sys/dev/netif/igb/if_igb.c

   1 /*
   2  * Copyright (c) 2001-2011, 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/toeplitz.h>
  59 #include <net/toeplitz2.h>
  60 #include <net/vlan/if_vlan_var.h>
  61 #include <net/vlan/if_vlan_ether.h>
  62 #include <net/if_poll.h>
  63
  64 #include <netinet/in_systm.h>
  65 #include <netinet/in.h>
  66 #include <netinet/ip.h>
  67 #include <netinet/tcp.h>
  68 #include <netinet/udp.h>
  69
  70 #include <bus/pci/pcivar.h>
  71 #include <bus/pci/pcireg.h>
  72
  73 #include <dev/netif/ig_hal/e1000_api.h>
  74 #include <dev/netif/ig_hal/e1000_82575.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
 125         /* required last entry */
 126         IGB_DEVICE_NULL
 127 };
 128
 129 static int      igb_probe(device_t);
 130 static int      igb_attach(device_t);
 131 static int      igb_detach(device_t);
 132 static int      igb_shutdown(device_t);
 133 static int      igb_suspend(device_t);
 134 static int      igb_resume(device_t);
 135
 136 static boolean_t igb_is_valid_ether_addr(const uint8_t *);
 137 static void     igb_setup_ifp(struct igb_softc *);
 138 static boolean_t igb_txcsum_ctx(struct igb_tx_ring *, struct mbuf *);
 139 static int      igb_tso_pullup(struct igb_tx_ring *, struct mbuf **);
 140 static void     igb_tso_ctx(struct igb_tx_ring *, struct mbuf *, uint32_t *);
 141 static void     igb_add_sysctl(struct igb_softc *);
 142 static int      igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS);
 143 static int      igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS);
 144 static int      igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS);
 145 static void     igb_set_ring_inuse(struct igb_softc *, boolean_t);
 146 #ifdef IFPOLL_ENABLE
 147 static int      igb_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS);
 148 static int      igb_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS);
 149 #endif
 150
 151 static void     igb_vf_init_stats(struct igb_softc *);
 152 static void     igb_reset(struct igb_softc *);
 153 static void     igb_update_stats_counters(struct igb_softc *);
 154 static void     igb_update_vf_stats_counters(struct igb_softc *);
 155 static void     igb_update_link_status(struct igb_softc *);
 156 static void     igb_init_tx_unit(struct igb_softc *);
 157 static void     igb_init_rx_unit(struct igb_softc *);
 158
 159 static void     igb_set_vlan(struct igb_softc *);
 160 static void     igb_set_multi(struct igb_softc *);
 161 static void     igb_set_promisc(struct igb_softc *);
 162 static void     igb_disable_promisc(struct igb_softc *);
 163
 164 static int      igb_alloc_rings(struct igb_softc *);
 165 static void     igb_free_rings(struct igb_softc *);
 166 static int      igb_create_tx_ring(struct igb_tx_ring *);
 167 static int      igb_create_rx_ring(struct igb_rx_ring *);
 168 static void     igb_free_tx_ring(struct igb_tx_ring *);
 169 static void     igb_free_rx_ring(struct igb_rx_ring *);
 170 static void     igb_destroy_tx_ring(struct igb_tx_ring *, int);
 171 static void     igb_destroy_rx_ring(struct igb_rx_ring *, int);
 172 static void     igb_init_tx_ring(struct igb_tx_ring *);
 173 static int      igb_init_rx_ring(struct igb_rx_ring *);
 174 static int      igb_newbuf(struct igb_rx_ring *, int, boolean_t);
 175 static int      igb_encap(struct igb_tx_ring *, struct mbuf **, int *, int *);
 176 static void     igb_rx_refresh(struct igb_rx_ring *, int);
 177
 178 static void     igb_stop(struct igb_softc *);
 179 static void     igb_init(void *);
 180 static int      igb_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 181 static void     igb_media_status(struct ifnet *, struct ifmediareq *);
 182 static int      igb_media_change(struct ifnet *);
 183 static void     igb_timer(void *);
 184 static void     igb_watchdog(struct ifnet *);
 185 static void     igb_start(struct ifnet *, struct ifaltq_subque *);
 186 #ifdef IFPOLL_ENABLE
 187 static void     igb_npoll(struct ifnet *, struct ifpoll_info *);
 188 static void     igb_npoll_rx(struct ifnet *, void *, int);
 189 static void     igb_npoll_tx(struct ifnet *, void *, int);
 190 static void     igb_npoll_status(struct ifnet *);
 191 #endif
 192 static void     igb_serialize(struct ifnet *, enum ifnet_serialize);
 193 static void     igb_deserialize(struct ifnet *, enum ifnet_serialize);
 194 static int      igb_tryserialize(struct ifnet *, enum ifnet_serialize);
 195 #ifdef INVARIANTS
 196 static void     igb_serialize_assert(struct ifnet *, enum ifnet_serialize,
 197                     boolean_t);
 198 #endif
 199
 200 static void     igb_intr(void *);
 201 static void     igb_intr_shared(void *);
 202 static void     igb_rxeof(struct igb_rx_ring *, int);
 203 static void     igb_txeof(struct igb_tx_ring *);
 204 static void     igb_set_eitr(struct igb_softc *, int, int);
 205 static void     igb_enable_intr(struct igb_softc *);
 206 static void     igb_disable_intr(struct igb_softc *);
 207 static void     igb_init_unshared_intr(struct igb_softc *);
 208 static void     igb_init_intr(struct igb_softc *);
 209 static int      igb_setup_intr(struct igb_softc *);
 210 static void     igb_set_txintr_mask(struct igb_tx_ring *, int *, int);
 211 static void     igb_set_rxintr_mask(struct igb_rx_ring *, int *, int);
 212 static void     igb_set_intr_mask(struct igb_softc *);
 213 static int      igb_alloc_intr(struct igb_softc *);
 214 static void     igb_free_intr(struct igb_softc *);
 215 static void     igb_teardown_intr(struct igb_softc *);
 216 static void     igb_msix_try_alloc(struct igb_softc *);
 217 static void     igb_msix_free(struct igb_softc *, boolean_t);
 218 static int      igb_msix_setup(struct igb_softc *);
 219 static void     igb_msix_teardown(struct igb_softc *, int);
 220 static void     igb_msix_rx(void *);
 221 static void     igb_msix_tx(void *);
 222 static void     igb_msix_status(void *);
 223
 224 /* Management and WOL Support */
 225 static void     igb_get_mgmt(struct igb_softc *);
 226 static void     igb_rel_mgmt(struct igb_softc *);
 227 static void     igb_get_hw_control(struct igb_softc *);
 228 static void     igb_rel_hw_control(struct igb_softc *);
 229 static void     igb_enable_wol(device_t);
 230
 231 static device_method_t igb_methods[] = {
 232         /* Device interface */
 233         DEVMETHOD(device_probe,         igb_probe),
 234         DEVMETHOD(device_attach,        igb_attach),
 235         DEVMETHOD(device_detach,        igb_detach),
 236         DEVMETHOD(device_shutdown,      igb_shutdown),
 237         DEVMETHOD(device_suspend,       igb_suspend),
 238         DEVMETHOD(device_resume,        igb_resume),
 239         { 0, 0 }
 240 };
 241
 242 static driver_t igb_driver = {
 243         "igb",
 244         igb_methods,
 245         sizeof(struct igb_softc),
 246 };
 247
 248 static devclass_t igb_devclass;
 249
 250 DECLARE_DUMMY_MODULE(if_igb);
 251 MODULE_DEPEND(igb, ig_hal, 1, 1, 1);
 252 DRIVER_MODULE(if_igb, pci, igb_driver, igb_devclass, NULL, NULL);
 253
 254 static int      igb_rxd = IGB_DEFAULT_RXD;
 255 static int      igb_txd = IGB_DEFAULT_TXD;
 256 static int      igb_rxr = 0;
 257 static int      igb_msi_enable = 1;
 258 static int      igb_msix_enable = 1;
 259 static int      igb_eee_disabled = 1;   /* Energy Efficient Ethernet */
 260 static int      igb_fc_setting = e1000_fc_full;
 261
 262 /*
 263  * DMA Coalescing, only for i350 - default to off,
 264  * this feature is for power savings
 265  */
 266 static int      igb_dma_coalesce = 0;
 267
 268 TUNABLE_INT("hw.igb.rxd", &igb_rxd);
 269 TUNABLE_INT("hw.igb.txd", &igb_txd);
 270 TUNABLE_INT("hw.igb.rxr", &igb_rxr);
 271 TUNABLE_INT("hw.igb.msi.enable", &igb_msi_enable);
 272 TUNABLE_INT("hw.igb.msix.enable", &igb_msix_enable);
 273 TUNABLE_INT("hw.igb.fc_setting", &igb_fc_setting);
 274
 275 /* i350 specific */
 276 TUNABLE_INT("hw.igb.eee_disabled", &igb_eee_disabled);
 277 TUNABLE_INT("hw.igb.dma_coalesce", &igb_dma_coalesce);
 278
 279 static __inline void
 280 igb_rxcsum(uint32_t staterr, struct mbuf *mp)
 281 {
 282         /* Ignore Checksum bit is set */
 283         if (staterr & E1000_RXD_STAT_IXSM)
 284                 return;
 285
 286         if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
 287             E1000_RXD_STAT_IPCS)
 288                 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
 289
 290         if (staterr & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
 291                 if ((staterr & E1000_RXDEXT_STATERR_TCPE) == 0) {
 292                         mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
 293                             CSUM_PSEUDO_HDR | CSUM_FRAG_NOT_CHECKED;
 294                         mp->m_pkthdr.csum_data = htons(0xffff);
 295                 }
 296         }
 297 }
 298
 299 static __inline struct pktinfo *
 300 igb_rssinfo(struct mbuf *m, struct pktinfo *pi,
 301     uint32_t hash, uint32_t hashtype, uint32_t staterr)
 302 {
 303         switch (hashtype) {
 304         case E1000_RXDADV_RSSTYPE_IPV4_TCP:
 305                 pi->pi_netisr = NETISR_IP;
 306                 pi->pi_flags = 0;
 307                 pi->pi_l3proto = IPPROTO_TCP;
 308                 break;
 309
 310         case E1000_RXDADV_RSSTYPE_IPV4:
 311                 if (staterr & E1000_RXD_STAT_IXSM)
 312                         return NULL;
 313
 314                 if ((staterr &
 315                      (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
 316                     E1000_RXD_STAT_TCPCS) {
 317                         pi->pi_netisr = NETISR_IP;
 318                         pi->pi_flags = 0;
 319                         pi->pi_l3proto = IPPROTO_UDP;
 320                         break;
 321                 }
 322                 /* FALL THROUGH */
 323         default:
 324                 return NULL;
 325         }
 326
 327         m->m_flags |= M_HASH;
 328         m->m_pkthdr.hash = toeplitz_hash(hash);
 329         return pi;
 330 }
 331
 332 static int
 333 igb_probe(device_t dev)
 334 {
 335         const struct igb_device *d;
 336         uint16_t vid, did;
 337
 338         vid = pci_get_vendor(dev);
 339         did = pci_get_device(dev);
 340
 341         for (d = igb_devices; d->desc != NULL; ++d) {
 342                 if (vid == d->vid && did == d->did) {
 343                         device_set_desc(dev, d->desc);
 344                         return 0;
 345                 }
 346         }
 347         return ENXIO;
 348 }
 349
 350 static int
 351 igb_attach(device_t dev)
 352 {
 353         struct igb_softc *sc = device_get_softc(dev);
 354         uint16_t eeprom_data;
 355         int error = 0, i, j, ring_max;
 356 #ifdef IFPOLL_ENABLE
 357         int offset, offset_def;
 358 #endif
 359
 360 #ifdef notyet
 361         /* SYSCTL stuff */
 362         SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
 363             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
 364             OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
 365             igb_sysctl_nvm_info, "I", "NVM Information");
 366         SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
 367             SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
 368             OID_AUTO, "flow_control", CTLTYPE_INT|CTLFLAG_RW,
 369             adapter, 0, igb_set_flowcntl, "I", "Flow Control");
 370 #endif
 371
 372         callout_init_mp(&sc->timer);
 373         lwkt_serialize_init(&sc->main_serialize);
 374
 375         if_initname(&sc->arpcom.ac_if, device_get_name(dev),
 376             device_get_unit(dev));
 377         sc->dev = sc->osdep.dev = dev;
 378
 379         /*
 380          * Determine hardware and mac type
 381          */
 382         sc->hw.vendor_id = pci_get_vendor(dev);
 383         sc->hw.device_id = pci_get_device(dev);
 384         sc->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
 385         sc->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2);
 386         sc->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2);
 387
 388         if (e1000_set_mac_type(&sc->hw))
 389                 return ENXIO;
 390
 391         /* Are we a VF device? */
 392         if (sc->hw.mac.type == e1000_vfadapt ||
 393             sc->hw.mac.type == e1000_vfadapt_i350)
 394                 sc->vf_ifp = 1;
 395         else
 396                 sc->vf_ifp = 0;
 397
 398         /*
 399          * Configure total supported RX/TX ring count
 400          */
 401         switch (sc->hw.mac.type) {
 402         case e1000_82575:
 403                 ring_max = IGB_MAX_RING_82575;
 404                 break;
 405         case e1000_82580:
 406                 ring_max = IGB_MAX_RING_82580;
 407                 break;
 408         case e1000_i350:
 409                 ring_max = IGB_MAX_RING_I350;
 410                 break;
 411         case e1000_82576:
 412                 ring_max = IGB_MAX_RING_82576;
 413                 break;
 414         default:
 415                 ring_max = IGB_MIN_RING;
 416                 break;
 417         }
 418         sc->rx_ring_cnt = device_getenv_int(dev, "rxr", igb_rxr);
 419         sc->rx_ring_cnt = if_ring_count2(sc->rx_ring_cnt, ring_max);
 420 #ifdef IGB_RSS_DEBUG
 421         sc->rx_ring_cnt = device_getenv_int(dev, "rxr_debug", sc->rx_ring_cnt);
 422 #endif
 423         sc->rx_ring_inuse = sc->rx_ring_cnt;
 424         sc->tx_ring_cnt = 1; /* XXX */
 425
 426         if (sc->hw.mac.type == e1000_82575)
 427                 sc->flags |= IGB_FLAG_TSO_IPLEN0;
 428
 429         /* Enable bus mastering */
 430         pci_enable_busmaster(dev);
 431
 432         /*
 433          * Allocate IO memory
 434          */
 435         sc->mem_rid = PCIR_BAR(0);
 436         sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
 437             RF_ACTIVE);
 438         if (sc->mem_res == NULL) {
 439                 device_printf(dev, "Unable to allocate bus resource: memory\n");
 440                 error = ENXIO;
 441                 goto failed;
 442         }
 443         sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->mem_res);
 444         sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->mem_res);
 445
 446         sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
 447
 448         /* Save PCI command register for Shared Code */
 449         sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
 450         sc->hw.back = &sc->osdep;
 451
 452         /* Do Shared Code initialization */
 453         if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
 454                 device_printf(dev, "Setup of Shared code failed\n");
 455                 error = ENXIO;
 456                 goto failed;
 457         }
 458
 459         e1000_get_bus_info(&sc->hw);
 460
 461         sc->hw.mac.autoneg = DO_AUTO_NEG;
 462         sc->hw.phy.autoneg_wait_to_complete = FALSE;
 463         sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
 464
 465         /* Copper options */
 466         if (sc->hw.phy.media_type == e1000_media_type_copper) {
 467                 sc->hw.phy.mdix = AUTO_ALL_MODES;
 468                 sc->hw.phy.disable_polarity_correction = FALSE;
 469                 sc->hw.phy.ms_type = IGB_MASTER_SLAVE;
 470         }
 471
 472         /* Set the frame limits assuming  standard ethernet sized frames. */
 473         sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
 474
 475         /* Allocate RX/TX rings */
 476         error = igb_alloc_rings(sc);
 477         if (error)
 478                 goto failed;
 479
 480 #ifdef IFPOLL_ENABLE
 481         /*
 482          * NPOLLING RX CPU offset
 483          */
 484         if (sc->rx_ring_cnt == ncpus2) {
 485                 offset = 0;
 486         } else {
 487                 offset_def = (sc->rx_ring_cnt * device_get_unit(dev)) % ncpus2;
 488                 offset = device_getenv_int(dev, "npoll.rxoff", offset_def);
 489                 if (offset >= ncpus2 ||
 490                     offset % sc->rx_ring_cnt != 0) {
 491                         device_printf(dev, "invalid npoll.rxoff %d, use %d\n",
 492                             offset, offset_def);
 493                         offset = offset_def;
 494                 }
 495         }
 496         sc->rx_npoll_off = offset;
 497
 498         /*
 499          * NPOLLING TX CPU offset
 500          */
 501         offset_def = sc->rx_npoll_off;
 502         offset = device_getenv_int(dev, "npoll.txoff", offset_def);
 503         if (offset >= ncpus2) {
 504                 device_printf(dev, "invalid npoll.txoff %d, use %d\n",
 505                     offset, offset_def);
 506                 offset = offset_def;
 507         }
 508         sc->tx_npoll_off = offset;
 509 #endif
 510
 511         /* Allocate interrupt */
 512         error = igb_alloc_intr(sc);
 513         if (error)
 514                 goto failed;
 515
 516         /*
 517          * Setup serializers
 518          */
 519         i = 0;
 520         sc->serializes[i++] = &sc->main_serialize;
 521
 522         sc->tx_serialize = i;
 523         for (j = 0; j < sc->tx_ring_cnt; ++j)
 524                 sc->serializes[i++] = &sc->tx_rings[j].tx_serialize;
 525
 526         sc->rx_serialize = i;
 527         for (j = 0; j < sc->rx_ring_cnt; ++j)
 528                 sc->serializes[i++] = &sc->rx_rings[j].rx_serialize;
 529
 530         sc->serialize_cnt = i;
 531         KKASSERT(sc->serialize_cnt <= IGB_NSERIALIZE);
 532
 533         /* Allocate the appropriate stats memory */
 534         if (sc->vf_ifp) {
 535                 sc->stats = kmalloc(sizeof(struct e1000_vf_stats), M_DEVBUF,
 536                     M_WAITOK | M_ZERO);
 537                 igb_vf_init_stats(sc);
 538         } else {
 539                 sc->stats = kmalloc(sizeof(struct e1000_hw_stats), M_DEVBUF,
 540                     M_WAITOK | M_ZERO);
 541         }
 542
 543         /* Allocate multicast array memory. */
 544         sc->mta = kmalloc(ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES,
 545             M_DEVBUF, M_WAITOK);
 546
 547         /* Some adapter-specific advanced features */
 548         if (sc->hw.mac.type >= e1000_i350) {
 549 #ifdef notyet
 550                 igb_set_sysctl_value(adapter, "dma_coalesce",
 551                     "configure dma coalesce",
 552                     &adapter->dma_coalesce, igb_dma_coalesce);
 553                 igb_set_sysctl_value(adapter, "eee_disabled",
 554                     "enable Energy Efficient Ethernet",
 555                     &adapter->hw.dev_spec._82575.eee_disable,
 556                     igb_eee_disabled);
 557 #else
 558                 sc->dma_coalesce = igb_dma_coalesce;
 559                 sc->hw.dev_spec._82575.eee_disable = igb_eee_disabled;
 560 #endif
 561                 e1000_set_eee_i350(&sc->hw);
 562         }
 563
 564         /*
 565          * Start from a known state, this is important in reading the nvm and
 566          * mac from that.
