sys/dev/netif/igb/if_igb.c

   1 /*
   2  * Copyright (c) 2001-2013, Intel Corporation
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions are met:
   7  *
   8  *  1. Redistributions of source code must retain the above copyright notice,
   9  *     this list of conditions and the following disclaimer.
  10  *
  11  *  2. Redistributions in binary form must reproduce the above copyright
  12  *     notice, this list of conditions and the following disclaimer in the
  13  *     documentation and/or other materials provided with the distribution.
  14  *
  15  *  3. Neither the name of the Intel Corporation nor the names of its
  16  *     contributors may be used to endorse or promote products derived from
  17  *     this software without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  23  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 #include "opt_ifpoll.h"
  33 #include "opt_igb.h"
  34
  35 #include <sys/param.h>
  36 #include <sys/bus.h>
  37 #include <sys/endian.h>
  38 #include <sys/interrupt.h>
  39 #include <sys/kernel.h>
  40 #include <sys/malloc.h>
  41 #include <sys/mbuf.h>
  42 #include <sys/proc.h>
  43 #include <sys/rman.h>
  44 #include <sys/serialize.h>
  45 #include <sys/serialize2.h>
  46 #include <sys/socket.h>
  47 #include <sys/sockio.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/systm.h>
  50
  51 #include <net/bpf.h>
  52 #include <net/ethernet.h>
  53 #include <net/if.h>
  54 #include <net/if_arp.h>
  55 #include <net/if_dl.h>
  56 #include <net/if_media.h>
  57 #include <net/ifq_var.h>
  58 #include <net/if_ringmap.h>
  59 #include <net/toeplitz.h>
  60 #include <net/toeplitz2.h>
  61 #include <net/vlan/if_vlan_var.h>
  62 #include <net/vlan/if_vlan_ether.h>
  63 #include <net/if_poll.h>
  64
  65 #include <netinet/in_systm.h>
  66 #include <netinet/in.h>
  67 #include <netinet/ip.h>
  68
  69 #include <bus/pci/pcivar.h>
  70 #include <bus/pci/pcireg.h>
  71
  72 #include <dev/netif/ig_hal/e1000_api.h>
  73 #include <dev/netif/ig_hal/e1000_82575.h>
  74 #include <dev/netif/ig_hal/e1000_dragonfly.h>
  75 #include <dev/netif/igb/if_igb.h>
  76
  77 #ifdef IGB_RSS_DEBUG
  78 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) \
  79 do { \
  80         if (sc->rss_debug >= lvl) \
  81                 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
  82 } while (0)
  83 #else   /* !IGB_RSS_DEBUG */
  84 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...)      ((void)0)
  85 #endif  /* IGB_RSS_DEBUG */
  86
  87 #define IGB_NAME        "Intel(R) PRO/1000 "
  88 #define IGB_DEVICE(id)  \
  89         { IGB_VENDOR_ID, E1000_DEV_ID_##id, IGB_NAME #id }
  90 #define IGB_DEVICE_NULL { 0, 0, NULL }
  91
  92 static struct igb_device {
  93         uint16_t        vid;
  94         uint16_t        did;
  95         const char      *desc;
  96 } igb_devices[] = {
  97         IGB_DEVICE(82575EB_COPPER),
  98         IGB_DEVICE(82575EB_FIBER_SERDES),
  99         IGB_DEVICE(82575GB_QUAD_COPPER),
 100         IGB_DEVICE(82576),
 101         IGB_DEVICE(82576_NS),
 102         IGB_DEVICE(82576_NS_SERDES),
 103         IGB_DEVICE(82576_FIBER),
 104         IGB_DEVICE(82576_SERDES),
 105         IGB_DEVICE(82576_SERDES_QUAD),
 106         IGB_DEVICE(82576_QUAD_COPPER),
 107         IGB_DEVICE(82576_QUAD_COPPER_ET2),
 108         IGB_DEVICE(82576_VF),
 109         IGB_DEVICE(82580_COPPER),
 110         IGB_DEVICE(82580_FIBER),
 111         IGB_DEVICE(82580_SERDES),
 112         IGB_DEVICE(82580_SGMII),
 113         IGB_DEVICE(82580_COPPER_DUAL),
 114         IGB_DEVICE(82580_QUAD_FIBER),
 115         IGB_DEVICE(DH89XXCC_SERDES),
 116         IGB_DEVICE(DH89XXCC_SGMII),
 117         IGB_DEVICE(DH89XXCC_SFP),
 118         IGB_DEVICE(DH89XXCC_BACKPLANE),
 119         IGB_DEVICE(I350_COPPER),
 120         IGB_DEVICE(I350_FIBER),
 121         IGB_DEVICE(I350_SERDES),
 122         IGB_DEVICE(I350_SGMII),
 123         IGB_DEVICE(I350_VF),
 124         IGB_DEVICE(I210_COPPER),
 125         IGB_DEVICE(I210_COPPER_IT),
 126         IGB_DEVICE(I210_COPPER_OEM1),
 127         IGB_DEVICE(I210_COPPER_FLASHLESS),
 128         IGB_DEVICE(I210_SERDES_FLASHLESS),
 129         IGB_DEVICE(I210_FIBER),
 130         IGB_DEVICE(I210_SERDES),
 131         IGB_DEVICE(I210_SGMII),
 132         IGB_DEVICE(I211_COPPER),
 133         IGB_DEVICE(I354_BACKPLANE_1GBPS),
 134         IGB_DEVICE(I354_BACKPLANE_2_5GBPS),
 135         IGB_DEVICE(I354_SGMII),
 136
 137         /* required last entry */
 138         IGB_DEVICE_NULL
 139 };
 140
 141 static int      igb_probe(device_t);
 142 static int      igb_attach(device_t);
 143 static int      igb_detach(device_t);
 144 static int      igb_shutdown(device_t);
 145 static int      igb_suspend(device_t);
 146 static int      igb_resume(device_t);
 147
 148 static boolean_t igb_is_valid_ether_addr(const uint8_t *);
 149 static void     igb_setup_ifp(struct igb_softc *);
 150 static boolean_t igb_txcsum_ctx(struct igb_tx_ring *, struct mbuf *);
 151 static int      igb_tso_pullup(struct igb_tx_ring *, struct mbuf **);
 152 static void     igb_tso_ctx(struct igb_tx_ring *, struct mbuf *, uint32_t *);
 153 static void     igb_add_sysctl(struct igb_softc *);
 154 static void     igb_add_intr_rate_sysctl(struct igb_softc *, int,
 155                     const char *, const char *);
 156 static int      igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS);
 157 static int      igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS);
 158 static int      igb_sysctl_tx_wreg_nsegs(SYSCTL_HANDLER_ARGS);
 159 static int      igb_sysctl_rx_wreg_nsegs(SYSCTL_HANDLER_ARGS);
 160 static void     igb_set_ring_inuse(struct igb_softc *, boolean_t);
 161 static int      igb_get_rxring_inuse(const struct igb_softc *, boolean_t);
 162 static int      igb_get_txring_inuse(const struct igb_softc *, boolean_t);
 163 static void     igb_set_timer_cpuid(struct igb_softc *, boolean_t);
 164
 165 static void     igb_vf_init_stats(struct igb_softc *);
 166 static void     igb_reset(struct igb_softc *, boolean_t);
 167 static void     igb_update_stats_counters(struct igb_softc *);
 168 static void     igb_update_vf_stats_counters(struct igb_softc *);
 169 static void     igb_update_link_status(struct igb_softc *);
 170 static void     igb_init_tx_unit(struct igb_softc *);
 171 static void     igb_init_rx_unit(struct igb_softc *, boolean_t);
 172 static void     igb_init_dmac(struct igb_softc *, uint32_t);
 173 static void     igb_reg_dump(struct igb_softc *);
 174 static int      igb_sysctl_reg_dump(SYSCTL_HANDLER_ARGS);
 175
 176 static void     igb_set_vlan(struct igb_softc *);
 177 static void     igb_set_multi(struct igb_softc *);
 178 static void     igb_set_promisc(struct igb_softc *);
 179 static void     igb_disable_promisc(struct igb_softc *);
 180
 181 static int      igb_get_ring_max(const struct igb_softc *);
 182 static void     igb_get_rxring_cnt(const struct igb_softc *, int *, int *);
 183 static void     igb_get_txring_cnt(const struct igb_softc *, int *, int *);
 184 static int      igb_alloc_rings(struct igb_softc *);
 185 static void     igb_free_rings(struct igb_softc *);
 186 static int      igb_create_tx_ring(struct igb_tx_ring *);
 187 static int      igb_create_rx_ring(struct igb_rx_ring *);
 188 static void     igb_free_tx_ring(struct igb_tx_ring *);
 189 static void     igb_free_rx_ring(struct igb_rx_ring *);
 190 static void     igb_destroy_tx_ring(struct igb_tx_ring *, int);
 191 static void     igb_destroy_rx_ring(struct igb_rx_ring *, int);
 192 static void     igb_init_tx_ring(struct igb_tx_ring *);
 193 static int      igb_init_rx_ring(struct igb_rx_ring *);
 194 static int      igb_newbuf(struct igb_rx_ring *, int, boolean_t);
 195 static int      igb_encap(struct igb_tx_ring *, struct mbuf **, int *, int *);
 196 static void     igb_rx_refresh(struct igb_rx_ring *, int);
 197 static void     igb_setup_serialize(struct igb_softc *);
 198
 199 static void     igb_stop(struct igb_softc *);
 200 static void     igb_init(void *);
 201 static int      igb_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 202 static void     igb_media_status(struct ifnet *, struct ifmediareq *);
 203 static int      igb_media_change(struct ifnet *);
 204 static void     igb_timer(void *);
 205 static void     igb_watchdog(struct ifaltq_subque *);
 206 static void     igb_start(struct ifnet *, struct ifaltq_subque *);
 207 #ifdef IFPOLL_ENABLE
 208 static void     igb_npoll(struct ifnet *, struct ifpoll_info *);
 209 static void     igb_npoll_rx(struct ifnet *, void *, int);
 210 static void     igb_npoll_tx(struct ifnet *, void *, int);
 211 static void     igb_npoll_status(struct ifnet *);
 212 #endif
 213 static void     igb_serialize(struct ifnet *, enum ifnet_serialize);
 214 static void     igb_deserialize(struct ifnet *, enum ifnet_serialize);
 215 static int      igb_tryserialize(struct ifnet *, enum ifnet_serialize);
 216 #ifdef INVARIANTS
 217 static void     igb_serialize_assert(struct ifnet *, enum ifnet_serialize,
 218                     boolean_t);
 219 #endif
 220
 221 static void     igb_intr(void *);
 222 static void     igb_intr_shared(void *);
 223 static void     igb_rxeof(struct igb_rx_ring *, int);
 224 static void     igb_txeof(struct igb_tx_ring *, int);
 225 static void     igb_txgc(struct igb_tx_ring *);
 226 static void     igb_txgc_timer(void *);
 227 static void     igb_set_eitr(struct igb_softc *, int, int);
 228 static void     igb_enable_intr(struct igb_softc *);
 229 static void     igb_disable_intr(struct igb_softc *);
 230 static void     igb_init_unshared_intr(struct igb_softc *);
 231 static void     igb_init_intr(struct igb_softc *);
 232 static int      igb_setup_intr(struct igb_softc *);
 233 static void     igb_set_txintr_mask(struct igb_tx_ring *, int *, int);
 234 static void     igb_set_rxintr_mask(struct igb_rx_ring *, int *, int);
 235 static void     igb_set_intr_mask(struct igb_softc *);
 236 static int      igb_alloc_intr(struct igb_softc *);
 237 static void     igb_free_intr(struct igb_softc *);
 238 static void     igb_teardown_intr(struct igb_softc *, int);
 239 static void     igb_alloc_msix(struct igb_softc *);
 240 static void     igb_free_msix(struct igb_softc *, boolean_t);
 241 static void     igb_msix_rx(void *);
 242 static void     igb_msix_tx(void *);
 243 static void     igb_msix_status(void *);
 244 static void     igb_msix_rxtx(void *);
 245
 246 /* Management and WOL Support */
 247 static void     igb_get_mgmt(struct igb_softc *);
 248 static void     igb_rel_mgmt(struct igb_softc *);
 249 static void     igb_get_hw_control(struct igb_softc *);
 250 static void     igb_rel_hw_control(struct igb_softc *);
 251 static void     igb_enable_wol(struct igb_softc *);
 252 static int      igb_enable_phy_wol(struct igb_softc *);
 253
 254 static device_method_t igb_methods[] = {
 255         /* Device interface */
 256         DEVMETHOD(device_probe,         igb_probe),
 257         DEVMETHOD(device_attach,        igb_attach),
 258         DEVMETHOD(device_detach,        igb_detach),
 259         DEVMETHOD(device_shutdown,      igb_shutdown),
 260         DEVMETHOD(device_suspend,       igb_suspend),
 261         DEVMETHOD(device_resume,        igb_resume),
 262         DEVMETHOD_END
 263 };
 264
 265 static driver_t igb_driver = {
 266         "igb",
 267         igb_methods,
 268         sizeof(struct igb_softc),
 269 };
 270
 271 static devclass_t igb_devclass;
 272
 273 DECLARE_DUMMY_MODULE(if_igb);
 274 MODULE_DEPEND(igb, ig_hal, 1, 1, 1);
 275 DRIVER_MODULE(if_igb, pci, igb_driver, igb_devclass, NULL, NULL);
 276
 277 static int      igb_rxd = IGB_DEFAULT_RXD;
 278 static int      igb_txd = IGB_DEFAULT_TXD;
 279 static int      igb_rxr = 0;
 280 static int      igb_txr = 0;
 281 static int      igb_msi_enable = 1;
 282 static int      igb_msix_enable = 1;
 283 static int      igb_eee_disabled = 1;   /* Energy Efficient Ethernet */
 284
 285 static char     igb_flowctrl[IFM_ETH_FC_STRLEN] = IFM_ETH_FC_NONE;
 286
 287 /*
 288  * DMA Coalescing, only for i350 - default to off,
 289  * this feature is for power savings
 290  */
 291 static int      igb_dma_coalesce = 0;
 292
 293 TUNABLE_INT("hw.igb.rxd", &igb_rxd);
 294 TUNABLE_INT("hw.igb.txd", &igb_txd);
 295 TUNABLE_INT("hw.igb.rxr", &igb_rxr);
 296 TUNABLE_INT("hw.igb.txr", &igb_txr);
 297 TUNABLE_INT("hw.igb.msi.enable", &igb_msi_enable);
 298 TUNABLE_INT("hw.igb.msix.enable", &igb_msix_enable);
 299 TUNABLE_STR("hw.igb.flow_ctrl", igb_flowctrl, sizeof(igb_flowctrl));
 300
 301 /* i350 specific */
 302 TUNABLE_INT("hw.igb.eee_disabled", &igb_eee_disabled);
 303 TUNABLE_INT("hw.igb.dma_coalesce", &igb_dma_coalesce);
 304
 305 static __inline void
 306 igb_tx_intr(struct igb_tx_ring *txr, int hdr)
 307 {
 308
 309         igb_txeof(txr, hdr);
 310         if (!ifsq_is_empty(txr->ifsq))
 311                 ifsq_devstart(txr->ifsq);
 312 }
 313
 314 static __inline void
 315 igb_try_txgc(struct igb_tx_ring *txr, int16_t dec)
 316 {
 317
 318         if (txr->tx_running > 0) {
 319                 txr->tx_running -= dec;
 320                 if (txr->tx_running <= 0 && txr->tx_nmbuf &&
 321                     txr->tx_avail < txr->num_tx_desc &&
 322                     txr->tx_avail + txr->intr_nsegs > txr->num_tx_desc)
 323                         igb_txgc(txr);
 324         }
 325 }
 326
 327 static void
 328 igb_txgc_timer(void *xtxr)
 329 {
 330         struct igb_tx_ring *txr = xtxr;
 331         struct ifnet *ifp = &txr->sc->arpcom.ac_if;
 332
 333         if ((ifp->if_flags & (IFF_RUNNING | IFF_UP | IFF_NPOLLING)) !=
 334             (IFF_RUNNING | IFF_UP))
 335                 return;
 336
 337         if (!lwkt_serialize_try(&txr->tx_serialize))
 338                 goto done;
 339
 340         if ((ifp->if_flags & (IFF_RUNNING | IFF_UP | IFF_NPOLLING)) !=
 341             (IFF_RUNNING | IFF_UP)) {
 342                 lwkt_serialize_exit(&txr->tx_serialize);
 343                 return;
 344         }
 345         igb_try_txgc(txr, IGB_TX_RUNNING_DEC);
 346
 347         lwkt_serialize_exit(&txr->tx_serialize);
 348 done:
 349         callout_reset(&txr->tx_gc_timer, 1, igb_txgc_timer, txr);
 350 }
 351
 352 static __inline void
 353 igb_free_txbuf(struct igb_tx_ring *txr, struct igb_tx_buf *txbuf)
 354 {
 355
 356         KKASSERT(txbuf->m_head != NULL);
 357         KKASSERT(txr->tx_nmbuf > 0);
 358         txr->tx_nmbuf--;
 359
 360         bus_dmamap_unload(txr->tx_tag, txbuf->map);
 361         m_freem(txbuf->m_head);
 362         txbuf->m_head = NULL;
 363 }
 364
 365 static __inline void
 366 igb_rxcsum(uint32_t staterr, struct mbuf *mp)
 367 {
 368         /* Ignore Checksum bit is set */
 369         if (staterr & E1000_RXD_STAT_IXSM)
 370                 return;
 371
 372         if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
 373             E1000_RXD_STAT_IPCS)
 374                 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 375
 376         if (staterr & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
 377                 if ((staterr & E1000_RXDEXT_STATERR_TCPE) == 0) {
 378                         mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 379                             CSUM_PSEUDO_HDR | CSUM_FRAG_NOT_CHECKED;
 380                         mp->m_pkthdr.csum_data = htons(0xffff);
 381                 }
 382         }
 383 }
 384
 385 static __inline struct pktinfo *
 386 igb_rssinfo(struct mbuf *m, struct pktinfo *pi,
 387     uint32_t hash, uint32_t hashtype, uint32_t staterr)
 388 {
 389         switch (hashtype) {
 390         case E1000_RXDADV_RSSTYPE_IPV4_TCP:
 391                 pi->pi_netisr = NETISR_IP;
 392                 pi->pi_flags = 0;
 393                 pi->pi_l3proto = IPPROTO_TCP;
 394                 break;
 395
 396         case E1000_RXDADV_RSSTYPE_IPV4:
 397                 if (staterr & E1000_RXD_STAT_IXSM)
 398                         return NULL;
 399
 400                 if ((staterr &
 401                      (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
 402                     E1000_RXD_STAT_TCPCS) {
 403                         pi->pi_netisr = NETISR_IP;
 404                         pi->pi_flags = 0;
 405                         pi->pi_l3proto = IPPROTO_UDP;
 406                         break;
 407                 }
 408                 /* FALL THROUGH */
 409         default:
 410                 return NULL;
 411         }
 412
 413         m_sethash(m, toeplitz_hash(hash));
 414         return pi;
 415 }
 416
 417 static int
 418 igb_get_ring_max(const struct igb_softc *sc)
 419 {
 420
 421         switch (sc->hw.mac.type) {
 422         case e1000_82575:
 423                 return (IGB_MAX_RING_82575);
 424
 425         case e1000_82576:
 426                 return (IGB_MAX_RING_82576);
 427
 428         case e1000_82580:
 429                 return (IGB_MAX_RING_82580);
 430
 431         case e1000_i350:
 432                 return (IGB_MAX_RING_I350);
 433
 434         case e1000_i354:
 435                 return (IGB_MAX_RING_I354);
 436
 437         case e1000_i210:
 438                 return (IGB_MAX_RING_I210);
 439
 440         case e1000_i211:
 441                 return (IGB_MAX_RING_I211);
 442
 443         default:
 444                 return (IGB_MIN_RING);
 445         }
 446 }
 447
 448 static void
 449 igb_get_rxring_cnt(const struct igb_softc *sc, int *ring_cnt, int *ring_max)
 450 {
 451
 452         *ring_max = igb_get_ring_max(sc);
 453         *ring_cnt = device_getenv_int(sc->dev, "rxr", igb_rxr);
 454 }
 455
 456 static void
 457 igb_get_txring_cnt(const struct igb_softc *sc, int *ring_cnt, int *ring_max)
 458 {
 459
 460         *ring_max = igb_get_ring_max(sc);
 461         *ring_cnt = device_getenv_int(sc->dev, "txr", igb_txr);
 462 }
 463
 464 static int
 465 igb_probe(device_t dev)
 466 {
 467         const struct igb_device *d;
 468         uint16_t vid, did;
 469
 470         vid = pci_get_vendor(dev);
 471         did = pci_get_device(dev);
 472
 473         for (d = igb_devices; d->desc != NULL; ++d) {
 474                 if (vid == d->vid && did == d->did) {
 475                         device_set_desc(dev, d->desc);
 476                         return 0;
 477                 }
 478         }
 479         return ENXIO;
 480 }
 481
 482 static int
 483 igb_attach(device_t dev)
 484 {
 485         struct igb_softc *sc = device_get_softc(dev);
 486         uint16_t eeprom_data;
 487         int error = 0, ring_max, ring_cnt;
 488         char flowctrl[IFM_ETH_FC_STRLEN];
 489
 490 #ifdef notyet
 491         /* SYSCTL stuff */
 492         SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
 493             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
 494             OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
 495             igb_sysctl_nvm_info, "I", "NVM Information");
 496 #endif
 497
 498         ifmedia_init(&sc->media, IFM_IMASK | IFM_ETH_FCMASK,
 499             igb_media_change, igb_media_status);
 500         callout_init_mp(&sc->timer);
 501         lwkt_serialize_init(&sc->main_serialize);
 502
 503         if_initname(&sc->arpcom.ac_if, device_get_name(dev),
 504             device_get_unit(dev));
 505         sc->dev = sc->osdep.dev = dev;
 506
 507         /* Enable bus mastering */
 508         pci_enable_busmaster(dev);
 509
 510         /*
 511          * Determine hardware and mac type
 512          */
 513         sc->hw.vendor_id = pci_get_vendor(dev);
 514         sc->hw.device_id = pci_get_device(dev);
 515         sc->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
 516         sc->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2);
 517         sc->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2);
 518
 519         if (e1000_set_mac_type(&sc->hw))
 520                 return ENXIO;
 521
 522         /* Are we a VF device? */
 523         if (sc->hw.mac.type == e1000_vfadapt ||
 524             sc->hw.mac.type == e1000_vfadapt_i350)
 525                 sc->vf_ifp = 1;
 526         else
 527                 sc->vf_ifp = 0;
 528
 529         /*
 530          * Configure total supported RX/TX ring count
 531          */
 532         igb_get_rxring_cnt(sc, &ring_cnt, &ring_max);
 533         sc->rx_rmap = if_ringmap_alloc(dev, ring_cnt, ring_max);
 534         igb_get_txring_cnt(sc, &ring_cnt, &ring_max);
 535         sc->tx_rmap = if_ringmap_alloc(dev, ring_cnt, ring_max);
 536         if_ringmap_match(dev, sc->rx_rmap, sc->tx_rmap);
 537
 538         sc->rx_ring_cnt = if_ringmap_count(sc->rx_rmap);
 539         sc->rx_ring_inuse = sc->rx_ring_cnt;
 540         sc->tx_ring_cnt = if_ringmap_count(sc->tx_rmap);
 541         sc->tx_ring_inuse = sc->tx_ring_cnt;
 542
 543         /* Setup flow control. */
 544         device_getenv_string(dev, "flow_ctrl", flowctrl, sizeof(flowctrl),
 545             igb_flowctrl);
 546         sc->ifm_flowctrl = ifmedia_str2ethfc(flowctrl);
 547
 548         /*
 549          * Allocate IO memory
 550          */
 551         sc->mem_rid = PCIR_BAR(0);
 552         sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
 553             RF_ACTIVE);
 554         if (sc->mem_res == NULL) {
 555                 device_printf(dev, "Unable to allocate bus resource: memory\n");
 556                 error = ENXIO;
 557                 goto failed;
 558         }
 559         sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->mem_res);
 560         sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->mem_res);
 561
 562         sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
 563
 564         /* Save PCI command register for Shared Code */
 565         sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
 566         sc->hw.back = &sc->osdep;
 567
 568         /* Do Shared Code initialization */
 569         if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
 570                 device_printf(dev, "Setup of Shared code failed\n");
 571                 error = ENXIO;
 572                 goto failed;
 573         }
 574
 575         e1000_get_bus_info(&sc->hw);
 576
 577         sc->hw.mac.autoneg = DO_AUTO_NEG;
 578         sc->hw.phy.autoneg_wait_to_complete = FALSE;
 579         sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
 580
 581         /* Copper options */
 582         if (sc->hw.phy.media_type == e1000_media_type_copper) {
 583                 sc->hw.phy.mdix = AUTO_ALL_MODES;
 584                 sc->hw.phy.disable_polarity_correction = FALSE;
 585                 sc->hw.phy.ms_type = IGB_MASTER_SLAVE;
 586         }
 587
 588         /* Set the frame limits assuming  standard ethernet sized frames. */
 589         sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
 590
 591         /* Allocate RX/TX rings */
 592         error = igb_alloc_rings(sc);
 593         if (error)
 594                 goto failed;
 595
 596         /* Allocate interrupt */
 597         error = igb_alloc_intr(sc);
 598         if (error)
 599                 goto failed;
 600
 601         /* Setup serializes */
 602         igb_setup_serialize(sc);
 603
 604         /* Allocate the appropriate stats memory */
 605         if (sc->vf_ifp) {
 606                 sc->stats = kmalloc(sizeof(struct e1000_vf_stats), M_DEVBUF,
 607                     M_WAITOK | M_ZERO);
 608                 igb_vf_init_stats(sc);
 609         } else {
 610                 sc->stats = kmalloc(sizeof(struct e1000_hw_stats), M_DEVBUF,
 611                     M_WAITOK | M_ZERO);
 612         }
 613
 614         /* Allocate multicast array memory. */
 615         sc->mta = kmalloc(ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES,
 616             M_DEVBUF, M_WAITOK);
 617
 618         /* Some adapter-specific advanced features */
 619         if (sc->hw.mac.type >= e1000_i350) {
 620 #ifdef notyet
 621                 igb_set_sysctl_value(adapter, "dma_coalesce",
 622                     "configure dma coalesce",
 623                     &adapter->dma_coalesce, igb_dma_coalesce);
 624                 igb_set_sysctl_value(adapter, "eee_disabled",
 625                     "enable Energy Efficient Ethernet",
 626                     &adapter->hw.dev_spec._82575.eee_disable,
 627                     igb_eee_disabled);
 628 #else
 629                 sc->dma_coalesce = igb_dma_coalesce;
 630                 sc->hw.dev_spec._82575.eee_disable = igb_eee_disabled;
 631 #endif
 632                 if (sc->hw.phy.media_type == e1000_media_type_copper) {
 633                         if (sc->hw.mac.type == e1000_i354)
 634                                 e1000_set_eee_i354(&sc->hw, TRUE, TRUE);
 635                         else
 636                                 e1000_set_eee_i350(&sc->hw, TRUE, TRUE);
 637                 }
 638         }
 639
 640         /*
 641          * Start from a known state, this is important in reading the nvm and
 642          * mac from that.
