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  * $DragonFly: src/sys/dev/netif/em/if_em.c,v 1.80 2008/09/17 08:51:29 sephe Exp $
  67  */
  68 /*
  69  * SERIALIZATION API RULES:
  70  *
  71  * - If the driver uses the same serializer for the interrupt as for the
  72  *   ifnet, most of the serialization will be done automatically for the
  73  *   driver.
  74  *
  75  * - ifmedia entry points will be serialized by the ifmedia code using the
  76  *   ifnet serializer.
  77  *
  78  * - if_* entry points except for if_input will be serialized by the IF
  79  *   and protocol layers.
  80  *
  81  * - The device driver must be sure to serialize access from timeout code
  82  *   installed by the device driver.
  83  *
  84  * - The device driver typically holds the serializer at the time it wishes
  85  *   to call if_input.
  86  *
  87  * - We must call lwkt_serialize_handler_enable() prior to enabling the
  88  *   hardware interrupt and lwkt_serialize_handler_disable() after disabling
  89  *   the hardware interrupt in order to avoid handler execution races from
  90  *   scheduled interrupt threads.
  91  *
  92  *   NOTE!  Since callers into the device driver hold the ifnet serializer,
  93  *   the device driver may be holding a serializer at the time it calls
  94  *   if_input even if it is not serializer-aware.
  95  */
  96
  97 #include "opt_polling.h"
  98
  99 #include <sys/param.h>
 100 #include <sys/bus.h>
 101 #include <sys/endian.h>
 102 #include <sys/interrupt.h>
 103 #include <sys/kernel.h>
 104 #include <sys/ktr.h>
 105 #include <sys/malloc.h>
 106 #include <sys/mbuf.h>
 107 #include <sys/proc.h>
 108 #include <sys/rman.h>
 109 #include <sys/serialize.h>
 110 #include <sys/socket.h>
 111 #include <sys/sockio.h>
 112 #include <sys/sysctl.h>
 113 #include <sys/systm.h>
 114
 115 #include <net/bpf.h>
 116 #include <net/ethernet.h>
 117 #include <net/if.h>
 118 #include <net/if_arp.h>
 119 #include <net/if_dl.h>
 120 #include <net/if_media.h>
 121 #include <net/ifq_var.h>
 122 #include <net/vlan/if_vlan_var.h>
 123 #include <net/vlan/if_vlan_ether.h>
 124
 125 #include <netinet/in_systm.h>
 126 #include <netinet/in.h>
 127 #include <netinet/ip.h>
 128 #include <netinet/tcp.h>
 129 #include <netinet/udp.h>
 130
 131 #include <bus/pci/pcivar.h>
 132 #include <bus/pci/pcireg.h>
 133
 134 #include <dev/netif/ig_hal/e1000_api.h>
 135 #include <dev/netif/ig_hal/e1000_82571.h>
 136 #include <dev/netif/em/if_em.h>
 137
 138 #define EM_NAME "Intel(R) PRO/1000 Network Connection "
 139 #define EM_VER  " 6.9.6"
 140
 141 #define _EM_DEVICE(id, ret)     \
 142         { EM_VENDOR_ID, E1000_DEV_ID_##id, ret, EM_NAME #id EM_VER }
 143 #define EM_EMX_DEVICE(id)       _EM_DEVICE(id, -100)
 144 #define EM_DEVICE(id)           _EM_DEVICE(id, 0)
 145 #define EM_DEVICE_NULL  { 0, 0, 0, NULL }
 146
 147 static const struct em_vendor_info em_vendor_info_array[] = {
 148         EM_DEVICE(82540EM),
 149         EM_DEVICE(82540EM_LOM),
 150         EM_DEVICE(82540EP),
 151         EM_DEVICE(82540EP_LOM),
 152         EM_DEVICE(82540EP_LP),
 153
 154         EM_DEVICE(82541EI),
 155         EM_DEVICE(82541ER),
 156         EM_DEVICE(82541ER_LOM),
 157         EM_DEVICE(82541EI_MOBILE),
 158         EM_DEVICE(82541GI),
 159         EM_DEVICE(82541GI_LF),
 160         EM_DEVICE(82541GI_MOBILE),
 161
 162         EM_DEVICE(82542),
 163
 164         EM_DEVICE(82543GC_FIBER),
 165         EM_DEVICE(82543GC_COPPER),
 166
 167         EM_DEVICE(82544EI_COPPER),
 168         EM_DEVICE(82544EI_FIBER),
 169         EM_DEVICE(82544GC_COPPER),
 170         EM_DEVICE(82544GC_LOM),
 171
 172         EM_DEVICE(82545EM_COPPER),
 173         EM_DEVICE(82545EM_FIBER),
 174         EM_DEVICE(82545GM_COPPER),
 175         EM_DEVICE(82545GM_FIBER),
 176         EM_DEVICE(82545GM_SERDES),
 177
 178         EM_DEVICE(82546EB_COPPER),
 179         EM_DEVICE(82546EB_FIBER),
 180         EM_DEVICE(82546EB_QUAD_COPPER),
 181         EM_DEVICE(82546GB_COPPER),
 182         EM_DEVICE(82546GB_FIBER),
 183         EM_DEVICE(82546GB_SERDES),
 184         EM_DEVICE(82546GB_PCIE),
 185         EM_DEVICE(82546GB_QUAD_COPPER),
 186         EM_DEVICE(82546GB_QUAD_COPPER_KSP3),
 187
 188         EM_DEVICE(82547EI),
 189         EM_DEVICE(82547EI_MOBILE),
 190         EM_DEVICE(82547GI),
 191
 192         EM_EMX_DEVICE(82571EB_COPPER),
 193         EM_EMX_DEVICE(82571EB_FIBER),
 194         EM_EMX_DEVICE(82571EB_SERDES),
 195         EM_EMX_DEVICE(82571EB_SERDES_DUAL),
 196         EM_EMX_DEVICE(82571EB_SERDES_QUAD),
 197         EM_EMX_DEVICE(82571EB_QUAD_COPPER),
 198         EM_EMX_DEVICE(82571EB_QUAD_COPPER_BP),
 199         EM_EMX_DEVICE(82571EB_QUAD_COPPER_LP),
 200         EM_EMX_DEVICE(82571EB_QUAD_FIBER),
 201         EM_EMX_DEVICE(82571PT_QUAD_COPPER),
 202
 203         EM_EMX_DEVICE(82572EI_COPPER),
 204         EM_EMX_DEVICE(82572EI_FIBER),
 205         EM_EMX_DEVICE(82572EI_SERDES),
 206         EM_EMX_DEVICE(82572EI),
 207
 208         EM_EMX_DEVICE(82573E),
 209         EM_EMX_DEVICE(82573E_IAMT),
 210         EM_EMX_DEVICE(82573L),
 211
 212         EM_EMX_DEVICE(80003ES2LAN_COPPER_SPT),
 213         EM_EMX_DEVICE(80003ES2LAN_SERDES_SPT),
 214         EM_EMX_DEVICE(80003ES2LAN_COPPER_DPT),
 215         EM_EMX_DEVICE(80003ES2LAN_SERDES_DPT),
 216
 217         EM_DEVICE(ICH8_IGP_M_AMT),
 218         EM_DEVICE(ICH8_IGP_AMT),
 219         EM_DEVICE(ICH8_IGP_C),
 220         EM_DEVICE(ICH8_IFE),
 221         EM_DEVICE(ICH8_IFE_GT),
 222         EM_DEVICE(ICH8_IFE_G),
 223         EM_DEVICE(ICH8_IGP_M),
 224
 225         EM_DEVICE(ICH9_IGP_M_AMT),
 226         EM_DEVICE(ICH9_IGP_AMT),
 227         EM_DEVICE(ICH9_IGP_C),
 228         EM_DEVICE(ICH9_IGP_M),
 229         EM_DEVICE(ICH9_IGP_M_V),
 230         EM_DEVICE(ICH9_IFE),
 231         EM_DEVICE(ICH9_IFE_GT),
 232         EM_DEVICE(ICH9_IFE_G),
 233         EM_DEVICE(ICH9_BM),
 234
 235         EM_EMX_DEVICE(82574L),
 236
 237         EM_DEVICE(ICH10_R_BM_LM),
 238         EM_DEVICE(ICH10_R_BM_LF),
 239         EM_DEVICE(ICH10_R_BM_V),
 240         EM_DEVICE(ICH10_D_BM_LM),
 241         EM_DEVICE(ICH10_D_BM_LF),
 242
 243         /* required last entry */
 244         EM_DEVICE_NULL
 245 };
 246
 247 static int      em_probe(device_t);
 248 static int      em_attach(device_t);
 249 static int      em_detach(device_t);
 250 static int      em_shutdown(device_t);
 251 static int      em_suspend(device_t);
 252 static int      em_resume(device_t);
 253
 254 static void     em_init(void *);
 255 static void     em_stop(struct adapter *);
 256 static int      em_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 257 static void     em_start(struct ifnet *);
 258 #ifdef DEVICE_POLLING
 259 static void     em_poll(struct ifnet *, enum poll_cmd, int);
 260 #endif
 261 static void     em_watchdog(struct ifnet *);
 262 static void     em_media_status(struct ifnet *, struct ifmediareq *);
 263 static int      em_media_change(struct ifnet *);
 264 static void     em_timer(void *);
 265
 266 static void     em_intr(void *);
 267 static void     em_rxeof(struct adapter *, int);
 268 static void     em_txeof(struct adapter *);
 269 static void     em_tx_collect(struct adapter *);
 270 static void     em_tx_purge(struct adapter *);
 271 static void     em_enable_intr(struct adapter *);
 272 static void     em_disable_intr(struct adapter *);
 273
 274 static int      em_dma_malloc(struct adapter *, bus_size_t,
 275                     struct em_dma_alloc *);
 276 static void     em_dma_free(struct adapter *, struct em_dma_alloc *);
 277 static void     em_init_tx_ring(struct adapter *);
 278 static int      em_init_rx_ring(struct adapter *);
 279 static int      em_create_tx_ring(struct adapter *);
 280 static int      em_create_rx_ring(struct adapter *);
 281 static void     em_destroy_tx_ring(struct adapter *, int);
 282 static void     em_destroy_rx_ring(struct adapter *, int);
 283 static int      em_newbuf(struct adapter *, int, int);
 284 static int      em_encap(struct adapter *, struct mbuf **);
 285 static void     em_rxcsum(struct adapter *, struct e1000_rx_desc *,
 286                     struct mbuf *);
 287 static int      em_txcsum_pullup(struct adapter *, struct mbuf **);
 288 static int      em_txcsum(struct adapter *, struct mbuf *,
 289                     uint32_t *, uint32_t *);
 290
 291 static int      em_get_hw_info(struct adapter *);
 292 static int      em_is_valid_eaddr(const uint8_t *);
 293 static int      em_alloc_pci_res(struct adapter *);
 294 static void     em_free_pci_res(struct adapter *);
 295 static int      em_hw_init(struct adapter *);
 296 static void     em_setup_ifp(struct adapter *);
 297 static void     em_init_tx_unit(struct adapter *);
 298 static void     em_init_rx_unit(struct adapter *);
 299 static void     em_update_stats(struct adapter *);
 300 static void     em_set_promisc(struct adapter *);
 301 static void     em_disable_promisc(struct adapter *);
 302 static void     em_set_multi(struct adapter *);
 303 static void     em_update_link_status(struct adapter *);
 304 static void     em_smartspeed(struct adapter *);
 305
 306 /* Hardware workarounds */
 307 static int      em_82547_fifo_workaround(struct adapter *, int);
 308 static void     em_82547_update_fifo_head(struct adapter *, int);
 309 static int      em_82547_tx_fifo_reset(struct adapter *);
 310 static void     em_82547_move_tail(void *);
 311 static void     em_82547_move_tail_serialized(struct adapter *);
 312 static uint32_t em_82544_fill_desc(bus_addr_t, uint32_t, PDESC_ARRAY);
 313
 314 static void     em_print_debug_info(struct adapter *);
 315 static void     em_print_nvm_info(struct adapter *);
 316 static void     em_print_hw_stats(struct adapter *);
 317
 318 static int      em_sysctl_stats(SYSCTL_HANDLER_ARGS);
 319 static int      em_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
 320 static int      em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS);
 321 static int      em_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS);
 322 static void     em_add_sysctl(struct adapter *adapter);
 323
 324 /* Management and WOL Support */
 325 static void     em_get_mgmt(struct adapter *);
 326 static void     em_rel_mgmt(struct adapter *);
 327 static void     em_get_hw_control(struct adapter *);
 328 static void     em_rel_hw_control(struct adapter *);
 329 static void     em_enable_wol(device_t);
 330
 331 static device_method_t em_methods[] = {
 332         /* Device interface */
 333         DEVMETHOD(device_probe,         em_probe),
 334         DEVMETHOD(device_attach,        em_attach),
 335         DEVMETHOD(device_detach,        em_detach),
 336         DEVMETHOD(device_shutdown,      em_shutdown),
 337         DEVMETHOD(device_suspend,       em_suspend),
 338         DEVMETHOD(device_resume,        em_resume),
 339         { 0, 0 }
 340 };
 341
 342 static driver_t em_driver = {
 343         "em",
 344         em_methods,
 345         sizeof(struct adapter),
 346 };
 347
 348 static devclass_t em_devclass;
 349
 350 DECLARE_DUMMY_MODULE(if_em);
 351 MODULE_DEPEND(em, ig_hal, 1, 1, 1);
 352 DRIVER_MODULE(if_em, pci, em_driver, em_devclass, 0, 0);
 353
 354 /*
 355  * Tunables
 356  */
 357 static int      em_int_throttle_ceil = EM_DEFAULT_ITR;
 358 static int      em_rxd = EM_DEFAULT_RXD;
 359 static int      em_txd = EM_DEFAULT_TXD;
 360 static int      em_smart_pwr_down = FALSE;
 361
 362 /* Controls whether promiscuous also shows bad packets */
 363 static int      em_debug_sbp = FALSE;
 364
 365 static int      em_82573_workaround = TRUE;
 366
 367 TUNABLE_INT("hw.em.int_throttle_ceil", &em_int_throttle_ceil);
 368 TUNABLE_INT("hw.em.rxd", &em_rxd);
 369 TUNABLE_INT("hw.em.txd", &em_txd);
 370 TUNABLE_INT("hw.em.smart_pwr_down", &em_smart_pwr_down);
 371 TUNABLE_INT("hw.em.sbp", &em_debug_sbp);
 372 TUNABLE_INT("hw.em.82573_workaround", &em_82573_workaround);
 373
 374 /* Global used in WOL setup with multiport cards */
 375 static int      em_global_quad_port_a = 0;
 376
 377 /* Set this to one to display debug statistics */
 378 static int      em_display_debug_stats = 0;
 379
 380 #if !defined(KTR_IF_EM)
 381 #define KTR_IF_EM       KTR_ALL
 382 #endif
 383 KTR_INFO_MASTER(if_em);
 384 KTR_INFO(KTR_IF_EM, if_em, intr_beg, 0, "intr begin", 0);
 385 KTR_INFO(KTR_IF_EM, if_em, intr_end, 1, "intr end", 0);
 386 KTR_INFO(KTR_IF_EM, if_em, pkt_receive, 4, "rx packet", 0);
 387 KTR_INFO(KTR_IF_EM, if_em, pkt_txqueue, 5, "tx packet", 0);
 388 KTR_INFO(KTR_IF_EM, if_em, pkt_txclean, 6, "tx clean", 0);
 389 #define logif(name)     KTR_LOG(if_em_ ## name)
 390
 391 static int
 392 em_probe(device_t dev)
 393 {
 394         const struct em_vendor_info *ent;
 395         uint16_t vid, did;
 396
 397         vid = pci_get_vendor(dev);
 398         did = pci_get_device(dev);
 399
 400         for (ent = em_vendor_info_array; ent->desc != NULL; ++ent) {
 401                 if (vid == ent->vendor_id && did == ent->device_id) {
 402                         device_set_desc(dev, ent->desc);
 403                         device_set_async_attach(dev, TRUE);
 404                         return (ent->ret);
 405                 }
 406         }
 407         return (ENXIO);
 408 }
 409
 410 static int
 411 em_attach(device_t dev)
 412 {
 413         struct adapter *adapter = device_get_softc(dev);
 414         struct ifnet *ifp = &adapter->arpcom.ac_if;
 415         int tsize, rsize;
 416         int error = 0;
 417         uint16_t eeprom_data, device_id;
 418
 419         adapter->dev = adapter->osdep.dev = dev;
 420
 421         callout_init(&adapter->timer);
 422         callout_init(&adapter->tx_fifo_timer);
 423
 424         /* Determine hardware and mac info */
 425         error = em_get_hw_info(adapter);
 426         if (error) {
 427                 device_printf(dev, "Identify hardware failed\n");
 428                 goto fail;
 429         }
 430
 431         /* Setup PCI resources */
 432         error = em_alloc_pci_res(adapter);
 433         if (error) {
 434                 device_printf(dev, "Allocation of PCI resources failed\n");
 435                 goto fail;
 436         }
 437
 438         /*
 439          * For ICH8 and family we need to map the flash memory,
 440          * and this must happen after the MAC is identified.
 441          */
 442         if (adapter->hw.mac.type == e1000_ich8lan ||
 443             adapter->hw.mac.type == e1000_ich10lan ||
 444             adapter->hw.mac.type == e1000_ich9lan) {
 445                 adapter->flash_rid = EM_BAR_FLASH;
 446
 447                 adapter->flash = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
 448                                         &adapter->flash_rid, RF_ACTIVE);
 449                 if (adapter->flash == NULL) {
 450                         device_printf(dev, "Mapping of Flash failed\n");
 451                         error = ENXIO;
 452                         goto fail;
 453                 }
 454                 adapter->osdep.flash_bus_space_tag =
 455                     rman_get_bustag(adapter->flash);
 456                 adapter->osdep.flash_bus_space_handle =
 457                     rman_get_bushandle(adapter->flash);
 458
 459                 /*
 460                  * This is used in the shared code
 461                  * XXX this goof is actually not used.
 462                  */
 463                 adapter->hw.flash_address = (uint8_t *)adapter->flash;
 464         }
 465
 466         /* Do Shared Code initialization */
 467         if (e1000_setup_init_funcs(&adapter->hw, TRUE)) {
 468                 device_printf(dev, "Setup of Shared code failed\n");
 469                 error = ENXIO;
 470                 goto fail;
 471         }
 472
 473         e1000_get_bus_info(&adapter->hw);
 474
 475         /*
 476          * Validate number of transmit and receive descriptors.  It
 477          * must not exceed hardware maximum, and must be multiple
 478          * of E1000_DBA_ALIGN.
