qemu/hw/net/e1000e_core.c
<<
>>
Prefs
   1/*
   2* Core code for QEMU e1000e emulation
   3*
   4* Software developer's manuals:
   5* http://www.intel.com/content/dam/doc/datasheet/82574l-gbe-controller-datasheet.pdf
   6*
   7* Copyright (c) 2015 Ravello Systems LTD (http://ravellosystems.com)
   8* Developed by Daynix Computing LTD (http://www.daynix.com)
   9*
  10* Authors:
  11* Dmitry Fleytman <dmitry@daynix.com>
  12* Leonid Bloch <leonid@daynix.com>
  13* Yan Vugenfirer <yan@daynix.com>
  14*
  15* Based on work done by:
  16* Nir Peleg, Tutis Systems Ltd. for Qumranet Inc.
  17* Copyright (c) 2008 Qumranet
  18* Based on work done by:
  19* Copyright (c) 2007 Dan Aloni
  20* Copyright (c) 2004 Antony T Curtis
  21*
  22* This library is free software; you can redistribute it and/or
  23* modify it under the terms of the GNU Lesser General Public
  24* License as published by the Free Software Foundation; either
  25* version 2 of the License, or (at your option) any later version.
  26*
  27* This library is distributed in the hope that it will be useful,
  28* but WITHOUT ANY WARRANTY; without even the implied warranty of
  29* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  30* Lesser General Public License for more details.
  31*
  32* You should have received a copy of the GNU Lesser General Public
  33* License along with this library; if not, see <http://www.gnu.org/licenses/>.
  34*/
  35
  36#include "qemu/osdep.h"
  37#include "sysemu/sysemu.h"
  38#include "net/net.h"
  39#include "net/tap.h"
  40#include "hw/pci/msi.h"
  41#include "hw/pci/msix.h"
  42
  43#include "net_tx_pkt.h"
  44#include "net_rx_pkt.h"
  45
  46#include "e1000x_common.h"
  47#include "e1000e_core.h"
  48
  49#include "trace.h"
  50
  51#define E1000E_MIN_XITR     (500) /* No more then 7813 interrupts per
  52                                     second according to spec 10.2.4.2 */
  53#define E1000E_MAX_TX_FRAGS (64)
  54
  55static inline void
  56e1000e_set_interrupt_cause(E1000ECore *core, uint32_t val);
  57
  58static inline void
  59e1000e_process_ts_option(E1000ECore *core, struct e1000_tx_desc *dp)
  60{
  61    if (le32_to_cpu(dp->upper.data) & E1000_TXD_EXTCMD_TSTAMP) {
  62        trace_e1000e_wrn_no_ts_support();
  63    }
  64}
  65
  66static inline void
  67e1000e_process_snap_option(E1000ECore *core, uint32_t cmd_and_length)
  68{
  69    if (cmd_and_length & E1000_TXD_CMD_SNAP) {
  70        trace_e1000e_wrn_no_snap_support();
  71    }
  72}
  73
  74static inline void
  75e1000e_raise_legacy_irq(E1000ECore *core)
  76{
  77    trace_e1000e_irq_legacy_notify(true);
  78    e1000x_inc_reg_if_not_full(core->mac, IAC);
  79    pci_set_irq(core->owner, 1);
  80}
  81
  82static inline void
  83e1000e_lower_legacy_irq(E1000ECore *core)
  84{
  85    trace_e1000e_irq_legacy_notify(false);
  86    pci_set_irq(core->owner, 0);
  87}
  88
  89static inline void
  90e1000e_intrmgr_rearm_timer(E1000IntrDelayTimer *timer)
  91{
  92    int64_t delay_ns = (int64_t) timer->core->mac[timer->delay_reg] *
  93                                 timer->delay_resolution_ns;
  94
  95    trace_e1000e_irq_rearm_timer(timer->delay_reg << 2, delay_ns);
  96
  97    timer_mod(timer->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + delay_ns);
  98
  99    timer->running = true;
 100}
 101
 102static void
 103e1000e_intmgr_timer_resume(E1000IntrDelayTimer *timer)
 104{
 105    if (timer->running) {
 106        e1000e_intrmgr_rearm_timer(timer);
 107    }
 108}
 109
 110static void
 111e1000e_intmgr_timer_pause(E1000IntrDelayTimer *timer)
 112{
 113    if (timer->running) {
 114        timer_del(timer->timer);
 115    }
 116}
 117
 118static inline void
 119e1000e_intrmgr_stop_timer(E1000IntrDelayTimer *timer)
 120{
 121    if (timer->running) {
 122        timer_del(timer->timer);
 123        timer->running = false;
 124    }
 125}
 126
 127static inline void
 128e1000e_intrmgr_fire_delayed_interrupts(E1000ECore *core)
 129{
 130    trace_e1000e_irq_fire_delayed_interrupts();
 131    e1000e_set_interrupt_cause(core, 0);
 132}
 133
 134static void
 135e1000e_intrmgr_on_timer(void *opaque)
 136{
 137    E1000IntrDelayTimer *timer = opaque;
 138
 139    trace_e1000e_irq_throttling_timer(timer->delay_reg << 2);
 140
 141    timer->running = false;
 142    e1000e_intrmgr_fire_delayed_interrupts(timer->core);
 143}
 144
 145static void
 146e1000e_intrmgr_on_throttling_timer(void *opaque)
 147{
 148    E1000IntrDelayTimer *timer = opaque;
 149
 150    assert(!msix_enabled(timer->core->owner));
 151
 152    timer->running = false;
 153
 154    if (!timer->core->itr_intr_pending) {
 155        trace_e1000e_irq_throttling_no_pending_interrupts();
 156        return;
 157    }
 158
 159    if (msi_enabled(timer->core->owner)) {
 160        trace_e1000e_irq_msi_notify_postponed();
 161        e1000e_set_interrupt_cause(timer->core, 0);
 162    } else {
 163        trace_e1000e_irq_legacy_notify_postponed();
 164        e1000e_set_interrupt_cause(timer->core, 0);
 165    }
 166}
 167
 168static void
 169e1000e_intrmgr_on_msix_throttling_timer(void *opaque)
 170{
 171    E1000IntrDelayTimer *timer = opaque;
 172    int idx = timer - &timer->core->eitr[0];
 173
 174    assert(msix_enabled(timer->core->owner));
 175
 176    timer->running = false;
 177
 178    if (!timer->core->eitr_intr_pending[idx]) {
 179        trace_e1000e_irq_throttling_no_pending_vec(idx);
 180        return;
 181    }
 182
 183    trace_e1000e_irq_msix_notify_postponed_vec(idx);
 184    msix_notify(timer->core->owner, idx);
 185}
 186
 187static void
 188e1000e_intrmgr_initialize_all_timers(E1000ECore *core, bool create)
 189{
 190    int i;
 191
 192    core->radv.delay_reg = RADV;
 193    core->rdtr.delay_reg = RDTR;
 194    core->raid.delay_reg = RAID;
 195    core->tadv.delay_reg = TADV;
 196    core->tidv.delay_reg = TIDV;
 197
 198    core->radv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 199    core->rdtr.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 200    core->raid.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 201    core->tadv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 202    core->tidv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 203
 204    core->radv.core = core;
 205    core->rdtr.core = core;
 206    core->raid.core = core;
 207    core->tadv.core = core;
 208    core->tidv.core = core;
 209
 210    core->itr.core = core;
 211    core->itr.delay_reg = ITR;
 212    core->itr.delay_resolution_ns = E1000_INTR_THROTTLING_NS_RES;
 213
 214    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 215        core->eitr[i].core = core;
 216        core->eitr[i].delay_reg = EITR + i;
 217        core->eitr[i].delay_resolution_ns = E1000_INTR_THROTTLING_NS_RES;
 218    }
 219
 220    if (!create) {
 221        return;
 222    }
 223
 224    core->radv.timer =
 225        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->radv);
 226    core->rdtr.timer =
 227        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->rdtr);
 228    core->raid.timer =
 229        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->raid);
 230
 231    core->tadv.timer =
 232        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->tadv);
 233    core->tidv.timer =
 234        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->tidv);
 235
 236    core->itr.timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
 237                                   e1000e_intrmgr_on_throttling_timer,
 238                                   &core->itr);
 239
 240    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 241        core->eitr[i].timer =
 242            timer_new_ns(QEMU_CLOCK_VIRTUAL,
 243                         e1000e_intrmgr_on_msix_throttling_timer,
 244                         &core->eitr[i]);
 245    }
 246}
 247
 248static inline void
 249e1000e_intrmgr_stop_delay_timers(E1000ECore *core)
 250{
 251    e1000e_intrmgr_stop_timer(&core->radv);
 252    e1000e_intrmgr_stop_timer(&core->rdtr);
 253    e1000e_intrmgr_stop_timer(&core->raid);
 254    e1000e_intrmgr_stop_timer(&core->tidv);
 255    e1000e_intrmgr_stop_timer(&core->tadv);
 256}
 257
 258static bool
 259e1000e_intrmgr_delay_rx_causes(E1000ECore *core, uint32_t *causes)
 260{
 261    uint32_t delayable_causes;
 262    uint32_t rdtr = core->mac[RDTR];
 263    uint32_t radv = core->mac[RADV];
 264    uint32_t raid = core->mac[RAID];
 265
 266    if (msix_enabled(core->owner)) {
 267        return false;
 268    }
 269
 270    delayable_causes = E1000_ICR_RXQ0 |
 271                       E1000_ICR_RXQ1 |
 272                       E1000_ICR_RXT0;
 273
 274    if (!(core->mac[RFCTL] & E1000_RFCTL_ACK_DIS)) {
 275        delayable_causes |= E1000_ICR_ACK;
 276    }
 277
 278    /* Clean up all causes that may be delayed */
 279    core->delayed_causes |= *causes & delayable_causes;
 280    *causes &= ~delayable_causes;
 281
 282    /* Check if delayed RX interrupts disabled by client
 283       or if there are causes that cannot be delayed */
 284    if ((rdtr == 0) || (*causes != 0)) {
 285        return false;
 286    }
 287
 288    /* Check if delayed RX ACK interrupts disabled by client
 289       and there is an ACK packet received */
 290    if ((raid == 0) && (core->delayed_causes & E1000_ICR_ACK)) {
 291        return false;
 292    }
 293
 294    /* All causes delayed */
 295    e1000e_intrmgr_rearm_timer(&core->rdtr);
 296
 297    if (!core->radv.running && (radv != 0)) {
 298        e1000e_intrmgr_rearm_timer(&core->radv);
 299    }
 300
 301    if (!core->raid.running && (core->delayed_causes & E1000_ICR_ACK)) {
 302        e1000e_intrmgr_rearm_timer(&core->raid);
 303    }
 304
 305    return true;
 306}
 307
 308static bool
 309e1000e_intrmgr_delay_tx_causes(E1000ECore *core, uint32_t *causes)
 310{
 311    static const uint32_t delayable_causes = E1000_ICR_TXQ0 |
 312                                             E1000_ICR_TXQ1 |
 313                                             E1000_ICR_TXQE |
 314                                             E1000_ICR_TXDW;
 315
 316    if (msix_enabled(core->owner)) {
 317        return false;
 318    }
 319
 320    /* Clean up all causes that may be delayed */
 321    core->delayed_causes |= *causes & delayable_causes;
 322    *causes &= ~delayable_causes;
 323
 324    /* If there are causes that cannot be delayed */
 325    if (*causes != 0) {
 326        return false;
 327    }
 328
 329    /* All causes delayed */
 330    e1000e_intrmgr_rearm_timer(&core->tidv);
 331
 332    if (!core->tadv.running && (core->mac[TADV] != 0)) {
 333        e1000e_intrmgr_rearm_timer(&core->tadv);
 334    }
 335
 336    return true;
 337}
 338
 339static uint32_t
 340e1000e_intmgr_collect_delayed_causes(E1000ECore *core)
 341{
 342    uint32_t res;
 343
 344    if (msix_enabled(core->owner)) {
 345        assert(core->delayed_causes == 0);
 346        return 0;
 347    }
 348
 349    res = core->delayed_causes;
 350    core->delayed_causes = 0;
 351
 352    e1000e_intrmgr_stop_delay_timers(core);
 353
 354    return res;
 355}
 356
 357static void
 358e1000e_intrmgr_fire_all_timers(E1000ECore *core)
 359{
 360    int i;
 361    uint32_t val = e1000e_intmgr_collect_delayed_causes(core);
 362
 363    trace_e1000e_irq_adding_delayed_causes(val, core->mac[ICR]);
 364    core->mac[ICR] |= val;
 365
 366    if (core->itr.running) {
 367        timer_del(core->itr.timer);
 368        e1000e_intrmgr_on_throttling_timer(&core->itr);
 369    }
 370
 371    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 372        if (core->eitr[i].running) {
 373            timer_del(core->eitr[i].timer);
 374            e1000e_intrmgr_on_msix_throttling_timer(&core->eitr[i]);
 375        }
 376    }
 377}
 378
 379static void
 380e1000e_intrmgr_resume(E1000ECore *core)
 381{
 382    int i;
 383
 384    e1000e_intmgr_timer_resume(&core->radv);
 385    e1000e_intmgr_timer_resume(&core->rdtr);
 386    e1000e_intmgr_timer_resume(&core->raid);
 387    e1000e_intmgr_timer_resume(&core->tidv);
 388    e1000e_intmgr_timer_resume(&core->tadv);
 389
 390    e1000e_intmgr_timer_resume(&core->itr);
 391
 392    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 393        e1000e_intmgr_timer_resume(&core->eitr[i]);
 394    }
 395}
 396
 397static void
 398e1000e_intrmgr_pause(E1000ECore *core)
 399{
 400    int i;
 401
 402    e1000e_intmgr_timer_pause(&core->radv);
 403    e1000e_intmgr_timer_pause(&core->rdtr);
 404    e1000e_intmgr_timer_pause(&core->raid);
 405    e1000e_intmgr_timer_pause(&core->tidv);
 406    e1000e_intmgr_timer_pause(&core->tadv);
 407
 408    e1000e_intmgr_timer_pause(&core->itr);
 409
 410    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 411        e1000e_intmgr_timer_pause(&core->eitr[i]);
 412    }
 413}
 414
 415static void
 416e1000e_intrmgr_reset(E1000ECore *core)
 417{
 418    int i;
 419
 420    core->delayed_causes = 0;
 421
 422    e1000e_intrmgr_stop_delay_timers(core);
 423
 424    e1000e_intrmgr_stop_timer(&core->itr);
 425
 426    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 427        e1000e_intrmgr_stop_timer(&core->eitr[i]);
 428    }
 429}
 430
 431static void
 432e1000e_intrmgr_pci_unint(E1000ECore *core)
 433{
 434    int i;
 435
 436    timer_del(core->radv.timer);
 437    timer_free(core->radv.timer);
 438    timer_del(core->rdtr.timer);
 439    timer_free(core->rdtr.timer);
 440    timer_del(core->raid.timer);
 441    timer_free(core->raid.timer);
 442
 443    timer_del(core->tadv.timer);
 444    timer_free(core->tadv.timer);
 445    timer_del(core->tidv.timer);
 446    timer_free(core->tidv.timer);
 447
 448    timer_del(core->itr.timer);
 449    timer_free(core->itr.timer);
 450
 451    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 452        timer_del(core->eitr[i].timer);
 453        timer_free(core->eitr[i].timer);
 454    }
 455}
 456
 457static void
 458e1000e_intrmgr_pci_realize(E1000ECore *core)
 459{
 460    e1000e_intrmgr_initialize_all_timers(core, true);
 461}
 462
 463static inline bool
 464e1000e_rx_csum_enabled(E1000ECore *core)
 465{
 466    return (core->mac[RXCSUM] & E1000_RXCSUM_PCSD) ? false : true;
 467}
 468
 469static inline bool
 470e1000e_rx_use_legacy_descriptor(E1000ECore *core)
 471{
 472    return (core->mac[RFCTL] & E1000_RFCTL_EXTEN) ? false : true;
 473}
 474
 475static inline bool
 476e1000e_rx_use_ps_descriptor(E1000ECore *core)
 477{
 478    return !e1000e_rx_use_legacy_descriptor(core) &&
 479           (core->mac[RCTL] & E1000_RCTL_DTYP_PS);
 480}
 481
 482static inline bool
 483e1000e_rss_enabled(E1000ECore *core)
 484{
 485    return E1000_MRQC_ENABLED(core->mac[MRQC]) &&
 486           !e1000e_rx_csum_enabled(core) &&
 487           !e1000e_rx_use_legacy_descriptor(core);
 488}
 489
 490typedef struct E1000E_RSSInfo_st {
 491    bool enabled;
 492    uint32_t hash;
 493    uint32_t queue;
 494    uint32_t type;
 495} E1000E_RSSInfo;
 496
 497static uint32_t
 498e1000e_rss_get_hash_type(E1000ECore *core, struct NetRxPkt *pkt)
 499{
 500    bool isip4, isip6, isudp, istcp;
 501
 502    assert(e1000e_rss_enabled(core));
 503
 504    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
 505
 506    if (isip4) {
 507        bool fragment = net_rx_pkt_get_ip4_info(pkt)->fragment;
 508
 509        trace_e1000e_rx_rss_ip4(fragment, istcp, core->mac[MRQC],
 510                                E1000_MRQC_EN_TCPIPV4(core->mac[MRQC]),
 511                                E1000_MRQC_EN_IPV4(core->mac[MRQC]));
 512
 513        if (!fragment && istcp && E1000_MRQC_EN_TCPIPV4(core->mac[MRQC])) {
 514            return E1000_MRQ_RSS_TYPE_IPV4TCP;
 515        }
 516
 517        if (E1000_MRQC_EN_IPV4(core->mac[MRQC])) {
 518            return E1000_MRQ_RSS_TYPE_IPV4;
 519        }
 520    } else if (isip6) {
 521        eth_ip6_hdr_info *ip6info = net_rx_pkt_get_ip6_info(pkt);
 522
 523        bool ex_dis = core->mac[RFCTL] & E1000_RFCTL_IPV6_EX_DIS;
 524        bool new_ex_dis = core->mac[RFCTL] & E1000_RFCTL_NEW_IPV6_EXT_DIS;
 525
 526        /*
 527         * Following two traces must not be combined because resulting
 528         * event will have 11 arguments totally and some trace backends
 529         * (at least "ust") have limitation of maximum 10 arguments per
 530         * event. Events with more arguments fail to compile for
 531         * backends like these.
