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 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->props.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->props.sum_needed & E1000_TXD_POPTS_TXSM) {
 643        net_tx_pkt_build_vheader(tx->tx_pkt, false, true, 0);
 644    }
 645
 646    if (tx->props.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->props.sum_needed = le32_to_cpu(dp->upper.data) >> 8;
 719        tx->props.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->props.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->props.sum_needed = 0;
 751        tx->props.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}
 811
 812static inline uint64_t
 813e1000e_ring_base(E1000ECore *core, const E1000E_RingInfo *r)
 814{
 815    uint64_t bah = core->mac[r->dbah];
 816    uint64_t bal = core->mac[r->dbal];
 817
 818    return (bah << 32) + bal;
 819}
 820
 821static inline uint64_t
 822e1000e_ring_head_descr(E1000ECore *core, const E1000E_RingInfo *r)
 823{
 824    return e1000e_ring_base(core, r) + E1000_RING_DESC_LEN * core->mac[r->dh];
 825}
 826
 827static inline void
 828e1000e_ring_advance(E1000ECore *core, const E1000E_RingInfo *r, uint32_t count)
 829{
 830    core->mac[r->dh] += count;
 831
 832    if (core->mac[r->dh] * E1000_RING_DESC_LEN >= core->mac[r->dlen]) {
 833        core->mac[r->dh] = 0;
 834    }
 835}
 836
 837static inline uint32_t
 838e1000e_ring_free_descr_num(E1000ECore *core, const E1000E_RingInfo *r)
 839{
 840    trace_e1000e_ring_free_space(r->idx, core->mac[r->dlen],
 841                                 core->mac[r->dh],  core->mac[r->dt]);
 842
 843    if (core->mac[r->dh] <= core->mac[r->dt]) {
 844        return core->mac[r->dt] - core->mac[r->dh];
 845    }
 846
 847    if (core->mac[r->dh] > core->mac[r->dt]) {
 848        return core->mac[r->dlen] / E1000_RING_DESC_LEN +
 849               core->mac[r->dt] - core->mac[r->dh];
 850    }
 851
 852    g_assert_not_reached();
 853    return 0;
 854}
 855
 856static inline bool
 857e1000e_ring_enabled(E1000ECore *core, const E1000E_RingInfo *r)
 858{
 859    return core->mac[r->dlen] > 0;
 860}
 861
 862static inline uint32_t
 863e1000e_ring_len(E1000ECore *core, const E1000E_RingInfo *r)
 864{
 865    return core->mac[r->dlen];
 866}
 867
 868typedef struct E1000E_TxRing_st {
 869    const E1000E_RingInfo *i;
 870    struct e1000e_tx *tx;
 871} E1000E_TxRing;
 872
 873static inline int
 874e1000e_mq_queue_idx(int base_reg_idx, int reg_idx)
 875{
 876    return (reg_idx - base_reg_idx) / (0x100 >> 2);
 877}
 878
 879static inline void
 880e1000e_tx_ring_init(E1000ECore *core, E1000E_TxRing *txr, int idx)
 881{
 882    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 883        { TDBAH,  TDBAL,  TDLEN,  TDH,  TDT, 0 },
 884        { TDBAH1, TDBAL1, TDLEN1, TDH1, TDT1, 1 }
 885    };
 886
 887    assert(idx < ARRAY_SIZE(i));
 888
 889    txr->i     = &i[idx];
 890    txr->tx    = &core->tx[idx];
 891}
 892
 893typedef struct E1000E_RxRing_st {
 894    const E1000E_RingInfo *i;
 895} E1000E_RxRing;
 896
 897static inline void
 898e1000e_rx_ring_init(E1000ECore *core, E1000E_RxRing *rxr, int idx)
 899{
 900    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 901        { RDBAH0, RDBAL0, RDLEN0, RDH0, RDT0, 0 },
 902        { RDBAH1, RDBAL1, RDLEN1, RDH1, RDT1, 1 }
 903    };
 904
 905    assert(idx < ARRAY_SIZE(i));
 906
 907    rxr->i      = &i[idx];
 908}
 909
 910static void
 911e1000e_start_xmit(E1000ECore *core, const E1000E_TxRing *txr)
 912{
 913    dma_addr_t base;
 914    struct e1000_tx_desc desc;