 643          */
 644         e1000_reset_hw(&sc->hw);
 645
 646         /* Make sure we have a good EEPROM before we read from it */
 647         if (sc->hw.mac.type != e1000_i210 && sc->hw.mac.type != e1000_i211 &&
 648             e1000_validate_nvm_checksum(&sc->hw) < 0) {
 649                 /*
 650                  * Some PCI-E parts fail the first check due to
 651                  * the link being in sleep state, call it again,
 652                  * if it fails a second time its a real issue.
 653                  */
 654                 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
 655                         device_printf(dev,
 656                             "The EEPROM Checksum Is Not Valid\n");
 657                         error = EIO;
 658                         goto failed;
 659                 }
 660         }
 661
 662         /* Copy the permanent MAC address out of the EEPROM */
 663         if (e1000_read_mac_addr(&sc->hw) < 0) {
 664                 device_printf(dev, "EEPROM read error while reading MAC"
 665                     " address\n");
 666                 error = EIO;
 667                 goto failed;
 668         }
 669         if (!igb_is_valid_ether_addr(sc->hw.mac.addr)) {
 670                 device_printf(dev, "Invalid MAC address\n");
 671                 error = EIO;
 672                 goto failed;
 673         }
 674
 675         /* Setup OS specific network interface */
 676         igb_setup_ifp(sc);
 677
 678         /* Add sysctl tree, must after igb_setup_ifp() */
 679         igb_add_sysctl(sc);
 680
 681         /* Now get a good starting state */
 682         igb_reset(sc, FALSE);
 683
 684         /* Initialize statistics */
 685         igb_update_stats_counters(sc);
 686
 687         sc->hw.mac.get_link_status = 1;
 688         igb_update_link_status(sc);
 689
 690         /* Indicate SOL/IDER usage */
 691         if (e1000_check_reset_block(&sc->hw)) {
 692                 device_printf(dev,
 693                     "PHY reset is blocked due to SOL/IDER session.\n");
 694         }
 695
 696         /* Determine if we have to control management hardware */
 697         if (e1000_enable_mng_pass_thru(&sc->hw))
 698                 sc->flags |= IGB_FLAG_HAS_MGMT;
 699
 700         /*
 701          * Setup Wake-on-Lan
 702          */
 703         /* APME bit in EEPROM is mapped to WUC.APME */
 704         eeprom_data = E1000_READ_REG(&sc->hw, E1000_WUC) & E1000_WUC_APME;
 705         if (eeprom_data) {
 706                 /* XXX E1000_WUFC_MC always be cleared from E1000_WUC. */
 707                 sc->wol = E1000_WUFC_MAG | E1000_WUFC_MC;
 708                 device_printf(dev, "has WOL\n");
 709         }
 710
 711 #ifdef notyet
 712         /* Register for VLAN events */
 713         adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
 714              igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
 715         adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
 716              igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
 717 #endif
 718
 719 #ifdef notyet
 720         igb_add_hw_stats(adapter);
 721 #endif
 722
 723         /*
 724          * Disable interrupt to prevent spurious interrupts (line based
 725          * interrupt, MSI or even MSI-X), which had been observed on
 726          * several types of LOMs, from being handled.
 727          */
 728         igb_disable_intr(sc);
 729
 730         error = igb_setup_intr(sc);
 731         if (error) {
 732                 ether_ifdetach(&sc->arpcom.ac_if);
 733                 goto failed;
 734         }
 735         return 0;
 736
 737 failed:
 738         igb_detach(dev);
 739         return error;
 740 }
 741
 742 static int
 743 igb_detach(device_t dev)
 744 {
 745         struct igb_softc *sc = device_get_softc(dev);
 746
 747         if (device_is_attached(dev)) {
 748                 struct ifnet *ifp = &sc->arpcom.ac_if;
 749
 750                 ifnet_serialize_all(ifp);
 751
 752                 igb_stop(sc);
 753
 754                 e1000_phy_hw_reset(&sc->hw);
 755
 756                 /* Give control back to firmware */
 757                 igb_rel_mgmt(sc);
 758                 igb_rel_hw_control(sc);
 759                 igb_enable_wol(sc);
 760
 761                 igb_teardown_intr(sc, sc->intr_cnt);
 762
 763                 ifnet_deserialize_all(ifp);
 764
 765                 ether_ifdetach(ifp);
 766         } else if (sc->mem_res != NULL) {
 767                 igb_rel_hw_control(sc);
 768         }
 769
 770         ifmedia_removeall(&sc->media);
 771         bus_generic_detach(dev);
 772
 773         igb_free_intr(sc);
 774
 775         if (sc->msix_mem_res != NULL) {
 776                 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_mem_rid,
 777                     sc->msix_mem_res);
 778         }
 779         if (sc->mem_res != NULL) {
 780                 bus_release_resource(dev, SYS_RES_MEMORY, sc->mem_rid,
 781                     sc->mem_res);
 782         }
 783
 784         igb_free_rings(sc);
 785
 786         if (sc->mta != NULL)
 787                 kfree(sc->mta, M_DEVBUF);
 788         if (sc->stats != NULL)
 789                 kfree(sc->stats, M_DEVBUF);
 790         if (sc->serializes != NULL)
 791                 kfree(sc->serializes, M_DEVBUF);
 792         if (sc->rx_rmap != NULL)
 793                 if_ringmap_free(sc->rx_rmap);
 794         if (sc->rx_rmap_intr != NULL)
 795                 if_ringmap_free(sc->rx_rmap_intr);
 796         if (sc->tx_rmap != NULL)
 797                 if_ringmap_free(sc->tx_rmap);
 798         if (sc->tx_rmap_intr != NULL)
 799                 if_ringmap_free(sc->tx_rmap_intr);
 800
 801         return 0;
 802 }
 803
 804 static int
 805 igb_shutdown(device_t dev)
 806 {
 807         return igb_suspend(dev);
 808 }
 809
 810 static int
 811 igb_suspend(device_t dev)
 812 {
 813         struct igb_softc *sc = device_get_softc(dev);
 814         struct ifnet *ifp = &sc->arpcom.ac_if;
 815
 816         ifnet_serialize_all(ifp);
 817
 818         igb_stop(sc);
 819
 820         igb_rel_mgmt(sc);
 821         igb_rel_hw_control(sc);
 822         igb_enable_wol(sc);
 823
 824         ifnet_deserialize_all(ifp);
 825
 826         return bus_generic_suspend(dev);
 827 }
 828
 829 static int
 830 igb_resume(device_t dev)
 831 {
 832         struct igb_softc *sc = device_get_softc(dev);
 833         struct ifnet *ifp = &sc->arpcom.ac_if;
 834         int i;
 835
 836         ifnet_serialize_all(ifp);
 837
 838         igb_init(sc);
 839         igb_get_mgmt(sc);
 840
 841         for (i = 0; i < sc->tx_ring_inuse; ++i)
 842                 ifsq_devstart_sched(sc->tx_rings[i].ifsq);
 843
 844         ifnet_deserialize_all(ifp);
 845
 846         return bus_generic_resume(dev);
 847 }
 848
 849 static int
 850 igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
 851 {
 852         struct igb_softc *sc = ifp->if_softc;
 853         struct ifreq *ifr = (struct ifreq *)data;
 854         int max_frame_size, mask, reinit;
 855         int error = 0;
 856
 857         ASSERT_IFNET_SERIALIZED_ALL(ifp);
 858
 859         switch (command) {
 860         case SIOCSIFMTU:
 861                 max_frame_size = 9234;
 862                 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
 863                     ETHER_CRC_LEN) {
 864                         error = EINVAL;
 865                         break;
 866                 }
 867
 868                 ifp->if_mtu = ifr->ifr_mtu;
 869                 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
 870                     ETHER_CRC_LEN;
 871
 872                 if (ifp->if_flags & IFF_RUNNING)
 873                         igb_init(sc);
 874                 break;
 875
 876         case SIOCSIFFLAGS:
 877                 if (ifp->if_flags & IFF_UP) {
 878                         if (ifp->if_flags & IFF_RUNNING) {
 879                                 if ((ifp->if_flags ^ sc->if_flags) &
 880                                     (IFF_PROMISC | IFF_ALLMULTI)) {
 881                                         igb_disable_promisc(sc);
 882                                         igb_set_promisc(sc);
 883                                 }
 884                         } else {
 885                                 igb_init(sc);
 886                         }
 887                 } else if (ifp->if_flags & IFF_RUNNING) {
 888                         igb_stop(sc);
 889                 }
 890                 sc->if_flags = ifp->if_flags;
 891                 break;
 892
 893         case SIOCADDMULTI:
 894         case SIOCDELMULTI:
 895                 if (ifp->if_flags & IFF_RUNNING) {
 896                         igb_disable_intr(sc);
 897                         igb_set_multi(sc);
 898 #ifdef IFPOLL_ENABLE
 899                         if (!(ifp->if_flags & IFF_NPOLLING))
 900 #endif
 901                                 igb_enable_intr(sc);
 902                 }
 903                 break;
 904
 905         case SIOCSIFMEDIA:
 906                 /* Check SOL/IDER usage */
 907                 if (e1000_check_reset_block(&sc->hw)) {
 908                         if_printf(ifp, "Media change is "
 909                             "blocked due to SOL/IDER session.\n");
 910                         break;
 911                 }
 912                 /* FALL THROUGH */
 913
 914         case SIOCGIFMEDIA:
 915                 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
 916                 break;
 917
 918         case SIOCSIFCAP:
 919                 reinit = 0;
 920                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
 921                 if (mask & IFCAP_RXCSUM) {
 922                         ifp->if_capenable ^= IFCAP_RXCSUM;
 923                         reinit = 1;
 924                 }
 925                 if (mask & IFCAP_VLAN_HWTAGGING) {
 926                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
 927                         reinit = 1;
 928                 }
 929                 if (mask & IFCAP_TXCSUM) {
 930                         ifp->if_capenable ^= IFCAP_TXCSUM;
 931                         if (ifp->if_capenable & IFCAP_TXCSUM)
 932                                 ifp->if_hwassist |= IGB_CSUM_FEATURES;
 933                         else
 934                                 ifp->if_hwassist &= ~IGB_CSUM_FEATURES;
 935                 }
 936                 if (mask & IFCAP_TSO) {
 937                         ifp->if_capenable ^= IFCAP_TSO;
 938                         if (ifp->if_capenable & IFCAP_TSO)
 939                                 ifp->if_hwassist |= CSUM_TSO;
 940                         else
 941                                 ifp->if_hwassist &= ~CSUM_TSO;
 942                 }
 943                 if (mask & IFCAP_RSS)
 944                         ifp->if_capenable ^= IFCAP_RSS;
 945                 if (reinit && (ifp->if_flags & IFF_RUNNING))
 946                         igb_init(sc);
 947                 break;
 948
 949         default:
 950                 error = ether_ioctl(ifp, command, data);
 951                 break;
 952         }
 953         return error;
 954 }
 955
 956 static void
 957 igb_init(void *xsc)
 958 {
 959         struct igb_softc *sc = xsc;
 960         struct ifnet *ifp = &sc->arpcom.ac_if;
 961         boolean_t polling;
 962         int i;
 963
 964         ASSERT_IFNET_SERIALIZED_ALL(ifp);
 965
 966         igb_stop(sc);
 967
 968         /* Get the latest mac address, User can use a LAA */
 969         bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
 970
 971         /* Put the address into the Receive Address Array */
 972         e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
 973
 974         igb_reset(sc, FALSE);
 975         igb_update_link_status(sc);
 976
 977         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
 978
 979         /* Clear bad data from Rx FIFOs */
 980         e1000_rx_fifo_flush_82575(&sc->hw);
 981
 982         /* Configure for OS presence */
 983         igb_get_mgmt(sc);
 984
 985         polling = FALSE;
 986 #ifdef IFPOLL_ENABLE
 987         if (ifp->if_flags & IFF_NPOLLING)
 988                 polling = TRUE;
 989 #endif
 990
 991         /* Configured used RX/TX rings */
 992         igb_set_ring_inuse(sc, polling);
 993         ifq_set_subq_divisor(&ifp->if_snd, sc->tx_ring_inuse);
 994
 995         /* Initialize interrupt */
 996         igb_init_intr(sc);
 997
 998         /* Prepare transmit descriptors and buffers */
 999         for (i = 0; i < sc->tx_ring_inuse; ++i)
1000                 igb_init_tx_ring(&sc->tx_rings[i]);
1001         igb_init_tx_unit(sc);
1002
1003         /* Setup Multicast table */
1004         igb_set_multi(sc);
1005
1006 #if 0
1007         /*
1008          * Figure out the desired mbuf pool
1009          * for doing jumbo/packetsplit
1010          */
1011         if (adapter->max_frame_size <= 2048)
1012                 adapter->rx_mbuf_sz = MCLBYTES;
1013         else if (adapter->max_frame_size <= 4096)
1014                 adapter->rx_mbuf_sz = MJUMPAGESIZE;
1015         else
1016                 adapter->rx_mbuf_sz = MJUM9BYTES;
1017 #endif
1018
1019         /* Prepare receive descriptors and buffers */
1020         for (i = 0; i < sc->rx_ring_inuse; ++i) {
1021                 int error;
1022
1023                 error = igb_init_rx_ring(&sc->rx_rings[i]);
1024                 if (error) {
1025                         if_printf(ifp, "Could not setup receive structures\n");
1026                         igb_stop(sc);
1027                         return;
1028                 }
1029         }
1030         igb_init_rx_unit(sc, polling);
1031
1032         /* Enable VLAN support */
1033         if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
1034                 igb_set_vlan(sc);
1035
1036         /* Don't lose promiscuous settings */
1037         igb_set_promisc(sc);
1038
1039         /* Clear counters */
1040         e1000_clear_hw_cntrs_base_generic(&sc->hw);
1041
1042         /* This clears any pending interrupts */
1043         E1000_READ_REG(&sc->hw, E1000_ICR);
1044
1045         /*
1046          * Only enable interrupts if we are not polling, make sure
1047          * they are off otherwise.
1048          */
1049         if (polling) {
1050                 igb_disable_intr(sc);
1051         } else {
1052                 igb_enable_intr(sc);
1053                 E1000_WRITE_REG(&sc->hw, E1000_ICS, E1000_ICS_LSC);
1054         }
1055
1056         /* Set Energy Efficient Ethernet */
1057         if (sc->hw.phy.media_type == e1000_media_type_copper) {
1058                 if (sc->hw.mac.type == e1000_i354)
1059                         e1000_set_eee_i354(&sc->hw, TRUE, TRUE);
1060                 else
1061                         e1000_set_eee_i350(&sc->hw, TRUE, TRUE);
1062         }
1063
1064         ifp->if_flags |= IFF_RUNNING;
1065         for (i = 0; i < sc->tx_ring_inuse; ++i) {
1066                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1067
1068                 ifsq_clr_oactive(txr->ifsq);
1069                 ifsq_watchdog_start(&txr->tx_watchdog);
1070
1071                 if (!polling) {
1072                         callout_reset_bycpu(&txr->tx_gc_timer, 1,
1073                             igb_txgc_timer, txr, txr->tx_intr_cpuid);
1074                 }
1075         }
1076
1077         igb_set_timer_cpuid(sc, polling);
1078         callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
1079 }
1080
1081 static void
1082 igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1083 {
1084         struct igb_softc *sc = ifp->if_softc;
1085
1086         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1087
1088         if ((ifp->if_flags & IFF_RUNNING) == 0)
1089                 sc->hw.mac.get_link_status = 1;
1090         igb_update_link_status(sc);
1091
1092         ifmr->ifm_status = IFM_AVALID;
1093         ifmr->ifm_active = IFM_ETHER;
1094
1095         if (!sc->link_active) {
1096                 if (sc->hw.mac.autoneg)
1097                         ifmr->ifm_active |= IFM_NONE;
1098                 else
1099                         ifmr->ifm_active |= sc->media.ifm_media;
1100                 return;
1101         }
1102
1103         ifmr->ifm_status |= IFM_ACTIVE;
1104         if (sc->ifm_flowctrl & IFM_ETH_FORCEPAUSE)
1105                 ifmr->ifm_active |= sc->ifm_flowctrl;
1106
1107         switch (sc->link_speed) {
1108         case 10:
1109                 ifmr->ifm_active |= IFM_10_T;
1110                 break;
1111
1112         case 100:
1113                 /*
1114                  * Support for 100Mb SFP - these are Fiber
1115                  * but the media type appears as serdes
1116                  */
1117                 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1118                     sc->hw.phy.media_type == e1000_media_type_internal_serdes)
1119                         ifmr->ifm_active |= IFM_100_FX;
1120                 else
1121                         ifmr->ifm_active |= IFM_100_TX;
1122                 break;
1123
1124         case 1000:
1125                 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1126                     sc->hw.phy.media_type == e1000_media_type_internal_serdes)
1127                         ifmr->ifm_active |= IFM_1000_SX;
1128                 else
1129                         ifmr->ifm_active |= IFM_1000_T;
1130                 break;
1131
1132         case 2500:
1133                 ifmr->ifm_active |= IFM_2500_SX;
1134                 break;
1135         }
1136
1137         if (sc->link_duplex == FULL_DUPLEX)
1138                 ifmr->ifm_active |= IFM_FDX;
1139         else
1140                 ifmr->ifm_active |= IFM_HDX;
1141
1142         if (sc->link_duplex == FULL_DUPLEX)
1143                 ifmr->ifm_active |= e1000_fc2ifmedia(sc->hw.fc.current_mode);
1144 }
1145
1146 static int
1147 igb_media_change(struct ifnet *ifp)
1148 {
1149         struct igb_softc *sc = ifp->if_softc;
1150         struct ifmedia *ifm = &sc->media;
1151
1152         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1153
1154         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1155                 return EINVAL;
1156
1157         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1158         case IFM_AUTO:
1159                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1160                 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1161                 break;
1162
1163         case IFM_1000_SX:
1164         case IFM_1000_T:
1165                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1166                 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1167                 break;
1168
1169         case IFM_100_TX:
1170                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_FDX) {
1171                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1172                 } else {
1173                         if (IFM_OPTIONS(ifm->ifm_media) &
1174                             (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
1175                                 if (bootverbose) {
1176                                         if_printf(ifp, "Flow control is not "
1177                                             "allowed for half-duplex\n");
1178                                 }
1179                                 return EINVAL;
1180                         }
1181                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1182                 }
1183                 sc->hw.mac.autoneg = FALSE;
1184                 sc->hw.phy.autoneg_advertised = 0;
1185                 break;
1186
1187         case IFM_10_T:
1188                 if (IFM_OPTIONS(ifm->ifm_media) & IFM_FDX) {
1189                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1190                 } else {
1191                         if (IFM_OPTIONS(ifm->ifm_media) &
1192                             (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) {
1193                                 if (bootverbose) {
1194                                         if_printf(ifp, "Flow control is not "
1195                                             "allowed for half-duplex\n");
1196                                 }
1197                                 return EINVAL;
1198                         }
1199                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1200                 }
1201                 sc->hw.mac.autoneg = FALSE;
1202                 sc->hw.phy.autoneg_advertised = 0;
1203                 break;
1204
1205         default:
1206                 if (bootverbose) {
1207                         if_printf(ifp, "Unsupported media type %d\n",
1208                             IFM_SUBTYPE(ifm->ifm_media));
1209                 }
1210                 return EINVAL;
1211         }
1212         sc->ifm_flowctrl = ifm->ifm_media & IFM_ETH_FCMASK;
1213
1214         if (ifp->if_flags & IFF_RUNNING)
1215                 igb_init(sc);
1216
1217         return 0;
1218 }
1219
1220 static void
1221 igb_set_promisc(struct igb_softc *sc)
1222 {
1223         struct ifnet *ifp = &sc->arpcom.ac_if;
1224         struct e1000_hw *hw = &sc->hw;
1225         uint32_t reg;
1226
1227         if (sc->vf_ifp) {
1228                 e1000_promisc_set_vf(hw, e1000_promisc_enabled);
1229                 return;
1230         }
1231
1232         reg = E1000_READ_REG(hw, E1000_RCTL);
1233         if (ifp->if_flags & IFF_PROMISC) {
1234                 reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1235                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1236         } else if (ifp->if_flags & IFF_ALLMULTI) {
1237                 reg |= E1000_RCTL_MPE;
1238                 reg &= ~E1000_RCTL_UPE;
1239                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1240         }
1241 }
1242
1243 static void
1244 igb_disable_promisc(struct igb_softc *sc)
1245 {
1246         struct e1000_hw *hw = &sc->hw;
1247         struct ifnet *ifp = &sc->arpcom.ac_if;
1248         uint32_t reg;
1249         int mcnt = 0;
1250
1251         if (sc->vf_ifp) {
1252                 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
1253                 return;
1254         }
1255         reg = E1000_READ_REG(hw, E1000_RCTL);
1256         reg &= ~E1000_RCTL_UPE;
1257         if (ifp->if_flags & IFF_ALLMULTI) {
1258                 mcnt = MAX_NUM_MULTICAST_ADDRESSES;
1259         } else {
1260                 struct  ifmultiaddr *ifma;
1261                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1262                         if (ifma->ifma_addr->sa_family != AF_LINK)
1263                                 continue;
1264                         if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1265                                 break;
1266                         mcnt++;
1267                 }
1268         }
1269         /* Don't disable if in MAX groups */
1270         if (mcnt < MAX_NUM_MULTICAST_ADDRESSES)
1271                 reg &= ~E1000_RCTL_MPE;
1272         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1273 }
1274
1275 static void
1276 igb_set_multi(struct igb_softc *sc)
1277 {
1278         struct ifnet *ifp = &sc->arpcom.ac_if;
1279         struct ifmultiaddr *ifma;
1280         uint32_t reg_rctl = 0;
1281         uint8_t *mta;
1282         int mcnt = 0;
1283
1284         mta = sc->mta;
1285         bzero(mta, ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
1286
1287         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1288                 if (ifma->ifma_addr->sa_family != AF_LINK)
1289                         continue;
1290
1291                 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1292                         break;
1293
1294                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1295                     &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
1296                 mcnt++;
1297         }
1298
1299         if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1300                 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1301                 reg_rctl |= E1000_RCTL_MPE;
1302                 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1303         } else {
1304                 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1305         }
1306 }
1307
1308 static void
1309 igb_timer(void *xsc)
1310 {
1311         struct igb_softc *sc = xsc;
1312
1313         lwkt_serialize_enter(&sc->main_serialize);
1314
1315         igb_update_link_status(sc);
1316         igb_update_stats_counters(sc);
1317
1318         callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
1319
1320         lwkt_serialize_exit(&sc->main_serialize);
1321 }
1322
1323 static void
1324 igb_update_link_status(struct igb_softc *sc)
1325 {
1326         struct ifnet *ifp = &sc->arpcom.ac_if;
1327         struct e1000_hw *hw = &sc->hw;
1328         uint32_t link_check, thstat, ctrl;
1329
1330         link_check = thstat = ctrl = 0;
1331
1332         /* Get the cached link value or read for real */
1333         switch (hw->phy.media_type) {
1334         case e1000_media_type_copper:
1335                 if (hw->mac.get_link_status) {
1336                         /* Do the work to read phy */
1337                         e1000_check_for_link(hw);
1338                         link_check = !hw->mac.get_link_status;
1339                 } else {
1340                         link_check = TRUE;
1341                 }
1342                 break;
1343
1344         case e1000_media_type_fiber:
1345                 e1000_check_for_link(hw);
1346                 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1347                 break;
1348
1349         case e1000_media_type_internal_serdes:
1350                 e1000_check_for_link(hw);
1351                 link_check = hw->mac.serdes_has_link;
1352                 break;
1353
1354         /* VF device is type_unknown */
1355         case e1000_media_type_unknown:
1356                 e1000_check_for_link(hw);
1357                 link_check = !hw->mac.get_link_status;
1358                 /* Fall thru */
1359         default:
1360                 break;
1361         }
1362
1363         /* Check for thermal downshift or shutdown */
1364         if (hw->mac.type == e1000_i350) {
1365                 thstat = E1000_READ_REG(hw, E1000_THSTAT);
1366                 ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
1367         }
1368
1369         /* Now we check if a transition has happened */
1370         if (link_check && sc->link_active == 0) {
1371                 e1000_get_speed_and_duplex(hw,
1372                     &sc->link_speed, &sc->link_duplex);
1373                 if (bootverbose) {
1374                         char flowctrl[IFM_ETH_FC_STRLEN];
1375
1376                         /* Get the flow control for display */
1377                         e1000_fc2str(hw->fc.current_mode, flowctrl,
1378                             sizeof(flowctrl));
1379
1380                         if_printf(ifp, "Link is up %d Mbps %s, "
1381                             "Flow control: %s\n",
1382                             sc->link_speed,
1383                             sc->link_duplex == FULL_DUPLEX ?