 532         */
 533        trace_e1000e_rx_rss_ip6_rfctl(core->mac[RFCTL]);
 534        trace_e1000e_rx_rss_ip6(ex_dis, new_ex_dis, istcp,
 535                                ip6info->has_ext_hdrs,
 536                                ip6info->rss_ex_dst_valid,
 537                                ip6info->rss_ex_src_valid,
 538                                core->mac[MRQC],
 539                                E1000_MRQC_EN_TCPIPV6(core->mac[MRQC]),
 540                                E1000_MRQC_EN_IPV6EX(core->mac[MRQC]),
 541                                E1000_MRQC_EN_IPV6(core->mac[MRQC]));
 542
 543        if ((!ex_dis || !ip6info->has_ext_hdrs) &&
 544            (!new_ex_dis || !(ip6info->rss_ex_dst_valid ||
 545                              ip6info->rss_ex_src_valid))) {
 546
 547            if (istcp && !ip6info->fragment &&
 548                E1000_MRQC_EN_TCPIPV6(core->mac[MRQC])) {
 549                return E1000_MRQ_RSS_TYPE_IPV6TCP;
 550            }
 551
 552            if (E1000_MRQC_EN_IPV6EX(core->mac[MRQC])) {
 553                return E1000_MRQ_RSS_TYPE_IPV6EX;
 554            }
 555
 556        }
 557
 558        if (E1000_MRQC_EN_IPV6(core->mac[MRQC])) {
 559            return E1000_MRQ_RSS_TYPE_IPV6;
 560        }
 561
 562    }
 563
 564    return E1000_MRQ_RSS_TYPE_NONE;
 565}
 566
 567static uint32_t
 568e1000e_rss_calc_hash(E1000ECore *core,
 569                     struct NetRxPkt *pkt,
 570                     E1000E_RSSInfo *info)
 571{
 572    NetRxPktRssType type;
 573
 574    assert(e1000e_rss_enabled(core));
 575
 576    switch (info->type) {
 577    case E1000_MRQ_RSS_TYPE_IPV4:
 578        type = NetPktRssIpV4;
 579        break;
 580    case E1000_MRQ_RSS_TYPE_IPV4TCP:
 581        type = NetPktRssIpV4Tcp;
 582        break;
 583    case E1000_MRQ_RSS_TYPE_IPV6TCP:
 584        type = NetPktRssIpV6Tcp;
 585        break;
 586    case E1000_MRQ_RSS_TYPE_IPV6:
 587        type = NetPktRssIpV6;
 588        break;
 589    case E1000_MRQ_RSS_TYPE_IPV6EX:
 590        type = NetPktRssIpV6Ex;
 591        break;
 592    default:
 593        assert(false);
 594        return 0;
 595    }
 596
 597    return net_rx_pkt_calc_rss_hash(pkt, type, (uint8_t *) &core->mac[RSSRK]);
 598}
 599
 600static void
 601e1000e_rss_parse_packet(E1000ECore *core,
 602                        struct NetRxPkt *pkt,
 603                        E1000E_RSSInfo *info)
 604{
 605    trace_e1000e_rx_rss_started();
 606
 607    if (!e1000e_rss_enabled(core)) {
 608        info->enabled = false;
 609        info->hash = 0;
 610        info->queue = 0;
 611        info->type = 0;
 612        trace_e1000e_rx_rss_disabled();
 613        return;
 614    }
 615
 616    info->enabled = true;
 617
 618    info->type = e1000e_rss_get_hash_type(core, pkt);
 619
 620    trace_e1000e_rx_rss_type(info->type);
 621
 622    if (info->type == E1000_MRQ_RSS_TYPE_NONE) {
 623        info->hash = 0;
 624        info->queue = 0;
 625        return;
 626    }
 627
 628    info->hash = e1000e_rss_calc_hash(core, pkt, info);
 629    info->queue = E1000_RSS_QUEUE(&core->mac[RETA], info->hash);
 630}
 631
 632static void
 633e1000e_setup_tx_offloads(E1000ECore *core, struct e1000e_tx *tx)
 634{
 635    if (tx->props.tse && tx->cptse) {
 636        net_tx_pkt_build_vheader(tx->tx_pkt, true, true, tx->props.mss);
 637        net_tx_pkt_update_ip_checksums(tx->tx_pkt);
 638        e1000x_inc_reg_if_not_full(core->mac, TSCTC);
 639        return;
 640    }
 641
 642    if (tx->sum_needed & E1000_TXD_POPTS_TXSM) {
 643        net_tx_pkt_build_vheader(tx->tx_pkt, false, true, 0);
 644    }
 645
 646    if (tx->sum_needed & E1000_TXD_POPTS_IXSM) {
 647        net_tx_pkt_update_ip_hdr_checksum(tx->tx_pkt);
 648    }
 649}
 650
 651static bool
 652e1000e_tx_pkt_send(E1000ECore *core, struct e1000e_tx *tx, int queue_index)
 653{
 654    int target_queue = MIN(core->max_queue_num, queue_index);
 655    NetClientState *queue = qemu_get_subqueue(core->owner_nic, target_queue);
 656
 657    e1000e_setup_tx_offloads(core, tx);
 658
 659    net_tx_pkt_dump(tx->tx_pkt);
 660
 661    if ((core->phy[0][PHY_CTRL] & MII_CR_LOOPBACK) ||
 662        ((core->mac[RCTL] & E1000_RCTL_LBM_MAC) == E1000_RCTL_LBM_MAC)) {
 663        return net_tx_pkt_send_loopback(tx->tx_pkt, queue);
 664    } else {
 665        return net_tx_pkt_send(tx->tx_pkt, queue);
 666    }
 667}
 668
 669static void
 670e1000e_on_tx_done_update_stats(E1000ECore *core, struct NetTxPkt *tx_pkt)
 671{
 672    static const int PTCregs[6] = { PTC64, PTC127, PTC255, PTC511,
 673                                    PTC1023, PTC1522 };
 674
 675    size_t tot_len = net_tx_pkt_get_total_len(tx_pkt);
 676
 677    e1000x_increase_size_stats(core->mac, PTCregs, tot_len);
 678    e1000x_inc_reg_if_not_full(core->mac, TPT);
 679    e1000x_grow_8reg_if_not_full(core->mac, TOTL, tot_len);
 680
 681    switch (net_tx_pkt_get_packet_type(tx_pkt)) {
 682    case ETH_PKT_BCAST:
 683        e1000x_inc_reg_if_not_full(core->mac, BPTC);
 684        break;
 685    case ETH_PKT_MCAST:
 686        e1000x_inc_reg_if_not_full(core->mac, MPTC);
 687        break;
 688    case ETH_PKT_UCAST:
 689        break;
 690    default:
 691        g_assert_not_reached();
 692    }
 693
 694    core->mac[GPTC] = core->mac[TPT];
 695    core->mac[GOTCL] = core->mac[TOTL];
 696    core->mac[GOTCH] = core->mac[TOTH];
 697}
 698
 699static void
 700e1000e_process_tx_desc(E1000ECore *core,
 701                       struct e1000e_tx *tx,
 702                       struct e1000_tx_desc *dp,
 703                       int queue_index)
 704{
 705    uint32_t txd_lower = le32_to_cpu(dp->lower.data);
 706    uint32_t dtype = txd_lower & (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D);
 707    unsigned int split_size = txd_lower & 0xffff;
 708    uint64_t addr;
 709    struct e1000_context_desc *xp = (struct e1000_context_desc *)dp;
 710    bool eop = txd_lower & E1000_TXD_CMD_EOP;
 711
 712    if (dtype == E1000_TXD_CMD_DEXT) { /* context descriptor */
 713        e1000x_read_tx_ctx_descr(xp, &tx->props);
 714        e1000e_process_snap_option(core, le32_to_cpu(xp->cmd_and_length));
 715        return;
 716    } else if (dtype == (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D)) {
 717        /* data descriptor */
 718        tx->sum_needed = le32_to_cpu(dp->upper.data) >> 8;
 719        tx->cptse = (txd_lower & E1000_TXD_CMD_TSE) ? 1 : 0;
 720        e1000e_process_ts_option(core, dp);
 721    } else {
 722        /* legacy descriptor */
 723        e1000e_process_ts_option(core, dp);
 724        tx->cptse = 0;
 725    }
 726
 727    addr = le64_to_cpu(dp->buffer_addr);
 728
 729    if (!tx->skip_cp) {
 730        if (!net_tx_pkt_add_raw_fragment(tx->tx_pkt, addr, split_size)) {
 731            tx->skip_cp = true;
 732        }
 733    }
 734
 735    if (eop) {
 736        if (!tx->skip_cp && net_tx_pkt_parse(tx->tx_pkt)) {
 737            if (e1000x_vlan_enabled(core->mac) &&
 738                e1000x_is_vlan_txd(txd_lower)) {
 739                net_tx_pkt_setup_vlan_header_ex(tx->tx_pkt,
 740                    le16_to_cpu(dp->upper.fields.special), core->vet);
 741            }
 742            if (e1000e_tx_pkt_send(core, tx, queue_index)) {
 743                e1000e_on_tx_done_update_stats(core, tx->tx_pkt);
 744            }
 745        }
 746
 747        tx->skip_cp = false;
 748        net_tx_pkt_reset(tx->tx_pkt);
 749
 750        tx->sum_needed = 0;
 751        tx->cptse = 0;
 752    }
 753}
 754
 755static inline uint32_t
 756e1000e_tx_wb_interrupt_cause(E1000ECore *core, int queue_idx)
 757{
 758    if (!msix_enabled(core->owner)) {
 759        return E1000_ICR_TXDW;
 760    }
 761
 762    return (queue_idx == 0) ? E1000_ICR_TXQ0 : E1000_ICR_TXQ1;
 763}
 764
 765static inline uint32_t
 766e1000e_rx_wb_interrupt_cause(E1000ECore *core, int queue_idx,
 767                             bool min_threshold_hit)
 768{
 769    if (!msix_enabled(core->owner)) {
 770        return E1000_ICS_RXT0 | (min_threshold_hit ? E1000_ICS_RXDMT0 : 0);
 771    }
 772
 773    return (queue_idx == 0) ? E1000_ICR_RXQ0 : E1000_ICR_RXQ1;
 774}
 775
 776static uint32_t
 777e1000e_txdesc_writeback(E1000ECore *core, dma_addr_t base,
 778                        struct e1000_tx_desc *dp, bool *ide, int queue_idx)
 779{
 780    uint32_t txd_upper, txd_lower = le32_to_cpu(dp->lower.data);
 781
 782    if (!(txd_lower & E1000_TXD_CMD_RS) &&
 783        !(core->mac[IVAR] & E1000_IVAR_TX_INT_EVERY_WB)) {
 784        return 0;
 785    }
 786
 787    *ide = (txd_lower & E1000_TXD_CMD_IDE) ? true : false;
 788
 789    txd_upper = le32_to_cpu(dp->upper.data) | E1000_TXD_STAT_DD;
 790
 791    dp->upper.data = cpu_to_le32(txd_upper);
 792    pci_dma_write(core->owner, base + ((char *)&dp->upper - (char *)dp),
 793                  &dp->upper, sizeof(dp->upper));
 794    return e1000e_tx_wb_interrupt_cause(core, queue_idx);
 795}
 796
 797typedef struct E1000E_RingInfo_st {
 798    int dbah;
 799    int dbal;
 800    int dlen;
 801    int dh;
 802    int dt;
 803    int idx;
 804} E1000E_RingInfo;
 805
 806static inline bool
 807e1000e_ring_empty(E1000ECore *core, const E1000E_RingInfo *r)
 808{
 809    return core->mac[r->dh] == core->mac[r->dt] ||
 810                core->mac[r->dt] >= core->mac[r->dlen] / E1000_RING_DESC_LEN;
 811}
 812
 813static inline uint64_t
 814e1000e_ring_base(E1000ECore *core, const E1000E_RingInfo *r)
 815{
 816    uint64_t bah = core->mac[r->dbah];
 817    uint64_t bal = core->mac[r->dbal];
 818
 819    return (bah << 32) + bal;
 820}
 821
 822static inline uint64_t
 823e1000e_ring_head_descr(E1000ECore *core, const E1000E_RingInfo *r)
 824{
 825    return e1000e_ring_base(core, r) + E1000_RING_DESC_LEN * core->mac[r->dh];
 826}
 827
 828static inline void
 829e1000e_ring_advance(E1000ECore *core, const E1000E_RingInfo *r, uint32_t count)
 830{
 831    core->mac[r->dh] += count;
 832
 833    if (core->mac[r->dh] * E1000_RING_DESC_LEN >= core->mac[r->dlen]) {
 834        core->mac[r->dh] = 0;
 835    }
 836}
 837
 838static inline uint32_t
 839e1000e_ring_free_descr_num(E1000ECore *core, const E1000E_RingInfo *r)
 840{
 841    trace_e1000e_ring_free_space(r->idx, core->mac[r->dlen],
 842                                 core->mac[r->dh],  core->mac[r->dt]);
 843
 844    if (core->mac[r->dh] <= core->mac[r->dt]) {
 845        return core->mac[r->dt] - core->mac[r->dh];
 846    }
 847
 848    if (core->mac[r->dh] > core->mac[r->dt]) {
 849        return core->mac[r->dlen] / E1000_RING_DESC_LEN +
 850               core->mac[r->dt] - core->mac[r->dh];
 851    }
 852
 853    g_assert_not_reached();
 854    return 0;
 855}
 856
 857static inline bool
 858e1000e_ring_enabled(E1000ECore *core, const E1000E_RingInfo *r)
 859{
 860    return core->mac[r->dlen] > 0;
 861}
 862
 863static inline uint32_t
 864e1000e_ring_len(E1000ECore *core, const E1000E_RingInfo *r)
 865{
 866    return core->mac[r->dlen];
 867}
 868
 869typedef struct E1000E_TxRing_st {
 870    const E1000E_RingInfo *i;
 871    struct e1000e_tx *tx;
 872} E1000E_TxRing;
 873
 874static inline int
 875e1000e_mq_queue_idx(int base_reg_idx, int reg_idx)
 876{
 877    return (reg_idx - base_reg_idx) / (0x100 >> 2);
 878}
 879
 880static inline void
 881e1000e_tx_ring_init(E1000ECore *core, E1000E_TxRing *txr, int idx)
 882{
 883    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 884        { TDBAH,  TDBAL,  TDLEN,  TDH,  TDT, 0 },
 885        { TDBAH1, TDBAL1, TDLEN1, TDH1, TDT1, 1 }
 886    };
 887
 888    assert(idx < ARRAY_SIZE(i));
 889
 890    txr->i     = &i[idx];
 891    txr->tx    = &core->tx[idx];
 892}
 893
 894typedef struct E1000E_RxRing_st {
 895    const E1000E_RingInfo *i;
 896} E1000E_RxRing;
 897
 898static inline void
 899e1000e_rx_ring_init(E1000ECore *core, E1000E_RxRing *rxr, int idx)
 900{
 901    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 902        { RDBAH0, RDBAL0, RDLEN0, RDH0, RDT0, 0 },
 903        { RDBAH1, RDBAL1, RDLEN1, RDH1, RDT1, 1 }
 904    };
 905
 906    assert(idx < ARRAY_SIZE(i));
 907
 908    rxr->i      = &i[idx];
 909}
 910
 911static void
 912e1000e_start_xmit(E1000ECore *core, const E1000E_TxRing *txr)
 913{
 914    dma_addr_t base;
 915    struct e1000_tx_desc desc;
 916    bool ide = false;
 917    const E1000E_RingInfo *txi = txr->i;
 918    uint32_t cause = E1000_ICS_TXQE;
 919
 920    if (!(core->mac[TCTL] & E1000_TCTL_EN)) {
 921        trace_e1000e_tx_disabled();
 922        return;
 923    }
 924
 925    while (!e1000e_ring_empty(core, txi)) {
 926        base = e1000e_ring_head_descr(core, txi);
 927
 928        pci_dma_read(core->owner, base, &desc, sizeof(desc));
 929
 930        trace_e1000e_tx_descr((void *)(intptr_t)desc.buffer_addr,
 931                              desc.lower.data, desc.upper.data);
 932
 933        e1000e_process_tx_desc(core, txr->tx, &desc, txi->idx);
 934        cause |= e1000e_txdesc_writeback(core, base, &desc, &ide, txi->idx);
 935
 936        e1000e_ring_advance(core, txi, 1);
 937    }
 938
 939    if (!ide || !e1000e_intrmgr_delay_tx_causes(core, &cause)) {
 940        e1000e_set_interrupt_cause(core, cause);
 941    }
 942}
 943
 944static bool
 945e1000e_has_rxbufs(E1000ECore *core, const E1000E_RingInfo *r,
 946                  size_t total_size)
 947{
 948    uint32_t bufs = e1000e_ring_free_descr_num(core, r);
 949
 950    trace_e1000e_rx_has_buffers(r->idx, bufs, total_size,
 951                                core->rx_desc_buf_size);
 952
 953    return total_size <= bufs / (core->rx_desc_len / E1000_MIN_RX_DESC_LEN) *
 954                         core->rx_desc_buf_size;
 955}
 956
 957void
 958e1000e_start_recv(E1000ECore *core)
 959{
 960    int i;
 961
 962    trace_e1000e_rx_start_recv();
 963
 964    for (i = 0; i <= core->max_queue_num; i++) {
 965        qemu_flush_queued_packets(qemu_get_subqueue(core->owner_nic, i));
 966    }
 967}
 968
 969int
 970e1000e_can_receive(E1000ECore *core)
 971{
 972    int i;
 973
 974    if (!e1000x_rx_ready(core->owner, core->mac)) {
 975        return false;
 976    }
 977
 978    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
 979        E1000E_RxRing rxr;
 980
 981        e1000e_rx_ring_init(core, &rxr, i);
 982        if (e1000e_ring_enabled(core, rxr.i) &&