 567          */
 568         e1000_reset_hw(&sc->hw);
 569
 570         /* Make sure we have a good EEPROM before we read from it */
 571         if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
 572                 /*
 573                  * Some PCI-E parts fail the first check due to
 574                  * the link being in sleep state, call it again,
 575                  * if it fails a second time its a real issue.
 576                  */
 577                 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
 578                         device_printf(dev,
 579                             "The EEPROM Checksum Is Not Valid\n");
 580                         error = EIO;
 581                         goto failed;
 582                 }
 583         }
 584
 585         /* Copy the permanent MAC address out of the EEPROM */
 586         if (e1000_read_mac_addr(&sc->hw) < 0) {
 587                 device_printf(dev, "EEPROM read error while reading MAC"
 588                     " address\n");
 589                 error = EIO;
 590                 goto failed;
 591         }
 592         if (!igb_is_valid_ether_addr(sc->hw.mac.addr)) {
 593                 device_printf(dev, "Invalid MAC address\n");
 594                 error = EIO;
 595                 goto failed;
 596         }
 597
 598         /* Setup OS specific network interface */
 599         igb_setup_ifp(sc);
 600
 601         /* Add sysctl tree, must after igb_setup_ifp() */
 602         igb_add_sysctl(sc);
 603
 604         /* Now get a good starting state */
 605         igb_reset(sc);
 606
 607         /* Initialize statistics */
 608         igb_update_stats_counters(sc);
 609
 610         sc->hw.mac.get_link_status = 1;
 611         igb_update_link_status(sc);
 612
 613         /* Indicate SOL/IDER usage */
 614         if (e1000_check_reset_block(&sc->hw)) {
 615                 device_printf(dev,
 616                     "PHY reset is blocked due to SOL/IDER session.\n");
 617         }
 618
 619         /* Determine if we have to control management hardware */
 620         if (e1000_enable_mng_pass_thru(&sc->hw))
 621                 sc->flags |= IGB_FLAG_HAS_MGMT;
 622
 623         /*
 624          * Setup Wake-on-Lan
 625          */
 626         /* APME bit in EEPROM is mapped to WUC.APME */
 627         eeprom_data = E1000_READ_REG(&sc->hw, E1000_WUC) & E1000_WUC_APME;
 628         if (eeprom_data)
 629                 sc->wol = E1000_WUFC_MAG;
 630         /* XXX disable WOL */
 631         sc->wol = 0;
 632
 633 #ifdef notyet
 634         /* Register for VLAN events */
 635         adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
 636              igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
 637         adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
 638              igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
 639 #endif
 640
 641 #ifdef notyet
 642         igb_add_hw_stats(adapter);
 643 #endif
 644
 645         error = igb_setup_intr(sc);
 646         if (error) {
 647                 ether_ifdetach(&sc->arpcom.ac_if);
 648                 goto failed;
 649         }
 650
 651         for (i = 0; i < sc->tx_ring_cnt; ++i) {
 652                 struct ifaltq_subque *ifsq =
 653                     ifq_get_subq(&sc->arpcom.ac_if.if_snd, i);
 654                 struct igb_tx_ring *txr = &sc->tx_rings[i];
 655
 656                 ifsq_set_cpuid(ifsq, txr->tx_intr_cpuid);
 657                 ifsq_set_priv(ifsq, txr);
 658                 txr->ifsq = ifsq;
 659         }
 660
 661         return 0;
 662
 663 failed:
 664         igb_detach(dev);
 665         return error;
 666 }
 667
 668 static int
 669 igb_detach(device_t dev)
 670 {
 671         struct igb_softc *sc = device_get_softc(dev);
 672
 673         if (device_is_attached(dev)) {
 674                 struct ifnet *ifp = &sc->arpcom.ac_if;
 675
 676                 ifnet_serialize_all(ifp);
 677
 678                 igb_stop(sc);
 679
 680                 e1000_phy_hw_reset(&sc->hw);
 681
 682                 /* Give control back to firmware */
 683                 igb_rel_mgmt(sc);
 684                 igb_rel_hw_control(sc);
 685
 686                 if (sc->wol) {
 687                         E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
 688                         E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
 689                         igb_enable_wol(dev);
 690                 }
 691
 692                 igb_teardown_intr(sc);
 693
 694                 ifnet_deserialize_all(ifp);
 695
 696                 ether_ifdetach(ifp);
 697         } else if (sc->mem_res != NULL) {
 698                 igb_rel_hw_control(sc);
 699         }
 700         bus_generic_detach(dev);
 701
 702         if (sc->sysctl_tree != NULL)
 703                 sysctl_ctx_free(&sc->sysctl_ctx);
 704
 705         igb_free_intr(sc);
 706
 707         if (sc->msix_mem_res != NULL) {
 708                 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_mem_rid,
 709                     sc->msix_mem_res);
 710         }
 711         if (sc->mem_res != NULL) {
 712                 bus_release_resource(dev, SYS_RES_MEMORY, sc->mem_rid,
 713                     sc->mem_res);
 714         }
 715
 716         igb_free_rings(sc);
 717
 718         if (sc->mta != NULL)
 719                 kfree(sc->mta, M_DEVBUF);
 720         if (sc->stats != NULL)
 721                 kfree(sc->stats, M_DEVBUF);
 722
 723         return 0;
 724 }
 725
 726 static int
 727 igb_shutdown(device_t dev)
 728 {
 729         return igb_suspend(dev);
 730 }
 731
 732 static int
 733 igb_suspend(device_t dev)
 734 {
 735         struct igb_softc *sc = device_get_softc(dev);
 736         struct ifnet *ifp = &sc->arpcom.ac_if;
 737
 738         ifnet_serialize_all(ifp);
 739
 740         igb_stop(sc);
 741
 742         igb_rel_mgmt(sc);
 743         igb_rel_hw_control(sc);
 744
 745         if (sc->wol) {
 746                 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
 747                 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
 748                 igb_enable_wol(dev);
 749         }
 750
 751         ifnet_deserialize_all(ifp);
 752
 753         return bus_generic_suspend(dev);
 754 }
 755
 756 static int
 757 igb_resume(device_t dev)
 758 {
 759         struct igb_softc *sc = device_get_softc(dev);
 760         struct ifnet *ifp = &sc->arpcom.ac_if;
 761         int i;
 762
 763         ifnet_serialize_all(ifp);
 764
 765         igb_init(sc);
 766         igb_get_mgmt(sc);
 767
 768         for (i = 0; i < sc->tx_ring_cnt; ++i)
 769                 ifsq_devstart(sc->tx_rings[i].ifsq);
 770
 771         ifnet_deserialize_all(ifp);
 772
 773         return bus_generic_resume(dev);
 774 }
 775
 776 static int
 777 igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
 778 {
 779         struct igb_softc *sc = ifp->if_softc;
 780         struct ifreq *ifr = (struct ifreq *)data;
 781         int max_frame_size, mask, reinit;
 782         int error = 0;
 783
 784         ASSERT_IFNET_SERIALIZED_ALL(ifp);
 785
 786         switch (command) {
 787         case SIOCSIFMTU:
 788                 max_frame_size = 9234;
 789                 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
 790                     ETHER_CRC_LEN) {
 791                         error = EINVAL;
 792                         break;
 793                 }
 794
 795                 ifp->if_mtu = ifr->ifr_mtu;
 796                 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
 797                     ETHER_CRC_LEN;
 798
 799                 if (ifp->if_flags & IFF_RUNNING)
 800                         igb_init(sc);
 801                 break;
 802
 803         case SIOCSIFFLAGS:
 804                 if (ifp->if_flags & IFF_UP) {
 805                         if (ifp->if_flags & IFF_RUNNING) {
 806                                 if ((ifp->if_flags ^ sc->if_flags) &
 807                                     (IFF_PROMISC | IFF_ALLMULTI)) {
 808                                         igb_disable_promisc(sc);
 809                                         igb_set_promisc(sc);
 810                                 }
 811                         } else {
 812                                 igb_init(sc);
 813                         }
 814                 } else if (ifp->if_flags & IFF_RUNNING) {
 815                         igb_stop(sc);
 816                 }
 817                 sc->if_flags = ifp->if_flags;
 818                 break;
 819
 820         case SIOCADDMULTI:
 821         case SIOCDELMULTI:
 822                 if (ifp->if_flags & IFF_RUNNING) {
 823                         igb_disable_intr(sc);
 824                         igb_set_multi(sc);
 825 #ifdef IFPOLL_ENABLE
 826                         if (!(ifp->if_flags & IFF_NPOLLING))
 827 #endif
 828                                 igb_enable_intr(sc);
 829                 }
 830                 break;
 831
 832         case SIOCSIFMEDIA:
 833                 /*
 834                  * As the speed/duplex settings are being
 835                  * changed, we need toreset the PHY.
 836                  */
 837                 sc->hw.phy.reset_disable = FALSE;
 838
 839                 /* Check SOL/IDER usage */
 840                 if (e1000_check_reset_block(&sc->hw)) {
 841                         if_printf(ifp, "Media change is "
 842                             "blocked due to SOL/IDER session.\n");
 843                         break;
 844                 }
 845                 /* FALL THROUGH */
 846
 847         case SIOCGIFMEDIA:
 848                 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
 849                 break;
 850
 851         case SIOCSIFCAP:
 852                 reinit = 0;
 853                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
 854                 if (mask & IFCAP_RXCSUM) {
 855                         ifp->if_capenable ^= IFCAP_RXCSUM;
 856                         reinit = 1;
 857                 }
 858                 if (mask & IFCAP_VLAN_HWTAGGING) {
 859                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
 860                         reinit = 1;
 861                 }
 862                 if (mask & IFCAP_TXCSUM) {
 863                         ifp->if_capenable ^= IFCAP_TXCSUM;
 864                         if (ifp->if_capenable & IFCAP_TXCSUM)
 865                                 ifp->if_hwassist |= IGB_CSUM_FEATURES;
 866                         else
 867                                 ifp->if_hwassist &= ~IGB_CSUM_FEATURES;
 868                 }
 869                 if (mask & IFCAP_TSO) {
 870                         ifp->if_capenable ^= IFCAP_TSO;
 871                         if (ifp->if_capenable & IFCAP_TSO)
 872                                 ifp->if_hwassist |= CSUM_TSO;
 873                         else
 874                                 ifp->if_hwassist &= ~CSUM_TSO;
 875                 }
 876                 if (mask & IFCAP_RSS)
 877                         ifp->if_capenable ^= IFCAP_RSS;
 878                 if (reinit && (ifp->if_flags & IFF_RUNNING))
 879                         igb_init(sc);
 880                 break;
 881
 882         default:
 883                 error = ether_ioctl(ifp, command, data);
 884                 break;
 885         }
 886         return error;
 887 }
 888
 889 static void
 890 igb_init(void *xsc)
 891 {
 892         struct igb_softc *sc = xsc;
 893         struct ifnet *ifp = &sc->arpcom.ac_if;
 894         boolean_t polling;
 895         int i;
 896
 897         ASSERT_IFNET_SERIALIZED_ALL(ifp);
 898
 899         igb_stop(sc);
 900
 901         /* Get the latest mac address, User can use a LAA */
 902         bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
 903
 904         /* Put the address into the Receive Address Array */
 905         e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
 906
 907         igb_reset(sc);
 908         igb_update_link_status(sc);
 909
 910         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
 911
 912         /* Configure for OS presence */
 913         igb_get_mgmt(sc);
 914
 915         polling = FALSE;
 916 #ifdef IFPOLL_ENABLE
 917         if (ifp->if_flags & IFF_NPOLLING)
 918                 polling = TRUE;
 919 #endif
 920
 921         /* Configured used RX/TX rings */
 922         igb_set_ring_inuse(sc, polling);
 923
 924         /* Initialize interrupt */
 925         igb_init_intr(sc);
 926
 927         /* Prepare transmit descriptors and buffers */
 928         for (i = 0; i < sc->tx_ring_cnt; ++i)
 929                 igb_init_tx_ring(&sc->tx_rings[i]);
 930         igb_init_tx_unit(sc);
 931
 932         /* Setup Multicast table */
 933         igb_set_multi(sc);
 934
 935 #if 0
 936         /*
 937          * Figure out the desired mbuf pool
 938          * for doing jumbo/packetsplit
 939          */
 940         if (adapter->max_frame_size <= 2048)
 941                 adapter->rx_mbuf_sz = MCLBYTES;
 942         else if (adapter->max_frame_size <= 4096)
 943                 adapter->rx_mbuf_sz = MJUMPAGESIZE;
 944         else
 945                 adapter->rx_mbuf_sz = MJUM9BYTES;
 946 #endif
 947
 948         /* Prepare receive descriptors and buffers */
 949         for (i = 0; i < sc->rx_ring_inuse; ++i) {
 950                 int error;
 951
 952                 error = igb_init_rx_ring(&sc->rx_rings[i]);
 953                 if (error) {
 954                         if_printf(ifp, "Could not setup receive structures\n");
 955                         igb_stop(sc);
 956                         return;
 957                 }
 958         }
 959         igb_init_rx_unit(sc);
 960
 961         /* Enable VLAN support */
 962         if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
 963                 igb_set_vlan(sc);
 964
 965         /* Don't lose promiscuous settings */
 966         igb_set_promisc(sc);
 967
 968         ifp->if_flags |= IFF_RUNNING;
 969         for (i = 0; i < sc->tx_ring_cnt; ++i)
 970                 ifsq_clr_oactive(sc->tx_rings[i].ifsq);
 971
 972         if (polling || sc->intr_type == PCI_INTR_TYPE_MSIX)
 973                 sc->timer_cpuid = 0; /* XXX fixed */
 974         else
 975                 sc->timer_cpuid = rman_get_cpuid(sc->intr_res);
 976         callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
 977         e1000_clear_hw_cntrs_base_generic(&sc->hw);
 978
 979         /* This clears any pending interrupts */
 980         E1000_READ_REG(&sc->hw, E1000_ICR);
 981
 982         /*
 983          * Only enable interrupts if we are not polling, make sure
 984          * they are off otherwise.
 985          */
 986         if (polling) {
 987                 igb_disable_intr(sc);
 988         } else {
 989                 igb_enable_intr(sc);
 990                 E1000_WRITE_REG(&sc->hw, E1000_ICS, E1000_ICS_LSC);
 991         }
 992
 993         /* Set Energy Efficient Ethernet */
 994         e1000_set_eee_i350(&sc->hw);
 995
 996         /* Don't reset the phy next time init gets called */
 997         sc->hw.phy.reset_disable = TRUE;
 998 }
 999
1000 static void
1001 igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1002 {
1003         struct igb_softc *sc = ifp->if_softc;
1004         u_char fiber_type = IFM_1000_SX;
1005
1006         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1007
1008         igb_update_link_status(sc);
1009
1010         ifmr->ifm_status = IFM_AVALID;
1011         ifmr->ifm_active = IFM_ETHER;
1012
1013         if (!sc->link_active)
1014                 return;
1015
1016         ifmr->ifm_status |= IFM_ACTIVE;
1017
1018         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1019             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1020                 ifmr->ifm_active |= fiber_type | IFM_FDX;
1021         } else {
1022                 switch (sc->link_speed) {
1023                 case 10:
1024                         ifmr->ifm_active |= IFM_10_T;
1025                         break;
1026
1027                 case 100:
1028                         ifmr->ifm_active |= IFM_100_TX;
1029                         break;
1030
1031                 case 1000:
1032                         ifmr->ifm_active |= IFM_1000_T;
1033                         break;
1034                 }
1035                 if (sc->link_duplex == FULL_DUPLEX)
1036                         ifmr->ifm_active |= IFM_FDX;
1037                 else
1038                         ifmr->ifm_active |= IFM_HDX;
1039         }
1040 }
1041
1042 static int
1043 igb_media_change(struct ifnet *ifp)
1044 {
1045         struct igb_softc *sc = ifp->if_softc;
1046         struct ifmedia *ifm = &sc->media;
1047
1048         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1049
1050         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1051                 return EINVAL;
1052
1053         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1054         case IFM_AUTO:
1055                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1056                 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1057                 break;
1058
1059         case IFM_1000_LX:
1060         case IFM_1000_SX:
1061         case IFM_1000_T:
1062                 sc->hw.mac.autoneg = DO_AUTO_NEG;
1063                 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1064                 break;
1065
1066         case IFM_100_TX:
1067                 sc->hw.mac.autoneg = FALSE;
1068                 sc->hw.phy.autoneg_advertised = 0;
1069                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1070                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1071                 else
1072                         sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1073                 break;
1074
1075         case IFM_10_T:
1076                 sc->hw.mac.autoneg = FALSE;
1077                 sc->hw.phy.autoneg_advertised = 0;
1078                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1079                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1080                 else
1081                         sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1082                 break;
1083
1084         default:
1085                 if_printf(ifp, "Unsupported media type\n");
1086                 break;
1087         }
1088
1089         igb_init(sc);
1090
1091         return 0;
1092 }
1093
1094 static void
1095 igb_set_promisc(struct igb_softc *sc)
1096 {
1097         struct ifnet *ifp = &sc->arpcom.ac_if;
1098         struct e1000_hw *hw = &sc->hw;
1099         uint32_t reg;
1100
1101         if (sc->vf_ifp) {
1102                 e1000_promisc_set_vf(hw, e1000_promisc_enabled);
1103                 return;
1104         }
1105
1106         reg = E1000_READ_REG(hw, E1000_RCTL);
1107         if (ifp->if_flags & IFF_PROMISC) {
1108                 reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1109                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1110         } else if (ifp->if_flags & IFF_ALLMULTI) {
1111                 reg |= E1000_RCTL_MPE;
1112                 reg &= ~E1000_RCTL_UPE;
1113                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1114         }
1115 }
1116
1117 static void
1118 igb_disable_promisc(struct igb_softc *sc)
1119 {
1120         struct e1000_hw *hw = &sc->hw;
1121         uint32_t reg;
1122
1123         if (sc->vf_ifp) {
1124                 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
1125                 return;
1126         }
1127         reg = E1000_READ_REG(hw, E1000_RCTL);
1128         reg &= ~E1000_RCTL_UPE;
1129         reg &= ~E1000_RCTL_MPE;
1130         E1000_WRITE_REG(hw, E1000_RCTL, reg);
1131 }
1132
1133 static void
1134 igb_set_multi(struct igb_softc *sc)
1135 {
1136         struct ifnet *ifp = &sc->arpcom.ac_if;
1137         struct ifmultiaddr *ifma;
1138         uint32_t reg_rctl = 0;
1139         uint8_t *mta;
1140         int mcnt = 0;
1141
1142         mta = sc->mta;
1143         bzero(mta, ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
1144
1145         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1146                 if (ifma->ifma_addr->sa_family != AF_LINK)
1147                         continue;
1148
1149                 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1150                         break;
1151
1152                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1153                     &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
1154                 mcnt++;
1155         }
1156
1157         if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1158                 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1159                 reg_rctl |= E1000_RCTL_MPE;
1160                 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1161         } else {
1162                 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1163         }
1164 }
1165
1166 static void
1167 igb_timer(void *xsc)
1168 {
1169         struct igb_softc *sc = xsc;
1170
1171         lwkt_serialize_enter(&sc->main_serialize);
1172
1173         igb_update_link_status(sc);
1174         igb_update_stats_counters(sc);
1175
1176         callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
1177
1178         lwkt_serialize_exit(&sc->main_serialize);
1179 }
1180
1181 static void
1182 igb_update_link_status(struct igb_softc *sc)
1183 {
1184         struct ifnet *ifp = &sc->arpcom.ac_if;
1185         struct e1000_hw *hw = &sc->hw;
1186         uint32_t link_check, thstat, ctrl;
1187
1188         link_check = thstat = ctrl = 0;
1189
1190         /* Get the cached link value or read for real */
1191         switch (hw->phy.media_type) {
1192         case e1000_media_type_copper:
1193                 if (hw->mac.get_link_status) {
1194                         /* Do the work to read phy */
1195                         e1000_check_for_link(hw);
1196                         link_check = !hw->mac.get_link_status;
1197                 } else {
1198                         link_check = TRUE;
1199                 }
1200                 break;
1201
1202         case e1000_media_type_fiber:
1203                 e1000_check_for_link(hw);
1204                 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1205                 break;
1206
1207         case e1000_media_type_internal_serdes:
1208                 e1000_check_for_link(hw);
1209                 link_check = hw->mac.serdes_has_link;
1210                 break;
1211
1212         /* VF device is type_unknown */
1213         case e1000_media_type_unknown:
1214                 e1000_check_for_link(hw);
1215                 link_check = !hw->mac.get_link_status;
1216                 /* Fall thru */
1217         default:
1218                 break;
1219         }
1220
1221         /* Check for thermal downshift or shutdown */
1222         if (hw->mac.type == e1000_i350) {
1223                 thstat = E1000_READ_REG(hw, E1000_THSTAT);
1224                 ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
1225         }
1226
1227         /* Now we check if a transition has happened */
1228         if (link_check && sc->link_active == 0) {
1229                 e1000_get_speed_and_duplex(hw,
1230                     &sc->link_speed, &sc->link_duplex);
1231                 if (bootverbose) {
1232                         if_printf(ifp, "Link is up %d Mbps %s\n",
1233                             sc->link_speed,
1234                             sc->link_duplex == FULL_DUPLEX ?