 479          */
 480         if ((em_txd * sizeof(struct e1000_tx_desc)) % EM_DBA_ALIGN != 0 ||
 481             (adapter->hw.mac.type >= e1000_82544 && em_txd > EM_MAX_TXD) ||
 482             (adapter->hw.mac.type < e1000_82544 && em_txd > EM_MAX_TXD_82543) ||
 483             em_txd < EM_MIN_TXD) {
 484                 device_printf(dev, "Using %d TX descriptors instead of %d!\n",
 485                     EM_DEFAULT_TXD, em_txd);
 486                 adapter->num_tx_desc = EM_DEFAULT_TXD;
 487         } else {
 488                 adapter->num_tx_desc = em_txd;
 489         }
 490         if ((em_rxd * sizeof(struct e1000_rx_desc)) % EM_DBA_ALIGN != 0 ||
 491             (adapter->hw.mac.type >= e1000_82544 && em_rxd > EM_MAX_RXD) ||
 492             (adapter->hw.mac.type < e1000_82544 && em_rxd > EM_MAX_RXD_82543) ||
 493             em_rxd < EM_MIN_RXD) {
 494                 device_printf(dev, "Using %d RX descriptors instead of %d!\n",
 495                     EM_DEFAULT_RXD, em_rxd);
 496                 adapter->num_rx_desc = EM_DEFAULT_RXD;
 497         } else {
 498                 adapter->num_rx_desc = em_rxd;
 499         }
 500
 501         adapter->hw.mac.autoneg = DO_AUTO_NEG;
 502         adapter->hw.phy.autoneg_wait_to_complete = FALSE;
 503         adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
 504         adapter->rx_buffer_len = MCLBYTES;
 505
 506         /*
 507          * Interrupt throttle rate
 508          */
 509         if (em_int_throttle_ceil == 0) {
 510                 adapter->int_throttle_ceil = 0;
 511         } else {
 512                 int throttle = em_int_throttle_ceil;
 513
 514                 if (throttle < 0)
 515                         throttle = EM_DEFAULT_ITR;
 516
 517                 /* Recalculate the tunable value to get the exact frequency. */
 518                 throttle = 1000000000 / 256 / throttle;
 519
 520                 /* Upper 16bits of ITR is reserved and should be zero */
 521                 if (throttle & 0xffff0000)
 522                         throttle = 1000000000 / 256 / EM_DEFAULT_ITR;
 523
 524                 adapter->int_throttle_ceil = 1000000000 / 256 / throttle;
 525         }
 526
 527         e1000_init_script_state_82541(&adapter->hw, TRUE);
 528         e1000_set_tbi_compatibility_82543(&adapter->hw, TRUE);
 529
 530         /* Copper options */
 531         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
 532                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
 533                 adapter->hw.phy.disable_polarity_correction = FALSE;
 534                 adapter->hw.phy.ms_type = EM_MASTER_SLAVE;
 535         }
 536
 537         /* Set the frame limits assuming standard ethernet sized frames. */
 538         adapter->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
 539         adapter->min_frame_size = ETH_ZLEN + ETHER_CRC_LEN;
 540
 541         /* This controls when hardware reports transmit completion status. */
 542         adapter->hw.mac.report_tx_early = 1;
 543
 544         /*
 545          * Create top level busdma tag
 546          */
 547         error = bus_dma_tag_create(NULL, 1, 0,
 548                         BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
 549                         NULL, NULL,
 550                         BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
 551                         0, &adapter->parent_dtag);
 552         if (error) {
 553                 device_printf(dev, "could not create top level DMA tag\n");
 554                 goto fail;
 555         }
 556
 557         /*
 558          * Allocate Transmit Descriptor ring
 559          */
 560         tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc),
 561                          EM_DBA_ALIGN);
 562         error = em_dma_malloc(adapter, tsize, &adapter->txdma);
 563         if (error) {
 564                 device_printf(dev, "Unable to allocate tx_desc memory\n");
 565                 goto fail;
 566         }
 567         adapter->tx_desc_base = adapter->txdma.dma_vaddr;
 568
 569         /*
 570          * Allocate Receive Descriptor ring
 571          */
 572         rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc),
 573                          EM_DBA_ALIGN);
 574         error = em_dma_malloc(adapter, rsize, &adapter->rxdma);
 575         if (error) {
 576                 device_printf(dev, "Unable to allocate rx_desc memory\n");
 577                 goto fail;
 578         }
 579         adapter->rx_desc_base = adapter->rxdma.dma_vaddr;
 580
 581         /* Make sure we have a good EEPROM before we read from it */
 582         if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
 583                 /*
 584                  * Some PCI-E parts fail the first check due to
 585                  * the link being in sleep state, call it again,
 586                  * if it fails a second time its a real issue.
 587                  */
 588                 if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
 589                         device_printf(dev,
 590                             "The EEPROM Checksum Is Not Valid\n");
 591                         error = EIO;
 592                         goto fail;
 593                 }
 594         }
 595
 596         /* Initialize the hardware */
 597         error = em_hw_init(adapter);
 598         if (error) {
 599                 device_printf(dev, "Unable to initialize the hardware\n");
 600                 goto fail;
 601         }
 602
 603         /* Copy the permanent MAC address out of the EEPROM */
 604         if (e1000_read_mac_addr(&adapter->hw) < 0) {
 605                 device_printf(dev, "EEPROM read error while reading MAC"
 606                     " address\n");
 607                 error = EIO;
 608                 goto fail;
 609         }
 610         if (!em_is_valid_eaddr(adapter->hw.mac.addr)) {
 611                 device_printf(dev, "Invalid MAC address\n");
 612                 error = EIO;
 613                 goto fail;
 614         }
 615
 616         /* Allocate transmit descriptors and buffers */
 617         error = em_create_tx_ring(adapter);
 618         if (error) {
 619                 device_printf(dev, "Could not setup transmit structures\n");
 620                 goto fail;
 621         }
 622
 623         /* Allocate receive descriptors and buffers */
 624         error = em_create_rx_ring(adapter);
 625         if (error) {
 626                 device_printf(dev, "Could not setup receive structures\n");
 627                 goto fail;
 628         }
 629
 630         /* Manually turn off all interrupts */
 631         E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
 632
 633         /* Setup OS specific network interface */
 634         em_setup_ifp(adapter);
 635
 636         /* Add sysctl tree, must after em_setup_ifp() */
 637         em_add_sysctl(adapter);
 638
 639         /* Initialize statistics */
 640         em_update_stats(adapter);
 641
 642         adapter->hw.mac.get_link_status = 1;
 643         em_update_link_status(adapter);
 644
 645         /* Indicate SOL/IDER usage */
 646         if (e1000_check_reset_block(&adapter->hw)) {
 647                 device_printf(dev,
 648                     "PHY reset is blocked due to SOL/IDER session.\n");
 649         }
 650
 651         /* Determine if we have to control management hardware */
 652         adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw);
 653
 654         /*
 655          * Setup Wake-on-Lan
 656          */
 657         switch (adapter->hw.mac.type) {
 658         case e1000_82542:
 659         case e1000_82543:
 660                 break;
 661
 662         case e1000_82546:
 663         case e1000_82546_rev_3:
 664         case e1000_82571:
 665         case e1000_80003es2lan:
 666                 if (adapter->hw.bus.func == 1) {
 667                         e1000_read_nvm(&adapter->hw,
 668                             NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 669                 } else {
 670                         e1000_read_nvm(&adapter->hw,
 671                             NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
 672                 }
 673                 eeprom_data &= EM_EEPROM_APME;
 674                 break;
 675
 676         default:
 677                 /* APME bit in EEPROM is mapped to WUC.APME */
 678                 eeprom_data =
 679                     E1000_READ_REG(&adapter->hw, E1000_WUC) & E1000_WUC_APME;
 680                 break;
 681         }
 682         if (eeprom_data)
 683                 adapter->wol = E1000_WUFC_MAG;
 684         /*
 685          * We have the eeprom settings, now apply the special cases
 686          * where the eeprom may be wrong or the board won't support
 687          * wake on lan on a particular port
 688          */
 689         device_id = pci_get_device(dev);
 690         switch (device_id) {
 691         case E1000_DEV_ID_82546GB_PCIE:
 692                 adapter->wol = 0;
 693                 break;
 694
 695         case E1000_DEV_ID_82546EB_FIBER:
 696         case E1000_DEV_ID_82546GB_FIBER:
 697         case E1000_DEV_ID_82571EB_FIBER:
 698                 /*
 699                  * Wake events only supported on port A for dual fiber
 700                  * regardless of eeprom setting
 701                  */
 702                 if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
 703                     E1000_STATUS_FUNC_1)
 704                         adapter->wol = 0;
 705                 break;
 706
 707         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
 708         case E1000_DEV_ID_82571EB_QUAD_COPPER:
 709         case E1000_DEV_ID_82571EB_QUAD_FIBER:
 710         case E1000_DEV_ID_82571EB_QUAD_COPPER_LP:
 711                 /* if quad port adapter, disable WoL on all but port A */
 712                 if (em_global_quad_port_a != 0)
 713                         adapter->wol = 0;
 714                 /* Reset for multiple quad port adapters */
 715                 if (++em_global_quad_port_a == 4)
 716                         em_global_quad_port_a = 0;
 717                 break;
 718         }
 719
 720         /* XXX disable wol */
 721         adapter->wol = 0;
 722
 723         /* Do we need workaround for 82544 PCI-X adapter? */
 724         if (adapter->hw.bus.type == e1000_bus_type_pcix &&
 725             adapter->hw.mac.type == e1000_82544)
 726                 adapter->pcix_82544 = TRUE;
 727         else
 728                 adapter->pcix_82544 = FALSE;
 729
 730         if (adapter->pcix_82544) {
 731                 /*
 732                  * 82544 on PCI-X may split one TX segment
 733                  * into two TX descs, so we double its number
 734                  * of spare TX desc here.
 735                  */
 736                 adapter->spare_tx_desc = 2 * EM_TX_SPARE;
 737         } else {
 738                 adapter->spare_tx_desc = EM_TX_SPARE;
 739         }
 740
 741         /*
 742          * Keep following relationship between spare_tx_desc, oact_tx_desc
 743          * and tx_int_nsegs:
 744          * (spare_tx_desc + EM_TX_RESERVED) <=
 745          * oact_tx_desc <= EM_TX_OACTIVE_MAX <= tx_int_nsegs
 746          */
 747         adapter->oact_tx_desc = adapter->num_tx_desc / 8;
 748         if (adapter->oact_tx_desc > EM_TX_OACTIVE_MAX)
 749                 adapter->oact_tx_desc = EM_TX_OACTIVE_MAX;
 750         if (adapter->oact_tx_desc < adapter->spare_tx_desc + EM_TX_RESERVED)
 751                 adapter->oact_tx_desc = adapter->spare_tx_desc + EM_TX_RESERVED;
 752
 753         adapter->tx_int_nsegs = adapter->num_tx_desc / 16;
 754         if (adapter->tx_int_nsegs < adapter->oact_tx_desc)
 755                 adapter->tx_int_nsegs = adapter->oact_tx_desc;
 756
 757         error = bus_setup_intr(dev, adapter->intr_res, INTR_MPSAFE,
 758                                em_intr, adapter, &adapter->intr_tag,
 759                                ifp->if_serializer);
 760         if (error) {
 761                 device_printf(dev, "Failed to register interrupt handler");
 762                 ether_ifdetach(&adapter->arpcom.ac_if);
 763                 goto fail;
 764         }
 765
 766         ifp->if_cpuid = ithread_cpuid(rman_get_start(adapter->intr_res));
 767         KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
 768         return (0);
 769 fail:
 770         em_detach(dev);
 771         return (error);
 772 }
 773
 774 static int
 775 em_detach(device_t dev)
 776 {
 777         struct adapter *adapter = device_get_softc(dev);
 778
 779         if (device_is_attached(dev)) {
 780                 struct ifnet *ifp = &adapter->arpcom.ac_if;
 781
 782                 lwkt_serialize_enter(ifp->if_serializer);
 783
 784                 em_stop(adapter);
 785
 786                 e1000_phy_hw_reset(&adapter->hw);
 787
 788                 em_rel_mgmt(adapter);
 789
 790                 if ((adapter->hw.mac.type == e1000_82573 ||
 791                      adapter->hw.mac.type == e1000_ich8lan ||
 792                      adapter->hw.mac.type == e1000_ich10lan ||
 793                      adapter->hw.mac.type == e1000_ich9lan) &&
 794                     e1000_check_mng_mode(&adapter->hw))
 795                         em_rel_hw_control(adapter);
 796
 797                 if (adapter->wol) {
 798                         E1000_WRITE_REG(&adapter->hw, E1000_WUC,
 799                                         E1000_WUC_PME_EN);
 800                         E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
 801                         em_enable_wol(dev);
 802                 }
 803
 804                 bus_teardown_intr(dev, adapter->intr_res, adapter->intr_tag);
 805
 806                 lwkt_serialize_exit(ifp->if_serializer);
 807
 808                 ether_ifdetach(ifp);
 809         }
 810         bus_generic_detach(dev);
 811
 812         em_free_pci_res(adapter);
 813
 814         em_destroy_tx_ring(adapter, adapter->num_tx_desc);
 815         em_destroy_rx_ring(adapter, adapter->num_rx_desc);
 816
 817         /* Free Transmit Descriptor ring */
 818         if (adapter->tx_desc_base)
 819                 em_dma_free(adapter, &adapter->txdma);
 820
 821         /* Free Receive Descriptor ring */
 822         if (adapter->rx_desc_base)
 823                 em_dma_free(adapter, &adapter->rxdma);
 824
 825         /* Free top level busdma tag */
 826         if (adapter->parent_dtag != NULL)
 827                 bus_dma_tag_destroy(adapter->parent_dtag);
 828
 829         /* Free sysctl tree */
 830         if (adapter->sysctl_tree != NULL)
 831                 sysctl_ctx_free(&adapter->sysctl_ctx);
 832
 833         return (0);
 834 }
 835
 836 static int
 837 em_shutdown(device_t dev)
 838 {
 839         return em_suspend(dev);
 840 }
 841
 842 static int
 843 em_suspend(device_t dev)
 844 {
 845         struct adapter *adapter = device_get_softc(dev);
 846         struct ifnet *ifp = &adapter->arpcom.ac_if;
 847
 848         lwkt_serialize_enter(ifp->if_serializer);
 849
 850         em_stop(adapter);
 851
 852         em_rel_mgmt(adapter);
 853
 854         if ((adapter->hw.mac.type == e1000_82573 ||
 855              adapter->hw.mac.type == e1000_ich8lan ||
 856              adapter->hw.mac.type == e1000_ich10lan ||
 857              adapter->hw.mac.type == e1000_ich9lan) &&
 858             e1000_check_mng_mode(&adapter->hw))
 859                 em_rel_hw_control(adapter);
 860
 861         if (adapter->wol) {
 862                 E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
 863                 E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
 864                 em_enable_wol(dev);
 865         }
 866
 867         lwkt_serialize_exit(ifp->if_serializer);
 868
 869         return bus_generic_suspend(dev);
 870 }
 871
 872 static int
 873 em_resume(device_t dev)
 874 {
 875         struct adapter *adapter = device_get_softc(dev);
 876         struct ifnet *ifp = &adapter->arpcom.ac_if;
 877
 878         lwkt_serialize_enter(ifp->if_serializer);
 879
 880         em_init(adapter);
 881         em_get_mgmt(adapter);
 882         if_devstart(ifp);
 883
 884         lwkt_serialize_exit(ifp->if_serializer);
 885
 886         return bus_generic_resume(dev);
 887 }
 888
 889 static void
 890 em_start(struct ifnet *ifp)
 891 {
 892         struct adapter *adapter = ifp->if_softc;
 893         struct mbuf *m_head;
 894
 895         ASSERT_SERIALIZED(ifp->if_serializer);
 896
 897         if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
 898                 return;
 899
 900         if (!adapter->link_active) {
 901                 ifq_purge(&ifp->if_snd);
 902                 return;
 903         }
 904
 905         while (!ifq_is_empty(&ifp->if_snd)) {
 906                 /* Now do we at least have a minimal? */
 907                 if (EM_IS_OACTIVE(adapter)) {
 908                         em_tx_collect(adapter);
 909                         if (EM_IS_OACTIVE(adapter)) {
 910                                 ifp->if_flags |= IFF_OACTIVE;
 911                                 adapter->no_tx_desc_avail1++;
 912                                 break;
 913                         }
 914                 }
 915
 916                 logif(pkt_txqueue);
 917                 m_head = ifq_dequeue(&ifp->if_snd, NULL);
 918                 if (m_head == NULL)
 919                         break;
 920
 921                 if (em_encap(adapter, &m_head)) {
 922                         ifp->if_oerrors++;
 923                         em_tx_collect(adapter);
 924                         continue;
 925                 }
 926
 927                 /* Send a copy of the frame to the BPF listener */
 928                 ETHER_BPF_MTAP(ifp, m_head);
 929
 930                 /* Set timeout in case hardware has problems transmitting. */
 931                 ifp->if_timer = EM_TX_TIMEOUT;
 932         }
 933 }
 934
 935 static int
 936 em_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
 937 {
 938         struct adapter *adapter = ifp->if_softc;
 939         struct ifreq *ifr = (struct ifreq *)data;
 940         uint16_t eeprom_data = 0;
 941         int max_frame_size, mask, reinit;
 942         int error = 0;
 943
 944         ASSERT_SERIALIZED(ifp->if_serializer);
 945
 946         switch (command) {
 947         case SIOCSIFMTU:
 948                 switch (adapter->hw.mac.type) {
 949                 case e1000_82573:
 950                         /*
 951                          * 82573 only supports jumbo frames
 952                          * if ASPM is disabled.
