prototypes */ 164 static int oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe); 165 static int oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe); 166 static void oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe); 167 static void oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq); 168 static uint16_t oce_rq_handler_lro(void *arg); 169 static void oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2); 170 static void oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2); 171 static void oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m); 172 173 /* Helper function prototypes in this file */ 174 static int oce_attach_ifp(POCE_SOFTC sc); 175 static void oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag); 176 static void oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag); 177 static int oce_vid_config(POCE_SOFTC sc); 178 static void oce_mac_addr_set(POCE_SOFTC sc); 179 static int oce_handle_passthrough(struct ifnet *ifp, caddr_t data); 180 static void oce_local_timer(void *arg); 181 static void oce_if_deactivate(POCE_SOFTC sc); 182 static void oce_if_activate(POCE_SOFTC sc); 183 static void setup_max_queues_want(POCE_SOFTC sc); 184 static void update_queues_got(POCE_SOFTC sc); 185 static void process_link_state(POCE_SOFTC sc, 186 struct oce_async_cqe_link_state *acqe); 187 static int oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m); 188 static void oce_get_config(POCE_SOFTC sc); 189 static struct mbuf *oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete); 190 static void oce_read_env_variables(POCE_SOFTC sc); 191 192 /* IP specific */ 193 #if defined(INET6) || defined(INET) 194 static int oce_init_lro(POCE_SOFTC sc); 195 static struct mbuf * oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp); 196 #endif 197 198 static device_method_t oce_dispatch[] = { 199 DEVMETHOD(device_probe, oce_probe), 200 DEVMETHOD(device_attach, oce_attach), 201 DEVMETHOD(device_detach, oce_detach), 202 DEVMETHOD(device_shutdown, oce_shutdown), 203 204 DEVMETHOD_END 205 }; 206 207 static driver_t oce_driver = { 208 "oce", 209 oce_dispatch, 210 sizeof(OCE_SOFTC) 211 }; 212 static devclass_t oce_devclass; 213 214 /* global vars */ 215 const char component_revision[32] = {"///" COMPONENT_REVISION "///"}; 216 217 /* Module capabilites and parameters */ 218 uint32_t oce_max_rsp_handled = OCE_MAX_RSP_HANDLED; 219 uint32_t oce_enable_rss = OCE_MODCAP_RSS; 220 uint32_t oce_rq_buf_size = 2048; 221 222 TUNABLE_INT("hw.oce.max_rsp_handled", &oce_max_rsp_handled); 223 TUNABLE_INT("hw.oce.enable_rss", &oce_enable_rss); 224 225 /* Supported devices table */ 226 static uint32_t supportedDevices[] = { 227 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE2, 228 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE3, 229 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_BE3, 230 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201, 231 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201_VF, 232 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_SH 233 }; 234 235 DRIVER_MODULE(oce, pci, oce_driver, oce_devclass, 0, 0); 236 MODULE_PNP_INFO("W32:vendor/device", pci, oce, supportedDevices, 237 nitems(supportedDevices)); 238 MODULE_DEPEND(oce, pci, 1, 1, 1); 239 MODULE_DEPEND(oce, ether, 1, 1, 1); 240 MODULE_VERSION(oce, 1); 241 242 POCE_SOFTC softc_head = NULL; 243 POCE_SOFTC softc_tail = NULL; 244 245 struct oce_rdma_if *oce_rdma_if = NULL; 246 247 /***************************************************************************** 248 * Driver entry points functions * 249 *****************************************************************************/ 250 251 static int 252 oce_probe(device_t dev) 253 { 254 uint16_t vendor = 0; 255 uint16_t device = 0; 256 int i = 0; 257 char str[256] = {0}; 258 POCE_SOFTC sc; 259 260 sc = device_get_softc(dev); 261 bzero(sc, sizeof(OCE_SOFTC)); 262 sc->dev = dev; 263 264 vendor = pci_get_vendor(dev); 265 device = pci_get_device(dev); 266 267 for (i = 0; i < (sizeof(supportedDevices) / sizeof(uint32_t)); i++) { 268 if (vendor == ((supportedDevices[i] >> 16) & 0xffff)) { 269 if (device == (supportedDevices[i] & 0xffff)) { 270 sprintf(str, "%s:%s", "Emulex CNA NIC function", 271 component_revision); 272 device_set_desc_copy(dev, str); 273 274 switch (device) { 275 case PCI_PRODUCT_BE2: 276 sc->flags |= OCE_FLAGS_BE2; 277 break; 278 case PCI_PRODUCT_BE3: 279 sc->flags |= OCE_FLAGS_BE3; 280 break; 281 case PCI_PRODUCT_XE201: 282 case PCI_PRODUCT_XE201_VF: 283 sc->flags |= OCE_FLAGS_XE201; 284 break; 285 case PCI_PRODUCT_SH: 286 sc->flags |= OCE_FLAGS_SH; 287 break; 288 default: 289 return ENXIO; 290 } 291 return BUS_PROBE_DEFAULT; 292 } 293 } 294 } 295 296 return ENXIO; 297 } 298 299 static int 300 oce_attach(device_t dev) 301 { 302 POCE_SOFTC sc; 303 int rc = 0; 304 305 sc = device_get_softc(dev); 306 307 rc = oce_hw_pci_alloc(sc); 308 if (rc) 309 return rc; 310 311 sc->tx_ring_size = OCE_TX_RING_SIZE; 312 sc->rx_ring_size = OCE_RX_RING_SIZE; 313 /* receive fragment size should be multiple of 2K */ 314 sc->rq_frag_size = ((oce_rq_buf_size / 2048) * 2048); 315 sc->flow_control = OCE_DEFAULT_FLOW_CONTROL; 316 sc->promisc = OCE_DEFAULT_PROMISCUOUS; 317 318 LOCK_CREATE(&sc->bmbx_lock, "Mailbox_lock"); 319 LOCK_CREATE(&sc->dev_lock, "Device_lock"); 320 321 /* initialise the hardware */ 322 rc = oce_hw_init(sc); 323 if (rc) 324 goto pci_res_free; 325 326 oce_read_env_variables(sc); 327 328 oce_get_config(sc); 329 330 setup_max_queues_want(sc); 331 332 rc = oce_setup_intr(sc); 333 if (rc) 334 goto mbox_free; 335 336 rc = oce_queue_init_all(sc); 337 if (rc) 338 goto intr_free; 339 340 rc = oce_attach_ifp(sc); 341 if (rc) 342 goto queues_free; 343 344 #if defined(INET6) || defined(INET) 345 rc = oce_init_lro(sc); 346 if (rc) 347 goto ifp_free; 348 #endif 349 350 rc = oce_hw_start(sc); 351 if (rc) 352 goto lro_free; 353 354 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 355 oce_add_vlan, sc, EVENTHANDLER_PRI_FIRST); 356 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 357 oce_del_vlan, sc, EVENTHANDLER_PRI_FIRST); 358 359 rc = oce_stats_init(sc); 360 if (rc) 361 goto vlan_free; 362 363 oce_add_sysctls(sc); 364 365 callout_init(&sc->timer, CALLOUT_MPSAFE); 366 rc = callout_reset(&sc->timer, 2 * hz, oce_local_timer, sc); 367 if (rc) 368 goto stats_free; 369 370 sc->next =NULL; 371 if (softc_tail != NULL) { 372 softc_tail->next = sc; 373 } else { 374 softc_head = sc; 375 } 376 softc_tail = sc; 377 378 return 0; 379 380 stats_free: 381 callout_drain(&sc->timer); 382 oce_stats_free(sc); 383 vlan_free: 384 if (sc->vlan_attach) 385 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 386 if (sc->vlan_detach) 387 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 388 oce_hw_intr_disable(sc); 389 lro_free: 390 #if defined(INET6) || defined(INET) 391 oce_free_lro(sc); 392 ifp_free: 393 #endif 394 ether_ifdetach(sc->ifp); 395 if_free(sc->ifp); 396 queues_free: 397 oce_queue_release_all(sc); 398 intr_free: 399 oce_intr_free(sc); 400 mbox_free: 401 oce_dma_free(sc, &sc->bsmbx); 402 pci_res_free: 403 oce_hw_pci_free(sc); 404 LOCK_DESTROY(&sc->dev_lock); 405 LOCK_DESTROY(&sc->bmbx_lock); 406 return rc; 407 408 } 409 410 static int 411 oce_detach(device_t dev) 412 { 413 POCE_SOFTC sc = device_get_softc(dev); 414 POCE_SOFTC poce_sc_tmp, *ppoce_sc_tmp1, poce_sc_tmp2 = NULL; 415 416 poce_sc_tmp = softc_head; 417 ppoce_sc_tmp1 = &softc_head; 418 while (poce_sc_tmp != NULL) { 419 if (poce_sc_tmp == sc) { 420 *ppoce_sc_tmp1 = sc->next; 421 if (sc->next == NULL) { 422 softc_tail = poce_sc_tmp2; 423 } 424 break; 425 } 426 poce_sc_tmp2 = poce_sc_tmp; 427 ppoce_sc_tmp1 = &poce_sc_tmp->next; 428 poce_sc_tmp = poce_sc_tmp->next; 429 } 430 431 LOCK(&sc->dev_lock); 432 oce_if_deactivate(sc); 433 UNLOCK(&sc->dev_lock); 434 435 callout_drain(&sc->timer); 436 437 if (sc->vlan_attach != NULL) 438 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 439 if (sc->vlan_detach != NULL) 440 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 441 442 ether_ifdetach(sc->ifp); 443 444 if_free(sc->ifp); 445 446 oce_hw_shutdown(sc); 447 448 bus_generic_detach(dev); 449 450 return 0; 451 } 452 453 static int 454 oce_shutdown(device_t dev) 455 { 456 int rc; 457 458 rc = oce_detach(dev); 459 460 return rc; 461 } 462 463 static int 464 oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data) 465 { 466 struct ifreq *ifr = (struct ifreq *)data; 467 POCE_SOFTC sc = ifp->if_softc; 468 struct ifi2creq i2c; 469 uint8_t offset = 0; 470 int rc = 0; 471 uint32_t u; 472 473 switch (command) { 474 case SIOCGIFMEDIA: 475 rc = ifmedia_ioctl(ifp, ifr, &sc->media, command); 476 break; 477 478 case SIOCSIFMTU: 479 if (ifr->ifr_mtu > OCE_MAX_MTU) 480 rc = EINVAL; 481 else 482 ifp->if_mtu = ifr->ifr_mtu; 483 break; 484 485 case SIOCSIFFLAGS: 486 if (ifp->if_flags & IFF_UP) { 487 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 488 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING; 489 oce_init(sc); 490 } 491 device_printf(sc->dev, "Interface Up\n"); 492 } else { 493 LOCK(&sc->dev_lock); 494 495 sc->ifp->if_drv_flags &= 496 ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 497 oce_if_deactivate(sc); 498 499 UNLOCK(&sc->dev_lock); 500 501 device_printf(sc->dev, "Interface Down\n"); 502 } 503 504 if ((ifp->if_flags & IFF_PROMISC) && !sc->promisc) { 505 if (!oce_rxf_set_promiscuous(sc, (1 | (1 << 1)))) 506 sc->promisc = TRUE; 507 } else if (!