sys/dev/netif/em/if_em.c

   1 /*
   2  * Copyright (c) 2004 Joerg Sonnenberger <joerg@bec.de>.  All rights reserved.
   3  *
   4  * Copyright (c) 2001-2008, Intel Corporation
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions are met:
   9  *
  10  *  1. Redistributions of source code must retain the above copyright notice,
  11  *     this list of conditions and the following disclaimer.
  12  *
  13  *  2. Redistributions in binary form must reproduce the above copyright
  14  *     notice, this list of conditions and the following disclaimer in the
  15  *     documentation and/or other materials provided with the distribution.
  16  *
  17  *  3. Neither the name of the Intel Corporation nor the names of its
  18  *     contributors may be used to endorse or promote products derived from
  19  *     this software without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  22  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  25  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  *
  34  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
  35  *
  36  * This code is derived from software contributed to The DragonFly Project
  37  * by Matthew Dillon <dillon@backplane.com>
  38  *
  39  * Redistribution and use in source and binary forms, with or without
  40  * modification, are permitted provided that the following conditions
  41  * are met:
  42  *
  43  * 1. Redistributions of source code must retain the above copyright
  44  *    notice, this list of conditions and the following disclaimer.
  45  * 2. Redistributions in binary form must reproduce the above copyright
  46  *    notice, this list of conditions and the following disclaimer in
  47  *    the documentation and/or other materials provided with the
  48  *    distribution.
  49  * 3. Neither the name of The DragonFly Project nor the names of its
  50  *    contributors may be used to endorse or promote products derived
  51  *    from this software without specific, prior written permission.
  52  *
  53  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  54  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  55  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  56  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  57  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  58  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  59  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  60  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  61  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  62  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  63  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  64  * SUCH DAMAGE.
  65  *
  66  */
  67 /*
  68  * SERIALIZATION API RULES:
  69  *
  70  * - If the driver uses the same serializer for the interrupt as for the
  71  *   ifnet, most of the serialization will be done automatically for the
  72  *   driver.
  73  *
  74  * - ifmedia entry points will be serialized by the ifmedia code using the
  75  *   ifnet serializer.
  76  *
  77  * - if_* entry points except for if_input will be serialized by the IF
  78  *   and protocol layers.
  79  *
  80  * - The device driver must be sure to serialize access from timeout code
  81  *   installed by the device driver.
  82  *
  83  * - The device driver typically holds the serializer at the time it wishes
  84  *   to call if_input.
  85  *
  86  * - We must call lwkt_serialize_handler_enable() prior to enabling the
  87  *   hardware interrupt and lwkt_serialize_handler_disable() after disabling
  88  *   the hardware interrupt in order to avoid handler execution races from
  89  *   scheduled interrupt threads.
  90  *
  91  *   NOTE!  Since callers into the device driver hold the ifnet serializer,
  92  *   the device driver may be holding a serializer at the time it calls
  93  *   if_input even if it is not serializer-aware.
  94  */
  95
  96 #include "opt_polling.h"
  97
  98 #include <sys/param.h>
  99 #include <sys/bus.h>
 100 #include <sys/endian.h>
 101 #include <sys/interrupt.h>
 102 #include <sys/kernel.h>
 103 #include <sys/ktr.h>
 104 #include <sys/malloc.h>
 105 #include <sys/mbuf.h>
 106 #include <sys/proc.h>
 107 #include <sys/rman.h>
 108 #include <sys/serialize.h>
 109 #include <sys/socket.h>
 110 #include <sys/sockio.h>
 111 #include <sys/sysctl.h>
 112 #include <sys/systm.h>
 113
 114 #include <net/bpf.h>
 115 #include <net/ethernet.h>
 116 #include <net/if.h>
 117 #include <net/if_arp.h>
 118 #include <net/if_dl.h>
 119 #include <net/if_media.h>
 120 #include <net/ifq_var.h>
 121 #include <net/vlan/if_vlan_var.h>
 122 #include <net/vlan/if_vlan_ether.h>
 123
 124 #include <netinet/in_systm.h>
 125 #include <netinet/in.h>
 126 #include <netinet/ip.h>
 127 #include <netinet/tcp.h>
 128 #include <netinet/udp.h>
 129
 130 #include <bus/pci/pcivar.h>
 131 #include <bus/pci/pcireg.h>
 132
 133 #include <dev/netif/ig_hal/e1000_api.h>
 134 #include <dev/netif/ig_hal/e1000_82571.h>
 135 #include <dev/netif/em/if_em.h>
 136
 137 #define EM_NAME "Intel(R) PRO/1000 Network Connection "
 138 #define EM_VER  " 6.9.6"
 139
 140 #define _EM_DEVICE(id, ret)     \
 141         { EM_VENDOR_ID, E1000_DEV_ID_##id, ret, EM_NAME #id EM_VER }
 142 #define EM_EMX_DEVICE(id)       _EM_DEVICE(id, -100)
 143 #define EM_DEVICE(id)           _EM_DEVICE(id, 0)
 144 #define EM_DEVICE_NULL  { 0, 0, 0, NULL }
 145
 146 static const struct em_vendor_info em_vendor_info_array[] = {
 147         EM_DEVICE(82540EM),
 148         EM_DEVICE(82540EM_LOM),
 149         EM_DEVICE(82540EP),
 150         EM_DEVICE(82540EP_LOM),
 151         EM_DEVICE(82540EP_LP),
 152
 153         EM_DEVICE(82541EI),
 154         EM_DEVICE(82541ER),
 155         EM_DEVICE(82541ER_LOM),
 156         EM_DEVICE(82541EI_MOBILE),
 157         EM_DEVICE(82541GI),
 158         EM_DEVICE(82541GI_LF),
 159         EM_DEVICE(82541GI_MOBILE),
 160
 161         EM_DEVICE(82542),
 162
 163         EM_DEVICE(82543GC_FIBER),
 164         EM_DEVICE(82543GC_COPPER),
 165
 166         EM_DEVICE(82544EI_COPPER),
 167         EM_DEVICE(82544EI_FIBER),
 168         EM_DEVICE(82544GC_COPPER),
 169         EM_DEVICE(82544GC_LOM),
 170
 171         EM_DEVICE(82545EM_COPPER),
 172         EM_DEVICE(82545EM_FIBER),
 173         EM_DEVICE(82545GM_COPPER),
 174         EM_DEVICE(82545GM_FIBER),
 175         EM_DEVICE(82545GM_SERDES),
 176
 177         EM_DEVICE(82546EB_COPPER),
 178         EM_DEVICE(82546EB_FIBER),
 179         EM_DEVICE(82546EB_QUAD_COPPER),
 180         EM_DEVICE(82546GB_COPPER),
 181         EM_DEVICE(82546GB_FIBER),
 182         EM_DEVICE(82546GB_SERDES),
 183         EM_DEVICE(82546GB_PCIE),
 184         EM_DEVICE(82546GB_QUAD_COPPER),
 185         EM_DEVICE(82546GB_QUAD_COPPER_KSP3),
 186
 187         EM_DEVICE(82547EI),
 188         EM_DEVICE(82547EI_MOBILE),
 189         EM_DEVICE(82547GI),
 190
 191         EM_EMX_DEVICE(82571EB_COPPER),
 192         EM_EMX_DEVICE(82571EB_FIBER),
 193         EM_EMX_DEVICE(82571EB_SERDES),
 194         EM_EMX_DEVICE(82571EB_SERDES_DUAL),
 195         EM_EMX_DEVICE(82571EB_SERDES_QUAD),
 196         EM_EMX_DEVICE(82571EB_QUAD_COPPER),
 197         EM_EMX_DEVICE(82571EB_QUAD_COPPER_BP),
 198         EM_EMX_DEVICE(82571EB_QUAD_COPPER_LP),
 199         EM_EMX_DEVICE(82571EB_QUAD_FIBER),
 200         EM_EMX_DEVICE(82571PT_QUAD_COPPER),
 201
 202         EM_EMX_DEVICE(82572EI_COPPER),
 203         EM_EMX_DEVICE(82572EI_FIBER),
 204         EM_EMX_DEVICE(82572EI_SERDES),
 205         EM_EMX_DEVICE(82572EI),
 206
 207         EM_EMX_DEVICE(82573E),
 208         EM_EMX_DEVICE(82573E_IAMT),
 209         EM_EMX_DEVICE(82573L),
 210
 211         EM_EMX_DEVICE(80003ES2LAN_COPPER_SPT),
 212         EM_EMX_DEVICE(80003ES2LAN_SERDES_SPT),
 213         EM_EMX_DEVICE(80003ES2LAN_COPPER_DPT),
 214         EM_EMX_DEVICE(80003ES2LAN_SERDES_DPT),
 215
 216         EM_DEVICE(ICH8_IGP_M_AMT),
 217         EM_DEVICE(ICH8_IGP_AMT),
 218         EM_DEVICE(ICH8_IGP_C),
 219         EM_DEVICE(ICH8_IFE),
 220         EM_DEVICE(ICH8_IFE_GT),
 221         EM_DEVICE(ICH8_IFE_G),
 222         EM_DEVICE(ICH8_IGP_M),
 223
 224         EM_DEVICE(ICH9_IGP_M_AMT),
 225         EM_DEVICE(ICH9_IGP_AMT),
 226         EM_DEVICE(ICH9_IGP_C),
 227         EM_DEVICE(ICH9_IGP_M),
 228         EM_DEVICE(ICH9_IGP_M_V),
 229         EM_DEVICE(ICH9_IFE),
 230         EM_DEVICE(ICH9_IFE_GT),
 231         EM_DEVICE(ICH9_IFE_G),
 232         EM_DEVICE(ICH9_BM),
 233
 234         EM_EMX_DEVICE(82574L),
 235
 236         EM_DEVICE(ICH10_R_BM_LM),
 237         EM_DEVICE(ICH10_R_BM_LF),
 238         EM_DEVICE(ICH10_R_BM_V),
 239         EM_DEVICE(ICH10_D_BM_LM),
 240         EM_DEVICE(ICH10_D_BM_LF),
 241
 242         /* required last entry */
 243         EM_DEVICE_NULL
 244 };
 245
 246 static int      em_probe(device_t);
 247 static int      em_attach(device_t);
 248 static int      em_detach(device_t);
 249 static int      em_shutdown(device_t);
 250 static int      em_suspend(device_t);
 251 static int      em_resume(device_t);
 252
 253 static void     em_init(void *);
 254 static void     em_stop(struct adapter *);
 255 static int      em_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 256 static void     em_start(struct ifnet *);
 257 #ifdef DEVICE_POLLING
 258 static void     em_poll(struct ifnet *, enum poll_cmd, int);
 259 #endif
 260 static void     em_watchdog(struct ifnet *);
 261 static void     em_media_status(struct ifnet *, struct ifmediareq *);
 262 static int      em_media_change(struct ifnet *);
 263 static void     em_timer(void *);
 264
 265 static void     em_intr(void *);
 266 static void     em_rxeof(struct adapter *, int);
 267 static void     em_txeof(struct adapter *);
 268 static void     em_tx_collect(struct adapter *);
 269 static void     em_tx_purge(struct adapter *);
 270 static void     em_enable_intr(struct adapter *);
 271 static void     em_disable_intr(struct adapter *);
 272
 273 static int      em_dma_malloc(struct adapter *, bus_size_t,
 274                     struct em_dma_alloc *);
 275 static void     em_dma_free(struct adapter *, struct em_dma_alloc *);
 276 static void     em_init_tx_ring(struct adapter *);
 277 static int      em_init_rx_ring(struct adapter *);
 278 static int      em_create_tx_ring(struct adapter *);
 279 static int      em_create_rx_ring(struct adapter *);
 280 static void     em_destroy_tx_ring(struct adapter *, int);
 281 static void     em_destroy_rx_ring(struct adapter *, int);
 282 static int      em_newbuf(struct adapter *, int, int);
 283 static int      em_encap(struct adapter *, struct mbuf **);
 284 static void     em_rxcsum(struct adapter *, struct e1000_rx_desc *,
 285                     struct mbuf *);
 286 static int      em_txcsum_pullup(struct adapter *, struct mbuf **);
 287 static int      em_txcsum(struct adapter *, struct mbuf *,
 288                     uint32_t *, uint32_t *);
 289
 290 static int      em_get_hw_info(struct adapter *);
 291 static int      em_is_valid_eaddr(const uint8_t *);
 292 static int      em_alloc_pci_res(struct adapter *);
 293 static void     em_free_pci_res(struct adapter *);
 294 static int      em_hw_init(struct adapter *);
 295 static void     em_setup_ifp(struct adapter *);
 296 static void     em_init_tx_unit(struct adapter *);
 297 static void     em_init_rx_unit(struct adapter *);
 298 static void     em_update_stats(struct adapter *);
 299 static void     em_set_promisc(struct adapter *);
 300 static void     em_disable_promisc(struct adapter *);
 301 static void     em_set_multi(struct adapter *);
 302 static void     em_update_link_status(struct adapter *);
 303 static void     em_smartspeed(struct adapter *);
 304
 305 /* Hardware workarounds */
 306 static int      em_82547_fifo_workaround(struct adapter *, int);
 307 static void     em_82547_update_fifo_head(struct adapter *, int);
 308 static int      em_82547_tx_fifo_reset(struct adapter *);
 309 static void     em_82547_move_tail(void *);
 310 static void     em_82547_move_tail_serialized(struct adapter *);
 311 static uint32_t em_82544_fill_desc(bus_addr_t, uint32_t, PDESC_ARRAY);
 312
 313 static void     em_print_debug_info(struct adapter *);
 314 static void     em_print_nvm_info(struct adapter *);
 315 static void     em_print_hw_stats(struct adapter *);
 316
 317 static int      em_sysctl_stats(SYSCTL_HANDLER_ARGS);
 318 static int      em_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
 319 static int      em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS);
 320 static int      em_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS);
 321 static void     em_add_sysctl(struct adapter *adapter);
 322
 323 /* Management and WOL Support */
 324 static void     em_get_mgmt(struct adapter *);
 325 static void     em_rel_mgmt(struct adapter *);
 326 static void     em_get_hw_control(struct adapter *);
 327 static void     em_rel_hw_control(struct adapter *);
 328 static void     em_enable_wol(device_t);
 329
 330 static device_method_t em_methods[] = {
 331         /* Device interface */
 332         DEVMETHOD(device_probe,         em_probe),
 333         DEVMETHOD(device_attach,        em_attach),
 334         DEVMETHOD(device_detach,        em_detach),
 335         DEVMETHOD(device_shutdown,      em_shutdown),
 336         DEVMETHOD(device_suspend,       em_suspend),
 337         DEVMETHOD(device_resume,        em_resume),
 338         { 0, 0 }
 339 };
 340
 341 static driver_t em_driver = {
 342         "em",
 343         em_methods,
 344         sizeof(struct adapter),
 345 };
 346
 347 static devclass_t em_devclass;
 348
 349 DECLARE_DUMMY_MODULE(if_em);
 350 MODULE_DEPEND(em, ig_hal, 1, 1, 1);
 351 DRIVER_MODULE(if_em, pci, em_driver, em_devclass, NULL, NULL);
 352
 353 /*
 354  * Tunables
 355  */
 356 static int      em_int_throttle_ceil = EM_DEFAULT_ITR;
 357 static int      em_rxd = EM_DEFAULT_RXD;
 358 static int      em_txd = EM_DEFAULT_TXD;
 359 static int      em_smart_pwr_down = FALSE;
 360
 361 /* Controls whether promiscuous also shows bad packets */
 362 static int      em_debug_sbp = FALSE;
 363
 364 static int      em_82573_workaround = TRUE;
 365
 366 TUNABLE_INT("hw.em.int_throttle_ceil", &em_int_throttle_ceil);
 367 TUNABLE_INT("hw.em.rxd", &em_rxd);
 368 TUNABLE_INT("hw.em.txd", &em_txd);
 369 TUNABLE_INT("hw.em.smart_pwr_down", &em_smart_pwr_down);
 370 TUNABLE_INT("hw.em.sbp", &em_debug_sbp);
 371 TUNABLE_INT("hw.em.82573_workaround", &em_82573_workaround);
 372
 373 /* Global used in WOL setup with multiport cards */
 374 static int      em_global_quad_port_a = 0;
 375
 376 /* Set this to one to display debug statistics */
 377 static int      em_display_debug_stats = 0;
 378
 379 #if !defined(KTR_IF_EM)
 380 #define KTR_IF_EM       KTR_ALL
 381 #endif
 382 KTR_INFO_MASTER(if_em);
 383 KTR_INFO(KTR_IF_EM, if_em, intr_beg, 0, "intr begin", 0);
 384 KTR_INFO(KTR_IF_EM, if_em, intr_end, 1, "intr end", 0);
 385 KTR_INFO(KTR_IF_EM, if_em, pkt_receive, 4, "rx packet", 0);
 386 KTR_INFO(KTR_IF_EM, if_em, pkt_txqueue, 5, "tx packet", 0);
 387 KTR_INFO(KTR_IF_EM, if_em, pkt_txclean, 6, "tx clean", 0);
 388 #define logif(name)     KTR_LOG(if_em_ ## name)
 389
 390 static int
 391 em_probe(device_t dev)
 392 {
 393         const struct em_vendor_info *ent;
 394         uint16_t vid, did;
 395
 396         vid = pci_get_vendor(dev);
 397         did = pci_get_device(dev);
 398
 399         for (ent = em_vendor_info_array; ent->desc != NULL; ++ent) {
 400                 if (vid == ent->vendor_id && did == ent->device_id) {
 401                         device_set_desc(dev, ent->desc);
 402                         device_set_async_attach(dev, TRUE);
 403                         return (ent->ret);
 404                 }
 405         }
 406         return (ENXIO);
 407 }
 408
 409 static int
 410 em_attach(device_t dev)
 411 {
 412         struct adapter *adapter = device_get_softc(dev);
 413         struct ifnet *ifp = &adapter->arpcom.ac_if;
 414         int tsize, rsize;
 415         int error = 0;
 416         uint16_t eeprom_data, device_id;
 417
 418         adapter->dev = adapter->osdep.dev = dev;
 419
 420         callout_init(&adapter->timer);
 421         callout_init(&adapter->tx_fifo_timer);
 422
 423         /* Determine hardware and mac info */
 424         error = em_get_hw_info(adapter);
 425         if (error) {
 426                 device_printf(dev, "Identify hardware failed\n");
 427                 goto fail;
 428         }
 429
 430         /* Setup PCI resources */
 431         error = em_alloc_pci_res(adapter);
 432         if (error) {
 433                 device_printf(dev, "Allocation of PCI resources failed\n");
 434                 goto fail;
 435         }
 436
 437         /*
 438          * For ICH8 and family we need to map the flash memory,
 439          * and this must happen after the MAC is identified.