 983            e1000e_has_rxbufs(core, rxr.i, 1)) {
 984            trace_e1000e_rx_can_recv();
 985            return true;
 986        }
 987    }
 988
 989    trace_e1000e_rx_can_recv_rings_full();
 990    return false;
 991}
 992
 993ssize_t
 994e1000e_receive(E1000ECore *core, const uint8_t *buf, size_t size)
 995{
 996    const struct iovec iov = {
 997        .iov_base = (uint8_t *)buf,
 998        .iov_len = size
 999    };
1000
1001    return e1000e_receive_iov(core, &iov, 1);
1002}
1003
1004static inline bool
1005e1000e_rx_l3_cso_enabled(E1000ECore *core)
1006{
1007    return !!(core->mac[RXCSUM] & E1000_RXCSUM_IPOFLD);
1008}
1009
1010static inline bool
1011e1000e_rx_l4_cso_enabled(E1000ECore *core)
1012{
1013    return !!(core->mac[RXCSUM] & E1000_RXCSUM_TUOFLD);
1014}
1015
1016static bool
1017e1000e_receive_filter(E1000ECore *core, const uint8_t *buf, int size)
1018{
1019    uint32_t rctl = core->mac[RCTL];
1020
1021    if (e1000x_is_vlan_packet(buf, core->vet) &&
1022        e1000x_vlan_rx_filter_enabled(core->mac)) {
1023        uint16_t vid = lduw_be_p(buf + 14);
1024        uint32_t vfta = ldl_le_p((uint32_t *)(core->mac + VFTA) +
1025                                 ((vid >> 5) & 0x7f));
1026        if ((vfta & (1 << (vid & 0x1f))) == 0) {
1027            trace_e1000e_rx_flt_vlan_mismatch(vid);
1028            return false;
1029        } else {
1030            trace_e1000e_rx_flt_vlan_match(vid);
1031        }
1032    }
1033
1034    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1035    case ETH_PKT_UCAST:
1036        if (rctl & E1000_RCTL_UPE) {
1037            return true; /* promiscuous ucast */
1038        }
1039        break;
1040
1041    case ETH_PKT_BCAST:
1042        if (rctl & E1000_RCTL_BAM) {
1043            return true; /* broadcast enabled */
1044        }
1045        break;
1046
1047    case ETH_PKT_MCAST:
1048        if (rctl & E1000_RCTL_MPE) {
1049            return true; /* promiscuous mcast */
1050        }
1051        break;
1052
1053    default:
1054        g_assert_not_reached();
1055    }
1056
1057    return e1000x_rx_group_filter(core->mac, buf);
1058}
1059
1060static inline void
1061e1000e_read_lgcy_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1062{
1063    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1064    *buff_addr = le64_to_cpu(d->buffer_addr);
1065}
1066
1067static inline void
1068e1000e_read_ext_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1069{
1070    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1071    *buff_addr = le64_to_cpu(d->read.buffer_addr);
1072}
1073
1074static inline void
1075e1000e_read_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1076                        hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1077{
1078    int i;
1079    union e1000_rx_desc_packet_split *d =
1080        (union e1000_rx_desc_packet_split *) desc;
1081
1082    for (i = 0; i < MAX_PS_BUFFERS; i++) {
1083        (*buff_addr)[i] = le64_to_cpu(d->read.buffer_addr[i]);
1084    }
1085
1086    trace_e1000e_rx_desc_ps_read((*buff_addr)[0], (*buff_addr)[1],
1087                                 (*buff_addr)[2], (*buff_addr)[3]);
1088}
1089
1090static inline void
1091e1000e_read_rx_descr(E1000ECore *core, uint8_t *desc,
1092                     hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1093{
1094    if (e1000e_rx_use_legacy_descriptor(core)) {
1095        e1000e_read_lgcy_rx_descr(core, desc, &(*buff_addr)[0]);
1096        (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1097    } else {
1098        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1099            e1000e_read_ps_rx_descr(core, desc, buff_addr);
1100        } else {
1101            e1000e_read_ext_rx_descr(core, desc, &(*buff_addr)[0]);
1102            (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1103        }
1104    }
1105}
1106
1107static void
1108e1000e_verify_csum_in_sw(E1000ECore *core,
1109                         struct NetRxPkt *pkt,
1110                         uint32_t *status_flags,
1111                         bool istcp, bool isudp)
1112{
1113    bool csum_valid;
1114    uint32_t csum_error;
1115
1116    if (e1000e_rx_l3_cso_enabled(core)) {
1117        if (!net_rx_pkt_validate_l3_csum(pkt, &csum_valid)) {
1118            trace_e1000e_rx_metadata_l3_csum_validation_failed();
1119        } else {
1120            csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_IPE;
1121            *status_flags |= E1000_RXD_STAT_IPCS | csum_error;
1122        }
1123    } else {
1124        trace_e1000e_rx_metadata_l3_cso_disabled();
1125    }
1126
1127    if (!e1000e_rx_l4_cso_enabled(core)) {
1128        trace_e1000e_rx_metadata_l4_cso_disabled();
1129        return;
1130    }
1131
1132    if (!net_rx_pkt_validate_l4_csum(pkt, &csum_valid)) {
1133        trace_e1000e_rx_metadata_l4_csum_validation_failed();
1134        return;
1135    }
1136
1137    csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_TCPE;
1138
1139    if (istcp) {
1140        *status_flags |= E1000_RXD_STAT_TCPCS |
1141                         csum_error;
1142    } else if (isudp) {
1143        *status_flags |= E1000_RXD_STAT_TCPCS |
1144                         E1000_RXD_STAT_UDPCS |
1145                         csum_error;
1146    }
1147}
1148
1149static inline bool
1150e1000e_is_tcp_ack(E1000ECore *core, struct NetRxPkt *rx_pkt)
1151{
1152    if (!net_rx_pkt_is_tcp_ack(rx_pkt)) {
1153        return false;
1154    }
1155
1156    if (core->mac[RFCTL] & E1000_RFCTL_ACK_DATA_DIS) {
1157        return !net_rx_pkt_has_tcp_data(rx_pkt);
1158    }
1159
1160    return true;
1161}
1162
1163static void
1164e1000e_build_rx_metadata(E1000ECore *core,
1165                         struct NetRxPkt *pkt,
1166                         bool is_eop,
1167                         const E1000E_RSSInfo *rss_info,
1168                         uint32_t *rss, uint32_t *mrq,
1169                         uint32_t *status_flags,
1170                         uint16_t *ip_id,
1171                         uint16_t *vlan_tag)
1172{
1173    struct virtio_net_hdr *vhdr;
1174    bool isip4, isip6, istcp, isudp;
1175    uint32_t pkt_type;
1176
1177    *status_flags = E1000_RXD_STAT_DD;
1178
1179    /* No additional metadata needed for non-EOP descriptors */
1180    if (!is_eop) {
1181        goto func_exit;
1182    }
1183
1184    *status_flags |= E1000_RXD_STAT_EOP;
1185
1186    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1187    trace_e1000e_rx_metadata_protocols(isip4, isip6, isudp, istcp);
1188
1189    /* VLAN state */
1190    if (net_rx_pkt_is_vlan_stripped(pkt)) {
1191        *status_flags |= E1000_RXD_STAT_VP;
1192        *vlan_tag = cpu_to_le16(net_rx_pkt_get_vlan_tag(pkt));
1193        trace_e1000e_rx_metadata_vlan(*vlan_tag);
1194    }
1195
1196    /* Packet parsing results */
1197    if ((core->mac[RXCSUM] & E1000_RXCSUM_PCSD) != 0) {
1198        if (rss_info->enabled) {
1199            *rss = cpu_to_le32(rss_info->hash);
1200            *mrq = cpu_to_le32(rss_info->type | (rss_info->queue << 8));
1201            trace_e1000e_rx_metadata_rss(*rss, *mrq);
1202        }
1203    } else if (isip4) {
1204            *status_flags |= E1000_RXD_STAT_IPIDV;
1205            *ip_id = cpu_to_le16(net_rx_pkt_get_ip_id(pkt));
1206            trace_e1000e_rx_metadata_ip_id(*ip_id);
1207    }
1208
1209    if (istcp && e1000e_is_tcp_ack(core, pkt)) {
1210        *status_flags |= E1000_RXD_STAT_ACK;
1211        trace_e1000e_rx_metadata_ack();
1212    }
1213
1214    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_DIS)) {
1215        trace_e1000e_rx_metadata_ipv6_filtering_disabled();
1216        pkt_type = E1000_RXD_PKT_MAC;
1217    } else if (istcp || isudp) {
1218        pkt_type = isip4 ? E1000_RXD_PKT_IP4_XDP : E1000_RXD_PKT_IP6_XDP;
1219    } else if (isip4 || isip6) {
1220        pkt_type = isip4 ? E1000_RXD_PKT_IP4 : E1000_RXD_PKT_IP6;
1221    } else {
1222        pkt_type = E1000_RXD_PKT_MAC;
1223    }
1224
1225    *status_flags |= E1000_RXD_PKT_TYPE(pkt_type);
1226    trace_e1000e_rx_metadata_pkt_type(pkt_type);
1227
1228    /* RX CSO information */
1229    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_XSUM_DIS)) {
1230        trace_e1000e_rx_metadata_ipv6_sum_disabled();
1231        goto func_exit;
1232    }
1233
1234    if (!net_rx_pkt_has_virt_hdr(pkt)) {
1235        trace_e1000e_rx_metadata_no_virthdr();
1236        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1237        goto func_exit;
1238    }
1239
1240    vhdr = net_rx_pkt_get_vhdr(pkt);
1241
1242    if (!(vhdr->flags & VIRTIO_NET_HDR_F_DATA_VALID) &&
1243        !(vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
1244        trace_e1000e_rx_metadata_virthdr_no_csum_info();
1245        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1246        goto func_exit;
1247    }
1248
1249    if (e1000e_rx_l3_cso_enabled(core)) {
1250        *status_flags |= isip4 ? E1000_RXD_STAT_IPCS : 0;
1251    } else {
1252        trace_e1000e_rx_metadata_l3_cso_disabled();
1253    }
1254
1255    if (e1000e_rx_l4_cso_enabled(core)) {
1256        if (istcp) {
1257            *status_flags |= E1000_RXD_STAT_TCPCS;
1258        } else if (isudp) {
1259            *status_flags |= E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS;
1260        }
1261    } else {
1262        trace_e1000e_rx_metadata_l4_cso_disabled();
1263    }
1264
1265    trace_e1000e_rx_metadata_status_flags(*status_flags);
1266
1267func_exit:
1268    *status_flags = cpu_to_le32(*status_flags);
1269}
1270
1271static inline void
1272e1000e_write_lgcy_rx_descr(E1000ECore *core, uint8_t *desc,
1273                           struct NetRxPkt *pkt,
1274                           const E1000E_RSSInfo *rss_info,
1275                           uint16_t length)
1276{
1277    uint32_t status_flags, rss, mrq;
1278    uint16_t ip_id;
1279
1280    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1281
1282    assert(!rss_info->enabled);
1283
1284    d->length = cpu_to_le16(length);
1285    d->csum = 0;
1286
1287    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1288                             rss_info,
1289                             &rss, &mrq,
1290                             &status_flags, &ip_id,
1291                             &d->special);
1292    d->errors = (uint8_t) (le32_to_cpu(status_flags) >> 24);
1293    d->status = (uint8_t) le32_to_cpu(status_flags);
1294    d->special = 0;
1295}
1296
1297static inline void
1298e1000e_write_ext_rx_descr(E1000ECore *core, uint8_t *desc,
1299                          struct NetRxPkt *pkt,
1300                          const E1000E_RSSInfo *rss_info,
1301                          uint16_t length)
1302{
1303    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1304
1305    memset(&d->wb, 0, sizeof(d->wb));
1306
1307    d->wb.upper.length = cpu_to_le16(length);
1308
1309    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1310                             rss_info,
1311                             &d->wb.lower.hi_dword.rss,
1312                             &d->wb.lower.mrq,
1313                             &d->wb.upper.status_error,
1314                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1315                             &d->wb.upper.vlan);
1316}
1317
1318static inline void
1319e1000e_write_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1320                         struct NetRxPkt *pkt,
1321                         const E1000E_RSSInfo *rss_info,
1322                         size_t ps_hdr_len,
1323                         uint16_t(*written)[MAX_PS_BUFFERS])
1324{
1325    int i;
1326    union e1000_rx_desc_packet_split *d =
1327        (union e1000_rx_desc_packet_split *) desc;
1328
1329    memset(&d->wb, 0, sizeof(d->wb));
1330
1331    d->wb.middle.length0 = cpu_to_le16((*written)[0]);
1332
1333    for (i = 0; i < PS_PAGE_BUFFERS; i++) {
1334        d->wb.upper.length[i] = cpu_to_le16((*written)[i + 1]);
1335    }
1336
1337    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1338                             rss_info,
1339                             &d->wb.lower.hi_dword.rss,
1340                             &d->wb.lower.mrq,
1341                             &d->wb.middle.status_error,
1342                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1343                             &d->wb.middle.vlan);
1344
1345    d->wb.upper.header_status =
1346        cpu_to_le16(ps_hdr_len | (ps_hdr_len ? E1000_RXDPS_HDRSTAT_HDRSP : 0));
1347
1348    trace_e1000e_rx_desc_ps_write((*written)[0], (*written)[1],
1349                                  (*written)[2], (*written)[3]);
1350}
1351
1352static inline void
1353e1000e_write_rx_descr(E1000ECore *core, uint8_t *desc,
1354struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info,
1355    size_t ps_hdr_len, uint16_t(*written)[MAX_PS_BUFFERS])
1356{
1357    if (e1000e_rx_use_legacy_descriptor(core)) {
1358        assert(ps_hdr_len == 0);
1359        e1000e_write_lgcy_rx_descr(core, desc, pkt, rss_info, (*written)[0]);
1360    } else {
1361        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1362            e1000e_write_ps_rx_descr(core, desc, pkt, rss_info,
1363                                      ps_hdr_len, written);
1364        } else {
1365            assert(ps_hdr_len == 0);
1366            e1000e_write_ext_rx_descr(core, desc, pkt, rss_info,
1367                                       (*written)[0]);
1368        }
1369    }
1370}
1371
1372typedef struct e1000e_ba_state_st {
1373    uint16_t written[MAX_PS_BUFFERS];
1374    uint8_t cur_idx;
1375} e1000e_ba_state;
1376
1377static inline void
1378e1000e_write_hdr_to_rx_buffers(E1000ECore *core,
1379                               hwaddr (*ba)[MAX_PS_BUFFERS],
1380                               e1000e_ba_state *bastate,
1381                               const char *data,
1382                               dma_addr_t data_len)
1383{
1384    assert(data_len <= core->rxbuf_sizes[0] - bastate->written[0]);
1385
1386    pci_dma_write(core->owner, (*ba)[0] + bastate->written[0], data, data_len);
1387    bastate->written[0] += data_len;
1388
1389    bastate->cur_idx = 1;
1390}
1391
1392static void
1393e1000e_write_to_rx_buffers(E1000ECore *core,
1394                           hwaddr (*ba)[MAX_PS_BUFFERS],
1395                           e1000e_ba_state *bastate,