(ifp->if_flags & IFF_PROMISC) && sc->promisc) { 508 if (!oce_rxf_set_promiscuous(sc, 0)) 509 sc->promisc = FALSE; 510 } 511 512 break; 513 514 case SIOCADDMULTI: 515 case SIOCDELMULTI: 516 rc = oce_hw_update_multicast(sc); 517 if (rc) 518 device_printf(sc->dev, 519 "Update multicast address failed\n"); 520 break; 521 522 case SIOCSIFCAP: 523 u = ifr->ifr_reqcap ^ ifp->if_capenable; 524 525 if (u & IFCAP_TXCSUM) { 526 ifp->if_capenable ^= IFCAP_TXCSUM; 527 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP); 528 529 if (IFCAP_TSO & ifp->if_capenable && 530 !(IFCAP_TXCSUM & ifp->if_capenable)) { 531 u &= ~IFCAP_TSO; 532 ifp->if_capenable &= ~IFCAP_TSO; 533 ifp->if_hwassist &= ~CSUM_TSO; 534 if_printf(ifp, 535 "TSO disabled due to -txcsum.\n"); 536 } 537 } 538 539 if (u & IFCAP_RXCSUM) 540 ifp->if_capenable ^= IFCAP_RXCSUM; 541 542 if (u & IFCAP_TSO4) { 543 ifp->if_capenable ^= IFCAP_TSO4; 544 545 if (IFCAP_TSO & ifp->if_capenable) { 546 if (IFCAP_TXCSUM & ifp->if_capenable) 547 ifp->if_hwassist |= CSUM_TSO; 548 else { 549 ifp->if_capenable &= ~IFCAP_TSO; 550 ifp->if_hwassist &= ~CSUM_TSO; 551 if_printf(ifp, 552 "Enable txcsum first.\n"); 553 rc = EAGAIN; 554 } 555 } else 556 ifp->if_hwassist &= ~CSUM_TSO; 557 } 558 559 if (u & IFCAP_VLAN_HWTAGGING) 560 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 561 562 if (u & IFCAP_VLAN_HWFILTER) { 563 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 564 oce_vid_config(sc); 565 } 566 #if defined(INET6) || defined(INET) 567 if (u & IFCAP_LRO) { 568 ifp->if_capenable ^= IFCAP_LRO; 569 if(sc->enable_hwlro) { 570 if(ifp->if_capenable & IFCAP_LRO) { 571 rc = oce_mbox_nic_set_iface_lro_config(sc, 1); 572 }else { 573 rc = oce_mbox_nic_set_iface_lro_config(sc, 0); 574 } 575 } 576 } 577 #endif 578 579 break; 580 581 case SIOCGI2C: 582 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c)); 583 if (rc) 584 break; 585 586 if (i2c.dev_addr == PAGE_NUM_A0) { 587 offset = i2c.offset; 588 } else if (i2c.dev_addr == PAGE_NUM_A2) { 589 offset = TRANSCEIVER_A0_SIZE + i2c.offset; 590 } else { 591 rc = EINVAL; 592 break; 593 } 594 595 if (i2c.len > sizeof(i2c.data) || 596 i2c.len + offset > sizeof(sfp_vpd_dump_buffer)) { 597 rc = EINVAL; 598 break; 599 } 600 601 rc = oce_mbox_read_transrecv_data(sc, i2c.dev_addr); 602 if (rc) { 603 rc = -rc; 604 break; 605 } 606 607 memcpy(&i2c.data[0], &sfp_vpd_dump_buffer[offset], i2c.len); 608 609 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c)); 610 break; 611 612 case SIOCGPRIVATE_0: 613 rc = priv_check(curthread, PRIV_DRIVER); 614 if (rc != 0) 615 break; 616 rc = oce_handle_passthrough(ifp, data); 617 break; 618 default: 619 rc = ether_ioctl(ifp, command, data); 620 break; 621 } 622 623 return rc; 624 } 625 626 static void 627 oce_init(void *arg) 628 { 629 POCE_SOFTC sc = arg; 630 631 LOCK(&sc->dev_lock); 632 633 if (sc->ifp->if_flags & IFF_UP) { 634 oce_if_deactivate(sc); 635 oce_if_activate(sc); 636 } 637 638 UNLOCK(&sc->dev_lock); 639 640 } 641 642 static int 643 oce_multiq_start(struct ifnet *ifp, struct mbuf *m) 644 { 645 POCE_SOFTC sc = ifp->if_softc; 646 struct oce_wq *wq = NULL; 647 int queue_index = 0; 648 int status = 0; 649 650 if (!sc->link_status) 651 return ENXIO; 652 653 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 654 queue_index = m->m_pkthdr.flowid % sc->nwqs; 655 656 wq = sc->wq[queue_index]; 657 658 LOCK(&wq->tx_lock); 659 status = oce_multiq_transmit(ifp, m, wq); 660 UNLOCK(&wq->tx_lock); 661 662 return status; 663 664 } 665 666 static void 667 oce_multiq_flush(struct ifnet *ifp) 668 { 669 POCE_SOFTC sc = ifp->if_softc; 670 struct mbuf *m; 671 int i = 0; 672 673 for (i = 0; i < sc->nwqs; i++) { 674 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL) 675 m_freem(m); 676 } 677 if_qflush(ifp); 678 } 679 680 /***************************************************************************** 681 * Driver interrupt routines functions * 682 *****************************************************************************/ 683 684 static void 685 oce_intr(void *arg, int pending) 686 { 687 688 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg; 689 POCE_SOFTC sc = ii->sc; 690 struct oce_eq *eq = ii->eq; 691 struct oce_eqe *eqe; 692 struct oce_cq *cq = NULL; 693 int i, num_eqes = 0; 694 695 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map, 696 BUS_DMASYNC_POSTWRITE); 697 do { 698 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe); 699 if (eqe->evnt == 0) 700 break; 701 eqe->evnt = 0; 702 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map, 703 BUS_DMASYNC_POSTWRITE); 704 RING_GET(eq->ring, 1); 705 num_eqes++; 706 707 } while (TRUE); 708 709 if (!num_eqes) 710 goto eq_arm; /* Spurious */ 711 712 /* Clear EQ entries, but dont arm */ 713 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE); 714 715 /* Process TX, RX and MCC. But dont arm CQ*/ 716 for (i = 0; i < eq->cq_valid; i++) { 717 cq = eq->cq[i]; 718 (*cq->cq_handler)(cq->cb_arg); 719 } 720 721 /* Arm all cqs connected to this EQ */ 722 for (i = 0; i < eq->cq_valid; i++) { 723 cq = eq->cq[i]; 724 oce_arm_cq(sc, cq->cq_id, 0, TRUE); 725 } 726 727 eq_arm: 728 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE); 729 730 return; 731 } 732 733 static int 734 oce_setup_intr(POCE_SOFTC sc) 735 { 736 int rc = 0, use_intx = 0; 737 int vector = 0, req_vectors = 0; 738 int tot_req_vectors, tot_vectors; 739 740 if (is_rss_enabled(sc)) 741 req_vectors = MAX((sc->nrqs - 1), sc->nwqs); 742 else 743 req_vectors = 1; 744 745 tot_req_vectors = req_vectors; 746 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) { 747 if (req_vectors > 1) { 748 tot_req_vectors += OCE_RDMA_VECTORS; 749 sc->roce_intr_count = OCE_RDMA_VECTORS; 750 } 751 } 752 753 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) { 754 sc->intr_count = req_vectors; 755 tot_vectors = tot_req_vectors; 756 rc = pci_alloc_msix(sc->dev, &tot_vectors); 757 if (rc != 0) { 758 use_intx = 1; 759 pci_release_msi(sc->dev); 760 } else { 761 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) { 762 if (tot_vectors < tot_req_vectors) { 763 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) { 764 sc->roce_intr_count = (tot_vectors / 2); 765 } 766 sc->intr_count = tot_vectors - sc->roce_intr_count; 767 } 768 } else { 769 sc->intr_count = tot_vectors; 770 } 771 sc->flags |= OCE_FLAGS_USING_MSIX; 772 } 773 } else 774 use_intx = 1; 775 776 if (use_intx) 777 sc->intr_count = 1; 778 779 /* Scale number of queues based on intr we got */ 780 update_queues_got(sc); 781 782 if (use_intx) { 783 device_printf(sc->dev, "Using legacy interrupt\n"); 784 rc = oce_alloc_intr(sc, vector, oce_intr); 785 if (rc) 786 goto error; 787 } else { 788 for (; vector < sc->intr_count; vector++) { 789 rc = oce_alloc_intr(sc, vector, oce_intr); 790 if (rc) 791 goto error; 792 } 793 } 794 795 return 0; 796 error: 797 oce_intr_free(sc); 798 return rc; 799 } 800 801 static int 802 oce_fast_isr(void *arg) 803 { 804 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg; 805 POCE_SOFTC sc = ii->sc; 806 807 if (ii->eq == NULL) 808 return FILTER_STRAY; 809 810 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE); 811 812 taskqueue_enqueue(ii->tq, &ii->task); 813 814 ii->eq->intr++; 815 816 return FILTER_HANDLED; 817 } 818 819 static int 820 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending)) 821 { 822 POCE_INTR_INFO ii; 823 int rc = 0, rr; 824 825 if (vector >= OCE_MAX_EQ) 826 return (EINVAL); 827 828 ii = &sc->intrs[vector]; 829 830 /* Set the resource id for the interrupt. 831 * MSIx is vector + 1 for the resource id, 832 * INTx is 0 for the resource id. 833 */ 834 if (sc->flags & OCE_FLAGS_USING_MSIX) 835 rr = vector + 1; 836 else 837 rr = 0; 838 ii->intr_res = bus_alloc_resource_any(sc->dev, 839 SYS_RES_IRQ, 840 &rr, RF_ACTIVE|RF_SHAREABLE); 841 ii->irq_rr = rr; 842 if (ii->intr_res == NULL) { 843 device_printf(sc->dev, 844 "Could not allocate interrupt\n"); 845 rc = ENXIO; 846 return rc; 847 } 848 849 TASK_INIT(&ii->task, 0, isr, ii); 850 ii->vector = vector; 851 sprintf(ii->task_name, "oce_task[%d]", ii->vector); 852 ii->tq = taskqueue_create_fast(ii->task_name, 853 M_NOWAIT, 854 taskqueue_thread_enqueue, 855 &ii->tq); 856 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq", 857 device_get_nameunit(sc->dev)); 858 859 ii->sc = sc; 860 rc = bus_setup_intr(sc->dev, 861 ii->intr_res, 862 INTR_TYPE_NET, 863 oce_fast_isr, NULL, ii, &ii->tag); 864 return rc; 865 866 } 867 868 void 869 oce_intr_free(POCE_SOFTC sc) 870 { 871 int i = 0; 872 873 for (i = 0; i < sc->intr_count; i++) { 874 875 if (sc->intrs[i].tag != NULL) 876 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res, 877 sc->intrs[i].tag); 878 if (sc->intrs[i].tq != NULL) 879 taskqueue_free(sc->intrs[i].tq); 880 881 if (sc->intrs[i].intr_res != NULL) 882 bus_release_resource(sc->dev, SYS_RES_IRQ, 883 sc->intrs[i].irq_rr, 884 sc->intrs[i].intr_res); 885 sc->intrs[i].tag = NULL; 886 sc->intrs[i].intr_res = NULL; 887 } 888 889 if (sc->flags & OCE_FLAGS_USING_MSIX) 890 pci_release_msi(sc->dev); 891 892 } 893 894 /****************************************************************************** 895 * Media callbacks functions * 896 ******************************************************************************/ 897 898 static void 899 oce_media_status(struct ifnet *ifp, struct ifmediareq *req) 900 { 901 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc; 902 903 req->ifm_status = IFM_AVALID; 904 req->ifm_active = IFM_ETHER; 905 906 if (sc->link_status == 