 953                          */
 954                         e1000_read_nvm(&adapter->hw,
 955                             NVM_INIT_3GIO_3, 1, &eeprom_data);
 956                         if (eeprom_data & NVM_WORD1A_ASPM_MASK) {
 957                                 max_frame_size = ETHER_MAX_LEN;
 958                                 break;
 959                         }
 960                         /* FALL THROUGH */
 961
 962                 /* Limit Jumbo Frame size */
 963                 case e1000_82571:
 964                 case e1000_82572:
 965                 case e1000_ich9lan:
 966                 case e1000_ich10lan:
 967                 case e1000_82574:
 968                 case e1000_80003es2lan:
 969                         max_frame_size = 9234;
 970                         break;
 971
 972                 /* Adapters that do not support jumbo frames */
 973                 case e1000_82542:
 974                 case e1000_ich8lan:
 975                         max_frame_size = ETHER_MAX_LEN;
 976                         break;
 977
 978                 default:
 979                         max_frame_size = MAX_JUMBO_FRAME_SIZE;
 980                         break;
 981                 }
 982                 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
 983                     ETHER_CRC_LEN) {
 984                         error = EINVAL;
 985                         break;
 986                 }
 987
 988                 ifp->if_mtu = ifr->ifr_mtu;
 989                 adapter->max_frame_size =
 990                     ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
 991
 992                 if (ifp->if_flags & IFF_RUNNING)
 993                         em_init(adapter);
 994                 break;
 995
 996         case SIOCSIFFLAGS:
 997                 if (ifp->if_flags & IFF_UP) {
 998                         if ((ifp->if_flags & IFF_RUNNING)) {
 999                                 if ((ifp->if_flags ^ adapter->if_flags) &
1000                                     (IFF_PROMISC | IFF_ALLMULTI)) {
1001                                         em_disable_promisc(adapter);
1002                                         em_set_promisc(adapter);
1003                                 }
1004                         } else {
1005                                 em_init(adapter);
1006                         }
1007                 } else if (ifp->if_flags & IFF_RUNNING) {
1008                         em_stop(adapter);
1009                 }
1010                 adapter->if_flags = ifp->if_flags;
1011                 break;
1012
1013         case SIOCADDMULTI:
1014         case SIOCDELMULTI:
1015                 if (ifp->if_flags & IFF_RUNNING) {
1016                         em_disable_intr(adapter);
1017                         em_set_multi(adapter);
1018                         if (adapter->hw.mac.type == e1000_82542 &&
1019                             adapter->hw.revision_id == E1000_REVISION_2)
1020                                 em_init_rx_unit(adapter);
1021 #ifdef DEVICE_POLLING
1022                         if (!(ifp->if_flags & IFF_POLLING))
1023 #endif
1024                                 em_enable_intr(adapter);
1025                 }
1026                 break;
1027
1028         case SIOCSIFMEDIA:
1029                 /* Check SOL/IDER usage */
1030                 if (e1000_check_reset_block(&adapter->hw)) {
1031                         device_printf(adapter->dev, "Media change is"
1032                             " blocked due to SOL/IDER session.\n");
1033                         break;
1034                 }
1035                 /* FALL THROUGH */
1036
1037         case SIOCGIFMEDIA:
1038                 error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
1039                 break;
1040
1041         case SIOCSIFCAP:
1042                 reinit = 0;
1043                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1044                 if (mask & IFCAP_HWCSUM) {
1045                         ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
1046                         reinit = 1;
1047                 }
1048                 if (mask & IFCAP_VLAN_HWTAGGING) {
1049                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1050                         reinit = 1;
1051                 }
1052                 if (reinit && (ifp->if_flags & IFF_RUNNING))
1053                         em_init(adapter);
1054                 break;
1055
1056         default:
1057                 error = ether_ioctl(ifp, command, data);
1058                 break;
1059         }
1060         return (error);
1061 }
1062
1063 static void
1064 em_watchdog(struct ifnet *ifp)
1065 {
1066         struct adapter *adapter = ifp->if_softc;
1067
1068         ASSERT_SERIALIZED(ifp->if_serializer);
1069
1070         /*
1071          * The timer is set to 5 every time start queues a packet.
1072          * Then txeof keeps resetting it as long as it cleans at
1073          * least one descriptor.
1074          * Finally, anytime all descriptors are clean the timer is
1075          * set to 0.
1076          */
1077
1078         if (E1000_READ_REG(&adapter->hw, E1000_TDT(0)) ==
1079             E1000_READ_REG(&adapter->hw, E1000_TDH(0))) {
1080                 /*
1081                  * If we reach here, all TX jobs are completed and
1082                  * the TX engine should have been idled for some time.
1083                  * We don't need to call if_devstart() here.
1084                  */
1085                 ifp->if_flags &= ~IFF_OACTIVE;
1086                 ifp->if_timer = 0;
1087                 return;
1088         }
1089
1090         /*
1091          * If we are in this routine because of pause frames, then
1092          * don't reset the hardware.
1093          */
1094         if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
1095             E1000_STATUS_TXOFF) {
1096                 ifp->if_timer = EM_TX_TIMEOUT;
1097                 return;
1098         }
1099
1100         if (e1000_check_for_link(&adapter->hw) == 0)
1101                 if_printf(ifp, "watchdog timeout -- resetting\n");
1102
1103         ifp->if_oerrors++;
1104         adapter->watchdog_events++;
1105
1106         em_init(adapter);
1107
1108         if (!ifq_is_empty(&ifp->if_snd))
1109                 if_devstart(ifp);
1110 }
1111
1112 static void
1113 em_init(void *xsc)
1114 {
1115         struct adapter *adapter = xsc;
1116         struct ifnet *ifp = &adapter->arpcom.ac_if;
1117         device_t dev = adapter->dev;
1118         uint32_t pba;
1119
1120         ASSERT_SERIALIZED(ifp->if_serializer);
1121
1122         em_stop(adapter);
1123
1124         /*
1125          * Packet Buffer Allocation (PBA)
1126          * Writing PBA sets the receive portion of the buffer
1127          * the remainder is used for the transmit buffer.
1128          *
1129          * Devices before the 82547 had a Packet Buffer of 64K.
1130          *   Default allocation: PBA=48K for Rx, leaving 16K for Tx.
1131          * After the 82547 the buffer was reduced to 40K.
1132          *   Default allocation: PBA=30K for Rx, leaving 10K for Tx.
1133          *   Note: default does not leave enough room for Jumbo Frame >10k.
1134          */
1135         switch (adapter->hw.mac.type) {
1136         case e1000_82547:
1137         case e1000_82547_rev_2: /* 82547: Total Packet Buffer is 40K */
1138                 if (adapter->max_frame_size > 8192)
1139                         pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
1140                 else
1141                         pba = E1000_PBA_30K; /* 30K for Rx, 10K for Tx */
1142                 adapter->tx_fifo_head = 0;
1143                 adapter->tx_head_addr = pba << EM_TX_HEAD_ADDR_SHIFT;
1144                 adapter->tx_fifo_size =
1145                     (E1000_PBA_40K - pba) << EM_PBA_BYTES_SHIFT;
1146                 break;
1147
1148         /* Total Packet Buffer on these is 48K */
1149         case e1000_82571:
1150         case e1000_82572:
1151         case e1000_80003es2lan:
1152                 pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1153                 break;
1154
1155         case e1000_82573: /* 82573: Total Packet Buffer is 32K */
1156                 pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
1157                 break;
1158
1159         case e1000_82574:
1160                 pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
1161                 break;
1162
1163         case e1000_ich9lan:
1164         case e1000_ich10lan:
1165 #define E1000_PBA_10K   0x000A
1166                 pba = E1000_PBA_10K;
1167                 break;
1168
1169         case e1000_ich8lan:
1170                 pba = E1000_PBA_8K;
1171                 break;
1172
1173         default:
1174                 /* Devices before 82547 had a Packet Buffer of 64K.   */
1175                 if (adapter->max_frame_size > 8192)
1176                         pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
1177                 else
1178                         pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
1179         }
1180         E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba);
1181
1182         /* Get the latest mac address, User can use a LAA */
1183         bcopy(IF_LLADDR(ifp), adapter->hw.mac.addr, ETHER_ADDR_LEN);
1184
1185         /* Put the address into the Receive Address Array */
1186         e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1187
1188         /*
1189          * With the 82571 adapter, RAR[0] may be overwritten
1190          * when the other port is reset, we make a duplicate
1191          * in RAR[14] for that eventuality, this assures
1192          * the interface continues to function.
1193          */
1194         if (adapter->hw.mac.type == e1000_82571) {
1195                 e1000_set_laa_state_82571(&adapter->hw, TRUE);
1196                 e1000_rar_set(&adapter->hw, adapter->hw.mac.addr,
1197                     E1000_RAR_ENTRIES - 1);
1198         }
1199
1200         /* Initialize the hardware */
1201         if (em_hw_init(adapter)) {
1202                 device_printf(dev, "Unable to initialize the hardware\n");
1203                 /* XXX em_stop()? */
1204                 return;
1205         }
1206         em_update_link_status(adapter);
1207
1208         /* Setup VLAN support, basic and offload if available */
1209         E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
1210
1211         if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1212                 uint32_t ctrl;
1213
1214                 ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
1215                 ctrl |= E1000_CTRL_VME;
1216                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
1217         }
1218
1219         /* Set hardware offload abilities */
1220         if (ifp->if_capenable & IFCAP_TXCSUM)
1221                 ifp->if_hwassist = EM_CSUM_FEATURES;
1222         else
1223                 ifp->if_hwassist = 0;
1224
1225         /* Configure for OS presence */
1226         em_get_mgmt(adapter);
1227
1228         /* Prepare transmit descriptors and buffers */
1229         em_init_tx_ring(adapter);
1230         em_init_tx_unit(adapter);
1231
1232         /* Setup Multicast table */
1233         em_set_multi(adapter);
1234
1235         /* Prepare receive descriptors and buffers */
1236         if (em_init_rx_ring(adapter)) {
1237                 device_printf(dev, "Could not setup receive structures\n");
1238                 em_stop(adapter);
1239                 return;
1240         }
1241         em_init_rx_unit(adapter);
1242
1243         /* Don't lose promiscuous settings */
1244         em_set_promisc(adapter);
1245
1246         ifp->if_flags |= IFF_RUNNING;
1247         ifp->if_flags &= ~IFF_OACTIVE;
1248
1249         callout_reset(&adapter->timer, hz, em_timer, adapter);
1250         e1000_clear_hw_cntrs_base_generic(&adapter->hw);
1251
1252         /* MSI/X configuration for 82574 */
1253         if (adapter->hw.mac.type == e1000_82574) {
1254                 int tmp;
1255
1256                 tmp = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
1257                 tmp |= E1000_CTRL_EXT_PBA_CLR;
1258                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, tmp);
1259                 /*
1260                  * Set the IVAR - interrupt vector routing.
1261                  * Each nibble represents a vector, high bit
1262                  * is enable, other 3 bits are the MSIX table
1263                  * entry, we map RXQ0 to 0, TXQ0 to 1, and
1264                  * Link (other) to 2, hence the magic number.
1265                  */
1266                 E1000_WRITE_REG(&adapter->hw, E1000_IVAR, 0x800A0908);
1267         }
1268
1269 #ifdef DEVICE_POLLING
1270         /*
1271          * Only enable interrupts if we are not polling, make sure
1272          * they are off otherwise.
1273          */
1274         if (ifp->if_flags & IFF_POLLING)
1275                 em_disable_intr(adapter);
1276         else
1277 #endif /* DEVICE_POLLING */
1278                 em_enable_intr(adapter);
1279
1280         /* Don't reset the phy next time init gets called */
1281         adapter->hw.phy.reset_disable = TRUE;
1282 }
1283
1284 #ifdef DEVICE_POLLING
1285
1286 static void
1287 em_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1288 {
1289         struct adapter *adapter = ifp->if_softc;
1290         uint32_t reg_icr;
1291
1292         ASSERT_SERIALIZED(ifp->if_serializer);
1293
1294         switch (cmd) {
1295         case POLL_REGISTER:
1296                 em_disable_intr(adapter);
1297                 break;
1298
1299         case POLL_DEREGISTER:
1300                 em_enable_intr(adapter);
1301                 break;
1302
1303         case POLL_AND_CHECK_STATUS:
1304                 reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
1305                 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1306                         callout_stop(&adapter->timer);
1307                         adapter->hw.mac.get_link_status = 1;
1308                         em_update_link_status(adapter);
1309                         callout_reset(&adapter->timer, hz, em_timer, adapter);
1310                 }
1311                 /* FALL THROUGH */
1312         case POLL_ONLY:
1313                 if (ifp->if_flags & IFF_RUNNING) {
1314                         em_rxeof(adapter, count);
1315                         em_txeof(adapter);
1316
1317                         if (!ifq_is_empty(&ifp->if_snd))
1318                                 if_devstart(ifp);
1319                 }
1320                 break;
1321         }
1322 }
1323
1324 #endif /* DEVICE_POLLING */
1325
1326 static void
1327 em_intr(void *xsc)
1328 {
1329         struct adapter *adapter = xsc;
1330         struct ifnet *ifp = &adapter->arpcom.ac_if;
1331         uint32_t reg_icr;
1332
1333         logif(intr_beg);
1334         ASSERT_SERIALIZED(ifp->if_serializer);
1335
1336         reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
1337
1338         if ((adapter->hw.mac.type >= e1000_82571 &&
1339              (reg_icr & E1000_ICR_INT_ASSERTED) == 0) ||
1340             reg_icr == 0) {
1341                 logif(intr_end);
1342                 return;
1343         }
1344
1345         /*
1346          * XXX: some laptops trigger several spurious interrupts
1347          * on em(4) when in the resume cycle. The ICR register
1348          * reports all-ones value in this case. Processing such
1349          * interrupts would lead to a freeze. I don't know why.
1350          */
1351         if (reg_icr == 0xffffffff) {
1352                 logif(intr_end);
1353                 return;
1354         }
1355
1356         if (ifp->if_flags & IFF_RUNNING) {
1357                 if (reg_icr &
1358                     (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO))
1359                         em_rxeof(adapter, -1);
1360                 if (reg_icr & E1000_ICR_TXDW) {
1361                         em_txeof(adapter);
1362                         if (!ifq_is_empty(&ifp->if_snd))
1363                                 if_devstart(ifp);
1364                 }
1365         }
1366
1367         /* Link status change */
1368         if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1369                 callout_stop(&adapter->timer);
1370                 adapter->hw.mac.get_link_status = 1;
1371                 em_update_link_status(adapter);
1372
1373                 /* Deal with TX cruft when link lost */
1374                 em_tx_purge(adapter);
1375
1376                 callout_reset(&adapter->timer, hz, em_timer, adapter);
1377         }
1378
1379         if (reg_icr & E1000_ICR_RXO)
1380                 adapter->rx_overruns++;
1381
1382         logif(intr_end);
1383 }
1384
1385 static void
1386 em_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1387 {
1388         struct adapter *adapter = ifp->if_softc;
1389         u_char fiber_type = IFM_1000_SX;
1390
1391         ASSERT_SERIALIZED(ifp->if_serializer);
1392
1393         em_update_link_status(adapter);
1394
1395         ifmr->ifm_status = IFM_AVALID;
1396         ifmr->ifm_active = IFM_ETHER;
1397
1398         if (!adapter->link_active)
1399                 return;
1400
1401         ifmr->ifm_status |= IFM_ACTIVE;
1402
1403         if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
1404             adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
1405                 if (adapter->hw.mac.type == e1000_82545)
1406                         fiber_type = IFM_1000_LX;
1407                 ifmr->ifm_active |= fiber_type | IFM_FDX;
1408         } else {
1409                 switch (adapter->link_speed) {
1410                 case 10:
1411                         ifmr->ifm_active |= IFM_10_T;
1412                         break;
1413                 case 100:
1414                         ifmr->ifm_active |= IFM_100_TX;
1415                         break;
1416
1417                 case 1000:
1418                         ifmr->ifm_active |= IFM_1000_T;
1419                         break;
1420                 }
1421                 if (adapter->link_duplex == FULL_DUPLEX)
1422                         ifmr->ifm_active |= IFM_FDX;
1423                 else
1424                         ifmr->ifm_active |= IFM_HDX;
1425         }
1426 }
1427
1428 static int
1429 em_media_change(struct ifnet *ifp)
1430 {
1431         struct adapter *adapter = ifp->if_softc;
1432         struct ifmedia *ifm = &adapter->media;
1433
1434         ASSERT_SERIALIZED(ifp->if_serializer);
1435
1436         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1437                 return (EINVAL);
1438
1439         switch (IFM_SUBTYPE(ifm->ifm_media)) {
1440         case IFM_AUTO:
1441                 adapter->hw.mac.autoneg = DO_AUTO_NEG;
1442                 adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1443                 break;
1444
1445         case IFM_1000_LX:
1446         case IFM_1000_SX:
1447         case IFM_1000_T:
1448                 adapter->hw.mac.autoneg = DO_AUTO_NEG;
1449                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1450                 break;
1451
1452         case IFM_100_TX:
1453                 adapter->hw.mac.autoneg = FALSE;
1454                 adapter->hw.phy.autoneg_advertised = 0;
1455                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1456                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1457                 else
1458                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1459                 break;
1460
1461         case IFM_10_T:
1462                 adapter->hw.mac.autoneg = FALSE;
1463                 adapter->hw.phy.autoneg_advertised = 0;
1464                 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1465                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1466                 else
1467                         adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1468                 break;
1469
1470         default:
1471                 if_printf(ifp, "Unsupported media type\n");
1472                 break;
1473         }
1474
1475         /*
1476          * As the speed/duplex settings my have changed we need to
1477          * reset the PHY.
1478          */
1479         adapter->hw.phy.reset_disable = FALSE;
1480
1481         em_init(adapter);
1482
1483         return (0);
1484 }
1485
1486 static int
1487 em_encap(struct adapter *adapter, struct mbuf **m_headp)
1488 {
1489         bus_dma_segment_t segs[EM_MAX_SCATTER];
1490         bus_dmamap_t map;
1491         struct em_buffer *tx_buffer, *tx_buffer_mapped;
1492         struct e1000_tx_desc *ctxd = NULL;
1493         struct mbuf *m_head = *m_headp;
1494         uint32_t txd_upper, txd_lower, txd_used, cmd = 0;
1495         int maxsegs, nsegs, i, j, first, last = 0, error;
1496
1497         if (m_head->m_len < EM_TXCSUM_MINHL &&
1498             (m_head->m_flags & EM_CSUM_FEATURES)) {
1499                 /*
1500                  * Make sure that ethernet header and ip.ip_hl are in
1501                  * contiguous memory, since if TXCSUM is enabled, later
1502                  * TX context descriptor's setup need to access ip.ip_hl.
1503                  */
1504                 error = em_txcsum_pullup(adapter, m_headp);
1505                 if (error) {
1506                         KKASSERT(*m_headp == NULL);
1507                         return error;
1508                 }
1509                 m_head = *m_headp;
1510         }
1511
1512         txd_upper = txd_lower = 0;
1513         txd_used = 0;
1514
1515         /*
1516          * Capture the first descriptor index, this descriptor
1517          * will have the index of the EOP which is the only one
1518          * that now gets a DONE bit writeback.