1384                             "Full Duplex" : "Half Duplex",
1385                             flowctrl);
1386                 }
1387                 if (sc->ifm_flowctrl & IFM_ETH_FORCEPAUSE)
1388                         e1000_force_flowctrl(hw, sc->ifm_flowctrl);
1389                 sc->link_active = 1;
1390
1391                 ifp->if_baudrate = sc->link_speed * 1000000;
1392                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1393                     (thstat & E1000_THSTAT_LINK_THROTTLE))
1394                         if_printf(ifp, "Link: thermal downshift\n");
1395                 /* Delay Link Up for Phy update */
1396                 if ((hw->mac.type == e1000_i210 ||
1397                      hw->mac.type == e1000_i211) &&
1398                     hw->phy.id == I210_I_PHY_ID)
1399                         msec_delay(IGB_I210_LINK_DELAY);
1400                 /*
1401                  * Reset if the media type changed.
1402                  * Support AutoMediaDetect for Marvell M88 PHY in i354.
1403                  */
1404                 if (hw->dev_spec._82575.media_changed) {
1405                         hw->dev_spec._82575.media_changed = FALSE;
1406                         igb_reset(sc, TRUE);
1407                 }
1408                 /* This can sleep */
1409                 ifp->if_link_state = LINK_STATE_UP;
1410                 if_link_state_change(ifp);
1411         } else if (!link_check && sc->link_active == 1) {
1412                 ifp->if_baudrate = sc->link_speed = 0;
1413                 sc->link_duplex = 0;
1414                 if (bootverbose)
1415                         if_printf(ifp, "Link is Down\n");
1416                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1417                     (thstat & E1000_THSTAT_PWR_DOWN))
1418                         if_printf(ifp, "Link: thermal shutdown\n");
1419                 sc->link_active = 0;
1420                 /* This can sleep */
1421                 ifp->if_link_state = LINK_STATE_DOWN;
1422                 if_link_state_change(ifp);
1423         }
1424 }
1425
1426 static void
1427 igb_stop(struct igb_softc *sc)
1428 {
1429         struct ifnet *ifp = &sc->arpcom.ac_if;
1430         int i;
1431
1432         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1433
1434         igb_disable_intr(sc);
1435
1436         callout_stop(&sc->timer);
1437
1438         ifp->if_flags &= ~IFF_RUNNING;
1439         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1440                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1441
1442                 ifsq_clr_oactive(txr->ifsq);
1443                 ifsq_watchdog_stop(&txr->tx_watchdog);
1444                 txr->tx_flags &= ~IGB_TXFLAG_ENABLED;
1445
1446                 txr->tx_running = 0;
1447                 callout_stop(&txr->tx_gc_timer);
1448         }
1449
1450         e1000_reset_hw(&sc->hw);
1451         E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1452
1453         e1000_led_off(&sc->hw);
1454         e1000_cleanup_led(&sc->hw);
1455
1456         for (i = 0; i < sc->tx_ring_cnt; ++i)
1457                 igb_free_tx_ring(&sc->tx_rings[i]);
1458         for (i = 0; i < sc->rx_ring_cnt; ++i)
1459                 igb_free_rx_ring(&sc->rx_rings[i]);
1460 }
1461
1462 static void
1463 igb_reset(struct igb_softc *sc, boolean_t media_reset)
1464 {
1465         struct ifnet *ifp = &sc->arpcom.ac_if;
1466         struct e1000_hw *hw = &sc->hw;
1467         struct e1000_fc_info *fc = &hw->fc;
1468         uint32_t pba = 0;
1469         uint16_t hwm;
1470
1471         /* Let the firmware know the OS is in control */
1472         igb_get_hw_control(sc);
1473
1474         /*
1475          * Packet Buffer Allocation (PBA)
1476          * Writing PBA sets the receive portion of the buffer
1477          * the remainder is used for the transmit buffer.
1478          */
1479         switch (hw->mac.type) {
1480         case e1000_82575:
1481                 pba = E1000_PBA_32K;
1482                 break;
1483
1484         case e1000_82576:
1485         case e1000_vfadapt:
1486                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1487                 pba &= E1000_RXPBS_SIZE_MASK_82576;
1488                 break;
1489
1490         case e1000_82580:
1491         case e1000_i350:
1492         case e1000_i354:
1493         case e1000_vfadapt_i350:
1494                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1495                 pba = e1000_rxpbs_adjust_82580(pba);
1496                 break;
1497
1498         case e1000_i210:
1499         case e1000_i211:
1500                 pba = E1000_PBA_34K;
1501                 break;
1502
1503         default:
1504                 break;
1505         }
1506
1507         /* Special needs in case of Jumbo frames */
1508         if (hw->mac.type == e1000_82575 && ifp->if_mtu > ETHERMTU) {
1509                 uint32_t tx_space, min_tx, min_rx;
1510
1511                 pba = E1000_READ_REG(hw, E1000_PBA);
1512                 tx_space = pba >> 16;
1513                 pba &= 0xffff;
1514
1515                 min_tx = (sc->max_frame_size +
1516                     sizeof(struct e1000_tx_desc) - ETHER_CRC_LEN) * 2;
1517                 min_tx = roundup2(min_tx, 1024);
1518                 min_tx >>= 10;
1519                 min_rx = sc->max_frame_size;
1520                 min_rx = roundup2(min_rx, 1024);
1521                 min_rx >>= 10;
1522                 if (tx_space < min_tx && (min_tx - tx_space) < pba) {
1523                         pba = pba - (min_tx - tx_space);
1524                         /*
1525                          * if short on rx space, rx wins
1526                          * and must trump tx adjustment
1527                          */
1528                         if (pba < min_rx)
1529                                 pba = min_rx;
1530                 }
1531                 E1000_WRITE_REG(hw, E1000_PBA, pba);
1532         }
1533
1534         /*
1535          * These parameters control the automatic generation (Tx) and
1536          * response (Rx) to Ethernet PAUSE frames.
1537          * - High water mark should allow for at least two frames to be
1538          *   received after sending an XOFF.
1539          * - Low water mark works best when it is very near the high water mark.
1540          *   This allows the receiver to restart by sending XON when it has
1541          *   drained a bit.
1542          */
1543         hwm = min(((pba << 10) * 9 / 10),
1544             ((pba << 10) - 2 * sc->max_frame_size));
1545
1546         if (hw->mac.type < e1000_82576) {
1547                 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1548                 fc->low_water = fc->high_water - 8;
1549         } else {
1550                 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1551                 fc->low_water = fc->high_water - 16;
1552         }
1553         fc->pause_time = IGB_FC_PAUSE_TIME;
1554         fc->send_xon = TRUE;
1555         fc->requested_mode = e1000_ifmedia2fc(sc->ifm_flowctrl);
1556
1557         /* Issue a global reset */
1558         e1000_reset_hw(hw);
1559         E1000_WRITE_REG(hw, E1000_WUC, 0);
1560
1561         /* Reset for AutoMediaDetect */
1562         if (media_reset) {
1563                 e1000_setup_init_funcs(hw, TRUE);
1564                 e1000_get_bus_info(hw);
1565         }
1566
1567         if (e1000_init_hw(hw) < 0)
1568                 if_printf(ifp, "Hardware Initialization Failed\n");
1569
1570         /* Setup DMA Coalescing */
1571         igb_init_dmac(sc, pba);
1572
1573         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1574         e1000_get_phy_info(hw);
1575         e1000_check_for_link(hw);
1576 }
1577
1578 static void
1579 igb_setup_ifp(struct igb_softc *sc)
1580 {
1581         struct ifnet *ifp = &sc->arpcom.ac_if;
1582         int i;
1583
1584         ifp->if_softc = sc;
1585         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1586         ifp->if_init = igb_init;
1587         ifp->if_ioctl = igb_ioctl;
1588         ifp->if_start = igb_start;
1589         ifp->if_serialize = igb_serialize;
1590         ifp->if_deserialize = igb_deserialize;
1591         ifp->if_tryserialize = igb_tryserialize;
1592 #ifdef INVARIANTS
1593         ifp->if_serialize_assert = igb_serialize_assert;
1594 #endif
1595 #ifdef IFPOLL_ENABLE
1596         ifp->if_npoll = igb_npoll;
1597 #endif
1598
1599         ifp->if_nmbclusters = sc->rx_ring_cnt * sc->rx_rings[0].num_rx_desc;
1600
1601         ifq_set_maxlen(&ifp->if_snd, sc->tx_rings[0].num_tx_desc - 1);
1602         ifq_set_ready(&ifp->if_snd);
1603         ifq_set_subq_cnt(&ifp->if_snd, sc->tx_ring_cnt);
1604
1605         ifp->if_mapsubq = ifq_mapsubq_modulo;
1606         ifq_set_subq_divisor(&ifp->if_snd, 1);
1607
1608         ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1609
1610         ifp->if_capabilities =
1611             IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_TSO;
1612         if (IGB_ENABLE_HWRSS(sc))
1613                 ifp->if_capabilities |= IFCAP_RSS;
1614         ifp->if_capenable = ifp->if_capabilities;
1615         ifp->if_hwassist = IGB_CSUM_FEATURES | CSUM_TSO;
1616
1617         /*
1618          * Tell the upper layer(s) we support long frames
1619          */
1620         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1621
1622         /* Setup TX rings and subqueues */
1623         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1624                 struct ifaltq_subque *ifsq = ifq_get_subq(&ifp->if_snd, i);
1625                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1626
1627                 ifsq_set_cpuid(ifsq, txr->tx_intr_cpuid);
1628                 ifsq_set_priv(ifsq, txr);
1629                 ifsq_set_hw_serialize(ifsq, &txr->tx_serialize);
1630                 txr->ifsq = ifsq;
1631
1632                 ifsq_watchdog_init(&txr->tx_watchdog, ifsq, igb_watchdog);
1633         }
1634
1635         /*
1636          * Specify the media types supported by this adapter and register
1637          * callbacks to update media and link information
1638          */
1639         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1640             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1641                 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1642                     0, NULL);
1643         } else {
1644                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1645                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1646                     0, NULL);
1647                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1648                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1649                     0, NULL);
1650                 if (sc->hw.phy.type != e1000_phy_ife) {
1651                         ifmedia_add(&sc->media,
1652                             IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1653                 }
1654         }
1655         ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1656         ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO | sc->ifm_flowctrl);
1657 }
1658
1659 static void
1660 igb_add_sysctl(struct igb_softc *sc)
1661 {
1662         struct sysctl_ctx_list *ctx;
1663         struct sysctl_oid *tree;
1664         char node[32];
1665         int i;
1666
1667         ctx = device_get_sysctl_ctx(sc->dev);
1668         tree = device_get_sysctl_tree(sc->dev);
1669         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1670             OID_AUTO, "rxr", CTLFLAG_RD, &sc->rx_ring_cnt, 0, "# of RX rings");
1671         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1672             OID_AUTO, "rxr_inuse", CTLFLAG_RD, &sc->rx_ring_inuse, 0,
1673             "# of RX rings used");
1674         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1675             OID_AUTO, "txr", CTLFLAG_RD, &sc->tx_ring_cnt, 0, "# of TX rings");
1676         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1677             OID_AUTO, "txr_inuse", CTLFLAG_RD, &sc->tx_ring_inuse, 0,
1678             "# of TX rings used");
1679         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1680             OID_AUTO, "rxd", CTLFLAG_RD, &sc->rx_rings[0].num_rx_desc, 0,
1681             "# of RX descs");
1682         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1683             OID_AUTO, "txd", CTLFLAG_RD, &sc->tx_rings[0].num_tx_desc, 0,
1684             "# of TX descs");
1685
1686 #define IGB_ADD_INTR_RATE_SYSCTL(sc, use, name) \
1687 do { \
1688         igb_add_intr_rate_sysctl(sc, IGB_INTR_USE_##use, #name "_intr_rate", \
1689             #use " interrupt rate"); \
1690 } while (0)
1691
1692         IGB_ADD_INTR_RATE_SYSCTL(sc, RXTX, rxtx);
1693         IGB_ADD_INTR_RATE_SYSCTL(sc, RX, rx);
1694         IGB_ADD_INTR_RATE_SYSCTL(sc, TX, tx);
1695         IGB_ADD_INTR_RATE_SYSCTL(sc, STATUS, sts);
1696
1697 #undef IGB_ADD_INTR_RATE_SYSCTL
1698
1699         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1700             OID_AUTO, "tx_intr_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1701             sc, 0, igb_sysctl_tx_intr_nsegs, "I",
1702             "# of segments per TX interrupt");
1703
1704         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1705             OID_AUTO, "tx_wreg_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1706             sc, 0, igb_sysctl_tx_wreg_nsegs, "I",
1707             "# of segments sent before write to hardware register");
1708
1709         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1710             OID_AUTO, "rx_wreg_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1711             sc, 0, igb_sysctl_rx_wreg_nsegs, "I",
1712             "# of segments received before write to hardware register");
1713
1714         if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
1715                 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1716                     OID_AUTO, "tx_msix_cpumap", CTLTYPE_OPAQUE | CTLFLAG_RD,
1717                     sc->tx_rmap_intr, 0, if_ringmap_cpumap_sysctl, "I",
1718                     "TX MSI-X CPU map");
1719                 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1720                     OID_AUTO, "rx_msix_cpumap", CTLTYPE_OPAQUE | CTLFLAG_RD,
1721                     sc->rx_rmap_intr, 0, if_ringmap_cpumap_sysctl, "I",
1722                     "RX MSI-X CPU map");
1723         }
1724 #ifdef IFPOLL_ENABLE
1725         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1726             OID_AUTO, "tx_poll_cpumap", CTLTYPE_OPAQUE | CTLFLAG_RD,
1727             sc->tx_rmap, 0, if_ringmap_cpumap_sysctl, "I",
1728             "TX polling CPU map");
1729         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1730             OID_AUTO, "rx_poll_cpumap", CTLTYPE_OPAQUE | CTLFLAG_RD,
1731             sc->rx_rmap, 0, if_ringmap_cpumap_sysctl, "I",
1732             "RX polling CPU map");
1733 #endif
1734
1735 #ifdef IGB_RSS_DEBUG
1736         SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree),
1737             OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug, 0,
1738             "RSS debug level");
1739         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1740                 ksnprintf(node, sizeof(node), "rx%d_pkt", i);
1741                 SYSCTL_ADD_ULONG(ctx,
1742                     SYSCTL_CHILDREN(tree), OID_AUTO, node,
1743                     CTLFLAG_RW, &sc->rx_rings[i].rx_packets, "RXed packets");
1744         }
1745 #endif
1746         for  (i = 0; i < sc->tx_ring_cnt; ++i) {
1747                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1748
1749 #ifdef IGB_TSS_DEBUG
1750                 ksnprintf(node, sizeof(node), "tx%d_pkt", i);
1751                 SYSCTL_ADD_ULONG(ctx, SYSCTL_CHILDREN(tree), OID_AUTO, node,
1752                     CTLFLAG_RW, &txr->tx_packets, "TXed packets");
1753 #endif
1754                 ksnprintf(node, sizeof(node), "tx%d_nmbuf", i);
1755                 SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO, node,
1756                     CTLFLAG_RD, &txr->tx_nmbuf, 0, "# of pending TX mbufs");
1757
1758                 ksnprintf(node, sizeof(node), "tx%d_gc", i);
1759                 SYSCTL_ADD_ULONG(ctx, SYSCTL_CHILDREN(tree), OID_AUTO, node,
1760                     CTLFLAG_RW, &txr->tx_gc, "# of TX desc GC");
1761         }
1762
1763         SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
1764             OID_AUTO, "dumpreg", CTLTYPE_INT | CTLFLAG_RW,
1765             sc, 0, igb_sysctl_reg_dump, "I", "dump registers");
1766 }
1767
1768 static int
1769 igb_alloc_rings(struct igb_softc *sc)
1770 {
1771         int error, i;
1772
1773         /*
1774          * Create top level busdma tag
1775          */
1776         error = bus_dma_tag_create(NULL, 1, 0,
1777             BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1778             BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1779             &sc->parent_tag);
1780         if (error) {
1781                 device_printf(sc->dev, "could not create top level DMA tag\n");
1782                 return error;
1783         }
1784
1785         /*
1786          * Allocate TX descriptor rings and buffers
1787          */
1788         sc->tx_rings = kmalloc_cachealign(
1789             sizeof(struct igb_tx_ring) * sc->tx_ring_cnt,
1790             M_DEVBUF, M_WAITOK | M_ZERO);
1791         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1792                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1793
1794                 /* Set up some basics */
1795                 txr->sc = sc;
1796                 txr->me = i;
1797                 txr->tx_intr_cpuid = -1;
1798                 lwkt_serialize_init(&txr->tx_serialize);
1799                 callout_init_mp(&txr->tx_gc_timer);
1800
1801                 error = igb_create_tx_ring(txr);
1802                 if (error)
1803                         return error;
1804         }
1805
1806         /*
1807          * Allocate RX descriptor rings and buffers
1808          */
1809         sc->rx_rings = kmalloc_cachealign(
1810             sizeof(struct igb_rx_ring) * sc->rx_ring_cnt,
1811             M_DEVBUF, M_WAITOK | M_ZERO);
1812         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1813                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1814
1815                 /* Set up some basics */
1816                 rxr->sc = sc;
1817                 rxr->me = i;
1818                 lwkt_serialize_init(&rxr->rx_serialize);
1819
1820                 error = igb_create_rx_ring(rxr);
1821                 if (error)
1822                         return error;
1823         }
1824
1825         return 0;
1826 }
1827
1828 static void
1829 igb_free_rings(struct igb_softc *sc)
1830 {
1831         int i;
1832
1833         if (sc->tx_rings != NULL) {
1834                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1835                         struct igb_tx_ring *txr = &sc->tx_rings[i];
1836
1837                         igb_destroy_tx_ring(txr, txr->num_tx_desc);
1838                 }
1839                 kfree(sc->tx_rings, M_DEVBUF);
1840         }
1841
1842         if (sc->rx_rings != NULL) {
1843                 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1844                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
1845
1846                         igb_destroy_rx_ring(rxr, rxr->num_rx_desc);
1847                 }
1848                 kfree(sc->rx_rings, M_DEVBUF);
1849         }
1850 }
1851
1852 static int
1853 igb_create_tx_ring(struct igb_tx_ring *txr)
1854 {
1855         int tsize, error, i, ntxd;
1856
1857         /*
1858          * Validate number of transmit descriptors. It must not exceed
1859          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1860          */
1861         ntxd = device_getenv_int(txr->sc->dev, "txd", igb_txd);
1862         if ((ntxd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN != 0 ||
1863             ntxd > IGB_MAX_TXD || ntxd < IGB_MIN_TXD) {
1864                 device_printf(txr->sc->dev,
1865                     "Using %d TX descriptors instead of %d!\n",
1866                     IGB_DEFAULT_TXD, ntxd);
1867                 txr->num_tx_desc = IGB_DEFAULT_TXD;
1868         } else {
1869                 txr->num_tx_desc = ntxd;
1870         }
1871
1872         /*
1873          * Allocate TX descriptor ring
1874          */
1875         tsize = roundup2(txr->num_tx_desc * sizeof(union e1000_adv_tx_desc),
1876             IGB_DBA_ALIGN);
1877         txr->txdma.dma_vaddr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1878             IGB_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1879             &txr->txdma.dma_tag, &txr->txdma.dma_map, &txr->txdma.dma_paddr);
1880         if (txr->txdma.dma_vaddr == NULL) {
1881                 device_printf(txr->sc->dev,
1882                     "Unable to allocate TX Descriptor memory\n");
1883                 return ENOMEM;
1884         }
1885         txr->tx_base = txr->txdma.dma_vaddr;
1886         bzero(txr->tx_base, tsize);
1887
1888         tsize = __VM_CACHELINE_ALIGN(
1889             sizeof(struct igb_tx_buf) * txr->num_tx_desc);
1890         txr->tx_buf = kmalloc_cachealign(tsize, M_DEVBUF, M_WAITOK | M_ZERO);
1891
1892         /*
1893          * Allocate TX head write-back buffer
1894          */
1895         txr->tx_hdr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1896             __VM_CACHELINE_SIZE, __VM_CACHELINE_SIZE, BUS_DMA_WAITOK,
1897             &txr->tx_hdr_dtag, &txr->tx_hdr_dmap, &txr->tx_hdr_paddr);
1898         if (txr->tx_hdr == NULL) {
1899                 device_printf(txr->sc->dev,
1900                     "Unable to allocate TX head write-back buffer\n");
1901                 return ENOMEM;
1902         }
1903
1904         /*
1905          * Create DMA tag for TX buffers
1906          */
1907         error = bus_dma_tag_create(txr->sc->parent_tag,
1908             1, 0,               /* alignment, bounds */
1909             BUS_SPACE_MAXADDR,  /* lowaddr */
1910             BUS_SPACE_MAXADDR,  /* highaddr */
1911             NULL, NULL,         /* filter, filterarg */
1912             IGB_TSO_SIZE,       /* maxsize */
1913             IGB_MAX_SCATTER,    /* nsegments */
1914             PAGE_SIZE,          /* maxsegsize */
1915             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1916             BUS_DMA_ONEBPAGE,   /* flags */
1917             &txr->tx_tag);
1918         if (error) {
1919                 device_printf(txr->sc->dev, "Unable to allocate TX DMA tag\n");
1920                 kfree(txr->tx_buf, M_DEVBUF);
1921                 txr->tx_buf = NULL;
1922                 return error;
1923         }
1924
1925         /*
1926          * Create DMA maps for TX buffers
1927          */
1928         for (i = 0; i < txr->num_tx_desc; ++i) {
1929                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1930
1931                 error = bus_dmamap_create(txr->tx_tag,
1932                     BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE, &txbuf->map);
1933                 if (error) {
1934                         device_printf(txr->sc->dev,
1935                             "Unable to create TX DMA map\n");
1936                         igb_destroy_tx_ring(txr, i);
1937                         return error;
1938                 }
1939         }
1940
1941         if (txr->sc->hw.mac.type == e1000_82575)
1942                 txr->tx_flags |= IGB_TXFLAG_TSO_IPLEN0;
1943
1944         /*
1945          * Initialize various watermark
1946          */
1947         if (txr->sc->hw.mac.type == e1000_82575) {
1948                 /*
1949                  * There no ways to GC pending TX mbufs in 'header
1950                  * write back' mode with reduced # of RS TX descs,
1951                  * since TDH does _not_ move for 82575.