 440          */
 441         if (adapter->hw.mac.type == e1000_ich8lan ||
 442             adapter->hw.mac.type == e1000_ich10lan ||
 443             adapter->hw.mac.type == e1000_ich9lan) {
 444                 adapter->flash_rid = EM_BAR_FLASH;
 445
 446                 adapter->flash = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
 447                                         &adapter->flash_rid, RF_ACTIVE);
 448                 if (adapter->flash == NULL) {
 449                         device_printf(dev, "Mapping of Flash failed\n");
 450                         error = ENXIO;
 451                         goto fail;
 452                 }
 453                 adapter->osdep.flash_bus_space_tag =
 454                     rman_get_bustag(adapter->flash);
 455                 adapter->osdep.flash_bus_space_handle =
 456                     rman_get_bushandle(adapter->flash);
 457
 458                 /*
 459                  * This is used in the shared code
 460                  * XXX this goof is actually not used.
 461                  */
 462                 adapter->hw.flash_address = (uint8_t *)adapter->flash;
 463         }
 464
 465         /* Do Shared Code initialization */
 466         if (e1000_setup_init_funcs(&adapter->hw, TRUE)) {
 467                 device_printf(dev, "Setup of Shared code failed\n");
 468                 error = ENXIO;
 469                 goto fail;
 470         }
 471
 472         e1000_get_bus_info(&adapter->hw);
 473
 474         /*
 475          * Validate number of transmit and receive descriptors.  It
 476          * must not exceed hardware maximum, and must be multiple
 477          * of E1000_DBA_ALIGN.
 478          */
 479         if ((em_txd * sizeof(struct e1000_tx_desc)) % EM_DBA_ALIGN != 0 ||
 480             (adapter->hw.mac.type >= e1000_82544 && em_txd > EM_MAX_TXD) ||
 481             (adapter->hw.mac.type < e1000_82544 && em_txd > EM_MAX_TXD_82543) ||
 482             em_txd < EM_MIN_TXD) {
 483                 device_printf(dev, "Using %d TX descriptors instead of %d!\n",
 484                     EM_DEFAULT_TXD, em_txd);
 485                 adapter->num_tx_desc = EM_DEFAULT_TXD;
 486         } else {
 487                 adapter->num_tx_desc = em_txd;
 488         }
 489         if ((em_rxd * sizeof(struct e1000_rx_desc)) % EM_DBA_ALIGN != 0 ||
 490             (adapter->hw.mac.type >= e1000_82544 && em_rxd > EM_MAX_RXD) ||
 491             (adapter->hw.mac.type < e1000_82544 && em_rxd > EM_MAX_RXD_82543) ||
 492             em_rxd < EM_MIN_RXD) {
 493                 device_printf(dev, "Using %d RX descriptors instead of %d!\n",
 494                     EM_DEFAULT_RXD, em_rxd);
 495                 adapter->num_rx_desc = EM_DEFAULT_RXD;
 496         } else {
 497                 adapter->num_rx_desc = em_rxd;
 498         }
 499
 500         adapter->hw.mac.autoneg = DO_AUTO_NEG;
 501         adapter->hw.phy.autoneg_wait_to_complete = FALSE;
 502         adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
 503         adapter->rx_buffer_len = MCLBYTES;
 504
 505         /*
 506          * Interrupt throttle rate
 507          */
 508         if (em_int_throttle_ceil == 0) {
 509                 adapter->int_throttle_ceil = 0;
 510         } else {
 511                 int throttle = em_int_throttle_ceil;
 512
 513                 if (throttle < 0)
 514                         throttle = EM_DEFAULT_ITR;
 515
 516                 /* Recalculate the tunable value to get the exact frequency. */
 517                 throttle = 1000000000 / 256 / throttle;
 518
 519                 /* Upper 16bits of ITR is reserved and should be zero */
 520                 if (throttle & 0xffff0000)
 521                         throttle = 1000000000 / 256 / EM_DEFAULT_ITR;
 522
 523                 adapter->int_throttle_ceil = 1000000000 / 256 / throttle;
 524         }
 525
 526         e1000_init_script_state_82541(&adapter->hw, TRUE);
 527         e1000_set_tbi_compatibility_82543(&adapter->hw, TRUE);
 528
 529         /* Copper options */
 530         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
 531                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
 532                 adapter->hw.phy.disable_polarity_correction = FALSE;
 533                 adapter->hw.phy.ms_type = EM_MASTER_SLAVE;
 534         }
 535
 536         /* Set the frame limits assuming standard ethernet sized frames. */
 537         adapter->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
 538         adapter->min_frame_size = ETH_ZLEN + ETHER_CRC_LEN;
 539
 540         /* This controls when hardware reports transmit completion status. */
 541         adapter->hw.mac.report_tx_early = 1;
 542
 543         /*
 544          * Create top level busdma tag
 545          */
 546         error = bus_dma_tag_create(NULL, 1, 0,
 547                         BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
 548                         NULL, NULL,
 549                         BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
 550                         0, &adapter->parent_dtag);
 551         if (error) {
 552                 device_printf(dev, "could not create top level DMA tag\n");
 553                 goto fail;
 554         }
 555
 556         /*
 557          * Allocate Transmit Descriptor ring
 558          */
 559         tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc),
 560                          EM_DBA_ALIGN);
 561         error = em_dma_malloc(adapter, tsize, &adapter->txdma);
 562         if (error) {
 563                 device_printf(dev, "Unable to allocate tx_desc memory\n");
 564                 goto fail;
 565         }
 566         adapter->tx_desc_base = adapter->txdma.dma_vaddr;
 567
 568         /*
 569          * Allocate Receive Descriptor ring
 570          */
 571         rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc),
 572                          EM_DBA_ALIGN);
 573         error = em_dma_malloc(adapter, rsize, &adapter->rxdma);
 574         if (error) {
 575                 device_printf(dev, "Unable to allocate rx_desc memory\n");
 576                 goto fail;
 577         }
 578         adapter->rx_desc_base = adapter->rxdma.dma_vaddr;
 579
 580         /* Make sure we have a good EEPROM before we read from it */
 581         if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
 582                 /*
 583                  * Some PCI-E parts fail the first check due to
 584                  * the link being in sleep state, call it again,
 585                  * if it fails a second time its a real issue.
 586                  */
 587                 if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
 588                         device_printf(dev,
 589                             "The EEPROM Checksum Is Not Valid\n");
 590                         error = EIO;
 591                         goto fail;
 592                 }
 593         }
 594
 595         /* Initialize the hardware */
 596         error = em_hw_init(adapter);
 597         if (error) {
 598                 device_printf(dev, "Unable to initialize the hardware\n");
 599                 goto fail;
 600         }
 601
 602         /* Copy the permanent MAC address out of the EEPROM */
 603         if (e1000_read_mac_addr(&adapter->hw) < 0) {
 604                 device_printf(dev, "EEPROM read error while reading MAC"
 605                     " address\n");
 606                 error = EIO;
 607                 goto fail;
 608         }
 609         if (!em_is_valid_eaddr(adapter->hw.mac.addr)) {
 610                 device_printf(dev, "Invalid MAC address\n");
 611                 error = EIO;
 612                 goto fail;
 613         }
 614
 615         /* Allocate transmit descriptors and buffers */
 616         error = em_create_tx_ring(adapter);
 617         if (error) {
 618                 device_printf(dev, "Could not setup transmit structures\n");
 619                 goto fail;
 620         }
 621
 622         /* Allocate receive descriptors and buffers */
 623         error = em_create_rx_ring(adapter);
 624         if (error) {
 625                 device_printf(dev, "Could not setup receive structures\n");
 626                 goto fail;
 627         }
 628
 629         /* Manually turn off all interrupts */
 630         E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
 631
 632         /* Setup OS specific network interface */
 633         em_setup_ifp(adapter);
 634
 635         /* Add sysctl tree, must after em_setup_ifp() */
 636         em_add_sysctl(adapter);
 637
 638         /* Initialize statistics */
 639         em_update_stats(adapter);
 640
 641         adapter->hw.mac.get_link_status = 1;
 642         em_update_link_status(adapter);
 643
 644         /* Indicate SOL/IDER usage */
 645         if (e1000_check_reset_block(&adapter->hw)) {
 646                 device_printf(dev,
 647                     "PHY reset is blocked due to SOL/IDER session.\n");
 648         }
 649
 650         /* Determine if we have to control management hardware */
 651         adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw);
 652
 653         /*
 654          * Setup Wake-on-Lan
 655          */
 656         switch (adapter->hw.mac.type) {
 657         case e1000_82542:
 658         case e1000_82543:
 659                 break;
 660
 661         case e1000_82546:
 662         case e1000_82546_rev_3:
 663         case e1000_82571:
 664         case e1000_80003es2lan:
 665                 if (adapter->hw.bus.func == 1) {
 666                         e1000_read_nvm(&adapter->hw,
 667                             NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 668                 } else {
 669                         e1000_read_nvm(&adapter->hw,
 670                             NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
 671                 }
 672                 eeprom_data &= EM_EEPROM_APME;
 673                 break;
 674
 675         default:
 676                 /* APME bit in EEPROM is mapped to WUC.APME */
 677                 eeprom_data =
 678                     E1000_READ_REG(&adapter->hw, E1000_WUC) & E1000_WUC_APME;
 679                 break;
 680         }
 681         if (eeprom_data)
 682                 adapter->wol = E1000_WUFC_MAG;
 683         /*
 684          * We have the eeprom settings, now apply the special cases
 685          * where the eeprom may be wrong or the board won't support
 686          * wake on lan on a particular port
 687          */
 688         device_id = pci_get_device(dev);
 689         switch (device_id) {
 690         case E1000_DEV_ID_82546GB_PCIE:
 691                 adapter->wol = 0;
 692                 break;
 693
 694         case E1000_DEV_ID_82546EB_FIBER:
 695         case E1000_DEV_ID_82546GB_FIBER:
 696         case E1000_DEV_ID_82571EB_FIBER:
 697                 /*
 698                  * Wake events only supported on port A for dual fiber
 699                  * regardless of eeprom setting
 700                  */
 701                 if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
 702                     E1000_STATUS_FUNC_1)
 703                         adapter->wol = 0;
 704                 break;
 705
 706         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
 707         case E1000_DEV_ID_82571EB_QUAD_COPPER:
 708         case E1000_DEV_ID_82571EB_QUAD_FIBER:
 709         case E1000_DEV_ID_82571EB_QUAD_COPPER_LP:
 710                 /* if quad port adapter, disable WoL on all but port A */
 711                 if (em_global_quad_port_a != 0)
 712                         adapter->wol = 0;
 713                 /* Reset for multiple quad port adapters */
 714                 if (++em_global_quad_port_a == 4)
 715                         em_global_quad_port_a = 0;
 716                 break;
 717         }
 718
 719         /* XXX disable wol */
 720         adapter->wol = 0;
 721
 722         /* Do we need workaround for 82544 PCI-X adapter? */
 723         if (adapter->hw.bus.type == e1000_bus_type_pcix &&
 724             adapter->hw.mac.type == e1000_82544)
 725                 adapter->pcix_82544 = TRUE;
 726         else
 727                 adapter->pcix_82544 = FALSE;
 728
 729         if (adapter->pcix_82544) {
 730                 /*
 731                  * 82544 on PCI-X may split one TX segment
 732                  * into two TX descs, so we double its number
 733                  * of spare TX desc here.
 734                  */
 735                 adapter->spare_tx_desc = 2 * EM_TX_SPARE;
 736         } else {
 737                 adapter->spare_tx_desc = EM_TX_SPARE;
 738         }
 739
 740         /*
 741          * Keep following relationship between spare_tx_desc, oact_tx_desc
 742          * and tx_int_nsegs:
 743          * (spare_tx_desc + EM_TX_RESERVED) <=
 744          * oact_tx_desc <= EM_TX_OACTIVE_MAX <= tx_int_nsegs
 745          */
 746         adapter->oact_tx_desc = adapter->num_tx_desc / 8;
 747         if (adapter->oact_tx_desc > EM_TX_OACTIVE_MAX)
 748                 adapter->oact_tx_desc = EM_TX_OACTIVE_MAX;
 749         if (adapter->oact_tx_desc < adapter->spare_tx_desc + EM_TX_RESERVED)
 750                 adapter->oact_tx_desc = adapter->spare_tx_desc + EM_TX_RESERVED;
 751
 752         adapter->tx_int_nsegs = adapter->num_tx_desc / 16;
 753         if (adapter->tx_int_nsegs < adapter->oact_tx_desc)
 754                 adapter->tx_int_nsegs = adapter->oact_tx_desc;
 755
 756         error = bus_setup_intr(dev, adapter->intr_res, INTR_MPSAFE,
 757                                em_intr, adapter, &adapter->intr_tag,
 758                                ifp->if_serializer);
 759         if (error) {
 760                 device_printf(dev, "Failed to register interrupt handler");
 761                 ether_ifdetach(&adapter->arpcom.ac_if);
 762                 goto fail;
 763         }
 764
 765         ifp->if_cpuid = ithread_cpuid(rman_get_start(adapter->intr_res));
 766         KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
 767         return (0);
 768 fail:
 769         em_detach(dev);
 770         return (error);
 771 }
 772
 773 static int
 774 em_detach(device_t dev)
 775 {
 776         struct adapter *adapter = device_get_softc(dev);
 777
 778         if (device_is_attached(dev)) {
 779                 struct ifnet *ifp = &adapter->arpcom.ac_if;
 780
 781                 lwkt_serialize_enter(ifp->if_serializer);
 782
 783                 em_stop(adapter);
 784
 785                 e1000_phy_hw_reset(&adapter->hw);
 786
 787                 em_rel_mgmt(adapter);
 788
 789                 if ((adapter->hw.mac.type == e1000_82573 ||
 790                      adapter->hw.mac.type == e1000_ich8lan ||
 791                      adapter->hw.mac.type == e1000_ich10lan ||
 792                      adapter->hw.mac.type == e1000_ich9lan) &&
 793                     e1000_check_mng_mode(&adapter->hw))
 794                         em_rel_hw_control(adapter);
 795
 796                 if (adapter->wol) {
 797                         E1000_WRITE_REG(&adapter->hw, E1000_WUC,
 798                                         E1000_WUC_PME_EN);
 799                         E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
 800                         em_enable_wol(dev);
 801                 }
 802
 803                 bus_teardown_intr(dev, adapter->intr_res, adapter->intr_tag);
 804
 805                 lwkt_serialize_exit(ifp->if_serializer);
 806
 807                 ether_ifdetach(ifp);
 808         }
 809         bus_generic_detach(dev);
 810
 811         em_free_pci_res(adapter);
 812
 813         em_destroy_tx_ring(adapter, adapter->num_tx_desc);
 814         em_destroy_rx_ring(adapter, adapter->num_rx_desc);
 815
 816         /* Free Transmit Descriptor ring */
 817         if (adapter->tx_desc_base)
 818                 em_dma_free(adapter, &adapter->txdma);
 819
 820         /* Free Receive Descriptor ring */
 821         if (adapter->rx_desc_base)
 822                 em_dma_free(adapter, &adapter->rxdma);
 823
 824         /* Free top level busdma tag */
 825         if (adapter->parent_dtag != NULL)
 826                 bus_dma_tag_destroy(adapter->parent_dtag);
 827
 828         /* Free sysctl tree */
 829         if (adapter->sysctl_tree != NULL)
 830                 sysctl_ctx_free(&adapter->sysctl_ctx);
 831
 832         return (0);
 833 }
 834
 835 static int
 836 em_shutdown(device_t dev)
 837 {
 838         return em_suspend(dev);
 839 }
 840
 841 static int
 842 em_suspend(device_t dev)
 843 {
 844         struct adapter *adapter = device_get_softc(dev);
 845         struct ifnet *ifp = &adapter->arpcom.ac_if;
 846
 847         lwkt_serialize_enter(ifp->if_serializer);
 848
 849         em_stop(adapter);
 850
 851         em_rel_mgmt(adapter);
 852
 853         if ((adapter->hw.mac.type == e1000_82573 ||
 854              adapter->hw.mac.type == e1000_ich8lan ||
 855              adapter->hw.mac.type == e1000_ich10lan ||
 856              adapter->hw.mac.type == e1000_ich9lan) &&
 857             e1000_check_mng_mode(&adapter->hw))
 858                 em_rel_hw_control(adapter);
 859
 860         if (adapter->wol) {
 861                 E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
 862                 E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
 863                 em_enable_wol(dev);
 864         }
 865
 866         lwkt_serialize_exit(ifp->if_serializer);
 867
 868         return bus_generic_suspend(dev);
 869 }
 870
 871 static int
 872 em_resume(device_t dev)
 873 {
 874         struct adapter *adapter = device_get_softc(dev);
 875         struct ifnet *ifp = &adapter->arpcom.ac_if;
 876
 877         lwkt_serialize_enter(ifp->if_serializer);
 878
 879         em_init(adapter);
 880         em_get_mgmt(adapter);
 881         if_devstart(ifp);
 882
 883         lwkt_serialize_exit(ifp->if_serializer);
 884
 885         return bus_generic_resume(dev);
 886 }
 887
 888 static void
 889 em_start(struct ifnet *ifp)
 890 {
 891         struct adapter *adapter = ifp->if_softc;
 892         struct mbuf *m_head;
 893
 894         ASSERT_SERIALIZED(ifp->if_serializer);
 895
 896         if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
 897                 return;
 898
 899         if (!adapter->link_active) {
 900                 ifq_purge(&ifp->if_snd);
 901                 return;
 902         }
 903
 904         while (!ifq_is_empty(&ifp->if_snd)) {
 905                 /* Now do we at least have a minimal? */
 906                 if (EM_IS_OACTIVE(adapter)) {
 907                         em_tx_collect(adapter);
 908                         if (EM_IS_OACTIVE(adapter)) {
 909                                 ifp->if_flags |= IFF_OACTIVE;
 910                                 adapter->no_tx_desc_avail1++;
 911                                 break;
 912                         }
 913                 }
 914
 915                 logif(pkt_txqueue);
 916                 m_head = ifq_dequeue(&ifp->if_snd, NULL);
 917                 if (m_head == NULL)
 918                         break;
 919
 920                 if (em_encap(adapter, &m_head)) {
 921                         ifp->if_oerrors++;
 922                         em_tx_collect(adapter);
 923                         continue;
 924                 }
 925
 926                 /* Send a copy of the frame to the BPF listener */
 927                 ETHER_BPF_MTAP(ifp, m_head);
 928
 929                 /* Set timeout in case hardware has problems transmitting. */
 930                 ifp->if_timer = EM_TX_TIMEOUT;
 931         }
 932 }
 933
 934 static int
 935 em_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
 936 {
 937         struct adapter *adapter = ifp->if_softc;
 938         struct ifreq *ifr = (struct ifreq *)data;
 939         uint16_t eeprom_data = 0;
 940         int max_frame_size, mask, reinit;
 941         int error = 0;
 942
 943         ASSERT_SERIALIZED(ifp->if_serializer);
 944
 945         switch (command) {
 946         case SIOCSIFMTU:
 947                 switch (adapter->hw.mac.type) {
 948                 case e1000_82573:
 949                         /*
 950                          * 82573 only supports jumbo frames
 951                          * if ASPM is disabled.