1519          */
1520         first = adapter->next_avail_tx_desc;
1521         tx_buffer = &adapter->tx_buffer_area[first];
1522         tx_buffer_mapped = tx_buffer;
1523         map = tx_buffer->map;
1524
1525         maxsegs = adapter->num_tx_desc_avail - EM_TX_RESERVED;
1526         KASSERT(maxsegs >= adapter->spare_tx_desc,
1527                 ("not enough spare TX desc\n"));
1528         if (adapter->pcix_82544) {
1529                 /* Half it; see the comment in em_attach() */
1530                 maxsegs >>= 1;
1531         }
1532         if (maxsegs > EM_MAX_SCATTER)
1533                 maxsegs = EM_MAX_SCATTER;
1534
1535         error = bus_dmamap_load_mbuf_defrag(adapter->txtag, map, m_headp,
1536                         segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1537         if (error) {
1538                 if (error == ENOBUFS)
1539                         adapter->mbuf_alloc_failed++;
1540                 else
1541                         adapter->no_tx_dma_setup++;
1542
1543                 m_freem(*m_headp);
1544                 *m_headp = NULL;
1545                 return error;
1546         }
1547         bus_dmamap_sync(adapter->txtag, map, BUS_DMASYNC_PREWRITE);
1548
1549         m_head = *m_headp;
1550         adapter->tx_nsegs += nsegs;
1551
1552         if (m_head->m_pkthdr.csum_flags & EM_CSUM_FEATURES) {
1553                 /* TX csum offloading will consume one TX desc */
1554                 adapter->tx_nsegs += em_txcsum(adapter, m_head,
1555                                                &txd_upper, &txd_lower);
1556         }
1557         i = adapter->next_avail_tx_desc;
1558
1559         /* Set up our transmit descriptors */
1560         for (j = 0; j < nsegs; j++) {
1561                 /* If adapter is 82544 and on PCIX bus */
1562                 if(adapter->pcix_82544) {
1563                         DESC_ARRAY desc_array;
1564                         uint32_t array_elements, counter;
1565
1566                         /*
1567                          * Check the Address and Length combination and
1568                          * split the data accordingly
1569                          */
1570                         array_elements = em_82544_fill_desc(segs[j].ds_addr,
1571                                                 segs[j].ds_len, &desc_array);
1572                         for (counter = 0; counter < array_elements; counter++) {
1573                                 KKASSERT(txd_used < adapter->num_tx_desc_avail);
1574
1575                                 tx_buffer = &adapter->tx_buffer_area[i];
1576                                 ctxd = &adapter->tx_desc_base[i];
1577
1578                                 ctxd->buffer_addr = htole64(
1579                                     desc_array.descriptor[counter].address);
1580                                 ctxd->lower.data = htole32(
1581                                     E1000_TXD_CMD_IFCS | txd_lower |
1582                                     desc_array.descriptor[counter].length);
1583                                 ctxd->upper.data = htole32(txd_upper);
1584
1585                                 last = i;
1586                                 if (++i == adapter->num_tx_desc)
1587                                         i = 0;
1588
1589                                 txd_used++;
1590                         }
1591                 } else {
1592                         tx_buffer = &adapter->tx_buffer_area[i];
1593                         ctxd = &adapter->tx_desc_base[i];
1594
1595                         ctxd->buffer_addr = htole64(segs[j].ds_addr);
1596                         ctxd->lower.data = htole32(E1000_TXD_CMD_IFCS |
1597                                                    txd_lower | segs[j].ds_len);
1598                         ctxd->upper.data = htole32(txd_upper);
1599
1600                         last = i;
1601                         if (++i == adapter->num_tx_desc)
1602                                 i = 0;
1603                 }
1604         }
1605
1606         adapter->next_avail_tx_desc = i;
1607         if (adapter->pcix_82544) {
1608                 KKASSERT(adapter->num_tx_desc_avail > txd_used);
1609                 adapter->num_tx_desc_avail -= txd_used;
1610         } else {
1611                 KKASSERT(adapter->num_tx_desc_avail > nsegs);
1612                 adapter->num_tx_desc_avail -= nsegs;
1613         }
1614
1615         /* Handle VLAN tag */
1616         if (m_head->m_flags & M_VLANTAG) {
1617                 /* Set the vlan id. */
1618                 ctxd->upper.fields.special =
1619                     htole16(m_head->m_pkthdr.ether_vlantag);
1620
1621                 /* Tell hardware to add tag */
1622                 ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE);
1623         }
1624
1625         tx_buffer->m_head = m_head;
1626         tx_buffer_mapped->map = tx_buffer->map;
1627         tx_buffer->map = map;
1628
1629         if (adapter->tx_nsegs >= adapter->tx_int_nsegs) {
1630                 adapter->tx_nsegs = 0;
1631
1632                 /*
1633                  * Report Status (RS) is turned on
1634                  * every tx_int_nsegs descriptors.
1635                  */
1636                 cmd = E1000_TXD_CMD_RS;
1637
1638                 /*
1639                  * Keep track of the descriptor, which will
1640                  * be written back by hardware.
1641                  */
1642                 adapter->tx_dd[adapter->tx_dd_tail] = last;
1643                 EM_INC_TXDD_IDX(adapter->tx_dd_tail);
1644                 KKASSERT(adapter->tx_dd_tail != adapter->tx_dd_head);
1645         }
1646
1647         /*
1648          * Last Descriptor of Packet needs End Of Packet (EOP)
1649          */
1650         ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | cmd);
1651
1652         /*
1653          * Advance the Transmit Descriptor Tail (TDT), this tells the E1000
1654          * that this frame is available to transmit.
1655          */
1656         if (adapter->hw.mac.type == e1000_82547 &&
1657             adapter->link_duplex == HALF_DUPLEX) {
1658                 em_82547_move_tail_serialized(adapter);
1659         } else {
1660                 E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), i);
1661                 if (adapter->hw.mac.type == e1000_82547) {
1662                         em_82547_update_fifo_head(adapter,
1663                             m_head->m_pkthdr.len);
1664                 }
1665         }
1666         return (0);
1667 }
1668
1669 /*
1670  * 82547 workaround to avoid controller hang in half-duplex environment.
1671  * The workaround is to avoid queuing a large packet that would span
1672  * the internal Tx FIFO ring boundary.  We need to reset the FIFO pointers
1673  * in this case.  We do that only when FIFO is quiescent.
1674  */
1675 static void
1676 em_82547_move_tail_serialized(struct adapter *adapter)
1677 {
1678         struct e1000_tx_desc *tx_desc;
1679         uint16_t hw_tdt, sw_tdt, length = 0;
1680         bool eop = 0;
1681
1682         ASSERT_SERIALIZED(adapter->arpcom.ac_if.if_serializer);
1683
1684         hw_tdt = E1000_READ_REG(&adapter->hw, E1000_TDT(0));
1685         sw_tdt = adapter->next_avail_tx_desc;
1686
1687         while (hw_tdt != sw_tdt) {
1688                 tx_desc = &adapter->tx_desc_base[hw_tdt];
1689                 length += tx_desc->lower.flags.length;
1690                 eop = tx_desc->lower.data & E1000_TXD_CMD_EOP;
1691                 if (++hw_tdt == adapter->num_tx_desc)
1692                         hw_tdt = 0;
1693
1694                 if (eop) {
1695                         if (em_82547_fifo_workaround(adapter, length)) {
1696                                 adapter->tx_fifo_wrk_cnt++;
1697                                 callout_reset(&adapter->tx_fifo_timer, 1,
1698                                         em_82547_move_tail, adapter);
1699                                 break;
1700                         }
1701                         E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), hw_tdt);
1702                         em_82547_update_fifo_head(adapter, length);
1703                         length = 0;
1704                 }
1705         }
1706 }
1707
1708 static void
1709 em_82547_move_tail(void *xsc)
1710 {
1711         struct adapter *adapter = xsc;
1712         struct ifnet *ifp = &adapter->arpcom.ac_if;
1713
1714         lwkt_serialize_enter(ifp->if_serializer);
1715         em_82547_move_tail_serialized(adapter);
1716         lwkt_serialize_exit(ifp->if_serializer);
1717 }
1718
1719 static int
1720 em_82547_fifo_workaround(struct adapter *adapter, int len)
1721 {
1722         int fifo_space, fifo_pkt_len;
1723
1724         fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1725
1726         if (adapter->link_duplex == HALF_DUPLEX) {
1727                 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
1728
1729                 if (fifo_pkt_len >= (EM_82547_PKT_THRESH + fifo_space)) {
1730                         if (em_82547_tx_fifo_reset(adapter))
1731                                 return (0);
1732                         else
1733                                 return (1);
1734                 }
1735         }
1736         return (0);
1737 }
1738
1739 static void
1740 em_82547_update_fifo_head(struct adapter *adapter, int len)
1741 {
1742         int fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1743
1744         /* tx_fifo_head is always 16 byte aligned */
1745         adapter->tx_fifo_head += fifo_pkt_len;
1746         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
1747                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
1748 }
1749
1750 static int
1751 em_82547_tx_fifo_reset(struct adapter *adapter)
1752 {
1753         uint32_t tctl;
1754
1755         if ((E1000_READ_REG(&adapter->hw, E1000_TDT(0)) ==
1756              E1000_READ_REG(&adapter->hw, E1000_TDH(0))) &&
1757             (E1000_READ_REG(&adapter->hw, E1000_TDFT) ==
1758              E1000_READ_REG(&adapter->hw, E1000_TDFH)) &&
1759             (E1000_READ_REG(&adapter->hw, E1000_TDFTS) ==
1760              E1000_READ_REG(&adapter->hw, E1000_TDFHS)) &&
1761             (E1000_READ_REG(&adapter->hw, E1000_TDFPC) == 0)) {
1762                 /* Disable TX unit */
1763                 tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
1764                 E1000_WRITE_REG(&adapter->hw, E1000_TCTL,
1765                     tctl & ~E1000_TCTL_EN);
1766
1767                 /* Reset FIFO pointers */
1768                 E1000_WRITE_REG(&adapter->hw, E1000_TDFT,
1769                     adapter->tx_head_addr);
1770                 E1000_WRITE_REG(&adapter->hw, E1000_TDFH,
1771                     adapter->tx_head_addr);
1772                 E1000_WRITE_REG(&adapter->hw, E1000_TDFTS,
1773                     adapter->tx_head_addr);
1774                 E1000_WRITE_REG(&adapter->hw, E1000_TDFHS,
1775                     adapter->tx_head_addr);
1776
1777                 /* Re-enable TX unit */
1778                 E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
1779                 E1000_WRITE_FLUSH(&adapter->hw);
1780
1781                 adapter->tx_fifo_head = 0;
1782                 adapter->tx_fifo_reset_cnt++;
1783
1784                 return (TRUE);
1785         } else {
1786                 return (FALSE);
1787         }
1788 }
1789
1790 static void
1791 em_set_promisc(struct adapter *adapter)
1792 {
1793         struct ifnet *ifp = &adapter->arpcom.ac_if;
1794         uint32_t reg_rctl;
1795
1796         reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1797
1798         if (ifp->if_flags & IFF_PROMISC) {
1799                 reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1800                 /* Turn this on if you want to see bad packets */
1801                 if (em_debug_sbp)
1802                         reg_rctl |= E1000_RCTL_SBP;
1803                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1804         } else if (ifp->if_flags & IFF_ALLMULTI) {
1805                 reg_rctl |= E1000_RCTL_MPE;
1806                 reg_rctl &= ~E1000_RCTL_UPE;
1807                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1808         }
1809 }
1810
1811 static void
1812 em_disable_promisc(struct adapter *adapter)
1813 {
1814         uint32_t reg_rctl;
1815
1816         reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1817
1818         reg_rctl &= ~E1000_RCTL_UPE;
1819         reg_rctl &= ~E1000_RCTL_MPE;
1820         reg_rctl &= ~E1000_RCTL_SBP;
1821         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1822 }
1823
1824 static void
1825 em_set_multi(struct adapter *adapter)
1826 {
1827         struct ifnet *ifp = &adapter->arpcom.ac_if;
1828         struct ifmultiaddr *ifma;
1829         uint32_t reg_rctl = 0;
1830         uint8_t  mta[512]; /* Largest MTS is 4096 bits */
1831         int mcnt = 0;
1832
1833         if (adapter->hw.mac.type == e1000_82542 &&
1834             adapter->hw.revision_id == E1000_REVISION_2) {
1835                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1836                 if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1837                         e1000_pci_clear_mwi(&adapter->hw);
1838                 reg_rctl |= E1000_RCTL_RST;
1839                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1840                 msec_delay(5);
1841         }
1842
1843         LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1844                 if (ifma->ifma_addr->sa_family != AF_LINK)
1845                         continue;
1846
1847                 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1848                         break;
1849
1850                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1851                     &mta[mcnt * ETHER_ADDR_LEN], ETHER_ADDR_LEN);
1852                 mcnt++;
1853         }
1854
1855         if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1856                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1857                 reg_rctl |= E1000_RCTL_MPE;
1858                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1859         } else {
1860                 e1000_update_mc_addr_list(&adapter->hw, mta,
1861                     mcnt, 1, adapter->hw.mac.rar_entry_count);
1862         }
1863
1864         if (adapter->hw.mac.type == e1000_82542 &&
1865             adapter->hw.revision_id == E1000_REVISION_2) {
1866                 reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
1867                 reg_rctl &= ~E1000_RCTL_RST;
1868                 E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
1869                 msec_delay(5);
1870                 if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1871                         e1000_pci_set_mwi(&adapter->hw);
1872         }
1873 }
1874
1875 /*
1876  * This routine checks for link status and updates statistics.
1877  */
1878 static void
1879 em_timer(void *xsc)
1880 {
1881         struct adapter *adapter = xsc;
1882         struct ifnet *ifp = &adapter->arpcom.ac_if;
1883
1884         lwkt_serialize_enter(ifp->if_serializer);
1885
1886         em_update_link_status(adapter);
1887         em_update_stats(adapter);
1888
1889         /* Reset LAA into RAR[0] on 82571 */
1890         if (e1000_get_laa_state_82571(&adapter->hw) == TRUE)
1891                 e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1892
1893         if (em_display_debug_stats && (ifp->if_flags & IFF_RUNNING))
1894                 em_print_hw_stats(adapter);
1895
1896         em_smartspeed(adapter);
1897
1898         callout_reset(&adapter->timer, hz, em_timer, adapter);
1899
1900         lwkt_serialize_exit(ifp->if_serializer);
1901 }
1902
1903 static void
1904 em_update_link_status(struct adapter *adapter)
1905 {
1906         struct e1000_hw *hw = &adapter->hw;
1907         struct ifnet *ifp = &adapter->arpcom.ac_if;
1908         device_t dev = adapter->dev;
1909         uint32_t link_check = 0;
1910
1911         /* Get the cached link value or read phy for real */
1912         switch (hw->phy.media_type) {
1913         case e1000_media_type_copper:
1914                 if (hw->mac.get_link_status) {
1915                         /* Do the work to read phy */
1916                         e1000_check_for_link(hw);
1917                         link_check = !hw->mac.get_link_status;
1918                         if (link_check) /* ESB2 fix */
1919                                 e1000_cfg_on_link_up(hw);
1920                 } else {
1921                         link_check = TRUE;
1922                 }
1923                 break;
1924
1925         case e1000_media_type_fiber:
1926                 e1000_check_for_link(hw);
1927                 link_check =
1928                         E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1929                 break;
1930
1931         case e1000_media_type_internal_serdes:
1932                 e1000_check_for_link(hw);
1933                 link_check = adapter->hw.mac.serdes_has_link;
1934                 break;
1935
1936         case e1000_media_type_unknown:
1937         default:
1938                 break;
1939         }
1940
1941         /* Now check for a transition */
1942         if (link_check && adapter->link_active == 0) {
1943                 e1000_get_speed_and_duplex(hw, &adapter->link_speed,
1944                     &adapter->link_duplex);
1945
1946                 /*
1947                  * Check if we should enable/disable SPEED_MODE bit on
1948                  * 82571/82572
1949                  */
1950                 if (hw->mac.type == e1000_82571 ||
1951                     hw->mac.type == e1000_82572) {
1952                         int tarc0;
1953
1954                         tarc0 = E1000_READ_REG(hw, E1000_TARC(0));
1955                         if (adapter->link_speed != SPEED_1000)
1956                                 tarc0 &= ~SPEED_MODE_BIT;
1957                         else
1958                                 tarc0 |= SPEED_MODE_BIT;
1959                         E1000_WRITE_REG(hw, E1000_TARC(0), tarc0);
1960                 }
1961                 if (bootverbose) {
1962                         device_printf(dev, "Link is up %d Mbps %s\n",
1963                             adapter->link_speed,
1964                             ((adapter->link_duplex == FULL_DUPLEX) ?