1) 907 req->ifm_status |= IFM_ACTIVE; 908 else 909 return; 910 911 switch (sc->link_speed) { 912 case 1: /* 10 Mbps */ 913 req->ifm_active |= IFM_10_T | IFM_FDX; 914 sc->speed = 10; 915 break; 916 case 2: /* 100 Mbps */ 917 req->ifm_active |= IFM_100_TX | IFM_FDX; 918 sc->speed = 100; 919 break; 920 case 3: /* 1 Gbps */ 921 req->ifm_active |= IFM_1000_T | IFM_FDX; 922 sc->speed = 1000; 923 break; 924 case 4: /* 10 Gbps */ 925 req->ifm_active |= IFM_10G_SR | IFM_FDX; 926 sc->speed = 10000; 927 break; 928 case 5: /* 20 Gbps */ 929 req->ifm_active |= IFM_10G_SR | IFM_FDX; 930 sc->speed = 20000; 931 break; 932 case 6: /* 25 Gbps */ 933 req->ifm_active |= IFM_10G_SR | IFM_FDX; 934 sc->speed = 25000; 935 break; 936 case 7: /* 40 Gbps */ 937 req->ifm_active |= IFM_40G_SR4 | IFM_FDX; 938 sc->speed = 40000; 939 break; 940 default: 941 sc->speed = 0; 942 break; 943 } 944 945 return; 946 } 947 948 int 949 oce_media_change(struct ifnet *ifp) 950 { 951 return 0; 952 } 953 954 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc, 955 struct mbuf *m, boolean_t *os2bmc, 956 struct mbuf **m_new) 957 { 958 struct ether_header *eh = NULL; 959 960 eh = mtod(m, struct ether_header *); 961 962 if (!is_os2bmc_enabled(sc) || *os2bmc) { 963 *os2bmc = FALSE; 964 goto done; 965 } 966 if (!ETHER_IS_MULTICAST(eh->ether_dhost)) 967 goto done; 968 969 if (is_mc_allowed_on_bmc(sc, eh) || 970 is_bc_allowed_on_bmc(sc, eh) || 971 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) { 972 *os2bmc = TRUE; 973 goto done; 974 } 975 976 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) { 977 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 978 uint8_t nexthdr = ip6->ip6_nxt; 979 if (nexthdr == IPPROTO_ICMPV6) { 980 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1); 981 switch (icmp6->icmp6_type) { 982 case ND_ROUTER_ADVERT: 983 *os2bmc = is_ipv6_ra_filt_enabled(sc); 984 goto done; 985 case ND_NEIGHBOR_ADVERT: 986 *os2bmc = is_ipv6_na_filt_enabled(sc); 987 goto done; 988 default: 989 break; 990 } 991 } 992 } 993 994 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) { 995 struct ip *ip = mtod(m, struct ip *); 996 int iphlen = ip->ip_hl << 2; 997 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen); 998 switch (uh->uh_dport) { 999 case DHCP_CLIENT_PORT: 1000 *os2bmc = is_dhcp_client_filt_enabled(sc); 1001 goto done; 1002 case DHCP_SERVER_PORT: 1003 *os2bmc = is_dhcp_srvr_filt_enabled(sc); 1004 goto done; 1005 case NET_BIOS_PORT1: 1006 case NET_BIOS_PORT2: 1007 *os2bmc = is_nbios_filt_enabled(sc); 1008 goto done; 1009 case DHCPV6_RAS_PORT: 1010 *os2bmc = is_ipv6_ras_filt_enabled(sc); 1011 goto done; 1012 default: 1013 break; 1014 } 1015 } 1016 done: 1017 if (*os2bmc) { 1018 *m_new = m_dup(m, M_NOWAIT); 1019 if (!*m_new) { 1020 *os2bmc = FALSE; 1021 return; 1022 } 1023 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL); 1024 } 1025 } 1026 1027 /***************************************************************************** 1028 * Transmit routines functions * 1029 *****************************************************************************/ 1030 1031 static int 1032 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index) 1033 { 1034 int rc = 0, i, retry_cnt = 0; 1035 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS]; 1036 struct mbuf *m, *m_temp, *m_new = NULL; 1037 struct oce_wq *wq = sc->wq[wq_index]; 1038 struct oce_packet_desc *pd; 1039 struct oce_nic_hdr_wqe *nichdr; 1040 struct oce_nic_frag_wqe *nicfrag; 1041 struct ether_header *eh = NULL; 1042 int num_wqes; 1043 uint32_t reg_value; 1044 boolean_t complete = TRUE; 1045 boolean_t os2bmc = FALSE; 1046 1047 m = *mpp; 1048 if (!m) 1049 return EINVAL; 1050 1051 if (!(m->m_flags & M_PKTHDR)) { 1052 rc = ENXIO; 1053 goto free_ret; 1054 } 1055 1056 /* Don't allow non-TSO packets longer than MTU */ 1057 if (!is_tso_pkt(m)) { 1058 eh = mtod(m, struct ether_header *); 1059 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE)) 1060 goto free_ret; 1061 } 1062 1063 if(oce_tx_asic_stall_verify(sc, m)) { 1064 m = oce_insert_vlan_tag(sc, m, &complete); 1065 if(!m) { 1066 device_printf(sc->dev, "Insertion unsuccessful\n"); 1067 return 0; 1068 } 1069 } 1070 1071 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less 1072 * may cause a transmit stall on that port. So the work-around is to 1073 * pad short packets (<= 32 bytes) to a 36-byte length. 1074 */ 1075 if(IS_SH(sc) || IS_XE201(sc) ) { 1076 if(m->m_pkthdr.len <= 32) { 1077 char buf[36]; 1078 bzero((void *)buf, 36); 1079 m_append(m, (36 - m->m_pkthdr.len), buf); 1080 } 1081 } 1082 1083 tx_start: 1084 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1085 /* consolidate packet buffers for TSO/LSO segment offload */ 1086 #if defined(INET6) || defined(INET) 1087 m = oce_tso_setup(sc, mpp); 1088 #else 1089 m = NULL; 1090 #endif 1091 if (m == NULL) { 1092 rc = ENXIO; 1093 goto free_ret; 1094 } 1095 } 1096 1097 pd = &wq->pckts[wq->pkt_desc_head]; 1098 1099 retry: 1100 rc = bus_dmamap_load_mbuf_sg(wq->tag, 1101 pd->map, 1102 m, segs, &pd->nsegs, BUS_DMA_NOWAIT); 1103 if (rc == 0) { 1104 num_wqes = pd->nsegs + 1; 1105 if (IS_BE(sc) || IS_SH(sc)) { 1106 /*Dummy required only for BE3.*/ 1107 if (num_wqes & 1) 1108 num_wqes++; 1109 } 1110 if (num_wqes >= RING_NUM_FREE(wq->ring)) { 1111 bus_dmamap_unload(wq->tag, pd->map); 1112 return EBUSY; 1113 } 1114 atomic_store_rel_int(&wq->pkt_desc_head, 1115 (wq->pkt_desc_head + 1) % \ 1116 OCE_WQ_PACKET_ARRAY_SIZE); 1117 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE); 1118 pd->mbuf = m; 1119 1120 nichdr = 1121 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe); 1122 nichdr->u0.dw[0] = 0; 1123 nichdr->u0.dw[1] = 0; 1124 nichdr->u0.dw[2] = 0; 1125 nichdr->u0.dw[3] = 0; 1126 1127 nichdr->u0.s.complete = complete; 1128 nichdr->u0.s.mgmt = os2bmc; 1129 nichdr->u0.s.event = 1; 1130 nichdr->u0.s.crc = 1; 1131 nichdr->u0.s.forward = 0; 1132 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0; 1133 nichdr->u0.s.udpcs = 1134 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0; 1135 nichdr->u0.s.tcpcs = 1136 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0; 1137 nichdr->u0.s.num_wqe = num_wqes; 1138 nichdr->u0.s.total_length = m->m_pkthdr.len; 1139 1140 if (m->m_flags & M_VLANTAG) { 1141 nichdr->u0.s.vlan = 1; /*Vlan present*/ 1142 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag; 1143 } 1144 1145 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1146 if (m->m_pkthdr.tso_segsz) { 1147 nichdr->u0.s.lso = 1; 1148 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz; 1149 } 1150 if (!IS_BE(sc) || !IS_SH(sc)) 1151 nichdr->u0.s.ipcs = 1; 1152 } 1153 1154 RING_PUT(wq->ring, 1); 1155 atomic_add_int(&wq->ring->num_used, 1); 1156 1157 for (i = 0; i < pd->nsegs; i++) { 1158 nicfrag = 1159 RING_GET_PRODUCER_ITEM_VA(wq->ring, 1160 struct oce_nic_frag_wqe); 1161 nicfrag->u0.s.rsvd0 = 0; 1162 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr); 1163 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr); 1164 nicfrag->u0.s.frag_len = segs[i].ds_len; 1165 pd->wqe_idx = wq->ring->pidx; 1166 RING_PUT(wq->ring, 1); 1167 atomic_add_int(&wq->ring->num_used, 1); 1168 } 1169 if (num_wqes > (pd->nsegs + 1)) { 1170 nicfrag = 1171 RING_GET_PRODUCER_ITEM_VA(wq->ring, 1172 struct oce_nic_frag_wqe); 1173 nicfrag->u0.dw[0] = 0; 1174 nicfrag->u0.dw[1] = 0; 1175 nicfrag->u0.dw[2] = 0; 1176 nicfrag->u0.dw[3] = 0; 1177 pd->wqe_idx = wq->ring->pidx; 1178 RING_PUT(wq->ring, 1); 1179 atomic_add_int(&wq->ring->num_used, 1); 1180 pd->nsegs++; 1181 } 1182 1183 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1); 1184 wq->tx_stats.tx_reqs++; 1185 wq->tx_stats.tx_wrbs += num_wqes; 1186 wq->tx_stats.tx_bytes += m->m_pkthdr.len; 1187 wq->tx_stats.tx_pkts++; 1188 1189 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map, 1190 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); 1191 reg_value = (num_wqes << 16) | wq->wq_id; 1192 1193 /* if os2bmc is not enabled or if the pkt is already tagged as 1194 bmc, do nothing 1195 */ 1196 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new); 1197 1198 if_inc_counter(sc->ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); 1199 if (m->m_flags & M_MCAST) 1200 if_inc_counter(sc->ifp, IFCOUNTER_OMCASTS, 1); 1201 ETHER_BPF_MTAP(sc->ifp, m); 1202 1203 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value); 1204 1205 } else if (rc == EFBIG) { 1206 if (retry_cnt == 0) { 1207 m_temp = m_defrag(m, M_NOWAIT); 1208 if (m_temp == NULL) 1209 goto free_ret; 1210 m = m_temp; 1211 *mpp = m_temp; 1212 retry_cnt = retry_cnt + 1; 1213 goto retry; 1214 } else 1215 goto free_ret; 1216 } else if (rc == ENOMEM) 1217 return rc; 1218 else 1219 goto free_ret; 1220 1221 if (os2bmc) { 1222 m = m_new; 1223 goto tx_start; 1224 } 1225 1226 return 0; 1227 1228 free_ret: 1229 m_freem(*mpp); 1230 *mpp = NULL; 1231 return rc; 1232 } 1233 1234 static void 1235 oce_process_tx_completion(struct oce_wq *wq) 1236 { 1237 struct oce_packet_desc *pd; 1238 POCE_SOFTC sc = (POCE_SOFTC) wq->parent; 1239 struct mbuf *m; 1240 1241 pd = &wq->pckts[wq->pkt_desc_tail]; 1242 atomic_store_rel_int(&wq->pkt_desc_tail, 1243 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE); 1244 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1); 1245 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE); 1246 bus_dmamap_unload(wq->tag, pd->map); 1247 1248 m = pd->mbuf; 1249 m_freem(m); 1250 pd->mbuf = NULL; 1251 1252 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) { 