1396                           const char *data,
1397                           dma_addr_t data_len)
1398{
1399    while (data_len > 0) {
1400        uint32_t cur_buf_len = core->rxbuf_sizes[bastate->cur_idx];
1401        uint32_t cur_buf_bytes_left = cur_buf_len -
1402                                      bastate->written[bastate->cur_idx];
1403        uint32_t bytes_to_write = MIN(data_len, cur_buf_bytes_left);
1404
1405        trace_e1000e_rx_desc_buff_write(bastate->cur_idx,
1406                                        (*ba)[bastate->cur_idx],
1407                                        bastate->written[bastate->cur_idx],
1408                                        data,
1409                                        bytes_to_write);
1410
1411        pci_dma_write(core->owner,
1412            (*ba)[bastate->cur_idx] + bastate->written[bastate->cur_idx],
1413            data, bytes_to_write);
1414
1415        bastate->written[bastate->cur_idx] += bytes_to_write;
1416        data += bytes_to_write;
1417        data_len -= bytes_to_write;
1418
1419        if (bastate->written[bastate->cur_idx] == cur_buf_len) {
1420            bastate->cur_idx++;
1421        }
1422
1423        assert(bastate->cur_idx < MAX_PS_BUFFERS);
1424    }
1425}
1426
1427static void
1428e1000e_update_rx_stats(E1000ECore *core,
1429                       size_t data_size,
1430                       size_t data_fcs_size)
1431{
1432    e1000x_update_rx_total_stats(core->mac, data_size, data_fcs_size);
1433
1434    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1435    case ETH_PKT_BCAST:
1436        e1000x_inc_reg_if_not_full(core->mac, BPRC);
1437        break;
1438
1439    case ETH_PKT_MCAST:
1440        e1000x_inc_reg_if_not_full(core->mac, MPRC);
1441        break;
1442
1443    default:
1444        break;
1445    }
1446}
1447
1448static inline bool
1449e1000e_rx_descr_threshold_hit(E1000ECore *core, const E1000E_RingInfo *rxi)
1450{
1451    return e1000e_ring_free_descr_num(core, rxi) ==
1452           e1000e_ring_len(core, rxi) >> core->rxbuf_min_shift;
1453}
1454
1455static bool
1456e1000e_do_ps(E1000ECore *core, struct NetRxPkt *pkt, size_t *hdr_len)
1457{
1458    bool isip4, isip6, isudp, istcp;
1459    bool fragment;
1460
1461    if (!e1000e_rx_use_ps_descriptor(core)) {
1462        return false;
1463    }
1464
1465    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1466
1467    if (isip4) {
1468        fragment = net_rx_pkt_get_ip4_info(pkt)->fragment;
1469    } else if (isip6) {
1470        fragment = net_rx_pkt_get_ip6_info(pkt)->fragment;
1471    } else {
1472        return false;
1473    }
1474
1475    if (fragment && (core->mac[RFCTL] & E1000_RFCTL_IPFRSP_DIS)) {
1476        return false;
1477    }
1478
1479    if (!fragment && (isudp || istcp)) {
1480        *hdr_len = net_rx_pkt_get_l5_hdr_offset(pkt);
1481    } else {
1482        *hdr_len = net_rx_pkt_get_l4_hdr_offset(pkt);
1483    }
1484
1485    if ((*hdr_len > core->rxbuf_sizes[0]) ||
1486        (*hdr_len > net_rx_pkt_get_total_len(pkt))) {
1487        return false;
1488    }
1489
1490    return true;
1491}
1492
1493static void
1494e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt,
1495                             const E1000E_RxRing *rxr,
1496                             const E1000E_RSSInfo *rss_info)
1497{
1498    PCIDevice *d = core->owner;
1499    dma_addr_t base;
1500    uint8_t desc[E1000_MAX_RX_DESC_LEN];
1501    size_t desc_size;
1502    size_t desc_offset = 0;
1503    size_t iov_ofs = 0;
1504
1505    struct iovec *iov = net_rx_pkt_get_iovec(pkt);
1506    size_t size = net_rx_pkt_get_total_len(pkt);
1507    size_t total_size = size + e1000x_fcs_len(core->mac);
1508    const E1000E_RingInfo *rxi;
1509    size_t ps_hdr_len = 0;
1510    bool do_ps = e1000e_do_ps(core, pkt, &ps_hdr_len);
1511    bool is_first = true;
1512
1513    rxi = rxr->i;
1514
1515    do {
1516        hwaddr ba[MAX_PS_BUFFERS];
1517        e1000e_ba_state bastate = { { 0 } };
1518        bool is_last = false;
1519
1520        desc_size = total_size - desc_offset;
1521
1522        if (desc_size > core->rx_desc_buf_size) {
1523            desc_size = core->rx_desc_buf_size;
1524        }
1525
1526        if (e1000e_ring_empty(core, rxi)) {
1527            return;
1528        }
1529
1530        base = e1000e_ring_head_descr(core, rxi);
1531
1532        pci_dma_read(d, base, &desc, core->rx_desc_len);
1533
1534        trace_e1000e_rx_descr(rxi->idx, base, core->rx_desc_len);
1535
1536        e1000e_read_rx_descr(core, desc, &ba);
1537
1538        if (ba[0]) {
1539            if (desc_offset < size) {
1540                static const uint32_t fcs_pad;
1541                size_t iov_copy;
1542                size_t copy_size = size - desc_offset;
1543                if (copy_size > core->rx_desc_buf_size) {
1544                    copy_size = core->rx_desc_buf_size;
1545                }
1546
1547                /* For PS mode copy the packet header first */
1548                if (do_ps) {
1549                    if (is_first) {
1550                        size_t ps_hdr_copied = 0;
1551                        do {
1552                            iov_copy = MIN(ps_hdr_len - ps_hdr_copied,
1553                                           iov->iov_len - iov_ofs);
1554
1555                            e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1556                                                      iov->iov_base, iov_copy);
1557
1558                            copy_size -= iov_copy;
1559                            ps_hdr_copied += iov_copy;
1560
1561                            iov_ofs += iov_copy;
1562                            if (iov_ofs == iov->iov_len) {
1563                                iov++;
1564                                iov_ofs = 0;
1565                            }
1566                        } while (ps_hdr_copied < ps_hdr_len);
1567
1568                        is_first = false;
1569                    } else {
1570                        /* Leave buffer 0 of each descriptor except first */
1571                        /* empty as per spec 7.1.5.1                      */
1572                        e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1573                                                       NULL, 0);
1574                    }
1575                }
1576
1577                /* Copy packet payload */
1578                while (copy_size) {
1579                    iov_copy = MIN(copy_size, iov->iov_len - iov_ofs);
1580
1581                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1582                                            iov->iov_base + iov_ofs, iov_copy);
1583
1584                    copy_size -= iov_copy;
1585                    iov_ofs += iov_copy;
1586                    if (iov_ofs == iov->iov_len) {
1587                        iov++;
1588                        iov_ofs = 0;
1589                    }
1590                }
1591
1592                if (desc_offset + desc_size >= total_size) {
1593                    /* Simulate FCS checksum presence in the last descriptor */
1594                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1595                          (const char *) &fcs_pad, e1000x_fcs_len(core->mac));
1596                }
1597            }
1598            desc_offset += desc_size;
1599            if (desc_offset >= total_size) {
1600                is_last = true;
1601            }
1602        } else { /* as per intel docs; skip descriptors with null buf addr */
1603            trace_e1000e_rx_null_descriptor();
1604        }
1605
1606        e1000e_write_rx_descr(core, desc, is_last ? core->rx_pkt : NULL,
1607                           rss_info, do_ps ? ps_hdr_len : 0, &bastate.written);
1608        pci_dma_write(d, base, &desc, core->rx_desc_len);
1609
1610        e1000e_ring_advance(core, rxi,
1611                            core->rx_desc_len / E1000_MIN_RX_DESC_LEN);
1612
1613    } while (desc_offset < total_size);
1614
1615    e1000e_update_rx_stats(core, size, total_size);
1616}
1617
1618static inline void
1619e1000e_rx_fix_l4_csum(E1000ECore *core, struct NetRxPkt *pkt)
1620{
1621    if (net_rx_pkt_has_virt_hdr(pkt)) {
1622        struct virtio_net_hdr *vhdr = net_rx_pkt_get_vhdr(pkt);
1623
1624        if (vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1625            net_rx_pkt_fix_l4_csum(pkt);
1626        }
1627    }
1628}
1629
1630ssize_t
1631e1000e_receive_iov(E1000ECore *core, const struct iovec *iov, int iovcnt)
1632{
1633    static const int maximum_ethernet_hdr_len = (14 + 4);
1634    /* Min. octets in an ethernet frame sans FCS */
1635    static const int min_buf_size = 60;
1636
1637    uint32_t n = 0;
1638    uint8_t min_buf[min_buf_size];
1639    struct iovec min_iov;
1640    uint8_t *filter_buf;
1641    size_t size, orig_size;
1642    size_t iov_ofs = 0;
1643    E1000E_RxRing rxr;
1644    E1000E_RSSInfo rss_info;
1645    size_t total_size;
1646    ssize_t retval;
1647    bool rdmts_hit;
1648
1649    trace_e1000e_rx_receive_iov(iovcnt);
1650
1651    if (!e1000x_hw_rx_enabled(core->mac)) {
1652        return -1;
1653    }
1654
1655    /* Pull virtio header in */
1656    if (core->has_vnet) {
1657        net_rx_pkt_set_vhdr_iovec(core->rx_pkt, iov, iovcnt);
1658        iov_ofs = sizeof(struct virtio_net_hdr);
1659    }
1660
1661    filter_buf = iov->iov_base + iov_ofs;
1662    orig_size = iov_size(iov, iovcnt);
1663    size = orig_size - iov_ofs;
1664
1665    /* Pad to minimum Ethernet frame length */
1666    if (size < sizeof(min_buf)) {
1667        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, size);
1668        memset(&min_buf[size], 0, sizeof(min_buf) - size);
1669        e1000x_inc_reg_if_not_full(core->mac, RUC);
1670        min_iov.iov_base = filter_buf = min_buf;
1671        min_iov.iov_len = size = sizeof(min_buf);
1672        iovcnt = 1;
1673        iov = &min_iov;
1674        iov_ofs = 0;
1675    } else if (iov->iov_len < maximum_ethernet_hdr_len) {
1676        /* This is very unlikely, but may happen. */
1677        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, maximum_ethernet_hdr_len);
1678        filter_buf = min_buf;
1679    }
1680
1681    /* Discard oversized packets if !LPE and !SBP. */
1682    if (e1000x_is_oversized(core->mac, size)) {
1683        return orig_size;
1684    }
1685
1686    net_rx_pkt_set_packet_type(core->rx_pkt,
1687        get_eth_packet_type(PKT_GET_ETH_HDR(filter_buf)));
1688
1689    if (!e1000e_receive_filter(core, filter_buf, size)) {
1690        trace_e1000e_rx_flt_dropped();
1691        return orig_size;
1692    }
1693
1694    net_rx_pkt_attach_iovec_ex(core->rx_pkt, iov, iovcnt, iov_ofs,
1695                               e1000x_vlan_enabled(core->mac), core->vet);
1696
1697    e1000e_rss_parse_packet(core, core->rx_pkt, &rss_info);
1698    e1000e_rx_ring_init(core, &rxr, rss_info.queue);
1699
1700    trace_e1000e_rx_rss_dispatched_to_queue(rxr.i->idx);
1701
1702    total_size = net_rx_pkt_get_total_len(core->rx_pkt) +
1703        e1000x_fcs_len(core->mac);
1704
1705    if (e1000e_has_rxbufs(core, rxr.i, total_size)) {
1706        e1000e_rx_fix_l4_csum(core, core->rx_pkt);
1707
1708        e1000e_write_packet_to_guest(core, core->rx_pkt, &rxr, &rss_info);
1709
1710        retval = orig_size;
1711
1712        /* Perform small receive detection (RSRPD) */
1713        if (total_size < core->mac[RSRPD]) {
1714            n |= E1000_ICS_SRPD;
1715        }
1716
1717        /* Perform ACK receive detection */
1718        if  (!(core->mac[RFCTL] & E1000_RFCTL_ACK_DIS) &&
1719             (e1000e_is_tcp_ack(core, core->rx_pkt))) {
1720            n |= E1000_ICS_ACK;
1721        }
1722
1723        /* Check if receive descriptor minimum threshold hit */
1724        rdmts_hit = e1000e_rx_descr_threshold_hit(core, rxr.i);
1725        n |= e1000e_rx_wb_interrupt_cause(core, rxr.i->idx, rdmts_hit);
1726
1727        trace_e1000e_rx_written_to_guest(n);
1728    } else {
1729        n |= E1000_ICS_RXO;
1730        retval = 0;
1731
1732        trace_e1000e_rx_not_written_to_guest(n);
1733    }
1734
1735    if (!e1000e_intrmgr_delay_rx_causes(core, &n)) {
1736        trace_e1000e_rx_interrupt_set(n);
1737        e1000e_set_interrupt_cause(core, n);
1738    } else {
1739        trace_e1000e_rx_interrupt_delayed(n);
1740    }
1741
1742    return retval;
1743}
1744
1745static inline bool
1746e1000e_have_autoneg(E1000ECore *core)
1747{
1748    return core->phy[0][PHY_CTRL] & MII_CR_AUTO_NEG_EN;
1749}
1750
1751static void e1000e_update_flowctl_status(E1000ECore *core)
1752{
1753    if (e1000e_have_autoneg(core) &&
1754        core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE) {
1755        trace_e1000e_link_autoneg_flowctl(true);
1756        core->mac[CTRL] |= E1000_CTRL_TFCE | E1000_CTRL_RFCE;
1757    } else {
1758        trace_e1000e_link_autoneg_flowctl(false);
1759    }
1760}
1761
1762static inline void
1763e1000e_link_down(E1000ECore *core)
1764{
1765    e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1766    e1000e_update_flowctl_status(core);
1767}
1768
1769static inline void
1770e1000e_set_phy_ctrl(E1000ECore *core, int index, uint16_t val)
1771{
1772    /* bits 0-5 reserved; MII_CR_[RESTART_AUTO_NEG,RESET] are self clearing */
1773    core->phy[0][PHY_CTRL] = val & ~(0x3f |
1774                                     MII_CR_RESET |
1775                                     MII_CR_RESTART_AUTO_NEG);
1776
1777    if ((val & MII_CR_RESTART_AUTO_NEG) &&
1778        e1000e_have_autoneg(core)) {
1779        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1780    }
1781}
1782
1783static void
1784e1000e_set_phy_oem_bits(E1000ECore *core, int index, uint16_t val)
1785{
1786    core->phy[0][PHY_OEM_BITS] = val & ~BIT(10);
1787
1788    if (val & BIT(10)) {
1789        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1790    }
1791}
1792
1793static void
1794e1000e_set_phy_page(E1000ECore *core, int index, uint16_t val)
1795{
1796    core->phy[0][PHY_PAGE] = val & PHY_PAGE_RW_MASK;
1797}
1798
1799void
1800e1000e_core_set_link_status(E1000ECore *core)
1801{
1802    NetClientState *nc = qemu_get_queue(core->owner_nic);
1803    uint32_t old_status = core->mac[STATUS];
1804
1805    trace_e1000e_link_status_changed(nc->link_down ? false : true);
1806
1807    if (nc->link_down) {