 952                          */
 953                         e1000_read_nvm(&adapter->hw,
 954                             NVM_INIT_3GIO_3, 1, &eeprom_data);
 955                         if (eeprom_data & NVM_WORD1A_ASPM_MASK) {
 956                                 max_frame_size = ETHER_MAX_LEN;
 957                                 break;
 958                         }
 959                         /* FALL THROUGH */
 960
 961                 /* Limit Jumbo Frame size */
 962                 case e1000_82571:
 963                 case e1000_82572:
 964                 case e1000_ich9lan:
 965                 case e1000_ich10lan:
 966                 case e1000_82574:
 967                 case e1000_80003es2lan:
 968                         max_frame_size = 9234;
 969                         break;
 970
 971                 /* Adapters that do not support jumbo frames */
 972                 case e1000_82542:
 973                 case e1000_ich8lan:
 974                         max_frame_size = ETHER_MAX_LEN;
 975                         break;
 976
 977                 default:
 978                         max_frame_size = MAX_JUMBO_FRAME_SIZE;
 979                         break;
 980                 }
 981                 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
 982                     ETHER_CRC_LEN) {
 983                         error = EINVAL;
 984                         break;
 985                 }
 986
 987                 ifp->if_mtu = ifr->ifr_mtu;
 988                 adapter->max_frame_size =
 989                     ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
 990
 991                 if (ifp->if_flags & IFF_RUNNING)
 992                         em_init(adapter);
 993                 break;
 994
 995         case SIOCSIFFLAGS:
 996                 if (ifp->if_flags & IFF_UP) {
 997                         if ((ifp->if_flags & IFF_RUNNING)) {
 998                                 if ((ifp->if_flags ^ adapter->if_flags) &
 999                                     (IFF_PROMISC | IFF_ALLMULTI)) {
1000                                         em_disable_promisc(adapter);
1001                                         em_set_promisc(adapter);
1002                                 }
1003                         } else {
1004                                 em_init(adapter);
1005                         }
1006                 } else if (ifp->if_flags & IFF_RUNNING) {
1007                         em_stop(adapter);
1008                 }
1009                 adapter->if_flags = ifp->if_flags;
1010                 break;
1011
1012         case SIOCADDMULTI:
1013         case SIOCDELMULTI:
1014                 if (ifp->if_flags & IFF_RUNNING) {
1015                         em_disable_intr(adapter);
1016                         em_set_multi(adapter);
1017                         if (adapter->hw.mac.type == e1000_82542 &&
1018                             adapter->hw.revision_id == E1000_REVISION_2)
1019                                 em_init_rx_unit(adapter);
1020 #ifdef DEVICE_POLLING
1021                         if (!(ifp->if_flags & IFF_POLLING))
1022 #endif
1023                                 em_enable_intr(adapter);
1024                 }
1025                 break;
1026
1027         case SIOCSIFMEDIA:
1028                 /* Check SOL/IDER usage */
1029                 if (e1000_check_reset_block(&adapter->hw)) {
1030                         device_printf(adapter->dev, "Media change is"
1031                             " blocked due to SOL/IDER session.\n");
1032                         break;
1033                 }
1034                 /* FALL THROUGH */
1035
1036         case SIOCGIFMEDIA:
1037                 error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
1038                 break;
1039
1040         case SIOCSIFCAP:
1041                 reinit = 0;
1042                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1043                 if (mask & IFCAP_HWCSUM) {
1044                         ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
1045                         reinit = 1;
1046                 }
1047                 if (mask & IFCAP_VLAN_HWTAGGING) {
1048                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1049                         reinit = 1;
1050                 }
1051                 if (reinit && (ifp->if_flags & IFF_RUNNING))
1052                         em_init(adapter);
1053                 break;
1054
1055         default:
1056                 error = ether_ioctl(ifp, command, data);
1057                 break;
1058         }
1059         return (error);
1060 }
1061
1062 static void
1063 em_watchdog(struct ifnet *ifp)
1064 {
1065         struct adapter *adapter = ifp->if_softc;
1066
1067         ASSERT_SERIALIZED(ifp->if_serializer);
1068
1069         /*
1070          * The timer is set to 5 every time start queues a packet.
1071          * Then txeof keeps resetting it as long as it cleans at
1072          * least one descriptor.
1073          * Finally, anytime all descriptors are clean the timer is
1074          * set to 0.
1075          */
1076
1077         if (E1000_READ_REG(&adapter->hw, E1000_TDT(0)) ==
1078             E1000_READ_REG(&adapter->hw, E1000_TDH(0))) {
1079                 /*
1080                  * If we reach here, all TX jobs are completed and
1081                  * the TX engine should have been idled for some time.
1082                  * We don't need to call if_devstart() here.
1083                  */
1084                 ifp->if_flags &= ~IFF_OACTIVE;
1085                 ifp->if_timer = 0;
1086                 return;
1087         }
1088
1089         /*
1090          * If we are in this routine because of pause frames, then
1091          * don't reset the hardware.
1092          */
1093         if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
1094             E1000_STATUS_TXOFF) {
1095                 ifp->if_timer = EM_TX_TIMEOUT;
1096                 return;
1097         }
1098
1099         if (e1000_check_for_link(&adapter->hw) == 0)
1100                 if_printf(ifp, "watchdog timeout -- resetting\n");
1101
1102         ifp->if_oerrors++;
1103         adapter->watchdog_events++;
1104
1105         em_init(adapter);
1106
1107         if (!ifq_is_empty(&ifp->if_snd))
1108                 if_devstart(ifp);
1109 }
1110
1111 static void
1112 em_init(void *xsc)
1113 {
1114         struct adapter *adapter = xsc;
1115         struct ifnet *ifp = &adapter->arpcom.ac_if;
1116         device_t dev = adapter->dev;
1117         uint32_t pba;
1118
1119         ASSERT_SERIALIZED(ifp->if_serializer);
1120
1121         em_stop(adapter);
1122
1123         /*
1124          * Packet Buffer Allocation (PBA)
1125          * Writing PBA sets the receive portion of the buffer
1126          * the remainder is used for the transmit buffer.
1127          *
1128          * Devices before the 82547 had a Packet Buffer of 64K.
1129          *   Default allocation: PBA=48K for Rx, leaving 16K for Tx.
1130          * After the 82547 the buffer was reduced to 40K.
1131          *   Default allocation: PBA=30K for Rx, leaving 10K for Tx.
1132          *   Note: default does not leave enough room for Jumbo Frame >10k.
1133          */
1134         switch (adapter->hw.mac.type) {
1135         case e1000_82547:
1136         case e1000_82547_rev_2: /* 82547: Total Packet Buffer is 40K */
1137                 if (adapter->max_frame_size > 8192)
1138                         pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
1139                 else
1140                         pba = E1000_PBA_30K; /* 30K for Rx, 10K for Tx */
1141                 adapter->tx_fifo_head = 0;
1142                 adapter->tx_head_addr = pba << EM_TX_HEAD_ADDR_SHIFT;
1143                 adapter->tx_fifo_size =
1144                     (E1000_PBA_40K - pba) << EM_PBA_BYTES_SHIFT;
1145                 break;
1146
1147         /* Total Packet Buffer on these is 48K */
1148         case e1000_82571:
1149         case e1000_82572:
1150         case e1000_80003es2lan:
1151                 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1152                 break;
1153
1154         case e1000_82573: /* 82573: Total Packet Buffer is 32K */
1155                 pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
1156                 break;
1157
1158         case e1000_82574:
1159                 pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
1160                 break;
1161
1162         case e1000_ich9lan:
1163         case e1000_ich10lan:
1164 #define E1000_PBA_10K   0x000A
1165                 pba = E1000_PBA_10K;
1166                 break;
1167
1168         case e1000_ich8lan:
1169                 pba = E1000_PBA_8K;
1170                 break;
1171
1172         default:
1173                 /* Devices before 82547 had a Packet Buffer of 64K.   */
1174                 if (adapter->max_frame_size > 8192)
1175                         pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
1176                 else
1177                         pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
1178         }
1179         E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba);
1180
1181         /* Get the latest mac address, User can use a LAA */
1182         bcopy(IF_LLADDR(ifp), adapter->hw.mac.addr, ETHER_ADDR_LEN);
1183
1184         /* Put the address into the Receive Address Array */
1185         e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1186
1187         /*
1188          * With the 82571 adapter, RAR[0] may be overwritten
1189          * when the other port is reset, we make a duplicate
1190          * in RAR[14] for that eventuality, this assures
1191          * the interface continues to function.
1192          */
1193         if (adapter->hw.mac.type == e1000_82571) {
1194                 e1000_set_laa_state_82571(&adapter->hw, TRUE);
1195                 e1000_rar_set(&adapter->hw, adapter->hw.mac.addr,
1196                     E1000_RAR_ENTRIES - 1);
1197         }
1198
1199         /* Initialize the hardware */
1200         if (em_hw_init(adapter)) {
1201                 device_printf(dev, "Unable to initialize the hardware\n");
1202                 /* XXX em_stop()? */
1203                 return;
1204         }
1205         em_update_link_status(adapter);
1206
1207         /* Setup VLAN support, basic and offload if available */
1208         E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
1209
1210         if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1211                 uint32_t ctrl;
1212
1213                 ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
1214                 ctrl |= E1000_CTRL_VME;
1215                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
1216         }
1217
1218         /* Set hardware offload abilities */
1219         if (ifp->if_capenable & IFCAP_TXCSUM)
1220                 ifp->if_hwassist = EM_CSUM_FEATURES;
1221         else
1222                 ifp->if_hwassist = 0;
1223
1224         /* Configure for OS presence */
1225         em_get_mgmt(adapter);
1226
1227         /* Prepare transmit descriptors and buffers */
1228         em_init_tx_ring(adapter);
1229         em_init_tx_unit(adapter);
1230
1231         /* Setup Multicast table */
1232         em_set_multi(adapter);
1233
1234         /* Prepare receive descriptors and buffers */
1235         if (em_init_rx_ring(adapter)) {
1236                 device_printf(dev, "Could not setup receive structures\n");
1237                 em_stop(adapter);
1238                 return;
1239         }
1240         em_init_rx_unit(adapter);
1241
1242         /* Don't lose promiscuous settings */
1243         em_set_promisc(adapter);
1244
1245         ifp->if_flags |= IFF_RUNNING;
1246         ifp->if_flags &= ~IFF_OACTIVE;
1247
1248         callout_reset(&adapter->timer, hz, em_timer, adapter);
1249         e1000_clear_hw_cntrs_base_generic(&adapter->hw);
1250
1251         /* MSI/X configuration for 82574 */
1252         if (adapter->hw.mac.type == e1000_82574) {
1253                 int tmp;
1254
1255                 tmp = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
1256                 tmp |= E1000_CTRL_EXT_PBA_CLR;
1257                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, tmp);
1258                 /*
1259                  * Set the IVAR - interrupt vector routing.
1260                  * Each nibble represents a vector, high bit
1261                  * is enable, other 3 bits are the MSIX table
1262                  * entry, we map RXQ0 to 0, TXQ0 to 1, and
1263                  * Link (other) to 2, hence the magic number.
1264                  */
1265                 E1000_WRITE_REG(&adapter->hw, E1000_IVAR, 0x800A0908);
1266         }
1267
1268 #ifdef DEVICE_POLLING
1269         /*
1270          * Only enable interrupts if we are not polling, make sure
1271          * they are off otherwise.
1272          */
1273         if (ifp->if_flags & IFF_POLLING)
1274                 em_disable_intr(adapter);
1275         else
1276 #endif /* DEVICE_POLLING */
1277                 em_enable_intr(adapter);
1278
1279         /* Don't reset the phy next time init gets called */
1280         adapter->hw.phy.reset_disable = TRUE;
1281 }
1282
1283 #ifdef DEVICE_POLLING
1284
1285 static void
1286 em_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1287 {
1288         struct adapter *adapter = ifp->if_softc;
1289         uint32_t reg_icr;
1290
1291         ASSERT_SERIALIZED(ifp->if_serializer);
1292
1293         switch (cmd) {
1294         case POLL_REGISTER:
1295                 em_disable_intr(adapter);
1296                 break;
1297
1298         case POLL_DEREGISTER:
1299                 em_enable_intr(adapter);
1300                 break;
1301
1302         case POLL_AND_CHECK_STATUS:
1303                 reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
1304                 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1305                         callout_stop(&adapter->timer);
1306                         adapter->hw.mac.get_link_status = 1;
1307                         em_update_link_status(adapter);
1308                         callout_reset(&adapter->timer, hz, em_timer, adapter);
1309                 }
1310                 /* FALL THROUGH */
1311         case POLL_ONLY:
1312                 if (ifp->if_flags & IFF_RUNNING) {
1313                         em_rxeof(adapter, count);
1314                         em_txeof(adapter);
1315
1316                         if (!ifq_is_empty(&ifp->if_snd))
1317                                 if_devstart(ifp);
1318                 }
1319                 break;
1320         }
1321 }
1322
1323 #endif /* DEVICE_POLLING */
1324
1325 static void
1326 em_intr(void *xsc)
1327 {
1328         struct adapter *adapter = xsc;
1329         struct ifnet *ifp = &adapter->arpcom.ac_if;
1330         uint32_t reg_icr;
1331
1332         logif(intr_beg);
1333         ASSERT_SERIALIZED(ifp->if_serializer);
1334
1335         reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
1336
1337         if ((adapter->hw.mac.type >= e1000_82571 &&
1338              (reg_icr & E1000_ICR_INT_ASSERTED) == 0) ||
1339             reg_icr == 0) {
1340                 logif(intr_end);
1341                 return;
1342         }
1343
1344         /*
1345          * XXX: some laptops trigger several spurious interrupts
1346          * on em(4) when in the resume cycle. The ICR register
1347          * reports all-ones value in this case. Processing such
1348          * interrupts would lead to a freeze. I don't know why.
1349          */
1350         if (reg_icr == 0xffffffff) {
1351                 logif(intr_end);
1352                 return;
1353         }
1354
1355         if (ifp->if_flags & IFF_RUNNING) {
1356                 if (reg_icr &
1357                     (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO))
1358                         em_rxeof(adapter, -1);
1359                 if (reg_icr & E1000_ICR_TXDW) {
1360                         em_txeof(adapter);
1361                         if (!ifq_is_empty(&ifp->if_snd))
1362                                 if_devstart(ifp);
1363                 }
1364         }
1365
1366         /* Link status change */
1367         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1368                 callout_stop(&adapter->timer);
1369                 adapter->hw.mac.get_link_status = 1;
1370                 em_update_link_status(adapter);
1371
1372                 /* Deal with TX cruft when link lost */
1373                 em_tx_purge(adapter);
1374
1375                 callout_reset(&adapter->timer, hz, em_timer, adapter);
1376         }
1377
1378         if (reg_icr & E1000_ICR_RXO)
1379                 adapter->rx_overruns++;
1380
1381         logif(intr_end);
1382 }
1383
1384 static void
1385 em_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1386 {
1387         struct adapter *adapter = ifp->if_softc;
1388         u_char fiber_type = IFM_1000_SX;
1389
1390         ASSERT_SERIALIZED(ifp->if_serializer);
1391
1392         em_update_link_status(adapter);
1393
1394         ifmr->ifm_status = IFM_AVALID;
1395         ifmr->ifm_active = IFM_ETHER;
1396
1397         if (!adapter->link_active)
1398                 return;
1399
1400         ifmr->ifm_status |= IFM_ACTIVE;
1401
1402         if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
1403             adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
1404                 if (adapter->hw.mac.type == e1000_82545)
1405                         fiber_type = IFM_1000_LX;
1406                 ifmr->ifm_active |= fiber_type | IFM_FDX;
1407         } else {
1408                 switch (adapter->link_speed) {
1409                 case 10:
1410                         ifmr->ifm_active |= IFM_10_T;
1411                         break;
1412                 case 100:
1413                         ifmr->ifm_active |= IFM_100_TX;
1414                         break;
1415
1416                 case 1000:
1417                         ifmr->ifm_active |= IFM_1000_T;
1418                         break;
1419                 }
1420                 if (adapter->link_duplex == FULL_DUPLEX)
1421                         ifmr->ifm_active |= IFM_FDX;
1422                 else
1423                         ifmr->ifm_active |= IFM_HDX;
1424         }
1425 }
1426
1427 static int
1428 em_media_change(struct ifnet *ifp)
1429 {
1430         struct adapter *adapter = ifp->if_softc;
1431         struct ifmedia *ifm = &adapter->media;
1432
1433         ASSERT_SERIALIZED(ifp->if_serializer);
1434
1435         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1436                 return (EINVAL);
1437
1438         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1439         case IFM_AUTO:
1440                 adapter->hw.mac.autoneg = DO_AUTO_NEG;
1441                 adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1442                 break;
1443
1444         case IFM_1000_LX:
1445         case IFM_1000_SX:
1446         case IFM_1000_T:
1447                 adapter->hw.mac.autoneg = DO_AUTO_NEG;
1448                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1449                 break;
1450
1451         case IFM_100_TX:
1452                 adapter->hw.mac.autoneg = FALSE;
1453                 adapter->hw.phy.autoneg_advertised = 0;
1454                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1455                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1456                 else
1457                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1458                 break;
1459
1460         case IFM_10_T:
1461                 adapter->hw.mac.autoneg = FALSE;
1462                 adapter->hw.phy.autoneg_advertised = 0;
1463                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1464                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1465                 else
1466                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1467                 break;
1468
1469         default:
1470                 if_printf(ifp, "Unsupported media type\n");
1471                 break;
1472         }
1473
1474         /*
1475          * As the speed/duplex settings my have changed we need to
1476          * reset the PHY.
1477          */
1478         adapter->hw.phy.reset_disable = FALSE;
1479
1480         em_init(adapter);
1481
1482         return (0);
1483 }
1484
1485 static int
1486 em_encap(struct adapter *adapter, struct mbuf **m_headp)
1487 {
1488         bus_dma_segment_t segs[EM_MAX_SCATTER];
1489         bus_dmamap_t map;
1490         struct em_buffer *tx_buffer, *tx_buffer_mapped;
1491         struct e1000_tx_desc *ctxd = NULL;
1492         struct mbuf *m_head = *m_headp;
1493         uint32_t txd_upper, txd_lower, txd_used, cmd = 0;
1494         int maxsegs, nsegs, i, j, first, last = 0, error;
1495
1496         if (m_head->m_len < EM_TXCSUM_MINHL &&
1497             (m_head->m_flags & EM_CSUM_FEATURES)) {
1498                 /*
1499                  * Make sure that ethernet header and ip.ip_hl are in
1500                  * contiguous memory, since if TXCSUM is enabled, later
1501                  * TX context descriptor's setup need to access ip.ip_hl.