1253 if (wq->ring->num_used < (wq->ring->num_items / 2)) { 1254 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE); 1255 oce_tx_restart(sc, wq); 1256 } 1257 } 1258 } 1259 1260 static void 1261 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq) 1262 { 1263 1264 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING) 1265 return; 1266 1267 if (!drbr_empty(sc->ifp, wq->br)) 1268 taskqueue_enqueue(taskqueue_swi, &wq->txtask); 1269 1270 } 1271 1272 #if defined(INET6) || defined(INET) 1273 static struct mbuf * 1274 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp) 1275 { 1276 struct mbuf *m; 1277 #ifdef INET 1278 struct ip *ip; 1279 #endif 1280 #ifdef INET6 1281 struct ip6_hdr *ip6; 1282 #endif 1283 struct ether_vlan_header *eh; 1284 struct tcphdr *th; 1285 uint16_t etype; 1286 int total_len = 0, ehdrlen = 0; 1287 1288 m = *mpp; 1289 1290 if (M_WRITABLE(m) == 0) { 1291 m = m_dup(*mpp, M_NOWAIT); 1292 if (!m) 1293 return NULL; 1294 m_freem(*mpp); 1295 *mpp = m; 1296 } 1297 1298 eh = mtod(m, struct ether_vlan_header *); 1299 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 1300 etype = ntohs(eh->evl_proto); 1301 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; 1302 } else { 1303 etype = ntohs(eh->evl_encap_proto); 1304 ehdrlen = ETHER_HDR_LEN; 1305 } 1306 1307 switch (etype) { 1308 #ifdef INET 1309 case ETHERTYPE_IP: 1310 ip = (struct ip *)(m->m_data + ehdrlen); 1311 if (ip->ip_p != IPPROTO_TCP) 1312 return NULL; 1313 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1314 1315 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2); 1316 break; 1317 #endif 1318 #ifdef INET6 1319 case ETHERTYPE_IPV6: 1320 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen); 1321 if (ip6->ip6_nxt != IPPROTO_TCP) 1322 return NULL; 1323 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr)); 1324 1325 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2); 1326 break; 1327 #endif 1328 default: 1329 return NULL; 1330 } 1331 1332 m = m_pullup(m, total_len); 1333 *mpp = m; 1334 return m; 1335 } 1336 #endif /* INET6 || INET */ 1337 1338 void 1339 oce_tx_task(void *arg, int npending) 1340 { 1341 struct oce_wq *wq = arg; 1342 POCE_SOFTC sc = wq->parent; 1343 struct ifnet *ifp = sc->ifp; 1344 int rc = 0; 1345 1346 LOCK(&wq->tx_lock); 1347 rc = oce_multiq_transmit(ifp, NULL, wq); 1348 if (rc) { 1349 device_printf(sc->dev, 1350 "TX[%d] restart failed\n", wq->queue_index); 1351 } 1352 UNLOCK(&wq->tx_lock); 1353 } 1354 1355 void 1356 oce_start(struct ifnet *ifp) 1357 { 1358 POCE_SOFTC sc = ifp->if_softc; 1359 struct mbuf *m; 1360 int rc = 0; 1361 int def_q = 0; /* Defualt tx queue is 0*/ 1362 1363 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 1364 IFF_DRV_RUNNING) 1365 return; 1366 1367 if (!sc->link_status) 1368 return; 1369 1370 while (true) { 1371 IF_DEQUEUE(&sc->ifp->if_snd, m); 1372 if (m == NULL) 1373 break; 1374 1375 LOCK(&sc->wq[def_q]->tx_lock); 1376 rc = oce_tx(sc, &m, def_q); 1377 UNLOCK(&sc->wq[def_q]->tx_lock); 1378 if (rc) { 1379 if (m != NULL) { 1380 sc->wq[def_q]->tx_stats.tx_stops ++; 1381 ifp->if_drv_flags |= IFF_DRV_OACTIVE; 1382 IFQ_DRV_PREPEND(&ifp->if_snd, m); 1383 m = NULL; 1384 } 1385 break; 1386 } 1387 } 1388 } 1389 1390 /* Handle the Completion Queue for transmit */ 1391 uint16_t 1392 oce_wq_handler(void *arg) 1393 { 1394 struct oce_wq *wq = (struct oce_wq *)arg; 1395 POCE_SOFTC sc = wq->parent; 1396 struct oce_cq *cq = wq->cq; 1397 struct oce_nic_tx_cqe *cqe; 1398 int num_cqes = 0; 1399 1400 LOCK(&wq->tx_compl_lock); 1401 bus_dmamap_sync(cq->ring->dma.tag, 1402 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 1403 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe); 1404 while (cqe->u0.dw[3]) { 1405 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe)); 1406 1407 wq->ring->cidx = cqe->u0.s.wqe_index + 1; 1408 if (wq->ring->cidx >= wq->ring->num_items) 1409 wq->ring->cidx -= wq->ring->num_items; 1410 1411 oce_process_tx_completion(wq); 1412 wq->tx_stats.tx_compl++; 1413 cqe->u0.dw[3] = 0; 1414 RING_GET(cq->ring, 1); 1415 bus_dmamap_sync(cq->ring->dma.tag, 1416 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 1417 cqe = 1418 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe); 1419 num_cqes++; 1420 } 1421 1422 if (num_cqes) 1423 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE); 1424 1425 UNLOCK(&wq->tx_compl_lock); 1426 return num_cqes; 1427 } 1428 1429 static int 1430 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq) 1431 { 1432 POCE_SOFTC sc = ifp->if_softc; 1433 int status = 0, queue_index = 0; 1434 struct mbuf *next = NULL; 1435 struct buf_ring *br = NULL; 1436 1437 br = wq->br; 1438 queue_index = wq->queue_index; 1439 1440 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 1441 IFF_DRV_RUNNING) { 1442 if (m != NULL) 1443 status = drbr_enqueue(ifp, br, m); 1444 return status; 1445 } 1446 1447 if (m != NULL) { 1448 if ((status = drbr_enqueue(ifp, br, m)) != 0) 1449 return status; 1450 } 1451 while ((next = drbr_peek(ifp, br)) != NULL) { 1452 if (oce_tx(sc, &next, queue_index)) { 1453 if (next == NULL) { 1454 drbr_advance(ifp, br); 1455 } else { 1456 drbr_putback(ifp, br, next); 1457 wq->tx_stats.tx_stops ++; 1458 ifp->if_drv_flags |= IFF_DRV_OACTIVE; 1459 } 1460 break; 1461 } 1462 drbr_advance(ifp, br); 1463 } 1464 1465 return 0; 1466 } 1467 1468 /***************************************************************************** 1469 * Receive routines functions * 1470 *****************************************************************************/ 1471 1472 static void 1473 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2) 1474 { 1475 uint32_t *p; 1476 struct ether_header *eh = NULL; 1477 struct tcphdr *tcp_hdr = NULL; 1478 struct ip *ip4_hdr = NULL; 1479 struct ip6_hdr *ip6 = NULL; 1480 uint32_t payload_len = 0; 1481 1482 eh = mtod(m, struct ether_header *); 1483 /* correct IP header */ 1484 if(!cqe2->ipv6_frame) { 1485 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header)); 1486 ip4_hdr->ip_ttl = cqe2->frame_lifespan; 1487 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header)); 1488 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip)); 1489 }else { 1490 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header)); 1491 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan; 1492 payload_len = cqe2->coalesced_size - sizeof(struct ether_header) 1493 - sizeof(struct ip6_hdr); 1494 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len); 1495 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr)); 1496 } 1497 1498 /* correct tcp header */ 1499 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num); 1500 if(cqe2->push) { 1501 tcp_hdr->th_flags |= TH_PUSH; 1502 } 1503 tcp_hdr->th_win = htons(cqe2->tcp_window); 1504 tcp_hdr->th_sum = 0xffff; 1505 if(cqe2->ts_opt) { 1506 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2); 1507 *p = cqe1->tcp_timestamp_val; 1508 *(p+1) = cqe1->tcp_timestamp_ecr; 1509 } 1510 1511 return; 1512 } 1513 1514 static void 1515 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m) 1516 { 1517 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1518 uint32_t i = 0, frag_len = 0; 1519 uint32_t len = cqe_info->pkt_size; 1520 struct oce_packet_desc *pd; 1521 struct mbuf *tail = NULL; 1522 1523 for (i = 0; i < cqe_info->num_frags; i++) { 1524 if (rq->ring->cidx == rq->ring->pidx) { 1525 device_printf(sc->dev, 1526 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n"); 1527 return; 1528 } 1529 pd = &rq->pckts[rq->ring->cidx]; 1530 1531 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE); 1532 bus_dmamap_unload(rq->tag, pd->map); 1533 RING_GET(rq->ring, 1); 1534 rq->pending--; 1535 1536 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len; 1537 pd->mbuf->m_len = frag_len; 1538 1539 if (tail != NULL) { 1540 /* additional fragments */ 1541 pd->mbuf->m_flags &= ~M_PKTHDR; 1542 tail->m_next = pd->mbuf; 1543 if(rq->islro) 1544 tail->m_nextpkt = NULL; 1545 tail = pd->mbuf; 1546 } else { 1547 /* first fragment, fill out much of the packet header */ 1548 pd->mbuf->m_pkthdr.len = len; 1549 if(rq->islro) 1550 pd->mbuf->m_nextpkt = NULL; 1551 pd->mbuf->m_pkthdr.csum_flags = 0; 1552 if (IF_CSUM_ENABLED(sc)) { 1553 if (cqe_info->l4_cksum_pass) { 1554 if(!cqe_info->ipv6_frame) { /* IPV4 */ 1555 pd->mbuf->m_pkthdr.csum_flags |= 1556 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 1557 }else { /* IPV6 frame */ 1558 if(rq->islro) { 1559 pd->mbuf->m_pkthdr.csum_flags |= 1560 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 1561 } 1562 } 1563 pd->mbuf->m_pkthdr.csum_data = 0xffff; 1564 } 1565 if (cqe_info->ip_cksum_pass) { 1566 pd->mbuf->m_pkthdr.csum_flags |= 1567 (CSUM_IP_CHECKED|CSUM_IP_VALID); 1568 } 1569 } 1570 *m = tail = pd->mbuf; 1571 } 1572 pd->mbuf = NULL; 1573 len -= frag_len; 1574 } 1575 1576 return; 1577 } 1578 1579 static void 1580 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2) 1581 { 1582 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1583 struct nic_hwlro_cqe_part1 *cqe1 = NULL; 1584 struct mbuf *m = NULL; 1585 struct oce_common_cqe_info cq_info; 1586 1587 /* parse cqe */ 1588 if(cqe2 == NULL) { 1589 cq_info.pkt_size = cqe->pkt_size; 1590 cq_info.vtag = cqe->vlan_tag; 1591 cq_info.l4_cksum_pass = cqe->l4_cksum_pass; 1592 cq_info.ip_cksum_pass = cqe->ip_cksum_pass; 1593 cq_info.ipv6_frame = cqe->ipv6_frame; 1594 cq_info.vtp = cqe->vtp; 1595 cq_info.qnq = cqe->qnq; 1596 }else { 1597 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe; 