1965                             "Full Duplex" : "Half Duplex"));
1966                 }
1967                 adapter->link_active = 1;
1968                 adapter->smartspeed = 0;
1969                 ifp->if_baudrate = adapter->link_speed * 1000000;
1970                 ifp->if_link_state = LINK_STATE_UP;
1971                 if_link_state_change(ifp);
1972         } else if (!link_check && adapter->link_active == 1) {
1973                 ifp->if_baudrate = adapter->link_speed = 0;
1974                 adapter->link_duplex = 0;
1975                 if (bootverbose)
1976                         device_printf(dev, "Link is Down\n");
1977                 adapter->link_active = 0;
1978 #if 0
1979                 /* Link down, disable watchdog */
1980                 if->if_timer = 0;
1981 #endif
1982                 ifp->if_link_state = LINK_STATE_DOWN;
1983                 if_link_state_change(ifp);
1984         }
1985 }
1986
1987 static void
1988 em_stop(struct adapter *adapter)
1989 {
1990         struct ifnet *ifp = &adapter->arpcom.ac_if;
1991         int i;
1992
1993         ASSERT_SERIALIZED(ifp->if_serializer);
1994
1995         em_disable_intr(adapter);
1996
1997         callout_stop(&adapter->timer);
1998         callout_stop(&adapter->tx_fifo_timer);
1999
2000         ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2001         ifp->if_timer = 0;
2002
2003         e1000_reset_hw(&adapter->hw);
2004         if (adapter->hw.mac.type >= e1000_82544)
2005                 E1000_WRITE_REG(&adapter->hw, E1000_WUC, 0);
2006
2007         for (i = 0; i < adapter->num_tx_desc; i++) {
2008                 struct em_buffer *tx_buffer = &adapter->tx_buffer_area[i];
2009
2010                 if (tx_buffer->m_head != NULL) {
2011                         bus_dmamap_unload(adapter->txtag, tx_buffer->map);
2012                         m_freem(tx_buffer->m_head);
2013                         tx_buffer->m_head = NULL;
2014                 }
2015         }
2016
2017         for (i = 0; i < adapter->num_rx_desc; i++) {
2018                 struct em_buffer *rx_buffer = &adapter->rx_buffer_area[i];
2019
2020                 if (rx_buffer->m_head != NULL) {
2021                         bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
2022                         m_freem(rx_buffer->m_head);
2023                         rx_buffer->m_head = NULL;
2024                 }
2025         }
2026
2027         if (adapter->fmp != NULL)
2028                 m_freem(adapter->fmp);
2029         adapter->fmp = NULL;
2030         adapter->lmp = NULL;
2031
2032         adapter->csum_flags = 0;
2033         adapter->csum_ehlen = 0;
2034         adapter->csum_iphlen = 0;
2035
2036         adapter->tx_dd_head = 0;
2037         adapter->tx_dd_tail = 0;
2038         adapter->tx_nsegs = 0;
2039 }
2040
2041 static int
2042 em_get_hw_info(struct adapter *adapter)
2043 {
2044         device_t dev = adapter->dev;
2045
2046         /* Save off the information about this board */
2047         adapter->hw.vendor_id = pci_get_vendor(dev);
2048         adapter->hw.device_id = pci_get_device(dev);
2049         adapter->hw.revision_id = pci_get_revid(dev);
2050         adapter->hw.subsystem_vendor_id = pci_get_subvendor(dev);
2051         adapter->hw.subsystem_device_id = pci_get_subdevice(dev);
2052
2053         /* Do Shared Code Init and Setup */
2054         if (e1000_set_mac_type(&adapter->hw))
2055                 return ENXIO;
2056         return 0;
2057 }
2058
2059 static int
2060 em_alloc_pci_res(struct adapter *adapter)
2061 {
2062         device_t dev = adapter->dev;
2063         int val, rid;
2064
2065         /* Enable bus mastering */
2066         pci_enable_busmaster(dev);
2067
2068         adapter->memory_rid = EM_BAR_MEM;
2069         adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
2070                                 &adapter->memory_rid, RF_ACTIVE);
2071         if (adapter->memory == NULL) {
2072                 device_printf(dev, "Unable to allocate bus resource: memory\n");
2073                 return (ENXIO);
2074         }
2075         adapter->osdep.mem_bus_space_tag =
2076             rman_get_bustag(adapter->memory);
2077         adapter->osdep.mem_bus_space_handle =
2078             rman_get_bushandle(adapter->memory);
2079
2080         /* XXX This is quite goofy, it is not actually used */
2081         adapter->hw.hw_addr = (uint8_t *)&adapter->osdep.mem_bus_space_handle;
2082
2083         /* Only older adapters use IO mapping */
2084         if (adapter->hw.mac.type > e1000_82543 &&
2085             adapter->hw.mac.type < e1000_82571) {
2086                 /* Figure our where our IO BAR is ? */
2087                 for (rid = PCIR_BAR(0); rid < PCIR_CARDBUSCIS;) {
2088                         val = pci_read_config(dev, rid, 4);
2089                         if (EM_BAR_TYPE(val) == EM_BAR_TYPE_IO) {
2090                                 adapter->io_rid = rid;
2091                                 break;
2092                         }
2093                         rid += 4;
2094                         /* check for 64bit BAR */
2095                         if (EM_BAR_MEM_TYPE(val) == EM_BAR_MEM_TYPE_64BIT)
2096                                 rid += 4;
2097                 }
2098                 if (rid >= PCIR_CARDBUSCIS) {
2099                         device_printf(dev, "Unable to locate IO BAR\n");
2100                         return (ENXIO);
2101                 }
2102                 adapter->ioport = bus_alloc_resource_any(dev, SYS_RES_IOPORT,
2103                                         &adapter->io_rid, RF_ACTIVE);
2104                 if (adapter->ioport == NULL) {
2105                         device_printf(dev, "Unable to allocate bus resource: "
2106                             "ioport\n");
2107                         return (ENXIO);
2108                 }
2109                 adapter->hw.io_base = 0;
2110                 adapter->osdep.io_bus_space_tag =
2111                     rman_get_bustag(adapter->ioport);
2112                 adapter->osdep.io_bus_space_handle =
2113                     rman_get_bushandle(adapter->ioport);
2114         }
2115
2116         adapter->intr_rid = 0;
2117         adapter->intr_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
2118                                 &adapter->intr_rid,
2119                                 RF_SHAREABLE | RF_ACTIVE);
2120         if (adapter->intr_res == NULL) {
2121                 device_printf(dev, "Unable to allocate bus resource: "
2122                     "interrupt\n");
2123                 return (ENXIO);
2124         }
2125
2126         adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
2127         adapter->hw.back = &adapter->osdep;
2128         return (0);
2129 }
2130
2131 static void
2132 em_free_pci_res(struct adapter *adapter)
2133 {
2134         device_t dev = adapter->dev;
2135
2136         if (adapter->intr_res != NULL) {
2137                 bus_release_resource(dev, SYS_RES_IRQ,
2138                     adapter->intr_rid, adapter->intr_res);
2139         }
2140
2141         if (adapter->memory != NULL) {
2142                 bus_release_resource(dev, SYS_RES_MEMORY,
2143                     adapter->memory_rid, adapter->memory);
2144         }
2145
2146         if (adapter->flash != NULL) {
2147                 bus_release_resource(dev, SYS_RES_MEMORY,
2148                     adapter->flash_rid, adapter->flash);
2149         }
2150
2151         if (adapter->ioport != NULL) {
2152                 bus_release_resource(dev, SYS_RES_IOPORT,
2153                     adapter->io_rid, adapter->ioport);
2154         }
2155 }
2156
2157 static int
2158 em_hw_init(struct adapter *adapter)
2159 {
2160         device_t dev = adapter->dev;
2161         uint16_t rx_buffer_size;
2162
2163         /* Issue a global reset */
2164         e1000_reset_hw(&adapter->hw);
2165
2166         /* Get control from any management/hw control */
2167         if ((adapter->hw.mac.type == e1000_82573 ||
2168              adapter->hw.mac.type == e1000_ich8lan ||
2169              adapter->hw.mac.type == e1000_ich10lan ||
2170              adapter->hw.mac.type == e1000_ich9lan) &&
2171             e1000_check_mng_mode(&adapter->hw))
2172                 em_get_hw_control(adapter);
2173
2174         /* When hardware is reset, fifo_head is also reset */
2175         adapter->tx_fifo_head = 0;
2176
2177         /* Set up smart power down as default off on newer adapters. */
2178         if (!em_smart_pwr_down &&
2179             (adapter->hw.mac.type == e1000_82571 ||
2180              adapter->hw.mac.type == e1000_82572)) {
2181                 uint16_t phy_tmp = 0;
2182
2183                 /* Speed up time to link by disabling smart power down. */
2184                 e1000_read_phy_reg(&adapter->hw,
2185                     IGP02E1000_PHY_POWER_MGMT, &phy_tmp);
2186                 phy_tmp &= ~IGP02E1000_PM_SPD;
2187                 e1000_write_phy_reg(&adapter->hw,
2188                     IGP02E1000_PHY_POWER_MGMT, phy_tmp);
2189         }
2190
2191         /*
2192          * These parameters control the automatic generation (Tx) and
2193          * response (Rx) to Ethernet PAUSE frames.
2194          * - High water mark should allow for at least two frames to be
2195          *   received after sending an XOFF.
2196          * - Low water mark works best when it is very near the high water mark.
2197          *   This allows the receiver to restart by sending XON when it has
2198          *   drained a bit. Here we use an arbitary value of 1500 which will
2199          *   restart after one full frame is pulled from the buffer. There
2200          *   could be several smaller frames in the buffer and if so they will
2201          *   not trigger the XON until their total number reduces the buffer
2202          *   by 1500.
2203          * - The pause time is fairly large at 1000 x 512ns = 512 usec.
2204          */
2205         rx_buffer_size =
2206                 (E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff) << 10;
2207
2208         adapter->hw.fc.high_water = rx_buffer_size -
2209                                     roundup2(adapter->max_frame_size, 1024);
2210         adapter->hw.fc.low_water = adapter->hw.fc.high_water - 1500;
2211
2212         if (adapter->hw.mac.type == e1000_80003es2lan)
2213                 adapter->hw.fc.pause_time = 0xFFFF;
2214         else
2215                 adapter->hw.fc.pause_time = EM_FC_PAUSE_TIME;
2216         adapter->hw.fc.send_xon = TRUE;
2217         adapter->hw.fc.requested_mode = e1000_fc_full;
2218
2219         if (e1000_init_hw(&adapter->hw) < 0) {
2220                 device_printf(dev, "Hardware Initialization Failed\n");
2221                 return (EIO);
2222         }
2223
2224         e1000_check_for_link(&adapter->hw);
2225
2226         return (0);
2227 }
2228
2229 static void
2230 em_setup_ifp(struct adapter *adapter)
2231 {
2232         struct ifnet *ifp = &adapter->arpcom.ac_if;
2233
2234         if_initname(ifp, device_get_name(adapter->dev),
2235                     device_get_unit(adapter->dev));
2236         ifp->if_softc = adapter;
2237         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2238         ifp->if_init =  em_init;
2239         ifp->if_ioctl = em_ioctl;
2240         ifp->if_start = em_start;
2241 #ifdef DEVICE_POLLING
2242         ifp->if_poll = em_poll;
2243 #endif
2244         ifp->if_watchdog = em_watchdog;
2245         ifq_set_maxlen(&ifp->if_snd, adapter->num_tx_desc - 1);
2246         ifq_set_ready(&ifp->if_snd);
2247
2248         ether_ifattach(ifp, adapter->hw.mac.addr, NULL);
2249
2250         if (adapter->hw.mac.type >= e1000_82543)
2251                 ifp->if_capabilities = IFCAP_HWCSUM;
2252
2253         ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2254         ifp->if_capenable = ifp->if_capabilities;
2255
2256         if (ifp->if_capenable & IFCAP_TXCSUM)
2257                 ifp->if_hwassist = EM_CSUM_FEATURES;
2258
2259         /*
2260          * Tell the upper layer(s) we support long frames.
2261          */
2262         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
2263
2264         /*
2265          * Specify the media types supported by this adapter and register
2266          * callbacks to update media and link information
2267          */
2268         ifmedia_init(&adapter->media, IFM_IMASK,
2269                      em_media_change, em_media_status);
2270         if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
2271             adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
2272                 u_char fiber_type = IFM_1000_SX; /* default type */
2273
2274                 if (adapter->hw.mac.type == e1000_82545)
2275                         fiber_type = IFM_1000_LX;
2276                 ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX,
2277                             0, NULL);
2278                 ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL);
2279         } else {
2280                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
2281                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
2282                             0, NULL);
2283                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
2284                             0, NULL);
2285                 ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
2286                             0, NULL);
2287                 if (adapter->hw.phy.type != e1000_phy_ife) {
2288                         ifmedia_add(&adapter->media,
2289                                 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
2290                         ifmedia_add(&adapter->media,
2291                                 IFM_ETHER | IFM_1000_T, 0, NULL);
2292                 }
2293         }
2294         ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2295         ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2296 }
2297
2298
2299 /*
2300  * Workaround for SmartSpeed on 82541 and 82547 controllers
2301  */
2302 static void
2303 em_smartspeed(struct adapter *adapter)
2304 {
2305         uint16_t phy_tmp;
2306
2307         if (adapter->link_active || adapter->hw.phy.type != e1000_phy_igp ||
2308             adapter->hw.mac.autoneg == 0 ||
2309             (adapter->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0)
2310                 return;
2311
2312         if (adapter->smartspeed == 0) {
2313                 /*
2314                  * If Master/Slave config fault is asserted twice,
2315                  * we assume back-to-back
2316                  */
2317                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2318                 if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
2319                         return;
2320                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2321                 if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
2322                         e1000_read_phy_reg(&adapter->hw,
2323                             PHY_1000T_CTRL, &phy_tmp);
2324                         if (phy_tmp & CR_1000T_MS_ENABLE) {
2325                                 phy_tmp &= ~CR_1000T_MS_ENABLE;
2326                                 e1000_write_phy_reg(&adapter->hw,
2327                                     PHY_1000T_CTRL, phy_tmp);
2328                                 adapter->smartspeed++;
2329                                 if (adapter->hw.mac.autoneg &&
2330                                     !e1000_phy_setup_autoneg(&adapter->hw) &&
2331                                     !e1000_read_phy_reg(&adapter->hw,
2332                                      PHY_CONTROL, &phy_tmp)) {
2333                                         phy_tmp |= MII_CR_AUTO_NEG_EN |
2334                                                    MII_CR_RESTART_AUTO_NEG;
2335                                         e1000_write_phy_reg(&adapter->hw,
2336                                             PHY_CONTROL, phy_tmp);
2337                                 }
2338                         }
2339                 }
2340                 return;
2341         } else if (adapter->smartspeed == EM_SMARTSPEED_DOWNSHIFT) {
2342                 /* If still no link, perhaps using 2/3 pair cable */
2343                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp);
2344                 phy_tmp |= CR_1000T_MS_ENABLE;
2345                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp);
2346                 if (adapter->hw.mac.autoneg &&
2347                     !e1000_phy_setup_autoneg(&adapter->hw) &&
2348                     !e1000_read_phy_reg(&adapter->hw, PHY_CONTROL, &phy_tmp)) {
2349                         phy_tmp |= MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG;
2350                         e1000_write_phy_reg(&adapter->hw, PHY_CONTROL, phy_tmp);
2351                 }
2352         }
2353
2354         /* Restart process after EM_SMARTSPEED_MAX iterations */
2355         if (adapter->smartspeed++ == EM_SMARTSPEED_MAX)
2356                 adapter->smartspeed = 0;
2357 }
2358
2359 static int
2360 em_dma_malloc(struct adapter *adapter, bus_size_t size,
2361               struct em_dma_alloc *dma)
2362 {
2363         dma->dma_vaddr = bus_dmamem_coherent_any(adapter->parent_dtag,
2364                                 EM_DBA_ALIGN, size, BUS_DMA_WAITOK,
2365                                 &dma->dma_tag, &dma->dma_map,
2366                                 &dma->dma_paddr);
2367         if (dma->dma_vaddr == NULL)
2368                 return ENOMEM;
2369         else
2370                 return 0;
2371 }
2372
2373 static void
2374 em_dma_free(struct adapter *adapter, struct em_dma_alloc *dma)
2375 {
2376         if (dma->dma_tag == NULL)
2377                 return;
2378         bus_dmamap_unload(dma->dma_tag, dma->dma_map);
2379         bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
2380         bus_dma_tag_destroy(dma->dma_tag);
2381 }
2382
2383 static int
2384 em_create_tx_ring(struct adapter *adapter)
2385 {
2386         device_t dev = adapter->dev;
2387         struct em_buffer *tx_buffer;
2388         int error, i;
2389
2390         adapter->tx_buffer_area =
2391                 kmalloc(sizeof(struct em_buffer) * adapter->num_tx_desc,
2392                         M_DEVBUF, M_WAITOK | M_ZERO);
2393
2394         /*
2395          * Create DMA tags for tx buffers
2396          */
2397         error = bus_dma_tag_create(adapter->parent_dtag, /* parent */
2398                         1, 0,                   /* alignment, bounds */
2399                         BUS_SPACE_MAXADDR,      /* lowaddr */
2400                         BUS_SPACE_MAXADDR,      /* highaddr */
2401                         NULL, NULL,             /* filter, filterarg */
2402                         EM_TSO_SIZE,            /* maxsize */
2403                         EM_MAX_SCATTER,         /* nsegments */
2404                         EM_MAX_SEGSIZE,         /* maxsegsize */
2405                         BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
2406                         BUS_DMA_ONEBPAGE,       /* flags */
2407                         &adapter->txtag);
2408         if (error) {
2409                 device_printf(dev, "Unable to allocate TX DMA tag\n");
2410                 kfree(adapter->tx_buffer_area, M_DEVBUF);
2411                 adapter->tx_buffer_area = NULL;
2412                 return error;
2413         }
2414
2415         /*
2416          * Create DMA maps for tx buffers
2417          */
2418         for (i = 0; i < adapter->num_tx_desc; i++) {
2419                 tx_buffer = &adapter->tx_buffer_area[i];
2420
2421                 error = bus_dmamap_create(adapter->txtag,
2422                                           BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
2423                                           &tx_buffer->map);
2424                 if (error) {
2425                         device_printf(dev, "Unable to create TX DMA map\n");
2426                         em_destroy_tx_ring(adapter, i);
2427                         return error;
2428                 }
2429         }
2430         return (0);
2431 }
2432
2433 static void
2434 em_init_tx_ring(struct adapter *adapter)
2435 {
2436         /* Clear the old ring contents */
2437         bzero(adapter->tx_desc_base,
2438             (sizeof(struct e1000_tx_desc)) * adapter->num_tx_desc);
2439
2440         /* Reset state */
2441         adapter->next_avail_tx_desc = 0;
2442         adapter->next_tx_to_clean = 0;
2443         adapter->num_tx_desc_avail = adapter->num_tx_desc;
2444 }
2445
2446 static void
2447 em_init_tx_unit(struct adapter *adapter)
2448 {
2449         uint32_t tctl, tarc, tipg = 0;
2450         uint64_t bus_addr;
2451
2452         /* Setup the Base and Length of the Tx Descriptor Ring */
2453         bus_addr = adapter->txdma.dma_paddr;
2454         E1000_WRITE_REG(&adapter->hw, E1000_TDLEN(0),
2455             adapter->num_tx_desc * sizeof(struct e1000_tx_desc));
2456         E1000_WRITE_REG(&adapter->hw, E1000_TDBAH(0),
2457             (uint32_t)(bus_addr >> 32));
2458         E1000_WRITE_REG(&adapter->hw, E1000_TDBAL(0),
2459             (uint32_t)bus_addr);
2460         /* Setup the HW Tx Head and Tail descriptor pointers */
2461         E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), 0);
2462         E1000_WRITE_REG(&adapter->hw, E1000_TDH(0), 0);
2463
2464         /* Set the default values for the Tx Inter Packet Gap timer */
2465         switch (adapter->hw.mac.type) {
2466         case e1000_82542:
2467                 tipg = DEFAULT_82542_TIPG_IPGT;
2468                 tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2469                 tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2470                 break;
2471
2472         case e1000_80003es2lan:
2473                 tipg = DEFAULT_82543_TIPG_IPGR1;
2474                 tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 <<
2475                     E1000_TIPG_IPGR2_SHIFT;
2476                 break;
2477
2478         default:
2479                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
2480                     adapter->hw.phy.media_type ==
2481                     e1000_media_type_internal_serdes)
2482                         tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
2483                 else
2484                         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
2485                 tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2486                 tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2487                 break;
2488         }
2489
2490         E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg);
2491
2492         /* NOTE: 0 is not allowed for TIDV */
2493         E1000_WRITE_REG(&adapter->hw, E1000_TIDV, 1);
2494         if(adapter->hw.mac.type >= e1000_82540)
2495                 E1000_WRITE_REG(&adapter->hw, E1000_TADV, 0);
2496
2497         if (adapter->hw.mac.type == e1000_82571 ||
2498             adapter->hw.mac.type == e1000_82572) {
2499                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
2500                 tarc |= SPEED_MODE_BIT;
2501                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
2502         } else if (adapter->hw.mac.type == e1000_80003es2lan) {
2503                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
2504                 tarc |= 1;
2505                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
2506                 tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(1));
2507                 tarc |= 1;
2508                 E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc);
2509         }
2510
2511         /* Program the Transmit Control Register */
2512         tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
2513         tctl &= ~E1000_TCTL_CT;
2514         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
2515                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2516
2517         if (adapter->hw.mac.type >= e1000_82571)
2518                 tctl |= E1000_TCTL_MULR;
2519
2520         /* This write will effectively turn on the transmit unit. */
2521         E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
2522 }
2523
2524 static void
2525 em_destroy_tx_ring(struct adapter *adapter, int ndesc)
2526 {
2527         struct em_buffer *tx_buffer;
2528         int i;
2529
2530         if (adapter->tx_buffer_area == NULL)
2531                 return;
2532
2533         for (i = 0; i < ndesc; i++) {
2534                 tx_buffer = &adapter->tx_buffer_area[i];
2535
2536                 KKASSERT(tx_buffer->m_head == NULL);
2537                 bus_dmamap_destroy(adapter->txtag, tx_buffer->map);
2538         }
2539         bus_dma_tag_destroy(adapter->txtag);
2540
2541         kfree(adapter->tx_buffer_area, M_DEVBUF);
2542         adapter->tx_buffer_area = NULL;
2543 }
2544
2545 /*
2546  * The offload context needs to be set when we transfer the first
2547  * packet of a particular protocol (TCP/UDP).  This routine has been
2548  * enhanced to deal with inserted VLAN headers.