1502                  */
1503                 error = em_txcsum_pullup(adapter, m_headp);
1504                 if (error) {
1505                         KKASSERT(*m_headp == NULL);
1506                         return error;
1507                 }
1508                 m_head = *m_headp;
1509         }
1510
1511         txd_upper = txd_lower = 0;
1512         txd_used = 0;
1513
1514         /*
1515          * Capture the first descriptor index, this descriptor
1516          * will have the index of the EOP which is the only one
1517          * that now gets a DONE bit writeback.
1518          */
1519         first = adapter->next_avail_tx_desc;
1520         tx_buffer = &adapter->tx_buffer_area[first];
1521         tx_buffer_mapped = tx_buffer;
1522         map = tx_buffer->map;
1523
1524         maxsegs = adapter->num_tx_desc_avail - EM_TX_RESERVED;
1525         KASSERT(maxsegs >= adapter->spare_tx_desc,
1526                 ("not enough spare TX desc\n"));
1527         if (adapter->pcix_82544) {
1528                 /* Half it; see the comment in em_attach() */
1529                 maxsegs >>= 1;
1530         }
1531         if (maxsegs > EM_MAX_SCATTER)
1532                 maxsegs = EM_MAX_SCATTER;
1533
1534         error = bus_dmamap_load_mbuf_defrag(adapter->txtag, map, m_headp,
1535                         segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1536         if (error) {
1537                 if (error == ENOBUFS)
1538                         adapter->mbuf_alloc_failed++;
1539                 else
1540                         adapter->no_tx_dma_setup++;
1541
1542                 m_freem(*m_headp);
1543                 *m_headp = NULL;
1544                 return error;
1545         }
1546         bus_dmamap_sync(adapter->txtag, map, BUS_DMASYNC_PREWRITE);
1547
1548         m_head = *m_headp;
1549         adapter->tx_nsegs += nsegs;
1550
1551         if (m_head->m_pkthdr.csum_flags & EM_CSUM_FEATURES) {
1552                 /* TX csum offloading will consume one TX desc */
1553                 adapter->tx_nsegs += em_txcsum(adapter, m_head,
1554                                                &txd_upper, &txd_lower);
1555         }
1556         i = adapter->next_avail_tx_desc;
1557
1558         /* Set up our transmit descriptors */
1559         for (j = 0; j < nsegs; j++) {
1560                 /* If adapter is 82544 and on PCIX bus */
1561                 if(adapter->pcix_82544) {
1562                         DESC_ARRAY desc_array;
1563                         uint32_t array_elements, counter;
1564
1565                         /*
1566                          * Check the Address and Length combination and
1567                          * split the data accordingly
1568                          */
1569                         array_elements = em_82544_fill_desc(segs[j].ds_addr,
1570                                                 segs[j].ds_len, &desc_array);
1571                         for (counter = 0; counter < array_elements; counter++) {
1572                                 KKASSERT(txd_used < adapter->num_tx_desc_avail);
1573
1574                                 tx_buffer = &adapter->tx_buffer_area[i];
1575                                 ctxd = &adapter->tx_desc_base[i];
1576
1577                                 ctxd->buffer_addr = htole64(
1578                                     desc_array.descriptor[counter].address);
1579                                 ctxd->lower.data = htole32(
1580                                     E1000_TXD_CMD_IFCS | txd_lower |
1581                                     desc_array.descriptor[counter].length);
1582                                 ctxd->upper.data = htole32(txd_upper);
1583
1584                                 last = i;
1585                                 if (++i == adapter->num_tx_desc)
1586                                         i = 0;
1587
1588                                 txd_used++;
1589                         }
1590                 } else {
1591                         tx_buffer = &adapter->tx_buffer_area[i];
1592                         ctxd = &adapter->tx_desc_base[i];
1593
1594                         ctxd->buffer_addr = htole64(segs[j].ds_addr);
1595                         ctxd->lower.data = htole32(E1000_TXD_CMD_IFCS |
1596                                                    txd_lower | segs[j].ds_len);
1597                         ctxd->upper.data = htole32(txd_upper);
1598
1599                         last = i;
1600                         if (++i == adapter->num_tx_desc)
1601                                 i = 0;
1602                 }
1603         }
1604
1605         adapter->next_avail_tx_desc = i;
1606         if (adapter->pcix_82544) {
1607                 KKASSERT(adapter->num_tx_desc_avail > txd_used);
1608                 adapter->num_tx_desc_avail -= txd_used;
1609         } else {
1610                 KKASSERT(adapter->num_tx_desc_avail > nsegs);
1611                 adapter->num_tx_desc_avail -= nsegs;
1612         }
1613
1614         /* Handle VLAN tag */
1615         if (m_head->m_flags & M_VLANTAG) {
1616                 /* Set the vlan id. */
1617                 ctxd->upper.fields.special =
1618                     htole16(m_head->m_pkthdr.ether_vlantag);
1619
1620                 /* Tell hardware to add tag */
1621                 ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE);
1622         }
1623
1624         tx_buffer->m_head = m_head;
1625         tx_buffer_mapped->map = tx_buffer->map;
1626         tx_buffer->map = map;
1627
1628         if (adapter->tx_nsegs >= adapter->tx_int_nsegs) {
1629                 adapter->tx_nsegs = 0;
1630
1631                 /*
1632                  * Report Status (RS) is turned on
1633                  * every tx_int_nsegs descriptors.
1634                  */
1635                 cmd = E1000_TXD_CMD_RS;
1636
1637                 /*
1638                  * Keep track of the descriptor, which will
1639                  * be written back by hardware.
1640                  */
1641                 adapter->tx_dd[adapter->tx_dd_tail] = last;
1642                 EM_INC_TXDD_IDX(adapter->tx_dd_tail);
1643                 KKASSERT(adapter->tx_dd_tail != adapter->tx_dd_head);
1644         }
1645
1646         /*
1647          * Last Descriptor of Packet needs End Of Packet (EOP)
1648          */
1649         ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | cmd);
1650
1651         /*
1652          * Advance the Transmit Descriptor Tail (TDT), this tells the E1000
1653          * that this frame is available to transmit.
1654          */
1655         if (adapter->hw.mac.type == e1000_82547 &&
1656             adapter->link_duplex == HALF_DUPLEX) {
1657                 em_82547_move_tail_serialized(adapter);
1658         } else {
1659                 E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), i);
1660                 if (adapter->hw.mac.type == e1000_82547) {
1661                         em_82547_update_fifo_head(adapter,
1662                             m_head->m_pkthdr.len);
1663                 }
1664         }
1665         return (0);
1666 }
1667
1668 /*
1669  * 82547 workaround to avoid controller hang in half-duplex environment.
1670  * The workaround is to avoid queuing a large packet that would span
1671  * the internal Tx FIFO ring boundary.  We need to reset the FIFO pointers
1672  * in this case.  We do that only when FIFO is quiescent.
1673  */
1674 static void
1675 em_82547_move_tail_serialized(struct adapter *adapter)
1676 {
1677         struct e1000_tx_desc *tx_desc;
1678         uint16_t hw_tdt, sw_tdt, length = 0;
1679         bool eop = 0;
1680
1681         ASSERT_SERIALIZED(adapter->arpcom.ac_if.if_serializer);
1682
1683         hw_tdt = E1000_READ_REG(&adapter->hw, E1000_TDT(0));
1684         sw_tdt = adapter->next_avail_tx_desc;
1685
1686         while (hw_tdt != sw_tdt) {
1687                 tx_desc = &adapter->tx_desc_base[hw_tdt];
1688                 length += tx_desc->lower.flags.length;
1689                 eop = tx_desc->lower.data & E1000_TXD_CMD_EOP;
1690                 if (++hw_tdt == adapter->num_tx_desc)
1691                         hw_tdt = 0;
1692
1693                 if (eop) {
1694                         if (em_82547_fifo_workaround(adapter, length)) {
1695                                 adapter->tx_fifo_wrk_cnt++;
1696                                 callout_reset(&adapter->tx_fifo_timer, 1,
1697                                         em_82547_move_tail, adapter);
1698                                 break;
1699                         }
1700                         E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), hw_tdt);
1701                         em_82547_update_fifo_head(adapter, length);
1702                         length = 0;
1703                 }
1704         }
1705 }
1706
1707 static void
1708 em_82547_move_tail(void *xsc)
1709 {
1710         struct adapter *adapter = xsc;
1711         struct ifnet *ifp = &adapter->arpcom.ac_if;
1712
1713         lwkt_serialize_enter(ifp->if_serializer);
1714         em_82547_move_tail_serialized(adapter);
1715         lwkt_serialize_exit(ifp->if_serializer);
1716 }
1717
1718 static int
1719 em_82547_fifo_workaround(struct adapter *adapter, int len)
1720 {
1721         int fifo_space, fifo_pkt_len;
1722
1723         fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1724
1725         if (adapter->link_duplex == HALF_DUPLEX) {
1726                 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
1727
1728                 if (fifo_pkt_len >= (EM_82547_PKT_THRESH + fifo_space)) {
1729                         if (em_82547_tx_fifo_reset(adapter))
1730                                 return (0);
1731                         else
1732                                 return (1);
1733                 }
1734         }
1735         return (0);
1736 }
1737
1738 static void
1739 em_82547_update_fifo_head(struct adapter *adapter, int len)
1740 {
1741         int fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1742
1743         /* tx_fifo_head is always 16 byte aligned */
1744         adapter->tx_fifo_head += fifo_pkt_len;
1745         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
1746                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
1747 }
1748
1749 static int
1750 em_82547_tx_fifo_reset(struct adapter *adapter)
1751 {
1752         uint32_t tctl;
1753
1754         if ((E1000_READ_REG(&adapter->hw, E1000_TDT(0)) ==
1755              E1000_READ_REG(&adapter->hw, E1000_TDH(0))) &&
1756             (E1000_READ_REG(&adapter->hw, E1000_TDFT) ==
1757              E1000_READ_REG(&adapter->hw, E1000_TDFH)) &&
1758             (E1000_READ_REG(&adapter->hw, E1000_TDFTS) ==
1759              E1000_READ_REG(&adapter->hw, E1000_TDFHS)) &&
1760             (E1000_READ_REG(&adapter->hw, E1000_TDFPC) == 0)) {
1761                 /* Disable TX unit */
1762                 tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
1763                 E1000_WRITE_REG(&adapter->hw, E1000_TCTL,
1764                     tctl & ~E1000_TCTL_EN);
1765
1766                 /* Reset FIFO pointers */
1767                 E1000_WRITE_REG(&adapter->hw, E1000_TDFT,
1768                     adapter->tx_head_addr);
1769                 E1000_WRITE_REG(&adapter->hw, E1000_TDFH,
1770                     adapter->tx_head_addr);
1771                 E1000_WRITE_REG(&adapter->hw, E1000_TDFTS,
1772                     adapter->tx_head_addr);
1773                 E1000_WRITE_REG(&adapter->hw, E1000_TDFHS,
1774                     adapter->tx_head_addr);
1775
1776                 /* Re-enable TX unit */
1777                 E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
1778                 E1000_WRITE_FLUSH(&adapter->hw);
1779
1780                 adapter->tx_fifo_head = 0;
1781                 adapter->tx_fifo_reset_cnt++;
1782
1783                 return (TRUE);
1784         } else {
1785                 return (FALSE);
1786         }
1787 }
1788
1789 static void
1790 em_set_promisc(struct adapter *adapter)
1791 {
1792         struct ifnet *ifp = &adapter->arpcom.ac_if;
1793         uint32_t reg_rctl;
1794
1795         reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1796
1797         if (ifp->if_flags & IFF_PROMISC) {
1798                 reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1799                 /* Turn this on if you want to see bad packets */
1800                 if (em_debug_sbp)
1801                         reg_rctl |= E1000_RCTL_SBP;
1802                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1803         } else if (ifp->if_flags & IFF_ALLMULTI) {
1804                 reg_rctl |= E1000_RCTL_MPE;
1805                 reg_rctl &= ~E1000_RCTL_UPE;
1806                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1807         }
1808 }
1809
1810 static void
1811 em_disable_promisc(struct adapter *adapter)
1812 {
1813         uint32_t reg_rctl;
1814
1815         reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1816
1817         reg_rctl &= ~E1000_RCTL_UPE;
1818         reg_rctl &= ~E1000_RCTL_MPE;
1819         reg_rctl &= ~E1000_RCTL_SBP;
1820         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1821 }
1822
1823 static void
1824 em_set_multi(struct adapter *adapter)
1825 {
1826         struct ifnet *ifp = &adapter->arpcom.ac_if;
1827         struct ifmultiaddr *ifma;
1828         uint32_t reg_rctl = 0;
1829         uint8_t  mta[512]; /* Largest MTS is 4096 bits */
1830         int mcnt = 0;
1831
1832         if (adapter->hw.mac.type == e1000_82542 &&
1833             adapter->hw.revision_id == E1000_REVISION_2) {
1834                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1835                 if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1836                         e1000_pci_clear_mwi(&adapter->hw);
1837                 reg_rctl |= E1000_RCTL_RST;
1838                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1839                 msec_delay(5);
1840         }
1841
1842         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1843                 if (ifma->ifma_addr->sa_family != AF_LINK)
1844                         continue;
1845
1846                 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1847                         break;
1848
1849                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1850                     &mta[mcnt * ETHER_ADDR_LEN], ETHER_ADDR_LEN);
1851                 mcnt++;
1852         }
1853
1854         if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1855                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1856                 reg_rctl |= E1000_RCTL_MPE;
1857                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1858         } else {
1859                 e1000_update_mc_addr_list(&adapter->hw, mta,
1860                     mcnt, 1, adapter->hw.mac.rar_entry_count);
1861         }
1862
1863         if (adapter->hw.mac.type == e1000_82542 &&
1864             adapter->hw.revision_id == E1000_REVISION_2) {
1865                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1866                 reg_rctl &= ~E1000_RCTL_RST;
1867                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1868                 msec_delay(5);
1869                 if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1870                         e1000_pci_set_mwi(&adapter->hw);
1871         }
1872 }
1873
1874 /*
1875  * This routine checks for link status and updates statistics.
1876  */
1877 static void
1878 em_timer(void *xsc)
1879 {
1880         struct adapter *adapter = xsc;
1881         struct ifnet *ifp = &adapter->arpcom.ac_if;
1882
1883         lwkt_serialize_enter(ifp->if_serializer);
1884
1885         em_update_link_status(adapter);
1886         em_update_stats(adapter);
1887
1888         /* Reset LAA into RAR[0] on 82571 */
1889         if (e1000_get_laa_state_82571(&adapter->hw) == TRUE)
1890                 e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1891
1892         if (em_display_debug_stats && (ifp->if_flags & IFF_RUNNING))
1893                 em_print_hw_stats(adapter);
1894
1895         em_smartspeed(adapter);
1896
1897         callout_reset(&adapter->timer, hz, em_timer, adapter);
1898
1899         lwkt_serialize_exit(ifp->if_serializer);
1900 }
1901
1902 static void
1903 em_update_link_status(struct adapter *adapter)
1904 {
1905         struct e1000_hw *hw = &adapter->hw;
1906         struct ifnet *ifp = &adapter->arpcom.ac_if;
1907         device_t dev = adapter->dev;
1908         uint32_t link_check = 0;
1909
1910         /* Get the cached link value or read phy for real */
1911         switch (hw->phy.media_type) {
1912         case e1000_media_type_copper:
1913                 if (hw->mac.get_link_status) {
1914                         /* Do the work to read phy */
1915                         e1000_check_for_link(hw);
1916                         link_check = !hw->mac.get_link_status;
1917                         if (link_check) /* ESB2 fix */
1918                                 e1000_cfg_on_link_up(hw);
1919                 } else {
1920                         link_check = TRUE;
1921                 }
1922                 break;
1923
1924         case e1000_media_type_fiber:
1925                 e1000_check_for_link(hw);
1926                 link_check =
1927                         E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1928                 break;
1929
1930         case e1000_media_type_internal_serdes:
1931                 e1000_check_for_link(hw);
1932                 link_check = adapter->hw.mac.serdes_has_link;
1933                 break;
1934
1935         case e1000_media_type_unknown:
1936         default:
1937                 break;
1938         }
1939
1940         /* Now check for a transition */
1941         if (link_check && adapter->link_active == 0) {
1942                 e1000_get_speed_and_duplex(hw, &adapter->link_speed,
1943                     &adapter->link_duplex);
1944
1945                 /*
1946                  * Check if we should enable/disable SPEED_MODE bit on
1947                  * 82571/82572
1948                  */
1949                 if (hw->mac.type == e1000_82571 ||
1950                     hw->mac.type == e1000_82572) {
1951                         int tarc0;
1952
1953                         tarc0 = E1000_READ_REG(hw, E1000_TARC(0));
1954                         if (adapter->link_speed != SPEED_1000)
1955                                 tarc0 &= ~SPEED_MODE_BIT;
1956                         else
1957                                 tarc0 |= SPEED_MODE_BIT;
1958                         E1000_WRITE_REG(hw, E1000_TARC(0), tarc0);
1959                 }
1960                 if (bootverbose) {
1961                         device_printf(dev, "Link is up %d Mbps %s\n",
1962                             adapter->link_speed,
1963                             ((adapter->link_duplex == FULL_DUPLEX) ?