1952                  */
1953                 txr->intr_nsegs = 1;
1954         } else {
1955                 txr->intr_nsegs = txr->num_tx_desc / 16;
1956         }
1957         txr->wreg_nsegs = IGB_DEF_TXWREG_NSEGS;
1958
1959         return 0;
1960 }
1961
1962 static void
1963 igb_free_tx_ring(struct igb_tx_ring *txr)
1964 {
1965         int i;
1966
1967         for (i = 0; i < txr->num_tx_desc; ++i) {
1968                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1969
1970                 if (txbuf->m_head != NULL)
1971                         igb_free_txbuf(txr, txbuf);
1972         }
1973 }
1974
1975 static void
1976 igb_destroy_tx_ring(struct igb_tx_ring *txr, int ndesc)
1977 {
1978         int i;
1979
1980         if (txr->txdma.dma_vaddr != NULL) {
1981                 bus_dmamap_unload(txr->txdma.dma_tag, txr->txdma.dma_map);
1982                 bus_dmamem_free(txr->txdma.dma_tag, txr->txdma.dma_vaddr,
1983                     txr->txdma.dma_map);
1984                 bus_dma_tag_destroy(txr->txdma.dma_tag);
1985                 txr->txdma.dma_vaddr = NULL;
1986         }
1987
1988         if (txr->tx_hdr != NULL) {
1989                 bus_dmamap_unload(txr->tx_hdr_dtag, txr->tx_hdr_dmap);
1990                 bus_dmamem_free(txr->tx_hdr_dtag, txr->tx_hdr,
1991                     txr->tx_hdr_dmap);
1992                 bus_dma_tag_destroy(txr->tx_hdr_dtag);
1993                 txr->tx_hdr = NULL;
1994         }
1995
1996         if (txr->tx_buf == NULL)
1997                 return;
1998
1999         for (i = 0; i < ndesc; ++i) {
2000                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
2001
2002                 KKASSERT(txbuf->m_head == NULL);
2003                 bus_dmamap_destroy(txr->tx_tag, txbuf->map);
2004         }
2005         bus_dma_tag_destroy(txr->tx_tag);
2006
2007         kfree(txr->tx_buf, M_DEVBUF);
2008         txr->tx_buf = NULL;
2009 }
2010
2011 static void
2012 igb_init_tx_ring(struct igb_tx_ring *txr)
2013 {
2014         /* Clear the old descriptor contents */
2015         bzero(txr->tx_base,
2016             sizeof(union e1000_adv_tx_desc) * txr->num_tx_desc);
2017
2018         /* Clear TX head write-back buffer */
2019         *(txr->tx_hdr) = 0;
2020
2021         /* Reset indices */
2022         txr->next_avail_desc = 0;
2023         txr->next_to_clean = 0;
2024         txr->tx_nsegs = 0;
2025         txr->tx_running = 0;
2026         txr->tx_nmbuf = 0;
2027
2028         /* Set number of descriptors available */
2029         txr->tx_avail = txr->num_tx_desc;
2030
2031         /* Enable this TX ring */
2032         txr->tx_flags |= IGB_TXFLAG_ENABLED;
2033 }
2034
2035 static void
2036 igb_init_tx_unit(struct igb_softc *sc)
2037 {
2038         struct e1000_hw *hw = &sc->hw;
2039         uint32_t tctl;
2040         int i;
2041
2042         /* Setup the Tx Descriptor Rings */
2043         for (i = 0; i < sc->tx_ring_inuse; ++i) {
2044                 struct igb_tx_ring *txr = &sc->tx_rings[i];
2045                 uint64_t bus_addr = txr->txdma.dma_paddr;
2046                 uint64_t hdr_paddr = txr->tx_hdr_paddr;
2047                 uint32_t txdctl = 0;
2048                 uint32_t dca_txctrl;
2049
2050                 E1000_WRITE_REG(hw, E1000_TDLEN(i),
2051                     txr->num_tx_desc * sizeof(struct e1000_tx_desc));
2052                 E1000_WRITE_REG(hw, E1000_TDBAH(i),
2053                     (uint32_t)(bus_addr >> 32));
2054                 E1000_WRITE_REG(hw, E1000_TDBAL(i),
2055                     (uint32_t)bus_addr);
2056
2057                 /* Setup the HW Tx Head and Tail descriptor pointers */
2058                 E1000_WRITE_REG(hw, E1000_TDT(i), 0);
2059                 E1000_WRITE_REG(hw, E1000_TDH(i), 0);
2060
2061                 dca_txctrl = E1000_READ_REG(hw, E1000_DCA_TXCTRL(i));
2062                 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2063                 E1000_WRITE_REG(hw, E1000_DCA_TXCTRL(i), dca_txctrl);
2064
2065                 /*
2066                  * Don't set WB_on_EITR:
2067                  * - 82575 does not have it
2068                  * - It almost has no effect on 82576, see:
2069                  *   82576 specification update errata #26
2070                  * - It causes unnecessary bus traffic
2071                  */
2072                 E1000_WRITE_REG(hw, E1000_TDWBAH(i),
2073                     (uint32_t)(hdr_paddr >> 32));
2074                 E1000_WRITE_REG(hw, E1000_TDWBAL(i),
2075                     ((uint32_t)hdr_paddr) | E1000_TX_HEAD_WB_ENABLE);
2076
2077                 /*
2078                  * WTHRESH is ignored by the hardware, since header
2079                  * write back mode is used.
2080                  */
2081                 txdctl |= IGB_TX_PTHRESH;
2082                 txdctl |= IGB_TX_HTHRESH << 8;
2083                 txdctl |= IGB_TX_WTHRESH << 16;
2084                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2085                 E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
2086         }
2087
2088         if (sc->vf_ifp)
2089                 return;
2090
2091         e1000_config_collision_dist(hw);
2092
2093         /* Program the Transmit Control Register */
2094         tctl = E1000_READ_REG(hw, E1000_TCTL);
2095         tctl &= ~E1000_TCTL_CT;
2096         tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
2097             (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
2098
2099         /* This write will effectively turn on the transmit unit. */
2100         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
2101 }
2102
2103 static boolean_t
2104 igb_txcsum_ctx(struct igb_tx_ring *txr, struct mbuf *mp)
2105 {
2106         struct e1000_adv_tx_context_desc *TXD;
2107         uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
2108         int ehdrlen, ctxd, ip_hlen = 0;
2109         boolean_t offload = TRUE;
2110
2111         if ((mp->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) == 0)
2112                 offload = FALSE;
2113
2114         vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
2115
2116         ctxd = txr->next_avail_desc;
2117         TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
2118
2119         /*
2120          * In advanced descriptors the vlan tag must
2121          * be placed into the context descriptor, thus
2122          * we need to be here just for that setup.
2123          */
2124         if (mp->m_flags & M_VLANTAG) {
2125                 uint16_t vlantag;
2126
2127                 vlantag = htole16(mp->m_pkthdr.ether_vlantag);
2128                 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
2129         } else if (!offload) {
2130                 return FALSE;
2131         }
2132
2133         ehdrlen = mp->m_pkthdr.csum_lhlen;
2134         KASSERT(ehdrlen > 0, ("invalid ether hlen"));
2135
2136         /* Set the ether header length */
2137         vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
2138         if (mp->m_pkthdr.csum_flags & CSUM_IP) {
2139                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
2140                 ip_hlen = mp->m_pkthdr.csum_iphlen;
2141                 KASSERT(ip_hlen > 0, ("invalid ip hlen"));
2142         }
2143         vlan_macip_lens |= ip_hlen;
2144
2145         type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
2146         if (mp->m_pkthdr.csum_flags & CSUM_TCP)
2147                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
2148         else if (mp->m_pkthdr.csum_flags & CSUM_UDP)
2149                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
2150
2151         /*
2152          * 82575 needs the TX context index added; the queue
2153          * index is used as TX context index here.
2154          */
2155         if (txr->sc->hw.mac.type == e1000_82575)
2156                 mss_l4len_idx = txr->me << 4;
2157
2158         /* Now copy bits into descriptor */
2159         TXD->vlan_macip_lens = htole32(vlan_macip_lens);
2160         TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
2161         TXD->seqnum_seed = htole32(0);
2162         TXD->mss_l4len_idx = htole32(mss_l4len_idx);
2163
2164         /* We've consumed the first desc, adjust counters */
2165         if (++ctxd == txr->num_tx_desc)
2166                 ctxd = 0;
2167         txr->next_avail_desc = ctxd;
2168         --txr->tx_avail;
2169
2170         return offload;
2171 }
2172
2173 static void
2174 igb_txeof(struct igb_tx_ring *txr, int hdr)
2175 {
2176         int first, avail;
2177
2178         if (txr->tx_avail == txr->num_tx_desc)
2179                 return;
2180
2181         first = txr->next_to_clean;
2182         if (first == hdr)
2183                 return;
2184
2185         avail = txr->tx_avail;
2186         while (first != hdr) {
2187                 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
2188
2189                 KKASSERT(avail < txr->num_tx_desc);
2190                 ++avail;
2191
2192                 if (txbuf->m_head)
2193                         igb_free_txbuf(txr, txbuf);
2194
2195                 if (++first == txr->num_tx_desc)
2196                         first = 0;
2197         }
2198         txr->next_to_clean = first;
2199         txr->tx_avail = avail;
2200
2201         /*
2202          * If we have a minimum free, clear OACTIVE
2203          * to tell the stack that it is OK to send packets.
2204          */
2205         if (txr->tx_avail > IGB_MAX_SCATTER + IGB_TX_RESERVED) {
2206                 ifsq_clr_oactive(txr->ifsq);
2207
2208                 /*
2209                  * We have enough TX descriptors, turn off
2210                  * the watchdog.  We allow small amount of
2211                  * packets (roughly intr_nsegs) pending on
2212                  * the transmit ring.
2213                  */
2214                 txr->tx_watchdog.wd_timer = 0;
2215         }
2216         txr->tx_running = IGB_TX_RUNNING;
2217 }
2218
2219 static void
2220 igb_txgc(struct igb_tx_ring *txr)
2221 {
2222         int first, hdr;
2223 #ifdef INVARIANTS
2224         int avail;
2225 #endif
2226
2227         if (txr->tx_avail == txr->num_tx_desc)
2228                 return;
2229
2230         hdr = E1000_READ_REG(&txr->sc->hw, E1000_TDH(txr->me)),
2231         first = txr->next_to_clean;
2232         if (first == hdr)
2233                 goto done;
2234         txr->tx_gc++;
2235
2236 #ifdef INVARIANTS
2237         avail = txr->tx_avail;
2238 #endif
2239         while (first != hdr) {
2240                 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
2241
2242 #ifdef INVARIANTS
2243                 KKASSERT(avail < txr->num_tx_desc);
2244                 ++avail;
2245 #endif
2246                 if (txbuf->m_head)
2247                         igb_free_txbuf(txr, txbuf);
2248
2249                 if (++first == txr->num_tx_desc)
2250                         first = 0;
2251         }
2252 done:
2253         if (txr->tx_nmbuf)
2254                 txr->tx_running = IGB_TX_RUNNING;
2255 }
2256
2257 static int
2258 igb_create_rx_ring(struct igb_rx_ring *rxr)
2259 {
2260         int rsize, i, error, nrxd;
2261
2262         /*
2263          * Validate number of receive descriptors. It must not exceed
2264          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
2265          */
2266         nrxd = device_getenv_int(rxr->sc->dev, "rxd", igb_rxd);
2267         if ((nrxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN != 0 ||
2268             nrxd > IGB_MAX_RXD || nrxd < IGB_MIN_RXD) {
2269                 device_printf(rxr->sc->dev,
2270                     "Using %d RX descriptors instead of %d!\n",
2271                     IGB_DEFAULT_RXD, nrxd);
2272                 rxr->num_rx_desc = IGB_DEFAULT_RXD;
2273         } else {
2274                 rxr->num_rx_desc = nrxd;
2275         }
2276
2277         /*
2278          * Allocate RX descriptor ring
2279          */
2280         rsize = roundup2(rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc),
2281             IGB_DBA_ALIGN);
2282         rxr->rxdma.dma_vaddr = bus_dmamem_coherent_any(rxr->sc->parent_tag,
2283             IGB_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
2284             &rxr->rxdma.dma_tag, &rxr->rxdma.dma_map,
2285             &rxr->rxdma.dma_paddr);
2286         if (rxr->rxdma.dma_vaddr == NULL) {
2287                 device_printf(rxr->sc->dev,
2288                     "Unable to allocate RxDescriptor memory\n");
2289                 return ENOMEM;
2290         }
2291         rxr->rx_base = rxr->rxdma.dma_vaddr;
2292         bzero(rxr->rx_base, rsize);
2293
2294         rsize = __VM_CACHELINE_ALIGN(
2295             sizeof(struct igb_rx_buf) * rxr->num_rx_desc);
2296         rxr->rx_buf = kmalloc_cachealign(rsize, M_DEVBUF, M_WAITOK | M_ZERO);
2297
2298         /*
2299          * Create DMA tag for RX buffers
2300          */
2301         error = bus_dma_tag_create(rxr->sc->parent_tag,
2302             1, 0,               /* alignment, bounds */
2303             BUS_SPACE_MAXADDR,  /* lowaddr */
2304             BUS_SPACE_MAXADDR,  /* highaddr */
2305             NULL, NULL,         /* filter, filterarg */
2306             MCLBYTES,           /* maxsize */
2307             1,                  /* nsegments */
2308             MCLBYTES,           /* maxsegsize */
2309             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2310             &rxr->rx_tag);
2311         if (error) {
2312                 device_printf(rxr->sc->dev,
2313                     "Unable to create RX payload DMA tag\n");
2314                 kfree(rxr->rx_buf, M_DEVBUF);
2315                 rxr->rx_buf = NULL;
2316                 return error;
2317         }
2318
2319         /*
2320          * Create spare DMA map for RX buffers
2321          */
2322         error = bus_dmamap_create(rxr->rx_tag, BUS_DMA_WAITOK,
2323             &rxr->rx_sparemap);
2324         if (error) {
2325                 device_printf(rxr->sc->dev,
2326                     "Unable to create spare RX DMA maps\n");
2327                 bus_dma_tag_destroy(rxr->rx_tag);
2328                 kfree(rxr->rx_buf, M_DEVBUF);
2329                 rxr->rx_buf = NULL;
2330                 return error;
2331         }
2332
2333         /*
2334          * Create DMA maps for RX buffers
2335          */
2336         for (i = 0; i < rxr->num_rx_desc; i++) {
2337                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2338
2339                 error = bus_dmamap_create(rxr->rx_tag,
2340                     BUS_DMA_WAITOK, &rxbuf->map);
2341                 if (error) {
2342                         device_printf(rxr->sc->dev,
2343                             "Unable to create RX DMA maps\n");
2344                         igb_destroy_rx_ring(rxr, i);
2345                         return error;
2346                 }
2347         }
2348
2349         /*
2350          * Initialize various watermark
2351          */
2352         rxr->wreg_nsegs = IGB_DEF_RXWREG_NSEGS;
2353
2354         return 0;
2355 }
2356
2357 static void
2358 igb_free_rx_ring(struct igb_rx_ring *rxr)
2359 {
2360         int i;
2361
2362         for (i = 0; i < rxr->num_rx_desc; ++i) {
2363                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2364
2365                 if (rxbuf->m_head != NULL) {
2366                         bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2367                         m_freem(rxbuf->m_head);
2368                         rxbuf->m_head = NULL;
2369                 }
2370         }
2371
2372         if (rxr->fmp != NULL)
2373                 m_freem(rxr->fmp);
2374         rxr->fmp = NULL;
2375         rxr->lmp = NULL;
2376 }
2377
2378 static void
2379 igb_destroy_rx_ring(struct igb_rx_ring *rxr, int ndesc)
2380 {
2381         int i;
2382
2383         if (rxr->rxdma.dma_vaddr != NULL) {
2384                 bus_dmamap_unload(rxr->rxdma.dma_tag, rxr->rxdma.dma_map);
2385                 bus_dmamem_free(rxr->rxdma.dma_tag, rxr->rxdma.dma_vaddr,
2386                     rxr->rxdma.dma_map);
2387                 bus_dma_tag_destroy(rxr->rxdma.dma_tag);
2388                 rxr->rxdma.dma_vaddr = NULL;
2389         }
2390
2391         if (rxr->rx_buf == NULL)
2392                 return;
2393
2394         for (i = 0; i < ndesc; ++i) {
2395                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2396
2397                 KKASSERT(rxbuf->m_head == NULL);
2398                 bus_dmamap_destroy(rxr->rx_tag, rxbuf->map);
2399         }
2400         bus_dmamap_destroy(rxr->rx_tag, rxr->rx_sparemap);
2401         bus_dma_tag_destroy(rxr->rx_tag);
2402
2403         kfree(rxr->rx_buf, M_DEVBUF);
2404         rxr->rx_buf = NULL;
2405 }
2406
2407 static void
2408 igb_setup_rxdesc(union e1000_adv_rx_desc *rxd, const struct igb_rx_buf *rxbuf)
2409 {
2410         rxd->read.pkt_addr = htole64(rxbuf->paddr);
2411         rxd->wb.upper.status_error = 0;
2412 }
2413
2414 static int
2415 igb_newbuf(struct igb_rx_ring *rxr, int i, boolean_t wait)
2416 {
2417         struct mbuf *m;
2418         bus_dma_segment_t seg;
2419         bus_dmamap_t map;
2420         struct igb_rx_buf *rxbuf;
2421         int error, nseg;
2422
2423         m = m_getcl(wait ? M_WAITOK : M_NOWAIT, MT_DATA, M_PKTHDR);
2424         if (m == NULL) {
2425                 if (wait) {
2426                         if_printf(&rxr->sc->arpcom.ac_if,
2427                             "Unable to allocate RX mbuf\n");
2428                 }
2429                 return ENOBUFS;
2430         }
2431         m->m_len = m->m_pkthdr.len = MCLBYTES;
2432
2433         if (rxr->sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2434                 m_adj(m, ETHER_ALIGN);
2435
2436         error = bus_dmamap_load_mbuf_segment(rxr->rx_tag,
2437             rxr->rx_sparemap, m, &seg, 1, &nseg, BUS_DMA_NOWAIT);
2438         if (error) {
2439                 m_freem(m);
2440                 if (wait) {
2441                         if_printf(&rxr->sc->arpcom.ac_if,
2442                             "Unable to load RX mbuf\n");
2443                 }
2444                 return error;
2445         }
2446
2447         rxbuf = &rxr->rx_buf[i];
2448         if (rxbuf->m_head != NULL)
2449                 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2450
2451         map = rxbuf->map;
2452         rxbuf->map = rxr->rx_sparemap;
2453         rxr->rx_sparemap = map;
2454
2455         rxbuf->m_head = m;
2456         rxbuf->paddr = seg.ds_addr;
2457
2458         igb_setup_rxdesc(&rxr->rx_base[i], rxbuf);
2459         return 0;
2460 }
2461
2462 static int
2463 igb_init_rx_ring(struct igb_rx_ring *rxr)
2464 {
2465         int i;
2466
2467         /* Clear the ring contents */
2468         bzero(rxr->rx_base,
2469             rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc));
2470
2471         /* Now replenish the ring mbufs */
2472         for (i = 0; i < rxr->num_rx_desc; ++i) {
2473                 int error;
2474
2475                 error = igb_newbuf(rxr, i, TRUE);
2476                 if (error)
2477                         return error;
2478         }
2479
2480         /* Setup our descriptor indices */
2481         rxr->next_to_check = 0;
2482
2483         rxr->fmp = NULL;
2484         rxr->lmp = NULL;
2485         rxr->discard = FALSE;
2486
2487         return 0;
2488 }
2489
2490 static void
2491 igb_init_rx_unit(struct igb_softc *sc, boolean_t polling)
2492 {
2493         struct ifnet *ifp = &sc->arpcom.ac_if;
2494         struct e1000_hw *hw = &sc->hw;
2495         uint32_t rctl, rxcsum, srrctl = 0;
2496         int i;
2497
2498         /*
2499          * Make sure receives are disabled while setting
2500          * up the descriptor ring
2501          */
2502         rctl = E1000_READ_REG(hw, E1000_RCTL);
2503         E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2504
2505 #if 0
2506         /*
2507         ** Set up for header split
2508         */
2509         if (igb_header_split) {
2510                 /* Use a standard mbuf for the header */
2511                 srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2512                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2513         } else
2514 #endif
2515                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2516
2517         /*
2518         ** Set up for jumbo frames
2519         */
2520         if (ifp->if_mtu > ETHERMTU) {
2521                 rctl |= E1000_RCTL_LPE;
2522 #if 0
2523                 if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
2524                         srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2525                         rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
2526                 } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
2527                         srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2528                         rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
2529                 }
2530                 /* Set maximum packet len */
2531                 psize = adapter->max_frame_size;
2532                 /* are we on a vlan? */
2533                 if (adapter->ifp->if_vlantrunk != NULL)
2534                         psize += VLAN_TAG_SIZE;
2535                 E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
2536 #else
2537                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2538                 rctl |= E1000_RCTL_SZ_2048;
2539 #endif
2540         } else {
2541                 rctl &= ~E1000_RCTL_LPE;
2542                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2543                 rctl |= E1000_RCTL_SZ_2048;
2544         }
2545
2546         /*
2547          * If TX flow control is disabled and more the 1 RX rings
2548          * are enabled, enable DROP.
2549          *
2550          * This drops frames rather than hanging the RX MAC for all
2551          * RX rings.
2552          */
2553         if (sc->rx_ring_inuse > 1 &&
2554             (sc->ifm_flowctrl & IFM_ETH_TXPAUSE) == 0) {
2555                 srrctl |= E1000_SRRCTL_DROP_EN;
2556                 if (bootverbose)
2557                         if_printf(ifp, "enable RX drop\n");
2558         }
2559
2560         /* Setup the Base and Length of the Rx Descriptor Rings */
2561         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2562                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2563                 uint64_t bus_addr = rxr->rxdma.dma_paddr;
2564                 uint32_t rxdctl;
2565
2566                 E1000_WRITE_REG(hw, E1000_RDLEN(i),
2567                     rxr->num_rx_desc * sizeof(struct e1000_rx_desc));
2568                 E1000_WRITE_REG(hw, E1000_RDBAH(i),
2569                     (uint32_t)(bus_addr >> 32));
2570                 E1000_WRITE_REG(hw, E1000_RDBAL(i),
2571                     (uint32_t)bus_addr);
2572                 E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
2573                 /* Enable this Queue */
2574                 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
2575                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2576                 rxdctl &= 0xFFF00000;
2577                 rxdctl |= IGB_RX_PTHRESH;
2578                 rxdctl |= IGB_RX_HTHRESH << 8;
2579                 /*
2580                  * Don't set WTHRESH to a value above 1 on 82576, see:
2581                  * 82576 specification update errata #26
2582                  */
2583                 rxdctl |= IGB_RX_WTHRESH << 16;
2584                 E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
2585         }
2586
2587         rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2588         rxcsum &= ~(E1000_RXCSUM_PCSS_MASK | E1000_RXCSUM_IPPCSE);
2589
2590         /*
2591          * Receive Checksum Offload for TCP and UDP
2592          *
2593          * Checksum offloading is also enabled if multiple receive
2594          * queue is to be supported, since we need it to figure out
2595          * fragments.
2596          */
2597         if ((ifp->if_capenable & IFCAP_RXCSUM) || IGB_ENABLE_HWRSS(sc)) {
2598                 /*
2599                  * NOTE:
2600                  * PCSD must be enabled to enable multiple
2601                  * receive queues.
2602                  */
2603                 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2604                     E1000_RXCSUM_PCSD;
2605         } else {
2606                 rxcsum &= ~(E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2607                     E1000_RXCSUM_PCSD);
2608         }
2609         E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2610
2611         if (sc->rx_ring_inuse > 1) {
2612                 uint8_t key[IGB_NRSSRK * IGB_RSSRK_SIZE];
2613                 const struct if_ringmap *rm;
2614                 uint32_t reta_shift;
2615                 int j, r;
2616
2617                 /*
2618                  * NOTE:
2619                  * When we reach here, RSS has already been disabled
2620                  * in igb_stop(), so we could safely configure RSS key
2621                  * and redirect table.
2622                  */
2623
2624                 /*
2625                  * Configure RSS key
2626                  */
2627                 toeplitz_get_key(key, sizeof(key));
2628                 for (i = 0; i < IGB_NRSSRK; ++i) {
2629                         uint32_t rssrk;
2630
2631                         rssrk = IGB_RSSRK_VAL(key, i);
2632                         IGB_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2633
2634                         E1000_WRITE_REG(hw, E1000_RSSRK(i), rssrk);
2635                 }
2636
2637                 /*
2638                  * Configure RSS redirect table
2639                  */
2640                 if (polling)
2641                         rm = sc->rx_rmap;
2642                 else
2643                         rm = sc->rx_rmap_intr;
2644                 if_ringmap_rdrtable(rm, sc->rdr_table, IGB_RDRTABLE_SIZE);
2645
2646                 reta_shift = IGB_RETA_SHIFT;
2647                 if (hw->mac.type == e1000_82575)
2648                         reta_shift = IGB_RETA_SHIFT_82575;
2649
2650                 r = 0;
2651                 for (j = 0; j < IGB_NRETA; ++j) {
2652                         uint32_t reta = 0;
2653
2654                         for (i = 0; i < IGB_RETA_SIZE; ++i) {
2655                                 uint32_t q;
2656
2657                                 q = sc->rdr_table[r] << reta_shift;
2658                                 reta |= q << (8 * i);
2659                                 ++r;
2660                         }
2661                         IGB_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2662                         E1000_WRITE_REG(hw, E1000_RETA(j), reta);
2663                 }
2664
2665                 /*
2666                  * Enable multiple receive queues.
2667                  * Enable IPv4 RSS standard hash functions.