2549  *
2550  * If the new packet's ether header length, ip header length and
2551  * csum offloading type are same as the previous packet, we should
2552  * avoid allocating a new csum context descriptor; mainly to take
2553  * advantage of the pipeline effect of the TX data read request.
2554  *
2555  * This function returns number of TX descrptors allocated for
2556  * csum context.
2557  */
2558 static int
2559 em_txcsum(struct adapter *adapter, struct mbuf *mp,
2560           uint32_t *txd_upper, uint32_t *txd_lower)
2561 {
2562         struct e1000_context_desc *TXD;
2563         struct em_buffer *tx_buffer;
2564         struct ether_vlan_header *eh;
2565         struct ip *ip;
2566         int curr_txd, ehdrlen, csum_flags;
2567         uint32_t cmd, hdr_len, ip_hlen;
2568         uint16_t etype;
2569
2570         /*
2571          * Determine where frame payload starts.
2572          * Jump over vlan headers if already present,
2573          * helpful for QinQ too.
2574          */
2575         KASSERT(mp->m_len >= ETHER_HDR_LEN,
2576                 ("em_txcsum_pullup is not called (eh)?\n"));
2577         eh = mtod(mp, struct ether_vlan_header *);
2578         if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2579                 KASSERT(mp->m_len >= ETHER_HDR_LEN + EVL_ENCAPLEN,
2580                         ("em_txcsum_pullup is not called (evh)?\n"));
2581                 etype = ntohs(eh->evl_proto);
2582                 ehdrlen = ETHER_HDR_LEN + EVL_ENCAPLEN;
2583         } else {
2584                 etype = ntohs(eh->evl_encap_proto);
2585                 ehdrlen = ETHER_HDR_LEN;
2586         }
2587
2588         /*
2589          * We only support TCP/UDP for IPv4 for the moment.
2590          * TODO: Support SCTP too when it hits the tree.
2591          */
2592         if (etype != ETHERTYPE_IP)
2593                 return 0;
2594
2595         KASSERT(mp->m_len >= ehdrlen + EM_IPVHL_SIZE,
2596                 ("em_txcsum_pullup is not called (eh+ip_vhl)?\n"));
2597
2598         /* NOTE: We could only safely access ip.ip_vhl part */
2599         ip = (struct ip *)(mp->m_data + ehdrlen);
2600         ip_hlen = ip->ip_hl << 2;
2601
2602         csum_flags = mp->m_pkthdr.csum_flags & EM_CSUM_FEATURES;
2603
2604         if (adapter->csum_ehlen == ehdrlen &&
2605             adapter->csum_iphlen == ip_hlen &&
2606             adapter->csum_flags == csum_flags) {
2607                 /*
2608                  * Same csum offload context as the previous packets;
2609                  * just return.
2610                  */
2611                 *txd_upper = adapter->csum_txd_upper;
2612                 *txd_lower = adapter->csum_txd_lower;
2613                 return 0;
2614         }
2615
2616         /*
2617          * Setup a new csum offload context.
2618          */
2619
2620         curr_txd = adapter->next_avail_tx_desc;
2621         tx_buffer = &adapter->tx_buffer_area[curr_txd];
2622         TXD = (struct e1000_context_desc *)&adapter->tx_desc_base[curr_txd];
2623
2624         cmd = 0;
2625
2626         /* Setup of IP header checksum. */
2627         if (csum_flags & CSUM_IP) {
2628                 /*
2629                  * Start offset for header checksum calculation.
2630                  * End offset for header checksum calculation.
2631                  * Offset of place to put the checksum.
2632                  */
2633                 TXD->lower_setup.ip_fields.ipcss = ehdrlen;
2634                 TXD->lower_setup.ip_fields.ipcse =
2635                     htole16(ehdrlen + ip_hlen - 1);
2636                 TXD->lower_setup.ip_fields.ipcso =
2637                     ehdrlen + offsetof(struct ip, ip_sum);
2638                 cmd |= E1000_TXD_CMD_IP;
2639                 *txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2640         }
2641         hdr_len = ehdrlen + ip_hlen;
2642
2643         if (csum_flags & CSUM_TCP) {
2644                 /*
2645                  * Start offset for payload checksum calculation.
2646                  * End offset for payload checksum calculation.
2647                  * Offset of place to put the checksum.
2648                  */
2649                 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2650                 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2651                 TXD->upper_setup.tcp_fields.tucso =
2652                     hdr_len + offsetof(struct tcphdr, th_sum);
2653                 cmd |= E1000_TXD_CMD_TCP;
2654                 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2655         } else if (csum_flags & CSUM_UDP) {
2656                 /*
2657                  * Start offset for header checksum calculation.
2658                  * End offset for header checksum calculation.
2659                  * Offset of place to put the checksum.
2660                  */
2661                 TXD->upper_setup.tcp_fields.tucss = hdr_len;
2662                 TXD->upper_setup.tcp_fields.tucse = htole16(0);
2663                 TXD->upper_setup.tcp_fields.tucso =
2664                     hdr_len + offsetof(struct udphdr, uh_sum);
2665                 *txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2666         }
2667
2668         *txd_lower = E1000_TXD_CMD_DEXT |       /* Extended descr type */
2669                      E1000_TXD_DTYP_D;          /* Data descr */
2670
2671         /* Save the information for this csum offloading context */
2672         adapter->csum_ehlen = ehdrlen;
2673         adapter->csum_iphlen = ip_hlen;
2674         adapter->csum_flags = csum_flags;
2675         adapter->csum_txd_upper = *txd_upper;
2676         adapter->csum_txd_lower = *txd_lower;
2677
2678         TXD->tcp_seg_setup.data = htole32(0);
2679         TXD->cmd_and_length =
2680             htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd);
2681
2682         if (++curr_txd == adapter->num_tx_desc)
2683                 curr_txd = 0;
2684
2685         KKASSERT(adapter->num_tx_desc_avail > 0);
2686         adapter->num_tx_desc_avail--;
2687
2688         adapter->next_avail_tx_desc = curr_txd;
2689         return 1;
2690 }
2691
2692 static int
2693 em_txcsum_pullup(struct adapter *adapter, struct mbuf **m0)
2694 {
2695         struct mbuf *m = *m0;
2696         struct ether_header *eh;
2697         int len;
2698
2699         adapter->tx_csum_try_pullup++;
2700
2701         len = ETHER_HDR_LEN + EM_IPVHL_SIZE;
2702
2703         if (__predict_false(!M_WRITABLE(m))) {
2704                 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2705                         adapter->tx_csum_drop1++;
2706                         m_freem(m);
2707                         *m0 = NULL;
2708                         return ENOBUFS;
2709                 }
2710                 eh = mtod(m, struct ether_header *);
2711
2712                 if (eh->ether_type == htons(ETHERTYPE_VLAN))
2713                         len += EVL_ENCAPLEN;
2714
2715                 if (m->m_len < len) {
2716                         adapter->tx_csum_drop2++;
2717                         m_freem(m);
2718                         *m0 = NULL;
2719                         return ENOBUFS;
2720                 }
2721                 return 0;
2722         }
2723
2724         if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
2725                 adapter->tx_csum_pullup1++;
2726                 m = m_pullup(m, ETHER_HDR_LEN);
2727                 if (m == NULL) {
2728                         adapter->tx_csum_pullup1_failed++;
2729                         *m0 = NULL;
2730                         return ENOBUFS;
2731                 }
2732                 *m0 = m;
2733         }
2734         eh = mtod(m, struct ether_header *);
2735
2736         if (eh->ether_type == htons(ETHERTYPE_VLAN))
2737                 len += EVL_ENCAPLEN;
2738
2739         if (m->m_len < len) {
2740                 adapter->tx_csum_pullup2++;
2741                 m = m_pullup(m, len);
2742                 if (m == NULL) {
2743                         adapter->tx_csum_pullup2_failed++;
2744                         *m0 = NULL;
2745                         return ENOBUFS;
2746                 }
2747                 *m0 = m;
2748         }
2749         return 0;
2750 }
2751
2752 static void
2753 em_txeof(struct adapter *adapter)
2754 {
2755         struct ifnet *ifp = &adapter->arpcom.ac_if;
2756         struct em_buffer *tx_buffer;
2757         int first, num_avail;
2758
2759         if (adapter->tx_dd_head == adapter->tx_dd_tail)
2760                 return;
2761
2762         if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2763                 return;
2764
2765         num_avail = adapter->num_tx_desc_avail;
2766         first = adapter->next_tx_to_clean;
2767
2768         while (adapter->tx_dd_head != adapter->tx_dd_tail) {
2769                 struct e1000_tx_desc *tx_desc;
2770                 int dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2771
2772                 tx_desc = &adapter->tx_desc_base[dd_idx];
2773                 if (tx_desc->upper.fields.status & E1000_TXD_STAT_DD) {
2774                         EM_INC_TXDD_IDX(adapter->tx_dd_head);
2775
2776                         if (++dd_idx == adapter->num_tx_desc)
2777                                 dd_idx = 0;
2778
2779                         while (first != dd_idx) {
2780                                 logif(pkt_txclean);
2781
2782                                 num_avail++;
2783
2784                                 tx_buffer = &adapter->tx_buffer_area[first];
2785                                 if (tx_buffer->m_head) {
2786                                         ifp->if_opackets++;
2787                                         bus_dmamap_unload(adapter->txtag,
2788                                                           tx_buffer->map);
2789                                         m_freem(tx_buffer->m_head);
2790                                         tx_buffer->m_head = NULL;
2791                                 }
2792
2793                                 if (++first == adapter->num_tx_desc)
2794                                         first = 0;
2795                         }
2796                 } else {
2797                         break;
2798                 }
2799         }
2800         adapter->next_tx_to_clean = first;
2801         adapter->num_tx_desc_avail = num_avail;
2802
2803         if (adapter->tx_dd_head == adapter->tx_dd_tail) {
2804                 adapter->tx_dd_head = 0;
2805                 adapter->tx_dd_tail = 0;
2806         }
2807
2808         if (!EM_IS_OACTIVE(adapter)) {
2809                 ifp->if_flags &= ~IFF_OACTIVE;
2810
2811                 /* All clean, turn off the timer */
2812                 if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2813                         ifp->if_timer = 0;
2814         }
2815 }
2816
2817 static void
2818 em_tx_collect(struct adapter *adapter)
2819 {
2820         struct ifnet *ifp = &adapter->arpcom.ac_if;
2821         struct em_buffer *tx_buffer;
2822         int tdh, first, num_avail, dd_idx = -1;
2823
2824         if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2825                 return;
2826
2827         tdh = E1000_READ_REG(&adapter->hw, E1000_TDH(0));
2828         if (tdh == adapter->next_tx_to_clean)
2829                 return;
2830
2831         if (adapter->tx_dd_head != adapter->tx_dd_tail)
2832                 dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2833
2834         num_avail = adapter->num_tx_desc_avail;
2835         first = adapter->next_tx_to_clean;
2836
2837         while (first != tdh) {
2838                 logif(pkt_txclean);
2839
2840                 num_avail++;
2841
2842                 tx_buffer = &adapter->tx_buffer_area[first];
2843                 if (tx_buffer->m_head) {
2844                         ifp->if_opackets++;
2845                         bus_dmamap_unload(adapter->txtag,
2846                                           tx_buffer->map);
2847                         m_freem(tx_buffer->m_head);
2848                         tx_buffer->m_head = NULL;
2849                 }
2850
2851                 if (first == dd_idx) {
2852                         EM_INC_TXDD_IDX(adapter->tx_dd_head);
2853                         if (adapter->tx_dd_head == adapter->tx_dd_tail) {
2854                                 adapter->tx_dd_head = 0;
2855                                 adapter->tx_dd_tail = 0;
2856                                 dd_idx = -1;
2857                         } else {
2858                                 dd_idx = adapter->tx_dd[adapter->tx_dd_head];
2859                         }
2860                 }
2861
2862                 if (++first == adapter->num_tx_desc)
2863                         first = 0;
2864         }
2865         adapter->next_tx_to_clean = first;
2866         adapter->num_tx_desc_avail = num_avail;
2867
2868         if (!EM_IS_OACTIVE(adapter)) {
2869                 ifp->if_flags &= ~IFF_OACTIVE;
2870
2871                 /* All clean, turn off the timer */
2872                 if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2873                         ifp->if_timer = 0;
2874         }
2875 }
2876
2877 /*
2878  * When Link is lost sometimes there is work still in the TX ring
2879  * which will result in a watchdog, rather than allow that do an
2880  * attempted cleanup and then reinit here.  Note that this has been
2881  * seens mostly with fiber adapters.
2882  */
2883 static void
2884 em_tx_purge(struct adapter *adapter)
2885 {
2886         struct ifnet *ifp = &adapter->arpcom.ac_if;
2887
2888         if (!adapter->link_active && ifp->if_timer) {
2889                 em_tx_collect(adapter);
2890                 if (ifp->if_timer) {
2891                         if_printf(ifp, "Link lost, TX pending, reinit\n");
2892                         ifp->if_timer = 0;
2893                         em_init(adapter);
2894                 }
2895         }
2896 }
2897
2898 static int
2899 em_newbuf(struct adapter *adapter, int i, int init)
2900 {
2901         struct mbuf *m;
2902         bus_dma_segment_t seg;
2903         bus_dmamap_t map;
2904         struct em_buffer *rx_buffer;
2905         int error, nseg;
2906
2907         m = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2908         if (m == NULL) {
2909                 adapter->mbuf_cluster_failed++;
2910                 if (init) {
2911                         if_printf(&adapter->arpcom.ac_if,
2912                                   "Unable to allocate RX mbuf\n");
2913                 }
2914                 return (ENOBUFS);
2915         }
2916         m->m_len = m->m_pkthdr.len = MCLBYTES;
2917
2918         if (adapter->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2919                 m_adj(m, ETHER_ALIGN);
2920
2921         error = bus_dmamap_load_mbuf_segment(adapter->rxtag,
2922                         adapter->rx_sparemap, m,
2923                         &seg, 1, &nseg, BUS_DMA_NOWAIT);
2924         if (error) {
2925                 m_freem(m);
2926                 if (init) {
2927                         if_printf(&adapter->arpcom.ac_if,
2928                                   "Unable to load RX mbuf\n");
2929                 }
2930                 return (error);
2931         }
2932
2933         rx_buffer = &adapter->rx_buffer_area[i];
2934         if (rx_buffer->m_head != NULL)
2935                 bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
2936
2937         map = rx_buffer->map;
2938         rx_buffer->map = adapter->rx_sparemap;
2939         adapter->rx_sparemap = map;
2940
2941         rx_buffer->m_head = m;
2942
2943         adapter->rx_desc_base[i].buffer_addr = htole64(seg.ds_addr);
2944         return (0);
2945 }
2946
2947 static int
2948 em_create_rx_ring(struct adapter *adapter)
2949 {
2950         device_t dev = adapter->dev;
2951         struct em_buffer *rx_buffer;
2952         int i, error;
2953
2954         adapter->rx_buffer_area =
2955                 kmalloc(sizeof(struct em_buffer) * adapter->num_rx_desc,
2956                         M_DEVBUF, M_WAITOK | M_ZERO);
2957
2958         /*
2959          * Create DMA tag for rx buffers
2960          */
2961         error = bus_dma_tag_create(adapter->parent_dtag, /* parent */
2962                         1, 0,                   /* alignment, bounds */
2963                         BUS_SPACE_MAXADDR,      /* lowaddr */
2964                         BUS_SPACE_MAXADDR,      /* highaddr */
2965                         NULL, NULL,             /* filter, filterarg */
2966                         MCLBYTES,               /* maxsize */
2967                         1,                      /* nsegments */
2968                         MCLBYTES,               /* maxsegsize */
2969                         BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2970                         &adapter->rxtag);
2971         if (error) {
2972                 device_printf(dev, "Unable to allocate RX DMA tag\n");
2973                 kfree(adapter->rx_buffer_area, M_DEVBUF);
2974                 adapter->rx_buffer_area = NULL;
2975                 return error;
2976         }
2977
2978         /*
2979          * Create spare DMA map for rx buffers
2980          */
2981         error = bus_dmamap_create(adapter->rxtag, BUS_DMA_WAITOK,
2982                                   &adapter->rx_sparemap);
2983         if (error) {
2984                 device_printf(dev, "Unable to create spare RX DMA map\n");
2985                 bus_dma_tag_destroy(adapter->rxtag);
2986                 kfree(adapter->rx_buffer_area, M_DEVBUF);
2987                 adapter->rx_buffer_area = NULL;
2988                 return error;
2989         }
2990
2991         /*
2992          * Create DMA maps for rx buffers
2993          */
2994         for (i = 0; i < adapter->num_rx_desc; i++) {
2995                 rx_buffer = &adapter->rx_buffer_area[i];
2996
2997                 error = bus_dmamap_create(adapter->rxtag, BUS_DMA_WAITOK,
2998                                           &rx_buffer->map);
2999                 if (error) {
3000                         device_printf(dev, "Unable to create RX DMA map\n");
3001                         em_destroy_rx_ring(adapter, i);
3002                         return error;
3003                 }
3004         }
3005         return (0);
3006 }
3007
3008 static int
3009 em_init_rx_ring(struct adapter *adapter)
3010 {
3011         int i, error;
3012
3013         /* Reset descriptor ring */
3014         bzero(adapter->rx_desc_base,
3015             (sizeof(struct e1000_rx_desc)) * adapter->num_rx_desc);
3016
3017         /* Allocate new ones. */
3018         for (i = 0; i < adapter->num_rx_desc; i++) {
3019                 error = em_newbuf(adapter, i, 1);
3020                 if (error)
3021                         return (error);
3022         }
3023
3024         /* Setup our descriptor pointers */
3025         adapter->next_rx_desc_to_check = 0;
3026
3027         return (0);
3028 }
3029
3030 static void
3031 em_init_rx_unit(struct adapter *adapter)
3032 {
3033         struct ifnet *ifp = &adapter->arpcom.ac_if;
3034         uint64_t bus_addr;
3035         uint32_t rctl, rxcsum;
3036
3037         /*
3038          * Make sure receives are disabled while setting
3039          * up the descriptor ring
3040          */
3041         rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
3042         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
3043
3044         if (adapter->hw.mac.type >= e1000_82540) {
3045                 /*
3046                  * Set the interrupt throttling rate. Value is calculated
3047                  * as ITR = 1 / (INT_THROTTLE_CEIL * 256ns)
3048                  */
3049                 if (adapter->int_throttle_ceil) {
3050                         E1000_WRITE_REG(&adapter->hw, E1000_ITR,
3051                                 1000000000 / 256 / adapter->int_throttle_ceil);
3052                 } else {
3053                         E1000_WRITE_REG(&adapter->hw, E1000_ITR, 0);
3054                 }
3055         }
3056
3057         /* Disable accelerated ackknowledge */
3058         if (adapter->hw.mac.type == e1000_82574) {
3059                 E1000_WRITE_REG(&adapter->hw,
3060                     E1000_RFCTL, E1000_RFCTL_ACK_DIS);
3061         }
3062
3063         /* Setup the Base and Length of the Rx Descriptor Ring */
3064         bus_addr = adapter->rxdma.dma_paddr;
3065         E1000_WRITE_REG(&adapter->hw, E1000_RDLEN(0),
3066             adapter->num_rx_desc * sizeof(struct e1000_rx_desc));
3067         E1000_WRITE_REG(&adapter->hw, E1000_RDBAH(0),
3068             (uint32_t)(bus_addr >> 32));
3069         E1000_WRITE_REG(&adapter->hw, E1000_RDBAL(0),
3070             (uint32_t)bus_addr);
3071
3072         /* Setup the Receive Control Register */
3073         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3074         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
3075                 E1000_RCTL_RDMTS_HALF |
3076                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3077
3078         /* Make sure VLAN Filters are off */
3079         rctl &= ~E1000_RCTL_VFE;
3080
3081         if (e1000_tbi_sbp_enabled_82543(&adapter->hw))
3082                 rctl |= E1000_RCTL_SBP;
3083         else
3084                 rctl &= ~E1000_RCTL_SBP;
3085
3086         switch (adapter->rx_buffer_len) {
3087         default:
3088         case 2048:
3089                 rctl |= E1000_RCTL_SZ_2048;
3090                 break;
3091
3092         case 4096:
3093                 rctl |= E1000_RCTL_SZ_4096 |
3094                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3095                 break;
3096
3097         case 8192:
3098                 rctl |= E1000_RCTL_SZ_8192 |
3099                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3100                 break;
3101
3102         case 16384:
3103                 rctl |= E1000_RCTL_SZ_16384 |
3104                     E1000_RCTL_BSEX | E1000_RCTL_LPE;
3105                 break;
3106         }
3107
3108         if (ifp->if_mtu > ETHERMTU)
3109                 rctl |= E1000_RCTL_LPE;
3110         else
3111                 rctl &= ~E1000_RCTL_LPE;
3112
3113         /* Receive Checksum Offload for TCP and UDP */
3114         if (ifp->if_capenable & IFCAP_RXCSUM) {
3115                 rxcsum = E1000_READ_REG(&adapter->hw, E1000_RXCSUM);
3116                 rxcsum |= (E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL);
3117                 E1000_WRITE_REG(&adapter->hw, E1000_RXCSUM, rxcsum);
3118         }
3119
3120         /*
3121          * XXX TEMPORARY WORKAROUND: on some systems with 82573
3122          * long latencies are observed, like Lenovo X60. This
3123          * change eliminates the problem, but since having positive
3124          * values in RDTR is a known source of problems on other
3125          * platforms another solution is being sought.