1964                             "Full Duplex" : "Half Duplex"));
1965                 }
1966                 adapter->link_active = 1;
1967                 adapter->smartspeed = 0;
1968                 ifp->if_baudrate = adapter->link_speed * 1000000;
1969                 ifp->if_link_state = LINK_STATE_UP;
1970                 if_link_state_change(ifp);
1971         } else if (!link_check && adapter->link_active == 1) {
1972                 ifp->if_baudrate = adapter->link_speed = 0;
1973                 adapter->link_duplex = 0;
1974                 if (bootverbose)
1975                         device_printf(dev, "Link is Down\n");
1976                 adapter->link_active = 0;
1977 #if 0
1978                 /* Link down, disable watchdog */
1979                 if->if_timer = 0;
1980 #endif
1981                 ifp->if_link_state = LINK_STATE_DOWN;
1982                 if_link_state_change(ifp);
1983         }
1984 }
1985
1986 static void
1987 em_stop(struct adapter *adapter)
1988 {
1989         struct ifnet *ifp = &adapter->arpcom.ac_if;
1990         int i;
1991
1992         ASSERT_SERIALIZED(ifp->if_serializer);
1993
1994         em_disable_intr(adapter);
1995
1996         callout_stop(&adapter->timer);
1997         callout_stop(&adapter->tx_fifo_timer);
1998
1999         ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2000         ifp->if_timer = 0;
2001
2002         e1000_reset_hw(&adapter->hw);
2003         if (adapter->hw.mac.type >= e1000_82544)
2004                 E1000_WRITE_REG(&adapter->hw, E1000_WUC, 0);
2005
2006         for (i = 0; i < adapter->num_tx_desc; i++) {
2007                 struct em_buffer *tx_buffer = &adapter->tx_buffer_area[i];
2008
2009                 if (tx_buffer->m_head != NULL) {
2010                         bus_dmamap_unload(adapter->txtag, tx_buffer->map);
2011                         m_freem(tx_buffer->m_head);
2012                         tx_buffer->m_head = NULL;
2013                 }
2014         }
2015
2016         for (i = 0; i < adapter->num_rx_desc; i++) {
2017                 struct em_buffer *rx_buffer = &adapter->rx_buffer_area[i];
2018
2019                 if (rx_buffer->m_head != NULL) {
2020                         bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
2021                         m_freem(rx_buffer->m_head);
2022                         rx_buffer->m_head = NULL;
2023                 }
2024         }
2025
2026         if (adapter->fmp != NULL)
2027                 m_freem(adapter->fmp);
2028         adapter->fmp = NULL;
2029         adapter->lmp = NULL;
2030
2031         adapter->csum_flags = 0;
2032         adapter->csum_ehlen = 0;
2033         adapter->csum_iphlen = 0;
2034
2035         adapter->tx_dd_head = 0;
2036         adapter->tx_dd_tail = 0;
2037         adapter->tx_nsegs = 0;
2038 }
2039
2040 static int
2041 em_get_hw_info(struct adapter *adapter)
2042 {
2043         device_t dev = adapter->dev;
2044
2045         /* Save off the information about this board */
2046         adapter->hw.vendor_id = pci_get_vendor(dev);
2047         adapter->hw.device_id = pci_get_device(dev);
2048         adapter->hw.revision_id = pci_get_revid(dev);
2049         adapter->hw.subsystem_vendor_id = pci_get_subvendor(dev);
2050         adapter->hw.subsystem_device_id = pci_get_subdevice(dev);
2051
2052         /* Do Shared Code Init and Setup */
2053         if (e1000_set_mac_type(&adapter->hw))
2054                 return ENXIO;
2055         return 0;
2056 }
2057
2058 static int
2059 em_alloc_pci_res(struct adapter *adapter)
2060 {
2061         device_t dev = adapter->dev;
2062         int val, rid;
2063
2064         /* Enable bus mastering */
2065         pci_enable_busmaster(dev);
2066
2067         adapter->memory_rid = EM_BAR_MEM;
2068         adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
2069                                 &adapter->memory_rid, RF_ACTIVE);
2070         if (adapter->memory == NULL) {
2071                 device_printf(dev, "Unable to allocate bus resource: memory\n");
2072                 return (ENXIO);
2073         }
2074         adapter->osdep.mem_bus_space_tag =
2075             rman_get_bustag(adapter->memory);
2076         adapter->osdep.mem_bus_space_handle =
2077             rman_get_bushandle(adapter->memory);
2078
2079         /* XXX This is quite goofy, it is not actually used */
2080         adapter->hw.hw_addr = (uint8_t *)&adapter->osdep.mem_bus_space_handle;
2081
2082         /* Only older adapters use IO mapping */
2083         if (adapter->hw.mac.type > e1000_82543 &&
2084             adapter->hw.mac.type < e1000_82571) {
2085                 /* Figure our where our IO BAR is ? */
2086                 for (rid = PCIR_BAR(0); rid < PCIR_CARDBUSCIS;) {
2087                         val = pci_read_config(dev, rid, 4);
2088                         if (EM_BAR_TYPE(val) == EM_BAR_TYPE_IO) {
2089                                 adapter->io_rid = rid;
2090                                 break;
2091                         }
2092                         rid += 4;
2093                         /* check for 64bit BAR */
2094                         if (EM_BAR_MEM_TYPE(val) == EM_BAR_MEM_TYPE_64BIT)
2095                                 rid += 4;
2096                 }
2097                 if (rid >= PCIR_CARDBUSCIS) {
2098                         device_printf(dev, "Unable to locate IO BAR\n");
2099                         return (ENXIO);
2100                 }
2101                 adapter->ioport = bus_alloc_resource_any(dev, SYS_RES_IOPORT,
2102                                         &adapter->io_rid, RF_ACTIVE);
2103                 if (adapter->ioport == NULL) {
2104                         device_printf(dev, "Unable to allocate bus resource: "
2105                             "ioport\n");
2106                         return (ENXIO);
2107                 }
2108                 adapter->hw.io_base = 0;
2109                 adapter->osdep.io_bus_space_tag =
2110                     rman_get_bustag(adapter->ioport);
2111                 adapter->osdep.io_bus_space_handle =
2112                     rman_get_bushandle(adapter->ioport);
2113         }
2114
2115         adapter->intr_rid = 0;
2116         adapter->intr_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
2117                                 &adapter->intr_rid,
2118                                 RF_SHAREABLE | RF_ACTIVE);
2119         if (adapter->intr_res == NULL) {
2120                 device_printf(dev, "Unable to allocate bus resource: "
2121                     "interrupt\n");
2122                 return (ENXIO);
2123         }
2124
2125         adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
2126         adapter->hw.back = &adapter->osdep;
2127         return (0);
2128 }
2129
2130 static void
2131 em_free_pci_res(struct adapter *adapter)
2132 {
2133         device_t dev = adapter->dev;
2134
2135         if (adapter->intr_res != NULL) {
2136                 bus_release_resource(dev, SYS_RES_IRQ,
2137                     adapter->intr_rid, adapter->intr_res);
2138         }
2139
2140         if (adapter->memory != NULL) {
2141                 bus_release_resource(dev, SYS_RES_MEMORY,
2142                     adapter->memory_rid, adapter->memory);
2143         }
2144
2145         if (adapter->flash != NULL) {
2146                 bus_release_resource(dev, SYS_RES_MEMORY,
2147                     adapter->flash_rid, adapter->flash);
2148         }
2149
2150         if (adapter->ioport != NULL) {
2151                 bus_release_resource(dev, SYS_RES_IOPORT,
2152                     adapter->io_rid, adapter->ioport);
2153         }
2154 }
2155
2156 static int
2157 em_hw_init(struct adapter *adapter)
2158 {
2159         device_t dev = adapter->dev;
2160         uint16_t rx_buffer_size;
2161
2162         /* Issue a global reset */
2163         e1000_reset_hw(&adapter->hw);
2164
2165         /* Get control from any management/hw control */
2166         if ((adapter->hw.mac.type == e1000_82573 ||
2167              adapter->hw.mac.type == e1000_ich8lan ||
2168              adapter->hw.mac.type == e1000_ich10lan ||
2169              adapter->hw.mac.type == e1000_ich9lan) &&
2170             e1000_check_mng_mode(&adapter->hw))
2171                 em_get_hw_control(adapter);
2172
2173         /* When hardware is reset, fifo_head is also reset */
2174         adapter->tx_fifo_head = 0;
2175
2176         /* Set up smart power down as default off on newer adapters. */
2177         if (!em_smart_pwr_down &&
2178             (adapter->hw.mac.type == e1000_82571 ||
2179              adapter->hw.mac.type == e1000_82572)) {
2180                 uint16_t phy_tmp = 0;
2181
2182                 /* Speed up time to link by disabling smart power down. */
2183                 e1000_read_phy_reg(&adapter->hw,
2184                     IGP02E1000_PHY_POWER_MGMT, &phy_tmp);
2185                 phy_tmp &= ~IGP02E1000_PM_SPD;
2186                 e1000_write_phy_reg(&adapter->hw,
2187                     IGP02E1000_PHY_POWER_MGMT, phy_tmp);
2188         }
2189
2190         /*
2191          * These parameters control the automatic generation (Tx) and
2192          * response (Rx) to Ethernet PAUSE frames.
2193          * - High water mark should allow for at least two frames to be
2194          *   received after sending an XOFF.
2195          * - Low water mark works best when it is very near the high water mark.
2196          *   This allows the receiver to restart by sending XON when it has
2197          *   drained a bit. Here we use an arbitary value of 1500 which will
2198          *   restart after one full frame is pulled from the buffer. There
2199          *   could be several smaller frames in the buffer and if so they will
2200          *   not trigger the XON until their total number reduces the buffer
2201          *   by 1500.
2202          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
2203          */
2204         rx_buffer_size =
2205                 (E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff) << 10;
2206
2207         adapter->hw.fc.high_water = rx_buffer_size -
2208                                     roundup2(adapter->max_frame_size, 1024);
2209         adapter->hw.fc.low_water = adapter->hw.fc.high_water - 1500;
2210
2211         if (adapter->hw.mac.type == e1000_80003es2lan)
2212                 adapter->hw.fc.pause_time = 0xFFFF;
2213         else
2214                 adapter->hw.fc.pause_time = EM_FC_PAUSE_TIME;
2215         adapter->hw.fc.send_xon = TRUE;
2216         adapter->hw.fc.requested_mode = e1000_fc_full;
2217
2218         if (e1000_init_hw(&adapter->hw) < 0) {
2219                 device_printf(dev, "Hardware Initialization Failed\n");
2220                 return (EIO);
2221         }
2222
2223         e1000_check_for_link(&adapter->hw);
2224
2225         return (0);
2226 }
2227
2228 static void
2229 em_setup_ifp(struct adapter *adapter)
2230 {
2231         struct ifnet *ifp = &adapter->arpcom.ac_if;
2232
2233         if_initname(ifp, device_get_name(adapter->dev),
2234                     device_get_unit(adapter->dev));
2235         ifp->if_softc = adapter;
2236         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2237         ifp->if_init =  em_init;
2238         ifp->if_ioctl = em_ioctl;
2239         ifp->if_start = em_start;
2240 #ifdef DEVICE_POLLING
2241         ifp->if_poll = em_poll;
2242 #endif
2243         ifp->if_watchdog = em_watchdog;
2244         ifq_set_maxlen(&ifp->if_snd, adapter->num_tx_desc - 1);
2245         ifq_set_ready(&ifp->if_snd);
2246
2247         ether_ifattach(ifp, adapter->hw.mac.addr, NULL);
2248
2249         if (adapter->hw.mac.type >= e1000_82543)
2250                 ifp->if_capabilities = IFCAP_HWCSUM;
2251
2252         ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2253         ifp->if_capenable = ifp->if_capabilities;
2254
2255         if (ifp->if_capenable & IFCAP_TXCSUM)
2256                 ifp->if_hwassist = EM_CSUM_FEATURES;
2257
2258         /*
2259          * Tell the upper layer(s) we support long frames.
2260          */
2261         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
2262
2263         /*
2264          * Specify the media types supported by this adapter and register
2265          * callbacks to update media and link information
2266          */
2267         ifmedia_init(&adapter->media, IFM_IMASK,
2268                      em_media_change, em_media_status);
2269         if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
2270             adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
2271                 u_char fiber_type = IFM_1000_SX; /* default type */
2272
2273                 if (adapter->hw.mac.type == e1000_82545)
2274                         fiber_type = IFM_1000_LX;
2275                 ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX,
2276                             0, NULL);
2277                 ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL);
2278         } else {
2279                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
2280                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
2281                             0, NULL);
2282                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
2283                             0, NULL);
2284                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
2285                             0, NULL);
2286                 if (adapter->hw.phy.type != e1000_phy_ife) {
2287                         ifmedia_add(&adapter->media,
2288                                 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
2289                         ifmedia_add(&adapter->media,
2290                                 IFM_ETHER | IFM_1000_T, 0, NULL);
2291                 }
2292         }
2293         ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2294         ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2295 }
2296
2297
2298 /*
2299  * Workaround for SmartSpeed on 82541 and 82547 controllers
2300  */
2301 static void
2302 em_smartspeed(struct adapter *adapter)
2303 {
2304         uint16_t phy_tmp;
2305
2306         if (adapter->link_active || adapter->hw.phy.type != e1000_phy_igp ||
2307             adapter->hw.mac.autoneg == 0 ||
2308             (adapter->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0)
2309                 return;
2310
2311         if (adapter->smartspeed == 0) {
2312                 /*
2313                  * If Master/Slave config fault is asserted twice,
2314                  * we assume back-to-back
2315                  */
2316                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2317                 if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
2318                         return;
2319                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2320                 if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
2321                         e1000_read_phy_reg(&adapter->hw,
2322                             PHY_1000T_CTRL, &phy_tmp);
2323                         if (phy_tmp & CR_1000T_MS_ENABLE) {
2324                                 phy_tmp &= ~CR_1000T_MS_ENABLE;
2325                                 e1000_write_phy_reg(&adapter->hw,
2326                                     PHY_1000T_CTRL, phy_tmp);
2327                                 adapter->smartspeed++;
2328                                 if (adapter->hw.mac.autoneg &&
2329                                     !e1000_phy_setup_autoneg(&adapter->hw) &&
2330                                     !e1000_read_phy_reg(&adapter->hw,
2331                                      PHY_CONTROL, &phy_tmp)) {
2332                                         phy_tmp |= MII_CR_AUTO_NEG_EN |
2333                                                    MII_CR_RESTART_AUTO_NEG;
2334                                         e1000_write_phy_reg(&adapter->hw,
2335                                             PHY_CONTROL, phy_tmp);
2336                                 }
2337                         }
2338                 }
2339                 return;
2340         } else if (adapter->smartspeed == EM_SMARTSPEED_DOWNSHIFT) {
2341                 /* If still no link, perhaps using 2/3 pair cable */
2342                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp);
2343                 phy_tmp |= CR_1000T_MS_ENABLE;
2344                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp);
2345                 if (adapter->hw.mac.autoneg &&
2346                     !e1000_phy_setup_autoneg(&adapter->hw) &&
2347                     !e1000_read_phy_reg(&adapter->hw, PHY_CONTROL, &phy_tmp)) {
2348                         phy_tmp |= MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG;
2349                         e1000_write_phy_reg(&adapter->hw, PHY_CONTROL, phy_tmp);
2350                 }
2351         }
2352
2353         /* Restart process after EM_SMARTSPEED_MAX iterations */
2354         if (adapter->smartspeed++ == EM_SMARTSPEED_MAX)
2355                 adapter->smartspeed = 0;
2356 }
2357
2358 static int
2359 em_dma_malloc(struct adapter *adapter, bus_size_t size,
2360               struct em_dma_alloc *dma)
2361 {
2362         dma->dma_vaddr = bus_dmamem_coherent_any(adapter->parent_dtag,
2363                                 EM_DBA_ALIGN, size, BUS_DMA_WAITOK,
2364                                 &dma->dma_tag, &dma->dma_map,
2365                                 &dma->dma_paddr);
2366         if (dma->dma_vaddr == NULL)
2367                 return ENOMEM;
2368         else
2369                 return 0;
2370 }
2371
2372 static void
2373 em_dma_free(struct adapter *adapter, struct em_dma_alloc *dma)
2374 {
2375         if (dma->dma_tag == NULL)
2376                 return;
2377         bus_dmamap_unload(dma->dma_tag, dma->dma_map);
2378         bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
2379         bus_dma_tag_destroy(dma->dma_tag);
2380 }
2381
2382 static int
2383 em_create_tx_ring(struct adapter *adapter)
2384 {
2385         device_t dev = adapter->dev;
2386         struct em_buffer *tx_buffer;
2387         int error, i;
2388
2389         adapter->tx_buffer_area =
2390                 kmalloc(sizeof(struct em_buffer) * adapter->num_tx_desc,
2391                         M_DEVBUF, M_WAITOK | M_ZERO);
2392
2393         /*
2394          * Create DMA tags for tx buffers
2395          */
2396         error = bus_dma_tag_create(adapter->parent_dtag, /* parent */
2397                         1, 0,                   /* alignment, bounds */
2398                         BUS_SPACE_MAXADDR,      /* lowaddr */
2399                         BUS_SPACE_MAXADDR,      /* highaddr */
2400                         NULL, NULL,             /* filter, filterarg */
2401                         EM_TSO_SIZE,            /* maxsize */
2402                         EM_MAX_SCATTER,         /* nsegments */
2403                         EM_MAX_SEGSIZE,         /* maxsegsize */
2404                         BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
2405                         BUS_DMA_ONEBPAGE,       /* flags */
2406                         &adapter->txtag);
2407         if (error) {
2408                 device_printf(dev, "Unable to allocate TX DMA tag\n");
2409                 kfree(adapter->tx_buffer_area, M_DEVBUF);
2410                 adapter->tx_buffer_area = NULL;
2411                 return error;
2412         }
2413
2414         /*
2415          * Create DMA maps for tx buffers
2416          */
2417         for (i = 0; i < adapter->num_tx_desc; i++) {
2418                 tx_buffer = &adapter->tx_buffer_area[i];
2419
2420                 error = bus_dmamap_create(adapter->txtag,
2421                                           BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
2422                                           &tx_buffer->map);
2423                 if (error) {
2424                         device_printf(dev, "Unable to create TX DMA map\n");
2425                         em_destroy_tx_ring(adapter, i);
2426                         return error;
2427                 }
2428         }
2429         return (0);
2430 }
2431
2432 static void
2433 em_init_tx_ring(struct adapter *adapter)
2434 {
2435         /* Clear the old ring contents */
2436         bzero(adapter->tx_desc_base,
2437             (sizeof(struct e1000_tx_desc)) * adapter->num_tx_desc);
2438
2439         /* Reset state */
2440         adapter->next_avail_tx_desc = 0;
2441         adapter->next_tx_to_clean = 0;
2442         adapter->num_tx_desc_avail = adapter->num_tx_desc;
2443 }
2444
2445 static void
2446 em_init_tx_unit(struct adapter *adapter)
2447 {
2448         uint32_t tctl, tarc, tipg = 0;
2449         uint64_t bus_addr;
2450
2451         /* Setup the Base and Length of the Tx Descriptor Ring */
2452         bus_addr = adapter->txdma.dma_paddr;
2453         E1000_WRITE_REG(&adapter->hw, E1000_TDLEN(0),
2454             adapter->num_tx_desc * sizeof(struct e1000_tx_desc));
2455         E1000_WRITE_REG(&adapter->hw, E1000_TDBAH(0),
2456             (uint32_t)(bus_addr >> 32));
2457         E1000_WRITE_REG(&adapter->hw, E1000_TDBAL(0),
2458             (uint32_t)bus_addr);
2459         /* Setup the HW Tx Head and Tail descriptor pointers */
2460         E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), 0);
2461         E1000_WRITE_REG(&adapter->hw, E1000_TDH(0), 0);
2462
2463         /* Set the default values for the Tx Inter Packet Gap timer */
2464         switch (adapter->hw.mac.type) {
2465         case e1000_82542:
2466                 tipg = DEFAULT_82542_TIPG_IPGT;
2467                 tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2468                 tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2469                 break;
2470
2471         case e1000_80003es2lan:
2472                 tipg = DEFAULT_82543_TIPG_IPGR1;
2473                 tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 <<
2474                     E1000_TIPG_IPGR2_SHIFT;
2475                 break;
2476
2477         default:
2478                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
2479                     adapter->hw.phy.media_type ==
2480                     e1000_media_type_internal_serdes)
2481                         tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
2482                 else
2483                         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
2484                 tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2485                 tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2486                 break;
2487         }
2488
2489         E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg);
2490
2491         /* NOTE: 0 is not allowed for TIDV */
2492         E1000_WRITE_REG(&adapter->hw, E1000_TIDV, 1);
2493         if(adapter->hw.mac.type >= e1000_82540)
2494                 E1000_WRITE_REG(&adapter->hw, E1000_TADV, 0);
2495
2496         if (adapter->hw.mac.type == e1000_82571 ||
2497             adapter->hw.mac.type == e1000_82572) {
2498                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
2499                 tarc |= SPEED_MODE_BIT;
2500                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
2501         } else if (adapter->hw.mac.type == e1000_80003es2lan) {
2502                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
2503                 tarc |= 1;
2504                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
2505                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(1));
2506                 tarc |= 1;
2507                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc);
2508         }
2509
2510         /* Program the Transmit Control Register */
2511         tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
2512         tctl &= ~E1000_TCTL_CT;
2513         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
2514                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2515
2516         if (adapter->hw.mac.type >= e1000_82571)
2517                 tctl |= E1000_TCTL_MULR;
2518
2519         /* This write will effectively turn on the transmit unit. */
2520         E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
2521 }
2522
2523 static void
2524 em_destroy_tx_ring(struct adapter *adapter, int ndesc)
2525 {
2526         struct em_buffer *tx_buffer;
2527         int i;
2528
2529         if (adapter->tx_buffer_area == NULL)
2530                 return;
2531
2532         for (i = 0; i < ndesc; i++) {
2533                 tx_buffer = &adapter->tx_buffer_area[i];
2534
2535                 KKASSERT(tx_buffer->m_head == NULL);
2536                 bus_dmamap_destroy(adapter->txtag, tx_buffer->map);
2537         }
2538         bus_dma_tag_destroy(adapter->txtag);
2539
2540         kfree(adapter->tx_buffer_area, M_DEVBUF);
2541         adapter->tx_buffer_area = NULL;
2542 }
2543
2544 /*
2545  * The offload context needs to be set when we transfer the first
2546  * packet of a particular protocol (TCP/UDP).  This routine has been
2547  * enhanced to deal with inserted VLAN headers.