1598 cq_info.pkt_size = cqe2->coalesced_size; 1599 cq_info.vtag = cqe2->vlan_tag; 1600 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass; 1601 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass; 1602 cq_info.ipv6_frame = cqe2->ipv6_frame; 1603 cq_info.vtp = cqe2->vtp; 1604 cq_info.qnq = cqe1->qnq; 1605 } 1606 1607 cq_info.vtag = BSWAP_16(cq_info.vtag); 1608 1609 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size; 1610 if(cq_info.pkt_size % rq->cfg.frag_size) 1611 cq_info.num_frags++; 1612 1613 oce_rx_mbuf_chain(rq, &cq_info, &m); 1614 1615 if (m) { 1616 if(cqe2) { 1617 //assert(cqe2->valid != 0); 1618 1619 //assert(cqe2->cqe_type != 2); 1620 oce_correct_header(m, cqe1, cqe2); 1621 } 1622 1623 m->m_pkthdr.rcvif = sc->ifp; 1624 if (rq->queue_index) 1625 m->m_pkthdr.flowid = (rq->queue_index - 1); 1626 else 1627 m->m_pkthdr.flowid = rq->queue_index; 1628 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 1629 1630 /* This deternies if vlan tag is Valid */ 1631 if (cq_info.vtp) { 1632 if (sc->function_mode & FNM_FLEX10_MODE) { 1633 /* FLEX10. If QnQ is not set, neglect VLAN */ 1634 if (cq_info.qnq) { 1635 m->m_pkthdr.ether_vtag = cq_info.vtag; 1636 m->m_flags |= M_VLANTAG; 1637 } 1638 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) { 1639 /* In UMC mode generally pvid will be striped by 1640 hw. But in some cases we have seen it comes 1641 with pvid. So if pvid == vlan, neglect vlan. 1642 */ 1643 m->m_pkthdr.ether_vtag = cq_info.vtag; 1644 m->m_flags |= M_VLANTAG; 1645 } 1646 } 1647 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1); 1648 1649 (*sc->ifp->if_input) (sc->ifp, m); 1650 1651 /* Update rx stats per queue */ 1652 rq->rx_stats.rx_pkts++; 1653 rq->rx_stats.rx_bytes += cq_info.pkt_size; 1654 rq->rx_stats.rx_frags += cq_info.num_frags; 1655 rq->rx_stats.rx_ucast_pkts++; 1656 } 1657 return; 1658 } 1659 1660 static void 1661 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe) 1662 { 1663 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1664 int len; 1665 struct mbuf *m = NULL; 1666 struct oce_common_cqe_info cq_info; 1667 uint16_t vtag = 0; 1668 1669 /* Is it a flush compl that has no data */ 1670 if(!cqe->u0.s.num_fragments) 1671 goto exit; 1672 1673 len = cqe->u0.s.pkt_size; 1674 if (!len) { 1675 /*partial DMA workaround for Lancer*/ 1676 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments); 1677 goto exit; 1678 } 1679 1680 if (!oce_cqe_portid_valid(sc, cqe)) { 1681 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments); 1682 goto exit; 1683 } 1684 1685 /* Get vlan_tag value */ 1686 if(IS_BE(sc) || IS_SH(sc)) 1687 vtag = BSWAP_16(cqe->u0.s.vlan_tag); 1688 else 1689 vtag = cqe->u0.s.vlan_tag; 1690 1691 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass; 1692 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass; 1693 cq_info.ipv6_frame = cqe->u0.s.ip_ver; 1694 cq_info.num_frags = cqe->u0.s.num_fragments; 1695 cq_info.pkt_size = cqe->u0.s.pkt_size; 1696 1697 oce_rx_mbuf_chain(rq, &cq_info, &m); 1698 1699 if (m) { 1700 m->m_pkthdr.rcvif = sc->ifp; 1701 if (rq->queue_index) 1702 m->m_pkthdr.flowid = (rq->queue_index - 1); 1703 else 1704 m->m_pkthdr.flowid = rq->queue_index; 1705 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 1706 1707 /* This deternies if vlan tag is Valid */ 1708 if (oce_cqe_vtp_valid(sc, cqe)) { 1709 if (sc->function_mode & FNM_FLEX10_MODE) { 1710 /* FLEX10. If QnQ is not set, neglect VLAN */ 1711 if (cqe->u0.s.qnq) { 1712 m->m_pkthdr.ether_vtag = vtag; 1713 m->m_flags |= M_VLANTAG; 1714 } 1715 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) { 1716 /* In UMC mode generally pvid will be striped by 1717 hw. But in some cases we have seen it comes 1718 with pvid. So if pvid == vlan, neglect vlan. 1719 */ 1720 m->m_pkthdr.ether_vtag = vtag; 1721 m->m_flags |= M_VLANTAG; 1722 } 1723 } 1724 1725 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1); 1726 #if defined(INET6) || defined(INET) 1727 /* Try to queue to LRO */ 1728 if (IF_LRO_ENABLED(sc) && 1729 (cqe->u0.s.ip_cksum_pass) && 1730 (cqe->u0.s.l4_cksum_pass) && 1731 (!cqe->u0.s.ip_ver) && 1732 (rq->lro.lro_cnt != 0)) { 1733 if (tcp_lro_rx(&rq->lro, m, 0) == 0) { 1734 rq->lro_pkts_queued ++; 1735 goto post_done; 1736 } 1737 /* If LRO posting fails then try to post to STACK */ 1738 } 1739 #endif 1740 1741 (*sc->ifp->if_input) (sc->ifp, m); 1742 #if defined(INET6) || defined(INET) 1743 post_done: 1744 #endif 1745 /* Update rx stats per queue */ 1746 rq->rx_stats.rx_pkts++; 1747 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size; 1748 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments; 1749 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET) 1750 rq->rx_stats.rx_mcast_pkts++; 1751 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET) 1752 rq->rx_stats.rx_ucast_pkts++; 1753 } 1754 exit: 1755 return; 1756 } 1757 1758 void 1759 oce_discard_rx_comp(struct oce_rq *rq, int num_frags) 1760 { 1761 uint32_t i = 0; 1762 struct oce_packet_desc *pd; 1763 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1764 1765 for (i = 0; i < num_frags; i++) { 1766 if (rq->ring->cidx == rq->ring->pidx) { 1767 device_printf(sc->dev, 1768 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n"); 1769 return; 1770 } 1771 pd = &rq->pckts[rq->ring->cidx]; 1772 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE); 1773 bus_dmamap_unload(rq->tag, pd->map); 1774 if (pd->mbuf != NULL) { 1775 m_freem(pd->mbuf); 1776 pd->mbuf = NULL; 1777 } 1778 1779 RING_GET(rq->ring, 1); 1780 rq->pending--; 1781 } 1782 } 1783 1784 static int 1785 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe) 1786 { 1787 struct oce_nic_rx_cqe_v1 *cqe_v1; 1788 int vtp = 0; 1789 1790 if (sc->be3_native) { 1791 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe; 1792 vtp = cqe_v1->u0.s.vlan_tag_present; 1793 } else 1794 vtp = cqe->u0.s.vlan_tag_present; 1795 1796 return vtp; 1797 1798 } 1799 1800 static int 1801 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe) 1802 { 1803 struct oce_nic_rx_cqe_v1 *cqe_v1; 1804 int port_id = 0; 1805 1806 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) { 1807 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe; 1808 port_id = cqe_v1->u0.s.port; 1809 if (sc->port_id != port_id) 1810 return 0; 1811 } else 1812 ;/* For BE3 legacy and Lancer this is dummy */ 1813 1814 return 1; 1815 1816 } 1817 1818 #if defined(INET6) || defined(INET) 1819 void 1820 oce_rx_flush_lro(struct oce_rq *rq) 1821 { 1822 struct lro_ctrl *lro = &rq->lro; 1823 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1824 1825 if (!IF_LRO_ENABLED(sc)) 1826 return; 1827 1828 tcp_lro_flush_all(lro); 1829 rq->lro_pkts_queued = 0; 1830 1831 return; 1832 } 1833 1834 static int 1835 oce_init_lro(POCE_SOFTC sc) 1836 { 1837 struct lro_ctrl *lro = NULL; 1838 int i = 0, rc = 0; 1839 1840 for (i = 0; i < sc->nrqs; i++) { 1841 lro = &sc->rq[i]->lro; 1842 rc = tcp_lro_init(lro); 1843 if (rc != 0) { 1844 device_printf(sc->dev, "LRO init failed\n"); 1845 return rc; 1846 } 1847 lro->ifp = sc->ifp; 1848 } 1849 1850 return rc; 1851 } 1852 1853 void 1854 oce_free_lro(POCE_SOFTC sc) 1855 { 1856 struct lro_ctrl *lro = NULL; 1857 int i = 0; 1858 1859 for (i = 0; i < sc->nrqs; i++) { 1860 lro = &sc->rq[i]->lro; 1861 if (lro) 1862 tcp_lro_free(lro); 1863 } 1864 } 1865 #endif 1866 1867 int 1868 oce_alloc_rx_bufs(struct oce_rq *rq, int count) 1869 { 1870 POCE_SOFTC sc = (POCE_SOFTC) rq->parent; 1871 int i, rc; 1872 struct oce_packet_desc *pd; 1873 bus_dma_segment_t segs[6]; 1874 int nsegs, added = 0; 1875 struct oce_nic_rqe *rqe; 1876 pd_rxulp_db_t rxdb_reg; 1877 uint32_t val = 0; 1878 uint32_t oce_max_rq_posts = 64; 1879 1880 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t)); 1881 for (i = 0; i < count; i++) { 1882 pd = &rq->pckts[rq->ring->pidx]; 1883 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size); 1884 if (pd->mbuf == NULL) { 1885 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size); 1886 break; 1887 } 1888 pd->mbuf->m_nextpkt = NULL; 1889 1890 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size; 1891 1892 rc = bus_dmamap_load_mbuf_sg(rq->tag, 1893 pd->map, 1894 pd->mbuf, 1895 segs, &nsegs, BUS_DMA_NOWAIT); 1896 if (rc) { 1897 m_free(pd->mbuf); 1898 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc); 1899 break; 1900 } 1901 1902 if (nsegs != 1) { 1903 i--; 1904 continue; 1905 } 1906 1907 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD); 1908 1909 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe); 1910 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr); 1911 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr); 1912 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe)); 1913 RING_PUT(rq->ring, 1); 1914 added++; 1915 rq->pending++; 1916 } 1917 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS; 1918 if (added != 0) { 1919 for (i = added / oce_max_rq_posts; i > 0; i--) { 1920 rxdb_reg.bits.num_posted = oce_max_rq_posts; 1921 rxdb_reg.bits.qid = rq->rq_id; 1922 if(rq->islro) { 1923 val |= rq->rq_id & DB_LRO_RQ_ID_MASK; 1924 val |= oce_max_rq_posts << 16; 1925 OCE_WRITE_REG32(sc, db, DB_OFFSET, val); 1926 }else { 1927 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0); 1928 } 1929 added -= oce_max_rq_posts; 1930 } 1931 if (added > 0) { 1932 rxdb_reg.bits.qid = rq->rq_id; 1933 rxdb_reg.bits.num_posted = added; 1934 if(rq->islro) { 1935 val |= rq->rq_id & DB_LRO_RQ_ID_MASK; 1936 val |= added << 16; 1937 