1808        e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1809    } else {
1810        if (e1000e_have_autoneg(core) &&
1811            !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
1812            e1000x_restart_autoneg(core->mac, core->phy[0],
1813                                   core->autoneg_timer);
1814        } else {
1815            e1000x_update_regs_on_link_up(core->mac, core->phy[0]);
1816            e1000e_start_recv(core);
1817        }
1818    }
1819
1820    if (core->mac[STATUS] != old_status) {
1821        e1000e_set_interrupt_cause(core, E1000_ICR_LSC);
1822    }
1823}
1824
1825static void
1826e1000e_set_ctrl(E1000ECore *core, int index, uint32_t val)
1827{
1828    trace_e1000e_core_ctrl_write(index, val);
1829
1830    /* RST is self clearing */
1831    core->mac[CTRL] = val & ~E1000_CTRL_RST;
1832    core->mac[CTRL_DUP] = core->mac[CTRL];
1833
1834    trace_e1000e_link_set_params(
1835        !!(val & E1000_CTRL_ASDE),
1836        (val & E1000_CTRL_SPD_SEL) >> E1000_CTRL_SPD_SHIFT,
1837        !!(val & E1000_CTRL_FRCSPD),
1838        !!(val & E1000_CTRL_FRCDPX),
1839        !!(val & E1000_CTRL_RFCE),
1840        !!(val & E1000_CTRL_TFCE));
1841
1842    if (val & E1000_CTRL_RST) {
1843        trace_e1000e_core_ctrl_sw_reset();
1844        e1000x_reset_mac_addr(core->owner_nic, core->mac, core->permanent_mac);
1845    }
1846
1847    if (val & E1000_CTRL_PHY_RST) {
1848        trace_e1000e_core_ctrl_phy_reset();
1849        core->mac[STATUS] |= E1000_STATUS_PHYRA;
1850    }
1851}
1852
1853static void
1854e1000e_set_rfctl(E1000ECore *core, int index, uint32_t val)
1855{
1856    trace_e1000e_rx_set_rfctl(val);
1857
1858    if (!(val & E1000_RFCTL_ISCSI_DIS)) {
1859        trace_e1000e_wrn_iscsi_filtering_not_supported();
1860    }
1861
1862    if (!(val & E1000_RFCTL_NFSW_DIS)) {
1863        trace_e1000e_wrn_nfsw_filtering_not_supported();
1864    }
1865
1866    if (!(val & E1000_RFCTL_NFSR_DIS)) {
1867        trace_e1000e_wrn_nfsr_filtering_not_supported();
1868    }
1869
1870    core->mac[RFCTL] = val;
1871}
1872
1873static void
1874e1000e_calc_per_desc_buf_size(E1000ECore *core)
1875{
1876    int i;
1877    core->rx_desc_buf_size = 0;
1878
1879    for (i = 0; i < ARRAY_SIZE(core->rxbuf_sizes); i++) {
1880        core->rx_desc_buf_size += core->rxbuf_sizes[i];
1881    }
1882}
1883
1884static void
1885e1000e_parse_rxbufsize(E1000ECore *core)
1886{
1887    uint32_t rctl = core->mac[RCTL];
1888
1889    memset(core->rxbuf_sizes, 0, sizeof(core->rxbuf_sizes));
1890
1891    if (rctl & E1000_RCTL_DTYP_MASK) {
1892        uint32_t bsize;
1893
1894        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE0_MASK;
1895        core->rxbuf_sizes[0] = (bsize >> E1000_PSRCTL_BSIZE0_SHIFT) * 128;
1896
1897        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE1_MASK;
1898        core->rxbuf_sizes[1] = (bsize >> E1000_PSRCTL_BSIZE1_SHIFT) * 1024;
1899
1900        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE2_MASK;
1901        core->rxbuf_sizes[2] = (bsize >> E1000_PSRCTL_BSIZE2_SHIFT) * 1024;
1902
1903        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE3_MASK;
1904        core->rxbuf_sizes[3] = (bsize >> E1000_PSRCTL_BSIZE3_SHIFT) * 1024;
1905    } else if (rctl & E1000_RCTL_FLXBUF_MASK) {
1906        int flxbuf = rctl & E1000_RCTL_FLXBUF_MASK;
1907        core->rxbuf_sizes[0] = (flxbuf >> E1000_RCTL_FLXBUF_SHIFT) * 1024;
1908    } else {
1909        core->rxbuf_sizes[0] = e1000x_rxbufsize(rctl);
1910    }
1911
1912    trace_e1000e_rx_desc_buff_sizes(core->rxbuf_sizes[0], core->rxbuf_sizes[1],
1913                                    core->rxbuf_sizes[2], core->rxbuf_sizes[3]);
1914
1915    e1000e_calc_per_desc_buf_size(core);
1916}
1917
1918static void
1919e1000e_calc_rxdesclen(E1000ECore *core)
1920{
1921    if (e1000e_rx_use_legacy_descriptor(core)) {
1922        core->rx_desc_len = sizeof(struct e1000_rx_desc);
1923    } else {
1924        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1925            core->rx_desc_len = sizeof(union e1000_rx_desc_packet_split);
1926        } else {
1927            core->rx_desc_len = sizeof(union e1000_rx_desc_extended);
1928        }
1929    }
1930    trace_e1000e_rx_desc_len(core->rx_desc_len);
1931}
1932
1933static void
1934e1000e_set_rx_control(E1000ECore *core, int index, uint32_t val)
1935{
1936    core->mac[RCTL] = val;
1937    trace_e1000e_rx_set_rctl(core->mac[RCTL]);
1938
1939    if (val & E1000_RCTL_EN) {
1940        e1000e_parse_rxbufsize(core);
1941        e1000e_calc_rxdesclen(core);
1942        core->rxbuf_min_shift = ((val / E1000_RCTL_RDMTS_QUAT) & 3) + 1 +
1943                                E1000_RING_DESC_LEN_SHIFT;
1944
1945        e1000e_start_recv(core);
1946    }
1947}
1948
1949static
1950void(*e1000e_phyreg_writeops[E1000E_PHY_PAGES][E1000E_PHY_PAGE_SIZE])
1951(E1000ECore *, int, uint16_t) = {
1952    [0] = {
1953        [PHY_CTRL]     = e1000e_set_phy_ctrl,
1954        [PHY_PAGE]     = e1000e_set_phy_page,
1955        [PHY_OEM_BITS] = e1000e_set_phy_oem_bits
1956    }
1957};
1958
1959static inline void
1960e1000e_clear_ims_bits(E1000ECore *core, uint32_t bits)
1961{
1962    trace_e1000e_irq_clear_ims(bits, core->mac[IMS], core->mac[IMS] & ~bits);
1963    core->mac[IMS] &= ~bits;
1964}
1965
1966static inline bool
1967e1000e_postpone_interrupt(bool *interrupt_pending,
1968                           E1000IntrDelayTimer *timer)
1969{
1970    if (timer->running) {
1971        trace_e1000e_irq_postponed_by_xitr(timer->delay_reg << 2);
1972
1973        *interrupt_pending = true;
1974        return true;
1975    }
1976
1977    if (timer->core->mac[timer->delay_reg] != 0) {
1978        e1000e_intrmgr_rearm_timer(timer);
1979    }
1980
1981    return false;
1982}
1983
1984static inline bool
1985e1000e_itr_should_postpone(E1000ECore *core)
1986{
1987    return e1000e_postpone_interrupt(&core->itr_intr_pending, &core->itr);
1988}
1989
1990static inline bool
1991e1000e_eitr_should_postpone(E1000ECore *core, int idx)
1992{
1993    return e1000e_postpone_interrupt(&core->eitr_intr_pending[idx],
1994                                     &core->eitr[idx]);
1995}
1996
1997static void
1998e1000e_msix_notify_one(E1000ECore *core, uint32_t cause, uint32_t int_cfg)
1999{
2000    uint32_t effective_eiac;
2001
2002    if (E1000_IVAR_ENTRY_VALID(int_cfg)) {
2003        uint32_t vec = E1000_IVAR_ENTRY_VEC(int_cfg);
2004        if (vec < E1000E_MSIX_VEC_NUM) {
2005            if (!e1000e_eitr_should_postpone(core, vec)) {
2006                trace_e1000e_irq_msix_notify_vec(vec);
2007                msix_notify(core->owner, vec);
2008            }
2009        } else {
2010            trace_e1000e_wrn_msix_vec_wrong(cause, int_cfg);
2011        }
2012    } else {
2013        trace_e1000e_wrn_msix_invalid(cause, int_cfg);
2014    }
2015
2016    if (core->mac[CTRL_EXT] & E1000_CTRL_EXT_EIAME) {
2017        trace_e1000e_irq_iam_clear_eiame(core->mac[IAM], cause);
2018        core->mac[IAM] &= ~cause;
2019    }
2020
2021    trace_e1000e_irq_icr_clear_eiac(core->mac[ICR], core->mac[EIAC]);
2022
2023    effective_eiac = core->mac[EIAC] & cause;
2024
2025    if (effective_eiac == E1000_ICR_OTHER) {
2026        effective_eiac |= E1000_ICR_OTHER_CAUSES;
2027    }
2028
2029    core->mac[ICR] &= ~effective_eiac;
2030
2031    if (!(core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2032        core->mac[IMS] &= ~effective_eiac;
2033    }
2034}
2035
2036static void
2037e1000e_msix_notify(E1000ECore *core, uint32_t causes)
2038{
2039    if (causes & E1000_ICR_RXQ0) {
2040        e1000e_msix_notify_one(core, E1000_ICR_RXQ0,
2041                               E1000_IVAR_RXQ0(core->mac[IVAR]));
2042    }
2043
2044    if (causes & E1000_ICR_RXQ1) {
2045        e1000e_msix_notify_one(core, E1000_ICR_RXQ1,
2046                               E1000_IVAR_RXQ1(core->mac[IVAR]));
2047    }
2048
2049    if (causes & E1000_ICR_TXQ0) {
2050        e1000e_msix_notify_one(core, E1000_ICR_TXQ0,
2051                               E1000_IVAR_TXQ0(core->mac[IVAR]));
2052    }
2053
2054    if (causes & E1000_ICR_TXQ1) {
2055        e1000e_msix_notify_one(core, E1000_ICR_TXQ1,
2056                               E1000_IVAR_TXQ1(core->mac[IVAR]));
2057    }
2058
2059    if (causes & E1000_ICR_OTHER) {
2060        e1000e_msix_notify_one(core, E1000_ICR_OTHER,
2061                               E1000_IVAR_OTHER(core->mac[IVAR]));
2062    }
2063}
2064
2065static void
2066e1000e_msix_clear_one(E1000ECore *core, uint32_t cause, uint32_t int_cfg)
2067{
2068    if (E1000_IVAR_ENTRY_VALID(int_cfg)) {
2069        uint32_t vec = E1000_IVAR_ENTRY_VEC(int_cfg);
2070        if (vec < E1000E_MSIX_VEC_NUM) {
2071            trace_e1000e_irq_msix_pending_clearing(cause, int_cfg, vec);
2072            msix_clr_pending(core->owner, vec);
2073        } else {
2074            trace_e1000e_wrn_msix_vec_wrong(cause, int_cfg);
2075        }
2076    } else {
2077        trace_e1000e_wrn_msix_invalid(cause, int_cfg);
2078    }
2079}
2080
2081static void
2082e1000e_msix_clear(E1000ECore *core, uint32_t causes)
2083{
2084    if (causes & E1000_ICR_RXQ0) {
2085        e1000e_msix_clear_one(core, E1000_ICR_RXQ0,
2086                              E1000_IVAR_RXQ0(core->mac[IVAR]));
2087    }
2088
2089    if (causes & E1000_ICR_RXQ1) {
2090        e1000e_msix_clear_one(core, E1000_ICR_RXQ1,
2091                              E1000_IVAR_RXQ1(core->mac[IVAR]));
2092    }
2093
2094    if (causes & E1000_ICR_TXQ0) {
2095        e1000e_msix_clear_one(core, E1000_ICR_TXQ0,
2096                              E1000_IVAR_TXQ0(core->mac[IVAR]));
2097    }
2098
2099    if (causes & E1000_ICR_TXQ1) {
2100        e1000e_msix_clear_one(core, E1000_ICR_TXQ1,
2101                              E1000_IVAR_TXQ1(core->mac[IVAR]));
2102    }
2103
2104    if (causes & E1000_ICR_OTHER) {
2105        e1000e_msix_clear_one(core, E1000_ICR_OTHER,
2106                              E1000_IVAR_OTHER(core->mac[IVAR]));
2107    }
2108}
2109
2110static inline void
2111e1000e_fix_icr_asserted(E1000ECore *core)
2112{
2113    core->mac[ICR] &= ~E1000_ICR_ASSERTED;
2114    if (core->mac[ICR]) {
2115        core->mac[ICR] |= E1000_ICR_ASSERTED;
2116    }
2117
2118    trace_e1000e_irq_fix_icr_asserted(core->mac[ICR]);
2119}
2120
2121static void
2122e1000e_send_msi(E1000ECore *core, bool msix)
2123{
2124    uint32_t causes = core->mac[ICR] & core->mac[IMS] & ~E1000_ICR_ASSERTED;
2125
2126    if (msix) {
2127        e1000e_msix_notify(core, causes);
2128    } else {
2129        if (!e1000e_itr_should_postpone(core)) {
2130            trace_e1000e_irq_msi_notify(causes);
2131            msi_notify(core->owner, 0);
2132        }
2133    }
2134}
2135
2136static void
2137e1000e_update_interrupt_state(E1000ECore *core)
2138{
2139    bool interrupts_pending;
2140    bool is_msix = msix_enabled(core->owner);
2141
2142    /* Set ICR[OTHER] for MSI-X */
2143    if (is_msix) {
2144        if (core->mac[ICR] & E1000_ICR_OTHER_CAUSES) {
2145            core->mac[ICR] |= E1000_ICR_OTHER;
2146            trace_e1000e_irq_add_msi_other(core->mac[ICR]);
2147        }
2148    }
2149
2150    e1000e_fix_icr_asserted(core);
2151
2152    /*
2153     * Make sure ICR and ICS registers have the same value.
2154     * The spec says that the ICS register is write-only.  However in practice,
2155     * on real hardware ICS is readable, and for reads it has the same value as
2156     * ICR (except that ICS does not have the clear on read behaviour of ICR).
2157     *
2158     * The VxWorks PRO/1000 driver uses this behaviour.
2159     */
2160    core->mac[ICS] = core->mac[ICR];
2161
2162    interrupts_pending = (core->mac[IMS] & core->mac[ICR]) ? true : false;
2163
2164    trace_e1000e_irq_pending_interrupts(core->mac[ICR] & core->mac[IMS],
2165                                        core->mac[ICR], core->mac[IMS]);
2166
2167    if (is_msix || msi_enabled(core->owner)) {
2168        if (interrupts_pending) {
2169            e1000e_send_msi(core, is_msix);
2170        }
2171    } else {
2172        if (interrupts_pending) {
2173            if (!e1000e_itr_should_postpone(core)) {
2174                e1000e_raise_legacy_irq(core);
2175            }
2176        } else {
2177            e1000e_lower_legacy_irq(core);
2178        }
2179    }
2180}
2181
2182static void
2183e1000e_set_interrupt_cause(E1000ECore *core, uint32_t val)
2184{
2185    trace_e1000e_irq_set_cause_entry(val, core->mac[ICR]);
2186
2187    val |= e1000e_intmgr_collect_delayed_causes(core);
2188    core->mac[ICR] |= val;
2189
2190    trace_e1000e_irq_set_cause_exit(val, core->mac[ICR]);
2191
2192    e1000e_update_interrupt_state(core);
2193}
2194
2195static inline void
2196e1000e_autoneg_timer(void *opaque)
2197{
2198    E1000ECore *core = opaque;
2199    if (!qemu_get_queue(core->owner_nic)->link_down) {
2200        e1000x_update_regs_on_autoneg_done(core->mac, core->phy[0]);
2201        e1000e_start_recv(core);
2202
2203        e1000e_update_flowctl_status(core);
2204        /* signal link status change to the guest */
2205        e1000e_set_interrupt_cause(core, E1000_ICR_LSC);
2206    }
2207}
2208
2209static inline uint16_t
2210e1000e_get_reg_index_with_offset(const uint16_t *mac_reg_access, hwaddr addr)
2211{
2212    uint16_t index = (addr & 0x1ffff) >> 2;
2213    return index + (mac_reg_access[index] & 0xfffe);
2214}
2215
2216static const char e1000e_phy_regcap[E1000E_PHY_PAGES][0x20] = {
2217    [0] = {
2218        [PHY_CTRL]          = PHY_ANYPAGE | PHY_RW,
2219        [PHY_STATUS]        = PHY_ANYPAGE | PHY_R,
2220        [PHY_ID1]           = PHY_ANYPAGE | PHY_R,
2221        [PHY_ID2]           = PHY_ANYPAGE | PHY_R,
2222        [PHY_AUTONEG_ADV]   = PHY_ANYPAGE | PHY_RW,
2223        [PHY_LP_ABILITY]    = PHY_ANYPAGE | PHY_R,
2224        [PHY_AUTONEG_EXP]   = PHY_ANYPAGE | PHY_R,
2225        [PHY_NEXT_PAGE_TX]  = PHY_ANYPAGE | PHY_RW,
2226        [PHY_LP_NEXT_PAGE]  = PHY_ANYPAGE | PHY_R,
2227        [PHY_1000T_CTRL]    = PHY_ANYPAGE | PHY_RW,
2228        [PHY_1000T_STATUS]  = PHY_ANYPAGE | PHY_R,
2229        [PHY_EXT_STATUS]    = PHY_ANYPAGE | PHY_R,
2230        [PHY_PAGE]          = PHY_ANYPAGE | PHY_RW,
2231
2232        [PHY_COPPER_CTRL1]      = PHY_RW,
2233        [PHY_COPPER_STAT1]      = PHY_R,
2234        [PHY_COPPER_CTRL3]      = PHY_RW,