2548  *
2549  * If the new packet's ether header length, ip header length and
2550  * csum offloading type are same as the previous packet, we should
2551  * avoid allocating a new csum context descriptor; mainly to take
2552  * advantage of the pipeline effect of the TX data read request.
2553  *
2554  * This function returns number of TX descrptors allocated for
2555  * csum context.
2556  */
2557 static int
2558 em_txcsum(struct adapter *adapter, struct mbuf *mp,
2559           uint32_t *txd_upper, uint32_t *txd_lower)
2560 {
2561         struct e1000_context_desc *TXD;
2562         struct em_buffer *tx_buffer;
2563         struct ether_vlan_header *eh;
2564         struct ip *ip;
2565         int curr_txd, ehdrlen, csum_flags;
2566         uint32_t cmd, hdr_len, ip_hlen;
2567         uint16_t etype;
2568
2569         /*
2570          * Determine where frame payload starts.
2571          * Jump over vlan headers if already present,
2572          * helpful for QinQ too.
2573          */
2574         KASSERT(mp->m_len >= ETHER_HDR_LEN,
2575                 ("em_txcsum_pullup is not called (eh)?\n"));
2576         eh = mtod(mp, struct ether_vlan_header *);
2577         if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2578                 KASSERT(mp->m_len >= ETHER_HDR_LEN + EVL_ENCAPLEN,
2579                         ("em_txcsum_pullup is not called (evh)?\n"));
2580                 etype = ntohs(eh->evl_proto);
2581                 ehdrlen = ETHER_HDR_LEN + EVL_ENCAPLEN;
2582         } else {
2583                 etype = ntohs(eh->evl_encap_proto);
2584                 ehdrlen = ETHER_HDR_LEN;
2585         }
2586
2587         /*
2588          * We only support TCP/UDP for IPv4 for the moment.
2589          * TODO: Support SCTP too when it hits the tree.
2590          */
2591         if (etype != ETHERTYPE_IP)
2592                 return 0;
2593
2594         KASSERT(mp->m_len >= ehdrlen + EM_IPVHL_SIZE,
2595                 ("em_txcsum_pullup is not called (eh+ip_vhl)?\n"));
2596
2597         /* NOTE: We could only safely access ip.ip_vhl part */
2598         ip = (struct ip *)(mp->m_data + ehdrlen);
2599         ip_hlen = ip->ip_hl << 2;
2600
2601         csum_flags = mp->m_pkthdr.csum_flags & EM_CSUM_FEATURES;
2602
2603         if (adapter->csum_ehlen == ehdrlen &&
2604             adapter->csum_iphlen == ip_hlen &&
2605             adapter->csum_flags == csum_flags) {
2606                 /*
2607                  * Same csum offload context as the previous packets;
2608                  * just return.
2609                  */
2610                 *txd_upper = adapter->csum_txd_upper;
2611                 *txd_lower = adapter->csum_txd_lower;
2612                 return 0;
2613         }
2614
2615         /*
2616          * Setup a new csum offload context.
2617          */
2618
2619         curr_txd = adapter->next_avail_tx_desc;
2620         tx_buffer = &adapter->tx_buffer_area[curr_txd];
2621         TXD = (struct e1000_context_desc *)&adapter->tx_desc_base[curr_txd];
2622
2623         cmd = 0;
2624
2625         /* Setup of IP header checksum. */
2626         if (csum_flags & CSUM_IP) {
2627                 /*
2628                  * Start offset for header checksum calculation.
2629                  * End offset for header checksum calculation.
2630                  * Offset of place to put the checksum.
2631                  */
2632                 TXD->lower_setup.ip_fields.ipcss = ehdrlen;
2633                 TXD->lower_setup.ip_fields.ipcse =
2634                     htole16(ehdrlen + ip_hlen - 1);
2635                 TXD->lower_setup.ip_fields.ipcso =
2636                     ehdrlen + offsetof(struct ip, ip_sum);
2637                 cmd |= E1000_TXD_CMD_IP;
2638                 *txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2639         }
2640         hdr_len = ehdrlen + ip_hlen;
2641
2642         if (csum_flags & CSUM_TCP) {
2643                 /*
2644                  * Start offset for payload checksum calculation.
2645                  * End offset for payload checksum calculation.
2646                  * Offset of place to put the checksum.
2647                  */
2648                 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2649                 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2650                 TXD->upper_setup.tcp_fields.tucso =
2651                     hdr_len + offsetof(struct tcphdr, th_sum);
2652                 cmd |= E1000_TXD_CMD_TCP;
2653                 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2654         } else if (csum_flags & CSUM_UDP) {
2655                 /*
2656                  * Start offset for header checksum calculation.
2657                  * End offset for header checksum calculation.
2658                  * Offset of place to put the checksum.
2659                  */
2660                 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2661                 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2662                 TXD->upper_setup.tcp_fields.tucso =
2663                     hdr_len + offsetof(struct udphdr, uh_sum);
2664                 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2665         }
2666
2667         *txd_lower = E1000_TXD_CMD_DEXT |       /* Extended descr type */
2668                      E1000_TXD_DTYP_D;          /* Data descr */
2669
2670         /* Save the information for this csum offloading context */
2671         adapter->csum_ehlen = ehdrlen;
2672         adapter->csum_iphlen = ip_hlen;
2673         adapter->csum_flags = csum_flags;
2674         adapter->csum_txd_upper = *txd_upper;
2675         adapter->csum_txd_lower = *txd_lower;
2676
2677         TXD->tcp_seg_setup.data = htole32(0);
2678         TXD->cmd_and_length =
2679             htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
2680
2681         if (++curr_txd == adapter->num_tx_desc)
2682                 curr_txd = 0;
2683
2684         KKASSERT(adapter->num_tx_desc_avail > 0);
2685         adapter->num_tx_desc_avail--;
2686
2687         adapter->next_avail_tx_desc = curr_txd;
2688         return 1;
2689 }
2690
2691 static int
2692 em_txcsum_pullup(struct adapter *adapter, struct mbuf **m0)
2693 {
2694         struct mbuf *m = *m0;
2695         struct ether_header *eh;
2696         int len;
2697
2698         adapter->tx_csum_try_pullup++;
2699
2700         len = ETHER_HDR_LEN + EM_IPVHL_SIZE;
2701
2702         if (__predict_false(!M_WRITABLE(m))) {
2703                 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2704                         adapter->tx_csum_drop1++;
2705                         m_freem(m);
2706                         *m0 = NULL;
2707                         return ENOBUFS;
2708                 }
2709                 eh = mtod(m, struct ether_header *);
2710
2711                 if (eh->ether_type == htons(ETHERTYPE_VLAN))
2712                         len += EVL_ENCAPLEN;
2713
2714                 if (m->m_len < len) {
2715                         adapter->tx_csum_drop2++;
2716                         m_freem(m);
2717                         *m0 = NULL;
2718                         return ENOBUFS;
2719                 }
2720                 return 0;
2721         }
2722
2723         if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2724                 adapter->tx_csum_pullup1++;
2725                 m = m_pullup(m, ETHER_HDR_LEN);
2726                 if (m == NULL) {
2727                         adapter->tx_csum_pullup1_failed++;
2728                         *m0 = NULL;
2729                         return ENOBUFS;
2730                 }
2731                 *m0 = m;
2732         }
2733         eh = mtod(m, struct ether_header *);
2734
2735         if (eh->ether_type == htons(ETHERTYPE_VLAN))
2736                 len += EVL_ENCAPLEN;
2737
2738         if (m->m_len < len) {
2739                 adapter->tx_csum_pullup2++;
2740                 m = m_pullup(m, len);
2741                 if (m == NULL) {
2742                         adapter->tx_csum_pullup2_failed++;
2743                         *m0 = NULL;
2744                         return ENOBUFS;
2745                 }
2746                 *m0 = m;
2747         }
2748         return 0;
2749 }
2750
2751 static void
2752 em_txeof(struct adapter *adapter)
2753 {
2754         struct ifnet *ifp = &adapter->arpcom.ac_if;
2755         struct em_buffer *tx_buffer;
2756         int first, num_avail;
2757
2758         if (adapter->tx_dd_head == adapter->tx_dd_tail)
2759                 return;
2760
2761         if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2762                 return;
2763
2764         num_avail = adapter->num_tx_desc_avail;
2765         first = adapter->next_tx_to_clean;
2766
2767         while (adapter->tx_dd_head != adapter->tx_dd_tail) {
2768                 struct e1000_tx_desc *tx_desc;
2769                 int dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2770
2771                 tx_desc = &adapter->tx_desc_base[dd_idx];
2772                 if (tx_desc->upper.fields.status & E1000_TXD_STAT_DD) {
2773                         EM_INC_TXDD_IDX(adapter->tx_dd_head);
2774
2775                         if (++dd_idx == adapter->num_tx_desc)
2776                                 dd_idx = 0;
2777
2778                         while (first != dd_idx) {
2779                                 logif(pkt_txclean);
2780
2781                                 num_avail++;
2782
2783                                 tx_buffer = &adapter->tx_buffer_area[first];
2784                                 if (tx_buffer->m_head) {
2785                                         ifp->if_opackets++;
2786                                         bus_dmamap_unload(adapter->txtag,
2787                                                           tx_buffer->map);
2788                                         m_freem(tx_buffer->m_head);
2789                                         tx_buffer->m_head = NULL;
2790                                 }
2791
2792                                 if (++first == adapter->num_tx_desc)
2793                                         first = 0;
2794                         }
2795                 } else {
2796                         break;
2797                 }
2798         }
2799         adapter->next_tx_to_clean = first;
2800         adapter->num_tx_desc_avail = num_avail;
2801
2802         if (adapter->tx_dd_head == adapter->tx_dd_tail) {
2803                 adapter->tx_dd_head = 0;
2804                 adapter->tx_dd_tail = 0;
2805         }
2806
2807         if (!EM_IS_OACTIVE(adapter)) {
2808                 ifp->if_flags &= ~IFF_OACTIVE;
2809
2810                 /* All clean, turn off the timer */
2811                 if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2812                         ifp->if_timer = 0;
2813         }
2814 }
2815
2816 static void
2817 em_tx_collect(struct adapter *adapter)
2818 {
2819         struct ifnet *ifp = &adapter->arpcom.ac_if;
2820         struct em_buffer *tx_buffer;
2821         int tdh, first, num_avail, dd_idx = -1;
2822
2823         if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2824                 return;
2825
2826         tdh = E1000_READ_REG(&adapter->hw, E1000_TDH(0));
2827         if (tdh == adapter->next_tx_to_clean)
2828                 return;
2829
2830         if (adapter->tx_dd_head != adapter->tx_dd_tail)
2831                 dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2832
2833         num_avail = adapter->num_tx_desc_avail;
2834         first = adapter->next_tx_to_clean;
2835
2836         while (first != tdh) {
2837                 logif(pkt_txclean);
2838
2839                 num_avail++;
2840
2841                 tx_buffer = &adapter->tx_buffer_area[first];
2842                 if (tx_buffer->m_head) {
2843                         ifp->if_opackets++;
2844                         bus_dmamap_unload(adapter->txtag,
2845                                           tx_buffer->map);
2846                         m_freem(tx_buffer->m_head);
2847                         tx_buffer->m_head = NULL;
2848                 }
2849
2850                 if (first == dd_idx) {
2851                         EM_INC_TXDD_IDX(adapter->tx_dd_head);
2852                         if (adapter->tx_dd_head == adapter->tx_dd_tail) {
2853                                 adapter->tx_dd_head = 0;
2854                                 adapter->tx_dd_tail = 0;
2855                                 dd_idx = -1;
2856                         } else {
2857                                 dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2858                         }
2859                 }
2860
2861                 if (++first == adapter->num_tx_desc)
2862                         first = 0;
2863         }
2864         adapter->next_tx_to_clean = first;
2865         adapter->num_tx_desc_avail = num_avail;
2866
2867         if (!EM_IS_OACTIVE(adapter)) {
2868                 ifp->if_flags &= ~IFF_OACTIVE;
2869
2870                 /* All clean, turn off the timer */
2871                 if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2872                         ifp->if_timer = 0;
2873         }
2874 }
2875
2876 /*
2877  * When Link is lost sometimes there is work still in the TX ring
2878  * which will result in a watchdog, rather than allow that do an
2879  * attempted cleanup and then reinit here.  Note that this has been
2880  * seens mostly with fiber adapters.