3126          */
3127         if (em_82573_workaround && adapter->hw.mac.type == e1000_82573) {
3128                 E1000_WRITE_REG(&adapter->hw, E1000_RADV, EM_RADV_82573);
3129                 E1000_WRITE_REG(&adapter->hw, E1000_RDTR, EM_RDTR_82573);
3130         }
3131
3132         /*
3133          * Setup the HW Rx Head and Tail Descriptor Pointers
3134          */
3135         E1000_WRITE_REG(&adapter->hw, E1000_RDH(0), 0);
3136         E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), adapter->num_rx_desc - 1);
3137
3138         /* Enable Receives */
3139         E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl);
3140 }
3141
3142 static void
3143 em_destroy_rx_ring(struct adapter *adapter, int ndesc)
3144 {
3145         struct em_buffer *rx_buffer;
3146         int i;
3147
3148         if (adapter->rx_buffer_area == NULL)
3149                 return;
3150
3151         for (i = 0; i < ndesc; i++) {
3152                 rx_buffer = &adapter->rx_buffer_area[i];
3153
3154                 KKASSERT(rx_buffer->m_head == NULL);
3155                 bus_dmamap_destroy(adapter->rxtag, rx_buffer->map);
3156         }
3157         bus_dmamap_destroy(adapter->rxtag, adapter->rx_sparemap);
3158         bus_dma_tag_destroy(adapter->rxtag);
3159
3160         kfree(adapter->rx_buffer_area, M_DEVBUF);
3161         adapter->rx_buffer_area = NULL;
3162 }
3163
3164 static void
3165 em_rxeof(struct adapter *adapter, int count)
3166 {
3167         struct ifnet *ifp = &adapter->arpcom.ac_if;
3168         uint8_t status, accept_frame = 0, eop = 0;
3169         uint16_t len, desc_len, prev_len_adj;
3170         struct e1000_rx_desc *current_desc;
3171         struct mbuf *mp;
3172         int i;
3173         struct mbuf_chain chain[MAXCPU];
3174
3175         i = adapter->next_rx_desc_to_check;
3176         current_desc = &adapter->rx_desc_base[i];
3177
3178         if (!(current_desc->status & E1000_RXD_STAT_DD))
3179                 return;
3180
3181         ether_input_chain_init(chain);
3182
3183         while ((current_desc->status & E1000_RXD_STAT_DD) && count != 0) {
3184                 struct mbuf *m = NULL;
3185
3186                 logif(pkt_receive);
3187
3188                 mp = adapter->rx_buffer_area[i].m_head;
3189
3190                 /*
3191                  * Can't defer bus_dmamap_sync(9) because TBI_ACCEPT
3192                  * needs to access the last received byte in the mbuf.
3193                  */
3194                 bus_dmamap_sync(adapter->rxtag, adapter->rx_buffer_area[i].map,
3195                                 BUS_DMASYNC_POSTREAD);
3196
3197                 accept_frame = 1;
3198                 prev_len_adj = 0;
3199                 desc_len = le16toh(current_desc->length);
3200                 status = current_desc->status;
3201                 if (status & E1000_RXD_STAT_EOP) {
3202                         count--;
3203                         eop = 1;
3204                         if (desc_len < ETHER_CRC_LEN) {
3205                                 len = 0;
3206                                 prev_len_adj = ETHER_CRC_LEN - desc_len;
3207                         } else {
3208                                 len = desc_len - ETHER_CRC_LEN;
3209                         }
3210                 } else {
3211                         eop = 0;
3212                         len = desc_len;
3213                 }
3214
3215                 if (current_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
3216                         uint8_t last_byte;
3217                         uint32_t pkt_len = desc_len;
3218
3219                         if (adapter->fmp != NULL)
3220                                 pkt_len += adapter->fmp->m_pkthdr.len;
3221
3222                         last_byte = *(mtod(mp, caddr_t) + desc_len - 1);
3223                         if (TBI_ACCEPT(&adapter->hw, status,
3224                             current_desc->errors, pkt_len, last_byte,
3225                             adapter->min_frame_size, adapter->max_frame_size)) {
3226                                 e1000_tbi_adjust_stats_82543(&adapter->hw,
3227                                     &adapter->stats, pkt_len,
3228                                     adapter->hw.mac.addr,
3229                                     adapter->max_frame_size);
3230                                 if (len > 0)
3231                                         len--;
3232                         } else {
3233                                 accept_frame = 0;
3234                         }
3235                 }
3236
3237                 if (accept_frame) {
3238                         if (em_newbuf(adapter, i, 0) != 0) {
3239                                 ifp->if_iqdrops++;
3240                                 goto discard;
3241                         }
3242
3243                         /* Assign correct length to the current fragment */
3244                         mp->m_len = len;
3245
3246                         if (adapter->fmp == NULL) {
3247                                 mp->m_pkthdr.len = len;
3248                                 adapter->fmp = mp; /* Store the first mbuf */
3249                                 adapter->lmp = mp;
3250                         } else {
3251                                 /*
3252                                  * Chain mbuf's together
3253                                  */
3254
3255                                 /*
3256                                  * Adjust length of previous mbuf in chain if
3257                                  * we received less than 4 bytes in the last
3258                                  * descriptor.
3259                                  */
3260                                 if (prev_len_adj > 0) {
3261                                         adapter->lmp->m_len -= prev_len_adj;
3262                                         adapter->fmp->m_pkthdr.len -=
3263                                             prev_len_adj;
3264                                 }
3265                                 adapter->lmp->m_next = mp;
3266                                 adapter->lmp = adapter->lmp->m_next;
3267                                 adapter->fmp->m_pkthdr.len += len;
3268                         }
3269
3270                         if (eop) {
3271                                 adapter->fmp->m_pkthdr.rcvif = ifp;
3272                                 ifp->if_ipackets++;
3273
3274                                 if (ifp->if_capenable & IFCAP_RXCSUM) {
3275                                         em_rxcsum(adapter, current_desc,
3276                                                   adapter->fmp);
3277                                 }
3278
3279                                 if (status & E1000_RXD_STAT_VP) {
3280                                         adapter->fmp->m_pkthdr.ether_vlantag =
3281                                             (le16toh(current_desc->special) &
3282                                             E1000_RXD_SPC_VLAN_MASK);
3283                                         adapter->fmp->m_flags |= M_VLANTAG;
3284                                 }
3285                                 m = adapter->fmp;
3286                                 adapter->fmp = NULL;
3287                                 adapter->lmp = NULL;
3288                         }
3289                 } else {
3290                         ifp->if_ierrors++;
3291 discard:
3292 #ifdef foo
3293                         /* Reuse loaded DMA map and just update mbuf chain */
3294                         mp = adapter->rx_buffer_area[i].m_head;
3295                         mp->m_len = mp->m_pkthdr.len = MCLBYTES;
3296                         mp->m_data = mp->m_ext.ext_buf;
3297                         mp->m_next = NULL;
3298                         if (adapter->max_frame_size <= (MCLBYTES - ETHER_ALIGN))
3299                                 m_adj(mp, ETHER_ALIGN);
3300 #endif
3301                         if (adapter->fmp != NULL) {
3302                                 m_freem(adapter->fmp);
3303                                 adapter->fmp = NULL;
3304                                 adapter->lmp = NULL;
3305                         }
3306                         m = NULL;
3307                 }
3308
3309                 /* Zero out the receive descriptors status. */
3310                 current_desc->status = 0;
3311
3312                 if (m != NULL)
3313                         ether_input_chain(ifp, m, NULL, chain);
3314
3315                 /* Advance our pointers to the next descriptor. */
3316                 if (++i == adapter->num_rx_desc)
3317                         i = 0;
3318                 current_desc = &adapter->rx_desc_base[i];
3319         }
3320         adapter->next_rx_desc_to_check = i;
3321
3322         ether_input_dispatch(chain);
3323
3324         /* Advance the E1000's Receive Queue #0  "Tail Pointer". */
3325         if (--i < 0)
3326                 i = adapter->num_rx_desc - 1;
3327         E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), i);
3328 }
3329
3330 static void
3331 em_rxcsum(struct adapter *adapter, struct e1000_rx_desc *rx_desc,
3332           struct mbuf *mp)
3333 {
3334         /* 82543 or newer only */
3335         if (adapter->hw.mac.type < e1000_82543 ||
3336             /* Ignore Checksum bit is set */
3337             (rx_desc->status & E1000_RXD_STAT_IXSM))
3338                 return;
3339
3340         if ((rx_desc->status & E1000_RXD_STAT_IPCS) &&
3341             !(rx_desc->errors & E1000_RXD_ERR_IPE)) {
3342                 /* IP Checksum Good */
3343                 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
3344         }
3345
3346         if ((rx_desc->status & E1000_RXD_STAT_TCPCS) &&
3347             !(rx_desc->errors & E1000_RXD_ERR_TCPE)) {
3348                 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
3349                                            CSUM_PSEUDO_HDR |
3350                                            CSUM_FRAG_NOT_CHECKED;
3351                 mp->m_pkthdr.csum_data = htons(0xffff);
3352         }
3353 }
3354
3355 static void
3356 em_enable_intr(struct adapter *adapter)
3357 {
3358         lwkt_serialize_handler_enable(adapter->arpcom.ac_if.if_serializer);
3359         E1000_WRITE_REG(&adapter->hw, E1000_IMS, IMS_ENABLE_MASK);
3360 }
3361
3362 static void
3363 em_disable_intr(struct adapter *adapter)
3364 {
3365         uint32_t clear = 0xffffffff;
3366
3367         /*
3368          * The first version of 82542 had an errata where when link was forced
3369          * it would stay up even up even if the cable was disconnected.
3370          * Sequence errors were used to detect the disconnect and then the
3371          * driver would unforce the link.  This code in the in the ISR.  For
3372          * this to work correctly the Sequence error interrupt had to be
3373          * enabled all the time.
3374          */
3375         if (adapter->hw.mac.type == e1000_82542 &&
3376             adapter->hw.revision_id == E1000_REVISION_2)
3377                 clear &= ~E1000_IMC_RXSEQ;
3378
3379         E1000_WRITE_REG(&adapter->hw, E1000_IMC, clear);
3380
3381         lwkt_serialize_handler_disable(adapter->arpcom.ac_if.if_serializer);
3382 }
3383
3384 /*
3385  * Bit of a misnomer, what this really means is
3386  * to enable OS management of the system... aka
3387  * to disable special hardware management features
3388  */
3389 static void
3390 em_get_mgmt(struct adapter *adapter)
3391 {
3392         /* A shared code workaround */
3393 #define E1000_82542_MANC2H E1000_MANC2H
3394         if (adapter->has_manage) {
3395                 int manc2h = E1000_READ_REG(&adapter->hw, E1000_MANC2H);
3396                 int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
3397
3398                 /* disable hardware interception of ARP */
3399                 manc &= ~(E1000_MANC_ARP_EN);
3400
3401                 /* enable receiving management packets to the host */
3402                 if (adapter->hw.mac.type >= e1000_82571) {
3403                         manc |= E1000_MANC_EN_MNG2HOST;
3404 #define E1000_MNG2HOST_PORT_623 (1 << 5)
3405 #define E1000_MNG2HOST_PORT_664 (1 << 6)
3406                         manc2h |= E1000_MNG2HOST_PORT_623;
3407                         manc2h |= E1000_MNG2HOST_PORT_664;
3408                         E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h);
3409                 }
3410
3411                 E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
3412         }
3413 }
3414
3415 /*
3416  * Give control back to hardware management
3417  * controller if there is one.
3418  */
3419 static void
3420 em_rel_mgmt(struct adapter *adapter)
3421 {
3422         if (adapter->has_manage) {
3423                 int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
3424
3425                 /* re-enable hardware interception of ARP */
3426                 manc |= E1000_MANC_ARP_EN;
3427
3428                 if (adapter->hw.mac.type >= e1000_82571)
3429                         manc &= ~E1000_MANC_EN_MNG2HOST;
3430
3431                 E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
3432         }
3433 }
3434
3435 /*
3436  * em_get_hw_control() sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3437  * For ASF and Pass Through versions of f/w this means that
3438  * the driver is loaded.  For AMT version (only with 82573)
3439  * of the f/w this means that the network i/f is open.
3440  */
3441 static void
3442 em_get_hw_control(struct adapter *adapter)
3443 {
3444         uint32_t ctrl_ext, swsm;
3445
3446         /* Let firmware know the driver has taken over */
3447         switch (adapter->hw.mac.type) {
3448         case e1000_82573:
3449                 swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
3450                 E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
3451                     swsm | E1000_SWSM_DRV_LOAD);
3452                 break;
3453         case e1000_82571:
3454         case e1000_82572:
3455         case e1000_80003es2lan:
3456         case e1000_ich8lan:
3457         case e1000_ich9lan:
3458         case e1000_ich10lan:
3459                 ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
3460                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
3461                     ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
3462                 break;
3463         default:
3464                 break;
3465         }
3466 }
3467
3468 /*
3469  * em_rel_hw_control() resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
3470  * For ASF and Pass Through versions of f/w this means that the
3471  * driver is no longer loaded.  For AMT version (only with 82573)
3472  * of the f/w this means that the network i/f is closed.
3473  */
3474 static void
3475 em_rel_hw_control(struct adapter *adapter)
3476 {
3477         uint32_t ctrl_ext, swsm;
3478
3479         /* Let firmware taken over control of h/w */
3480         switch (adapter->hw.mac.type) {
3481         case e1000_82573:
3482                 swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
3483                 E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
3484                     swsm & ~E1000_SWSM_DRV_LOAD);
3485                 break;
3486
3487         case e1000_82571:
3488         case e1000_82572:
3489         case e1000_80003es2lan:
3490         case e1000_ich8lan:
3491         case e1000_ich9lan:
3492         case e1000_ich10lan:
3493                 ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
3494                 E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
3495                     ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
3496                 break;
3497
3498         default:
3499                 break;
3500         }
3501 }
3502
3503 static int
3504 em_is_valid_eaddr(const uint8_t *addr)
3505 {
3506         char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
3507
3508         if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3509                 return (FALSE);
3510
3511         return (TRUE);
3512 }
3513
3514 /*
3515  * Enable PCI Wake On Lan capability
3516  */
3517 void
3518 em_enable_wol(device_t dev)
3519 {
3520         uint16_t cap, status;
3521         uint8_t id;
3522
3523         /* First find the capabilities pointer*/
3524         cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
3525
3526         /* Read the PM Capabilities */
3527         id = pci_read_config(dev, cap, 1);
3528         if (id != PCIY_PMG)     /* Something wrong */
3529                 return;
3530
3531         /*
3532          * OK, we have the power capabilities,
3533          * so now get the status register
3534          */
3535         cap += PCIR_POWER_STATUS;
3536         status = pci_read_config(dev, cap, 2);
3537         status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
3538         pci_write_config(dev, cap, status, 2);
3539 }
3540
3541
3542 /*
3543  * 82544 Coexistence issue workaround.