2235        [PHY_RX_ERR_CNTR]       = PHY_R,
2236        [PHY_OEM_BITS]          = PHY_RW,
2237        [PHY_BIAS_1]            = PHY_RW,
2238        [PHY_BIAS_2]            = PHY_RW,
2239        [PHY_COPPER_INT_ENABLE] = PHY_RW,
2240        [PHY_COPPER_STAT2]      = PHY_R,
2241        [PHY_COPPER_CTRL2]      = PHY_RW
2242    },
2243    [2] = {
2244        [PHY_MAC_CTRL1]         = PHY_RW,
2245        [PHY_MAC_INT_ENABLE]    = PHY_RW,
2246        [PHY_MAC_STAT]          = PHY_R,
2247        [PHY_MAC_CTRL2]         = PHY_RW
2248    },
2249    [3] = {
2250        [PHY_LED_03_FUNC_CTRL1] = PHY_RW,
2251        [PHY_LED_03_POL_CTRL]   = PHY_RW,
2252        [PHY_LED_TIMER_CTRL]    = PHY_RW,
2253        [PHY_LED_45_CTRL]       = PHY_RW
2254    },
2255    [5] = {
2256        [PHY_1000T_SKEW]        = PHY_R,
2257        [PHY_1000T_SWAP]        = PHY_R
2258    },
2259    [6] = {
2260        [PHY_CRC_COUNTERS]      = PHY_R
2261    }
2262};
2263
2264static bool
2265e1000e_phy_reg_check_cap(E1000ECore *core, uint32_t addr,
2266                         char cap, uint8_t *page)
2267{
2268    *page =
2269        (e1000e_phy_regcap[0][addr] & PHY_ANYPAGE) ? 0
2270                                                    : core->phy[0][PHY_PAGE];
2271
2272    if (*page >= E1000E_PHY_PAGES) {
2273        return false;
2274    }
2275
2276    return e1000e_phy_regcap[*page][addr] & cap;
2277}
2278
2279static void
2280e1000e_phy_reg_write(E1000ECore *core, uint8_t page,
2281                     uint32_t addr, uint16_t data)
2282{
2283    assert(page < E1000E_PHY_PAGES);
2284    assert(addr < E1000E_PHY_PAGE_SIZE);
2285
2286    if (e1000e_phyreg_writeops[page][addr]) {
2287        e1000e_phyreg_writeops[page][addr](core, addr, data);
2288    } else {
2289        core->phy[page][addr] = data;
2290    }
2291}
2292
2293static void
2294e1000e_set_mdic(E1000ECore *core, int index, uint32_t val)
2295{
2296    uint32_t data = val & E1000_MDIC_DATA_MASK;
2297    uint32_t addr = ((val & E1000_MDIC_REG_MASK) >> E1000_MDIC_REG_SHIFT);
2298    uint8_t page;
2299
2300    if ((val & E1000_MDIC_PHY_MASK) >> E1000_MDIC_PHY_SHIFT != 1) { /* phy # */
2301        val = core->mac[MDIC] | E1000_MDIC_ERROR;
2302    } else if (val & E1000_MDIC_OP_READ) {
2303        if (!e1000e_phy_reg_check_cap(core, addr, PHY_R, &page)) {
2304            trace_e1000e_core_mdic_read_unhandled(page, addr);
2305            val |= E1000_MDIC_ERROR;
2306        } else {
2307            val = (val ^ data) | core->phy[page][addr];
2308            trace_e1000e_core_mdic_read(page, addr, val);
2309        }
2310    } else if (val & E1000_MDIC_OP_WRITE) {
2311        if (!e1000e_phy_reg_check_cap(core, addr, PHY_W, &page)) {
2312            trace_e1000e_core_mdic_write_unhandled(page, addr);
2313            val |= E1000_MDIC_ERROR;
2314        } else {
2315            trace_e1000e_core_mdic_write(page, addr, data);
2316            e1000e_phy_reg_write(core, page, addr, data);
2317        }
2318    }
2319    core->mac[MDIC] = val | E1000_MDIC_READY;
2320
2321    if (val & E1000_MDIC_INT_EN) {
2322        e1000e_set_interrupt_cause(core, E1000_ICR_MDAC);
2323    }
2324}
2325
2326static void
2327e1000e_set_rdt(E1000ECore *core, int index, uint32_t val)
2328{
2329    core->mac[index] = val & 0xffff;
2330    trace_e1000e_rx_set_rdt(e1000e_mq_queue_idx(RDT0, index), val);
2331    e1000e_start_recv(core);
2332}
2333
2334static void
2335e1000e_set_status(E1000ECore *core, int index, uint32_t val)
2336{
2337    if ((val & E1000_STATUS_PHYRA) == 0) {
2338        core->mac[index] &= ~E1000_STATUS_PHYRA;
2339    }
2340}
2341
2342static void
2343e1000e_set_ctrlext(E1000ECore *core, int index, uint32_t val)
2344{
2345    trace_e1000e_link_set_ext_params(!!(val & E1000_CTRL_EXT_ASDCHK),
2346                                     !!(val & E1000_CTRL_EXT_SPD_BYPS));
2347
2348    /* Zero self-clearing bits */
2349    val &= ~(E1000_CTRL_EXT_ASDCHK | E1000_CTRL_EXT_EE_RST);
2350    core->mac[CTRL_EXT] = val;
2351}
2352
2353static void
2354e1000e_set_pbaclr(E1000ECore *core, int index, uint32_t val)
2355{
2356    int i;
2357
2358    core->mac[PBACLR] = val & E1000_PBACLR_VALID_MASK;
2359
2360    if (!msix_enabled(core->owner)) {
2361        return;
2362    }
2363
2364    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
2365        if (core->mac[PBACLR] & BIT(i)) {
2366            msix_clr_pending(core->owner, i);
2367        }
2368    }
2369}
2370
2371static void
2372e1000e_set_fcrth(E1000ECore *core, int index, uint32_t val)
2373{
2374    core->mac[FCRTH] = val & 0xFFF8;
2375}
2376
2377static void
2378e1000e_set_fcrtl(E1000ECore *core, int index, uint32_t val)
2379{
2380    core->mac[FCRTL] = val & 0x8000FFF8;
2381}
2382
2383static inline void
2384e1000e_set_16bit(E1000ECore *core, int index, uint32_t val)
2385{
2386    core->mac[index] = val & 0xffff;
2387}
2388
2389static void
2390e1000e_set_12bit(E1000ECore *core, int index, uint32_t val)
2391{
2392    core->mac[index] = val & 0xfff;
2393}
2394
2395static void
2396e1000e_set_vet(E1000ECore *core, int index, uint32_t val)
2397{
2398    core->mac[VET] = val & 0xffff;
2399    core->vet = le16_to_cpu(core->mac[VET]);
2400    trace_e1000e_vlan_vet(core->vet);
2401}
2402
2403static void
2404e1000e_set_dlen(E1000ECore *core, int index, uint32_t val)
2405{
2406    core->mac[index] = val & E1000_XDLEN_MASK;
2407}
2408
2409static void
2410e1000e_set_dbal(E1000ECore *core, int index, uint32_t val)
2411{
2412    core->mac[index] = val & E1000_XDBAL_MASK;
2413}
2414
2415static void
2416e1000e_set_tctl(E1000ECore *core, int index, uint32_t val)
2417{
2418    E1000E_TxRing txr;
2419    core->mac[index] = val;
2420
2421    if (core->mac[TARC0] & E1000_TARC_ENABLE) {
2422        e1000e_tx_ring_init(core, &txr, 0);
2423        e1000e_start_xmit(core, &txr);
2424    }
2425
2426    if (core->mac[TARC1] & E1000_TARC_ENABLE) {
2427        e1000e_tx_ring_init(core, &txr, 1);
2428        e1000e_start_xmit(core, &txr);
2429    }
2430}
2431
2432static void
2433e1000e_set_tdt(E1000ECore *core, int index, uint32_t val)
2434{
2435    E1000E_TxRing txr;
2436    int qidx = e1000e_mq_queue_idx(TDT, index);
2437    uint32_t tarc_reg = (qidx == 0) ? TARC0 : TARC1;
2438
2439    core->mac[index] = val & 0xffff;
2440
2441    if (core->mac[tarc_reg] & E1000_TARC_ENABLE) {
2442        e1000e_tx_ring_init(core, &txr, qidx);
2443        e1000e_start_xmit(core, &txr);
2444    }
2445}
2446
2447static void
2448e1000e_set_ics(E1000ECore *core, int index, uint32_t val)
2449{
2450    trace_e1000e_irq_write_ics(val);
2451    e1000e_set_interrupt_cause(core, val);
2452}
2453
2454static void
2455e1000e_set_icr(E1000ECore *core, int index, uint32_t val)
2456{
2457    uint32_t icr = 0;
2458    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2459        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2460        trace_e1000e_irq_icr_process_iame();
2461        e1000e_clear_ims_bits(core, core->mac[IAM]);
2462    }
2463
2464    icr = core->mac[ICR] & ~val;
2465    /* Windows driver expects that the "receive overrun" bit and other
2466     * ones to be cleared when the "Other" bit (#24) is cleared.
2467     */
2468    icr = (val & E1000_ICR_OTHER) ? (icr & ~E1000_ICR_OTHER_CAUSES) : icr;
2469    trace_e1000e_irq_icr_write(val, core->mac[ICR], icr);
2470    core->mac[ICR] = icr;
2471    e1000e_update_interrupt_state(core);
2472}
2473
2474static void
2475e1000e_set_imc(E1000ECore *core, int index, uint32_t val)
2476{
2477    trace_e1000e_irq_ims_clear_set_imc(val);
2478    e1000e_clear_ims_bits(core, val);
2479    e1000e_update_interrupt_state(core);
2480}
2481
2482static void
2483e1000e_set_ims(E1000ECore *core, int index, uint32_t val)
2484{
2485    static const uint32_t ims_ext_mask =
2486        E1000_IMS_RXQ0 | E1000_IMS_RXQ1 |
2487        E1000_IMS_TXQ0 | E1000_IMS_TXQ1 |
2488        E1000_IMS_OTHER;
2489
2490    static const uint32_t ims_valid_mask =
2491        E1000_IMS_TXDW      | E1000_IMS_TXQE    | E1000_IMS_LSC  |
2492        E1000_IMS_RXDMT0    | E1000_IMS_RXO     | E1000_IMS_RXT0 |
2493        E1000_IMS_MDAC      | E1000_IMS_TXD_LOW | E1000_IMS_SRPD |
2494        E1000_IMS_ACK       | E1000_IMS_MNG     | E1000_IMS_RXQ0 |
2495        E1000_IMS_RXQ1      | E1000_IMS_TXQ0    | E1000_IMS_TXQ1 |
2496        E1000_IMS_OTHER;
2497
2498    uint32_t valid_val = val & ims_valid_mask;
2499
2500    trace_e1000e_irq_set_ims(val, core->mac[IMS], core->mac[IMS] | valid_val);
2501    core->mac[IMS] |= valid_val;
2502
2503    if ((valid_val & ims_ext_mask) &&
2504        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_PBA_CLR) &&
2505        msix_enabled(core->owner)) {
2506        e1000e_msix_clear(core, valid_val);
2507    }
2508
2509    if ((valid_val == ims_valid_mask) &&
2510        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_INT_TIMERS_CLEAR_ENA)) {
2511        trace_e1000e_irq_fire_all_timers(val);
2512        e1000e_intrmgr_fire_all_timers(core);
2513    }
2514
2515    e1000e_update_interrupt_state(core);
2516}
2517
2518static void
2519e1000e_set_rdtr(E1000ECore *core, int index, uint32_t val)
2520{
2521    e1000e_set_16bit(core, index, val);
2522
2523    if ((val & E1000_RDTR_FPD) && (core->rdtr.running)) {
2524        trace_e1000e_irq_rdtr_fpd_running();
2525        e1000e_intrmgr_fire_delayed_interrupts(core);
2526    } else {
2527        trace_e1000e_irq_rdtr_fpd_not_running();
2528    }
2529}
2530
2531static void
2532e1000e_set_tidv(E1000ECore *core, int index, uint32_t val)
2533{
2534    e1000e_set_16bit(core, index, val);
2535
2536    if ((val & E1000_TIDV_FPD) && (core->tidv.running)) {
2537        trace_e1000e_irq_tidv_fpd_running();
2538        e1000e_intrmgr_fire_delayed_interrupts(core);
2539    } else {
2540        trace_e1000e_irq_tidv_fpd_not_running();
2541    }
2542}
2543
2544static uint32_t
2545e1000e_mac_readreg(E1000ECore *core, int index)
2546{
2547    return core->mac[index];
2548}
2549
2550static uint32_t
2551e1000e_mac_ics_read(E1000ECore *core, int index)
2552{
2553    trace_e1000e_irq_read_ics(core->mac[ICS]);
2554    return core->mac[ICS];
2555}
2556
2557static uint32_t
2558e1000e_mac_ims_read(E1000ECore *core, int index)
2559{
2560    trace_e1000e_irq_read_ims(core->mac[IMS]);
2561    return core->mac[IMS];
2562}
2563
2564#define E1000E_LOW_BITS_READ_FUNC(num)                      \
2565    static uint32_t                                         \
2566    e1000e_mac_low##num##_read(E1000ECore *core, int index) \
2567    {                                                       \
2568        return core->mac[index] & (BIT(num) - 1);           \
2569    }                                                       \
2570
2571#define E1000E_LOW_BITS_READ(num)                           \
2572    e1000e_mac_low##num##_read
2573
2574E1000E_LOW_BITS_READ_FUNC(4);
2575E1000E_LOW_BITS_READ_FUNC(6);
2576E1000E_LOW_BITS_READ_FUNC(11);
2577E1000E_LOW_BITS_READ_FUNC(13);
2578E1000E_LOW_BITS_READ_FUNC(16);
2579
2580static uint32_t
2581e1000e_mac_swsm_read(E1000ECore *core, int index)
2582{
2583    uint32_t val = core->mac[SWSM];
2584    core->mac[SWSM] = val | 1;
2585    return val;
2586}
2587
2588static uint32_t
2589e1000e_mac_itr_read(E1000ECore *core, int index)
2590{
2591    return core->itr_guest_value;
2592}
2593
2594static uint32_t
2595e1000e_mac_eitr_read(E1000ECore *core, int index)
2596{
2597    return core->eitr_guest_value[index - EITR];
2598}
2599
2600static uint32_t
2601e1000e_mac_icr_read(E1000ECore *core, int index)
2602{
2603    uint32_t ret = core->mac[ICR];
2604    trace_e1000e_irq_icr_read_entry(ret);
2605
2606    if (core->mac[IMS] == 0) {
2607        trace_e1000e_irq_icr_clear_zero_ims();
2608        core->mac[ICR] = 0;
2609    }
2610
2611    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2612        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2613        trace_e1000e_irq_icr_clear_iame();
2614        core->mac[ICR] = 0;
2615        trace_e1000e_irq_icr_process_iame();
2616        e1000e_clear_ims_bits(core, core->mac[IAM]);
2617    }
2618
2619    trace_e1000e_irq_icr_read_exit(core->mac[ICR]);
2620    e1000e_update_interrupt_state(core);
2621    return ret;
2622}
2623
2624static uint32_t
2625e1000e_mac_read_clr4(E1000ECore *core, int index)
2626{
2627    uint32_t ret = core->mac[index];
2628
2629    core->mac[index] = 0;
2630    return ret;
2631}
2632
2633static uint32_t
2634e1000e_mac_read_clr8(E1000ECore *core, int index)
2635{
2636    uint32_t ret = core->mac[index];
2637
2638    core->mac[index] = 0;
2639    core->mac[index - 1] = 0;
2640    return ret;
2641}
2642
2643static uint32_t
2644e1000e_get_ctrl(E1000ECore *core, int index)
2645{
2646    uint32_t val = core->mac[CTRL];
2647
2648    trace_e1000e_link_read_params(
2649        !!(val & E1000_CTRL_ASDE),
2650        (val & E1000_CTRL_SPD_SEL) >> E1000_CTRL_SPD_SHIFT,
2651        !!(val & E1000_CTRL_FRCSPD),
2652        !!(val & E1000_CTRL_FRCDPX),
2653        !!(val & E1000_CTRL_RFCE),
2654        !!(val & E1000_CTRL_TFCE));
2655
2656    return val;
2657}
2658
2659static uint32_t
2660e1000e_get_status(E1000ECore *core, int index)
2661{
2662    uint32_t res = core->mac[STATUS];
2663
2664    if (!(core->mac[CTRL] & E1000_CTRL_GIO_MASTER_DISABLE)) {
2665        res |= E1000_STATUS_GIO_MASTER_ENABLE;
2666    }
2667
2668    if (core->mac[CTRL] & E1000_CTRL_FRCDPX) {
2669        res |= (core->mac[CTRL] & E1000_CTRL_FD) ? E1000_STATUS_FD : 0;
2670    } else {
2671        res |= E1000_STATUS_FD;
2672    }
2673
2674    if ((core->mac[CTRL] & E1000_CTRL_FRCSPD) ||
2675        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_SPD_BYPS)) {
2676        switch (core->mac[CTRL] & E1000_CTRL_SPD_SEL) {
2677        case E1000_CTRL_SPD_10:
2678            res |= E1000_STATUS_SPEED_10;
2679            break;
2680        case E1000_CTRL_SPD_100:
2681            res |= E1000_STATUS_SPEED_100;
2682            break;
2683        case E1000_CTRL_SPD_1000:
2684        default:
2685            res |= E1000_STATUS_SPEED_1000;
2686            break;
2687        }
2688    } else {
2689        res |= E1000_STATUS_SPEED_1000;
2690    }
2691
2692    trace_e1000e_link_status(
2693        !!(res & E1000_STATUS_LU),
2694        !!(res & E1000_STATUS_FD),
2695        (res & E1000_STATUS_SPEED_MASK) >> E1000_STATUS_SPEED_SHIFT,