1235                             "Full Duplex" : "Half Duplex");
1236                 }
1237                 sc->link_active = 1;
1238
1239                 ifp->if_baudrate = sc->link_speed * 1000000;
1240                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1241                     (thstat & E1000_THSTAT_LINK_THROTTLE))
1242                         if_printf(ifp, "Link: thermal downshift\n");
1243                 /* This can sleep */
1244                 ifp->if_link_state = LINK_STATE_UP;
1245                 if_link_state_change(ifp);
1246         } else if (!link_check && sc->link_active == 1) {
1247                 ifp->if_baudrate = sc->link_speed = 0;
1248                 sc->link_duplex = 0;
1249                 if (bootverbose)
1250                         if_printf(ifp, "Link is Down\n");
1251                 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1252                     (thstat & E1000_THSTAT_PWR_DOWN))
1253                         if_printf(ifp, "Link: thermal shutdown\n");
1254                 sc->link_active = 0;
1255                 /* This can sleep */
1256                 ifp->if_link_state = LINK_STATE_DOWN;
1257                 if_link_state_change(ifp);
1258         }
1259 }
1260
1261 static void
1262 igb_stop(struct igb_softc *sc)
1263 {
1264         struct ifnet *ifp = &sc->arpcom.ac_if;
1265         int i;
1266
1267         ASSERT_IFNET_SERIALIZED_ALL(ifp);
1268
1269         igb_disable_intr(sc);
1270
1271         callout_stop(&sc->timer);
1272
1273         ifp->if_flags &= ~IFF_RUNNING;
1274         for (i = 0; i < sc->tx_ring_cnt; ++i)
1275                 ifsq_clr_oactive(sc->tx_rings[i].ifsq);
1276         ifp->if_timer = 0;
1277
1278         e1000_reset_hw(&sc->hw);
1279         E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1280
1281         e1000_led_off(&sc->hw);
1282         e1000_cleanup_led(&sc->hw);
1283
1284         for (i = 0; i < sc->tx_ring_cnt; ++i)
1285                 igb_free_tx_ring(&sc->tx_rings[i]);
1286         for (i = 0; i < sc->rx_ring_cnt; ++i)
1287                 igb_free_rx_ring(&sc->rx_rings[i]);
1288 }
1289
1290 static void
1291 igb_reset(struct igb_softc *sc)
1292 {
1293         struct ifnet *ifp = &sc->arpcom.ac_if;
1294         struct e1000_hw *hw = &sc->hw;
1295         struct e1000_fc_info *fc = &hw->fc;
1296         uint32_t pba = 0;
1297         uint16_t hwm;
1298
1299         /* Let the firmware know the OS is in control */
1300         igb_get_hw_control(sc);
1301
1302         /*
1303          * Packet Buffer Allocation (PBA)
1304          * Writing PBA sets the receive portion of the buffer
1305          * the remainder is used for the transmit buffer.
1306          */
1307         switch (hw->mac.type) {
1308         case e1000_82575:
1309                 pba = E1000_PBA_32K;
1310                 break;
1311
1312         case e1000_82576:
1313         case e1000_vfadapt:
1314                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1315                 pba &= E1000_RXPBS_SIZE_MASK_82576;
1316                 break;
1317
1318         case e1000_82580:
1319         case e1000_i350:
1320         case e1000_vfadapt_i350:
1321                 pba = E1000_READ_REG(hw, E1000_RXPBS);
1322                 pba = e1000_rxpbs_adjust_82580(pba);
1323                 break;
1324                 /* XXX pba = E1000_PBA_35K; */
1325
1326         default:
1327                 break;
1328         }
1329
1330         /* Special needs in case of Jumbo frames */
1331         if (hw->mac.type == e1000_82575 && ifp->if_mtu > ETHERMTU) {
1332                 uint32_t tx_space, min_tx, min_rx;
1333
1334                 pba = E1000_READ_REG(hw, E1000_PBA);
1335                 tx_space = pba >> 16;
1336                 pba &= 0xffff;
1337
1338                 min_tx = (sc->max_frame_size +
1339                     sizeof(struct e1000_tx_desc) - ETHER_CRC_LEN) * 2;
1340                 min_tx = roundup2(min_tx, 1024);
1341                 min_tx >>= 10;
1342                 min_rx = sc->max_frame_size;
1343                 min_rx = roundup2(min_rx, 1024);
1344                 min_rx >>= 10;
1345                 if (tx_space < min_tx && (min_tx - tx_space) < pba) {
1346                         pba = pba - (min_tx - tx_space);
1347                         /*
1348                          * if short on rx space, rx wins
1349                          * and must trump tx adjustment
1350                          */
1351                         if (pba < min_rx)
1352                                 pba = min_rx;
1353                 }
1354                 E1000_WRITE_REG(hw, E1000_PBA, pba);
1355         }
1356
1357         /*
1358          * These parameters control the automatic generation (Tx) and
1359          * response (Rx) to Ethernet PAUSE frames.
1360          * - High water mark should allow for at least two frames to be
1361          *   received after sending an XOFF.
1362          * - Low water mark works best when it is very near the high water mark.
1363          *   This allows the receiver to restart by sending XON when it has
1364          *   drained a bit.
1365          */
1366         hwm = min(((pba << 10) * 9 / 10),
1367             ((pba << 10) - 2 * sc->max_frame_size));
1368
1369         if (hw->mac.type < e1000_82576) {
1370                 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1371                 fc->low_water = fc->high_water - 8;
1372         } else {
1373                 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1374                 fc->low_water = fc->high_water - 16;
1375         }
1376         fc->pause_time = IGB_FC_PAUSE_TIME;
1377         fc->send_xon = TRUE;
1378
1379         /* Issue a global reset */
1380         e1000_reset_hw(hw);
1381         E1000_WRITE_REG(hw, E1000_WUC, 0);
1382
1383         if (e1000_init_hw(hw) < 0)
1384                 if_printf(ifp, "Hardware Initialization Failed\n");
1385
1386         /* Setup DMA Coalescing */
1387         if (hw->mac.type == e1000_i350 && sc->dma_coalesce) {
1388                 uint32_t reg;
1389
1390                 hwm = (pba - 4) << 10;
1391                 reg = ((pba - 6) << E1000_DMACR_DMACTHR_SHIFT)
1392                     & E1000_DMACR_DMACTHR_MASK;
1393
1394                 /* transition to L0x or L1 if available..*/
1395                 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
1396
1397                 /* timer = +-1000 usec in 32usec intervals */
1398                 reg |= (1000 >> 5);
1399                 E1000_WRITE_REG(hw, E1000_DMACR, reg);
1400
1401                 /* No lower threshold */
1402                 E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
1403
1404                 /* set hwm to PBA -  2 * max frame size */
1405                 E1000_WRITE_REG(hw, E1000_FCRTC, hwm);
1406
1407                 /* Set the interval before transition */
1408                 reg = E1000_READ_REG(hw, E1000_DMCTLX);
1409                 reg |= 0x800000FF; /* 255 usec */
1410                 E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
1411
1412                 /* free space in tx packet buffer to wake from DMA coal */
1413                 E1000_WRITE_REG(hw, E1000_DMCTXTH,
1414                     (20480 - (2 * sc->max_frame_size)) >> 6);
1415
1416                 /* make low power state decision controlled by DMA coal */
1417                 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
1418                 E1000_WRITE_REG(hw, E1000_PCIEMISC,
1419                     reg | E1000_PCIEMISC_LX_DECISION);
1420                 if_printf(ifp, "DMA Coalescing enabled\n");
1421         }
1422
1423         E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1424         e1000_get_phy_info(hw);
1425         e1000_check_for_link(hw);
1426 }
1427
1428 static void
1429 igb_setup_ifp(struct igb_softc *sc)
1430 {
1431         struct ifnet *ifp = &sc->arpcom.ac_if;
1432
1433         ifp->if_softc = sc;
1434         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1435         ifp->if_init = igb_init;
1436         ifp->if_ioctl = igb_ioctl;
1437         ifp->if_start = igb_start;
1438         ifp->if_serialize = igb_serialize;
1439         ifp->if_deserialize = igb_deserialize;
1440         ifp->if_tryserialize = igb_tryserialize;
1441 #ifdef INVARIANTS
1442         ifp->if_serialize_assert = igb_serialize_assert;
1443 #endif
1444 #ifdef IFPOLL_ENABLE
1445         ifp->if_npoll = igb_npoll;
1446 #endif
1447         ifp->if_watchdog = igb_watchdog;
1448
1449         ifq_set_maxlen(&ifp->if_snd, sc->tx_rings[0].num_tx_desc - 1);
1450         ifq_set_ready(&ifp->if_snd);
1451
1452         ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1453
1454         ifp->if_capabilities =
1455             IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_TSO;
1456         if (IGB_ENABLE_HWRSS(sc))
1457                 ifp->if_capabilities |= IFCAP_RSS;
1458         ifp->if_capenable = ifp->if_capabilities;
1459         ifp->if_hwassist = IGB_CSUM_FEATURES | CSUM_TSO;
1460
1461         /*
1462          * Tell the upper layer(s) we support long frames
1463          */
1464         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1465
1466         /*
1467          * Specify the media types supported by this adapter and register
1468          * callbacks to update media and link information
1469          */
1470         ifmedia_init(&sc->media, IFM_IMASK, igb_media_change, igb_media_status);
1471         if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1472             sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1473                 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1474                     0, NULL);
1475                 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1476         } else {
1477                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1478                 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1479                     0, NULL);
1480                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1481                 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1482                     0, NULL);
1483                 if (sc->hw.phy.type != e1000_phy_ife) {
1484                         ifmedia_add(&sc->media,
1485                             IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1486                         ifmedia_add(&sc->media,
1487                             IFM_ETHER | IFM_1000_T, 0, NULL);
1488                 }
1489         }
1490         ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1491         ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1492 }
1493
1494 static void
1495 igb_add_sysctl(struct igb_softc *sc)
1496 {
1497         char node[32];
1498         int i;
1499
1500         sysctl_ctx_init(&sc->sysctl_ctx);
1501         sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
1502             SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
1503             device_get_nameunit(sc->dev), CTLFLAG_RD, 0, "");
1504         if (sc->sysctl_tree == NULL) {
1505                 device_printf(sc->dev, "can't add sysctl node\n");
1506                 return;
1507         }
1508
1509         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1510             OID_AUTO, "rxr", CTLFLAG_RD, &sc->rx_ring_cnt, 0, "# of RX rings");
1511         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1512             OID_AUTO, "rxr_inuse", CTLFLAG_RD, &sc->rx_ring_inuse, 0,
1513             "# of RX rings used");
1514         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1515             OID_AUTO, "rxd", CTLFLAG_RD, &sc->rx_rings[0].num_rx_desc, 0,
1516             "# of RX descs");
1517         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1518             OID_AUTO, "txd", CTLFLAG_RD, &sc->tx_rings[0].num_tx_desc, 0,
1519             "# of TX descs");
1520
1521         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
1522                 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1523                     SYSCTL_CHILDREN(sc->sysctl_tree),
1524                     OID_AUTO, "intr_rate", CTLTYPE_INT | CTLFLAG_RW,
1525                     sc, 0, igb_sysctl_intr_rate, "I", "interrupt rate");
1526         } else {
1527                 for (i = 0; i < sc->msix_cnt; ++i) {
1528                         struct igb_msix_data *msix = &sc->msix_data[i];
1529
1530                         ksnprintf(node, sizeof(node), "msix%d_rate", i);
1531                         SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1532                             SYSCTL_CHILDREN(sc->sysctl_tree),
1533                             OID_AUTO, node, CTLTYPE_INT | CTLFLAG_RW,
1534                             msix, 0, igb_sysctl_msix_rate, "I",
1535                             msix->msix_rate_desc);
1536                 }
1537         }
1538
1539         SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1540             OID_AUTO, "tx_intr_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1541             sc, 0, igb_sysctl_tx_intr_nsegs, "I",
1542             "# of segments per TX interrupt");
1543
1544         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1545             OID_AUTO, "tx_wreg_nsegs", CTLFLAG_RW,
1546             &sc->tx_rings[0].wreg_nsegs, 0,
1547             "# of segments before write to hardare register");
1548
1549 #ifdef IFPOLL_ENABLE
1550         SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1551             OID_AUTO, "npoll_rxoff", CTLTYPE_INT|CTLFLAG_RW,
1552             sc, 0, igb_sysctl_npoll_rxoff, "I", "NPOLLING RX cpu offset");
1553         SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1554             OID_AUTO, "npoll_txoff", CTLTYPE_INT|CTLFLAG_RW,
1555             sc, 0, igb_sysctl_npoll_txoff, "I", "NPOLLING TX cpu offset");
1556 #endif
1557
1558 #ifdef IGB_RSS_DEBUG
1559         SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1560             OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug, 0,
1561             "RSS debug level");
1562 #endif
1563         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1564 #ifdef IGB_RSS_DEBUG
1565                 ksnprintf(node, sizeof(node), "rx%d_pkt", i);
1566                 SYSCTL_ADD_ULONG(&sc->sysctl_ctx,
1567                     SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, node,
1568                     CTLFLAG_RW, &sc->rx_rings[i].rx_packets, "RXed packets");
1569 #endif
1570                 ksnprintf(node, sizeof(node), "rx%d_wreg", i);
1571                 SYSCTL_ADD_INT(&sc->sysctl_ctx,
1572                     SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, node,
1573                     CTLFLAG_RW, &sc->rx_rings[i].rx_wreg, 0,
1574                     "# of segments before write to hardare register");
1575         }
1576 }
1577
1578 static int
1579 igb_alloc_rings(struct igb_softc *sc)
1580 {
1581         int error, i;
1582
1583         /*
1584          * Create top level busdma tag
1585          */
1586         error = bus_dma_tag_create(NULL, 1, 0,
1587             BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1588             BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1589             &sc->parent_tag);
1590         if (error) {
1591                 device_printf(sc->dev, "could not create top level DMA tag\n");
1592                 return error;
1593         }
1594
1595         /*
1596          * Allocate TX descriptor rings and buffers
1597          */
1598         sc->tx_rings = kmalloc_cachealign(
1599             sizeof(struct igb_tx_ring) * sc->tx_ring_cnt,
1600             M_DEVBUF, M_WAITOK | M_ZERO);
1601         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1602                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1603
1604                 /* Set up some basics */
1605                 txr->sc = sc;
1606                 txr->me = i;
1607                 lwkt_serialize_init(&txr->tx_serialize);
1608
1609                 error = igb_create_tx_ring(txr);
1610                 if (error)
1611                         return error;
1612         }
1613
1614         /*
1615          * Allocate RX descriptor rings and buffers
1616          */
1617         sc->rx_rings = kmalloc_cachealign(
1618             sizeof(struct igb_rx_ring) * sc->rx_ring_cnt,
1619             M_DEVBUF, M_WAITOK | M_ZERO);
1620         for (i = 0; i < sc->rx_ring_cnt; ++i) {
1621                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1622
1623                 /* Set up some basics */
1624                 rxr->sc = sc;
1625                 rxr->me = i;
1626                 lwkt_serialize_init(&rxr->rx_serialize);
1627
1628                 error = igb_create_rx_ring(rxr);
1629                 if (error)
1630                         return error;
1631         }
1632
1633         return 0;
1634 }
1635
1636 static void
1637 igb_free_rings(struct igb_softc *sc)
1638 {
1639         int i;
1640
1641         if (sc->tx_rings != NULL) {
1642                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1643                         struct igb_tx_ring *txr = &sc->tx_rings[i];
1644
1645                         igb_destroy_tx_ring(txr, txr->num_tx_desc);
1646                 }
1647                 kfree(sc->tx_rings, M_DEVBUF);
1648         }
1649
1650         if (sc->rx_rings != NULL) {
1651                 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1652                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
1653
1654                         igb_destroy_rx_ring(rxr, rxr->num_rx_desc);
1655                 }
1656                 kfree(sc->rx_rings, M_DEVBUF);
1657         }
1658 }
1659
1660 static int
1661 igb_create_tx_ring(struct igb_tx_ring *txr)
1662 {
1663         int tsize, error, i, ntxd;
1664
1665         /*
1666          * Validate number of transmit descriptors. It must not exceed
1667          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1668          */
1669         ntxd = device_getenv_int(txr->sc->dev, "txd", igb_txd);
1670         if ((ntxd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN != 0 ||
1671             ntxd > IGB_MAX_TXD || ntxd < IGB_MIN_TXD) {
1672                 device_printf(txr->sc->dev,
1673                     "Using %d TX descriptors instead of %d!\n",
1674                     IGB_DEFAULT_TXD, ntxd);
1675                 txr->num_tx_desc = IGB_DEFAULT_TXD;
1676         } else {
1677                 txr->num_tx_desc = ntxd;
1678         }
1679
1680         /*
1681          * Allocate TX descriptor ring
1682          */
1683         tsize = roundup2(txr->num_tx_desc * sizeof(union e1000_adv_tx_desc),
1684             IGB_DBA_ALIGN);
1685         txr->txdma.dma_vaddr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1686             IGB_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1687             &txr->txdma.dma_tag, &txr->txdma.dma_map, &txr->txdma.dma_paddr);
1688         if (txr->txdma.dma_vaddr == NULL) {
1689                 device_printf(txr->sc->dev,
1690                     "Unable to allocate TX Descriptor memory\n");
1691                 return ENOMEM;
1692         }
1693         txr->tx_base = txr->txdma.dma_vaddr;
1694         bzero(txr->tx_base, tsize);
1695
1696         tsize = __VM_CACHELINE_ALIGN(
1697             sizeof(struct igb_tx_buf) * txr->num_tx_desc);
1698         txr->tx_buf = kmalloc_cachealign(tsize, M_DEVBUF, M_WAITOK | M_ZERO);
1699
1700         /*
1701          * Allocate TX head write-back buffer
1702          */
1703         txr->tx_hdr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1704             __VM_CACHELINE_SIZE, __VM_CACHELINE_SIZE, BUS_DMA_WAITOK,
1705             &txr->tx_hdr_dtag, &txr->tx_hdr_dmap, &txr->tx_hdr_paddr);
1706         if (txr->tx_hdr == NULL) {
1707                 device_printf(txr->sc->dev,
1708                     "Unable to allocate TX head write-back buffer\n");
1709                 return ENOMEM;
1710         }
1711
1712         /*
1713          * Create DMA tag for TX buffers
1714          */
1715         error = bus_dma_tag_create(txr->sc->parent_tag,
1716             1, 0,               /* alignment, bounds */
1717             BUS_SPACE_MAXADDR,  /* lowaddr */
1718             BUS_SPACE_MAXADDR,  /* highaddr */
1719             NULL, NULL,         /* filter, filterarg */
1720             IGB_TSO_SIZE,       /* maxsize */
1721             IGB_MAX_SCATTER,    /* nsegments */
1722             PAGE_SIZE,          /* maxsegsize */
1723             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1724             BUS_DMA_ONEBPAGE,   /* flags */
1725             &txr->tx_tag);
1726         if (error) {
1727                 device_printf(txr->sc->dev, "Unable to allocate TX DMA tag\n");
1728                 kfree(txr->tx_buf, M_DEVBUF);
1729                 txr->tx_buf = NULL;
1730                 return error;
1731         }
1732
1733         /*
1734          * Create DMA maps for TX buffers
1735          */
1736         for (i = 0; i < txr->num_tx_desc; ++i) {
1737                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1738
1739                 error = bus_dmamap_create(txr->tx_tag,
1740                     BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE, &txbuf->map);
1741                 if (error) {
1742                         device_printf(txr->sc->dev,
1743                             "Unable to create TX DMA map\n");
1744                         igb_destroy_tx_ring(txr, i);
1745                         return error;
1746                 }
1747         }
1748
1749         /*
1750          * Initialize various watermark
1751          */
1752         txr->spare_desc = IGB_TX_SPARE;
1753         txr->intr_nsegs = txr->num_tx_desc / 16;
1754         txr->wreg_nsegs = 8;
1755         txr->oact_hi_desc = txr->num_tx_desc / 2;
1756         txr->oact_lo_desc = txr->num_tx_desc / 8;
1757         if (txr->oact_lo_desc > IGB_TX_OACTIVE_MAX)
1758                 txr->oact_lo_desc = IGB_TX_OACTIVE_MAX;
1759         if (txr->oact_lo_desc < txr->spare_desc + IGB_TX_RESERVED)
1760                 txr->oact_lo_desc = txr->spare_desc + IGB_TX_RESERVED;