3544  *    There are 2 issues.
3545  *       1. Transmit Hang issue.
3546  *    To detect this issue, following equation can be used...
3547  *        SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3548  *        If SUM[3:0] is in between 1 to 4, we will have this issue.
3549  *
3550  *       2. DAC issue.
3551  *    To detect this issue, following equation can be used...
3552  *        SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3553  *        If SUM[3:0] is in between 9 to c, we will have this issue.
3554  *
3555  *    WORKAROUND:
3556  *        Make sure we do not have ending address
3557  *        as 1,2,3,4(Hang) or 9,a,b,c (DAC)
3558  */
3559 static uint32_t
3560 em_82544_fill_desc(bus_addr_t address, uint32_t length, PDESC_ARRAY desc_array)
3561 {
3562         uint32_t safe_terminator;
3563
3564         /*
3565          * Since issue is sensitive to length and address.
3566          * Let us first check the address...
3567          */
3568         if (length <= 4) {
3569                 desc_array->descriptor[0].address = address;
3570                 desc_array->descriptor[0].length = length;
3571                 desc_array->elements = 1;
3572                 return (desc_array->elements);
3573         }
3574
3575         safe_terminator =
3576         (uint32_t)((((uint32_t)address & 0x7) + (length & 0xF)) & 0xF);
3577
3578         /* If it does not fall between 0x1 to 0x4 and 0x9 to 0xC then return */
3579         if (safe_terminator == 0 ||
3580             (safe_terminator > 4 && safe_terminator < 9) ||
3581             (safe_terminator > 0xC && safe_terminator <= 0xF)) {
3582                 desc_array->descriptor[0].address = address;
3583                 desc_array->descriptor[0].length = length;
3584                 desc_array->elements = 1;
3585                 return (desc_array->elements);
3586         }
3587
3588         desc_array->descriptor[0].address = address;
3589         desc_array->descriptor[0].length = length - 4;
3590         desc_array->descriptor[1].address = address + (length - 4);
3591         desc_array->descriptor[1].length = 4;
3592         desc_array->elements = 2;
3593         return (desc_array->elements);
3594 }
3595
3596 static void
3597 em_update_stats(struct adapter *adapter)
3598 {
3599         struct ifnet *ifp = &adapter->arpcom.ac_if;
3600
3601         if (adapter->hw.phy.media_type == e1000_media_type_copper ||
3602             (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) {
3603                 adapter->stats.symerrs +=
3604                         E1000_READ_REG(&adapter->hw, E1000_SYMERRS);
3605                 adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC);
3606         }
3607         adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS);
3608         adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC);
3609         adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC);
3610         adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL);
3611
3612         adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC);
3613         adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL);
3614         adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC);
3615         adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC);
3616         adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC);
3617         adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC);
3618         adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC);
3619         adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, E1000_XOFFRXC);
3620         adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC);
3621         adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC);
3622         adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64);
3623         adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127);
3624         adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255);
3625         adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511);
3626         adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023);
3627         adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522);
3628         adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC);
3629         adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC);
3630         adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC);
3631         adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC);
3632
3633         /* For the 64-bit byte counters the low dword must be read first. */
3634         /* Both registers clear on the read of the high dword */
3635
3636         adapter->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCH);
3637         adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCH);
3638
3639         adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC);
3640         adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC);
3641         adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC);
3642         adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC);
3643         adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC);
3644
3645         adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH);
3646         adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH);
3647
3648         adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR);
3649         adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT);
3650         adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64);
3651         adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127);
3652         adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255);
3653         adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511);
3654         adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023);
3655         adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522);
3656         adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC);
3657         adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC);
3658
3659         if (adapter->hw.mac.type >= e1000_82543) {
3660                 adapter->stats.algnerrc +=
3661                 E1000_READ_REG(&adapter->hw, E1000_ALGNERRC);
3662                 adapter->stats.rxerrc +=
3663                 E1000_READ_REG(&adapter->hw, E1000_RXERRC);
3664                 adapter->stats.tncrs +=
3665                 E1000_READ_REG(&adapter->hw, E1000_TNCRS);
3666                 adapter->stats.cexterr +=
3667                 E1000_READ_REG(&adapter->hw, E1000_CEXTERR);
3668                 adapter->stats.tsctc +=
3669                 E1000_READ_REG(&adapter->hw, E1000_TSCTC);
3670                 adapter->stats.tsctfc +=
3671                 E1000_READ_REG(&adapter->hw, E1000_TSCTFC);
3672         }
3673
3674         ifp->if_collisions = adapter->stats.colc;
3675
3676         /* Rx Errors */
3677         ifp->if_ierrors =
3678             adapter->dropped_pkts + adapter->stats.rxerrc +
3679             adapter->stats.crcerrs + adapter->stats.algnerrc +
3680             adapter->stats.ruc + adapter->stats.roc +
3681             adapter->stats.mpc + adapter->stats.cexterr;
3682
3683         /* Tx Errors */
3684         ifp->if_oerrors =
3685             adapter->stats.ecol + adapter->stats.latecol +
3686             adapter->watchdog_events;
3687 }
3688
3689 static void
3690 em_print_debug_info(struct adapter *adapter)
3691 {
3692         device_t dev = adapter->dev;
3693         uint8_t *hw_addr = adapter->hw.hw_addr;
3694
3695         device_printf(dev, "Adapter hardware address = %p \n", hw_addr);
3696         device_printf(dev, "CTRL = 0x%x RCTL = 0x%x \n",
3697             E1000_READ_REG(&adapter->hw, E1000_CTRL),
3698             E1000_READ_REG(&adapter->hw, E1000_RCTL));
3699         device_printf(dev, "Packet buffer = Tx=%dk Rx=%dk \n",
3700             ((E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff0000) >> 16),\
3701             (E1000_READ_REG(&adapter->hw, E1000_PBA) & 0xffff) );
3702         device_printf(dev, "Flow control watermarks high = %d low = %d\n",
3703             adapter->hw.fc.high_water,
3704             adapter->hw.fc.low_water);
3705         device_printf(dev, "tx_int_delay = %d, tx_abs_int_delay = %d\n",
3706             E1000_READ_REG(&adapter->hw, E1000_TIDV),
3707             E1000_READ_REG(&adapter->hw, E1000_TADV));
3708         device_printf(dev, "rx_int_delay = %d, rx_abs_int_delay = %d\n",
3709             E1000_READ_REG(&adapter->hw, E1000_RDTR),
3710             E1000_READ_REG(&adapter->hw, E1000_RADV));
3711         device_printf(dev, "fifo workaround = %lld, fifo_reset_count = %lld\n",
3712             (long long)adapter->tx_fifo_wrk_cnt,
3713             (long long)adapter->tx_fifo_reset_cnt);
3714         device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
3715             E1000_READ_REG(&adapter->hw, E1000_TDH(0)),
3716             E1000_READ_REG(&adapter->hw, E1000_TDT(0)));
3717         device_printf(dev, "hw rdh = %d, hw rdt = %d\n",
3718             E1000_READ_REG(&adapter->hw, E1000_RDH(0)),
3719             E1000_READ_REG(&adapter->hw, E1000_RDT(0)));
3720         device_printf(dev, "Num Tx descriptors avail = %d\n",
3721             adapter->num_tx_desc_avail);
3722         device_printf(dev, "Tx Descriptors not avail1 = %ld\n",
3723             adapter->no_tx_desc_avail1);
3724         device_printf(dev, "Tx Descriptors not avail2 = %ld\n",
3725             adapter->no_tx_desc_avail2);
3726         device_printf(dev, "Std mbuf failed = %ld\n",
3727             adapter->mbuf_alloc_failed);
3728         device_printf(dev, "Std mbuf cluster failed = %ld\n",
3729             adapter->mbuf_cluster_failed);
3730         device_printf(dev, "Driver dropped packets = %ld\n",
3731             adapter->dropped_pkts);
3732         device_printf(dev, "Driver tx dma failure in encap = %ld\n",
3733             adapter->no_tx_dma_setup);
3734
3735         device_printf(dev, "TXCSUM try pullup = %lu\n",
3736             adapter->tx_csum_try_pullup);
3737         device_printf(dev, "TXCSUM m_pullup(eh) called = %lu\n",
3738             adapter->tx_csum_pullup1);
3739         device_printf(dev, "TXCSUM m_pullup(eh) failed = %lu\n",
3740             adapter->tx_csum_pullup1_failed);
3741         device_printf(dev, "TXCSUM m_pullup(eh+ip) called = %lu\n",
3742             adapter->tx_csum_pullup2);
3743         device_printf(dev, "TXCSUM m_pullup(eh+ip) failed = %lu\n",
3744             adapter->tx_csum_pullup2_failed);
3745         device_printf(dev, "TXCSUM non-writable(eh) droped = %lu\n",
3746             adapter->tx_csum_drop1);
3747         device_printf(dev, "TXCSUM non-writable(eh+ip) droped = %lu\n",
3748             adapter->tx_csum_drop2);
3749 }
3750
3751 static void
3752 em_print_hw_stats(struct adapter *adapter)
3753 {
3754         device_t dev = adapter->dev;
3755
3756         device_printf(dev, "Excessive collisions = %lld\n",
3757             (long long)adapter->stats.ecol);
3758 #if (DEBUG_HW > 0)  /* Dont output these errors normally */
3759         device_printf(dev, "Symbol errors = %lld\n",
3760             (long long)adapter->stats.symerrs);
3761 #endif
3762         device_printf(dev, "Sequence errors = %lld\n",
3763             (long long)adapter->stats.sec);
3764         device_printf(dev, "Defer count = %lld\n",
3765             (long long)adapter->stats.dc);
3766         device_printf(dev, "Missed Packets = %lld\n",
3767             (long long)adapter->stats.mpc);
3768         device_printf(dev, "Receive No Buffers = %lld\n",
3769             (long long)adapter->stats.rnbc);
3770         /* RLEC is inaccurate on some hardware, calculate our own. */
3771         device_printf(dev, "Receive Length Errors = %lld\n",
3772             ((long long)adapter->stats.roc + (long long)adapter->stats.ruc));
3773         device_printf(dev, "Receive errors = %lld\n",
3774             (long long)adapter->stats.rxerrc);
3775         device_printf(dev, "Crc errors = %lld\n",
3776             (long long)adapter->stats.crcerrs);
3777         device_printf(dev, "Alignment errors = %lld\n",
3778             (long long)adapter->stats.algnerrc);
3779         device_printf(dev, "Collision/Carrier extension errors = %lld\n",
3780             (long long)adapter->stats.cexterr);
3781         device_printf(dev, "RX overruns = %ld\n", adapter->rx_overruns);
3782         device_printf(dev, "watchdog timeouts = %ld\n",
3783             adapter->watchdog_events);
3784         device_printf(dev, "XON Rcvd = %lld\n",
3785             (long long)adapter->stats.xonrxc);
3786         device_printf(dev, "XON Xmtd = %lld\n",
3787             (long long)adapter->stats.xontxc);
3788         device_printf(dev, "XOFF Rcvd = %lld\n",
3789             (long long)adapter->stats.xoffrxc);
3790         device_printf(dev, "XOFF Xmtd = %lld\n",
3791             (long long)adapter->stats.xofftxc);
3792         device_printf(dev, "Good Packets Rcvd = %lld\n",
3793             (long long)adapter->stats.gprc);
3794         device_printf(dev, "Good Packets Xmtd = %lld\n",
3795             (long long)adapter->stats.gptc);
3796 }
3797
3798 static void
3799 em_print_nvm_info(struct adapter *adapter)
3800 {
3801         uint16_t eeprom_data;
3802         int i, j, row = 0;
3803
3804         /* Its a bit crude, but it gets the job done */
3805         kprintf("\nInterface EEPROM Dump:\n");
3806         kprintf("Offset\n0x0000  ");
3807         for (i = 0, j = 0; i < 32; i++, j++) {
3808                 if (j == 8) { /* Make the offset block */
3809                         j = 0; ++row;
3810                         kprintf("\n0x00%x0  ",row);
3811                 }
3812                 e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data);
3813                 kprintf("%04x ", eeprom_data);
3814         }
3815         kprintf("\n");
3816 }
3817
3818 static int
3819 em_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
3820 {
3821         struct adapter *adapter;
3822         struct ifnet *ifp;
3823         int error, result;
3824
3825         result = -1;
3826         error = sysctl_handle_int(oidp, &result, 0, req);
3827         if (error || !req->newptr)
3828                 return (error);
3829
3830         adapter = (struct adapter *)arg1;
3831         ifp = &adapter->arpcom.ac_if;
3832
3833         lwkt_serialize_enter(ifp->if_serializer);
3834
3835         if (result == 1)
3836                 em_print_debug_info(adapter);
3837
3838         /*
3839          * This value will cause a hex dump of the
3840          * first 32 16-bit words of the EEPROM to
3841          * the screen.
3842          */
3843         if (result == 2)
3844                 em_print_nvm_info(adapter);
3845
3846         lwkt_serialize_exit(ifp->if_serializer);
3847
3848         return (error);
3849 }
3850
3851 static int
3852 em_sysctl_stats(SYSCTL_HANDLER_ARGS)
3853 {
3854         int error, result;
3855
3856         result = -1;
3857         error = sysctl_handle_int(oidp, &result, 0, req);
3858         if (error || !req->newptr)
3859                 return (error);
3860
3861         if (result == 1) {
3862                 struct adapter *adapter = (struct adapter *)arg1;
3863                 struct ifnet *ifp = &adapter->arpcom.ac_if;
3864
3865                 lwkt_serialize_enter(ifp->if_serializer);
3866                 em_print_hw_stats(adapter);
3867                 lwkt_serialize_exit(ifp->if_serializer);
3868         }
3869         return (error);
3870 }
3871
3872 static void
3873 em_add_sysctl(struct adapter *adapter)
3874 {
3875         sysctl_ctx_init(&adapter->sysctl_ctx);
3876         adapter->sysctl_tree = SYSCTL_ADD_NODE(&adapter->sysctl_ctx,
3877                                         SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
3878                                         device_get_nameunit(adapter->dev),
3879                                         CTLFLAG_RD, 0, "");
3880         if (adapter->sysctl_tree == NULL) {
3881                 device_printf(adapter->dev, "can't add sysctl node\n");
3882         } else {
3883                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3884                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3885                     OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3886                     em_sysctl_debug_info, "I", "Debug Information");
3887
3888                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3889                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3890                     OID_AUTO, "stats", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3891                     em_sysctl_stats, "I", "Statistics");
3892
3893                 SYSCTL_ADD_INT(&adapter->sysctl_ctx,
3894                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3895                     OID_AUTO, "rxd", CTLFLAG_RD,
3896                     &adapter->num_rx_desc, 0, NULL);
3897                 SYSCTL_ADD_INT(&adapter->sysctl_ctx,
3898                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3899                     OID_AUTO, "txd", CTLFLAG_RD,
3900                     &adapter->num_tx_desc, 0, NULL);
3901
3902                 if (adapter->hw.mac.type >= e1000_82540) {
3903                         SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3904                             SYSCTL_CHILDREN(adapter->sysctl_tree),
3905                             OID_AUTO, "int_throttle_ceil",
3906                             CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3907                             em_sysctl_int_throttle, "I",
3908                             "interrupt throttling rate");
3909                 }
3910                 SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3911                     SYSCTL_CHILDREN(adapter->sysctl_tree),
3912                     OID_AUTO, "int_tx_nsegs",
3913                     CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
3914                     em_sysctl_int_tx_nsegs, "I",
3915                     "# segments per TX interrupt");
3916         }
3917 }
3918
3919 static int
3920 em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS)
3921 {
3922         struct adapter *adapter = (void *)arg1;
3923         struct ifnet *ifp = &adapter->arpcom.ac_if;
3924         int error, throttle;
3925
3926         throttle = adapter->int_throttle_ceil;
3927         error = sysctl_handle_int(oidp, &throttle, 0, req);
3928         if (error || req->newptr == NULL)
3929                 return error;
3930         if (throttle < 0 || throttle > 1000000000 / 256)
3931                 return EINVAL;
3932
3933         if (throttle) {
3934                 /*
3935                  * Set the interrupt throttling rate in 256ns increments,
3936                  * recalculate sysctl value assignment to get exact frequency.
3937                  */
3938                 throttle = 1000000000 / 256 / throttle;
3939
3940                 /* Upper 16bits of ITR is reserved and should be zero */
3941                 if (throttle & 0xffff0000)
3942                         return EINVAL;
3943         }
3944
3945         lwkt_serialize_enter(ifp->if_serializer);
3946
3947         if (throttle)
3948                 adapter->int_throttle_ceil = 1000000000 / 256 / throttle;
3949         else
3950                 adapter->int_throttle_ceil = 0;
3951
3952         if (ifp->if_flags & IFF_RUNNING)
3953                 E1000_WRITE_REG(&adapter->hw, E1000_ITR, throttle);
3954
3955         lwkt_serialize_exit(ifp->if_serializer);
3956
3957         if (bootverbose) {
3958                 if_printf(ifp, "Interrupt moderation set to %d/sec\n",
3959                           adapter->int_throttle_ceil);
3960         }
3961         return 0;
3962 }
3963
3964 static int
3965 em_sysctl_int_tx_nsegs(SYSCTL_HANDLER_ARGS)
3966 {
3967         struct adapter *adapter = (void *)arg1;
3968         struct ifnet *ifp = &adapter->arpcom.ac_if;
3969         int error, segs;
3970
3971         segs = adapter->tx_int_nsegs;
3972         error = sysctl_handle_int(oidp, &segs, 0, req);
3973         if (error || req->newptr == NULL)
3974                 return error;
3975         if (segs <= 0)
3976                 return EINVAL;
3977
3978         lwkt_serialize_enter(ifp->if_serializer);
3979
3980         /*
3981          * Don't allow int_tx_nsegs to become:
3982          * o  Less the oact_tx_desc
3983          * o  Too large that no TX desc will cause TX interrupt to
3984          *    be generated (OACTIVE will never recover)
3985          * o  Too small that will cause tx_dd[] overflow
3986          */
3987         if (segs < adapter->oact_tx_desc ||
3988             segs >= adapter->num_tx_desc - adapter->oact_tx_desc ||
3989             segs < adapter->num_tx_desc / EM_TXDD_SAFE) {
3990                 error = EINVAL;
3991         } else {
3992                 error = 0;
3993                 adapter->tx_int_nsegs = segs;
3994         }
3995
3996         lwkt_serialize_exit(ifp->if_serializer);
3997
3998         return error;
3999 }