2696        (res & E1000_STATUS_ASDV) >> E1000_STATUS_ASDV_SHIFT);
2697
2698    return res;
2699}
2700
2701static uint32_t
2702e1000e_get_tarc(E1000ECore *core, int index)
2703{
2704    return core->mac[index] & ((BIT(11) - 1) |
2705                                BIT(27)      |
2706                                BIT(28)      |
2707                                BIT(29)      |
2708                                BIT(30));
2709}
2710
2711static void
2712e1000e_mac_writereg(E1000ECore *core, int index, uint32_t val)
2713{
2714    core->mac[index] = val;
2715}
2716
2717static void
2718e1000e_mac_setmacaddr(E1000ECore *core, int index, uint32_t val)
2719{
2720    uint32_t macaddr[2];
2721
2722    core->mac[index] = val;
2723
2724    macaddr[0] = cpu_to_le32(core->mac[RA]);
2725    macaddr[1] = cpu_to_le32(core->mac[RA + 1]);
2726    qemu_format_nic_info_str(qemu_get_queue(core->owner_nic),
2727        (uint8_t *) macaddr);
2728
2729    trace_e1000e_mac_set_sw(MAC_ARG(macaddr));
2730}
2731
2732static void
2733e1000e_set_eecd(E1000ECore *core, int index, uint32_t val)
2734{
2735    static const uint32_t ro_bits = E1000_EECD_PRES          |
2736                                    E1000_EECD_AUTO_RD       |
2737                                    E1000_EECD_SIZE_EX_MASK;
2738
2739    core->mac[EECD] = (core->mac[EECD] & ro_bits) | (val & ~ro_bits);
2740}
2741
2742static void
2743e1000e_set_eerd(E1000ECore *core, int index, uint32_t val)
2744{
2745    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2746    uint32_t flags = 0;
2747    uint32_t data = 0;
2748
2749    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2750        data = core->eeprom[addr];
2751        flags = E1000_EERW_DONE;
2752    }
2753
2754    core->mac[EERD] = flags                           |
2755                      (addr << E1000_EERW_ADDR_SHIFT) |
2756                      (data << E1000_EERW_DATA_SHIFT);
2757}
2758
2759static void
2760e1000e_set_eewr(E1000ECore *core, int index, uint32_t val)
2761{
2762    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2763    uint32_t data = (val >> E1000_EERW_DATA_SHIFT) & E1000_EERW_DATA_MASK;
2764    uint32_t flags = 0;
2765
2766    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2767        core->eeprom[addr] = data;
2768        flags = E1000_EERW_DONE;
2769    }
2770
2771    core->mac[EERD] = flags                           |
2772                      (addr << E1000_EERW_ADDR_SHIFT) |
2773                      (data << E1000_EERW_DATA_SHIFT);
2774}
2775
2776static void
2777e1000e_set_rxdctl(E1000ECore *core, int index, uint32_t val)
2778{
2779    core->mac[RXDCTL] = core->mac[RXDCTL1] = val;
2780}
2781
2782static void
2783e1000e_set_itr(E1000ECore *core, int index, uint32_t val)
2784{
2785    uint32_t interval = val & 0xffff;
2786
2787    trace_e1000e_irq_itr_set(val);
2788
2789    core->itr_guest_value = interval;
2790    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2791}
2792
2793static void
2794e1000e_set_eitr(E1000ECore *core, int index, uint32_t val)
2795{
2796    uint32_t interval = val & 0xffff;
2797    uint32_t eitr_num = index - EITR;
2798
2799    trace_e1000e_irq_eitr_set(eitr_num, val);
2800
2801    core->eitr_guest_value[eitr_num] = interval;
2802    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2803}
2804
2805static void
2806e1000e_set_psrctl(E1000ECore *core, int index, uint32_t val)
2807{
2808    if ((val & E1000_PSRCTL_BSIZE0_MASK) == 0) {
2809        hw_error("e1000e: PSRCTL.BSIZE0 cannot be zero");
2810    }
2811
2812    if ((val & E1000_PSRCTL_BSIZE1_MASK) == 0) {
2813        hw_error("e1000e: PSRCTL.BSIZE1 cannot be zero");
2814    }
2815
2816    core->mac[PSRCTL] = val;
2817}
2818
2819static void
2820e1000e_update_rx_offloads(E1000ECore *core)
2821{
2822    int cso_state = e1000e_rx_l4_cso_enabled(core);
2823
2824    trace_e1000e_rx_set_cso(cso_state);
2825
2826    if (core->has_vnet) {
2827        qemu_set_offload(qemu_get_queue(core->owner_nic)->peer,
2828                         cso_state, 0, 0, 0, 0);
2829    }
2830}
2831
2832static void
2833e1000e_set_rxcsum(E1000ECore *core, int index, uint32_t val)
2834{
2835    core->mac[RXCSUM] = val;
2836    e1000e_update_rx_offloads(core);
2837}
2838
2839static void
2840e1000e_set_gcr(E1000ECore *core, int index, uint32_t val)
2841{
2842    uint32_t ro_bits = core->mac[GCR] & E1000_GCR_RO_BITS;
2843    core->mac[GCR] = (val & ~E1000_GCR_RO_BITS) | ro_bits;
2844}
2845
2846#define e1000e_getreg(x)    [x] = e1000e_mac_readreg
2847static uint32_t (*e1000e_macreg_readops[])(E1000ECore *, int) = {
2848    e1000e_getreg(PBA),
2849    e1000e_getreg(WUFC),
2850    e1000e_getreg(MANC),
2851    e1000e_getreg(TOTL),
2852    e1000e_getreg(RDT0),
2853    e1000e_getreg(RDBAH0),
2854    e1000e_getreg(TDBAL1),
2855    e1000e_getreg(RDLEN0),
2856    e1000e_getreg(RDH1),
2857    e1000e_getreg(LATECOL),
2858    e1000e_getreg(SEQEC),
2859    e1000e_getreg(XONTXC),
2860    e1000e_getreg(WUS),
2861    e1000e_getreg(GORCL),
2862    e1000e_getreg(MGTPRC),
2863    e1000e_getreg(EERD),
2864    e1000e_getreg(EIAC),
2865    e1000e_getreg(PSRCTL),
2866    e1000e_getreg(MANC2H),
2867    e1000e_getreg(RXCSUM),
2868    e1000e_getreg(GSCL_3),
2869    e1000e_getreg(GSCN_2),
2870    e1000e_getreg(RSRPD),
2871    e1000e_getreg(RDBAL1),
2872    e1000e_getreg(FCAH),
2873    e1000e_getreg(FCRTH),
2874    e1000e_getreg(FLOP),
2875    e1000e_getreg(FLASHT),
2876    e1000e_getreg(RXSTMPH),
2877    e1000e_getreg(TXSTMPL),
2878    e1000e_getreg(TIMADJL),
2879    e1000e_getreg(TXDCTL),
2880    e1000e_getreg(RDH0),
2881    e1000e_getreg(TDT1),
2882    e1000e_getreg(TNCRS),
2883    e1000e_getreg(RJC),
2884    e1000e_getreg(IAM),
2885    e1000e_getreg(GSCL_2),
2886    e1000e_getreg(RDBAH1),
2887    e1000e_getreg(FLSWDATA),
2888    e1000e_getreg(RXSATRH),
2889    e1000e_getreg(TIPG),
2890    e1000e_getreg(FLMNGCTL),
2891    e1000e_getreg(FLMNGCNT),
2892    e1000e_getreg(TSYNCTXCTL),
2893    e1000e_getreg(EXTCNF_SIZE),
2894    e1000e_getreg(EXTCNF_CTRL),
2895    e1000e_getreg(EEMNGDATA),
2896    e1000e_getreg(CTRL_EXT),
2897    e1000e_getreg(SYSTIMH),
2898    e1000e_getreg(EEMNGCTL),
2899    e1000e_getreg(FLMNGDATA),
2900    e1000e_getreg(TSYNCRXCTL),
2901    e1000e_getreg(TDH),
2902    e1000e_getreg(LEDCTL),
2903    e1000e_getreg(STATUS),
2904    e1000e_getreg(TCTL),
2905    e1000e_getreg(TDBAL),
2906    e1000e_getreg(TDLEN),
2907    e1000e_getreg(TDH1),
2908    e1000e_getreg(RADV),
2909    e1000e_getreg(ECOL),
2910    e1000e_getreg(DC),
2911    e1000e_getreg(RLEC),
2912    e1000e_getreg(XOFFTXC),
2913    e1000e_getreg(RFC),
2914    e1000e_getreg(RNBC),
2915    e1000e_getreg(MGTPTC),
2916    e1000e_getreg(TIMINCA),
2917    e1000e_getreg(RXCFGL),
2918    e1000e_getreg(MFUTP01),
2919    e1000e_getreg(FACTPS),
2920    e1000e_getreg(GSCL_1),
2921    e1000e_getreg(GSCN_0),
2922    e1000e_getreg(GCR2),
2923    e1000e_getreg(RDT1),
2924    e1000e_getreg(PBACLR),
2925    e1000e_getreg(FCTTV),
2926    e1000e_getreg(EEWR),
2927    e1000e_getreg(FLSWCTL),
2928    e1000e_getreg(RXDCTL1),
2929    e1000e_getreg(RXSATRL),
2930    e1000e_getreg(SYSTIML),
2931    e1000e_getreg(RXUDP),
2932    e1000e_getreg(TORL),
2933    e1000e_getreg(TDLEN1),
2934    e1000e_getreg(MCC),
2935    e1000e_getreg(WUC),
2936    e1000e_getreg(EECD),
2937    e1000e_getreg(MFUTP23),
2938    e1000e_getreg(RAID),
2939    e1000e_getreg(FCRTV),
2940    e1000e_getreg(TXDCTL1),
2941    e1000e_getreg(RCTL),
2942    e1000e_getreg(TDT),
2943    e1000e_getreg(MDIC),
2944    e1000e_getreg(FCRUC),
2945    e1000e_getreg(VET),
2946    e1000e_getreg(RDBAL0),
2947    e1000e_getreg(TDBAH1),
2948    e1000e_getreg(RDTR),
2949    e1000e_getreg(SCC),
2950    e1000e_getreg(COLC),
2951    e1000e_getreg(CEXTERR),
2952    e1000e_getreg(XOFFRXC),
2953    e1000e_getreg(IPAV),
2954    e1000e_getreg(GOTCL),
2955    e1000e_getreg(MGTPDC),
2956    e1000e_getreg(GCR),
2957    e1000e_getreg(IVAR),
2958    e1000e_getreg(POEMB),
2959    e1000e_getreg(MFVAL),
2960    e1000e_getreg(FUNCTAG),
2961    e1000e_getreg(GSCL_4),
2962    e1000e_getreg(GSCN_3),
2963    e1000e_getreg(MRQC),
2964    e1000e_getreg(RDLEN1),
2965    e1000e_getreg(FCT),
2966    e1000e_getreg(FLA),
2967    e1000e_getreg(FLOL),
2968    e1000e_getreg(RXDCTL),
2969    e1000e_getreg(RXSTMPL),
2970    e1000e_getreg(TXSTMPH),
2971    e1000e_getreg(TIMADJH),
2972    e1000e_getreg(FCRTL),
2973    e1000e_getreg(TDBAH),
2974    e1000e_getreg(TADV),
2975    e1000e_getreg(XONRXC),
2976    e1000e_getreg(TSCTFC),
2977    e1000e_getreg(RFCTL),
2978    e1000e_getreg(GSCN_1),
2979    e1000e_getreg(FCAL),
2980    e1000e_getreg(FLSWCNT),
2981
2982    [TOTH]    = e1000e_mac_read_clr8,
2983    [GOTCH]   = e1000e_mac_read_clr8,
2984    [PRC64]   = e1000e_mac_read_clr4,
2985    [PRC255]  = e1000e_mac_read_clr4,
2986    [PRC1023] = e1000e_mac_read_clr4,
2987    [PTC64]   = e1000e_mac_read_clr4,
2988    [PTC255]  = e1000e_mac_read_clr4,
2989    [PTC1023] = e1000e_mac_read_clr4,
2990    [GPRC]    = e1000e_mac_read_clr4,
2991    [TPT]     = e1000e_mac_read_clr4,
2992    [RUC]     = e1000e_mac_read_clr4,
2993    [BPRC]    = e1000e_mac_read_clr4,
2994    [MPTC]    = e1000e_mac_read_clr4,
2995    [IAC]     = e1000e_mac_read_clr4,
2996    [ICR]     = e1000e_mac_icr_read,
2997    [RDFH]    = E1000E_LOW_BITS_READ(13),
2998    [RDFHS]   = E1000E_LOW_BITS_READ(13),
2999    [RDFPC]   = E1000E_LOW_BITS_READ(13),
3000    [TDFH]    = E1000E_LOW_BITS_READ(13),
3001    [TDFHS]   = E1000E_LOW_BITS_READ(13),
3002    [STATUS]  = e1000e_get_status,
3003    [TARC0]   = e1000e_get_tarc,
3004    [PBS]     = E1000E_LOW_BITS_READ(6),
3005    [ICS]     = e1000e_mac_ics_read,
3006    [AIT]     = E1000E_LOW_BITS_READ(16),
3007    [TORH]    = e1000e_mac_read_clr8,
3008    [GORCH]   = e1000e_mac_read_clr8,
3009    [PRC127]  = e1000e_mac_read_clr4,
3010    [PRC511]  = e1000e_mac_read_clr4,
3011    [PRC1522] = e1000e_mac_read_clr4,
3012    [PTC127]  = e1000e_mac_read_clr4,
3013    [PTC511]  = e1000e_mac_read_clr4,
3014    [PTC1522] = e1000e_mac_read_clr4,
3015    [GPTC]    = e1000e_mac_read_clr4,
3016    [TPR]     = e1000e_mac_read_clr4,
3017    [ROC]     = e1000e_mac_read_clr4,
3018    [MPRC]    = e1000e_mac_read_clr4,
3019    [BPTC]    = e1000e_mac_read_clr4,
3020    [TSCTC]   = e1000e_mac_read_clr4,
3021    [ITR]     = e1000e_mac_itr_read,
3022    [RDFT]    = E1000E_LOW_BITS_READ(13),
3023    [RDFTS]   = E1000E_LOW_BITS_READ(13),
3024    [TDFPC]   = E1000E_LOW_BITS_READ(13),
3025    [TDFT]    = E1000E_LOW_BITS_READ(13),
3026    [TDFTS]   = E1000E_LOW_BITS_READ(13),
3027    [CTRL]    = e1000e_get_ctrl,
3028    [TARC1]   = e1000e_get_tarc,
3029    [SWSM]    = e1000e_mac_swsm_read,
3030    [IMS]     = e1000e_mac_ims_read,
3031
3032    [CRCERRS ... MPC]      = e1000e_mac_readreg,
3033    [IP6AT ... IP6AT + 3]  = e1000e_mac_readreg,
3034    [IP4AT ... IP4AT + 6]  = e1000e_mac_readreg,
3035    [RA ... RA + 31]       = e1000e_mac_readreg,
3036    [WUPM ... WUPM + 31]   = e1000e_mac_readreg,
3037    [MTA ... MTA + 127]    = e1000e_mac_readreg,
3038    [VFTA ... VFTA + 127]  = e1000e_mac_readreg,
3039    [FFMT ... FFMT + 254]  = E1000E_LOW_BITS_READ(4),
3040    [FFVT ... FFVT + 254]  = e1000e_mac_readreg,
3041    [MDEF ... MDEF + 7]    = e1000e_mac_readreg,
3042    [FFLT ... FFLT + 10]   = E1000E_LOW_BITS_READ(11),
3043    [FTFT ... FTFT + 254]  = e1000e_mac_readreg,
3044    [PBM ... PBM + 10239]  = e1000e_mac_readreg,
3045    [RETA ... RETA + 31]   = e1000e_mac_readreg,
3046    [RSSRK ... RSSRK + 31] = e1000e_mac_readreg,
3047    [MAVTV0 ... MAVTV3]    = e1000e_mac_readreg,
3048    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_mac_eitr_read
3049};
3050enum { E1000E_NREADOPS = ARRAY_SIZE(e1000e_macreg_readops) };
3051
3052#define e1000e_putreg(x)    [x] = e1000e_mac_writereg
3053static void (*e1000e_macreg_writeops[])(E1000ECore *, int, uint32_t) = {
3054    e1000e_putreg(PBA),
3055    e1000e_putreg(SWSM),
3056    e1000e_putreg(WUFC),
3057    e1000e_putreg(RDBAH1),
3058    e1000e_putreg(TDBAH),
3059    e1000e_putreg(TXDCTL),
3060    e1000e_putreg(RDBAH0),
3061    e1000e_putreg(LEDCTL),
3062    e1000e_putreg(FCAL),
3063    e1000e_putreg(FCRUC),
3064    e1000e_putreg(AIT),
3065    e1000e_putreg(TDFH),
3066    e1000e_putreg(TDFT),
3067    e1000e_putreg(TDFHS),
3068    e1000e_putreg(TDFTS),
3069    e1000e_putreg(TDFPC),
3070    e1000e_putreg(WUC),
3071    e1000e_putreg(WUS),
3072    e1000e_putreg(RDFH),
3073    e1000e_putreg(RDFT),
3074    e1000e_putreg(RDFHS),
3075    e1000e_putreg(RDFTS),
3076    e1000e_putreg(RDFPC),
3077    e1000e_putreg(IPAV),
3078    e1000e_putreg(TDBAH1),
3079    e1000e_putreg(TIMINCA),
3080    e1000e_putreg(IAM),
3081    e1000e_putreg(EIAC),
3082    e1000e_putreg(IVAR),
3083    e1000e_putreg(TARC0),
3084    e1000e_putreg(TARC1),
3085    e1000e_putreg(FLSWDATA),
3086    e1000e_putreg(POEMB),
3087    e1000e_putreg(PBS),
3088    e1000e_putreg(MFUTP01),
3089    e1000e_putreg(MFUTP23),
3090    e1000e_putreg(MANC),
3091    e1000e_putreg(MANC2H),
3092    e1000e_putreg(MFVAL),
3093    e1000e_putreg(EXTCNF_CTRL),
3094    e1000e_putreg(FACTPS),
3095    e1000e_putreg(FUNCTAG),
3096    e1000e_putreg(GSCL_1),
3097    e1000e_putreg(GSCL_2),
3098    e1000e_putreg(GSCL_3),
3099    e1000e_putreg(GSCL_4),
3100    e1000e_putreg(GSCN_0),
3101    e1000e_putreg(GSCN_1),
3102    e1000e_putreg(GSCN_2),
3103    e1000e_putreg(GSCN_3),
3104    e1000e_putreg(GCR2),
3105    e1000e_putreg(MRQC),
3106    e1000e_putreg(FLOP),
3107    e1000e_putreg(FLOL),
3108    e1000e_putreg(FLSWCTL),
3109    e1000e_putreg(FLSWCNT),
3110    e1000e_putreg(FLA),
3111    e1000e_putreg(RXDCTL1),
3112    e1000e_putreg(TXDCTL1),
3113    e1000e_putreg(TIPG),
3114    e1000e_putreg(RXSTMPH),
3115    e1000e_putreg(RXSTMPL),
3116    e1000e_putreg(RXSATRL),
3117    e1000e_putreg(RXSATRH),
3118    e1000e_putreg(TXSTMPL),
3119    e1000e_putreg(TXSTMPH),
3120    e1000e_putreg(SYSTIML),
3121    e1000e_putreg(SYSTIMH),
3122    e1000e_putreg(TIMADJL),
3123    e1000e_putreg(TIMADJH),
3124    e1000e_putreg(RXUDP),
3125    e1000e_putreg(RXCFGL),
3126    e1000e_putreg(TSYNCRXCTL),
3127    e1000e_putreg(TSYNCTXCTL),
3128    e1000e_putreg(FLSWDATA),
3129    e1000e_putreg(EXTCNF_SIZE),
3130    e1000e_putreg(EEMNGCTL),
3131    e1000e_putreg(RA),
3132
3133    [TDH1]     = e1000e_set_16bit,
3134    [TDT1]     = e1000e_set_tdt,
3135    [TCTL]     = e1000e_set_tctl,