2668                  * Disable RSS interrupt on 82575
2669                  */
2670                 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2671                                 E1000_MRQC_ENABLE_RSS_4Q |
2672                                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2673                                 E1000_MRQC_RSS_FIELD_IPV4);
2674         }
2675
2676         /* Setup the Receive Control Register */
2677         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2678         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2679             E1000_RCTL_RDMTS_HALF |
2680             (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2681         /* Strip CRC bytes. */
2682         rctl |= E1000_RCTL_SECRC;
2683         /* Make sure VLAN Filters are off */
2684         rctl &= ~E1000_RCTL_VFE;
2685         /* Don't store bad packets */
2686         rctl &= ~E1000_RCTL_SBP;
2687
2688         /* Enable Receives */
2689         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2690
2691         /*
2692          * Setup the HW Rx Head and Tail Descriptor Pointers
2693          *   - needs to be after enable
2694          */
2695         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2696                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2697
2698                 E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
2699                 E1000_WRITE_REG(hw, E1000_RDT(i), rxr->num_rx_desc - 1);
2700         }
2701 }
2702
2703 static void
2704 igb_rx_refresh(struct igb_rx_ring *rxr, int i)
2705 {
2706         if (--i < 0)
2707                 i = rxr->num_rx_desc - 1;
2708         E1000_WRITE_REG(&rxr->sc->hw, E1000_RDT(rxr->me), i);
2709 }
2710
2711 static void
2712 igb_rxeof(struct igb_rx_ring *rxr, int count)
2713 {
2714         struct ifnet *ifp = &rxr->sc->arpcom.ac_if;
2715         union e1000_adv_rx_desc *cur;
2716         uint32_t staterr;
2717         int i, ncoll = 0, cpuid = mycpuid;
2718
2719         i = rxr->next_to_check;
2720         cur = &rxr->rx_base[i];
2721         staterr = le32toh(cur->wb.upper.status_error);
2722
2723         if ((staterr & E1000_RXD_STAT_DD) == 0)
2724                 return;
2725
2726         while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
2727                 struct pktinfo *pi = NULL, pi0;
2728                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2729                 struct mbuf *m = NULL;
2730                 boolean_t eop;
2731
2732                 eop = (staterr & E1000_RXD_STAT_EOP) ? TRUE : FALSE;
2733                 if (eop)
2734                         --count;
2735
2736                 ++ncoll;
2737                 if ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) == 0 &&
2738                     !rxr->discard) {
2739                         struct mbuf *mp = rxbuf->m_head;
2740                         uint32_t hash, hashtype;
2741                         uint16_t vlan;
2742                         int len;
2743
2744                         len = le16toh(cur->wb.upper.length);
2745                         if ((rxr->sc->hw.mac.type == e1000_i350 ||
2746                              rxr->sc->hw.mac.type == e1000_i354) &&
2747                             (staterr & E1000_RXDEXT_STATERR_LB))
2748                                 vlan = be16toh(cur->wb.upper.vlan);
2749                         else
2750                                 vlan = le16toh(cur->wb.upper.vlan);
2751
2752                         hash = le32toh(cur->wb.lower.hi_dword.rss);
2753                         hashtype = le32toh(cur->wb.lower.lo_dword.data) &
2754                             E1000_RXDADV_RSSTYPE_MASK;
2755
2756                         IGB_RSS_DPRINTF(rxr->sc, 10,
2757                             "ring%d, hash 0x%08x, hashtype %u\n",
2758                             rxr->me, hash, hashtype);
2759
2760                         bus_dmamap_sync(rxr->rx_tag, rxbuf->map,
2761                             BUS_DMASYNC_POSTREAD);
2762
2763                         if (igb_newbuf(rxr, i, FALSE) != 0) {
2764                                 IFNET_STAT_INC(ifp, iqdrops, 1);
2765                                 goto discard;
2766                         }
2767
2768                         mp->m_len = len;
2769                         if (rxr->fmp == NULL) {
2770                                 mp->m_pkthdr.len = len;
2771                                 rxr->fmp = mp;
2772                                 rxr->lmp = mp;
2773                         } else {
2774                                 rxr->lmp->m_next = mp;
2775                                 rxr->lmp = rxr->lmp->m_next;
2776                                 rxr->fmp->m_pkthdr.len += len;
2777                         }
2778
2779                         if (eop) {
2780                                 m = rxr->fmp;
2781                                 rxr->fmp = NULL;
2782                                 rxr->lmp = NULL;
2783
2784                                 m->m_pkthdr.rcvif = ifp;
2785                                 IFNET_STAT_INC(ifp, ipackets, 1);
2786
2787                                 if (ifp->if_capenable & IFCAP_RXCSUM)
2788                                         igb_rxcsum(staterr, m);
2789
2790                                 if (staterr & E1000_RXD_STAT_VP) {
2791                                         m->m_pkthdr.ether_vlantag = vlan;
2792                                         m->m_flags |= M_VLANTAG;
2793                                 }
2794
2795                                 if (ifp->if_capenable & IFCAP_RSS) {
2796                                         pi = igb_rssinfo(m, &pi0,
2797                                             hash, hashtype, staterr);
2798                                 }
2799 #ifdef IGB_RSS_DEBUG
2800                                 rxr->rx_packets++;
2801 #endif
2802                         }
2803                 } else {
2804                         IFNET_STAT_INC(ifp, ierrors, 1);
2805 discard:
2806                         igb_setup_rxdesc(cur, rxbuf);
2807                         if (!eop)
2808                                 rxr->discard = TRUE;
2809                         else
2810                                 rxr->discard = FALSE;
2811                         if (rxr->fmp != NULL) {
2812                                 m_freem(rxr->fmp);
2813                                 rxr->fmp = NULL;
2814                                 rxr->lmp = NULL;
2815                         }
2816                         m = NULL;
2817                 }
2818
2819                 if (m != NULL)
2820                         ifp->if_input(ifp, m, pi, cpuid);
2821
2822                 /* Advance our pointers to the next descriptor. */
2823                 if (++i == rxr->num_rx_desc)
2824                         i = 0;
2825
2826                 if (ncoll >= rxr->wreg_nsegs) {
2827                         igb_rx_refresh(rxr, i);
2828                         ncoll = 0;
2829                 }
2830
2831                 cur = &rxr->rx_base[i];
2832                 staterr = le32toh(cur->wb.upper.status_error);
2833         }
2834         rxr->next_to_check = i;
2835
2836         if (ncoll > 0)
2837                 igb_rx_refresh(rxr, i);
2838 }
2839
2840
2841 static void
2842 igb_set_vlan(struct igb_softc *sc)
2843 {
2844         struct e1000_hw *hw = &sc->hw;
2845         uint32_t reg;
2846 #if 0
2847         struct ifnet *ifp = sc->arpcom.ac_if;
2848 #endif
2849
2850         if (sc->vf_ifp) {
2851                 e1000_rlpml_set_vf(hw, sc->max_frame_size + VLAN_TAG_SIZE);
2852                 return;
2853         }
2854
2855         reg = E1000_READ_REG(hw, E1000_CTRL);
2856         reg |= E1000_CTRL_VME;
2857         E1000_WRITE_REG(hw, E1000_CTRL, reg);
2858
2859 #if 0
2860         /* Enable the Filter Table */
2861         if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2862                 reg = E1000_READ_REG(hw, E1000_RCTL);
2863                 reg &= ~E1000_RCTL_CFIEN;
2864                 reg |= E1000_RCTL_VFE;
2865                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
2866         }
2867 #endif
2868
2869         /* Update the frame size */
2870         E1000_WRITE_REG(&sc->hw, E1000_RLPML,
2871             sc->max_frame_size + VLAN_TAG_SIZE);
2872
2873 #if 0
2874         /* Don't bother with table if no vlans */
2875         if ((adapter->num_vlans == 0) ||
2876             ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
2877                 return;
2878         /*
2879         ** A soft reset zero's out the VFTA, so
2880         ** we need to repopulate it now.
2881         */
2882         for (int i = 0; i < IGB_VFTA_SIZE; i++)
2883                 if (adapter->shadow_vfta[i] != 0) {
2884                         if (adapter->vf_ifp)
2885                                 e1000_vfta_set_vf(hw,
2886                                     adapter->shadow_vfta[i], TRUE);
2887                         else
2888                                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
2889                                  i, adapter->shadow_vfta[i]);
2890                 }
2891 #endif
2892 }
2893
2894 static void
2895 igb_enable_intr(struct igb_softc *sc)
2896 {
2897         int i;
2898
2899         for (i = 0; i < sc->intr_cnt; ++i)
2900                 lwkt_serialize_handler_enable(sc->intr_data[i].intr_serialize);
2901
2902         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2903                 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
2904                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, sc->intr_mask);
2905                 else
2906                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2907                 E1000_WRITE_REG(&sc->hw, E1000_EIAM, sc->intr_mask);
2908                 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2909                 E1000_WRITE_REG(&sc->hw, E1000_IMS, E1000_IMS_LSC);
2910         } else {
2911                 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
2912         }
2913         E1000_WRITE_FLUSH(&sc->hw);
2914 }
2915
2916 static void
2917 igb_disable_intr(struct igb_softc *sc)
2918 {
2919         int i;
2920
2921         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2922                 E1000_WRITE_REG(&sc->hw, E1000_EIMC, 0xffffffff);
2923                 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2924         }
2925         E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2926         E1000_WRITE_FLUSH(&sc->hw);
2927
2928         for (i = 0; i < sc->intr_cnt; ++i)
2929                 lwkt_serialize_handler_disable(sc->intr_data[i].intr_serialize);
2930 }
2931
2932 /*
2933  * Bit of a misnomer, what this really means is
2934  * to enable OS management of the system... aka
2935  * to disable special hardware management features
2936  */
2937 static void
2938 igb_get_mgmt(struct igb_softc *sc)
2939 {
2940         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2941                 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2942                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2943
2944                 /* disable hardware interception of ARP */
2945                 manc &= ~E1000_MANC_ARP_EN;
2946
2947                 /* enable receiving management packets to the host */
2948                 manc |= E1000_MANC_EN_MNG2HOST;
2949                 manc2h |= 1 << 5; /* Mng Port 623 */
2950                 manc2h |= 1 << 6; /* Mng Port 664 */
2951                 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
2952                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2953         }
2954 }
2955
2956 /*
2957  * Give control back to hardware management controller
2958  * if there is one.
2959  */
2960 static void
2961 igb_rel_mgmt(struct igb_softc *sc)
2962 {
2963         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2964                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2965
2966                 /* Re-enable hardware interception of ARP */
2967                 manc |= E1000_MANC_ARP_EN;
2968                 manc &= ~E1000_MANC_EN_MNG2HOST;
2969
2970                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2971         }
2972 }
2973
2974 /*
2975  * Sets CTRL_EXT:DRV_LOAD bit.
2976  *
2977  * For ASF and Pass Through versions of f/w this means that
2978  * the driver is loaded.
2979  */
2980 static void
2981 igb_get_hw_control(struct igb_softc *sc)
2982 {
2983         uint32_t ctrl_ext;
2984
2985         if (sc->vf_ifp)
2986                 return;
2987
2988         /* Let firmware know the driver has taken over */
2989         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2990         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2991             ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2992 }
2993
2994 /*
2995  * Resets CTRL_EXT:DRV_LOAD bit.
2996  *
2997  * For ASF and Pass Through versions of f/w this means that the
2998  * driver is no longer loaded.
2999  */
3000 static void
3001 igb_rel_hw_control(struct igb_softc *sc)
3002 {
3003         uint32_t ctrl_ext;
3004
3005         if (sc->vf_ifp)
3006                 return;
3007
3008         /* Let firmware taken over control of h/w */
3009         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3010         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
3011             ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
3012 }
3013
3014 static boolean_t
3015 igb_is_valid_ether_addr(const uint8_t *addr)
3016 {
3017         uint8_t zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
3018
3019         if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3020                 return FALSE;
3021         return TRUE;
3022 }
3023
3024 /*
3025  * Enable PCI Wake On Lan capability
3026  */
3027 static void
3028 igb_enable_wol(struct igb_softc *sc)
3029 {
3030         device_t dev = sc->dev;
3031         int error = 0;
3032         uint32_t pmc, ctrl;
3033         uint16_t status;
3034
3035         if (pci_find_extcap(dev, PCIY_PMG, &pmc) != 0) {
3036                 device_printf(dev, "no PMG\n");
3037                 return;
3038         }
3039
3040         /*
3041          * Set the type of wakeup.
3042          */
3043         sc->wol &= ~(E1000_WUFC_EX | E1000_WUFC_MC);
3044         if ((sc->wol & (E1000_WUFC_EX | E1000_WUFC_MAG | E1000_WUFC_MC)) == 0)
3045                 goto pme;
3046
3047         /*
3048          * Advertise the wakeup capabilities.
3049          */
3050         ctrl = E1000_READ_REG(&sc->hw, E1000_CTRL);
3051         ctrl |= (E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN3);
3052         E1000_WRITE_REG(&sc->hw, E1000_CTRL, ctrl);
3053
3054         /*
3055          * Keep the laser running on Fiber adapters.
3056          */
3057         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
3058             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
3059                 uint32_t ctrl_ext;
3060
3061                 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
3062                 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
3063                 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT, ctrl_ext);
3064         }
3065
3066         error = igb_enable_phy_wol(sc);
3067         if (error)
3068                 goto pme;
3069
3070         /* XXX will this happen? ich/pch specific. */
3071         if (sc->hw.phy.type == e1000_phy_igp_3)
3072                 e1000_igp3_phy_powerdown_workaround_ich8lan(&sc->hw);
3073
3074 pme:
3075         status = pci_read_config(dev, pmc + PCIR_POWER_STATUS, 2);
3076         status &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
3077         if (!error)
3078                 status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3079         pci_write_config(dev, pmc + PCIR_POWER_STATUS, status, 2);
3080 }
3081
3082 /*
3083  * WOL in the newer chipset interfaces (pchlan)
3084  * require thing to be copied into the phy
3085  */
3086 static int
3087 igb_enable_phy_wol(struct igb_softc *sc)
3088 {
3089         struct e1000_hw *hw = &sc->hw;
3090         uint32_t mreg;
3091         uint16_t preg;
3092         int ret = 0, i;
3093
3094         /* Copy MAC RARs to PHY RARs */
3095         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
3096
3097         /* Copy MAC MTA to PHY MTA */
3098         for (i = 0; i < hw->mac.mta_reg_count; i++) {
3099                 mreg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
3100                 e1000_write_phy_reg(hw, BM_MTA(i), (uint16_t)(mreg & 0xFFFF));
3101                 e1000_write_phy_reg(hw, BM_MTA(i) + 1,
3102                     (uint16_t)((mreg >> 16) & 0xFFFF));
3103         }
3104
3105         /* Configure PHY Rx Control register */
3106         e1000_read_phy_reg(hw, BM_RCTL, &preg);
3107         mreg = E1000_READ_REG(hw, E1000_RCTL);
3108         if (mreg & E1000_RCTL_UPE)
3109                 preg |= BM_RCTL_UPE;
3110         if (mreg & E1000_RCTL_MPE)
3111                 preg |= BM_RCTL_MPE;
3112         preg &= ~(BM_RCTL_MO_MASK);
3113         if (mreg & E1000_RCTL_MO_3) {
3114                 preg |= (((mreg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
3115                                 << BM_RCTL_MO_SHIFT);
3116         }
3117         if (mreg & E1000_RCTL_BAM)
3118                 preg |= BM_RCTL_BAM;
3119         if (mreg & E1000_RCTL_PMCF)
3120                 preg |= BM_RCTL_PMCF;
3121         mreg = E1000_READ_REG(hw, E1000_CTRL);
3122         if (mreg & E1000_CTRL_RFCE)
3123                 preg |= BM_RCTL_RFCE;
3124         e1000_write_phy_reg(&sc->hw, BM_RCTL, preg);
3125
3126         /* Enable PHY wakeup in MAC register. */
3127         E1000_WRITE_REG(hw, E1000_WUC,
3128             E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN | E1000_WUC_APME);
3129         E1000_WRITE_REG(hw, E1000_WUFC, sc->wol);
3130
3131         /* Configure and enable PHY wakeup in PHY registers */
3132         e1000_write_phy_reg(hw, BM_WUFC, sc->wol);
3133         e1000_write_phy_reg(hw, BM_WUC, E1000_WUC_PME_EN);
3134         /* Activate PHY wakeup */
3135         ret = hw->phy.ops.acquire(hw);
3136         if (ret) {
3137                 if_printf(&sc->arpcom.ac_if, "Could not acquire PHY\n");
3138                 return ret;
3139         }
3140         e1000_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3141                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
3142         ret = e1000_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &preg);
3143         if (ret) {
3144                 if_printf(&sc->arpcom.ac_if, "Could not read PHY page 769\n");
3145                 goto out;
3146         }
3147         preg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
3148         ret = e1000_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, preg);
3149         if (ret) {
3150                 if_printf(&sc->arpcom.ac_if,
3151                     "Could not set PHY Host Wakeup bit\n");
3152         }
3153 out:
3154         hw->phy.ops.release(hw);
3155         return ret;
3156 }
3157
3158 static void
3159 igb_update_stats_counters(struct igb_softc *sc)
3160 {
3161         struct e1000_hw *hw = &sc->hw;
3162         struct e1000_hw_stats *stats;
3163         struct ifnet *ifp = &sc->arpcom.ac_if;
3164
3165         /*
3166          * The virtual function adapter has only a
3167          * small controlled set of stats, do only
3168          * those and return.
3169          */
3170         if (sc->vf_ifp) {
3171                 igb_update_vf_stats_counters(sc);
3172                 return;
3173         }
3174         stats = sc->stats;
3175
3176         if (sc->hw.phy.media_type == e1000_media_type_copper ||
3177             (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
3178                 stats->symerrs +=
3179                     E1000_READ_REG(hw,E1000_SYMERRS);
3180                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
3181         }
3182
3183         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
3184         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
3185         stats->scc += E1000_READ_REG(hw, E1000_SCC);
3186         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
3187
3188         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
3189         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
3190         stats->colc += E1000_READ_REG(hw, E1000_COLC);
3191         stats->dc += E1000_READ_REG(hw, E1000_DC);
3192         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
3193         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
3194         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
3195
3196         /*
3197          * For watchdog management we need to know if we have been
3198          * paused during the last interval, so capture that here.
3199          */
3200         sc->pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
3201         stats->xoffrxc += sc->pause_frames;
3202         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
3203         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
3204         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
3205         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
3206         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
3207         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
3208         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
3209         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
3210         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
3211         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
3212         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
3213         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
3214
3215         /* For the 64-bit byte counters the low dword must be read first. */
3216         /* Both registers clear on the read of the high dword */
3217
3218         stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
3219             ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
3220         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
3221             ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
3222
3223         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
3224         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
3225         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
3226         stats->roc += E1000_READ_REG(hw, E1000_ROC);
3227         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
3228
3229         stats->mgprc += E1000_READ_REG(hw, E1000_MGTPRC);
3230         stats->mgpdc += E1000_READ_REG(hw, E1000_MGTPDC);
3231         stats->mgptc += E1000_READ_REG(hw, E1000_MGTPTC);
3232
3233         stats->tor += E1000_READ_REG(hw, E1000_TORL) +
3234             ((uint64_t)E1000_READ_REG(hw, E1000_TORH) << 32);
3235         stats->tot += E1000_READ_REG(hw, E1000_TOTL) +
3236             ((uint64_t)E1000_READ_REG(hw, E1000_TOTH) << 32);
3237
3238         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
3239         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
3240         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
3241         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
3242         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
3243         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
3244         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
3245         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
3246         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
3247         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
3248
3249         /* Interrupt Counts */
3250
3251         stats->iac += E1000_READ_REG(hw, E1000_IAC);
3252         stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
3253         stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
3254         stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
3255         stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
3256         stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
3257         stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
3258         stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
3259         stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
3260
3261         /* Host to Card Statistics */
3262
3263         stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
3264         stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
3265         stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
3266         stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
3267         stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
3268         stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
3269         stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
3270         stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
3271             ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32));
3272         stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
3273             ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
3274         stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
3275         stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
3276         stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
3277
3278         stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
3279         stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
3280         stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
3281         stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
3282         stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
3283         stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
3284
3285         IFNET_STAT_SET(ifp, collisions, stats->colc);
3286
3287         /* Rx Errors */
3288         IFNET_STAT_SET(ifp, ierrors,
3289             stats->rxerrc + stats->crcerrs + stats->algnerrc +
3290             stats->ruc + stats->roc + stats->mpc + stats->cexterr);
3291
3292         /* Tx Errors */
3293         IFNET_STAT_SET(ifp, oerrors,
3294             stats->ecol + stats->latecol + sc->watchdog_events);
3295
3296         /* Driver specific counters */
3297         sc->device_control = E1000_READ_REG(hw, E1000_CTRL);
3298         sc->rx_control = E1000_READ_REG(hw, E1000_RCTL);
3299         sc->int_mask = E1000_READ_REG(hw, E1000_IMS);
3300         sc->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
3301         sc->packet_buf_alloc_tx =
3302             ((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
3303         sc->packet_buf_alloc_rx =
3304             (E1000_READ_REG(hw, E1000_PBA) & 0xffff);
3305 }
3306
3307 static void
3308 igb_vf_init_stats(struct igb_softc *sc)
3309 {
3310         struct e1000_hw *hw = &sc->hw;
3311         struct e1000_vf_stats *stats;
3312
3313         stats = sc->stats;
3314         stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
3315         stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
3316         stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
3317         stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
3318         stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
3319 }
3320
3321 static void
3322 igb_update_vf_stats_counters(struct igb_softc *sc)
3323 {
3324         struct e1000_hw *hw = &sc->hw;
3325         struct e1000_vf_stats *stats;
3326
3327         if (sc->link_speed == 0)
3328                 return;
3329
3330         stats = sc->stats;
3331         UPDATE_VF_REG(E1000_VFGPRC, stats->last_gprc, stats->gprc);
3332         UPDATE_VF_REG(E1000_VFGORC, stats->last_gorc, stats->gorc);
3333         UPDATE_VF_REG(E1000_VFGPTC, stats->last_gptc, stats->gptc);
3334         UPDATE_VF_REG(E1000_VFGOTC, stats->last_gotc, stats->gotc);
3335         UPDATE_VF_REG(E1000_VFMPRC, stats->last_mprc, stats->mprc);
3336 }
3337
3338 #ifdef IFPOLL_ENABLE
3339
3340 static void
3341 igb_npoll_status(struct ifnet *ifp)
3342 {
3343         struct igb_softc *sc = ifp->if_softc;
3344         uint32_t reg_icr;
3345
3346         ASSERT_SERIALIZED(&sc->main_serialize);
3347
3348         reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3349         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3350                 sc->hw.mac.get_link_status = 1;
3351                 igb_update_link_status(sc);
3352         }
3353 }
3354
3355 static void
3356 igb_npoll_tx(struct ifnet *ifp, void *arg, int cycle __unused)
3357 {
3358         struct igb_tx_ring *txr = arg;
3359
3360         ASSERT_SERIALIZED(&txr->tx_serialize);
3361         igb_tx_intr(txr, *(txr->tx_hdr));
3362         igb_try_txgc(txr, 1);
3363 }
3364
3365 static void
3366 igb_npoll_rx(struct ifnet *ifp __unused, void *arg, int cycle)
3367 {
3368         struct igb_rx_ring *rxr = arg;
3369
3370         ASSERT_SERIALIZED(&rxr->rx_serialize);
3371
3372         igb_rxeof(rxr, cycle);
3373 }
3374
3375 static void
3376 igb_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3377 {
3378         struct igb_softc *sc = ifp->if_softc;
3379         int i, txr_cnt, rxr_cnt;
3380
3381         ASSERT_IFNET_SERIALIZED_ALL(ifp);
3382
3383         if (info) {
3384                 int cpu;
3385
3386                 info->ifpi_status.status_func = igb_npoll_status;
3387                 info->ifpi_status.serializer = &sc->main_serialize;
3388
3389                 txr_cnt = igb_get_txring_inuse(sc, TRUE);
3390                 for (i = 0; i < txr_cnt; ++i) {
3391                         struct igb_tx_ring *txr = &sc->tx_rings[i];
3392
3393                         cpu = if_ringmap_cpumap(sc->tx_rmap, i);
3394                         KKASSERT(cpu < netisr_ncpus);
3395                         info->ifpi_tx[cpu].poll_func = igb_npoll_tx;
3396                         info->ifpi_tx[cpu].arg = txr;
3397                         info->ifpi_tx[cpu].serializer = &txr->tx_serialize;
3398                         ifsq_set_cpuid(txr->ifsq, cpu);
3399                 }
3400
3401                 rxr_cnt = igb_get_rxring_inuse(sc, TRUE);
3402                 for (i = 0; i < rxr_cnt; ++i) {
3403                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
3404
3405                         cpu = if_ringmap_cpumap(sc->rx_rmap, i);
3406                         KKASSERT(cpu < netisr_ncpus);
3407                         info->ifpi_rx[cpu].poll_func = igb_npoll_rx;
3408                         info->ifpi_rx[cpu].arg = rxr;
3409                         info->ifpi_rx[cpu].serializer = &rxr->rx_serialize;
3410                 }
3411         } else {
3412                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3413                         struct igb_tx_ring *txr = &sc->tx_rings[i];
3414
3415                         ifsq_set_cpuid(txr->ifsq, txr->tx_intr_cpuid);
3416                 }
3417         }
3418         if (ifp->if_flags & IFF_RUNNING)
3419                 igb_init(sc);
3420 }
3421
3422 #endif /* IFPOLL_ENABLE */
3423
3424 static void
3425 igb_intr(void *xsc)
3426 {
3427         struct igb_softc *sc = xsc;
3428         struct ifnet *ifp = &sc->arpcom.ac_if;
3429         uint32_t eicr;
3430
3431         ASSERT_SERIALIZED(&sc->main_serialize);
3432
3433         eicr = E1000_READ_REG(&sc->hw, E1000_EICR);
3434
3435         if (eicr == 0)
3436                 return;
3437
3438         if (ifp->if_flags & IFF_RUNNING) {
3439                 struct igb_tx_ring *txr = &sc->tx_rings[0];
3440                 int i;
3441
3442                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3443                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
3444
3445                         if (eicr & rxr->rx_intr_mask) {
3446                                 lwkt_serialize_enter(&rxr->rx_serialize);
3447                                 igb_rxeof(rxr, -1);
3448                                 lwkt_serialize_exit(&rxr->rx_serialize);
3449                         }
3450                 }
3451
3452                 if (eicr & txr->tx_intr_mask) {
3453                         lwkt_serialize_enter(&txr->tx_serialize);
3454                         igb_tx_intr(txr, *(txr->tx_hdr));
3455                         lwkt_serialize_exit(&txr->tx_serialize);
3456                 }
3457         }
3458
3459         if (eicr & E1000_EICR_OTHER) {
3460                 uint32_t icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3461
3462                 /* Link status change */
3463                 if (icr & E1000_ICR_LSC) {
3464                         sc->hw.mac.get_link_status = 1;
3465                         igb_update_link_status(sc);
3466                 }
3467         }
3468
3469         /*
3470          * Reading EICR has the side effect to clear interrupt mask,
3471          * so all interrupts need to be enabled here.