2881  */
2882 static void
2883 em_tx_purge(struct adapter *adapter)
2884 {
2885         struct ifnet *ifp = &adapter->arpcom.ac_if;
2886
2887         if (!adapter->link_active && ifp->if_timer) {
2888                 em_tx_collect(adapter);
2889                 if (ifp->if_timer) {
2890                         if_printf(ifp, "Link lost, TX pending, reinit\n");
2891                         ifp->if_timer = 0;
2892                         em_init(adapter);
2893                 }
2894         }
2895 }
2896
2897 static int
2898 em_newbuf(struct adapter *adapter, int i, int init)
2899 {
2900         struct mbuf *m;
2901         bus_dma_segment_t seg;
2902         bus_dmamap_t map;
2903         struct em_buffer *rx_buffer;
2904         int error, nseg;
2905
2906         m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2907         if (m == NULL) {
2908                 adapter->mbuf_cluster_failed++;
2909                 if (init) {
2910                         if_printf(&adapter->arpcom.ac_if,
2911                                   "Unable to allocate RX mbuf\n");
2912                 }
2913                 return (ENOBUFS);
2914         }
2915         m->m_len = m->m_pkthdr.len = MCLBYTES;
2916
2917         if (adapter->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2918                 m_adj(m, ETHER_ALIGN);
2919
2920         error = bus_dmamap_load_mbuf_segment(adapter->rxtag,
2921                         adapter->rx_sparemap, m,
2922                         &seg, 1, &nseg, BUS_DMA_NOWAIT);
2923         if (error) {
2924                 m_freem(m);
2925                 if (init) {
2926                         if_printf(&adapter->arpcom.ac_if,
2927                                   "Unable to load RX mbuf\n");
2928                 }
2929                 return (error);
2930         }
2931
2932         rx_buffer = &adapter->rx_buffer_area[i];
2933         if (rx_buffer->m_head != NULL)
2934                 bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
2935
2936         map = rx_buffer->map;
2937         rx_buffer->map = adapter->rx_sparemap;
2938         adapter->rx_sparemap = map;
2939
2940         rx_buffer->m_head = m;
2941
2942         adapter->rx_desc_base[i].buffer_addr = htole64(seg.ds_addr);
2943         return (0);
2944 }
2945
2946 static int
2947 em_create_rx_ring(struct adapter *adapter)
2948 {
2949         device_t dev = adapter->dev;
2950         struct em_buffer *rx_buffer;
2951         int i, error;
2952
2953         adapter->rx_buffer_area =
2954                 kmalloc(sizeof(struct em_buffer) * adapter->num_rx_desc,
2955                         M_DEVBUF, M_WAITOK | M_ZERO);
2956
2957         /*
2958          * Create DMA tag for rx buffers
2959          */
2960         error = bus_dma_tag_create(adapter->parent_dtag, /* parent */
2961                         1, 0,                   /* alignment, bounds */
2962                         BUS_SPACE_MAXADDR,      /* lowaddr */
2963                         BUS_SPACE_MAXADDR,      /* highaddr */
2964                         NULL, NULL,             /* filter, filterarg */
2965                         MCLBYTES,               /* maxsize */
2966                         1,                      /* nsegments */
2967                         MCLBYTES,               /* maxsegsize */
2968                         BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2969                         &adapter->rxtag);
2970         if (error) {
2971                 device_printf(dev, "Unable to allocate RX DMA tag\n");
2972                 kfree(adapter->rx_buffer_area, M_DEVBUF);
2973                 adapter->rx_buffer_area = NULL;
2974                 return error;
2975         }
2976
2977         /*
2978          * Create spare DMA map for rx buffers
2979          */
2980         error = bus_dmamap_create(adapter->rxtag, BUS_DMA_WAITOK,
2981                                   &adapter->rx_sparemap);
2982         if (error) {
2983                 device_printf(dev, "Unable to create spare RX DMA map\n");
2984                 bus_dma_tag_destroy(adapter->rxtag);
2985                 kfree(adapter->rx_buffer_area, M_DEVBUF);
2986                 adapter->rx_buffer_area = NULL;
2987                 return error;
2988         }
2989
2990         /*
2991          * Create DMA maps for rx buffers
2992          */
2993         for (i = 0; i < adapter->num_rx_desc; i++) {
2994                 rx_buffer = &adapter->rx_buffer_area[i];
2995
2996                 error = bus_dmamap_create(adapter->rxtag, BUS_DMA_WAITOK,
2997                                           &rx_buffer->map);
2998                 if (error) {
2999                         device_printf(dev, "Unable to create RX DMA map\n");
3000                         em_destroy_rx_ring(adapter, i);
3001                         return error;
3002                 }
3003         }
3004         return (0);
3005 }
3006
3007 static int
3008 em_init_rx_ring(struct adapter *adapter)
3009 {
3010         int i, error;
3011
3012         /* Reset descriptor ring */
3013         bzero(adapter->rx_desc_base,
3014             (sizeof(struct e1000_rx_desc)) * adapter->num_rx_desc);
3015
3016         /* Allocate new ones. */
3017         for (i = 0; i < adapter->num_rx_desc; i++) {
3018                 error = em_newbuf(adapter, i, 1);
3019                 if (error)
3020                         return (error);
3021         }
3022
3023         /* Setup our descriptor pointers */
3024         adapter->next_rx_desc_to_check = 0;
3025
3026         return (0);
3027 }
3028
3029 static void
3030 em_init_rx_unit(struct adapter *adapter)
3031 {
3032         struct ifnet *ifp = &adapter->arpcom.ac_if;
3033         uint64_t bus_addr;
3034         uint32_t rctl, rxcsum;
3035
3036         /*
3037          * Make sure receives are disabled while setting
3038          * up the descriptor ring
3039          */
3040         rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
3041         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
3042
3043         if (adapter->hw.mac.type >= e1000_82540) {
3044                 /*
3045                  * Set the interrupt throttling rate. Value is calculated
3046                  * as ITR = 1 / (INT_THROTTLE_CEIL * 256ns)
3047                  */
3048                 if (adapter->int_throttle_ceil) {
3049                         E1000_WRITE_REG(&adapter->hw, E1000_ITR,
3050                                 1000000000 / 256 / adapter->int_throttle_ceil);
3051                 } else {
3052                         E1000_WRITE_REG(&adapter->hw, E1000_ITR, 0);
3053                 }
3054         }
3055
3056         /* Disable accelerated ackknowledge */
3057         if (adapter->hw.mac.type == e1000_82574) {
3058                 E1000_WRITE_REG(&adapter->hw,
3059                     E1000_RFCTL, E1000_RFCTL_ACK_DIS);
3060         }
3061
3062         /* Setup the Base and Length of the Rx Descriptor Ring */
3063         bus_addr = adapter->rxdma.dma_paddr;
3064         E1000_WRITE_REG(&adapter->hw, E1000_RDLEN(0),
3065             adapter->num_rx_desc * sizeof(struct e1000_rx_desc));
3066         E1000_WRITE_REG(&adapter->hw, E1000_RDBAH(0),
3067             (uint32_t)(bus_addr >> 32));
3068         E1000_WRITE_REG(&adapter->hw, E1000_RDBAL(0),
3069             (uint32_t)bus_addr);
3070
3071         /* Setup the Receive Control Register */
3072         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3073         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
3074                 E1000_RCTL_RDMTS_HALF |
3075                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3076
3077         /* Make sure VLAN Filters are off */
3078         rctl &= ~E1000_RCTL_VFE;
3079
3080         if (e1000_tbi_sbp_enabled_82543(&adapter->hw))
3081                 rctl |= E1000_RCTL_SBP;
3082         else
3083                 rctl &= ~E1000_RCTL_SBP;
3084
3085         switch (adapter->rx_buffer_len) {
3086         default:
3087         case 2048:
3088                 rctl |= E1000_RCTL_SZ_2048;
3089                 break;
3090
3091         case 4096:
3092                 rctl |= E1000_RCTL_SZ_4096 |
3093                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3094                 break;
3095
3096         case 8192:
3097                 rctl |= E1000_RCTL_SZ_8192 |
3098                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3099                 break;
3100
3101         case 16384:
3102                 rctl |= E1000_RCTL_SZ_16384 |
3103                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3104                 break;
3105         }
3106
3107         if (ifp->if_mtu > ETHERMTU)
3108                 rctl |= E1000_RCTL_LPE;
3109         else
3110                 rctl &= ~E1000_RCTL_LPE;
3111
3112         /* Receive Checksum Offload for TCP and UDP */
3113         if (ifp->if_capenable & IFCAP_RXCSUM) {
3114                 rxcsum = E1000_READ_REG(&adapter->hw, E1000_RXCSUM);
3115                 rxcsum |= (E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL);
3116                 E1000_WRITE_REG(&adapter->hw, E1000_RXCSUM, rxcsum);
3117         }
3118
3119         /*
3120          * XXX TEMPORARY WORKAROUND: on some systems with 82573
3121          * long latencies are observed, like Lenovo X60. This
3122          * change eliminates the problem, but since having positive
3123          * values in RDTR is a known source of problems on other
3124          * platforms another solution is being sought.
3125          */
3126         if (em_82573_workaround && adapter->hw.mac.type == e1000_82573) {
3127                 E1000_WRITE_REG(&adapter->hw, E1000_RADV, EM_RADV_82573);
3128                 E1000_WRITE_REG(&adapter->hw, E1000_RDTR, EM_RDTR_82573);
3129         }
3130
3131         /*
3132          * Setup the HW Rx Head and Tail Descriptor Pointers
3133          */
3134         E1000_WRITE_REG(&adapter->hw, E1000_RDH(0), 0);
3135         E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), adapter->num_rx_desc - 1);
3136
3137         /* Enable Receives */
3138         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl);
3139 }
3140
3141 static void
3142 em_destroy_rx_ring(struct adapter *adapter, int ndesc)
3143 {
3144         struct em_buffer *rx_buffer;
3145         int i;
3146
3147         if (adapter->rx_buffer_area == NULL)
3148                 return;
3149
3150         for (i = 0; i < ndesc; i++) {
3151                 rx_buffer = &adapter->rx_buffer_area[i];
3152
3153                 KKASSERT(rx_buffer->m_head == NULL);
3154                 bus_dmamap_destroy(adapter->rxtag, rx_buffer->map);
3155         }
3156         bus_dmamap_destroy(adapter->rxtag, adapter->rx_sparemap);
3157         bus_dma_tag_destroy(adapter->rxtag);
3158
3159         kfree(adapter->rx_buffer_area, M_DEVBUF);
3160         adapter->rx_buffer_area = NULL;
3161 }
3162
3163 static void
3164 em_rxeof(struct adapter *adapter, int count)
3165 {
3166         struct ifnet *ifp = &adapter->arpcom.ac_if;
3167         uint8_t status, accept_frame = 0, eop = 0;
3168         uint16_t len, desc_len, prev_len_adj;
3169         struct e1000_rx_desc *current_desc;
3170         struct mbuf *mp;
3171         int i;
3172         struct mbuf_chain chain[MAXCPU];
3173
3174         i = adapter->next_rx_desc_to_check;
3175         current_desc = &adapter->rx_desc_base[i];
3176
3177         if (!(current_desc->status & E1000_RXD_STAT_DD))
3178                 return;
3179
3180         ether_input_chain_init(chain);
3181
3182         while ((current_desc->status & E1000_RXD_STAT_DD) && count != 0) {
3183                 struct mbuf *m = NULL;
3184
3185                 logif(pkt_receive);
3186
3187                 mp = adapter->rx_buffer_area[i].m_head;
3188
3189                 /*
3190                  * Can't defer bus_dmamap_sync(9) because TBI_ACCEPT
3191                  * needs to access the last received byte in the mbuf.
3192                  */
3193                 bus_dmamap_sync(adapter->rxtag, adapter->rx_buffer_area[i].map,
3194                                 BUS_DMASYNC_POSTREAD);
3195
3196                 accept_frame = 1;
3197                 prev_len_adj = 0;
3198                 desc_len = le16toh(current_desc->length);
3199                 status = current_desc->status;
3200                 if (status & E1000_RXD_STAT_EOP) {
3201                         count--;
3202                         eop = 1;
3203                         if (desc_len < ETHER_CRC_LEN) {
3204                                 len = 0;
3205                                 prev_len_adj = ETHER_CRC_LEN - desc_len;
3206                         } else {
3207                                 len = desc_len - ETHER_CRC_LEN;
3208                         }
3209                 } else {
3210                         eop = 0;
3211                         len = desc_len;
3212                 }
3213
3214                 if (current_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
3215                         uint8_t last_byte;
3216                         uint32_t pkt_len = desc_len;
3217
3218                         if (adapter->fmp != NULL)
3219                                 pkt_len += adapter->fmp->m_pkthdr.len;
3220
3221                         last_byte = *(mtod(mp, caddr_t) + desc_len - 1);
3222                         if (TBI_ACCEPT(&adapter->hw, status,
3223                             current_desc->errors, pkt_len, last_byte,
3224                             adapter->min_frame_size, adapter->max_frame_size)) {
3225                                 e1000_tbi_adjust_stats_82543(&adapter->hw,
3226                                     &adapter->stats, pkt_len,
3227                                     adapter->hw.mac.addr,
3228                                     adapter->max_frame_size);
3229                                 if (len > 0)
3230                                         len--;
3231                         } else {
3232                                 accept_frame = 0;
3233                         }
3234                 }
3235
3236                 if (accept_frame) {
3237                         if (em_newbuf(adapter, i, 0) != 0) {
3238                                 ifp->if_iqdrops++;
3239                                 goto discard;
3240                         }
3241
3242                         /* Assign correct length to the current fragment */
3243                         mp->m_len = len;
3244
3245                         if (adapter->fmp == NULL) {
3246                                 mp->m_pkthdr.len = len;
3247                                 adapter->fmp = mp; /* Store the first mbuf */
3248                                 adapter->lmp = mp;
3249                         } else {
3250                                 /*
3251                                  * Chain mbuf's together
3252                                  */
3253
3254                                 /*
3255                                  * Adjust length of previous mbuf in chain if
3256                                  * we received less than 4 bytes in the last
3257                                  * descriptor.
3258                                  */
3259                                 if (prev_len_adj > 0) {
3260                                         adapter->lmp->m_len -= prev_len_adj;
3261                                         adapter->fmp->m_pkthdr.len -=
3262                                             prev_len_adj;
3263                                 }
3264                                 adapter->lmp->m_next = mp;
3265                                 adapter->lmp = adapter->lmp->m_next;
3266                                 adapter->fmp->m_pkthdr.len += len;
3267                         }
3268
3269                         if (eop) {
3270                                 adapter->fmp->m_pkthdr.rcvif = ifp;
3271                                 ifp->if_ipackets++;
3272
3273                                 if (ifp->if_capenable & IFCAP_RXCSUM) {
3274                                         em_rxcsum(adapter, current_desc,
3275                                                   adapter->fmp);
3276                                 }
3277
3278                                 if (status & E1000_RXD_STAT_VP) {
3279                                         adapter->fmp->m_pkthdr.ether_vlantag =
3280                                             (le16toh(current_desc->special) &
3281                                             E1000_RXD_SPC_VLAN_MASK);
3282                                         adapter->fmp->m_flags |= M_VLANTAG;
3283                                 }
3284                                 m = adapter->fmp;
3285                                 adapter->fmp = NULL;
3286                                 adapter->lmp = NULL;
3287                         }
3288                 } else {
3289                         ifp->if_ierrors++;
3290 discard:
3291 #ifdef foo
3292                         /* Reuse loaded DMA map and just update mbuf chain */
3293                         mp = adapter->rx_buffer_area[i].m_head;
3294                         mp->m_len = mp->m_pkthdr.len = MCLBYTES;
3295                         mp->m_data = mp->m_ext.ext_buf;
3296                         mp->m_next = NULL;
3297                         if (adapter->max_frame_size <= (MCLBYTES - ETHER_ALIGN))
3298                                 m_adj(mp, ETHER_ALIGN);
3299 #endif
3300                         if (adapter->fmp != NULL) {
3301                                 m_freem(adapter->fmp);
3302                                 adapter->fmp = NULL;
3303                                 adapter->lmp = NULL;
3304                         }
3305                         m = NULL;
3306                 }
3307
3308                 /* Zero out the receive descriptors status. */
3309                 current_desc->status = 0;
3310
3311                 if (m != NULL)
3312                         ether_input_chain(ifp, m, NULL, chain);
3313
3314                 /* Advance our pointers to the next descriptor. */
3315                 if (++i == adapter->num_rx_desc)
3316                         i = 0;
3317                 current_desc = &adapter->rx_desc_base[i];
3318         }
3319         adapter->next_rx_desc_to_check = i;
3320
3321         ether_input_dispatch(chain);
3322
3323         /* Advance the E1000's Receive Queue #0  "Tail Pointer". */
3324         if (--i < 0)
3325                 i = adapter->num_rx_desc - 1;
3326         E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), i);
3327 }
3328
3329 static void
3330 em_rxcsum(struct adapter *adapter, struct e1000_rx_desc *rx_desc,
3331           struct mbuf *mp)
3332 {
3333         /* 82543 or newer only */
3334         if (adapter->hw.mac.type < e1000_82543 ||
3335             /* Ignore Checksum bit is set */
3336             (rx_desc->status & E1000_RXD_STAT_IXSM))
3337                 return;
3338
3339         if ((rx_desc->status & E1000_RXD_STAT_IPCS) &&
3340             !(rx_desc->errors & E1000_RXD_ERR_IPE)) {
3341                 /* IP Checksum Good */
3342                 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
3343         }
3344
3345         if ((rx_desc->status & E1000_RXD_STAT_TCPCS) &&
3346             !(rx_desc->errors & E1000_RXD_ERR_TCPE)) {
3347                 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
3348                                            CSUM_PSEUDO_HDR |
3349                                            CSUM_FRAG_NOT_CHECKED;
3350                 mp->m_pkthdr.csum_data = htons(0xffff);
3351         }
3352 }
3353
3354 static void
3355 em_enable_intr(struct adapter *adapter)
3356 {
3357         lwkt_serialize_handler_enable(adapter->arpcom.ac_if.if_serializer);
3358         E1000_WRITE_REG(&adapter->hw, E1000_IMS, IMS_ENABLE_MASK);
3359 }
3360
3361 static void
3362 em_disable_intr(struct adapter *adapter)
3363 {
3364         uint32_t clear = 0xffffffff;
3365
3366         /*
3367          * The first version of 82542 had an errata where when link was forced
3368          * it would stay up even up even if the cable was disconnected.
3369          * Sequence errors were used to detect the disconnect and then the
3370          * driver would unforce the link.  This code in the in the ISR.  For
3371          * this to work correctly the Sequence error interrupt had to be
3372          * enabled all the time.
3373          */
3374         if (adapter->hw.mac.type == e1000_82542 &&
3375             adapter->hw.revision_id == E1000_REVISION_2)
3376                 clear &= ~E1000_IMC_RXSEQ;
3377
3378         E1000_WRITE_REG(&adapter->hw, E1000_IMC, clear);
3379
3380         lwkt_serialize_handler_disable(adapter->arpcom.ac_if.if_serializer);
3381 }
3382
3383 /*
3384  * Bit of a misnomer, what this really means is
3385  * to enable OS management of the system... aka
3386  * to disable special hardware management features
3387  */
3388 static void
3389 em_get_mgmt(struct adapter *adapter)
3390 {
3391         /* A shared code workaround */
3392 #define E1000_82542_MANC2H E1000_MANC2H
3393         if (adapter->has_manage) {
3394                 int manc2h = E1000_READ_REG(&adapter->hw, E1000_MANC2H);
3395                 int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
3396
3397                 /* disable hardware interception of ARP */
3398                 manc &= ~(E1000_MANC_ARP_EN);
3399
3400                 /* enable receiving management packets to the host */
3401                 if (adapter->hw.mac.type >= e1000_82571) {
3402                         manc |= E1000_MANC_EN_MNG2HOST;
3403 #define E1000_MNG2HOST_PORT_623 (1 << 5)
3404 #define E1000_MNG2HOST_PORT_664 (1 << 6)
3405                         manc2h |= E1000_MNG2HOST_PORT_623;
3406                         manc2h |= E1000_MNG2HOST_PORT_664;
3407                         E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h);
3408                 }
3409
3410                 E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
3411         }
3412 }
3413
3414 /*
3415  * Give control back to hardware management
3416  * controller if there is one.
3417  */
3418 static void
3419 em_rel_mgmt(struct adapter *adapter)
3420 {
3421         if (adapter->has_manage) {
3422                 int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
3423
3424                 /* re-enable hardware interception of ARP */
3425                 manc |= E1000_MANC_ARP_EN;
3426
3427                 if (adapter->hw.mac.type >= e1000_82571)
3428                         manc &= ~E1000_MANC_EN_MNG2HOST;
3429
3430                 E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
3431         }
3432 }
3433
3434 /*
3435  * em_get_hw_control() sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3436  * For ASF and Pass Through versions of f/w this means that
3437  * the driver is loaded.  For AMT version (only with 82573)
3438  * of the f/w this means that the network i/f is open.
3439  */
3440 static void
3441 em_get_hw_control(struct adapter *adapter)
3442 {
3443         uint32_t ctrl_ext, swsm;
3444
3445         /* Let firmware know the driver has taken over */
3446         switch (adapter->hw.mac.type) {
3447         case e1000_82573:
3448                 swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
3449                 E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
3450                     swsm | E1000_SWSM_DRV_LOAD);
3451                 break;
3452         case e1000_82571:
3453         case e1000_82572:
3454         case e1000_80003es2lan:
3455         case e1000_ich8lan:
3456         case e1000_ich9lan:
3457         case e1000_ich10lan:
3458                 ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
3459                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
3460                     ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
3461                 break;
3462         default:
3463                 break;
3464         }
3465 }
3466
3467 /*
3468  * em_rel_hw_control() resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3469  * For ASF and Pass Through versions of f/w this means that the
3470  * driver is no longer loaded.  For AMT version (only with 82573)
3471  * of the f/w this means that the network i/f is closed.
3472  */
3473 static void
3474 em_rel_hw_control(struct adapter *adapter)
3475 {
3476         uint32_t ctrl_ext, swsm;
3477
3478         /* Let firmware taken over control of h/w */
3479         switch (adapter->hw.mac.type) {
3480         case e1000_82573:
3481                 swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
3482                 E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
3483                     swsm & ~E1000_SWSM_DRV_LOAD);
3484                 break;
3485
3486         case e1000_82571:
3487         case e1000_82572:
3488         case e1000_80003es2lan:
3489         case e1000_ich8lan:
3490         case e1000_ich9lan:
3491         case e1000_ich10lan:
3492                 ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
3493                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
3494                     ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
3495                 break;
3496
3497         default:
3498                 break;
3499         }
3500 }
3501
3502 static int
3503 em_is_valid_eaddr(const uint8_t *addr)
3504 {
3505         char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
3506
3507         if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3508                 return (FALSE);
3509
3510         return (TRUE);
3511 }
3512
3513 /*
3514  * Enable PCI Wake On Lan capability
3515  */
3516 void
3517 em_enable_wol(device_t dev)
3518 {
3519         uint16_t cap, status;
3520         uint8_t id;
3521
3522         /* First find the capabilities pointer*/
3523         cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
3524
3525         /* Read the PM Capabilities */
3526         id = pci_read_config(dev, cap, 1);
3527         if (id != PCIY_PMG)     /* Something wrong */
3528                 return;
3529
3530         /*
3531          * OK, we have the power capabilities,
3532          * so now get the status register
3533          */
3534         cap += PCIR_POWER_STATUS;
3535         status = pci_read_config(dev, cap, 2);
3536         status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3537         pci_write_config(dev, cap, status, 2);
3538 }
3539
3540
3541 /*
3542  * 82544 Coexistence issue workaround.