1761
1762         return 0;
1763 }
1764
1765 static void
1766 igb_free_tx_ring(struct igb_tx_ring *txr)
1767 {
1768         int i;
1769
1770         for (i = 0; i < txr->num_tx_desc; ++i) {
1771                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1772
1773                 if (txbuf->m_head != NULL) {
1774                         bus_dmamap_unload(txr->tx_tag, txbuf->map);
1775                         m_freem(txbuf->m_head);
1776                         txbuf->m_head = NULL;
1777                 }
1778         }
1779 }
1780
1781 static void
1782 igb_destroy_tx_ring(struct igb_tx_ring *txr, int ndesc)
1783 {
1784         int i;
1785
1786         if (txr->txdma.dma_vaddr != NULL) {
1787                 bus_dmamap_unload(txr->txdma.dma_tag, txr->txdma.dma_map);
1788                 bus_dmamem_free(txr->txdma.dma_tag, txr->txdma.dma_vaddr,
1789                     txr->txdma.dma_map);
1790                 bus_dma_tag_destroy(txr->txdma.dma_tag);
1791                 txr->txdma.dma_vaddr = NULL;
1792         }
1793
1794         if (txr->tx_hdr != NULL) {
1795                 bus_dmamap_unload(txr->tx_hdr_dtag, txr->tx_hdr_dmap);
1796                 bus_dmamem_free(txr->tx_hdr_dtag, txr->tx_hdr,
1797                     txr->tx_hdr_dmap);
1798                 bus_dma_tag_destroy(txr->tx_hdr_dtag);
1799                 txr->tx_hdr = NULL;
1800         }
1801
1802         if (txr->tx_buf == NULL)
1803                 return;
1804
1805         for (i = 0; i < ndesc; ++i) {
1806                 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1807
1808                 KKASSERT(txbuf->m_head == NULL);
1809                 bus_dmamap_destroy(txr->tx_tag, txbuf->map);
1810         }
1811         bus_dma_tag_destroy(txr->tx_tag);
1812
1813         kfree(txr->tx_buf, M_DEVBUF);
1814         txr->tx_buf = NULL;
1815 }
1816
1817 static void
1818 igb_init_tx_ring(struct igb_tx_ring *txr)
1819 {
1820         /* Clear the old descriptor contents */
1821         bzero(txr->tx_base,
1822             sizeof(union e1000_adv_tx_desc) * txr->num_tx_desc);
1823
1824         /* Clear TX head write-back buffer */
1825         *(txr->tx_hdr) = 0;
1826
1827         /* Reset indices */
1828         txr->next_avail_desc = 0;
1829         txr->next_to_clean = 0;
1830         txr->tx_nsegs = 0;
1831
1832         /* Set number of descriptors available */
1833         txr->tx_avail = txr->num_tx_desc;
1834 }
1835
1836 static void
1837 igb_init_tx_unit(struct igb_softc *sc)
1838 {
1839         struct e1000_hw *hw = &sc->hw;
1840         uint32_t tctl;
1841         int i;
1842
1843         /* Setup the Tx Descriptor Rings */
1844         for (i = 0; i < sc->tx_ring_cnt; ++i) {
1845                 struct igb_tx_ring *txr = &sc->tx_rings[i];
1846                 uint64_t bus_addr = txr->txdma.dma_paddr;
1847                 uint64_t hdr_paddr = txr->tx_hdr_paddr;
1848                 uint32_t txdctl = 0;
1849                 uint32_t dca_txctrl;
1850
1851                 E1000_WRITE_REG(hw, E1000_TDLEN(i),
1852                     txr->num_tx_desc * sizeof(struct e1000_tx_desc));
1853                 E1000_WRITE_REG(hw, E1000_TDBAH(i),
1854                     (uint32_t)(bus_addr >> 32));
1855                 E1000_WRITE_REG(hw, E1000_TDBAL(i),
1856                     (uint32_t)bus_addr);
1857
1858                 /* Setup the HW Tx Head and Tail descriptor pointers */
1859                 E1000_WRITE_REG(hw, E1000_TDT(i), 0);
1860                 E1000_WRITE_REG(hw, E1000_TDH(i), 0);
1861
1862                 dca_txctrl = E1000_READ_REG(hw, E1000_DCA_TXCTRL(i));
1863                 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1864                 E1000_WRITE_REG(hw, E1000_DCA_TXCTRL(i), dca_txctrl);
1865
1866                 /*
1867                  * Don't set WB_on_EITR:
1868                  * - 82575 does not have it
1869                  * - It almost has no effect on 82576, see:
1870                  *   82576 specification update errata #26
1871                  * - It causes unnecessary bus traffic
1872                  */
1873                 E1000_WRITE_REG(hw, E1000_TDWBAH(i),
1874                     (uint32_t)(hdr_paddr >> 32));
1875                 E1000_WRITE_REG(hw, E1000_TDWBAL(i),
1876                     ((uint32_t)hdr_paddr) | E1000_TX_HEAD_WB_ENABLE);
1877
1878                 /*
1879                  * WTHRESH is ignored by the hardware, since header
1880                  * write back mode is used.
1881                  */
1882                 txdctl |= IGB_TX_PTHRESH;
1883                 txdctl |= IGB_TX_HTHRESH << 8;
1884                 txdctl |= IGB_TX_WTHRESH << 16;
1885                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1886                 E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
1887         }
1888
1889         if (sc->vf_ifp)
1890                 return;
1891
1892         e1000_config_collision_dist(hw);
1893
1894         /* Program the Transmit Control Register */
1895         tctl = E1000_READ_REG(hw, E1000_TCTL);
1896         tctl &= ~E1000_TCTL_CT;
1897         tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
1898             (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
1899
1900         /* This write will effectively turn on the transmit unit. */
1901         E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1902 }
1903
1904 static boolean_t
1905 igb_txcsum_ctx(struct igb_tx_ring *txr, struct mbuf *mp)
1906 {
1907         struct e1000_adv_tx_context_desc *TXD;
1908         uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
1909         int ehdrlen, ctxd, ip_hlen = 0;
1910         boolean_t offload = TRUE;
1911
1912         if ((mp->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) == 0)
1913                 offload = FALSE;
1914
1915         vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
1916
1917         ctxd = txr->next_avail_desc;
1918         TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
1919
1920         /*
1921          * In advanced descriptors the vlan tag must
1922          * be placed into the context descriptor, thus
1923          * we need to be here just for that setup.
1924          */
1925         if (mp->m_flags & M_VLANTAG) {
1926                 uint16_t vlantag;
1927
1928                 vlantag = htole16(mp->m_pkthdr.ether_vlantag);
1929                 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
1930         } else if (!offload) {
1931                 return FALSE;
1932         }
1933
1934         ehdrlen = mp->m_pkthdr.csum_lhlen;
1935         KASSERT(ehdrlen > 0, ("invalid ether hlen"));
1936
1937         /* Set the ether header length */
1938         vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
1939         if (mp->m_pkthdr.csum_flags & CSUM_IP) {
1940                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
1941                 ip_hlen = mp->m_pkthdr.csum_iphlen;
1942                 KASSERT(ip_hlen > 0, ("invalid ip hlen"));
1943         }
1944         vlan_macip_lens |= ip_hlen;
1945
1946         type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1947         if (mp->m_pkthdr.csum_flags & CSUM_TCP)
1948                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
1949         else if (mp->m_pkthdr.csum_flags & CSUM_UDP)
1950                 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
1951
1952         /* 82575 needs the queue index added */
1953         if (txr->sc->hw.mac.type == e1000_82575)
1954                 mss_l4len_idx = txr->me << 4;
1955
1956         /* Now copy bits into descriptor */
1957         TXD->vlan_macip_lens = htole32(vlan_macip_lens);
1958         TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
1959         TXD->seqnum_seed = htole32(0);
1960         TXD->mss_l4len_idx = htole32(mss_l4len_idx);
1961
1962         /* We've consumed the first desc, adjust counters */
1963         if (++ctxd == txr->num_tx_desc)
1964                 ctxd = 0;
1965         txr->next_avail_desc = ctxd;
1966         --txr->tx_avail;
1967
1968         return offload;
1969 }
1970
1971 static void
1972 igb_txeof(struct igb_tx_ring *txr)
1973 {
1974         struct ifnet *ifp = &txr->sc->arpcom.ac_if;
1975         int first, hdr, avail;
1976
1977         if (txr->tx_avail == txr->num_tx_desc)
1978                 return;
1979
1980         first = txr->next_to_clean;
1981         hdr = *(txr->tx_hdr);
1982
1983         if (first == hdr)
1984                 return;
1985
1986         avail = txr->tx_avail;
1987         while (first != hdr) {
1988                 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
1989
1990                 ++avail;
1991                 if (txbuf->m_head) {
1992                         bus_dmamap_unload(txr->tx_tag, txbuf->map);
1993                         m_freem(txbuf->m_head);
1994                         txbuf->m_head = NULL;
1995                         ++ifp->if_opackets;
1996                 }
1997                 if (++first == txr->num_tx_desc)
1998                         first = 0;
1999         }
2000         txr->next_to_clean = first;
2001         txr->tx_avail = avail;
2002
2003         /*
2004          * If we have a minimum free, clear OACTIVE
2005          * to tell the stack that it is OK to send packets.
2006          */
2007         if (IGB_IS_NOT_OACTIVE(txr)) {
2008                 ifsq_clr_oactive(txr->ifsq);
2009
2010                 /*
2011                  * We have enough TX descriptors, turn off
2012                  * the watchdog.  We allow small amount of
2013                  * packets (roughly intr_nsegs) pending on
2014                  * the transmit ring.
2015                  */
2016                 ifp->if_timer = 0;
2017         }
2018 }
2019
2020 static int
2021 igb_create_rx_ring(struct igb_rx_ring *rxr)
2022 {
2023         int rsize, i, error, nrxd;
2024
2025         /*
2026          * Validate number of receive descriptors. It must not exceed
2027          * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
2028          */
2029         nrxd = device_getenv_int(rxr->sc->dev, "rxd", igb_rxd);
2030         if ((nrxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN != 0 ||
2031             nrxd > IGB_MAX_RXD || nrxd < IGB_MIN_RXD) {
2032                 device_printf(rxr->sc->dev,
2033                     "Using %d RX descriptors instead of %d!\n",
2034                     IGB_DEFAULT_RXD, nrxd);
2035                 rxr->num_rx_desc = IGB_DEFAULT_RXD;
2036         } else {
2037                 rxr->num_rx_desc = nrxd;
2038         }
2039
2040         /*
2041          * Allocate RX descriptor ring
2042          */
2043         rsize = roundup2(rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc),
2044             IGB_DBA_ALIGN);
2045         rxr->rxdma.dma_vaddr = bus_dmamem_coherent_any(rxr->sc->parent_tag,
2046             IGB_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
2047             &rxr->rxdma.dma_tag, &rxr->rxdma.dma_map,
2048             &rxr->rxdma.dma_paddr);
2049         if (rxr->rxdma.dma_vaddr == NULL) {
2050                 device_printf(rxr->sc->dev,
2051                     "Unable to allocate RxDescriptor memory\n");
2052                 return ENOMEM;
2053         }
2054         rxr->rx_base = rxr->rxdma.dma_vaddr;
2055         bzero(rxr->rx_base, rsize);
2056
2057         rsize = __VM_CACHELINE_ALIGN(
2058             sizeof(struct igb_rx_buf) * rxr->num_rx_desc);
2059         rxr->rx_buf = kmalloc_cachealign(rsize, M_DEVBUF, M_WAITOK | M_ZERO);
2060
2061         /*
2062          * Create DMA tag for RX buffers
2063          */
2064         error = bus_dma_tag_create(rxr->sc->parent_tag,
2065             1, 0,               /* alignment, bounds */
2066             BUS_SPACE_MAXADDR,  /* lowaddr */
2067             BUS_SPACE_MAXADDR,  /* highaddr */
2068             NULL, NULL,         /* filter, filterarg */
2069             MCLBYTES,           /* maxsize */
2070             1,                  /* nsegments */
2071             MCLBYTES,           /* maxsegsize */
2072             BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2073             &rxr->rx_tag);
2074         if (error) {
2075                 device_printf(rxr->sc->dev,
2076                     "Unable to create RX payload DMA tag\n");
2077                 kfree(rxr->rx_buf, M_DEVBUF);
2078                 rxr->rx_buf = NULL;
2079                 return error;
2080         }
2081
2082         /*
2083          * Create spare DMA map for RX buffers
2084          */
2085         error = bus_dmamap_create(rxr->rx_tag, BUS_DMA_WAITOK,
2086             &rxr->rx_sparemap);
2087         if (error) {
2088                 device_printf(rxr->sc->dev,
2089                     "Unable to create spare RX DMA maps\n");
2090                 bus_dma_tag_destroy(rxr->rx_tag);
2091                 kfree(rxr->rx_buf, M_DEVBUF);
2092                 rxr->rx_buf = NULL;
2093                 return error;
2094         }
2095
2096         /*
2097          * Create DMA maps for RX buffers
2098          */
2099         for (i = 0; i < rxr->num_rx_desc; i++) {
2100                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2101
2102                 error = bus_dmamap_create(rxr->rx_tag,
2103                     BUS_DMA_WAITOK, &rxbuf->map);
2104                 if (error) {
2105                         device_printf(rxr->sc->dev,
2106                             "Unable to create RX DMA maps\n");
2107                         igb_destroy_rx_ring(rxr, i);
2108                         return error;
2109                 }
2110         }
2111
2112         /*
2113          * Initialize various watermark
2114          */
2115         rxr->rx_wreg = 32;
2116
2117         return 0;
2118 }
2119
2120 static void
2121 igb_free_rx_ring(struct igb_rx_ring *rxr)
2122 {
2123         int i;
2124
2125         for (i = 0; i < rxr->num_rx_desc; ++i) {
2126                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2127
2128                 if (rxbuf->m_head != NULL) {
2129                         bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2130                         m_freem(rxbuf->m_head);
2131                         rxbuf->m_head = NULL;
2132                 }
2133         }
2134
2135         if (rxr->fmp != NULL)
2136                 m_freem(rxr->fmp);
2137         rxr->fmp = NULL;
2138         rxr->lmp = NULL;
2139 }
2140
2141 static void
2142 igb_destroy_rx_ring(struct igb_rx_ring *rxr, int ndesc)
2143 {
2144         int i;
2145
2146         if (rxr->rxdma.dma_vaddr != NULL) {
2147                 bus_dmamap_unload(rxr->rxdma.dma_tag, rxr->rxdma.dma_map);
2148                 bus_dmamem_free(rxr->rxdma.dma_tag, rxr->rxdma.dma_vaddr,
2149                     rxr->rxdma.dma_map);
2150                 bus_dma_tag_destroy(rxr->rxdma.dma_tag);
2151                 rxr->rxdma.dma_vaddr = NULL;
2152         }
2153
2154         if (rxr->rx_buf == NULL)
2155                 return;
2156
2157         for (i = 0; i < ndesc; ++i) {
2158                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2159
2160                 KKASSERT(rxbuf->m_head == NULL);
2161                 bus_dmamap_destroy(rxr->rx_tag, rxbuf->map);
2162         }
2163         bus_dmamap_destroy(rxr->rx_tag, rxr->rx_sparemap);
2164         bus_dma_tag_destroy(rxr->rx_tag);
2165
2166         kfree(rxr->rx_buf, M_DEVBUF);
2167         rxr->rx_buf = NULL;
2168 }
2169
2170 static void
2171 igb_setup_rxdesc(union e1000_adv_rx_desc *rxd, const struct igb_rx_buf *rxbuf)
2172 {
2173         rxd->read.pkt_addr = htole64(rxbuf->paddr);
2174         rxd->wb.upper.status_error = 0;
2175 }
2176
2177 static int
2178 igb_newbuf(struct igb_rx_ring *rxr, int i, boolean_t wait)
2179 {
2180         struct mbuf *m;
2181         bus_dma_segment_t seg;
2182         bus_dmamap_t map;
2183         struct igb_rx_buf *rxbuf;
2184         int error, nseg;
2185
2186         m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2187         if (m == NULL) {
2188                 if (wait) {
2189                         if_printf(&rxr->sc->arpcom.ac_if,
2190                             "Unable to allocate RX mbuf\n");
2191                 }
2192                 return ENOBUFS;
2193         }
2194         m->m_len = m->m_pkthdr.len = MCLBYTES;
2195
2196         if (rxr->sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2197                 m_adj(m, ETHER_ALIGN);
2198
2199         error = bus_dmamap_load_mbuf_segment(rxr->rx_tag,
2200             rxr->rx_sparemap, m, &seg, 1, &nseg, BUS_DMA_NOWAIT);
2201         if (error) {
2202                 m_freem(m);
2203                 if (wait) {
2204                         if_printf(&rxr->sc->arpcom.ac_if,
2205                             "Unable to load RX mbuf\n");
2206                 }
2207                 return error;
2208         }
2209
2210         rxbuf = &rxr->rx_buf[i];
2211         if (rxbuf->m_head != NULL)
2212                 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2213
2214         map = rxbuf->map;
2215         rxbuf->map = rxr->rx_sparemap;
2216         rxr->rx_sparemap = map;
2217
2218         rxbuf->m_head = m;
2219         rxbuf->paddr = seg.ds_addr;
2220
2221         igb_setup_rxdesc(&rxr->rx_base[i], rxbuf);
2222         return 0;
2223 }
2224
2225 static int
2226 igb_init_rx_ring(struct igb_rx_ring *rxr)
2227 {
2228         int i;
2229
2230         /* Clear the ring contents */
2231         bzero(rxr->rx_base,
2232             rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc));
2233
2234         /* Now replenish the ring mbufs */
2235         for (i = 0; i < rxr->num_rx_desc; ++i) {
2236                 int error;
2237
2238                 error = igb_newbuf(rxr, i, TRUE);
2239                 if (error)
2240                         return error;
2241         }
2242
2243         /* Setup our descriptor indices */
2244         rxr->next_to_check = 0;
2245
2246         rxr->fmp = NULL;
2247         rxr->lmp = NULL;
2248         rxr->discard = FALSE;
2249
2250         return 0;
2251 }
2252
2253 static void
2254 igb_init_rx_unit(struct igb_softc *sc)
2255 {
2256         struct ifnet *ifp = &sc->arpcom.ac_if;
2257         struct e1000_hw *hw = &sc->hw;
2258         uint32_t rctl, rxcsum, srrctl = 0;
2259         int i;
2260
2261         /*
2262          * Make sure receives are disabled while setting
2263          * up the descriptor ring
2264          */
2265         rctl = E1000_READ_REG(hw, E1000_RCTL);
2266         E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2267
2268 #if 0
2269         /*
2270         ** Set up for header split
2271         */
2272         if (igb_header_split) {
2273                 /* Use a standard mbuf for the header */
2274                 srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2275                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2276         } else
2277 #endif
2278                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2279
2280         /*
2281         ** Set up for jumbo frames
2282         */
2283         if (ifp->if_mtu > ETHERMTU) {
2284                 rctl |= E1000_RCTL_LPE;
2285 #if 0
2286                 if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
2287                         srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2288                         rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
2289                 } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
2290                         srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2291                         rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
2292                 }
2293                 /* Set maximum packet len */
2294                 psize = adapter->max_frame_size;
2295                 /* are we on a vlan? */
2296                 if (adapter->ifp->if_vlantrunk != NULL)
2297                         psize += VLAN_TAG_SIZE;
2298                 E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
2299 #else
2300                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2301                 rctl |= E1000_RCTL_SZ_2048;
2302 #endif
2303         } else {
2304                 rctl &= ~E1000_RCTL_LPE;
2305                 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2306                 rctl |= E1000_RCTL_SZ_2048;
2307         }
2308
2309         /* Setup the Base and Length of the Rx Descriptor Rings */
2310         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2311                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2312                 uint64_t bus_addr = rxr->rxdma.dma_paddr;
2313                 uint32_t rxdctl;
2314
2315                 E1000_WRITE_REG(hw, E1000_RDLEN(i),
2316                     rxr->num_rx_desc * sizeof(struct e1000_rx_desc));
2317                 E1000_WRITE_REG(hw, E1000_RDBAH(i),
2318                     (uint32_t)(bus_addr >> 32));
2319                 E1000_WRITE_REG(hw, E1000_RDBAL(i),
2320                     (uint32_t)bus_addr);
2321                 E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
2322                 /* Enable this Queue */
2323                 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
2324                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2325                 rxdctl &= 0xFFF00000;
2326                 rxdctl |= IGB_RX_PTHRESH;
2327                 rxdctl |= IGB_RX_HTHRESH << 8;
2328                 /*
2329                  * Don't set WTHRESH to a value above 1 on 82576, see:
2330                  * 82576 specification update errata #26
2331                  */
2332                 rxdctl |= IGB_RX_WTHRESH << 16;
2333                 E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
2334         }
2335
2336         rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2337         rxcsum &= ~(E1000_RXCSUM_PCSS_MASK | E1000_RXCSUM_IPPCSE);
2338
2339         /*
2340          * Receive Checksum Offload for TCP and UDP
2341          *
2342          * Checksum offloading is also enabled if multiple receive
2343          * queue is to be supported, since we need it to figure out
2344          * fragments.