OCE_WRITE_REG32(sc, db, DB_OFFSET, val); 1938 }else { 1939 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0); 1940 } 1941 } 1942 } 1943 1944 return 0; 1945 } 1946 1947 static void 1948 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq) 1949 { 1950 if (num_cqes) { 1951 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE); 1952 if(!sc->enable_hwlro) { 1953 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1) 1954 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1)); 1955 }else { 1956 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64) 1957 oce_alloc_rx_bufs(rq, 64); 1958 } 1959 } 1960 1961 return; 1962 } 1963 1964 uint16_t 1965 oce_rq_handler_lro(void *arg) 1966 { 1967 struct oce_rq *rq = (struct oce_rq *)arg; 1968 struct oce_cq *cq = rq->cq; 1969 POCE_SOFTC sc = rq->parent; 1970 struct nic_hwlro_singleton_cqe *cqe; 1971 struct nic_hwlro_cqe_part2 *cqe2; 1972 int num_cqes = 0; 1973 1974 LOCK(&rq->rx_lock); 1975 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 1976 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe); 1977 while (cqe->valid) { 1978 if(cqe->cqe_type == 0) { /* singleton cqe */ 1979 /* we should not get singleton cqe after cqe1 on same rq */ 1980 if(rq->cqe_firstpart != NULL) { 1981 device_printf(sc->dev, "Got singleton cqe after cqe1 \n"); 1982 goto exit_rq_handler_lro; 1983 } 1984 if(cqe->error != 0) { 1985 rq->rx_stats.rxcp_err++; 1986 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1); 1987 } 1988 oce_rx_lro(rq, cqe, NULL); 1989 rq->rx_stats.rx_compl++; 1990 cqe->valid = 0; 1991 RING_GET(cq->ring, 1); 1992 num_cqes++; 1993 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled)) 1994 break; 1995 }else if(cqe->cqe_type == 0x1) { /* first part */ 1996 /* we should not get cqe1 after cqe1 on same rq */ 1997 if(rq->cqe_firstpart != NULL) { 1998 device_printf(sc->dev, "Got cqe1 after cqe1 \n"); 1999 goto exit_rq_handler_lro; 2000 } 2001 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe; 2002 RING_GET(cq->ring, 1); 2003 }else if(cqe->cqe_type == 0x2) { /* second part */ 2004 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe; 2005 if(cqe2->error != 0) { 2006 rq->rx_stats.rxcp_err++; 2007 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1); 2008 } 2009 /* We should not get cqe2 without cqe1 */ 2010 if(rq->cqe_firstpart == NULL) { 2011 device_printf(sc->dev, "Got cqe2 without cqe1 \n"); 2012 goto exit_rq_handler_lro; 2013 } 2014 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2); 2015 2016 rq->rx_stats.rx_compl++; 2017 rq->cqe_firstpart->valid = 0; 2018 cqe2->valid = 0; 2019 rq->cqe_firstpart = NULL; 2020 2021 RING_GET(cq->ring, 1); 2022 num_cqes += 2; 2023 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled)) 2024 break; 2025 } 2026 2027 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 2028 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe); 2029 } 2030 oce_check_rx_bufs(sc, num_cqes, rq); 2031 exit_rq_handler_lro: 2032 UNLOCK(&rq->rx_lock); 2033 return 0; 2034 } 2035 2036 /* Handle the Completion Queue for receive */ 2037 uint16_t 2038 oce_rq_handler(void *arg) 2039 { 2040 struct epoch_tracker et; 2041 struct oce_rq *rq = (struct oce_rq *)arg; 2042 struct oce_cq *cq = rq->cq; 2043 POCE_SOFTC sc = rq->parent; 2044 struct oce_nic_rx_cqe *cqe; 2045 int num_cqes = 0; 2046 2047 NET_EPOCH_ENTER(et); 2048 if(rq->islro) { 2049 oce_rq_handler_lro(arg); 2050 NET_EPOCH_EXIT(et); 2051 return 0; 2052 } 2053 LOCK(&rq->rx_lock); 2054 bus_dmamap_sync(cq->ring->dma.tag, 2055 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 2056 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe); 2057 while (cqe->u0.dw[2]) { 2058 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe)); 2059 2060 if (cqe->u0.s.error == 0) { 2061 oce_rx(rq, cqe); 2062 } else { 2063 rq->rx_stats.rxcp_err++; 2064 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1); 2065 /* Post L3/L4 errors to stack.*/ 2066 oce_rx(rq, cqe); 2067 } 2068 rq->rx_stats.rx_compl++; 2069 cqe->u0.dw[2] = 0; 2070 2071 #if defined(INET6) || defined(INET) 2072 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) { 2073 oce_rx_flush_lro(rq); 2074 } 2075 #endif 2076 2077 RING_GET(cq->ring, 1); 2078 bus_dmamap_sync(cq->ring->dma.tag, 2079 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 2080 cqe = 2081 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe); 2082 num_cqes++; 2083 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled)) 2084 break; 2085 } 2086 2087 #if defined(INET6) || defined(INET) 2088 if (IF_LRO_ENABLED(sc)) 2089 oce_rx_flush_lro(rq); 2090 #endif 2091 2092 oce_check_rx_bufs(sc, num_cqes, rq); 2093 UNLOCK(&rq->rx_lock); 2094 NET_EPOCH_EXIT(et); 2095 return 0; 2096 2097 } 2098 2099 /***************************************************************************** 2100 * Helper function prototypes in this file * 2101 *****************************************************************************/ 2102 2103 static int 2104 oce_attach_ifp(POCE_SOFTC sc) 2105 { 2106 2107 sc->ifp = if_alloc(IFT_ETHER); 2108 if (!sc->ifp) 2109 return ENOMEM; 2110 2111 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status); 2112 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL); 2113 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO); 2114 2115 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_KNOWSEPOCH; 2116 sc->ifp->if_ioctl = oce_ioctl; 2117 sc->ifp->if_start = oce_start; 2118 sc->ifp->if_init = oce_init; 2119 sc->ifp->if_mtu = ETHERMTU; 2120 sc->ifp->if_softc = sc; 2121 sc->ifp->if_transmit = oce_multiq_start; 2122 sc->ifp->if_qflush = oce_multiq_flush; 2123 2124 if_initname(sc->ifp, 2125 device_get_name(sc->dev), device_get_unit(sc->dev)); 2126 2127 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1; 2128 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen); 2129 IFQ_SET_READY(&sc->ifp->if_snd); 2130 2131 sc->ifp->if_hwassist = OCE_IF_HWASSIST; 2132 sc->ifp->if_hwassist |= CSUM_TSO; 2133 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP); 2134 2135 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES; 2136 sc->ifp->if_capabilities |= IFCAP_HWCSUM; 2137 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 2138 2139 #if defined(INET6) || defined(INET) 2140 sc->ifp->if_capabilities |= IFCAP_TSO; 2141 sc->ifp->if_capabilities |= IFCAP_LRO; 2142 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 2143 #endif 2144 2145 sc->ifp->if_capenable = sc->ifp->if_capabilities; 2146 sc->ifp->if_baudrate = IF_Gbps(10); 2147 2148 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); 2149 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS; 2150 sc->ifp->if_hw_tsomaxsegsize = 4096; 2151 2152 ether_ifattach(sc->ifp, sc->macaddr.mac_addr); 2153 2154 return 0; 2155 } 2156 2157 static void 2158 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag) 2159 { 2160 POCE_SOFTC sc = ifp->if_softc; 2161 2162 if (ifp->if_softc != arg) 2163 return; 2164 if ((vtag == 0) || (vtag > 4095)) 2165 return; 2166 2167 sc->vlan_tag[vtag] = 1; 2168 sc->vlans_added++; 2169 if (sc->vlans_added <= (sc->max_vlans + 1)) 2170 oce_vid_config(sc); 2171 } 2172 2173 static void 2174 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag) 2175 { 2176 POCE_SOFTC sc = ifp->if_softc; 2177 2178 if (ifp->if_softc != arg) 2179 return; 2180 if ((vtag == 0) || (vtag > 4095)) 2181 return; 2182 2183 sc->vlan_tag[vtag] = 0; 2184 sc->vlans_added--; 2185 oce_vid_config(sc); 2186 } 2187 2188 /* 2189 * A max of 64 vlans can be configured in BE. If the user configures 2190 * more, place the card in vlan promiscuous mode. 2191 */ 2192 static int 2193 oce_vid_config(POCE_SOFTC sc) 2194 { 2195 struct normal_vlan vtags[MAX_VLANFILTER_SIZE]; 2196 uint16_t ntags = 0, i; 2197 int status = 0; 2198 2199 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) && 2200 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) { 2201 for (i = 0; i < MAX_VLANS; i++) { 2202 if (sc->vlan_tag[i]) { 2203 vtags[ntags].vtag = i; 2204 ntags++; 2205 } 2206 } 2207 if (ntags) 2208 status = oce_config_vlan(sc, (uint8_t) sc->if_id, 2209 vtags, ntags, 1, 0); 2210 } else 2211 status = oce_config_vlan(sc, (uint8_t) sc->if_id, 2212 NULL, 0, 1, 1); 2213 return status; 2214 } 2215 2216 static void 2217 oce_mac_addr_set(POCE_SOFTC sc) 2218 { 2219 uint32_t old_pmac_id = sc->pmac_id; 2220 int status = 0; 2221 2222 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr, 2223 sc->macaddr.size_of_struct); 2224 if (!status) 2225 return; 2226 2227 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)), 2228 sc->if_id, &sc->pmac_id); 2229 if (!status) { 2230 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id); 2231 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr, 2232 sc->macaddr.size_of_struct); 2233 } 2234 if (status) 2235 device_printf(sc->dev, "Failed update macaddress\n"); 2236 2237 } 2238 2239 static int 2240 oce_handle_passthrough(struct ifnet *ifp, caddr_t data) 2241 { 2242 POCE_SOFTC sc = ifp->if_softc; 2243 struct ifreq *ifr = (struct ifreq *)data; 2244 int rc = ENXIO; 2245 char cookie[32] = {0}; 2246 void *priv_data = ifr_data_get_ptr(ifr); 2247 void *ioctl_ptr; 2248 uint32_t req_size; 2249 struct mbx_hdr req; 2250 OCE_DMA_MEM dma_mem; 2251 struct mbx_common_get_cntl_attr *fw_cmd; 2252 2253 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE))) 2254 return EFAULT; 2255 2256 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE))) 2257 return EINVAL; 2258 2259 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE); 2260 