3543  *    There are 2 issues.
3544  *       1. Transmit Hang issue.
3545  *    To detect this issue, following equation can be used...
3546  *        SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3547  *        If SUM[3:0] is in between 1 to 4, we will have this issue.
3548  *
3549  *       2. DAC issue.
3550  *    To detect this issue, following equation can be used...
3551  *        SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3552  *        If SUM[3:0] is in between 9 to c, we will have this issue.
3553  *
3554  *    WORKAROUND:
3555  *        Make sure we do not have ending address
3556  *        as 1,2,3,4(Hang) or 9,a,b,c (DAC)
3557  */
3558 static uint32_t
3559 em_82544_fill_desc(bus_addr_t address, uint32_t length, PDESC_ARRAY desc_array)
3560 {
3561         uint32_t safe_terminator;
3562
3563         /*
3564          * Since issue is sensitive to length and address.
3565          * Let us first check the address...
3566          */
3567         if (length <= 4) {
3568                 desc_array->descriptor[0].address = address;
3569                 desc_array->descriptor[0].length = length;
3570                 desc_array->elements = 1;
3571                 return (desc_array->elements);
3572         }
3573
3574         safe_terminator =
3575         (uint32_t)((((uint32_t)address & 0x7) + (length & 0xF)) & 0xF);
3576
3577         /* If it does not fall between 0x1 to 0x4 and 0x9 to 0xC then return */
3578         if (safe_terminator == 0 ||
3579             (safe_terminator > 4 && safe_terminator < 9) ||
3580             (safe_terminator > 0xC && safe_terminator <= 0xF)) {
3581                 desc_array->descriptor[0].address = address;
3582                 desc_array->descriptor[0].length = length;
3583                 desc_array->elements = 1;
3584                 return (desc_array->elements);
3585         }
3586
3587         desc_array->descriptor[0].address = address;
3588         desc_array->descriptor[0].length = length - 4;
3589         desc_array->descriptor[1].address = address + (length - 4);
3590         desc_array->descriptor[1].length = 4;
3591         desc_array->elements = 2;
3592         return (desc_array->elements);
3593 }
3594
3595 static void
3596 em_update_stats(struct adapter *adapter)
3597 {
3598         struct ifnet *ifp = &adapter->arpcom.ac_if;
3599
3600         if (adapter->hw.phy.media_type == e1000_media_type_copper ||
3601             (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) {
3602                 adapter->stats.symerrs +=
3603                         E1000_READ_REG(&adapter->hw, E1000_SYMERRS);
3604                 adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC);
3605         }
3606         adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS);
3607         adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC);
3608         adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC);
3609         adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL);
3610
3611         adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC);
3612         adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL);
3613         adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC);
3614         adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC);
3615         adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC);
3616         adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC);
3617         adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC);
3618         adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, E1000_XOFFRXC);
3619         adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC);
3620         adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC);
3621         adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64);
3622         adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127);
3623         adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255);
3624         adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511);
3625         adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023);
3626         adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522);
3627         adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC);
3628         adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC);
3629         adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC);
3630         adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC);
3631
3632         /* For the 64-bit byte counters the low dword must be read first. */
3633         /* Both registers clear on the read of the high dword */
3634
3635         adapter->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCH);
3636         adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCH);
3637
3638         adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC);
3639         adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC);
3640         adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC);
3641         adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC);
3642         adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC);
3643
3644         adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH);
3645         adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH);
3646
3647         adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR);
3648         adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT);
3649         adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64);
3650         adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127);
3651         adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255);
3652         adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511);
3653         adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023);
3654         adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522);
3655         adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC);
3656         adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC);
3657
3658         if (adapter->hw.mac.type >= e1000_82543) {
3659                 adapter->stats.algnerrc +=
3660                 E1000_READ_REG(&adapter->hw, E1000_ALGNERRC);
3661                 adapter->stats.rxerrc +=
3662                 E1000_READ_REG(&adapter->hw, E1000_RXERRC);
3663                 adapter->stats.tncrs +=
3664                 E1000_READ_REG(&adapter->hw, E1000_TNCRS);
3665                 adapter->stats.cexterr +=
3666                 E1000_READ_REG(&adapter->hw, E1000_CEXTERR);
3667                 adapter->stats.tsctc +=
3668                 E1000_READ_REG(&adapter->hw, E1000_TSCTC);
3669                 adapter->stats.tsctfc +=
3670                 E1000_READ_REG(&adapter->hw, E1000_TSCTFC);
3671         }
3672
3673         ifp->if_collisions = adapter->stats.colc;
3674
3675         /* Rx Errors */
3676         ifp->if_ierrors =
3677             adapter->dropped_pkts + adapter->stats.rxerrc +
3678             adapter->stats.crcerrs + adapter->stats.algnerrc +
3679             adapter->stats.ruc + adapter->stats.roc +
3680             adapter->stats.mpc + adapter->stats.cexterr;
3681
3682         /* Tx Errors */
3683         ifp->if_oerrors =
3684             adapter->stats.ecol + adapter->stats.latecol +
3685             adapter->watchdog_events;
3686 }
3687
3688 static void
3689 em_print_debug_info(struct adapter *adapter)
3690 {
3691         device_t dev = adapter->dev;
3692         uint8_t *hw_addr = adapter->hw.hw_addr;
3693
3694         device_printf(dev, "Adapter hardware address = %p \n", hw_addr);
3695         device_printf(dev, "CTRL = 0x%x RCTL = 0x%x \n",
3696             E1000_READ_REG(&adapter->hw, E1000_CTRL),
3697             E1000_READ_REG(&adapter->hw, E1000_RCTL));
3698         device_printf(dev, "Packet buffer = Tx=%dk Rx=%dk \n",
3699             ((E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff0000) >> 16),\
3700             (E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff) );
3701         device_printf(dev, "Flow control watermarks high = %d low = %d\n",
3702             adapter->hw.fc.high_water,
3703             adapter->hw.fc.low_water);
3704         device_printf(dev, "tx_int_delay = %d, tx_abs_int_delay = %d\n",
3705             E1000_READ_REG(&adapter->hw, E1000_TIDV),
3706             E1000_READ_REG(&adapter->hw, E1000_TADV));
3707         device_printf(dev, "rx_int_delay = %d, rx_abs_int_delay = %d\n",
3708             E1000_READ_REG(&adapter->hw, E1000_RDTR),
3709             E1000_READ_REG(&adapter->hw, E1000_RADV));
3710         device_printf(dev, "fifo workaround = %lld, fifo_reset_count = %lld\n",
3711             (long long)adapter->tx_fifo_wrk_cnt,
3712             (long long)adapter->tx_fifo_reset_cnt);
3713         device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
3714             E1000_READ_REG(&adapter->hw, E1000_TDH(0)),
3715             E1000_READ_REG(&adapter->hw, E1000_TDT(0)));
3716         device_printf(dev, "hw rdh = %d, hw rdt = %d\n",
3717             E1000_READ_REG(&adapter->hw, E1000_RDH(0)),
3718             E1000_READ_REG(&adapter->hw, E1000_RDT(0)));
3719         device_printf(dev, "Num Tx descriptors avail = %d\n",
3720             adapter->num_tx_desc_avail);
3721         device_printf(dev, "Tx Descriptors not avail1 = %ld\n",
3722             adapter->no_tx_desc_avail1);
3723         device_printf(dev, "Tx Descriptors not avail2 = %ld\n",
3724             adapter->no_tx_desc_avail2);
3725         device_printf(dev, "Std mbuf failed = %ld\n",
3726             adapter->mbuf_alloc_failed);
3727         device_printf(dev, "Std mbuf cluster failed = %ld\n",
3728             adapter->mbuf_cluster_failed);
3729         device_printf(dev, "Driver dropped packets = %ld\n",
3730             adapter->dropped_pkts);
3731         device_printf(dev, "Driver tx dma failure in encap = %ld\n",
3732             adapter->no_tx_dma_setup);
3733
3734         device_printf(dev, "TXCSUM try pullup = %lu\n",
3735             adapter->tx_csum_try_pullup);
3736         device_printf(dev, "TXCSUM m_pullup(eh) called = %lu\n",
3737             adapter->tx_csum_pullup1);
3738         device_printf(dev, "TXCSUM m_pullup(eh) failed = %lu\n",
3739             adapter->tx_csum_pullup1_failed);
3740         device_printf(dev, "TXCSUM m_pullup(eh+ip) called = %lu\n",
3741             adapter->tx_csum_pullup2);
3742         device_printf(dev, "TXCSUM m_pullup(eh+ip) failed = %lu\n",
3743             adapter->tx_csum_pullup2_failed);
3744         device_printf(dev, "TXCSUM non-writable(eh) droped = %lu\n",
3745             adapter->tx_csum_drop1);
3746         device_printf(dev, "TXCSUM non-writable(eh+ip) droped = %lu\n",
3747             adapter->tx_csum_drop2);
3748 }
3749
3750 static void
3751 em_print_hw_stats(struct adapter *adapter)
3752 {
3753         device_t dev = adapter->dev;
3754
3755         device_printf(dev, "Excessive collisions = %lld\n",
3756             (long long)adapter->stats.ecol);
3757 #if (DEBUG_HW > 0)  /* Dont output these errors normally */
3758         device_printf(dev, "Symbol errors = %lld\n",
3759             (long long)adapter->stats.symerrs);
3760 #endif
3761         device_printf(dev, "Sequence errors = %lld\n",
3762             (long long)adapter->stats.sec);
3763         device_printf(dev, "Defer count = %lld\n",
3764             (long long)adapter->stats.dc);
3765         device_printf(dev, "Missed Packets = %lld\n",
3766             (long long)adapter->stats.mpc);
3767         device_printf(dev, "Receive No Buffers = %lld\n",
3768             (long long)adapter->stats.rnbc);
3769         /* RLEC is inaccurate on some hardware, calculate our own. */
3770         device_printf(dev, "Receive Length Errors = %lld\n",
3771             ((long long)adapter->stats.roc + (long long)adapter->stats.ruc));
3772         device_printf(dev, "Receive errors = %lld\n",
3773             (long long)adapter->stats.rxerrc);
3774         device_printf(dev, "Crc errors = %lld\n",
3775             (long long)adapter->stats.crcerrs);
3776         device_printf(dev, "Alignment errors = %lld\n",
3777             (long long)adapter->stats.algnerrc);
3778         device_printf(dev, "Collision/Carrier extension errors = %lld\n",
3779             (long long)adapter->stats.cexterr);
3780         device_printf(dev, "RX overruns = %ld\n", adapter->rx_overruns);
3781         device_printf(dev, "watchdog timeouts = %ld\n",
3782             adapter->watchdog_events);
3783         device_printf(dev, "XON Rcvd = %lld\n",
3784             (long long)adapter->stats.xonrxc);
3785         device_printf(dev, "XON Xmtd = %lld\n",
3786             (long long)adapter->stats.xontxc);
3787         device_printf(dev, "XOFF Rcvd = %lld\n",
3788             (long long)adapter->stats.xoffrxc);
3789         device_printf(dev, "XOFF Xmtd = %lld\n",
3790             (long long)adapter->stats.xofftxc);
3791         device_printf(dev, "Good Packets Rcvd = %lld\n",
3792             (long long)adapter->stats.gprc);
3793         device_printf(dev, "Good Packets Xmtd = %lld\n",
3794             (long long)adapter->stats.gptc);
3795 }
3796
3797 static void
3798 em_print_nvm_info(struct adapter *adapter)
3799 {
3800         uint16_t eeprom_data;
3801         int i, j, row = 0;
3802
3803         /* Its a bit crude, but it gets the job done */
3804         kprintf("\nInterface EEPROM Dump:\n");
3805         kprintf("Offset\n0x0000  ");
3806         for (i = 0, j = 0; i < 32; i++, j++) {
3807                 if (j == 8) { /* Make the offset block */
3808                         j = 0; ++row;
3809                         kprintf("\n0x00%x0  ",row);
3810                 }
3811                 e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data);
3812                 kprintf("%04x ", eeprom_data);
3813         }
3814         kprintf("\n");
3815 }
3816
3817 static int
3818 em_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
3819 {
3820         struct adapter *adapter;
3821         struct ifnet *ifp;
3822         int error, result;
3823
3824         result = -1;
3825         error = sysctl_handle_int(oidp, &result, 0, req);
3826         if (error || !req->newptr)
3827                 return (error);
3828
3829         adapter = (struct adapter *)arg1;
3830         ifp = &adapter->arpcom.ac_if;
3831
3832         lwkt_serialize_enter(ifp->if_serializer);
3833
3834         if (result == 1)
3835                 em_print_debug_info(adapter);
3836
3837         /*
3838          * This value will cause a hex dump of the
3839          * first 32 16-bit words of the EEPROM to
3840          * the screen.
3841          */
3842         if (result == 2)
3843                 em_print_nvm_info(adapter);
3844
3845         lwkt_serialize_exit(ifp->if_serializer);
3846
3847         return (error);
3848 }
3849
3850 static int
3851 em_sysctl_stats(SYSCTL_HANDLER_ARGS)
3852 {
3853         int error, result;
3854
3855         result = -1;
3856         error = sysctl_handle_int(oidp, &result, 0, req);
3857         if (error || !req->newptr)
3858                 return (error);
3859
3860         if (result == 1) {
3861                 struct adapter *adapter = (struct adapter *)arg1;
3862                 struct ifnet *ifp = &adapter->arpcom.ac_if;
3863
3864                 lwkt_serialize_enter(ifp->if_serializer);
3865                 em_print_hw_stats(adapter);
3866                 lwkt_serialize_exit(ifp->if_serializer);
3867         }
3868         return (error);
3869 }
3870
3871 static void
3872 em_add_sysctl(struct adapter *adapter)
3873 {
3874         sysctl_ctx_init(&adapter->sysctl_ctx);
3875         adapter->sysctl_tree = SYSCTL_ADD_NODE(&adapter->sysctl_ctx,
3876                                         SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
3877                                         device_get_nameunit(adapter->dev),
3878                                         CTLFLAG_RD, 0, "");
3879         if (adapter->sysctl_tree == NULL) {
3880                 device_printf(adapter->dev, "can't add sysctl node\n");
3881         } else {
3882                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3883                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3884                     OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3885                     em_sysctl_debug_info, "I", "Debug Information");
3886
3887                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3888                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3889                     OID_AUTO, "stats", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3890                     em_sysctl_stats, "I", "Statistics");
3891
3892                 SYSCTL_ADD_INT(&adapter->sysctl_ctx,
3893                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3894                     OID_AUTO, "rxd", CTLFLAG_RD,
3895                     &adapter->num_rx_desc, 0, NULL);
3896                 SYSCTL_ADD_INT(&adapter->sysctl_ctx,
3897                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3898                     OID_AUTO, "txd", CTLFLAG_RD,
3899                     &adapter->num_tx_desc, 0, NULL);
3900
3901                 if (adapter->hw.mac.type >= e1000_82540) {
3902                         SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3903                             SYSCTL_CHILDREN(adapter->sysctl_tree),
3904                             OID_AUTO, "int_throttle_ceil",
3905                             CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3906                             em_sysctl_int_throttle, "I",
3907                             "interrupt throttling rate");
3908                 }
3909                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3910                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3911                     OID_AUTO, "int_tx_nsegs",
3912                     CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3913                     em_sysctl_int_tx_nsegs, "I",
3914                     "# segments per TX interrupt");
3915         }
3916 }
3917
3918 static int
3919 em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS)
3920 {
3921         struct adapter *adapter = (void *)arg1;
3922         struct ifnet *ifp = &adapter->arpcom.ac_if;
3923         int error, throttle;
3924
3925         throttle = adapter->int_throttle_ceil;
3926         error = sysctl_handle_int(oidp, &throttle, 0, req);
3927         if (error || req->newptr == NULL)
3928                 return error;
3929         if (throttle < 0 || throttle > 1000000000 / 256)
3930                 return EINVAL;
3931
3932         if (throttle) {
3933                 /*
3934                  * Set the interrupt throttling rate in 256ns increments,
3935                  * recalculate sysctl value assignment to get exact frequency.
3936                  */
3937                 throttle = 1000000000 / 256 / throttle;
3938
3939                 /* Upper 16bits of ITR is reserved and should be zero */
3940                 if (throttle & 0xffff0000)
3941                         return EINVAL;
3942         }
3943
3944         lwkt_serialize_enter(ifp->if_serializer);
3945
3946         if (throttle)
3947                 adapter->int_throttle_ceil = 1000000000 / 256 / throttle;
3948         else
3949                 adapter->int_throttle_ceil = 0;
3950
3951         if (ifp->if_flags & IFF_RUNNING)
3952                 E1000_WRITE_REG(&adapter->hw, E1000_ITR, throttle);
3953
3954         lwkt_serialize_exit(ifp->if_serializer);
3955
3956         if (bootverbose) {
3957                 if_printf(ifp, "Interrupt moderation set to %d/sec\n",
3958                           adapter->int_throttle_ceil);
3959         }
3960         return 0;
3961 }
3962
3963 static int
3964 em_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS)
3965 {
3966         struct adapter *adapter = (void *)arg1;
3967         struct ifnet *ifp = &adapter->arpcom.ac_if;
3968         int error, segs;
3969
3970         segs = adapter->tx_int_nsegs;
3971         error = sysctl_handle_int(oidp, &segs, 0, req);
3972         if (error || req->newptr == NULL)
3973                 return error;
3974         if (segs <= 0)
3975                 return EINVAL;
3976
3977         lwkt_serialize_enter(ifp->if_serializer);
3978
3979         /*
3980          * Don't allow int_tx_nsegs to become:
3981          * o  Less the oact_tx_desc
3982          * o  Too large that no TX desc will cause TX interrupt to
3983          *    be generated (OACTIVE will never recover)
3984          * o  Too small that will cause tx_dd[] overflow
3985          */
3986         if (segs < adapter->oact_tx_desc ||
3987             segs >= adapter->num_tx_desc - adapter->oact_tx_desc ||
3988             segs < adapter->num_tx_desc / EM_TXDD_SAFE) {
3989                 error = EINVAL;
3990         } else {
3991                 error = 0;
3992                 adapter->tx_int_nsegs = segs;
3993         }
3994
3995         lwkt_serialize_exit(ifp->if_serializer);
3996
3997         return error;
3998 }