2345          */
2346         if ((ifp->if_capenable & IFCAP_RXCSUM) || IGB_ENABLE_HWRSS(sc)) {
2347                 /*
2348                  * NOTE:
2349                  * PCSD must be enabled to enable multiple
2350                  * receive queues.
2351                  */
2352                 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2353                     E1000_RXCSUM_PCSD;
2354         } else {
2355                 rxcsum &= ~(E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2356                     E1000_RXCSUM_PCSD);
2357         }
2358         E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2359
2360         if (IGB_ENABLE_HWRSS(sc)) {
2361                 uint8_t key[IGB_NRSSRK * IGB_RSSRK_SIZE];
2362                 uint32_t reta_shift;
2363                 int j, r;
2364
2365                 /*
2366                  * NOTE:
2367                  * When we reach here, RSS has already been disabled
2368                  * in igb_stop(), so we could safely configure RSS key
2369                  * and redirect table.
2370                  */
2371
2372                 /*
2373                  * Configure RSS key
2374                  */
2375                 toeplitz_get_key(key, sizeof(key));
2376                 for (i = 0; i < IGB_NRSSRK; ++i) {
2377                         uint32_t rssrk;
2378
2379                         rssrk = IGB_RSSRK_VAL(key, i);
2380                         IGB_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2381
2382                         E1000_WRITE_REG(hw, E1000_RSSRK(i), rssrk);
2383                 }
2384
2385                 /*
2386                  * Configure RSS redirect table in following fashion:
2387                  * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
2388                  */
2389                 reta_shift = IGB_RETA_SHIFT;
2390                 if (hw->mac.type == e1000_82575)
2391                         reta_shift = IGB_RETA_SHIFT_82575;
2392
2393                 r = 0;
2394                 for (j = 0; j < IGB_NRETA; ++j) {
2395                         uint32_t reta = 0;
2396
2397                         for (i = 0; i < IGB_RETA_SIZE; ++i) {
2398                                 uint32_t q;
2399
2400                                 q = (r % sc->rx_ring_inuse) << reta_shift;
2401                                 reta |= q << (8 * i);
2402                                 ++r;
2403                         }
2404                         IGB_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2405                         E1000_WRITE_REG(hw, E1000_RETA(j), reta);
2406                 }
2407
2408                 /*
2409                  * Enable multiple receive queues.
2410                  * Enable IPv4 RSS standard hash functions.
2411                  * Disable RSS interrupt on 82575
2412                  */
2413                 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2414                                 E1000_MRQC_ENABLE_RSS_4Q |
2415                                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2416                                 E1000_MRQC_RSS_FIELD_IPV4);
2417         }
2418
2419         /* Setup the Receive Control Register */
2420         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2421         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2422             E1000_RCTL_RDMTS_HALF |
2423             (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2424         /* Strip CRC bytes. */
2425         rctl |= E1000_RCTL_SECRC;
2426         /* Make sure VLAN Filters are off */
2427         rctl &= ~E1000_RCTL_VFE;
2428         /* Don't store bad packets */
2429         rctl &= ~E1000_RCTL_SBP;
2430
2431         /* Enable Receives */
2432         E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2433
2434         /*
2435          * Setup the HW Rx Head and Tail Descriptor Pointers
2436          *   - needs to be after enable
2437          */
2438         for (i = 0; i < sc->rx_ring_inuse; ++i) {
2439                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2440
2441                 E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
2442                 E1000_WRITE_REG(hw, E1000_RDT(i), rxr->num_rx_desc - 1);
2443         }
2444 }
2445
2446 static void
2447 igb_rx_refresh(struct igb_rx_ring *rxr, int i)
2448 {
2449         if (--i < 0)
2450                 i = rxr->num_rx_desc - 1;
2451         E1000_WRITE_REG(&rxr->sc->hw, E1000_RDT(rxr->me), i);
2452 }
2453
2454 static void
2455 igb_rxeof(struct igb_rx_ring *rxr, int count)
2456 {
2457         struct ifnet *ifp = &rxr->sc->arpcom.ac_if;
2458         union e1000_adv_rx_desc *cur;
2459         uint32_t staterr;
2460         int i, ncoll = 0;
2461
2462         i = rxr->next_to_check;
2463         cur = &rxr->rx_base[i];
2464         staterr = le32toh(cur->wb.upper.status_error);
2465
2466         if ((staterr & E1000_RXD_STAT_DD) == 0)
2467                 return;
2468
2469         while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
2470                 struct pktinfo *pi = NULL, pi0;
2471                 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2472                 struct mbuf *m = NULL;
2473                 boolean_t eop;
2474
2475                 eop = (staterr & E1000_RXD_STAT_EOP) ? TRUE : FALSE;
2476                 if (eop)
2477                         --count;
2478
2479                 ++ncoll;
2480                 if ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) == 0 &&
2481                     !rxr->discard) {
2482                         struct mbuf *mp = rxbuf->m_head;
2483                         uint32_t hash, hashtype;
2484                         uint16_t vlan;
2485                         int len;
2486
2487                         len = le16toh(cur->wb.upper.length);
2488                         if (rxr->sc->hw.mac.type == e1000_i350 &&
2489                             (staterr & E1000_RXDEXT_STATERR_LB))
2490                                 vlan = be16toh(cur->wb.upper.vlan);
2491                         else
2492                                 vlan = le16toh(cur->wb.upper.vlan);
2493
2494                         hash = le32toh(cur->wb.lower.hi_dword.rss);
2495                         hashtype = le32toh(cur->wb.lower.lo_dword.data) &
2496                             E1000_RXDADV_RSSTYPE_MASK;
2497
2498                         IGB_RSS_DPRINTF(rxr->sc, 10,
2499                             "ring%d, hash 0x%08x, hashtype %u\n",
2500                             rxr->me, hash, hashtype);
2501
2502                         bus_dmamap_sync(rxr->rx_tag, rxbuf->map,
2503                             BUS_DMASYNC_POSTREAD);
2504
2505                         if (igb_newbuf(rxr, i, FALSE) != 0) {
2506                                 ifp->if_iqdrops++;
2507                                 goto discard;
2508                         }
2509
2510                         mp->m_len = len;
2511                         if (rxr->fmp == NULL) {
2512                                 mp->m_pkthdr.len = len;
2513                                 rxr->fmp = mp;
2514                                 rxr->lmp = mp;
2515                         } else {
2516                                 rxr->lmp->m_next = mp;
2517                                 rxr->lmp = rxr->lmp->m_next;
2518                                 rxr->fmp->m_pkthdr.len += len;
2519                         }
2520
2521                         if (eop) {
2522                                 m = rxr->fmp;
2523                                 rxr->fmp = NULL;
2524                                 rxr->lmp = NULL;
2525
2526                                 m->m_pkthdr.rcvif = ifp;
2527                                 ifp->if_ipackets++;
2528
2529                                 if (ifp->if_capenable & IFCAP_RXCSUM)
2530                                         igb_rxcsum(staterr, m);
2531
2532                                 if (staterr & E1000_RXD_STAT_VP) {
2533                                         m->m_pkthdr.ether_vlantag = vlan;
2534                                         m->m_flags |= M_VLANTAG;
2535                                 }
2536
2537                                 if (ifp->if_capenable & IFCAP_RSS) {
2538                                         pi = igb_rssinfo(m, &pi0,
2539                                             hash, hashtype, staterr);
2540                                 }
2541 #ifdef IGB_RSS_DEBUG
2542                                 rxr->rx_packets++;
2543 #endif
2544                         }
2545                 } else {
2546                         ifp->if_ierrors++;
2547 discard:
2548                         igb_setup_rxdesc(cur, rxbuf);
2549                         if (!eop)
2550                                 rxr->discard = TRUE;
2551                         else
2552                                 rxr->discard = FALSE;
2553                         if (rxr->fmp != NULL) {
2554                                 m_freem(rxr->fmp);
2555                                 rxr->fmp = NULL;
2556                                 rxr->lmp = NULL;
2557                         }
2558                         m = NULL;
2559                 }
2560
2561                 if (m != NULL)
2562                         ether_input_pkt(ifp, m, pi);
2563
2564                 /* Advance our pointers to the next descriptor. */
2565                 if (++i == rxr->num_rx_desc)
2566                         i = 0;
2567
2568                 if (ncoll >= rxr->rx_wreg) {
2569                         igb_rx_refresh(rxr, i);
2570                         ncoll = 0;
2571                 }
2572
2573                 cur = &rxr->rx_base[i];
2574                 staterr = le32toh(cur->wb.upper.status_error);
2575         }
2576         rxr->next_to_check = i;
2577
2578         if (ncoll > 0)
2579                 igb_rx_refresh(rxr, i);
2580 }
2581
2582
2583 static void
2584 igb_set_vlan(struct igb_softc *sc)
2585 {
2586         struct e1000_hw *hw = &sc->hw;
2587         uint32_t reg;
2588 #if 0
2589         struct ifnet *ifp = sc->arpcom.ac_if;
2590 #endif
2591
2592         if (sc->vf_ifp) {
2593                 e1000_rlpml_set_vf(hw, sc->max_frame_size + VLAN_TAG_SIZE);
2594                 return;
2595         }
2596
2597         reg = E1000_READ_REG(hw, E1000_CTRL);
2598         reg |= E1000_CTRL_VME;
2599         E1000_WRITE_REG(hw, E1000_CTRL, reg);
2600
2601 #if 0
2602         /* Enable the Filter Table */
2603         if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2604                 reg = E1000_READ_REG(hw, E1000_RCTL);
2605                 reg &= ~E1000_RCTL_CFIEN;
2606                 reg |= E1000_RCTL_VFE;
2607                 E1000_WRITE_REG(hw, E1000_RCTL, reg);
2608         }
2609 #endif
2610
2611         /* Update the frame size */
2612         E1000_WRITE_REG(&sc->hw, E1000_RLPML,
2613             sc->max_frame_size + VLAN_TAG_SIZE);
2614
2615 #if 0
2616         /* Don't bother with table if no vlans */
2617         if ((adapter->num_vlans == 0) ||
2618             ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
2619                 return;
2620         /*
2621         ** A soft reset zero's out the VFTA, so
2622         ** we need to repopulate it now.
2623         */
2624         for (int i = 0; i < IGB_VFTA_SIZE; i++)
2625                 if (adapter->shadow_vfta[i] != 0) {
2626                         if (adapter->vf_ifp)
2627                                 e1000_vfta_set_vf(hw,
2628                                     adapter->shadow_vfta[i], TRUE);
2629                         else
2630                                 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
2631                                  i, adapter->shadow_vfta[i]);
2632                 }
2633 #endif
2634 }
2635
2636 static void
2637 igb_enable_intr(struct igb_softc *sc)
2638 {
2639         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2640                 lwkt_serialize_handler_enable(&sc->main_serialize);
2641         } else {
2642                 int i;
2643
2644                 for (i = 0; i < sc->msix_cnt; ++i) {
2645                         lwkt_serialize_handler_enable(
2646                             sc->msix_data[i].msix_serialize);
2647                 }
2648         }
2649
2650         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2651                 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
2652                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, sc->intr_mask);
2653                 else
2654                         E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2655                 E1000_WRITE_REG(&sc->hw, E1000_EIAM, sc->intr_mask);
2656                 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2657                 E1000_WRITE_REG(&sc->hw, E1000_IMS, E1000_IMS_LSC);
2658         } else {
2659                 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
2660         }
2661         E1000_WRITE_FLUSH(&sc->hw);
2662 }
2663
2664 static void
2665 igb_disable_intr(struct igb_softc *sc)
2666 {
2667         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2668                 E1000_WRITE_REG(&sc->hw, E1000_EIMC, 0xffffffff);
2669                 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2670         }
2671         E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2672         E1000_WRITE_FLUSH(&sc->hw);
2673
2674         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2675                 lwkt_serialize_handler_disable(&sc->main_serialize);
2676         } else {
2677                 int i;
2678
2679                 for (i = 0; i < sc->msix_cnt; ++i) {
2680                         lwkt_serialize_handler_disable(
2681                             sc->msix_data[i].msix_serialize);
2682                 }
2683         }
2684 }
2685
2686 /*
2687  * Bit of a misnomer, what this really means is
2688  * to enable OS management of the system... aka
2689  * to disable special hardware management features
2690  */
2691 static void
2692 igb_get_mgmt(struct igb_softc *sc)
2693 {
2694         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2695                 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2696                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2697
2698                 /* disable hardware interception of ARP */
2699                 manc &= ~E1000_MANC_ARP_EN;
2700
2701                 /* enable receiving management packets to the host */
2702                 manc |= E1000_MANC_EN_MNG2HOST;
2703                 manc2h |= 1 << 5; /* Mng Port 623 */
2704                 manc2h |= 1 << 6; /* Mng Port 664 */
2705                 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
2706                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2707         }
2708 }
2709
2710 /*
2711  * Give control back to hardware management controller
2712  * if there is one.
2713  */
2714 static void
2715 igb_rel_mgmt(struct igb_softc *sc)
2716 {
2717         if (sc->flags & IGB_FLAG_HAS_MGMT) {
2718                 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2719
2720                 /* Re-enable hardware interception of ARP */
2721                 manc |= E1000_MANC_ARP_EN;
2722                 manc &= ~E1000_MANC_EN_MNG2HOST;
2723
2724                 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2725         }
2726 }
2727
2728 /*
2729  * Sets CTRL_EXT:DRV_LOAD bit.
2730  *
2731  * For ASF and Pass Through versions of f/w this means that
2732  * the driver is loaded.
2733  */
2734 static void
2735 igb_get_hw_control(struct igb_softc *sc)
2736 {
2737         uint32_t ctrl_ext;
2738
2739         if (sc->vf_ifp)
2740                 return;
2741
2742         /* Let firmware know the driver has taken over */
2743         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2744         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2745             ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2746 }
2747
2748 /*
2749  * Resets CTRL_EXT:DRV_LOAD bit.
2750  *
2751  * For ASF and Pass Through versions of f/w this means that the
2752  * driver is no longer loaded.
2753  */
2754 static void
2755 igb_rel_hw_control(struct igb_softc *sc)
2756 {
2757         uint32_t ctrl_ext;
2758
2759         if (sc->vf_ifp)
2760                 return;
2761
2762         /* Let firmware taken over control of h/w */
2763         ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2764         E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2765             ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2766 }
2767
2768 static int
2769 igb_is_valid_ether_addr(const uint8_t *addr)
2770 {
2771         uint8_t zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
2772
2773         if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
2774                 return FALSE;
2775         return TRUE;
2776 }
2777
2778 /*
2779  * Enable PCI Wake On Lan capability
2780  */
2781 static void
2782 igb_enable_wol(device_t dev)
2783 {
2784         uint16_t cap, status;
2785         uint8_t id;
2786
2787         /* First find the capabilities pointer*/
2788         cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
2789
2790         /* Read the PM Capabilities */
2791         id = pci_read_config(dev, cap, 1);
2792         if (id != PCIY_PMG)     /* Something wrong */
2793                 return;
2794
2795         /*
2796          * OK, we have the power capabilities,
2797          * so now get the status register
2798          */
2799         cap += PCIR_POWER_STATUS;
2800         status = pci_read_config(dev, cap, 2);
2801         status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2802         pci_write_config(dev, cap, status, 2);
2803 }
2804
2805 static void
2806 igb_update_stats_counters(struct igb_softc *sc)
2807 {
2808         struct e1000_hw *hw = &sc->hw;
2809         struct e1000_hw_stats *stats;
2810         struct ifnet *ifp = &sc->arpcom.ac_if;
2811
2812         /*
2813          * The virtual function adapter has only a
2814          * small controlled set of stats, do only
2815          * those and return.