3136    [TDT]      = e1000e_set_tdt,
3137    [MDIC]     = e1000e_set_mdic,
3138    [ICS]      = e1000e_set_ics,
3139    [TDH]      = e1000e_set_16bit,
3140    [RDH0]     = e1000e_set_16bit,
3141    [RDT0]     = e1000e_set_rdt,
3142    [IMC]      = e1000e_set_imc,
3143    [IMS]      = e1000e_set_ims,
3144    [ICR]      = e1000e_set_icr,
3145    [EECD]     = e1000e_set_eecd,
3146    [RCTL]     = e1000e_set_rx_control,
3147    [CTRL]     = e1000e_set_ctrl,
3148    [RDTR]     = e1000e_set_rdtr,
3149    [RADV]     = e1000e_set_16bit,
3150    [TADV]     = e1000e_set_16bit,
3151    [ITR]      = e1000e_set_itr,
3152    [EERD]     = e1000e_set_eerd,
3153    [GCR]      = e1000e_set_gcr,
3154    [PSRCTL]   = e1000e_set_psrctl,
3155    [RXCSUM]   = e1000e_set_rxcsum,
3156    [RAID]     = e1000e_set_16bit,
3157    [RSRPD]    = e1000e_set_12bit,
3158    [TIDV]     = e1000e_set_tidv,
3159    [TDLEN1]   = e1000e_set_dlen,
3160    [TDLEN]    = e1000e_set_dlen,
3161    [RDLEN0]   = e1000e_set_dlen,
3162    [RDLEN1]   = e1000e_set_dlen,
3163    [TDBAL]    = e1000e_set_dbal,
3164    [TDBAL1]   = e1000e_set_dbal,
3165    [RDBAL0]   = e1000e_set_dbal,
3166    [RDBAL1]   = e1000e_set_dbal,
3167    [RDH1]     = e1000e_set_16bit,
3168    [RDT1]     = e1000e_set_rdt,
3169    [STATUS]   = e1000e_set_status,
3170    [PBACLR]   = e1000e_set_pbaclr,
3171    [CTRL_EXT] = e1000e_set_ctrlext,
3172    [FCAH]     = e1000e_set_16bit,
3173    [FCT]      = e1000e_set_16bit,
3174    [FCTTV]    = e1000e_set_16bit,
3175    [FCRTV]    = e1000e_set_16bit,
3176    [FCRTH]    = e1000e_set_fcrth,
3177    [FCRTL]    = e1000e_set_fcrtl,
3178    [VET]      = e1000e_set_vet,
3179    [RXDCTL]   = e1000e_set_rxdctl,
3180    [FLASHT]   = e1000e_set_16bit,
3181    [EEWR]     = e1000e_set_eewr,
3182    [CTRL_DUP] = e1000e_set_ctrl,
3183    [RFCTL]    = e1000e_set_rfctl,
3184    [RA + 1]   = e1000e_mac_setmacaddr,
3185
3186    [IP6AT ... IP6AT + 3]    = e1000e_mac_writereg,
3187    [IP4AT ... IP4AT + 6]    = e1000e_mac_writereg,
3188    [RA + 2 ... RA + 31]     = e1000e_mac_writereg,
3189    [WUPM ... WUPM + 31]     = e1000e_mac_writereg,
3190    [MTA ... MTA + 127]      = e1000e_mac_writereg,
3191    [VFTA ... VFTA + 127]    = e1000e_mac_writereg,
3192    [FFMT ... FFMT + 254]    = e1000e_mac_writereg,
3193    [FFVT ... FFVT + 254]    = e1000e_mac_writereg,
3194    [PBM ... PBM + 10239]    = e1000e_mac_writereg,
3195    [MDEF ... MDEF + 7]      = e1000e_mac_writereg,
3196    [FFLT ... FFLT + 10]     = e1000e_mac_writereg,
3197    [FTFT ... FTFT + 254]    = e1000e_mac_writereg,
3198    [RETA ... RETA + 31]     = e1000e_mac_writereg,
3199    [RSSRK ... RSSRK + 31]   = e1000e_mac_writereg,
3200    [MAVTV0 ... MAVTV3]      = e1000e_mac_writereg,
3201    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_set_eitr
3202};
3203enum { E1000E_NWRITEOPS = ARRAY_SIZE(e1000e_macreg_writeops) };
3204
3205enum { MAC_ACCESS_PARTIAL = 1 };
3206
3207/* The array below combines alias offsets of the index values for the
3208 * MAC registers that have aliases, with the indication of not fully
3209 * implemented registers (lowest bit). This combination is possible
3210 * because all of the offsets are even. */
3211static const uint16_t mac_reg_access[E1000E_MAC_SIZE] = {
3212    /* Alias index offsets */
3213    [FCRTL_A] = 0x07fe, [FCRTH_A] = 0x0802,
3214    [RDH0_A]  = 0x09bc, [RDT0_A]  = 0x09bc, [RDTR_A] = 0x09c6,
3215    [RDFH_A]  = 0xe904, [RDFT_A]  = 0xe904,
3216    [TDH_A]   = 0x0cf8, [TDT_A]   = 0x0cf8, [TIDV_A] = 0x0cf8,
3217    [TDFH_A]  = 0xed00, [TDFT_A]  = 0xed00,
3218    [RA_A ... RA_A + 31]      = 0x14f0,
3219    [VFTA_A ... VFTA_A + 127] = 0x1400,
3220    [RDBAL0_A ... RDLEN0_A] = 0x09bc,
3221    [TDBAL_A ... TDLEN_A]   = 0x0cf8,
3222    /* Access options */
3223    [RDFH]  = MAC_ACCESS_PARTIAL,    [RDFT]  = MAC_ACCESS_PARTIAL,
3224    [RDFHS] = MAC_ACCESS_PARTIAL,    [RDFTS] = MAC_ACCESS_PARTIAL,
3225    [RDFPC] = MAC_ACCESS_PARTIAL,
3226    [TDFH]  = MAC_ACCESS_PARTIAL,    [TDFT]  = MAC_ACCESS_PARTIAL,
3227    [TDFHS] = MAC_ACCESS_PARTIAL,    [TDFTS] = MAC_ACCESS_PARTIAL,
3228    [TDFPC] = MAC_ACCESS_PARTIAL,    [EECD]  = MAC_ACCESS_PARTIAL,
3229    [PBM]   = MAC_ACCESS_PARTIAL,    [FLA]   = MAC_ACCESS_PARTIAL,
3230    [FCAL]  = MAC_ACCESS_PARTIAL,    [FCAH]  = MAC_ACCESS_PARTIAL,
3231    [FCT]   = MAC_ACCESS_PARTIAL,    [FCTTV] = MAC_ACCESS_PARTIAL,
3232    [FCRTV] = MAC_ACCESS_PARTIAL,    [FCRTL] = MAC_ACCESS_PARTIAL,
3233    [FCRTH] = MAC_ACCESS_PARTIAL,    [TXDCTL] = MAC_ACCESS_PARTIAL,
3234    [TXDCTL1] = MAC_ACCESS_PARTIAL,
3235    [MAVTV0 ... MAVTV3] = MAC_ACCESS_PARTIAL
3236};
3237
3238void
3239e1000e_core_write(E1000ECore *core, hwaddr addr, uint64_t val, unsigned size)
3240{
3241    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3242
3243    if (index < E1000E_NWRITEOPS && e1000e_macreg_writeops[index]) {
3244        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3245            trace_e1000e_wrn_regs_write_trivial(index << 2);
3246        }
3247        trace_e1000e_core_write(index << 2, size, val);
3248        e1000e_macreg_writeops[index](core, index, val);
3249    } else if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3250        trace_e1000e_wrn_regs_write_ro(index << 2, size, val);
3251    } else {
3252        trace_e1000e_wrn_regs_write_unknown(index << 2, size, val);
3253    }
3254}
3255
3256uint64_t
3257e1000e_core_read(E1000ECore *core, hwaddr addr, unsigned size)
3258{
3259    uint64_t val;
3260    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3261
3262    if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3263        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3264            trace_e1000e_wrn_regs_read_trivial(index << 2);
3265        }
3266        val = e1000e_macreg_readops[index](core, index);
3267        trace_e1000e_core_read(index << 2, size, val);
3268        return val;
3269    } else {
3270        trace_e1000e_wrn_regs_read_unknown(index << 2, size);
3271    }
3272    return 0;
3273}
3274
3275static inline void
3276e1000e_autoneg_pause(E1000ECore *core)
3277{
3278    timer_del(core->autoneg_timer);
3279}
3280
3281static void
3282e1000e_autoneg_resume(E1000ECore *core)
3283{
3284    if (e1000e_have_autoneg(core) &&
3285        !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
3286        qemu_get_queue(core->owner_nic)->link_down = false;
3287        timer_mod(core->autoneg_timer,
3288                  qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 500);
3289    }
3290}
3291
3292static void
3293e1000e_vm_state_change(void *opaque, int running, RunState state)
3294{
3295    E1000ECore *core = opaque;
3296
3297    if (running) {
3298        trace_e1000e_vm_state_running();
3299        e1000e_intrmgr_resume(core);
3300        e1000e_autoneg_resume(core);
3301    } else {
3302        trace_e1000e_vm_state_stopped();
3303        e1000e_autoneg_pause(core);
3304        e1000e_intrmgr_pause(core);
3305    }
3306}
3307
3308void
3309e1000e_core_pci_realize(E1000ECore     *core,
3310                        const uint16_t *eeprom_templ,
3311                        uint32_t        eeprom_size,
3312                        const uint8_t  *macaddr)
3313{
3314    int i;
3315
3316    core->autoneg_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
3317                                       e1000e_autoneg_timer, core);
3318    e1000e_intrmgr_pci_realize(core);
3319
3320    core->vmstate =
3321        qemu_add_vm_change_state_handler(e1000e_vm_state_change, core);
3322
3323    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3324        net_tx_pkt_init(&core->tx[i].tx_pkt, core->owner,
3325                        E1000E_MAX_TX_FRAGS, core->has_vnet);
3326    }
3327
3328    net_rx_pkt_init(&core->rx_pkt, core->has_vnet);
3329
3330    e1000x_core_prepare_eeprom(core->eeprom,
3331                               eeprom_templ,
3332                               eeprom_size,
3333                               PCI_DEVICE_GET_CLASS(core->owner)->device_id,
3334                               macaddr);
3335    e1000e_update_rx_offloads(core);
3336}
3337
3338void
3339e1000e_core_pci_uninit(E1000ECore *core)
3340{
3341    int i;
3342
3343    timer_del(core->autoneg_timer);
3344    timer_free(core->autoneg_timer);
3345
3346    e1000e_intrmgr_pci_unint(core);
3347
3348    qemu_del_vm_change_state_handler(core->vmstate);
3349
3350    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3351        net_tx_pkt_reset(core->tx[i].tx_pkt);
3352        net_tx_pkt_uninit(core->tx[i].tx_pkt);
3353    }
3354
3355    net_rx_pkt_uninit(core->rx_pkt);
3356}
3357
3358static const uint16_t
3359e1000e_phy_reg_init[E1000E_PHY_PAGES][E1000E_PHY_PAGE_SIZE] = {
3360    [0] = {
3361        [PHY_CTRL] =   MII_CR_SPEED_SELECT_MSB  |
3362                       MII_CR_FULL_DUPLEX       |
3363                       MII_CR_AUTO_NEG_EN,
3364
3365        [PHY_STATUS] = MII_SR_EXTENDED_CAPS     |
3366                       MII_SR_LINK_STATUS       |
3367                       MII_SR_AUTONEG_CAPS      |
3368                       MII_SR_PREAMBLE_SUPPRESS |
3369                       MII_SR_EXTENDED_STATUS   |
3370                       MII_SR_10T_HD_CAPS       |
3371                       MII_SR_10T_FD_CAPS       |
3372                       MII_SR_100X_HD_CAPS      |
3373                       MII_SR_100X_FD_CAPS,
3374
3375        [PHY_ID1]               = 0x141,
3376        [PHY_ID2]               = E1000_PHY_ID2_82574x,
3377        [PHY_AUTONEG_ADV]       = 0xde1,
3378        [PHY_LP_ABILITY]        = 0x7e0,
3379        [PHY_AUTONEG_EXP]       = BIT(2),
3380        [PHY_NEXT_PAGE_TX]      = BIT(0) | BIT(13),
3381        [PHY_1000T_CTRL]        = BIT(8) | BIT(9) | BIT(10) | BIT(11),
3382        [PHY_1000T_STATUS]      = 0x3c00,
3383        [PHY_EXT_STATUS]        = BIT(12) | BIT(13),
3384
3385        [PHY_COPPER_CTRL1]      = BIT(5) | BIT(6) | BIT(8) | BIT(9) |
3386                                  BIT(12) | BIT(13),
3387        [PHY_COPPER_STAT1]      = BIT(3) | BIT(10) | BIT(11) | BIT(13) | BIT(15)
3388    },
3389    [2] = {
3390        [PHY_MAC_CTRL1]         = BIT(3) | BIT(7),
3391        [PHY_MAC_CTRL2]         = BIT(1) | BIT(2) | BIT(6) | BIT(12)
3392    },
3393    [3] = {
3394        [PHY_LED_TIMER_CTRL]    = BIT(0) | BIT(2) | BIT(14)
3395    }
3396};
3397
3398static const uint32_t e1000e_mac_reg_init[] = {
3399    [PBA]           =     0x00140014,
3400    [LEDCTL]        =  BIT(1) | BIT(8) | BIT(9) | BIT(15) | BIT(17) | BIT(18),
3401    [EXTCNF_CTRL]   = BIT(3),
3402    [EEMNGCTL]      = BIT(31),
3403    [FLASHT]        = 0x2,
3404    [FLSWCTL]       = BIT(30) | BIT(31),
3405    [FLOL]          = BIT(0),
3406    [RXDCTL]        = BIT(16),
3407    [RXDCTL1]       = BIT(16),
3408    [TIPG]          = 0x8 | (0x8 << 10) | (0x6 << 20),
3409    [RXCFGL]        = 0x88F7,
3410    [RXUDP]         = 0x319,
3411    [CTRL]          = E1000_CTRL_FD | E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN0 |
3412                      E1000_CTRL_SPD_1000 | E1000_CTRL_SLU |
3413                      E1000_CTRL_ADVD3WUC,
3414    [STATUS]        =  E1000_STATUS_ASDV_1000 | E1000_STATUS_LU,
3415    [PSRCTL]        = (2 << E1000_PSRCTL_BSIZE0_SHIFT) |
3416                      (4 << E1000_PSRCTL_BSIZE1_SHIFT) |
3417                      (4 << E1000_PSRCTL_BSIZE2_SHIFT),
3418    [TARC0]         = 0x3 | E1000_TARC_ENABLE,
3419    [TARC1]         = 0x3 | E1000_TARC_ENABLE,
3420    [EECD]          = E1000_EECD_AUTO_RD | E1000_EECD_PRES,
3421    [EERD]          = E1000_EERW_DONE,
3422    [EEWR]          = E1000_EERW_DONE,
3423    [GCR]           = E1000_L0S_ADJUST |
3424                      E1000_L1_ENTRY_LATENCY_MSB |
3425                      E1000_L1_ENTRY_LATENCY_LSB,
3426    [TDFH]          = 0x600,
3427    [TDFT]          = 0x600,
3428    [TDFHS]         = 0x600,
3429    [TDFTS]         = 0x600,
3430    [POEMB]         = 0x30D,
3431    [PBS]           = 0x028,
3432    [MANC]          = E1000_MANC_DIS_IP_CHK_ARP,
3433    [FACTPS]        = E1000_FACTPS_LAN0_ON | 0x20000000,
3434    [SWSM]          = 1,
3435    [RXCSUM]        = E1000_RXCSUM_IPOFLD | E1000_RXCSUM_TUOFLD,
3436    [ITR]           = E1000E_MIN_XITR,
3437    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = E1000E_MIN_XITR,
3438};
3439
3440void
3441e1000e_core_reset(E1000ECore *core)
3442{
3443    int i;
3444
3445    timer_del(core->autoneg_timer);
3446
3447    e1000e_intrmgr_reset(core);
3448
3449    memset(core->phy, 0, sizeof core->phy);
3450    memmove(core->phy, e1000e_phy_reg_init, sizeof e1000e_phy_reg_init);
3451    memset(core->mac, 0, sizeof core->mac);
3452    memmove(core->mac, e1000e_mac_reg_init, sizeof e1000e_mac_reg_init);
3453
3454    core->rxbuf_min_shift = 1 + E1000_RING_DESC_LEN_SHIFT;
3455
3456    if (qemu_get_queue(core->owner_nic)->link_down) {
3457        e1000e_link_down(core);
3458    }
3459
3460    e1000x_reset_mac_addr(core->owner_nic, core->mac, core->permanent_mac);
3461
3462    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3463        net_tx_pkt_reset(core->tx[i].tx_pkt);
3464        memset(&core->tx[i].props, 0, sizeof(core->tx[i].props));
3465        core->tx[i].skip_cp = false;
3466    }
3467}
3468
3469void e1000e_core_pre_save(E1000ECore *core)
3470{
3471    int i;
3472    NetClientState *nc = qemu_get_queue(core->owner_nic);
3473
3474    /*
3475    * If link is down and auto-negotiation is supported and ongoing,
3476    * complete auto-negotiation immediately. This allows us to look
3477    * at MII_SR_AUTONEG_COMPLETE to infer link status on load.
3478    */
3479    if (nc->link_down && e1000e_have_autoneg(core)) {
3480        core->phy[0][PHY_STATUS] |= MII_SR_AUTONEG_COMPLETE;
3481        e1000e_update_flowctl_status(core);
3482    }
3483
3484    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3485        if (net_tx_pkt_has_fragments(core->tx[i].tx_pkt)) {
3486            core->tx[i].skip_cp = true;
3487        }
3488    }
3489}
3490
3491int
3492e1000e_core_post_load(E1000ECore *core)
3493{
3494    NetClientState *nc = qemu_get_queue(core->owner_nic);
3495
3496    /* nc.link_down can't be migrated, so infer link_down according
3497     * to link status bit in core.mac[STATUS].
3498     */
3499    nc->link_down = (core->mac[STATUS] & E1000_STATUS_LU) == 0;
3500
3501    return 0;
3502}
3503