3472          */
3473         E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
3474 }
3475
3476 static void
3477 igb_intr_shared(void *xsc)
3478 {
3479         struct igb_softc *sc = xsc;
3480         struct ifnet *ifp = &sc->arpcom.ac_if;
3481         uint32_t reg_icr;
3482
3483         ASSERT_SERIALIZED(&sc->main_serialize);
3484
3485         reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3486
3487         /* Hot eject?  */
3488         if (reg_icr == 0xffffffff)
3489                 return;
3490
3491         /* Definitely not our interrupt.  */
3492         if (reg_icr == 0x0)
3493                 return;
3494
3495         if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
3496                 return;
3497
3498         if (ifp->if_flags & IFF_RUNNING) {
3499                 if (reg_icr &
3500                     (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
3501                         int i;
3502
3503                         for (i = 0; i < sc->rx_ring_inuse; ++i) {
3504                                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3505
3506                                 lwkt_serialize_enter(&rxr->rx_serialize);
3507                                 igb_rxeof(rxr, -1);
3508                                 lwkt_serialize_exit(&rxr->rx_serialize);
3509                         }
3510                 }
3511
3512                 if (reg_icr & E1000_ICR_TXDW) {
3513                         struct igb_tx_ring *txr = &sc->tx_rings[0];
3514
3515                         lwkt_serialize_enter(&txr->tx_serialize);
3516                         igb_tx_intr(txr, *(txr->tx_hdr));
3517                         lwkt_serialize_exit(&txr->tx_serialize);
3518                 }
3519         }
3520
3521         /* Link status change */
3522         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3523                 sc->hw.mac.get_link_status = 1;
3524                 igb_update_link_status(sc);
3525         }
3526
3527         if (reg_icr & E1000_ICR_RXO)
3528                 sc->rx_overruns++;
3529 }
3530
3531 static int
3532 igb_encap(struct igb_tx_ring *txr, struct mbuf **m_headp,
3533     int *segs_used, int *idx)
3534 {
3535         bus_dma_segment_t segs[IGB_MAX_SCATTER];
3536         bus_dmamap_t map;
3537         struct igb_tx_buf *tx_buf, *tx_buf_mapped;
3538         union e1000_adv_tx_desc *txd = NULL;
3539         struct mbuf *m_head = *m_headp;
3540         uint32_t olinfo_status = 0, cmd_type_len = 0, cmd_rs = 0;
3541         int maxsegs, nsegs, i, j, error;
3542         uint32_t hdrlen = 0;
3543
3544         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3545                 error = igb_tso_pullup(txr, m_headp);
3546                 if (error)
3547                         return error;
3548                 m_head = *m_headp;
3549         }
3550
3551         /* Set basic descriptor constants */
3552         cmd_type_len |= E1000_ADVTXD_DTYP_DATA;
3553         cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
3554         if (m_head->m_flags & M_VLANTAG)
3555                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3556
3557         /*
3558          * Map the packet for DMA.
3559          */
3560         tx_buf = &txr->tx_buf[txr->next_avail_desc];
3561         tx_buf_mapped = tx_buf;
3562         map = tx_buf->map;
3563
3564         maxsegs = txr->tx_avail - IGB_TX_RESERVED;
3565         if (maxsegs > IGB_MAX_SCATTER)
3566                 maxsegs = IGB_MAX_SCATTER;
3567
3568         error = bus_dmamap_load_mbuf_defrag(txr->tx_tag, map, m_headp,
3569             segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3570         if (error) {
3571                 if (error == ENOBUFS)
3572                         txr->sc->mbuf_defrag_failed++;
3573                 else
3574                         txr->sc->no_tx_dma_setup++;
3575
3576                 m_freem(*m_headp);
3577                 *m_headp = NULL;
3578                 return error;
3579         }
3580         bus_dmamap_sync(txr->tx_tag, map, BUS_DMASYNC_PREWRITE);
3581
3582         m_head = *m_headp;
3583
3584         /*
3585          * Set up the TX context descriptor, if any hardware offloading is
3586          * needed.  This includes CSUM, VLAN, and TSO.  It will consume one
3587          * TX descriptor.
3588          *
3589          * Unlike these chips' predecessors (em/emx), TX context descriptor
3590          * will _not_ interfere TX data fetching pipelining.
3591          */
3592         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3593                 igb_tso_ctx(txr, m_head, &hdrlen);
3594                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3595                 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3596                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3597                 txr->tx_nsegs++;
3598                 (*segs_used)++;
3599         } else if (igb_txcsum_ctx(txr, m_head)) {
3600                 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3601                         olinfo_status |= (E1000_TXD_POPTS_IXSM << 8);
3602                 if (m_head->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_TCP))
3603                         olinfo_status |= (E1000_TXD_POPTS_TXSM << 8);
3604                 txr->tx_nsegs++;
3605                 (*segs_used)++;
3606         }
3607
3608         *segs_used += nsegs;
3609         txr->tx_nsegs += nsegs;
3610         if (txr->tx_nsegs >= txr->intr_nsegs) {
3611                 /*
3612                  * Report Status (RS) is turned on every intr_nsegs
3613                  * descriptors (roughly).
3614                  */
3615                 txr->tx_nsegs = 0;
3616                 cmd_rs = E1000_ADVTXD_DCMD_RS;
3617         }
3618
3619         /* Calculate payload length */
3620         olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
3621             << E1000_ADVTXD_PAYLEN_SHIFT);
3622
3623         /*
3624          * 82575 needs the TX context index added; the queue
3625          * index is used as TX context index here.
3626          */
3627         if (txr->sc->hw.mac.type == e1000_82575)
3628                 olinfo_status |= txr->me << 4;
3629
3630         /* Set up our transmit descriptors */
3631         i = txr->next_avail_desc;
3632         for (j = 0; j < nsegs; j++) {
3633                 bus_size_t seg_len;
3634                 bus_addr_t seg_addr;
3635
3636                 tx_buf = &txr->tx_buf[i];
3637                 txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
3638                 seg_addr = segs[j].ds_addr;
3639                 seg_len = segs[j].ds_len;
3640
3641                 txd->read.buffer_addr = htole64(seg_addr);
3642                 txd->read.cmd_type_len = htole32(cmd_type_len | seg_len);
3643                 txd->read.olinfo_status = htole32(olinfo_status);
3644                 if (++i == txr->num_tx_desc)
3645                         i = 0;
3646                 tx_buf->m_head = NULL;
3647         }
3648
3649         KASSERT(txr->tx_avail > nsegs, ("invalid avail TX desc\n"));
3650         txr->next_avail_desc = i;
3651         txr->tx_avail -= nsegs;
3652         txr->tx_nmbuf++;
3653
3654         tx_buf->m_head = m_head;
3655         tx_buf_mapped->map = tx_buf->map;
3656         tx_buf->map = map;
3657
3658         /*
3659          * Last Descriptor of Packet needs End Of Packet (EOP)
3660          */
3661         txd->read.cmd_type_len |= htole32(E1000_ADVTXD_DCMD_EOP | cmd_rs);
3662
3663         /*
3664          * Defer TDT updating, until enough descrptors are setup
3665          */
3666         *idx = i;
3667 #ifdef IGB_TSS_DEBUG
3668         ++txr->tx_packets;
3669 #endif
3670
3671         return 0;
3672 }
3673
3674 static void
3675 igb_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
3676 {
3677         struct igb_softc *sc = ifp->if_softc;
3678         struct igb_tx_ring *txr = ifsq_get_priv(ifsq);
3679         struct mbuf *m_head;
3680         int idx = -1, nsegs = 0;
3681
3682         KKASSERT(txr->ifsq == ifsq);
3683         ASSERT_SERIALIZED(&txr->tx_serialize);
3684
3685         if ((ifp->if_flags & IFF_RUNNING) == 0 || ifsq_is_oactive(ifsq))
3686                 return;
3687
3688         if (!sc->link_active || (txr->tx_flags & IGB_TXFLAG_ENABLED) == 0) {
3689                 ifsq_purge(ifsq);
3690                 return;
3691         }
3692
3693         while (!ifsq_is_empty(ifsq)) {
3694                 if (txr->tx_avail <= IGB_MAX_SCATTER + IGB_TX_RESERVED) {
3695                         ifsq_set_oactive(ifsq);
3696                         /* Set watchdog on */
3697                         txr->tx_watchdog.wd_timer = 5;
3698                         break;
3699                 }
3700
3701                 m_head = ifsq_dequeue(ifsq);
3702                 if (m_head == NULL)
3703                         break;
3704
3705                 if (igb_encap(txr, &m_head, &nsegs, &idx)) {
3706                         IFNET_STAT_INC(ifp, oerrors, 1);
3707                         continue;
3708                 }
3709
3710                 /*
3711                  * TX interrupt are aggressively aggregated, so increasing
3712                  * opackets at TX interrupt time will make the opackets
3713                  * statistics vastly inaccurate; we do the opackets increment
3714                  * now.
3715                  */
3716                 IFNET_STAT_INC(ifp, opackets, 1);
3717
3718                 if (nsegs >= txr->wreg_nsegs) {
3719                         E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3720                         idx = -1;
3721                         nsegs = 0;
3722                 }
3723
3724                 /* Send a copy of the frame to the BPF listener */
3725                 ETHER_BPF_MTAP(ifp, m_head);
3726         }
3727         if (idx >= 0)
3728                 E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3729         txr->tx_running = IGB_TX_RUNNING;
3730 }
3731
3732 static void
3733 igb_watchdog(struct ifaltq_subque *ifsq)
3734 {
3735         struct igb_tx_ring *txr = ifsq_get_priv(ifsq);
3736         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3737         struct igb_softc *sc = ifp->if_softc;
3738         int i;
3739
3740         KKASSERT(txr->ifsq == ifsq);
3741         ASSERT_IFNET_SERIALIZED_ALL(ifp);
3742
3743         /*
3744          * If flow control has paused us since last checking
3745          * it invalidates the watchdog timing, so dont run it.
3746          */
3747         if (sc->pause_frames) {
3748                 sc->pause_frames = 0;
3749                 txr->tx_watchdog.wd_timer = 5;
3750                 return;
3751         }
3752
3753         if_printf(ifp, "Watchdog timeout -- resetting\n");
3754         if_printf(ifp, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
3755             E1000_READ_REG(&sc->hw, E1000_TDH(txr->me)),
3756             E1000_READ_REG(&sc->hw, E1000_TDT(txr->me)));
3757         if_printf(ifp, "TX(%d) desc avail = %d, "
3758             "Next TX to Clean = %d\n",
3759             txr->me, txr->tx_avail, txr->next_to_clean);
3760
3761         IFNET_STAT_INC(ifp, oerrors, 1);
3762         sc->watchdog_events++;
3763
3764         igb_init(sc);
3765         for (i = 0; i < sc->tx_ring_inuse; ++i)
3766                 ifsq_devstart_sched(sc->tx_rings[i].ifsq);
3767 }
3768
3769 static void
3770 igb_set_eitr(struct igb_softc *sc, int idx, int rate)
3771 {
3772         uint32_t eitr = 0;
3773
3774         if (rate > 0) {
3775                 if (sc->hw.mac.type == e1000_82575) {
3776                         eitr = 1000000000 / 256 / rate;
3777                         /*
3778                          * NOTE:
3779                          * Document is wrong on the 2 bits left shift
3780                          */
3781                 } else {
3782                         eitr = 1000000 / rate;
3783                         eitr <<= IGB_EITR_INTVL_SHIFT;
3784                 }
3785
3786                 if (eitr == 0) {
3787                         /* Don't disable it */
3788                         eitr = 1 << IGB_EITR_INTVL_SHIFT;
3789                 } else if (eitr > IGB_EITR_INTVL_MASK) {
3790                         /* Don't allow it to be too large */
3791                         eitr = IGB_EITR_INTVL_MASK;
3792                 }
3793         }
3794         if (sc->hw.mac.type == e1000_82575)
3795                 eitr |= eitr << 16;
3796         else
3797                 eitr |= E1000_EITR_CNT_IGNR;
3798         E1000_WRITE_REG(&sc->hw, E1000_EITR(idx), eitr);
3799 }
3800
3801 static void
3802 igb_add_intr_rate_sysctl(struct igb_softc *sc, int use,
3803     const char *name, const char *desc)
3804 {
3805         int i;
3806
3807         for (i = 0; i < sc->intr_cnt; ++i) {
3808                 if (sc->intr_data[i].intr_use == use) {
3809                         SYSCTL_ADD_PROC(device_get_sysctl_ctx(sc->dev),
3810                             SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)),
3811                             OID_AUTO, name, CTLTYPE_INT | CTLFLAG_RW,
3812                             sc, use, igb_sysctl_intr_rate, "I", desc);
3813                         break;
3814                 }
3815         }
3816 }
3817
3818 static int
3819 igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS)
3820 {
3821         struct igb_softc *sc = (void *)arg1;
3822         int use = arg2;
3823         struct ifnet *ifp = &sc->arpcom.ac_if;
3824         int error, rate, i;
3825         struct igb_intr_data *intr;
3826
3827         rate = 0;
3828         for (i = 0; i < sc->intr_cnt; ++i) {
3829                 intr = &sc->intr_data[i];
3830                 if (intr->intr_use == use) {
3831                         rate = intr->intr_rate;
3832                         break;
3833                 }
3834         }
3835
3836         error = sysctl_handle_int(oidp, &rate, 0, req);
3837         if (error || req->newptr == NULL)
3838                 return error;
3839         if (rate <= 0)
3840                 return EINVAL;
3841
3842         ifnet_serialize_all(ifp);
3843
3844         for (i = 0; i < sc->intr_cnt; ++i) {
3845                 intr = &sc->intr_data[i];
3846                 if (intr->intr_use == use && intr->intr_rate != rate) {
3847                         intr->intr_rate = rate;
3848                         if (ifp->if_flags & IFF_RUNNING)
3849                                 igb_set_eitr(sc, i, rate);
3850                 }
3851         }
3852
3853         ifnet_deserialize_all(ifp);
3854
3855         return error;
3856 }
3857
3858 static int
3859 igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS)
3860 {
3861         struct igb_softc *sc = (void *)arg1;
3862         struct ifnet *ifp = &sc->arpcom.ac_if;
3863         struct igb_tx_ring *txr = &sc->tx_rings[0];
3864         int error, nsegs;
3865
3866         nsegs = txr->intr_nsegs;
3867         error = sysctl_handle_int(oidp, &nsegs, 0, req);
3868         if (error || req->newptr == NULL)
3869                 return error;
3870         if (nsegs <= 0)
3871                 return EINVAL;
3872
3873         ifnet_serialize_all(ifp);
3874
3875         if (nsegs >= txr->num_tx_desc - IGB_MAX_SCATTER - IGB_TX_RESERVED) {
3876                 error = EINVAL;
3877         } else {
3878                 int i;
3879
3880                 error = 0;
3881                 for (i = 0; i < sc->tx_ring_cnt; ++i)
3882                         sc->tx_rings[i].intr_nsegs = nsegs;
3883         }
3884
3885         ifnet_deserialize_all(ifp);
3886
3887         return error;
3888 }
3889
3890 static int
3891 igb_sysctl_rx_wreg_nsegs(SYSCTL_HANDLER_ARGS)
3892 {
3893         struct igb_softc *sc = (void *)arg1;
3894         struct ifnet *ifp = &sc->arpcom.ac_if;
3895         int error, nsegs, i;
3896
3897         nsegs = sc->rx_rings[0].wreg_nsegs;
3898         error = sysctl_handle_int(oidp, &nsegs, 0, req);
3899         if (error || req->newptr == NULL)
3900                 return error;
3901
3902         ifnet_serialize_all(ifp);
3903         for (i = 0; i < sc->rx_ring_cnt; ++i)
3904                 sc->rx_rings[i].wreg_nsegs = nsegs;
3905         ifnet_deserialize_all(ifp);
3906
3907         return 0;
3908 }
3909
3910 static int
3911 igb_sysctl_tx_wreg_nsegs(SYSCTL_HANDLER_ARGS)
3912 {
3913         struct igb_softc *sc = (void *)arg1;
3914         struct ifnet *ifp = &sc->arpcom.ac_if;
3915         int error, nsegs, i;
3916
3917         nsegs = sc->tx_rings[0].wreg_nsegs;
3918         error = sysctl_handle_int(oidp, &nsegs, 0, req);
3919         if (error || req->newptr == NULL)
3920                 return error;
3921
3922         ifnet_serialize_all(ifp);
3923         for (i = 0; i < sc->tx_ring_cnt; ++i)
3924                 sc->tx_rings[i].wreg_nsegs = nsegs;
3925         ifnet_deserialize_all(ifp);
3926
3927         return 0;
3928 }
3929
3930 static void
3931 igb_init_intr(struct igb_softc *sc)
3932 {
3933         int i;
3934
3935         igb_set_intr_mask(sc);
3936
3937         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0)
3938                 igb_init_unshared_intr(sc);
3939
3940         for (i = 0; i < sc->intr_cnt; ++i)
3941                 igb_set_eitr(sc, i, sc->intr_data[i].intr_rate);
3942 }
3943
3944 static void
3945 igb_init_unshared_intr(struct igb_softc *sc)
3946 {
3947         struct e1000_hw *hw = &sc->hw;
3948         const struct igb_rx_ring *rxr;
3949         const struct igb_tx_ring *txr;
3950         uint32_t ivar, index;
3951         int i;
3952
3953         /*
3954          * Enable extended mode
3955          */
3956         if (sc->hw.mac.type != e1000_82575) {
3957                 uint32_t gpie;
3958                 int ivar_max;
3959
3960                 gpie = E1000_GPIE_NSICR;
3961                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3962                         gpie |= E1000_GPIE_MSIX_MODE |
3963                             E1000_GPIE_EIAME |
3964                             E1000_GPIE_PBA;
3965                 }
3966                 E1000_WRITE_REG(hw, E1000_GPIE, gpie);
3967
3968                 /*
3969                  * Clear IVARs
3970                  */
3971                 switch (sc->hw.mac.type) {
3972                 case e1000_82576:
3973                         ivar_max = IGB_MAX_IVAR_82576;
3974                         break;
3975
3976                 case e1000_82580:
3977                         ivar_max = IGB_MAX_IVAR_82580;
3978                         break;
3979
3980                 case e1000_i350:
3981                         ivar_max = IGB_MAX_IVAR_I350;
3982                         break;
3983
3984                 case e1000_i354:
3985                         ivar_max = IGB_MAX_IVAR_I354;
3986                         break;
3987
3988                 case e1000_vfadapt:
3989                 case e1000_vfadapt_i350:
3990                         ivar_max = IGB_MAX_IVAR_VF;
3991                         break;
3992
3993                 case e1000_i210:
3994                         ivar_max = IGB_MAX_IVAR_I210;
3995                         break;
3996
3997                 case e1000_i211:
3998                         ivar_max = IGB_MAX_IVAR_I211;
3999                         break;
4000
4001                 default:
4002                         panic("unknown mac type %d\n", sc->hw.mac.type);
4003                 }
4004                 for (i = 0; i < ivar_max; ++i)
4005                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, i, 0);
4006                 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
4007         } else {
4008                 uint32_t tmp;
4009
4010                 KASSERT(sc->intr_type != PCI_INTR_TYPE_MSIX,
4011                     ("82575 w/ MSI-X"));
4012                 tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
4013                 tmp |= E1000_CTRL_EXT_IRCA;
4014                 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
4015         }
4016
4017         /*
4018          * Map TX/RX interrupts to EICR
4019          */
4020         switch (sc->hw.mac.type) {
4021         case e1000_82580:
4022         case e1000_i350:
4023         case e1000_i354:
4024         case e1000_vfadapt:
4025         case e1000_vfadapt_i350:
4026         case e1000_i210:
4027         case e1000_i211:
4028                 /* RX entries */
4029                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
4030                         rxr = &sc->rx_rings[i];
4031
4032                         index = i >> 1;
4033                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4034
4035                         if (i & 1) {
4036                                 ivar &= 0xff00ffff;
4037                                 ivar |=
4038                                 (rxr->rx_intr_vec | E1000_IVAR_VALID) << 16;
4039                         } else {
4040                                 ivar &= 0xffffff00;
4041                                 ivar |=
4042                                 (rxr->rx_intr_vec | E1000_IVAR_VALID);
4043                         }
4044                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4045                 }
4046                 /* TX entries */
4047                 for (i = 0; i < sc->tx_ring_inuse; ++i) {
4048                         txr = &sc->tx_rings[i];
4049
4050                         index = i >> 1;
4051                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4052
4053                         if (i & 1) {
4054                                 ivar &= 0x00ffffff;
4055                                 ivar |=
4056                                 (txr->tx_intr_vec | E1000_IVAR_VALID) << 24;
4057                         } else {
4058                                 ivar &= 0xffff00ff;
4059                                 ivar |=
4060                                 (txr->tx_intr_vec | E1000_IVAR_VALID) << 8;
4061                         }
4062                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4063                 }
4064                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
4065                         ivar = (sc->sts_msix_vec | E1000_IVAR_VALID) << 8;
4066                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4067                 }
4068                 break;
4069
4070         case e1000_82576:
4071                 /* RX entries */
4072                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
4073                         rxr = &sc->rx_rings[i];
4074
4075                         index = i & 0x7; /* Each IVAR has two entries */
4076                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4077
4078                         if (i < 8) {
4079                                 ivar &= 0xffffff00;
4080                                 ivar |=
4081                                 (rxr->rx_intr_vec | E1000_IVAR_VALID);
4082                         } else {
4083                                 ivar &= 0xff00ffff;
4084                                 ivar |=
4085                                 (rxr->rx_intr_vec | E1000_IVAR_VALID) << 16;
4086                         }
4087                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4088                 }
4089                 /* TX entries */
4090                 for (i = 0; i < sc->tx_ring_inuse; ++i) {
4091                         txr = &sc->tx_rings[i];
4092
4093                         index = i & 0x7; /* Each IVAR has two entries */
4094                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4095
4096                         if (i < 8) {
4097                                 ivar &= 0xffff00ff;
4098                                 ivar |=
4099                                 (txr->tx_intr_vec | E1000_IVAR_VALID) << 8;
4100                         } else {
4101                                 ivar &= 0x00ffffff;
4102                                 ivar |=
4103                                 (txr->tx_intr_vec | E1000_IVAR_VALID) << 24;
4104                         }
4105                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4106                 }
4107                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
4108                         ivar = (sc->sts_msix_vec | E1000_IVAR_VALID) << 8;
4109                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4110                 }
4111                 break;
4112
4113         case e1000_82575:
4114                 /*
4115                  * Enable necessary interrupt bits.
4116                  *
4117                  * The name of the register is confusing; in addition to
4118                  * configuring the first vector of MSI-X, it also configures
4119                  * which bits of EICR could be set by the hardware even when
4120                  * MSI or line interrupt is used; it thus controls interrupt
4121                  * generation.  It MUST be configured explicitly; the default
4122                  * value mentioned in the datasheet is wrong: RX queue0 and
4123                  * TX queue0 are NOT enabled by default.