2816          */
2817         if (sc->vf_ifp) {
2818                 igb_update_vf_stats_counters(sc);
2819                 return;
2820         }
2821         stats = sc->stats;
2822
2823         if (sc->hw.phy.media_type == e1000_media_type_copper ||
2824             (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
2825                 stats->symerrs +=
2826                     E1000_READ_REG(hw,E1000_SYMERRS);
2827                 stats->sec += E1000_READ_REG(hw, E1000_SEC);
2828         }
2829
2830         stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
2831         stats->mpc += E1000_READ_REG(hw, E1000_MPC);
2832         stats->scc += E1000_READ_REG(hw, E1000_SCC);
2833         stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
2834
2835         stats->mcc += E1000_READ_REG(hw, E1000_MCC);
2836         stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
2837         stats->colc += E1000_READ_REG(hw, E1000_COLC);
2838         stats->dc += E1000_READ_REG(hw, E1000_DC);
2839         stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
2840         stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
2841         stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
2842
2843         /*
2844          * For watchdog management we need to know if we have been
2845          * paused during the last interval, so capture that here.
2846          */
2847         sc->pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
2848         stats->xoffrxc += sc->pause_frames;
2849         stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
2850         stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
2851         stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
2852         stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
2853         stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
2854         stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
2855         stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
2856         stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
2857         stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
2858         stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
2859         stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
2860         stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
2861
2862         /* For the 64-bit byte counters the low dword must be read first. */
2863         /* Both registers clear on the read of the high dword */
2864
2865         stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
2866             ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
2867         stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
2868             ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
2869
2870         stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
2871         stats->ruc += E1000_READ_REG(hw, E1000_RUC);
2872         stats->rfc += E1000_READ_REG(hw, E1000_RFC);
2873         stats->roc += E1000_READ_REG(hw, E1000_ROC);
2874         stats->rjc += E1000_READ_REG(hw, E1000_RJC);
2875
2876         stats->tor += E1000_READ_REG(hw, E1000_TORH);
2877         stats->tot += E1000_READ_REG(hw, E1000_TOTH);
2878
2879         stats->tpr += E1000_READ_REG(hw, E1000_TPR);
2880         stats->tpt += E1000_READ_REG(hw, E1000_TPT);
2881         stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
2882         stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
2883         stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
2884         stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
2885         stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
2886         stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
2887         stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
2888         stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
2889
2890         /* Interrupt Counts */
2891
2892         stats->iac += E1000_READ_REG(hw, E1000_IAC);
2893         stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
2894         stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
2895         stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
2896         stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
2897         stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
2898         stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
2899         stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
2900         stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
2901
2902         /* Host to Card Statistics */
2903
2904         stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
2905         stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
2906         stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
2907         stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
2908         stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
2909         stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
2910         stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
2911         stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
2912             ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32));
2913         stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
2914             ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
2915         stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
2916         stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
2917         stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
2918
2919         stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
2920         stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
2921         stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
2922         stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
2923         stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
2924         stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
2925
2926         ifp->if_collisions = stats->colc;
2927
2928         /* Rx Errors */
2929         ifp->if_ierrors = stats->rxerrc + stats->crcerrs + stats->algnerrc +
2930             stats->ruc + stats->roc + stats->mpc + stats->cexterr;
2931
2932         /* Tx Errors */
2933         ifp->if_oerrors = stats->ecol + stats->latecol + sc->watchdog_events;
2934
2935         /* Driver specific counters */
2936         sc->device_control = E1000_READ_REG(hw, E1000_CTRL);
2937         sc->rx_control = E1000_READ_REG(hw, E1000_RCTL);
2938         sc->int_mask = E1000_READ_REG(hw, E1000_IMS);
2939         sc->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
2940         sc->packet_buf_alloc_tx =
2941             ((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
2942         sc->packet_buf_alloc_rx =
2943             (E1000_READ_REG(hw, E1000_PBA) & 0xffff);
2944 }
2945
2946 static void
2947 igb_vf_init_stats(struct igb_softc *sc)
2948 {
2949         struct e1000_hw *hw = &sc->hw;
2950         struct e1000_vf_stats *stats;
2951
2952         stats = sc->stats;
2953         stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
2954         stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
2955         stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
2956         stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
2957         stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
2958 }
2959
2960 static void
2961 igb_update_vf_stats_counters(struct igb_softc *sc)
2962 {
2963         struct e1000_hw *hw = &sc->hw;
2964         struct e1000_vf_stats *stats;
2965
2966         if (sc->link_speed == 0)
2967                 return;
2968
2969         stats = sc->stats;
2970         UPDATE_VF_REG(E1000_VFGPRC, stats->last_gprc, stats->gprc);
2971         UPDATE_VF_REG(E1000_VFGORC, stats->last_gorc, stats->gorc);
2972         UPDATE_VF_REG(E1000_VFGPTC, stats->last_gptc, stats->gptc);
2973         UPDATE_VF_REG(E1000_VFGOTC, stats->last_gotc, stats->gotc);
2974         UPDATE_VF_REG(E1000_VFMPRC, stats->last_mprc, stats->mprc);
2975 }
2976
2977 #ifdef IFPOLL_ENABLE
2978
2979 static void
2980 igb_npoll_status(struct ifnet *ifp)
2981 {
2982         struct igb_softc *sc = ifp->if_softc;
2983         uint32_t reg_icr;
2984
2985         ASSERT_SERIALIZED(&sc->main_serialize);
2986
2987         reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2988         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
2989                 sc->hw.mac.get_link_status = 1;
2990                 igb_update_link_status(sc);
2991         }
2992 }
2993
2994 static void
2995 igb_npoll_tx(struct ifnet *ifp, void *arg, int cycle __unused)
2996 {
2997         struct igb_tx_ring *txr = arg;
2998
2999         ASSERT_SERIALIZED(&txr->tx_serialize);
3000
3001         igb_txeof(txr);
3002         if (!ifsq_is_empty(txr->ifsq))
3003                 ifsq_devstart(txr->ifsq);
3004 }
3005
3006 static void
3007 igb_npoll_rx(struct ifnet *ifp __unused, void *arg, int cycle)
3008 {
3009         struct igb_rx_ring *rxr = arg;
3010
3011         ASSERT_SERIALIZED(&rxr->rx_serialize);
3012
3013         igb_rxeof(rxr, cycle);
3014 }
3015
3016 static void
3017 igb_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3018 {
3019         struct igb_softc *sc = ifp->if_softc;
3020         int i;
3021
3022         ASSERT_IFNET_SERIALIZED_ALL(ifp);
3023
3024         if (info) {
3025                 int off;
3026
3027                 info->ifpi_status.status_func = igb_npoll_status;
3028                 info->ifpi_status.serializer = &sc->main_serialize;
3029
3030                 off = sc->tx_npoll_off;
3031                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3032                         struct igb_tx_ring *txr = &sc->tx_rings[i];
3033                         int idx = i + off;
3034
3035                         KKASSERT(idx < ncpus2);
3036                         info->ifpi_tx[idx].poll_func = igb_npoll_tx;
3037                         info->ifpi_tx[idx].arg = txr;
3038                         info->ifpi_tx[idx].serializer = &txr->tx_serialize;
3039                         ifsq_set_cpuid(txr->ifsq, idx);
3040                 }
3041
3042                 off = sc->rx_npoll_off;
3043                 for (i = 0; i < sc->rx_ring_cnt; ++i) {
3044                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
3045                         int idx = i + off;
3046
3047                         KKASSERT(idx < ncpus2);
3048                         info->ifpi_rx[idx].poll_func = igb_npoll_rx;
3049                         info->ifpi_rx[idx].arg = rxr;
3050                         info->ifpi_rx[idx].serializer = &rxr->rx_serialize;
3051                 }
3052
3053                 if (ifp->if_flags & IFF_RUNNING) {
3054                         if (sc->rx_ring_inuse == sc->rx_ring_cnt)
3055                                 igb_disable_intr(sc);
3056                         else
3057                                 igb_init(sc);
3058                 }
3059         } else {
3060                 if (ifp->if_flags & IFF_RUNNING) {
3061                         if (sc->rx_ring_inuse == sc->rx_ring_cnt)
3062                                 igb_enable_intr(sc);
3063                         else
3064                                 igb_init(sc);
3065                 }
3066
3067                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3068                         struct igb_tx_ring *txr = &sc->tx_rings[i];
3069
3070                         ifsq_set_cpuid(txr->ifsq, txr->tx_intr_cpuid);
3071                 }
3072         }
3073 }
3074
3075 #endif /* IFPOLL_ENABLE */
3076
3077 static void
3078 igb_intr(void *xsc)
3079 {
3080         struct igb_softc *sc = xsc;
3081         struct ifnet *ifp = &sc->arpcom.ac_if;
3082         uint32_t eicr;
3083
3084         ASSERT_SERIALIZED(&sc->main_serialize);
3085
3086         eicr = E1000_READ_REG(&sc->hw, E1000_EICR);
3087
3088         if (eicr == 0)
3089                 return;
3090
3091         if (ifp->if_flags & IFF_RUNNING) {
3092                 struct igb_tx_ring *txr = &sc->tx_rings[0];
3093                 int i;
3094
3095                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3096                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
3097
3098                         if (eicr & rxr->rx_intr_mask) {
3099                                 lwkt_serialize_enter(&rxr->rx_serialize);
3100                                 igb_rxeof(rxr, -1);
3101                                 lwkt_serialize_exit(&rxr->rx_serialize);
3102                         }
3103                 }
3104
3105                 if (eicr & txr->tx_intr_mask) {
3106                         lwkt_serialize_enter(&txr->tx_serialize);
3107                         igb_txeof(txr);
3108                         if (!ifsq_is_empty(txr->ifsq))
3109                                 ifsq_devstart(txr->ifsq);
3110                         lwkt_serialize_exit(&txr->tx_serialize);
3111                 }
3112         }
3113
3114         if (eicr & E1000_EICR_OTHER) {
3115                 uint32_t icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3116
3117                 /* Link status change */
3118                 if (icr & E1000_ICR_LSC) {
3119                         sc->hw.mac.get_link_status = 1;
3120                         igb_update_link_status(sc);
3121                 }
3122         }
3123
3124         /*
3125          * Reading EICR has the side effect to clear interrupt mask,
3126          * so all interrupts need to be enabled here.
3127          */
3128         E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
3129 }
3130
3131 static void
3132 igb_intr_shared(void *xsc)
3133 {
3134         struct igb_softc *sc = xsc;
3135         struct ifnet *ifp = &sc->arpcom.ac_if;
3136         uint32_t reg_icr;
3137
3138         ASSERT_SERIALIZED(&sc->main_serialize);
3139
3140         reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3141
3142         /* Hot eject?  */
3143         if (reg_icr == 0xffffffff)
3144                 return;
3145
3146         /* Definitely not our interrupt.  */
3147         if (reg_icr == 0x0)
3148                 return;
3149
3150         if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
3151                 return;
3152
3153         if (ifp->if_flags & IFF_RUNNING) {
3154                 if (reg_icr &
3155                     (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
3156                         int i;
3157
3158                         for (i = 0; i < sc->rx_ring_inuse; ++i) {
3159                                 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3160
3161                                 lwkt_serialize_enter(&rxr->rx_serialize);
3162                                 igb_rxeof(rxr, -1);
3163                                 lwkt_serialize_exit(&rxr->rx_serialize);
3164                         }
3165                 }
3166
3167                 if (reg_icr & E1000_ICR_TXDW) {
3168                         struct igb_tx_ring *txr = &sc->tx_rings[0];
3169
3170                         lwkt_serialize_enter(&txr->tx_serialize);
3171                         igb_txeof(txr);
3172                         if (!ifsq_is_empty(txr->ifsq))
3173                                 ifsq_devstart(txr->ifsq);
3174                         lwkt_serialize_exit(&txr->tx_serialize);
3175                 }
3176         }
3177
3178         /* Link status change */
3179         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3180                 sc->hw.mac.get_link_status = 1;
3181                 igb_update_link_status(sc);
3182         }
3183
3184         if (reg_icr & E1000_ICR_RXO)
3185                 sc->rx_overruns++;
3186 }
3187
3188 static int
3189 igb_encap(struct igb_tx_ring *txr, struct mbuf **m_headp,
3190     int *segs_used, int *idx)
3191 {
3192         bus_dma_segment_t segs[IGB_MAX_SCATTER];
3193         bus_dmamap_t map;
3194         struct igb_tx_buf *tx_buf, *tx_buf_mapped;
3195         union e1000_adv_tx_desc *txd = NULL;
3196         struct mbuf *m_head = *m_headp;
3197         uint32_t olinfo_status = 0, cmd_type_len = 0, cmd_rs = 0;
3198         int maxsegs, nsegs, i, j, error;
3199         uint32_t hdrlen = 0;
3200
3201         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3202                 error = igb_tso_pullup(txr, m_headp);
3203                 if (error)
3204                         return error;
3205                 m_head = *m_headp;
3206         }
3207
3208         /* Set basic descriptor constants */
3209         cmd_type_len |= E1000_ADVTXD_DTYP_DATA;
3210         cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
3211         if (m_head->m_flags & M_VLANTAG)
3212                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3213
3214         /*
3215          * Map the packet for DMA.
3216          */
3217         tx_buf = &txr->tx_buf[txr->next_avail_desc];
3218         tx_buf_mapped = tx_buf;
3219         map = tx_buf->map;
3220
3221         maxsegs = txr->tx_avail - IGB_TX_RESERVED;
3222         KASSERT(maxsegs >= txr->spare_desc, ("not enough spare TX desc\n"));
3223         if (maxsegs > IGB_MAX_SCATTER)
3224                 maxsegs = IGB_MAX_SCATTER;
3225
3226         error = bus_dmamap_load_mbuf_defrag(txr->tx_tag, map, m_headp,
3227             segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3228         if (error) {
3229                 if (error == ENOBUFS)
3230                         txr->sc->mbuf_defrag_failed++;
3231                 else
3232                         txr->sc->no_tx_dma_setup++;
3233
3234                 m_freem(*m_headp);
3235                 *m_headp = NULL;
3236                 return error;
3237         }
3238         bus_dmamap_sync(txr->tx_tag, map, BUS_DMASYNC_PREWRITE);
3239
3240         m_head = *m_headp;
3241
3242         /*
3243          * Set up the TX context descriptor, if any hardware offloading is
3244          * needed.  This includes CSUM, VLAN, and TSO.  It will consume one
3245          * TX descriptor.
3246          *
3247          * Unlike these chips' predecessors (em/emx), TX context descriptor
3248          * will _not_ interfere TX data fetching pipelining.
3249          */
3250         if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3251                 igb_tso_ctx(txr, m_head, &hdrlen);
3252                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3253                 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3254                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3255                 txr->tx_nsegs++;
3256                 (*segs_used)++;
3257         } else if (igb_txcsum_ctx(txr, m_head)) {
3258                 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3259                         olinfo_status |= (E1000_TXD_POPTS_IXSM << 8);
3260                 if (m_head->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_TCP))
3261                         olinfo_status |= (E1000_TXD_POPTS_TXSM << 8);
3262                 txr->tx_nsegs++;
3263                 (*segs_used)++;
3264         }
3265
3266         *segs_used += nsegs;
3267         txr->tx_nsegs += nsegs;
3268         if (txr->tx_nsegs >= txr->intr_nsegs) {
3269                 /*
3270                  * Report Status (RS) is turned on every intr_nsegs
3271                  * descriptors (roughly).
3272                  */
3273                 txr->tx_nsegs = 0;
3274                 cmd_rs = E1000_ADVTXD_DCMD_RS;
3275         }
3276
3277         /* Calculate payload length */
3278         olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
3279             << E1000_ADVTXD_PAYLEN_SHIFT);
3280
3281         /* 82575 needs the queue index added */
3282         if (txr->sc->hw.mac.type == e1000_82575)
3283                 olinfo_status |= txr->me << 4;
3284
3285         /* Set up our transmit descriptors */
3286         i = txr->next_avail_desc;
3287         for (j = 0; j < nsegs; j++) {
3288                 bus_size_t seg_len;
3289                 bus_addr_t seg_addr;
3290
3291                 tx_buf = &txr->tx_buf[i];
3292                 txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
3293                 seg_addr = segs[j].ds_addr;
3294                 seg_len = segs[j].ds_len;
3295
3296                 txd->read.buffer_addr = htole64(seg_addr);
3297                 txd->read.cmd_type_len = htole32(cmd_type_len | seg_len);
3298                 txd->read.olinfo_status = htole32(olinfo_status);
3299                 if (++i == txr->num_tx_desc)
3300                         i = 0;
3301                 tx_buf->m_head = NULL;
3302         }
3303
3304         KASSERT(txr->tx_avail > nsegs, ("invalid avail TX desc\n"));
3305         txr->next_avail_desc = i;
3306         txr->tx_avail -= nsegs;
3307
3308         tx_buf->m_head = m_head;
3309         tx_buf_mapped->map = tx_buf->map;
3310         tx_buf->map = map;
3311
3312         /*
3313          * Last Descriptor of Packet needs End Of Packet (EOP)
3314          */
3315         txd->read.cmd_type_len |= htole32(E1000_ADVTXD_DCMD_EOP | cmd_rs);
3316
3317         /*
3318          * Defer TDT updating, until enough descrptors are setup
3319          */
3320         *idx = i;
3321         ++txr->tx_packets;
3322
3323         return 0;
3324 }
3325
3326 static void
3327 igb_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
3328 {
3329         struct igb_softc *sc = ifp->if_softc;
3330         struct igb_tx_ring *txr = ifsq_get_priv(ifsq);
3331         struct mbuf *m_head;
3332         int idx = -1, nsegs = 0;
3333
3334         KKASSERT(txr->ifsq == ifsq);
3335         ASSERT_SERIALIZED(&txr->tx_serialize);
3336
3337         if ((ifp->if_flags & IFF_RUNNING) == 0 || ifsq_is_oactive(ifsq))
3338                 return;
3339
3340         if (!sc->link_active) {
3341                 ifsq_purge(ifsq);
3342                 return;
3343         }
3344
3345         if (!IGB_IS_NOT_OACTIVE(txr))
3346                 igb_txeof(txr);
3347
3348         while (!ifsq_is_empty(ifsq)) {
3349                 if (IGB_IS_OACTIVE(txr)) {
3350                         ifsq_set_oactive(ifsq);
3351                         /* Set watchdog on */
3352                         ifp->if_timer = 5;
3353                         break;
3354                 }
3355
3356                 m_head = ifsq_dequeue(ifsq, NULL);
3357                 if (m_head == NULL)
3358                         break;
3359
3360                 if (igb_encap(txr, &m_head, &nsegs, &idx)) {
3361                         ifp->if_oerrors++;
3362                         continue;
3363                 }
3364
3365                 if (nsegs >= txr->wreg_nsegs) {
3366                         E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3367                         idx = -1;
3368                         nsegs = 0;
3369                 }
3370
3371                 /* Send a copy of the frame to the BPF listener */
3372                 ETHER_BPF_MTAP(ifp, m_head);
3373         }
3374         if (idx >= 0)
3375                 E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3376 }
3377
3378 static void
3379 igb_watchdog(struct ifnet *ifp)
3380 {
3381         struct igb_softc *sc = ifp->if_softc;
3382         struct igb_tx_ring *txr = &sc->tx_rings[0];
3383
3384         ASSERT_IFNET_SERIALIZED_ALL(ifp);
3385
3386         /*
3387          * If flow control has paused us since last checking
3388          * it invalidates the watchdog timing, so dont run it.