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr))) 2261 return EFAULT; 2262 2263 req_size = le32toh(req.u0.req.request_length); 2264 if (req_size > 65536) 2265 return EINVAL; 2266 2267 req_size += sizeof(struct mbx_hdr); 2268 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0); 2269 if (rc) 2270 return ENOMEM; 2271 2272 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) { 2273 rc = EFAULT; 2274 goto dma_free; 2275 } 2276 2277 rc = oce_pass_through_mbox(sc, &dma_mem, req_size); 2278 if (rc) { 2279 rc = EIO; 2280 goto dma_free; 2281 } 2282 2283 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size)) 2284 rc = EFAULT; 2285 2286 /* 2287 firmware is filling all the attributes for this ioctl except 2288 the driver version..so fill it 2289 */ 2290 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) { 2291 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr; 2292 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str, 2293 COMPONENT_REVISION, strlen(COMPONENT_REVISION)); 2294 } 2295 2296 dma_free: 2297 oce_dma_free(sc, &dma_mem); 2298 return rc; 2299 2300 } 2301 2302 static void 2303 oce_eqd_set_periodic(POCE_SOFTC sc) 2304 { 2305 struct oce_set_eqd set_eqd[OCE_MAX_EQ]; 2306 struct oce_aic_obj *aic; 2307 struct oce_eq *eqo; 2308 uint64_t now = 0, delta; 2309 int eqd, i, num = 0; 2310 uint32_t tx_reqs = 0, rxpkts = 0, pps; 2311 struct oce_wq *wq; 2312 struct oce_rq *rq; 2313 2314 #define ticks_to_msecs(t) (1000 * (t) / hz) 2315 2316 for (i = 0 ; i < sc->neqs; i++) { 2317 eqo = sc->eq[i]; 2318 aic = &sc->aic_obj[i]; 2319 /* When setting the static eq delay from the user space */ 2320 if (!aic->enable) { 2321 if (aic->ticks) 2322 aic->ticks = 0; 2323 eqd = aic->et_eqd; 2324 goto modify_eqd; 2325 } 2326 2327 if (i == 0) { 2328 rq = sc->rq[0]; 2329 rxpkts = rq->rx_stats.rx_pkts; 2330 } else 2331 rxpkts = 0; 2332 if (i + 1 < sc->nrqs) { 2333 rq = sc->rq[i + 1]; 2334 rxpkts += rq->rx_stats.rx_pkts; 2335 } 2336 if (i < sc->nwqs) { 2337 wq = sc->wq[i]; 2338 tx_reqs = wq->tx_stats.tx_reqs; 2339 } else 2340 tx_reqs = 0; 2341 now = ticks; 2342 2343 if (!aic->ticks || now < aic->ticks || 2344 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) { 2345 aic->prev_rxpkts = rxpkts; 2346 aic->prev_txreqs = tx_reqs; 2347 aic->ticks = now; 2348 continue; 2349 } 2350 2351 delta = ticks_to_msecs(now - aic->ticks); 2352 2353 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) + 2354 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta); 2355 eqd = (pps / 15000) << 2; 2356 if (eqd < 8) 2357 eqd = 0; 2358 2359 /* Make sure that the eq delay is in the known range */ 2360 eqd = min(eqd, aic->max_eqd); 2361 eqd = max(eqd, aic->min_eqd); 2362 2363 aic->prev_rxpkts = rxpkts; 2364 aic->prev_txreqs = tx_reqs; 2365 aic->ticks = now; 2366 2367 modify_eqd: 2368 if (eqd != aic->cur_eqd) { 2369 set_eqd[num].delay_multiplier = (eqd * 65)/100; 2370 set_eqd[num].eq_id = eqo->eq_id; 2371 aic->cur_eqd = eqd; 2372 num++; 2373 } 2374 } 2375 2376 /* Is there atleast one eq that needs to be modified? */ 2377 for(i = 0; i < num; i += 8) { 2378 if((num - i) >=8 ) 2379 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8); 2380 else 2381 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i)); 2382 } 2383 2384 } 2385 2386 static void oce_detect_hw_error(POCE_SOFTC sc) 2387 { 2388 2389 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0; 2390 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0; 2391 uint32_t i; 2392 2393 if (sc->hw_error) 2394 return; 2395 2396 if (IS_XE201(sc)) { 2397 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET); 2398 if (sliport_status & SLIPORT_STATUS_ERR_MASK) { 2399 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET); 2400 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET); 2401 } 2402 } else { 2403 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW); 2404 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH); 2405 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK); 2406 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK); 2407 2408 ue_low = (ue_low & ~ue_low_mask); 2409 ue_high = (ue_high & ~ue_high_mask); 2410 } 2411 2412 /* On certain platforms BE hardware can indicate spurious UEs. 2413 * Allow the h/w to stop working completely in case of a real UE. 2414 * Hence not setting the hw_error for UE detection. 2415 */ 2416 if (sliport_status & SLIPORT_STATUS_ERR_MASK) { 2417 sc->hw_error = TRUE; 2418 device_printf(sc->dev, "Error detected in the card\n"); 2419 } 2420 2421 if (sliport_status & SLIPORT_STATUS_ERR_MASK) { 2422 device_printf(sc->dev, 2423 "ERR: sliport status 0x%x\n", sliport_status); 2424 device_printf(sc->dev, 2425 "ERR: sliport error1 0x%x\n", sliport_err1); 2426 device_printf(sc->dev, 2427 "ERR: sliport error2 0x%x\n", sliport_err2); 2428 } 2429 2430 if (ue_low) { 2431 for (i = 0; ue_low; ue_low >>= 1, i++) { 2432 if (ue_low & 1) 2433 device_printf(sc->dev, "UE: %s bit set\n", 2434 ue_status_low_desc[i]); 2435 } 2436 } 2437 2438 if (ue_high) { 2439 for (i = 0; ue_high; ue_high >>= 1, i++) { 2440 if (ue_high & 1) 2441 device_printf(sc->dev, "UE: %s bit set\n", 2442 ue_status_hi_desc[i]); 2443 } 2444 } 2445 2446 } 2447 2448 static void 2449 oce_local_timer(void *arg) 2450 { 2451 POCE_SOFTC sc = arg; 2452 int i = 0; 2453 2454 oce_detect_hw_error(sc); 2455 oce_refresh_nic_stats(sc); 2456 oce_refresh_queue_stats(sc); 2457 oce_mac_addr_set(sc); 2458 2459 /* TX Watch Dog*/ 2460 for (i = 0; i < sc->nwqs; i++) 2461 oce_tx_restart(sc, sc->wq[i]); 2462 2463 /* calculate and set the eq delay for optimal interrupt rate */ 2464 if (IS_BE(sc) || IS_SH(sc)) 2465 oce_eqd_set_periodic(sc); 2466 2467 callout_reset(&sc->timer, hz, oce_local_timer, sc); 2468 } 2469 2470 static void 2471 oce_tx_compl_clean(POCE_SOFTC sc) 2472 { 2473 struct oce_wq *wq; 2474 int i = 0, timeo = 0, num_wqes = 0; 2475 int pending_txqs = sc->nwqs; 2476 2477 /* Stop polling for compls when HW has been silent for 10ms or 2478 * hw_error or no outstanding completions expected 2479 */ 2480 do { 2481 pending_txqs = sc->nwqs; 2482 2483 for_all_wq_queues(sc, wq, i) { 2484 num_wqes = oce_wq_handler(wq); 2485 2486 if(num_wqes) 2487 timeo = 0; 2488 2489 if(!wq->ring->num_used) 2490 pending_txqs--; 2491 } 2492 2493 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error) 2494 break; 2495 2496 DELAY(1000); 2497 } while (TRUE); 2498 2499 for_all_wq_queues(sc, wq, i) { 2500 while(wq->ring->num_used) { 2501 LOCK(&wq->tx_compl_lock); 2502 oce_process_tx_completion(wq); 2503 UNLOCK(&wq->tx_compl_lock); 2504 } 2505 } 2506 2507 } 2508 2509 /* NOTE : This should only be called holding 2510 * DEVICE_LOCK. 2511 */ 2512 static void 2513 oce_if_deactivate(POCE_SOFTC sc) 2514 { 2515 int i; 2516 struct oce_rq *rq; 2517 struct oce_wq *wq; 2518 struct oce_eq *eq; 2519 2520 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 2521 2522 oce_tx_compl_clean(sc); 2523 2524 /* Stop intrs and finish any bottom halves pending */ 2525 oce_hw_intr_disable(sc); 2526 2527 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire 2528 any other lock. So unlock device lock and require after 2529 completing taskqueue_drain. 2530 */ 2531 UNLOCK(&sc->dev_lock); 2532 for (i = 0; i < sc->intr_count; i++) { 2533 if (sc->intrs[i].tq != NULL) { 2534 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task); 2535 } 2536 } 2537 LOCK(&sc->dev_lock); 2538 2539 /* Delete RX queue in card with flush param */ 2540 oce_stop_rx(sc); 2541 2542 /* Invalidate any pending cq and eq entries*/ 2543 for_all_evnt_queues(sc, eq, i) 2544 oce_drain_eq(eq); 2545 for_all_rq_queues(sc, rq, i) 2546 oce_drain_rq_cq(rq); 2547 for_all_wq_queues(sc, wq, i) 2548 oce_drain_wq_cq(wq); 2549 2550 /* But still we need to get MCC aync events. 2551 So enable intrs and also arm first EQ 2552 */ 2553 oce_hw_intr_enable(sc); 2554 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE); 2555 2556 DELAY(10); 2557 } 2558 2559 static void 2560 oce_if_activate(POCE_SOFTC sc) 2561 { 2562 struct oce_eq *eq; 2563 struct oce_rq *rq; 2564 struct oce_wq *wq; 2565 int i, rc = 0; 2566 2567 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING; 2568 2569 oce_hw_intr_disable(sc); 2570 2571 oce_start_rx(sc); 2572 2573 for_all_rq_queues(sc, rq, i) { 2574 rc = oce_start_rq(rq); 2575 if (rc) 2576 device_printf(sc->dev, "Unable to start RX\n"); 2577 } 2578 2579 for_all_wq_queues(sc, wq, i) { 2580 rc = oce_start_wq(wq); 2581 if (rc) 2582 device_printf(sc->dev, "Unable to start TX\n"); 2583 } 2584 2585 for_all_evnt_queues(sc, eq, i) 2586 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE); 2587 2588 oce_hw_intr_enable(sc); 2589 2590 } 2591 2592 static void 2593 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe) 2594 { 2595 /* Update Link status */ 2596 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) == 2597 ASYNC_EVENT_LINK_UP) { 2598 sc->link_status = ASYNC_EVENT_LINK_UP; 2599 if_link_state_change(sc->ifp, LINK_STATE_UP); 2600 } else { 2601 sc->link_status = ASYNC_EVENT_LINK_DOWN; 2602 if_link_state_change(sc->ifp, LINK_STATE_DOWN); 2603 } 2604 } 2605 2606 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc, 2607 struct oce_async_evt_grp5_os2bmc *evt) 2608 { 2609 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc)); 2610 if (evt->u.s.mgmt_enable) 2611 sc->flags |= OCE_FLAGS_OS2BMC; 2612 else 2613 return; 2614 2615 sc->bmc_filt_mask = evt->u.s.arp_filter; 2616 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1); 2617 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2); 