4124                  */
4125                 E1000_WRITE_REG(&sc->hw, E1000_MSIXBM(0), sc->intr_mask);
4126                 break;
4127
4128         default:
4129                 panic("unknown mac type %d\n", sc->hw.mac.type);
4130         }
4131 }
4132
4133 static int
4134 igb_setup_intr(struct igb_softc *sc)
4135 {
4136         int i;
4137
4138         for (i = 0; i < sc->intr_cnt; ++i) {
4139                 struct igb_intr_data *intr = &sc->intr_data[i];
4140                 int error;
4141
4142                 error = bus_setup_intr_descr(sc->dev, intr->intr_res,
4143                     INTR_MPSAFE, intr->intr_func, intr->intr_funcarg,
4144                     &intr->intr_hand, intr->intr_serialize, intr->intr_desc);
4145                 if (error) {
4146                         device_printf(sc->dev, "can't setup %dth intr\n", i);
4147                         igb_teardown_intr(sc, i);
4148                         return error;
4149                 }
4150         }
4151         return 0;
4152 }
4153
4154 static void
4155 igb_set_txintr_mask(struct igb_tx_ring *txr, int *intr_vec0, int intr_vecmax)
4156 {
4157         if (txr->sc->hw.mac.type == e1000_82575) {
4158                 txr->tx_intr_vec = 0;   /* unused */
4159                 switch (txr->me) {
4160                 case 0:
4161                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE0;
4162                         break;
4163                 case 1:
4164                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE1;
4165                         break;
4166                 case 2:
4167                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE2;
4168                         break;
4169                 case 3:
4170                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE3;
4171                         break;
4172                 default:
4173                         panic("unsupported # of TX ring, %d\n", txr->me);
4174                 }
4175         } else {
4176                 int intr_vec = *intr_vec0;
4177
4178                 txr->tx_intr_vec = intr_vec % intr_vecmax;
4179                 txr->tx_intr_mask = 1 << txr->tx_intr_vec;
4180
4181                 *intr_vec0 = intr_vec + 1;
4182         }
4183 }
4184
4185 static void
4186 igb_set_rxintr_mask(struct igb_rx_ring *rxr, int *intr_vec0, int intr_vecmax)
4187 {
4188         if (rxr->sc->hw.mac.type == e1000_82575) {
4189                 rxr->rx_intr_vec = 0;   /* unused */
4190                 switch (rxr->me) {
4191                 case 0:
4192                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE0;
4193                         break;
4194                 case 1:
4195                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE1;
4196                         break;
4197                 case 2:
4198                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE2;
4199                         break;
4200                 case 3:
4201                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE3;
4202                         break;
4203                 default:
4204                         panic("unsupported # of RX ring, %d\n", rxr->me);
4205                 }
4206         } else {
4207                 int intr_vec = *intr_vec0;
4208
4209                 rxr->rx_intr_vec = intr_vec % intr_vecmax;
4210                 rxr->rx_intr_mask = 1 << rxr->rx_intr_vec;
4211
4212                 *intr_vec0 = intr_vec + 1;
4213         }
4214 }
4215
4216 static void
4217 igb_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
4218 {
4219         struct igb_softc *sc = ifp->if_softc;
4220
4221         ifnet_serialize_array_enter(sc->serializes, sc->serialize_cnt, slz);
4222 }
4223
4224 static void
4225 igb_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
4226 {
4227         struct igb_softc *sc = ifp->if_softc;
4228
4229         ifnet_serialize_array_exit(sc->serializes, sc->serialize_cnt, slz);
4230 }
4231
4232 static int
4233 igb_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
4234 {
4235         struct igb_softc *sc = ifp->if_softc;
4236
4237         return ifnet_serialize_array_try(sc->serializes, sc->serialize_cnt,
4238             slz);
4239 }
4240
4241 #ifdef INVARIANTS
4242
4243 static void
4244 igb_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
4245     boolean_t serialized)
4246 {
4247         struct igb_softc *sc = ifp->if_softc;
4248
4249         ifnet_serialize_array_assert(sc->serializes, sc->serialize_cnt,
4250             slz, serialized);
4251 }
4252
4253 #endif  /* INVARIANTS */
4254
4255 static void
4256 igb_set_intr_mask(struct igb_softc *sc)
4257 {
4258         int i;
4259
4260         sc->intr_mask = sc->sts_intr_mask;
4261         for (i = 0; i < sc->rx_ring_inuse; ++i)
4262                 sc->intr_mask |= sc->rx_rings[i].rx_intr_mask;
4263         for (i = 0; i < sc->tx_ring_inuse; ++i)
4264                 sc->intr_mask |= sc->tx_rings[i].tx_intr_mask;
4265         if (bootverbose) {
4266                 if_printf(&sc->arpcom.ac_if, "intr mask 0x%08x\n",
4267                     sc->intr_mask);
4268         }
4269 }
4270
4271 static int
4272 igb_alloc_intr(struct igb_softc *sc)
4273 {
4274         struct igb_tx_ring *txr;
4275         struct igb_intr_data *intr;
4276         int i, intr_vec, intr_vecmax;
4277         u_int intr_flags;
4278
4279         igb_alloc_msix(sc);
4280         if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
4281                 igb_set_ring_inuse(sc, FALSE);
4282                 goto done;
4283         }
4284
4285         /*
4286          * Reset some settings changed by igb_alloc_msix().
4287          */
4288         if (sc->rx_rmap_intr != NULL) {
4289                 if_ringmap_free(sc->rx_rmap_intr);
4290                 sc->rx_rmap_intr = NULL;
4291         }
4292         if (sc->tx_rmap_intr != NULL) {
4293                 if_ringmap_free(sc->tx_rmap_intr);
4294                 sc->tx_rmap_intr = NULL;
4295         }
4296         if (sc->intr_data != NULL) {
4297                 kfree(sc->intr_data, M_DEVBUF);
4298                 sc->intr_data = NULL;
4299         }
4300         for (i = 0; i < sc->tx_ring_cnt; ++i) {
4301                 txr = &sc->tx_rings[i];
4302                 txr->tx_intr_vec = 0;
4303                 txr->tx_intr_mask = 0;
4304                 txr->tx_intr_cpuid = -1;
4305         }
4306         for (i = 0; i < sc->rx_ring_cnt; ++i) {
4307                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
4308
4309                 rxr->rx_intr_vec = 0;
4310                 rxr->rx_intr_mask = 0;
4311                 rxr->rx_txr = NULL;
4312         }
4313
4314         sc->intr_cnt = 1;
4315         sc->intr_data = kmalloc(sizeof(struct igb_intr_data), M_DEVBUF,
4316             M_WAITOK | M_ZERO);
4317         intr = &sc->intr_data[0];
4318
4319         /*
4320          * Allocate MSI/legacy interrupt resource
4321          */
4322         sc->intr_type = pci_alloc_1intr(sc->dev, igb_msi_enable,
4323             &intr->intr_rid, &intr_flags);
4324
4325         if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
4326                 int unshared;
4327
4328                 unshared = device_getenv_int(sc->dev, "irq.unshared", 0);
4329                 if (!unshared) {
4330                         sc->flags |= IGB_FLAG_SHARED_INTR;
4331                         if (bootverbose)
4332                                 device_printf(sc->dev, "IRQ shared\n");
4333                 } else {
4334                         intr_flags &= ~RF_SHAREABLE;
4335                         if (bootverbose)
4336                                 device_printf(sc->dev, "IRQ unshared\n");
4337                 }
4338         }
4339
4340         intr->intr_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4341             &intr->intr_rid, intr_flags);
4342         if (intr->intr_res == NULL) {
4343                 device_printf(sc->dev, "Unable to allocate bus resource: "
4344                     "interrupt\n");
4345                 return ENXIO;
4346         }
4347
4348         intr->intr_serialize = &sc->main_serialize;
4349         intr->intr_cpuid = rman_get_cpuid(intr->intr_res);
4350         intr->intr_func = (sc->flags & IGB_FLAG_SHARED_INTR) ?
4351             igb_intr_shared : igb_intr;
4352         intr->intr_funcarg = sc;
4353         intr->intr_rate = IGB_INTR_RATE;
4354         intr->intr_use = IGB_INTR_USE_RXTX;
4355
4356         sc->tx_rings[0].tx_intr_cpuid = intr->intr_cpuid;
4357
4358         /*
4359          * Setup MSI/legacy interrupt mask
4360          */
4361         switch (sc->hw.mac.type) {
4362         case e1000_82575:
4363                 intr_vecmax = IGB_MAX_TXRXINT_82575;
4364                 break;
4365
4366         case e1000_82576:
4367                 intr_vecmax = IGB_MAX_TXRXINT_82576;
4368                 break;
4369
4370         case e1000_82580:
4371                 intr_vecmax = IGB_MAX_TXRXINT_82580;
4372                 break;
4373
4374         case e1000_i350:
4375                 intr_vecmax = IGB_MAX_TXRXINT_I350;
4376                 break;
4377
4378         case e1000_i354:
4379                 intr_vecmax = IGB_MAX_TXRXINT_I354;
4380                 break;
4381
4382         case e1000_i210:
4383                 intr_vecmax = IGB_MAX_TXRXINT_I210;
4384                 break;
4385
4386         case e1000_i211:
4387                 intr_vecmax = IGB_MAX_TXRXINT_I211;
4388                 break;
4389
4390         default:
4391                 intr_vecmax = IGB_MIN_TXRXINT;
4392                 break;
4393         }
4394         intr_vec = 0;
4395         for (i = 0; i < sc->tx_ring_cnt; ++i)
4396                 igb_set_txintr_mask(&sc->tx_rings[i], &intr_vec, intr_vecmax);
4397         for (i = 0; i < sc->rx_ring_cnt; ++i)
4398                 igb_set_rxintr_mask(&sc->rx_rings[i], &intr_vec, intr_vecmax);
4399         sc->sts_intr_mask = E1000_EICR_OTHER;
4400
4401         igb_set_ring_inuse(sc, FALSE);
4402         KKASSERT(sc->rx_ring_inuse <= IGB_MIN_RING_RSS);
4403         if (sc->rx_ring_inuse == IGB_MIN_RING_RSS) {
4404                 /*
4405                  * Allocate RX ring map for RSS setup.
4406                  */
4407                 sc->rx_rmap_intr = if_ringmap_alloc(sc->dev,
4408                     IGB_MIN_RING_RSS, IGB_MIN_RING_RSS);
4409                 KASSERT(if_ringmap_count(sc->rx_rmap_intr) ==
4410                     sc->rx_ring_inuse, ("RX ring inuse mismatch"));
4411         }
4412 done:
4413         igb_set_intr_mask(sc);
4414         for (i = 0; i < sc->tx_ring_cnt; ++i) {
4415                 txr = &sc->tx_rings[i];
4416                 if (txr->tx_intr_cpuid < 0)
4417                         txr->tx_intr_cpuid = 0;
4418         }
4419         return 0;
4420 }
4421
4422 static void
4423 igb_free_intr(struct igb_softc *sc)
4424 {
4425         if (sc->intr_data == NULL)
4426                 return;
4427
4428         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
4429                 struct igb_intr_data *intr = &sc->intr_data[0];
4430
4431                 KKASSERT(sc->intr_cnt == 1);
4432                 if (intr->intr_res != NULL) {
4433                         bus_release_resource(sc->dev, SYS_RES_IRQ,
4434                             intr->intr_rid, intr->intr_res);
4435                 }
4436                 if (sc->intr_type == PCI_INTR_TYPE_MSI)
4437                         pci_release_msi(sc->dev);
4438
4439                 kfree(sc->intr_data, M_DEVBUF);
4440         } else {
4441                 igb_free_msix(sc, TRUE);
4442         }
4443 }
4444
4445 static void
4446 igb_teardown_intr(struct igb_softc *sc, int intr_cnt)
4447 {
4448         int i;
4449
4450         if (sc->intr_data == NULL)
4451                 return;
4452
4453         for (i = 0; i < intr_cnt; ++i) {
4454                 struct igb_intr_data *intr = &sc->intr_data[i];
4455
4456                 bus_teardown_intr(sc->dev, intr->intr_res, intr->intr_hand);
4457         }
4458 }
4459
4460 static void
4461 igb_alloc_msix(struct igb_softc *sc)
4462 {
4463         int msix_enable, msix_cnt, msix_ring, alloc_cnt;
4464         int i, x, error;
4465         int ring_cnt, ring_cntmax;
4466         struct igb_intr_data *intr;
4467         boolean_t setup = FALSE;
4468
4469         /*
4470          * Don't enable MSI-X on 82575, see:
4471          * 82575 specification update errata #25
4472          */
4473         if (sc->hw.mac.type == e1000_82575)
4474                 return;
4475
4476         /* Don't enable MSI-X on VF */
4477         if (sc->vf_ifp)
4478                 return;
4479
4480         msix_enable = device_getenv_int(sc->dev, "msix.enable",
4481             igb_msix_enable);
4482         if (!msix_enable)
4483                 return;
4484
4485         msix_cnt = pci_msix_count(sc->dev);
4486 #ifdef IGB_MSIX_DEBUG
4487         msix_cnt = device_getenv_int(sc->dev, "msix.count", msix_cnt);
4488 #endif
4489         if (msix_cnt <= 1) {
4490                 /* One MSI-X model does not make sense. */
4491                 return;
4492         }
4493         if (bootverbose)
4494                 device_printf(sc->dev, "MSI-X count %d\n", msix_cnt);
4495         msix_ring = msix_cnt - 1; /* -1 for status */
4496
4497         /*
4498          * Configure # of RX/TX rings usable by MSI-X.
4499          */
4500         igb_get_rxring_cnt(sc, &ring_cnt, &ring_cntmax);
4501         if (ring_cntmax > msix_ring)
4502                 ring_cntmax = msix_ring;
4503         sc->rx_rmap_intr = if_ringmap_alloc(sc->dev, ring_cnt, ring_cntmax);
4504
4505         igb_get_txring_cnt(sc, &ring_cnt, &ring_cntmax);
4506         if (ring_cntmax > msix_ring)
4507                 ring_cntmax = msix_ring;
4508         sc->tx_rmap_intr = if_ringmap_alloc(sc->dev, ring_cnt, ring_cntmax);
4509
4510         if_ringmap_match(sc->dev, sc->rx_rmap_intr, sc->tx_rmap_intr);
4511         sc->rx_ring_msix = if_ringmap_count(sc->rx_rmap_intr);
4512         KASSERT(sc->rx_ring_msix <= sc->rx_ring_cnt,
4513             ("total RX ring count %d, MSI-X RX ring count %d",
4514              sc->rx_ring_cnt, sc->rx_ring_msix));
4515         sc->tx_ring_msix = if_ringmap_count(sc->tx_rmap_intr);
4516         KASSERT(sc->tx_ring_msix <= sc->tx_ring_cnt,
4517             ("total TX ring count %d, MSI-X TX ring count %d",
4518              sc->tx_ring_cnt, sc->tx_ring_msix));
4519
4520         /*
4521          * Aggregate TX/RX MSI-X
4522          */
4523         ring_cntmax = sc->rx_ring_msix;
4524         if (ring_cntmax < sc->tx_ring_msix)
4525                 ring_cntmax = sc->tx_ring_msix;
4526         KASSERT(ring_cntmax <= msix_ring,
4527             ("invalid ring count max %d, MSI-X count for rings %d",
4528              ring_cntmax, msix_ring));
4529
4530         alloc_cnt = ring_cntmax + 1; /* +1 for status */
4531         if (bootverbose) {
4532                 device_printf(sc->dev, "MSI-X alloc %d, "
4533                     "RX ring %d, TX ring %d\n", alloc_cnt,
4534                     sc->rx_ring_msix, sc->tx_ring_msix);
4535         }
4536
4537         sc->msix_mem_rid = PCIR_BAR(IGB_MSIX_BAR);
4538         sc->msix_mem_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
4539             &sc->msix_mem_rid, RF_ACTIVE);
4540         if (sc->msix_mem_res == NULL) {
4541                 sc->msix_mem_rid = PCIR_BAR(IGB_MSIX_BAR_ALT);
4542                 sc->msix_mem_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
4543                     &sc->msix_mem_rid, RF_ACTIVE);
4544                 if (sc->msix_mem_res == NULL) {
4545                         device_printf(sc->dev, "Unable to map MSI-X table\n");
4546                         return;
4547                 }
4548         }
4549
4550         sc->intr_cnt = alloc_cnt;
4551         sc->intr_data = kmalloc(sizeof(struct igb_intr_data) * sc->intr_cnt,
4552             M_DEVBUF, M_WAITOK | M_ZERO);
4553         for (x = 0; x < sc->intr_cnt; ++x) {
4554                 intr = &sc->intr_data[x];
4555                 intr->intr_rid = -1;
4556                 intr->intr_rate = IGB_INTR_RATE;
4557         }
4558
4559         x = 0;
4560         for (i = 0; i < sc->rx_ring_msix; ++i) {
4561                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
4562                 struct igb_tx_ring *txr = NULL;
4563                 int cpuid, j;
4564
4565                 KKASSERT(x < sc->intr_cnt);
4566                 rxr->rx_intr_vec = x;
4567                 rxr->rx_intr_mask = 1 << rxr->rx_intr_vec;
4568
4569                 cpuid = if_ringmap_cpumap(sc->rx_rmap_intr, i);
4570
4571                 /*
4572                  * Try finding TX ring to piggyback.
4573                  */
4574                 for (j = 0; j < sc->tx_ring_msix; ++j) {
4575                         if (cpuid ==
4576                             if_ringmap_cpumap(sc->tx_rmap_intr, j)) {
4577                                 txr = &sc->tx_rings[j];
4578                                 KKASSERT(txr->tx_intr_cpuid < 0);
4579                                 break;
4580                         }
4581                 }
4582                 rxr->rx_txr = txr;
4583
4584                 intr = &sc->intr_data[x++];
4585                 intr->intr_serialize = &rxr->rx_serialize;
4586                 intr->intr_cpuid = cpuid;
4587                 KKASSERT(intr->intr_cpuid < netisr_ncpus);
4588                 intr->intr_funcarg = rxr;
4589                 if (txr != NULL) {
4590                         intr->intr_func = igb_msix_rxtx;
4591                         intr->intr_use = IGB_INTR_USE_RXTX;
4592                         ksnprintf(intr->intr_desc0, sizeof(intr->intr_desc0),
4593                             "%s rx%dtx%d", device_get_nameunit(sc->dev),
4594                             i, txr->me);
4595
4596                         txr->tx_intr_vec = rxr->rx_intr_vec;
4597                         txr->tx_intr_mask = rxr->rx_intr_mask;
4598                         txr->tx_intr_cpuid = intr->intr_cpuid;
4599                 } else {
4600                         intr->intr_func = igb_msix_rx;
4601                         intr->intr_rate = IGB_MSIX_RX_RATE;
4602                         intr->intr_use = IGB_INTR_USE_RX;
4603
4604                         ksnprintf(intr->intr_desc0, sizeof(intr->intr_desc0),
4605                             "%s rx%d", device_get_nameunit(sc->dev), i);
4606                 }
4607                 intr->intr_desc = intr->intr_desc0;
4608         }
4609
4610         for (i = 0; i < sc->tx_ring_msix; ++i) {
4611                 struct igb_tx_ring *txr = &sc->tx_rings[i];
4612
4613                 if (txr->tx_intr_cpuid >= 0) {
4614                         /* Piggybacked by RX ring. */
4615                         continue;
4616                 }
4617
4618                 KKASSERT(x < sc->intr_cnt);
4619                 txr->tx_intr_vec = x;
4620                 txr->tx_intr_mask = 1 << txr->tx_intr_vec;
4621
4622                 intr = &sc->intr_data[x++];
4623                 intr->intr_serialize = &txr->tx_serialize;
4624                 intr->intr_func = igb_msix_tx;
4625                 intr->intr_funcarg = txr;
4626                 intr->intr_rate = IGB_MSIX_TX_RATE;
4627                 intr->intr_use = IGB_INTR_USE_TX;
4628
4629                 intr->intr_cpuid = if_ringmap_cpumap(sc->tx_rmap_intr, i);
4630                 KKASSERT(intr->intr_cpuid < netisr_ncpus);
4631                 txr->tx_intr_cpuid = intr->intr_cpuid;
4632
4633                 ksnprintf(intr->intr_desc0, sizeof(intr->intr_desc0), "%s tx%d",
4634                     device_get_nameunit(sc->dev), i);
4635                 intr->intr_desc = intr->intr_desc0;
4636         }
4637
4638         /*
4639          * Link status
4640          */
4641         KKASSERT(x < sc->intr_cnt);
4642         sc->sts_msix_vec = x;
4643         sc->sts_intr_mask = 1 << sc->sts_msix_vec;
4644
4645         intr = &sc->intr_data[x++];
4646         intr->intr_serialize = &sc->main_serialize;
4647         intr->intr_func = igb_msix_status;
4648         intr->intr_funcarg = sc;
4649         intr->intr_cpuid = 0;
4650         intr->intr_use = IGB_INTR_USE_STATUS;
4651
4652         ksnprintf(intr->intr_desc0, sizeof(intr->intr_desc0), "%s sts",
4653             device_get_nameunit(sc->dev));
4654         intr->intr_desc = intr->intr_desc0;
4655
4656         KKASSERT(x == sc->intr_cnt);
4657
4658         error = pci_setup_msix(sc->dev);
4659         if (error) {
4660                 device_printf(sc->dev, "Setup MSI-X failed\n");
4661                 goto back;
4662         }
4663         setup = TRUE;
4664
4665         for (i = 0; i < sc->intr_cnt; ++i) {
4666                 intr = &sc->intr_data[i];
4667
4668                 error = pci_alloc_msix_vector(sc->dev, i, &intr->intr_rid,
4669                     intr->intr_cpuid);
4670                 if (error) {
4671                         device_printf(sc->dev,
4672                             "Unable to allocate MSI-X %d on cpu%d\n", i,
4673                             intr->intr_cpuid);
4674                         goto back;
4675                 }
4676
4677                 intr->intr_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4678                     &intr->intr_rid, RF_ACTIVE);
4679                 if (intr->intr_res == NULL) {
4680                         device_printf(sc->dev,
4681                             "Unable to allocate MSI-X %d resource\n", i);
4682                         error = ENOMEM;
4683                         goto back;
4684                 }
4685         }
4686
4687         pci_enable_msix(sc->dev);
4688         sc->intr_type = PCI_INTR_TYPE_MSIX;
4689 back:
4690         if (error)
4691                 igb_free_msix(sc, setup);
4692 }
4693
4694 static void
4695 igb_free_msix(struct igb_softc *sc, boolean_t setup)
4696 {
4697         int i;
4698
4699         KKASSERT(sc->intr_cnt > 1);
4700
4701         for (i = 0; i < sc->intr_cnt; ++i) {
4702                 struct igb_intr_data *intr = &sc->intr_data[i];
4703
4704                 if (intr->intr_res != NULL) {
4705                         bus_release_resource(sc->dev, SYS_RES_IRQ,
4706                             intr->intr_rid, intr->intr_res);
4707                 }
4708                 if (intr->intr_rid >= 0)
4709                         pci_release_msix_vector(sc->dev, intr->intr_rid);
4710         }
4711         if (setup)
4712                 pci_teardown_msix(sc->dev);
4713
4714         sc->intr_cnt = 0;
4715         kfree(sc->intr_data, M_DEVBUF);
4716         sc->intr_data = NULL;
4717 }
4718
4719 static void
4720 igb_msix_rx(void *arg)
4721 {
4722         struct igb_rx_ring *rxr = arg;
4723
4724         ASSERT_SERIALIZED(&rxr->rx_serialize);
4725         igb_rxeof(rxr, -1);
4726
4727         E1000_WRITE_REG(&rxr->sc->hw, E1000_EIMS, rxr->rx_intr_mask);
4728 }
4729
4730 static void
4731 igb_msix_tx(void *arg)
4732 {
4733         struct igb_tx_ring *txr = arg;
4734
4735         ASSERT_SERIALIZED(&txr->tx_serialize);
4736
4737         igb_tx_intr(txr, *(txr->tx_hdr));
4738         E1000_WRITE_REG(&txr->sc->hw, E1000_EIMS, txr->tx_intr_mask);
4739 }
4740
4741 static void
4742 igb_msix_status(void *arg)
4743 {
4744         struct igb_softc *sc = arg;
4745         uint32_t icr;
4746
4747         ASSERT_SERIALIZED(&sc->main_serialize);
4748
4749         icr = E1000_READ_REG(&sc->hw, E1000_ICR);
4750         if (icr & E1000_ICR_LSC) {
4751                 sc->hw.mac.get_link_status = 1;
4752                 igb_update_link_status(sc);
4753         }
4754
4755         E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->sts_intr_mask);
4756 }
4757
4758 static void
4759 igb_set_ring_inuse(struct igb_softc *sc, boolean_t polling)
4760 {
4761         sc->rx_ring_inuse = igb_get_rxring_inuse(sc, polling);
4762         sc->tx_ring_inuse = igb_get_txring_inuse(sc, polling);
4763         if (bootverbose) {
4764                 if_printf(&sc->arpcom.ac_if, "RX rings %d/%d, TX rings %d/%d\n",
4765                     sc->rx_ring_inuse, sc->rx_ring_cnt,
4766                     sc->tx_ring_inuse, sc->tx_ring_cnt);
4767         }
4768 }
4769
4770 static int
4771 igb_get_rxring_inuse(const struct igb_softc *sc, boolean_t polling)
4772 {
4773         if (!IGB_ENABLE_HWRSS(sc))
4774                 return 1;
4775
4776         if (polling)
4777                 return sc->rx_ring_cnt;
4778         else if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4779                 return IGB_MIN_RING_RSS;
4780         else
4781                 return sc->rx_ring_msix;
4782 }
4783
4784 static int
4785 igb_get_txring_inuse(const struct igb_softc *sc, boolean_t polling)
4786 {
4787         if (!IGB_ENABLE_HWTSS(sc))
4788                 return 1;
4789
4790         if (polling)
4791                 return sc->tx_ring_cnt;
4792         else if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4793                 return IGB_MIN_RING;
4794         else
4795                 return sc->tx_ring_msix;
4796 }
4797
4798 static int
4799 igb_tso_pullup(struct igb_tx_ring *txr, struct mbuf **mp)
4800 {
4801         int hoff, iphlen, thoff;
4802         struct mbuf *m;
4803
4804         m = *mp;
4805         KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4806
4807         iphlen = m->m_pkthdr.csum_iphlen;
4808         thoff = m->m_pkthdr.csum_thlen;
4809         hoff = m->m_pkthdr.csum_lhlen;
4810
4811         KASSERT(iphlen > 0, ("invalid ip hlen"));
4812         KASSERT(thoff > 0, ("invalid tcp hlen"));
4813         KASSERT(hoff > 0, ("invalid ether hlen"));
4814
4815         if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4816                 m = m_pullup(m, hoff + iphlen + thoff);
4817                 if (m == NULL) {
4818                         *mp = NULL;
4819                         return ENOBUFS;
4820                 }
4821                 *mp = m;
4822         }
4823         if (txr->tx_flags & IGB_TXFLAG_TSO_IPLEN0) {
4824                 struct ip *ip;
4825
4826                 ip = mtodoff(m, struct ip *, hoff);
4827                 ip->ip_len = 0;
4828         }
4829
4830         return 0;
4831 }
4832
4833 static void
4834 igb_tso_ctx(struct igb_tx_ring *txr, struct mbuf *m, uint32_t *hlen)
4835 {
4836         struct e1000_adv_tx_context_desc *TXD;
4837         uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
4838         int hoff, ctxd, iphlen, thoff;
4839
4840         iphlen = m->m_pkthdr.csum_iphlen;
4841         thoff = m->m_pkthdr.csum_thlen;
4842         hoff = m->m_pkthdr.csum_lhlen;
4843
4844         vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
4845
4846         ctxd = txr->next_avail_desc;
4847         TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
4848
4849         if (m->m_flags & M_VLANTAG) {
4850                 uint16_t vlantag;
4851
4852                 vlantag = htole16(m->m_pkthdr.ether_vlantag);
4853                 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
4854         }
4855
4856         vlan_macip_lens |= (hoff << E1000_ADVTXD_MACLEN_SHIFT);
4857         vlan_macip_lens |= iphlen;
4858
4859         type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4860         type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
4861         type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
4862
4863         mss_l4len_idx |= (m->m_pkthdr.tso_segsz << E1000_ADVTXD_MSS_SHIFT);
4864         mss_l4len_idx |= (thoff << E1000_ADVTXD_L4LEN_SHIFT);
4865
4866         /*
4867          * 82575 needs the TX context index added; the queue
4868          * index is used as TX context index here.