3389          */
3390         if (sc->pause_frames) {
3391                 sc->pause_frames = 0;
3392                 ifp->if_timer = 5;
3393                 return;
3394         }
3395
3396         if_printf(ifp, "Watchdog timeout -- resetting\n");
3397         if_printf(ifp, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
3398             E1000_READ_REG(&sc->hw, E1000_TDH(txr->me)),
3399             E1000_READ_REG(&sc->hw, E1000_TDT(txr->me)));
3400         if_printf(ifp, "TX(%d) desc avail = %d, "
3401             "Next TX to Clean = %d\n",
3402             txr->me, txr->tx_avail, txr->next_to_clean);
3403
3404         ifp->if_oerrors++;
3405         sc->watchdog_events++;
3406
3407         igb_init(sc);
3408         if (!ifsq_is_empty(txr->ifsq))
3409                 ifsq_devstart(txr->ifsq);
3410 }
3411
3412 static void
3413 igb_set_eitr(struct igb_softc *sc, int idx, int rate)
3414 {
3415         uint32_t eitr = 0;
3416
3417         if (rate > 0) {
3418                 if (sc->hw.mac.type == e1000_82575) {
3419                         eitr = 1000000000 / 256 / rate;
3420                         /*
3421                          * NOTE:
3422                          * Document is wrong on the 2 bits left shift
3423                          */
3424                 } else {
3425                         eitr = 1000000 / rate;
3426                         eitr <<= IGB_EITR_INTVL_SHIFT;
3427                 }
3428
3429                 if (eitr == 0) {
3430                         /* Don't disable it */
3431                         eitr = 1 << IGB_EITR_INTVL_SHIFT;
3432                 } else if (eitr > IGB_EITR_INTVL_MASK) {
3433                         /* Don't allow it to be too large */
3434                         eitr = IGB_EITR_INTVL_MASK;
3435                 }
3436         }
3437         if (sc->hw.mac.type == e1000_82575)
3438                 eitr |= eitr << 16;
3439         else
3440                 eitr |= E1000_EITR_CNT_IGNR;
3441         E1000_WRITE_REG(&sc->hw, E1000_EITR(idx), eitr);
3442 }
3443
3444 static int
3445 igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS)
3446 {
3447         struct igb_softc *sc = (void *)arg1;
3448         struct ifnet *ifp = &sc->arpcom.ac_if;
3449         int error, intr_rate;
3450
3451         intr_rate = sc->intr_rate;
3452         error = sysctl_handle_int(oidp, &intr_rate, 0, req);
3453         if (error || req->newptr == NULL)
3454                 return error;
3455         if (intr_rate < 0)
3456                 return EINVAL;
3457
3458         ifnet_serialize_all(ifp);
3459
3460         sc->intr_rate = intr_rate;
3461         if (ifp->if_flags & IFF_RUNNING)
3462                 igb_set_eitr(sc, 0, sc->intr_rate);
3463
3464         if (bootverbose)
3465                 if_printf(ifp, "interrupt rate set to %d/sec\n", sc->intr_rate);
3466
3467         ifnet_deserialize_all(ifp);
3468
3469         return 0;
3470 }
3471
3472 static int
3473 igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS)
3474 {
3475         struct igb_msix_data *msix = (void *)arg1;
3476         struct igb_softc *sc = msix->msix_sc;
3477         struct ifnet *ifp = &sc->arpcom.ac_if;
3478         int error, msix_rate;
3479
3480         msix_rate = msix->msix_rate;
3481         error = sysctl_handle_int(oidp, &msix_rate, 0, req);
3482         if (error || req->newptr == NULL)
3483                 return error;
3484         if (msix_rate < 0)
3485                 return EINVAL;
3486
3487         lwkt_serialize_enter(msix->msix_serialize);
3488
3489         msix->msix_rate = msix_rate;
3490         if (ifp->if_flags & IFF_RUNNING)
3491                 igb_set_eitr(sc, msix->msix_vector, msix->msix_rate);
3492
3493         if (bootverbose) {
3494                 if_printf(ifp, "%s set to %d/sec\n", msix->msix_rate_desc,
3495                     msix->msix_rate);
3496         }
3497
3498         lwkt_serialize_exit(msix->msix_serialize);
3499
3500         return 0;
3501 }
3502
3503 static int
3504 igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS)
3505 {
3506         struct igb_softc *sc = (void *)arg1;
3507         struct ifnet *ifp = &sc->arpcom.ac_if;
3508         struct igb_tx_ring *txr = &sc->tx_rings[0];
3509         int error, nsegs;
3510
3511         nsegs = txr->intr_nsegs;
3512         error = sysctl_handle_int(oidp, &nsegs, 0, req);
3513         if (error || req->newptr == NULL)
3514                 return error;
3515         if (nsegs <= 0)
3516                 return EINVAL;
3517
3518         ifnet_serialize_all(ifp);
3519
3520         if (nsegs >= txr->num_tx_desc - txr->oact_lo_desc ||
3521             nsegs >= txr->oact_hi_desc - IGB_MAX_SCATTER) {
3522                 error = EINVAL;
3523         } else {
3524                 error = 0;
3525                 txr->intr_nsegs = nsegs;
3526         }
3527
3528         ifnet_deserialize_all(ifp);
3529
3530         return error;
3531 }
3532
3533 #ifdef IFPOLL_ENABLE
3534
3535 static int
3536 igb_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS)
3537 {
3538         struct igb_softc *sc = (void *)arg1;
3539         struct ifnet *ifp = &sc->arpcom.ac_if;
3540         int error, off;
3541
3542         off = sc->rx_npoll_off;
3543         error = sysctl_handle_int(oidp, &off, 0, req);
3544         if (error || req->newptr == NULL)
3545                 return error;
3546         if (off < 0)
3547                 return EINVAL;
3548
3549         ifnet_serialize_all(ifp);
3550         if (off >= ncpus2 || off % sc->rx_ring_cnt != 0) {
3551                 error = EINVAL;
3552         } else {
3553                 error = 0;
3554                 sc->rx_npoll_off = off;
3555         }
3556         ifnet_deserialize_all(ifp);
3557
3558         return error;
3559 }
3560
3561 static int
3562 igb_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS)
3563 {
3564         struct igb_softc *sc = (void *)arg1;
3565         struct ifnet *ifp = &sc->arpcom.ac_if;
3566         int error, off;
3567
3568         off = sc->tx_npoll_off;
3569         error = sysctl_handle_int(oidp, &off, 0, req);
3570         if (error || req->newptr == NULL)
3571                 return error;
3572         if (off < 0)
3573                 return EINVAL;
3574
3575         ifnet_serialize_all(ifp);
3576         if (off >= ncpus2) {
3577                 error = EINVAL;
3578         } else {
3579                 error = 0;
3580                 sc->tx_npoll_off = off;
3581         }
3582         ifnet_deserialize_all(ifp);
3583
3584         return error;
3585 }
3586
3587 #endif  /* IFPOLL_ENABLE */
3588
3589 static void
3590 igb_init_intr(struct igb_softc *sc)
3591 {
3592         igb_set_intr_mask(sc);
3593
3594         if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0)
3595                 igb_init_unshared_intr(sc);
3596
3597         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
3598                 igb_set_eitr(sc, 0, sc->intr_rate);
3599         } else {
3600                 int i;
3601
3602                 for (i = 0; i < sc->msix_cnt; ++i)
3603                         igb_set_eitr(sc, i, sc->msix_data[i].msix_rate);
3604         }
3605 }
3606
3607 static void
3608 igb_init_unshared_intr(struct igb_softc *sc)
3609 {
3610         struct e1000_hw *hw = &sc->hw;
3611         const struct igb_rx_ring *rxr;
3612         const struct igb_tx_ring *txr;
3613         uint32_t ivar, index;
3614         int i;
3615
3616         /*
3617          * Enable extended mode
3618          */
3619         if (sc->hw.mac.type != e1000_82575) {
3620                 uint32_t gpie;
3621                 int ivar_max;
3622
3623                 gpie = E1000_GPIE_NSICR;
3624                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3625                         gpie |= E1000_GPIE_MSIX_MODE |
3626                             E1000_GPIE_EIAME |
3627                             E1000_GPIE_PBA;
3628                 }
3629                 E1000_WRITE_REG(hw, E1000_GPIE, gpie);
3630
3631                 /*
3632                  * Clear IVARs
3633                  */
3634                 switch (sc->hw.mac.type) {
3635                 case e1000_82580:
3636                         ivar_max = IGB_MAX_IVAR_82580;
3637                         break;
3638
3639                 case e1000_i350:
3640                         ivar_max = IGB_MAX_IVAR_I350;
3641                         break;
3642
3643                 case e1000_vfadapt:
3644                 case e1000_vfadapt_i350:
3645                         ivar_max = IGB_MAX_IVAR_VF;
3646                         break;
3647
3648                 case e1000_82576:
3649                         ivar_max = IGB_MAX_IVAR_82576;
3650                         break;
3651
3652                 default:
3653                         panic("unknown mac type %d\n", sc->hw.mac.type);
3654                 }
3655                 for (i = 0; i < ivar_max; ++i)
3656                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, i, 0);
3657                 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
3658         } else {
3659                 uint32_t tmp;
3660
3661                 KASSERT(sc->intr_type != PCI_INTR_TYPE_MSIX,
3662                     ("82575 w/ MSI-X"));
3663                 tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
3664                 tmp |= E1000_CTRL_EXT_IRCA;
3665                 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
3666         }
3667
3668         /*
3669          * Map TX/RX interrupts to EICR
3670          */
3671         switch (sc->hw.mac.type) {
3672         case e1000_82580:
3673         case e1000_i350:
3674         case e1000_vfadapt:
3675         case e1000_vfadapt_i350:
3676                 /* RX entries */
3677                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3678                         rxr = &sc->rx_rings[i];
3679
3680                         index = i >> 1;
3681                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3682
3683                         if (i & 1) {
3684                                 ivar &= 0xff00ffff;
3685                                 ivar |=
3686                                 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3687                         } else {
3688                                 ivar &= 0xffffff00;
3689                                 ivar |=
3690                                 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3691                         }
3692                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3693                 }
3694                 /* TX entries */
3695                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3696                         txr = &sc->tx_rings[i];
3697
3698                         index = i >> 1;
3699                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3700
3701                         if (i & 1) {
3702                                 ivar &= 0x00ffffff;
3703                                 ivar |=
3704                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3705                         } else {
3706                                 ivar &= 0xffff00ff;
3707                                 ivar |=
3708                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3709                         }
3710                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3711                 }
3712                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3713                         ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3714                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3715                 }
3716                 break;
3717
3718         case e1000_82576:
3719                 /* RX entries */
3720                 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3721                         rxr = &sc->rx_rings[i];
3722
3723                         index = i & 0x7; /* Each IVAR has two entries */
3724                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3725
3726                         if (i < 8) {
3727                                 ivar &= 0xffffff00;
3728                                 ivar |=
3729                                 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3730                         } else {
3731                                 ivar &= 0xff00ffff;
3732                                 ivar |=
3733                                 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3734                         }
3735                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3736                 }
3737                 /* TX entries */
3738                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3739                         txr = &sc->tx_rings[i];
3740
3741                         index = i & 0x7; /* Each IVAR has two entries */
3742                         ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3743
3744                         if (i < 8) {
3745                                 ivar &= 0xffff00ff;
3746                                 ivar |=
3747                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3748                         } else {
3749                                 ivar &= 0x00ffffff;
3750                                 ivar |=
3751                                 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3752                         }
3753                         E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3754                 }
3755                 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3756                         ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3757                         E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3758                 }
3759                 break;
3760
3761         case e1000_82575:
3762                 /*
3763                  * Enable necessary interrupt bits.
3764                  *
3765                  * The name of the register is confusing; in addition to
3766                  * configuring the first vector of MSI-X, it also configures
3767                  * which bits of EICR could be set by the hardware even when
3768                  * MSI or line interrupt is used; it thus controls interrupt
3769                  * generation.  It MUST be configured explicitly; the default
3770                  * value mentioned in the datasheet is wrong: RX queue0 and
3771                  * TX queue0 are NOT enabled by default.
3772                  */
3773                 E1000_WRITE_REG(&sc->hw, E1000_MSIXBM(0), sc->intr_mask);
3774                 break;
3775
3776         default:
3777                 panic("unknown mac type %d\n", sc->hw.mac.type);
3778         }
3779 }
3780
3781 static int
3782 igb_setup_intr(struct igb_softc *sc)
3783 {
3784         int error;
3785
3786         if (sc->intr_type == PCI_INTR_TYPE_MSIX)
3787                 return igb_msix_setup(sc);
3788
3789         error = bus_setup_intr(sc->dev, sc->intr_res, INTR_MPSAFE,
3790             (sc->flags & IGB_FLAG_SHARED_INTR) ? igb_intr_shared : igb_intr,
3791             sc, &sc->intr_tag, &sc->main_serialize);
3792         if (error) {
3793                 device_printf(sc->dev, "Failed to register interrupt handler");
3794                 return error;
3795         }
3796         sc->tx_rings[0].tx_intr_cpuid = rman_get_cpuid(sc->intr_res);
3797
3798         return 0;
3799 }
3800
3801 static void
3802 igb_set_txintr_mask(struct igb_tx_ring *txr, int *intr_bit0, int intr_bitmax)
3803 {
3804         if (txr->sc->hw.mac.type == e1000_82575) {
3805                 txr->tx_intr_bit = 0;   /* unused */
3806                 switch (txr->me) {
3807                 case 0:
3808                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE0;
3809                         break;
3810                 case 1:
3811                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE1;
3812                         break;
3813                 case 2:
3814                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE2;
3815                         break;
3816                 case 3:
3817                         txr->tx_intr_mask = E1000_EICR_TX_QUEUE3;
3818                         break;
3819                 default:
3820                         panic("unsupported # of TX ring, %d\n", txr->me);
3821                 }
3822         } else {
3823                 int intr_bit = *intr_bit0;
3824
3825                 txr->tx_intr_bit = intr_bit % intr_bitmax;
3826                 txr->tx_intr_mask = 1 << txr->tx_intr_bit;
3827
3828                 *intr_bit0 = intr_bit + 1;
3829         }
3830 }
3831
3832 static void
3833 igb_set_rxintr_mask(struct igb_rx_ring *rxr, int *intr_bit0, int intr_bitmax)
3834 {
3835         if (rxr->sc->hw.mac.type == e1000_82575) {
3836                 rxr->rx_intr_bit = 0;   /* unused */
3837                 switch (rxr->me) {
3838                 case 0:
3839                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE0;
3840                         break;
3841                 case 1:
3842                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE1;
3843                         break;
3844                 case 2:
3845                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE2;
3846                         break;
3847                 case 3:
3848                         rxr->rx_intr_mask = E1000_EICR_RX_QUEUE3;
3849                         break;
3850                 default:
3851                         panic("unsupported # of RX ring, %d\n", rxr->me);
3852                 }
3853         } else {
3854                 int intr_bit = *intr_bit0;
3855
3856                 rxr->rx_intr_bit = intr_bit % intr_bitmax;
3857                 rxr->rx_intr_mask = 1 << rxr->rx_intr_bit;
3858
3859                 *intr_bit0 = intr_bit + 1;
3860         }
3861 }
3862
3863 static void
3864 igb_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3865 {
3866         struct igb_softc *sc = ifp->if_softc;
3867
3868         ifnet_serialize_array_enter(sc->serializes, sc->serialize_cnt,
3869             sc->tx_serialize, sc->rx_serialize, slz);
3870 }
3871
3872 static void
3873 igb_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3874 {
3875         struct igb_softc *sc = ifp->if_softc;
3876
3877         ifnet_serialize_array_exit(sc->serializes, sc->serialize_cnt,
3878             sc->tx_serialize, sc->rx_serialize, slz);
3879 }
3880
3881 static int
3882 igb_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3883 {
3884         struct igb_softc *sc = ifp->if_softc;
3885
3886         return ifnet_serialize_array_try(sc->serializes, sc->serialize_cnt,
3887             sc->tx_serialize, sc->rx_serialize, slz);
3888 }
3889
3890 #ifdef INVARIANTS
3891
3892 static void
3893 igb_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3894     boolean_t serialized)
3895 {
3896         struct igb_softc *sc = ifp->if_softc;
3897
3898         ifnet_serialize_array_assert(sc->serializes, sc->serialize_cnt,
3899             sc->tx_serialize, sc->rx_serialize, slz, serialized);
3900 }
3901
3902 #endif  /* INVARIANTS */
3903
3904 static void
3905 igb_set_intr_mask(struct igb_softc *sc)
3906 {
3907         int i;
3908
3909         sc->intr_mask = sc->sts_intr_mask;
3910         for (i = 0; i < sc->rx_ring_inuse; ++i)
3911                 sc->intr_mask |= sc->rx_rings[i].rx_intr_mask;
3912         for (i = 0; i < sc->tx_ring_cnt; ++i)
3913                 sc->intr_mask |= sc->tx_rings[i].tx_intr_mask;
3914         if (bootverbose) {
3915                 if_printf(&sc->arpcom.ac_if, "intr mask 0x%08x\n",
3916                     sc->intr_mask);
3917         }
3918 }
3919
3920 static int
3921 igb_alloc_intr(struct igb_softc *sc)
3922 {
3923         int i, intr_bit, intr_bitmax;
3924         u_int intr_flags;
3925
3926         igb_msix_try_alloc(sc);
3927         if (sc->intr_type == PCI_INTR_TYPE_MSIX)
3928                 goto done;
3929
3930         /*
3931          * Allocate MSI/legacy interrupt resource
3932          */
3933         sc->intr_type = pci_alloc_1intr(sc->dev, igb_msi_enable,
3934             &sc->intr_rid, &intr_flags);
3935
3936         if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
3937                 int unshared;
3938
3939                 unshared = device_getenv_int(sc->dev, "irq.unshared", 0);