2618 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3); 2619 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4); 2620 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5); 2621 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6); 2622 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7); 2623 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8); 2624 } 2625 2626 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe) 2627 { 2628 struct oce_async_event_grp5_pvid_state *gcqe; 2629 struct oce_async_evt_grp5_os2bmc *bmccqe; 2630 2631 switch (cqe->u0.s.async_type) { 2632 case ASYNC_EVENT_PVID_STATE: 2633 /* GRP5 PVID */ 2634 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe; 2635 if (gcqe->enabled) 2636 sc->pvid = gcqe->tag & VLAN_VID_MASK; 2637 else 2638 sc->pvid = 0; 2639 break; 2640 case ASYNC_EVENT_OS2BMC: 2641 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe; 2642 oce_async_grp5_osbmc_process(sc, bmccqe); 2643 break; 2644 default: 2645 break; 2646 } 2647 } 2648 2649 /* Handle the Completion Queue for the Mailbox/Async notifications */ 2650 uint16_t 2651 oce_mq_handler(void *arg) 2652 { 2653 struct oce_mq *mq = (struct oce_mq *)arg; 2654 POCE_SOFTC sc = mq->parent; 2655 struct oce_cq *cq = mq->cq; 2656 int num_cqes = 0, evt_type = 0, optype = 0; 2657 struct oce_mq_cqe *cqe; 2658 struct oce_async_cqe_link_state *acqe; 2659 struct oce_async_event_qnq *dbgcqe; 2660 2661 bus_dmamap_sync(cq->ring->dma.tag, 2662 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 2663 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe); 2664 2665 while (cqe->u0.dw[3]) { 2666 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe)); 2667 if (cqe->u0.s.async_event) { 2668 evt_type = cqe->u0.s.event_type; 2669 optype = cqe->u0.s.async_type; 2670 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) { 2671 /* Link status evt */ 2672 acqe = (struct oce_async_cqe_link_state *)cqe; 2673 process_link_state(sc, acqe); 2674 } else if (evt_type == ASYNC_EVENT_GRP5) { 2675 oce_process_grp5_events(sc, cqe); 2676 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG && 2677 optype == ASYNC_EVENT_DEBUG_QNQ) { 2678 dbgcqe = (struct oce_async_event_qnq *)cqe; 2679 if(dbgcqe->valid) 2680 sc->qnqid = dbgcqe->vlan_tag; 2681 sc->qnq_debug_event = TRUE; 2682 } 2683 } 2684 cqe->u0.dw[3] = 0; 2685 RING_GET(cq->ring, 1); 2686 bus_dmamap_sync(cq->ring->dma.tag, 2687 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE); 2688 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe); 2689 num_cqes++; 2690 } 2691 2692 if (num_cqes) 2693 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE); 2694 2695 return 0; 2696 } 2697 2698 static void 2699 setup_max_queues_want(POCE_SOFTC sc) 2700 { 2701 /* Check if it is FLEX machine. Is so dont use RSS */ 2702 if ((sc->function_mode & FNM_FLEX10_MODE) || 2703 (sc->function_mode & FNM_UMC_MODE) || 2704 (sc->function_mode & FNM_VNIC_MODE) || 2705 (!is_rss_enabled(sc)) || 2706 IS_BE2(sc)) { 2707 sc->nrqs = 1; 2708 sc->nwqs = 1; 2709 } else { 2710 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1; 2711 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs); 2712 } 2713 2714 if (IS_BE2(sc) && is_rss_enabled(sc)) 2715 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1; 2716 } 2717 2718 static void 2719 update_queues_got(POCE_SOFTC sc) 2720 { 2721 if (is_rss_enabled(sc)) { 2722 sc->nrqs = sc->intr_count + 1; 2723 sc->nwqs = sc->intr_count; 2724 } else { 2725 sc->nrqs = 1; 2726 sc->nwqs = 1; 2727 } 2728 2729 if (IS_BE2(sc)) 2730 sc->nwqs = 1; 2731 } 2732 2733 static int 2734 oce_check_ipv6_ext_hdr(struct mbuf *m) 2735 { 2736 struct ether_header *eh = mtod(m, struct ether_header *); 2737 caddr_t m_datatemp = m->m_data; 2738 2739 if (eh->ether_type == htons(ETHERTYPE_IPV6)) { 2740 m->m_data += sizeof(struct ether_header); 2741 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 2742 2743 if((ip6->ip6_nxt != IPPROTO_TCP) && \ 2744 (ip6->ip6_nxt != IPPROTO_UDP)){ 2745 struct ip6_ext *ip6e = NULL; 2746 m->m_data += sizeof(struct ip6_hdr); 2747 2748 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *); 2749 if(ip6e->ip6e_len == 0xff) { 2750 m->m_data = m_datatemp; 2751 return TRUE; 2752 } 2753 } 2754 m->m_data = m_datatemp; 2755 } 2756 return FALSE; 2757 } 2758 2759 static int 2760 is_be3_a1(POCE_SOFTC sc) 2761 { 2762 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) { 2763 return TRUE; 2764 } 2765 return FALSE; 2766 } 2767 2768 static struct mbuf * 2769 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete) 2770 { 2771 uint16_t vlan_tag = 0; 2772 2773 if(!M_WRITABLE(m)) 2774 return NULL; 2775 2776 /* Embed vlan tag in the packet if it is not part of it */ 2777 if(m->m_flags & M_VLANTAG) { 2778 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag); 2779 m->m_flags &= ~M_VLANTAG; 2780 } 2781 2782 /* if UMC, ignore vlan tag insertion and instead insert pvid */ 2783 if(sc->pvid) { 2784 if(!vlan_tag) 2785 vlan_tag = sc->pvid; 2786 if (complete) 2787 *complete = FALSE; 2788 } 2789 2790 if(vlan_tag) { 2791 m = ether_vlanencap(m, vlan_tag); 2792 } 2793 2794 if(sc->qnqid) { 2795 m = ether_vlanencap(m, sc->qnqid); 2796 2797 if (complete) 2798 *complete = FALSE; 2799 } 2800 return m; 2801 } 2802 2803 static int 2804 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m) 2805 { 2806 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \ 2807 oce_check_ipv6_ext_hdr(m)) { 2808 return TRUE; 2809 } 2810 return FALSE; 2811 } 2812 2813 static void 2814 oce_get_config(POCE_SOFTC sc) 2815 { 2816 int rc = 0; 2817 uint32_t max_rss = 0; 2818 2819 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native)) 2820 max_rss = OCE_LEGACY_MODE_RSS; 2821 else 2822 max_rss = OCE_MAX_RSS; 2823 2824 if (!IS_BE(sc)) { 2825 rc = oce_get_profile_config(sc, max_rss); 2826 if (rc) { 2827 sc->nwqs = OCE_MAX_WQ; 2828 sc->nrssqs = max_rss; 2829 sc->nrqs = sc->nrssqs + 1; 2830 } 2831 } 2832 else { /* For BE3 don't rely on fw for determining the resources */ 2833 sc->nrssqs = max_rss; 2834 sc->nrqs = sc->nrssqs + 1; 2835 sc->nwqs = OCE_MAX_WQ; 2836 sc->max_vlans = MAX_VLANFILTER_SIZE; 2837 } 2838 } 2839 2840 static void 2841 oce_rdma_close(void) 2842 { 2843 if (oce_rdma_if != NULL) { 2844 oce_rdma_if = NULL; 2845 } 2846 } 2847 2848 static void 2849 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr) 2850 { 2851 memcpy(macaddr, sc->macaddr.mac_addr, 6); 2852 } 2853 2854 int 2855 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if) 2856 { 2857 POCE_SOFTC sc; 2858 struct oce_dev_info di; 2859 int i; 2860 2861 if ((rdma_info == NULL) || (rdma_if == NULL)) { 2862 return -EINVAL; 2863 } 2864 2865 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) || 2866 (rdma_if->size != OCE_RDMA_IF_SIZE)) { 2867 return -ENXIO; 2868 } 2869 2870 rdma_info->close = oce_rdma_close; 2871 rdma_info->mbox_post = oce_mbox_post; 2872 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init; 2873 rdma_info->get_mac_addr = oce_get_mac_addr; 2874 2875 oce_rdma_if = rdma_if; 2876 2877 sc = softc_head; 2878 while (sc != NULL) { 2879 if (oce_rdma_if->announce != NULL) { 2880 memset(&di, 0, sizeof(di)); 2881 di.dev = sc->dev; 2882 di.softc = sc; 2883 di.ifp = sc->ifp; 2884 di.db_bhandle = sc->db_bhandle; 2885 di.db_btag = sc->db_btag; 2886 di.db_page_size = 4096; 2887 if (sc->flags & OCE_FLAGS_USING_MSIX) { 2888 di.intr_mode = OCE_INTERRUPT_MODE_MSIX; 2889 } else if (sc->flags & OCE_FLAGS_USING_MSI) { 2890 di.intr_mode = OCE_INTERRUPT_MODE_MSI; 2891 } else { 2892 di.intr_mode = OCE_INTERRUPT_MODE_INTX; 2893 } 2894 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk 2895 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) { 2896 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count; 2897 di.msix.start_vector = sc->intr_count; 2898 for (i=0; i<di.msix.num_vectors; i++) { 2899 di.msix.vector_list[i] = sc->intrs[i].vector; 2900 } 2901 } else { 2902 } 2903 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6); 2904 di.vendor_id = pci_get_vendor(sc->dev); 2905 di.dev_id = pci_get_device(sc->dev); 2906 2907 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) { 2908 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED; 2909 } 2910 2911 rdma_if->announce(&di); 2912 sc = sc->next; 2913 } 2914 } 2915 2916 return 0; 2917 } 2918 2919 static void 2920 oce_read_env_variables( POCE_SOFTC sc ) 2921 { 2922 char *value = NULL; 2923 int rc = 0; 2924 2925 /* read if user wants to enable hwlro or swlro */ 2926 //value = getenv("oce_enable_hwlro"); 2927 if(value && IS_SH(sc)) { 2928 sc->enable_hwlro = strtol(value, NULL, 10); 2929 if(sc->enable_hwlro) { 2930 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL); 2931 if(rc) { 2932 device_printf(sc->dev, "no hardware lro support\n"); 2933 device_printf(sc->dev, "software lro enabled\n"); 2934 sc->enable_hwlro = 0; 2935 }else { 2936 device_printf(sc->dev, "hardware lro enabled\n"); 2937 oce_max_rsp_handled = 32; 2938 } 2939 }else { 2940 device_printf(sc->dev, "software lro enabled\n"); 2941 } 2942 }else { 2943 sc->enable_hwlro = 0; 2944 } 2945 2946 /* read mbuf size */ 2947 //value = getenv("oce_rq_buf_size"); 2948 if(value && IS_SH(sc)) { 2949 oce_rq_buf_size = strtol(value, NULL, 10); 2950 switch(oce_rq_buf_size) { 2951 case 2048: 2952 case 4096: 2953 case 9216: 2954 case 16384: 2955 break; 2956 2957 default: 2958 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n"); 2959 oce_rq_buf_size = 2048; 2960 } 2961 } 2962 2963 return; 2964 } 2965