4869          */
4870         if (txr->sc->hw.mac.type == e1000_82575)
4871                 mss_l4len_idx |= txr->me << 4;
4872
4873         TXD->vlan_macip_lens = htole32(vlan_macip_lens);
4874         TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
4875         TXD->seqnum_seed = htole32(0);
4876         TXD->mss_l4len_idx = htole32(mss_l4len_idx);
4877
4878         /* We've consumed the first desc, adjust counters */
4879         if (++ctxd == txr->num_tx_desc)
4880                 ctxd = 0;
4881         txr->next_avail_desc = ctxd;
4882         --txr->tx_avail;
4883
4884         *hlen = hoff + iphlen + thoff;
4885 }
4886
4887 static void
4888 igb_setup_serialize(struct igb_softc *sc)
4889 {
4890         int i = 0, j;
4891
4892         /* Main + RX + TX */
4893         sc->serialize_cnt = 1 + sc->rx_ring_cnt + sc->tx_ring_cnt;
4894         sc->serializes =
4895             kmalloc(sc->serialize_cnt * sizeof(struct lwkt_serialize *),
4896                 M_DEVBUF, M_WAITOK | M_ZERO);
4897
4898         /*
4899          * Setup serializes
4900          *
4901          * NOTE: Order is critical
4902          */
4903
4904         KKASSERT(i < sc->serialize_cnt);
4905         sc->serializes[i++] = &sc->main_serialize;
4906
4907         for (j = 0; j < sc->rx_ring_cnt; ++j) {
4908                 KKASSERT(i < sc->serialize_cnt);
4909                 sc->serializes[i++] = &sc->rx_rings[j].rx_serialize;
4910         }
4911
4912         for (j = 0; j < sc->tx_ring_cnt; ++j) {
4913                 KKASSERT(i < sc->serialize_cnt);
4914                 sc->serializes[i++] = &sc->tx_rings[j].tx_serialize;
4915         }
4916
4917         KKASSERT(i == sc->serialize_cnt);
4918 }
4919
4920 static void
4921 igb_msix_rxtx(void *arg)
4922 {
4923         struct igb_rx_ring *rxr = arg;
4924         struct igb_tx_ring *txr;
4925         int hdr;
4926
4927         ASSERT_SERIALIZED(&rxr->rx_serialize);
4928
4929         igb_rxeof(rxr, -1);
4930
4931         /*
4932          * NOTE:
4933          * Since next_to_clean is only changed by igb_txeof(),
4934          * which is called only in interrupt handler, the
4935          * check w/o holding tx serializer is MPSAFE.
4936          */
4937         txr = rxr->rx_txr;
4938         hdr = *(txr->tx_hdr);
4939         if (hdr != txr->next_to_clean) {
4940                 lwkt_serialize_enter(&txr->tx_serialize);
4941                 igb_tx_intr(txr, hdr);
4942                 lwkt_serialize_exit(&txr->tx_serialize);
4943         }
4944
4945         E1000_WRITE_REG(&rxr->sc->hw, E1000_EIMS, rxr->rx_intr_mask);
4946 }
4947
4948 static void
4949 igb_set_timer_cpuid(struct igb_softc *sc, boolean_t polling)
4950 {
4951         if (polling || sc->intr_type == PCI_INTR_TYPE_MSIX)
4952                 sc->timer_cpuid = 0; /* XXX fixed */
4953         else
4954                 sc->timer_cpuid = rman_get_cpuid(sc->intr_data[0].intr_res);
4955 }
4956
4957 static void
4958 igb_init_dmac(struct igb_softc *sc, uint32_t pba)
4959 {
4960         struct e1000_hw *hw = &sc->hw;
4961         uint32_t reg;
4962
4963         if (hw->mac.type == e1000_i211)
4964                 return;
4965
4966         if (hw->mac.type > e1000_82580) {
4967                 uint32_t dmac;
4968                 uint16_t hwm;
4969
4970                 if (sc->dma_coalesce == 0) { /* Disabling it */
4971                         reg = ~E1000_DMACR_DMAC_EN;
4972                         E1000_WRITE_REG(hw, E1000_DMACR, reg);
4973                         return;
4974                 } else {
4975                         if_printf(&sc->arpcom.ac_if,
4976                             "DMA Coalescing enabled\n");
4977                 }
4978
4979                 /* Set starting threshold */
4980                 E1000_WRITE_REG(hw, E1000_DMCTXTH, 0);
4981
4982                 hwm = 64 * pba - sc->max_frame_size / 16;
4983                 if (hwm < 64 * (pba - 6))
4984                         hwm = 64 * (pba - 6);
4985                 reg = E1000_READ_REG(hw, E1000_FCRTC);
4986                 reg &= ~E1000_FCRTC_RTH_COAL_MASK;
4987                 reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
4988                     & E1000_FCRTC_RTH_COAL_MASK);
4989                 E1000_WRITE_REG(hw, E1000_FCRTC, reg);
4990
4991                 dmac = pba - sc->max_frame_size / 512;
4992                 if (dmac < pba - 10)
4993                         dmac = pba - 10;
4994                 reg = E1000_READ_REG(hw, E1000_DMACR);
4995                 reg &= ~E1000_DMACR_DMACTHR_MASK;
4996                 reg |= ((dmac << E1000_DMACR_DMACTHR_SHIFT)
4997                     & E1000_DMACR_DMACTHR_MASK);
4998
4999                 /* transition to L0x or L1 if available..*/
5000                 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
5001
5002                 /*
5003                  * Check if status is 2.5Gb backplane connection
5004                  * before configuration of watchdog timer, which
5005                  * is in msec values in 12.8usec intervals watchdog
5006                  * timer = msec values in 32usec intervals for non
5007                  * 2.5Gb connection.
5008                  */
5009                 if (hw->mac.type == e1000_i354) {
5010                         int status = E1000_READ_REG(hw, E1000_STATUS);
5011
5012                         if ((status & E1000_STATUS_2P5_SKU) &&
5013                             !(status & E1000_STATUS_2P5_SKU_OVER))
5014                                 reg |= ((sc->dma_coalesce * 5) >> 6);
5015                         else
5016                                 reg |= (sc->dma_coalesce >> 5);
5017                 } else {
5018                         reg |= (sc->dma_coalesce >> 5);
5019                 }
5020
5021                 E1000_WRITE_REG(hw, E1000_DMACR, reg);
5022
5023                 E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
5024
5025                 /* Set the interval before transition */
5026                 reg = E1000_READ_REG(hw, E1000_DMCTLX);
5027                 if (hw->mac.type == e1000_i350)
5028                         reg |= IGB_DMCTLX_DCFLUSH_DIS;
5029                 /*
5030                  * In 2.5Gb connection, TTLX unit is 0.4 usec, which
5031                  * is 0x4*2 = 0xA.  But delay is still 4 usec.
5032                  */
5033                 if (hw->mac.type == e1000_i354) {
5034                         int status = E1000_READ_REG(hw, E1000_STATUS);
5035
5036                         if ((status & E1000_STATUS_2P5_SKU) &&
5037                             !(status & E1000_STATUS_2P5_SKU_OVER))
5038                                 reg |= 0xA;
5039                         else
5040                                 reg |= 0x4;
5041                 } else {
5042                         reg |= 0x4;
5043                 }
5044                 E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
5045
5046                 /* Free space in tx packet buffer to wake from DMA coal */
5047                 E1000_WRITE_REG(hw, E1000_DMCTXTH,
5048                     (IGB_TXPBSIZE - (2 * sc->max_frame_size)) >> 6);
5049
5050                 /* Make low power state decision controlled by DMA coal */
5051                 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
5052                 reg &= ~E1000_PCIEMISC_LX_DECISION;
5053                 E1000_WRITE_REG(hw, E1000_PCIEMISC, reg);
5054         } else if (hw->mac.type == e1000_82580) {
5055                 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
5056                 E1000_WRITE_REG(hw, E1000_PCIEMISC,
5057                     reg & ~E1000_PCIEMISC_LX_DECISION);
5058                 E1000_WRITE_REG(hw, E1000_DMACR, 0);
5059         }
5060 }
5061
5062 static void
5063 igb_reg_dump(struct igb_softc *sc)
5064 {
5065         device_t dev = sc->dev;
5066         int col = 0;
5067
5068 #define DUMPREG(regno)  \
5069         kprintf(" %13s=%08x", #regno + 6, E1000_READ_REG(&sc->hw, regno));\
5070         if (++col == 3) {               \
5071                 kprintf("\n");          \
5072                 col = 0;                \
5073         }                               \
5074
5075         device_printf(dev, "REGISTER DUMP\n");
5076         DUMPREG(E1000_CTRL);
5077         DUMPREG(E1000_STATUS);
5078         DUMPREG(E1000_EECD);
5079         DUMPREG(E1000_EERD);
5080         DUMPREG(E1000_CTRL_EXT);
5081         DUMPREG(E1000_FLA);
5082         DUMPREG(E1000_MDIC);
5083         DUMPREG(E1000_SCTL);
5084         DUMPREG(E1000_FCAL);
5085         DUMPREG(E1000_FCAH);
5086         DUMPREG(E1000_FCT);
5087         DUMPREG(E1000_CONNSW);
5088         DUMPREG(E1000_VET);
5089         DUMPREG(E1000_ICR);
5090         DUMPREG(E1000_ITR);
5091         DUMPREG(E1000_IMS);
5092         DUMPREG(E1000_IVAR);
5093         DUMPREG(E1000_SVCR);
5094         DUMPREG(E1000_SVT);
5095         DUMPREG(E1000_LPIC);
5096         DUMPREG(E1000_RCTL);
5097         DUMPREG(E1000_FCTTV);
5098         DUMPREG(E1000_TXCW);
5099         DUMPREG(E1000_RXCW);
5100         DUMPREG(E1000_EIMS);
5101         DUMPREG(E1000_EIAC);
5102         DUMPREG(E1000_EIAM);
5103         DUMPREG(E1000_GPIE);
5104         DUMPREG(E1000_IVAR0);
5105         DUMPREG(E1000_IVAR_MISC);
5106         DUMPREG(E1000_TCTL);
5107         DUMPREG(E1000_TCTL_EXT);
5108         DUMPREG(E1000_TIPG);
5109         DUMPREG(E1000_TBT);
5110         DUMPREG(E1000_AIT);
5111         DUMPREG(E1000_LEDCTL);
5112         DUMPREG(E1000_EXTCNF_CTRL);
5113         DUMPREG(E1000_EXTCNF_SIZE);
5114         DUMPREG(E1000_PHY_CTRL);
5115         DUMPREG(E1000_PBA);
5116         DUMPREG(E1000_PBS);
5117         DUMPREG(E1000_PBECCSTS);
5118         DUMPREG(E1000_EEMNGCTL);
5119         DUMPREG(E1000_EEARBC);
5120         DUMPREG(E1000_FLASHT);
5121         DUMPREG(E1000_EEARBC_I210);
5122         DUMPREG(E1000_EEWR);
5123         DUMPREG(E1000_FLSWCTL);
5124         DUMPREG(E1000_FLSWDATA);
5125         DUMPREG(E1000_FLSWCNT);
5126         DUMPREG(E1000_FLOP);
5127         DUMPREG(E1000_I2CCMD);
5128         DUMPREG(E1000_I2CPARAMS);
5129         DUMPREG(E1000_WDSTP);
5130         DUMPREG(E1000_SWDSTS);
5131         DUMPREG(E1000_FRTIMER);
5132         DUMPREG(E1000_TCPTIMER);
5133         DUMPREG(E1000_VPDDIAG);
5134         DUMPREG(E1000_IMS_V2);
5135         DUMPREG(E1000_IAM_V2);
5136         DUMPREG(E1000_ERT);
5137         DUMPREG(E1000_FCRTL);
5138         DUMPREG(E1000_FCRTH);
5139         DUMPREG(E1000_PSRCTL);
5140         DUMPREG(E1000_RDFH);
5141         DUMPREG(E1000_RDFT);
5142         DUMPREG(E1000_RDFHS);
5143         DUMPREG(E1000_RDFTS);
5144         DUMPREG(E1000_RDFPC);
5145         DUMPREG(E1000_PBRTH);
5146         DUMPREG(E1000_FCRTV);
5147         DUMPREG(E1000_RDPUMB);
5148         DUMPREG(E1000_RDPUAD);
5149         DUMPREG(E1000_RDPUWD);
5150         DUMPREG(E1000_RDPURD);
5151         DUMPREG(E1000_RDPUCTL);
5152         DUMPREG(E1000_PBDIAG);
5153         DUMPREG(E1000_RXPBS);
5154         DUMPREG(E1000_IRPBS);
5155         DUMPREG(E1000_PBRWAC);
5156         DUMPREG(E1000_RDTR);
5157         DUMPREG(E1000_RADV);
5158         DUMPREG(E1000_SRWR);
5159         DUMPREG(E1000_I210_FLMNGCTL);
5160         DUMPREG(E1000_I210_FLMNGDATA);
5161         DUMPREG(E1000_I210_FLMNGCNT);
5162         DUMPREG(E1000_I210_FLSWCTL);
5163         DUMPREG(E1000_I210_FLSWDATA);
5164         DUMPREG(E1000_I210_FLSWCNT);
5165         DUMPREG(E1000_I210_FLA);
5166         DUMPREG(E1000_INVM_SIZE);
5167         DUMPREG(E1000_I210_TQAVCTRL);
5168         DUMPREG(E1000_RSRPD);
5169         DUMPREG(E1000_RAID);
5170         DUMPREG(E1000_TXDMAC);
5171         DUMPREG(E1000_KABGTXD);
5172         DUMPREG(E1000_PBSLAC);
5173         DUMPREG(E1000_TXPBS);
5174         DUMPREG(E1000_ITPBS);
5175         DUMPREG(E1000_TDFH);
5176         DUMPREG(E1000_TDFT);
5177         DUMPREG(E1000_TDFHS);
5178         DUMPREG(E1000_TDFTS);
5179         DUMPREG(E1000_TDFPC);
5180         DUMPREG(E1000_TDPUMB);
5181         DUMPREG(E1000_TDPUAD);
5182         DUMPREG(E1000_TDPUWD);
5183         DUMPREG(E1000_TDPURD);
5184         DUMPREG(E1000_TDPUCTL);
5185         DUMPREG(E1000_DTXCTL);
5186         DUMPREG(E1000_DTXTCPFLGL);
5187         DUMPREG(E1000_DTXTCPFLGH);
5188         DUMPREG(E1000_DTXMXSZRQ);
5189         DUMPREG(E1000_TIDV);
5190         DUMPREG(E1000_TADV);
5191         DUMPREG(E1000_TSPMT);
5192         DUMPREG(E1000_VFGPRC);
5193         DUMPREG(E1000_VFGORC);
5194         DUMPREG(E1000_VFMPRC);
5195         DUMPREG(E1000_VFGPTC);
5196         DUMPREG(E1000_VFGOTC);
5197         DUMPREG(E1000_VFGOTLBC);
5198         DUMPREG(E1000_VFGPTLBC);
5199         DUMPREG(E1000_VFGORLBC);
5200         DUMPREG(E1000_VFGPRLBC);
5201         DUMPREG(E1000_LSECTXCAP);
5202         DUMPREG(E1000_LSECRXCAP);
5203         DUMPREG(E1000_LSECTXCTRL);
5204         DUMPREG(E1000_LSECRXCTRL);
5205         DUMPREG(E1000_LSECTXSCL);
5206         DUMPREG(E1000_LSECTXSCH);
5207         DUMPREG(E1000_LSECTXSA);
5208         DUMPREG(E1000_LSECTXPN0);
5209         DUMPREG(E1000_LSECTXPN1);
5210         DUMPREG(E1000_LSECRXSCL);
5211         DUMPREG(E1000_LSECRXSCH);
5212         DUMPREG(E1000_IPSCTRL);
5213         DUMPREG(E1000_IPSRXCMD);
5214         DUMPREG(E1000_IPSRXIDX);
5215         DUMPREG(E1000_IPSRXSALT);
5216         DUMPREG(E1000_IPSRXSPI);
5217         DUMPREG(E1000_IPSTXSALT);
5218         DUMPREG(E1000_IPSTXIDX);
5219         DUMPREG(E1000_PCS_CFG0);
5220         DUMPREG(E1000_PCS_LCTL);
5221         DUMPREG(E1000_PCS_LSTAT);
5222         DUMPREG(E1000_PCS_ANADV);
5223         DUMPREG(E1000_PCS_LPAB);
5224         DUMPREG(E1000_PCS_NPTX);
5225         DUMPREG(E1000_PCS_LPABNP);
5226         DUMPREG(E1000_RXCSUM);
5227         DUMPREG(E1000_RLPML);
5228         DUMPREG(E1000_RFCTL);
5229         DUMPREG(E1000_MTA);
5230         DUMPREG(E1000_RA);
5231         DUMPREG(E1000_RA2);
5232         DUMPREG(E1000_VFTA);
5233         DUMPREG(E1000_VT_CTL);
5234         DUMPREG(E1000_CIAA);
5235         DUMPREG(E1000_CIAD);
5236         DUMPREG(E1000_VFQA0);
5237         DUMPREG(E1000_VFQA1);
5238         DUMPREG(E1000_WUC);
5239         DUMPREG(E1000_WUFC);
5240         DUMPREG(E1000_WUS);
5241         DUMPREG(E1000_MANC);
5242         DUMPREG(E1000_IPAV);
5243         DUMPREG(E1000_IP4AT);
5244         DUMPREG(E1000_IP6AT);
5245         DUMPREG(E1000_WUPL);
5246         DUMPREG(E1000_WUPM);
5247         DUMPREG(E1000_PBACL);
5248         DUMPREG(E1000_FFLT);
5249         DUMPREG(E1000_HOST_IF);
5250         DUMPREG(E1000_HIBBA);
5251         DUMPREG(E1000_KMRNCTRLSTA);
5252         DUMPREG(E1000_MANC2H);
5253         DUMPREG(E1000_CCMCTL);
5254         DUMPREG(E1000_GIOCTL);
5255         DUMPREG(E1000_SCCTL);
5256
5257 #define E1000_WCS       0x558C
5258         DUMPREG(E1000_WCS);
5259 #define E1000_GCR_EXT   0x586C
5260         DUMPREG(E1000_GCR_EXT);
5261         DUMPREG(E1000_GCR);
5262         DUMPREG(E1000_GCR2);
5263         DUMPREG(E1000_FACTPS);
5264         DUMPREG(E1000_DCA_ID);
5265         DUMPREG(E1000_DCA_CTRL);
5266         DUMPREG(E1000_UFUSE);
5267         DUMPREG(E1000_FFLT_DBG);
5268         DUMPREG(E1000_HICR);
5269         DUMPREG(E1000_FWSTS);
5270         DUMPREG(E1000_CPUVEC);
5271         DUMPREG(E1000_MRQC);
5272         DUMPREG(E1000_SWPBS);
5273         DUMPREG(E1000_MBVFICR);
5274         DUMPREG(E1000_MBVFIMR);
5275         DUMPREG(E1000_VFLRE);
5276         DUMPREG(E1000_VFRE);
5277         DUMPREG(E1000_VFTE);
5278         DUMPREG(E1000_QDE);
5279         DUMPREG(E1000_DTXSWC);
5280         DUMPREG(E1000_WVBR);
5281         DUMPREG(E1000_RPLOLR);
5282         DUMPREG(E1000_UTA);
5283         DUMPREG(E1000_IOVTCL);
5284         DUMPREG(E1000_VMRCTL);
5285         DUMPREG(E1000_VMRVLAN);
5286         DUMPREG(E1000_VMRVM);
5287         DUMPREG(E1000_LVMMC);
5288         DUMPREG(E1000_TXSWC);
5289         DUMPREG(E1000_SCCRL);
5290         DUMPREG(E1000_BSCTRH);
5291         DUMPREG(E1000_MSCTRH);
5292         DUMPREG(E1000_RXSTMPL);
5293         DUMPREG(E1000_RXSTMPH);
5294         DUMPREG(E1000_RXSATRL);
5295         DUMPREG(E1000_RXSATRH);
5296         DUMPREG(E1000_TXSTMPL);
5297         DUMPREG(E1000_TXSTMPH);
5298         DUMPREG(E1000_TIMINCA);
5299         DUMPREG(E1000_TIMADJL);
5300         DUMPREG(E1000_TIMADJH);
5301         DUMPREG(E1000_TSAUXC);
5302         DUMPREG(E1000_SYSSTMPL);
5303         DUMPREG(E1000_SYSSTMPH);
5304         DUMPREG(E1000_PLTSTMPL);
5305         DUMPREG(E1000_PLTSTMPH);
5306         DUMPREG(E1000_RXMTRL);
5307         DUMPREG(E1000_RXUDP);
5308         DUMPREG(E1000_SYSTIMR);
5309         DUMPREG(E1000_TSICR);
5310         DUMPREG(E1000_TSIM);
5311         DUMPREG(E1000_DMACR);
5312         DUMPREG(E1000_DMCTXTH);
5313         DUMPREG(E1000_DMCTLX);
5314         DUMPREG(E1000_DMCRTRH);
5315         DUMPREG(E1000_DMCCNT);
5316         DUMPREG(E1000_FCRTC);
5317         DUMPREG(E1000_PCIEMISC);
5318         DUMPREG(E1000_PCIEERRSTS);
5319         DUMPREG(E1000_IPCNFG);
5320         DUMPREG(E1000_LTRC);
5321         DUMPREG(E1000_EEER);
5322         DUMPREG(E1000_EEE_SU);
5323         DUMPREG(E1000_TLPIC);
5324         DUMPREG(E1000_RLPIC);
5325         if (++col != 1)
5326                 kprintf("\n");
5327         kprintf("\n");
5328 }
5329
5330 static int
5331 igb_sysctl_reg_dump(SYSCTL_HANDLER_ARGS)
5332 {
5333         struct igb_softc *sc = (void *)arg1;
5334         struct ifnet *ifp = &sc->arpcom.ac_if;
5335         int error, dump = 0;
5336
5337         error = sysctl_handle_int(oidp, &dump, 0, req);
5338         if (error || req->newptr == NULL)
5339                 return error;
5340         if (dump <= 0)
5341                 return EINVAL;
5342
5343         ifnet_serialize_all(ifp);
5344         igb_reg_dump(sc);
5345         ifnet_deserialize_all(ifp);
5346
5347         return error;
5348 }