3940                 if (!unshared) {
3941                         sc->flags |= IGB_FLAG_SHARED_INTR;
3942                         if (bootverbose)
3943                                 device_printf(sc->dev, "IRQ shared\n");
3944                 } else {
3945                         intr_flags &= ~RF_SHAREABLE;
3946                         if (bootverbose)
3947                                 device_printf(sc->dev, "IRQ unshared\n");
3948                 }
3949         }
3950
3951         sc->intr_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
3952             &sc->intr_rid, intr_flags);
3953         if (sc->intr_res == NULL) {
3954                 device_printf(sc->dev, "Unable to allocate bus resource: "
3955                     "interrupt\n");
3956                 return ENXIO;
3957         }
3958
3959         /*
3960          * Setup MSI/legacy interrupt mask
3961          */
3962         switch (sc->hw.mac.type) {
3963         case e1000_82575:
3964                 intr_bitmax = IGB_MAX_TXRXINT_82575;
3965                 break;
3966         case e1000_82580:
3967                 intr_bitmax = IGB_MAX_TXRXINT_82580;
3968                 break;
3969         case e1000_i350:
3970                 intr_bitmax = IGB_MAX_TXRXINT_I350;
3971                 break;
3972         case e1000_82576:
3973                 intr_bitmax = IGB_MAX_TXRXINT_82576;
3974                 break;
3975         default:
3976                 intr_bitmax = IGB_MIN_TXRXINT;
3977                 break;
3978         }
3979         intr_bit = 0;
3980         for (i = 0; i < sc->tx_ring_cnt; ++i)
3981                 igb_set_txintr_mask(&sc->tx_rings[i], &intr_bit, intr_bitmax);
3982         for (i = 0; i < sc->rx_ring_cnt; ++i)
3983                 igb_set_rxintr_mask(&sc->rx_rings[i], &intr_bit, intr_bitmax);
3984         sc->sts_intr_bit = 0;
3985         sc->sts_intr_mask = E1000_EICR_OTHER;
3986
3987         /* Initialize interrupt rate */
3988         sc->intr_rate = IGB_INTR_RATE;
3989 done:
3990         igb_set_ring_inuse(sc, FALSE);
3991         igb_set_intr_mask(sc);
3992         return 0;
3993 }
3994
3995 static void
3996 igb_free_intr(struct igb_softc *sc)
3997 {
3998         if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
3999                 if (sc->intr_res != NULL) {
4000                         bus_release_resource(sc->dev, SYS_RES_IRQ, sc->intr_rid,
4001                             sc->intr_res);
4002                 }
4003                 if (sc->intr_type == PCI_INTR_TYPE_MSI)
4004                         pci_release_msi(sc->dev);
4005         } else {
4006                 igb_msix_free(sc, TRUE);
4007         }
4008 }
4009
4010 static void
4011 igb_teardown_intr(struct igb_softc *sc)
4012 {
4013         if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4014                 bus_teardown_intr(sc->dev, sc->intr_res, sc->intr_tag);
4015         else
4016                 igb_msix_teardown(sc, sc->msix_cnt);
4017 }
4018
4019 static void
4020 igb_msix_try_alloc(struct igb_softc *sc)
4021 {
4022         int msix_enable, msix_cnt, msix_cnt2, alloc_cnt;
4023         int i, x, error;
4024         struct igb_msix_data *msix;
4025         boolean_t aggregate, setup = FALSE;
4026
4027         /*
4028          * Don't enable MSI-X on 82575, see:
4029          * 82575 specification update errata #25
4030          */
4031         if (sc->hw.mac.type == e1000_82575)
4032                 return;
4033
4034         /* Don't enable MSI-X on VF */
4035         if (sc->vf_ifp)
4036                 return;
4037
4038         msix_enable = device_getenv_int(sc->dev, "msix.enable",
4039             igb_msix_enable);
4040         if (!msix_enable)
4041                 return;
4042
4043         msix_cnt = pci_msix_count(sc->dev);
4044 #ifdef IGB_MSIX_DEBUG
4045         msix_cnt = device_getenv_int(sc->dev, "msix.count", msix_cnt);
4046 #endif
4047         if (msix_cnt <= 1) {
4048                 /* One MSI-X model does not make sense */
4049                 return;
4050         }
4051
4052         i = 0;
4053         while ((1 << (i + 1)) <= msix_cnt)
4054                 ++i;
4055         msix_cnt2 = 1 << i;
4056
4057         if (bootverbose) {
4058                 device_printf(sc->dev, "MSI-X count %d/%d\n",
4059                     msix_cnt2, msix_cnt);
4060         }
4061
4062         KKASSERT(msix_cnt2 <= msix_cnt);
4063         if (msix_cnt == msix_cnt2) {
4064                 /* We need at least one MSI-X for link status */
4065                 msix_cnt2 >>= 1;
4066                 if (msix_cnt2 <= 1) {
4067                         /* One MSI-X for RX/TX does not make sense */
4068                         device_printf(sc->dev, "not enough MSI-X for TX/RX, "
4069                             "MSI-X count %d/%d\n", msix_cnt2, msix_cnt);
4070                         return;
4071                 }
4072                 KKASSERT(msix_cnt > msix_cnt2);
4073
4074                 if (bootverbose) {
4075                         device_printf(sc->dev, "MSI-X count fixup %d/%d\n",
4076                             msix_cnt2, msix_cnt);
4077                 }
4078         }
4079
4080         sc->rx_ring_msix = sc->rx_ring_cnt;
4081         if (sc->rx_ring_msix > msix_cnt2)
4082                 sc->rx_ring_msix = msix_cnt2;
4083
4084         if (msix_cnt >= sc->tx_ring_cnt + sc->rx_ring_msix + 1) {
4085                 /*
4086                  * Independent TX/RX MSI-X
4087                  */
4088                 aggregate = FALSE;
4089                 if (bootverbose)
4090                         device_printf(sc->dev, "independent TX/RX MSI-X\n");
4091                 alloc_cnt = sc->tx_ring_cnt + sc->rx_ring_msix;
4092         } else {
4093                 /*
4094                  * Aggregate TX/RX MSI-X
4095                  */
4096                 aggregate = TRUE;
4097                 if (bootverbose)
4098                         device_printf(sc->dev, "aggregate TX/RX MSI-X\n");
4099                 alloc_cnt = msix_cnt2;
4100                 if (alloc_cnt > ncpus2)
4101                         alloc_cnt = ncpus2;
4102                 if (sc->rx_ring_msix > alloc_cnt)
4103                         sc->rx_ring_msix = alloc_cnt;
4104         }
4105         ++alloc_cnt;    /* For link status */
4106
4107         if (bootverbose) {
4108                 device_printf(sc->dev, "MSI-X alloc %d, RX ring %d\n",
4109                     alloc_cnt, sc->rx_ring_msix);
4110         }
4111
4112         sc->msix_mem_rid = PCIR_BAR(IGB_MSIX_BAR);
4113         sc->msix_mem_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
4114             &sc->msix_mem_rid, RF_ACTIVE);
4115         if (sc->msix_mem_res == NULL) {
4116                 device_printf(sc->dev, "Unable to map MSI-X table\n");
4117                 return;
4118         }
4119
4120         sc->msix_cnt = alloc_cnt;
4121         sc->msix_data = kmalloc_cachealign(
4122             sizeof(struct igb_msix_data) * sc->msix_cnt,
4123             M_DEVBUF, M_WAITOK | M_ZERO);
4124         for (x = 0; x < sc->msix_cnt; ++x) {
4125                 msix = &sc->msix_data[x];
4126
4127                 lwkt_serialize_init(&msix->msix_serialize0);
4128                 msix->msix_sc = sc;
4129                 msix->msix_rid = -1;
4130                 msix->msix_vector = x;
4131                 msix->msix_mask = 1 << msix->msix_vector;
4132                 msix->msix_rate = IGB_INTR_RATE;
4133         }
4134
4135         x = 0;
4136         if (!aggregate) {
4137                 int offset, offset_def;
4138
4139                 if (sc->rx_ring_msix == ncpus2) {
4140                         offset = 0;
4141                 } else {
4142                         offset_def = (sc->rx_ring_msix *
4143                             device_get_unit(sc->dev)) % ncpus2;
4144
4145                         offset = device_getenv_int(sc->dev,
4146                             "msix.rxoff", offset_def);
4147                         if (offset >= ncpus2 ||
4148                             offset % sc->rx_ring_msix != 0) {
4149                                 device_printf(sc->dev,
4150                                     "invalid msix.rxoff %d, use %d\n",
4151                                     offset, offset_def);
4152                                 offset = offset_def;
4153                         }
4154                 }
4155
4156                 /* RX rings */
4157                 for (i = 0; i < sc->rx_ring_msix; ++i) {
4158                         struct igb_rx_ring *rxr = &sc->rx_rings[i];
4159
4160                         KKASSERT(x < sc->msix_cnt);
4161                         msix = &sc->msix_data[x++];
4162                         rxr->rx_intr_bit = msix->msix_vector;
4163                         rxr->rx_intr_mask = msix->msix_mask;
4164
4165                         msix->msix_serialize = &rxr->rx_serialize;
4166                         msix->msix_func = igb_msix_rx;
4167                         msix->msix_arg = rxr;
4168                         msix->msix_cpuid = i + offset;
4169                         KKASSERT(msix->msix_cpuid < ncpus2);
4170                         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
4171                             "%s rx%d", device_get_nameunit(sc->dev), i);
4172                         msix->msix_rate = IGB_MSIX_RX_RATE;
4173                         ksnprintf(msix->msix_rate_desc,
4174                             sizeof(msix->msix_rate_desc),
4175                             "RX%d interrupt rate", i);
4176                 }
4177
4178                 offset_def = device_get_unit(sc->dev) % ncpus2;
4179                 offset = device_getenv_int(sc->dev, "msix.txoff", offset_def);
4180                 if (offset >= ncpus2) {
4181                         device_printf(sc->dev, "invalid msix.txoff %d, "
4182                             "use %d\n", offset, offset_def);
4183                         offset = offset_def;
4184                 }
4185
4186                 /* TX rings */
4187                 for (i = 0; i < sc->tx_ring_cnt; ++i) {
4188                         struct igb_tx_ring *txr = &sc->tx_rings[i];
4189
4190                         KKASSERT(x < sc->msix_cnt);
4191                         msix = &sc->msix_data[x++];
4192                         txr->tx_intr_bit = msix->msix_vector;
4193                         txr->tx_intr_mask = msix->msix_mask;
4194
4195                         msix->msix_serialize = &txr->tx_serialize;
4196                         msix->msix_func = igb_msix_tx;
4197                         msix->msix_arg = txr;
4198                         msix->msix_cpuid = i + offset;
4199                         txr->tx_intr_cpuid = msix->msix_cpuid;
4200                         KKASSERT(msix->msix_cpuid < ncpus2);
4201                         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
4202                             "%s tx%d", device_get_nameunit(sc->dev), i);
4203                         msix->msix_rate = IGB_MSIX_TX_RATE;
4204                         ksnprintf(msix->msix_rate_desc,
4205                             sizeof(msix->msix_rate_desc),
4206                             "TX%d interrupt rate", i);
4207                 }
4208         } else {
4209                 /* TODO */
4210                 error = EOPNOTSUPP;
4211                 goto back;
4212         }
4213
4214         /*
4215          * Link status
4216          */
4217         KKASSERT(x < sc->msix_cnt);
4218         msix = &sc->msix_data[x++];
4219         sc->sts_intr_bit = msix->msix_vector;
4220         sc->sts_intr_mask = msix->msix_mask;
4221
4222         msix->msix_serialize = &sc->main_serialize;
4223         msix->msix_func = igb_msix_status;
4224         msix->msix_arg = sc;
4225         msix->msix_cpuid = 0; /* TODO tunable */
4226         ksnprintf(msix->msix_desc, sizeof(msix->msix_desc), "%s sts",
4227             device_get_nameunit(sc->dev));
4228         ksnprintf(msix->msix_rate_desc, sizeof(msix->msix_rate_desc),
4229             "status interrupt rate");
4230
4231         KKASSERT(x == sc->msix_cnt);
4232
4233         error = pci_setup_msix(sc->dev);
4234         if (error) {
4235                 device_printf(sc->dev, "Setup MSI-X failed\n");
4236                 goto back;
4237         }
4238         setup = TRUE;
4239
4240         for (i = 0; i < sc->msix_cnt; ++i) {
4241                 msix = &sc->msix_data[i];
4242
4243                 error = pci_alloc_msix_vector(sc->dev, msix->msix_vector,
4244                     &msix->msix_rid, msix->msix_cpuid);
4245                 if (error) {
4246                         device_printf(sc->dev,
4247                             "Unable to allocate MSI-X %d on cpu%d\n",
4248                             msix->msix_vector, msix->msix_cpuid);
4249                         goto back;
4250                 }
4251
4252                 msix->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4253                     &msix->msix_rid, RF_ACTIVE);
4254                 if (msix->msix_res == NULL) {
4255                         device_printf(sc->dev,
4256                             "Unable to allocate MSI-X %d resource\n",
4257                             msix->msix_vector);
4258                         error = ENOMEM;
4259                         goto back;
4260                 }
4261         }
4262
4263         pci_enable_msix(sc->dev);
4264         sc->intr_type = PCI_INTR_TYPE_MSIX;
4265 back:
4266         if (error)
4267                 igb_msix_free(sc, setup);
4268 }
4269
4270 static void
4271 igb_msix_free(struct igb_softc *sc, boolean_t setup)
4272 {
4273         int i;
4274
4275         KKASSERT(sc->msix_cnt > 1);
4276
4277         for (i = 0; i < sc->msix_cnt; ++i) {
4278                 struct igb_msix_data *msix = &sc->msix_data[i];
4279
4280                 if (msix->msix_res != NULL) {
4281                         bus_release_resource(sc->dev, SYS_RES_IRQ,
4282                             msix->msix_rid, msix->msix_res);
4283                 }
4284                 if (msix->msix_rid >= 0)
4285                         pci_release_msix_vector(sc->dev, msix->msix_rid);
4286         }
4287         if (setup)
4288                 pci_teardown_msix(sc->dev);
4289
4290         sc->msix_cnt = 0;
4291         kfree(sc->msix_data, M_DEVBUF);
4292         sc->msix_data = NULL;
4293 }
4294
4295 static int
4296 igb_msix_setup(struct igb_softc *sc)
4297 {
4298         int i;
4299
4300         for (i = 0; i < sc->msix_cnt; ++i) {
4301                 struct igb_msix_data *msix = &sc->msix_data[i];
4302                 int error;
4303
4304                 error = bus_setup_intr_descr(sc->dev, msix->msix_res,
4305                     INTR_MPSAFE, msix->msix_func, msix->msix_arg,
4306                     &msix->msix_handle, msix->msix_serialize, msix->msix_desc);
4307                 if (error) {
4308                         device_printf(sc->dev, "could not set up %s "
4309                             "interrupt handler.\n", msix->msix_desc);
4310                         igb_msix_teardown(sc, i);
4311                         return error;
4312                 }
4313         }
4314         return 0;
4315 }
4316
4317 static void
4318 igb_msix_teardown(struct igb_softc *sc, int msix_cnt)
4319 {
4320         int i;
4321
4322         for (i = 0; i < msix_cnt; ++i) {
4323                 struct igb_msix_data *msix = &sc->msix_data[i];
4324
4325                 bus_teardown_intr(sc->dev, msix->msix_res, msix->msix_handle);
4326         }
4327 }
4328
4329 static void
4330 igb_msix_rx(void *arg)
4331 {
4332         struct igb_rx_ring *rxr = arg;
4333
4334         ASSERT_SERIALIZED(&rxr->rx_serialize);
4335         igb_rxeof(rxr, -1);
4336
4337         E1000_WRITE_REG(&rxr->sc->hw, E1000_EIMS, rxr->rx_intr_mask);
4338 }
4339
4340 static void
4341 igb_msix_tx(void *arg)
4342 {
4343         struct igb_tx_ring *txr = arg;
4344
4345         ASSERT_SERIALIZED(&txr->tx_serialize);
4346
4347         igb_txeof(txr);
4348         if (!ifsq_is_empty(txr->ifsq))
4349                 ifsq_devstart(txr->ifsq);
4350
4351         E1000_WRITE_REG(&txr->sc->hw, E1000_EIMS, txr->tx_intr_mask);
4352 }
4353
4354 static void
4355 igb_msix_status(void *arg)
4356 {
4357         struct igb_softc *sc = arg;
4358         uint32_t icr;
4359
4360         ASSERT_SERIALIZED(&sc->main_serialize);
4361
4362         icr = E1000_READ_REG(&sc->hw, E1000_ICR);
4363         if (icr & E1000_ICR_LSC) {
4364                 sc->hw.mac.get_link_status = 1;
4365                 igb_update_link_status(sc);
4366         }
4367
4368         E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->sts_intr_mask);
4369 }
4370
4371 static void
4372 igb_set_ring_inuse(struct igb_softc *sc, boolean_t polling)
4373 {
4374         if (!IGB_ENABLE_HWRSS(sc))
4375                 return;
4376
4377         if (polling)
4378                 sc->rx_ring_inuse = sc->rx_ring_cnt;
4379         else if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4380                 sc->rx_ring_inuse = IGB_MIN_RING_RSS;
4381         else
4382                 sc->rx_ring_inuse = sc->rx_ring_msix;
4383         if (bootverbose) {
4384                 if_printf(&sc->arpcom.ac_if, "RX rings %d/%d\n",
4385                     sc->rx_ring_inuse, sc->rx_ring_cnt);
4386         }
4387 }
4388
4389 static int
4390 igb_tso_pullup(struct igb_tx_ring *txr, struct mbuf **mp)
4391 {
4392         int hoff, iphlen, thoff;
4393         struct mbuf *m;
4394
4395         m = *mp;
4396         KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4397
4398         iphlen = m->m_pkthdr.csum_iphlen;
4399         thoff = m->m_pkthdr.csum_thlen;
4400         hoff = m->m_pkthdr.csum_lhlen;
4401
4402         KASSERT(iphlen > 0, ("invalid ip hlen"));
4403         KASSERT(thoff > 0, ("invalid tcp hlen"));
4404         KASSERT(hoff > 0, ("invalid ether hlen"));
4405
4406         if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4407                 m = m_pullup(m, hoff + iphlen + thoff);
4408                 if (m == NULL) {
4409                         *mp = NULL;
4410                         return ENOBUFS;
4411                 }
4412                 *mp = m;
4413         }
4414         if (txr->sc->flags & IGB_FLAG_TSO_IPLEN0) {
4415                 struct ip *ip;
4416
4417                 ip = mtodoff(m, struct ip *, hoff);
4418                 ip->ip_len = 0;
4419         }
4420
4421         return 0;
4422 }
4423
4424 static void
4425 igb_tso_ctx(struct igb_tx_ring *txr, struct mbuf *m, uint32_t *hlen)
4426 {
4427         struct e1000_adv_tx_context_desc *TXD;
4428         uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
4429         int hoff, ctxd, iphlen, thoff;
4430
4431         iphlen = m->m_pkthdr.csum_iphlen;
4432         thoff = m->m_pkthdr.csum_thlen;
4433         hoff = m->m_pkthdr.csum_lhlen;
4434
4435         vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
4436
4437         ctxd = txr->next_avail_desc;
4438         TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
4439
4440         if (m->m_flags & M_VLANTAG) {
4441                 uint16_t vlantag;
4442
4443                 vlantag = htole16(m->m_pkthdr.ether_vlantag);
4444                 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
4445         }
4446
4447         vlan_macip_lens |= (hoff << E1000_ADVTXD_MACLEN_SHIFT);
4448         vlan_macip_lens |= iphlen;
4449
4450         type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4451         type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
4452         type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
4453
4454         mss_l4len_idx |= (m->m_pkthdr.tso_segsz << E1000_ADVTXD_MSS_SHIFT);
4455         mss_l4len_idx |= (thoff << E1000_ADVTXD_L4LEN_SHIFT);
4456         /* 82575 needs the queue index added */
4457         if (txr->sc->hw.mac.type == e1000_82575)
4458                 mss_l4len_idx |= txr->me << 4;
4459
4460         TXD->vlan_macip_lens = htole32(vlan_macip_lens);
4461         TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
4462         TXD->seqnum_seed = htole32(0);
4463         TXD->mss_l4len_idx = htole32(mss_l4len_idx);
4464
4465         /* We've consumed the first desc, adjust counters */
4466         if (++ctxd == txr->num_tx_desc)
4467                 ctxd = 0;
4468         txr->next_avail_desc = ctxd;
4469         --txr->tx_avail;
4470
4471         *hlen = hoff + iphlen + thoff;
4472 }