 915    bool ide = false;
 916    const E1000E_RingInfo *txi = txr->i;
 917    uint32_t cause = E1000_ICS_TXQE;
 918
 919    if (!(core->mac[TCTL] & E1000_TCTL_EN)) {
 920        trace_e1000e_tx_disabled();
 921        return;
 922    }
 923
 924    while (!e1000e_ring_empty(core, txi)) {
 925        base = e1000e_ring_head_descr(core, txi);
 926
 927        pci_dma_read(core->owner, base, &desc, sizeof(desc));
 928
 929        trace_e1000e_tx_descr((void *)(intptr_t)desc.buffer_addr,
 930                              desc.lower.data, desc.upper.data);
 931
 932        e1000e_process_tx_desc(core, txr->tx, &desc, txi->idx);
 933        cause |= e1000e_txdesc_writeback(core, base, &desc, &ide, txi->idx);
 934
 935        e1000e_ring_advance(core, txi, 1);
 936    }
 937
 938    if (!ide || !e1000e_intrmgr_delay_tx_causes(core, &cause)) {
 939        e1000e_set_interrupt_cause(core, cause);
 940    }
 941}
 942
 943static bool
 944e1000e_has_rxbufs(E1000ECore *core, const E1000E_RingInfo *r,
 945                  size_t total_size)
 946{
 947    uint32_t bufs = e1000e_ring_free_descr_num(core, r);
 948
 949    trace_e1000e_rx_has_buffers(r->idx, bufs, total_size,
 950                                core->rx_desc_buf_size);
 951
 952    return total_size <= bufs / (core->rx_desc_len / E1000_MIN_RX_DESC_LEN) *
 953                         core->rx_desc_buf_size;
 954}
 955
 956static inline void
 957e1000e_start_recv(E1000ECore *core)
 958{
 959    int i;
 960
 961    trace_e1000e_rx_start_recv();
 962
 963    for (i = 0; i <= core->max_queue_num; i++) {
 964        qemu_flush_queued_packets(qemu_get_subqueue(core->owner_nic, i));
 965    }
 966}
 967
 968int
 969e1000e_can_receive(E1000ECore *core)
 970{
 971    int i;
 972
 973    if (!e1000x_rx_ready(core->owner, core->mac)) {
 974        return false;
 975    }
 976
 977    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
 978        E1000E_RxRing rxr;
 979
 980        e1000e_rx_ring_init(core, &rxr, i);
 981        if (e1000e_ring_enabled(core, rxr.i) &&
 982            e1000e_has_rxbufs(core, rxr.i, 1)) {
 983            trace_e1000e_rx_can_recv();
 984            return true;
 985        }
 986    }
 987
 988    trace_e1000e_rx_can_recv_rings_full();
 989    return false;
 990}
 991
 992ssize_t
 993e1000e_receive(E1000ECore *core, const uint8_t *buf, size_t size)
 994{
 995    const struct iovec iov = {
 996        .iov_base = (uint8_t *)buf,
 997        .iov_len = size
 998    };
 999
1000    return e1000e_receive_iov(core, &iov, 1);
1001}
1002
1003static inline bool
1004e1000e_rx_l3_cso_enabled(E1000ECore *core)
1005{
1006    return !!(core->mac[RXCSUM] & E1000_RXCSUM_IPOFLD);
1007}
1008
1009static inline bool
1010e1000e_rx_l4_cso_enabled(E1000ECore *core)
1011{
1012    return !!(core->mac[RXCSUM] & E1000_RXCSUM_TUOFLD);
1013}
1014
1015static bool
1016e1000e_receive_filter(E1000ECore *core, const uint8_t *buf, int size)
1017{
1018    uint32_t rctl = core->mac[RCTL];
1019
1020    if (e1000x_is_vlan_packet(buf, core->vet) &&
1021        e1000x_vlan_rx_filter_enabled(core->mac)) {
1022        uint16_t vid = lduw_be_p(buf + 14);
1023        uint32_t vfta = ldl_le_p((uint32_t *)(core->mac + VFTA) +
1024                                 ((vid >> 5) & 0x7f));
1025        if ((vfta & (1 << (vid & 0x1f))) == 0) {
1026            trace_e1000e_rx_flt_vlan_mismatch(vid);
1027            return false;
1028        } else {
1029            trace_e1000e_rx_flt_vlan_match(vid);
1030        }
1031    }
1032
1033    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1034    case ETH_PKT_UCAST:
1035        if (rctl & E1000_RCTL_UPE) {
1036            return true; /* promiscuous ucast */
1037        }
1038        break;
1039
1040    case ETH_PKT_BCAST:
1041        if (rctl & E1000_RCTL_BAM) {
1042            return true; /* broadcast enabled */
1043        }
1044        break;
1045
1046    case ETH_PKT_MCAST:
1047        if (rctl & E1000_RCTL_MPE) {
1048            return true; /* promiscuous mcast */
1049        }
1050        break;
1051
1052    default:
1053        g_assert_not_reached();
1054    }
1055
1056    return e1000x_rx_group_filter(core->mac, buf);
1057}
1058
1059static inline void
1060e1000e_read_lgcy_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1061{
1062    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1063    *buff_addr = le64_to_cpu(d->buffer_addr);
1064}
1065
1066static inline void
1067e1000e_read_ext_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1068{
1069    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1070    *buff_addr = le64_to_cpu(d->read.buffer_addr);
1071}
1072
1073static inline void
1074e1000e_read_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1075                        hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1076{
1077    int i;
1078    union e1000_rx_desc_packet_split *d =
1079        (union e1000_rx_desc_packet_split *) desc;
1080
1081    for (i = 0; i < MAX_PS_BUFFERS; i++) {
1082        (*buff_addr)[i] = le64_to_cpu(d->read.buffer_addr[i]);
1083    }
1084
1085    trace_e1000e_rx_desc_ps_read((*buff_addr)[0], (*buff_addr)[1],
1086                                 (*buff_addr)[2], (*buff_addr)[3]);
1087}
1088
1089static inline void
1090e1000e_read_rx_descr(E1000ECore *core, uint8_t *desc,
1091                     hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1092{
1093    if (e1000e_rx_use_legacy_descriptor(core)) {
1094        e1000e_read_lgcy_rx_descr(core, desc, &(*buff_addr)[0]);
1095        (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1096    } else {
1097        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1098            e1000e_read_ps_rx_descr(core, desc, buff_addr);
1099        } else {
1100            e1000e_read_ext_rx_descr(core, desc, &(*buff_addr)[0]);
1101            (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1102        }
1103    }
1104}
1105
1106static void
1107e1000e_verify_csum_in_sw(E1000ECore *core,
1108                         struct NetRxPkt *pkt,
1109                         uint32_t *status_flags,
1110                         bool istcp, bool isudp)
1111{
1112    bool csum_valid;
1113    uint32_t csum_error;
1114
1115    if (e1000e_rx_l3_cso_enabled(core)) {
1116        if (!net_rx_pkt_validate_l3_csum(pkt, &csum_valid)) {
1117            trace_e1000e_rx_metadata_l3_csum_validation_failed();
1118        } else {
1119            csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_IPE;
1120            *status_flags |= E1000_RXD_STAT_IPCS | csum_error;
1121        }
1122    } else {
1123        trace_e1000e_rx_metadata_l3_cso_disabled();
1124    }
1125
1126    if (!e1000e_rx_l4_cso_enabled(core)) {
1127        trace_e1000e_rx_metadata_l4_cso_disabled();
1128        return;
1129    }
1130
1131    if (!net_rx_pkt_validate_l4_csum(pkt, &csum_valid)) {
1132        trace_e1000e_rx_metadata_l4_csum_validation_failed();
1133        return;
1134    }
1135
1136    csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_TCPE;
1137
1138    if (istcp) {
1139        *status_flags |= E1000_RXD_STAT_TCPCS |
1140                         csum_error;
1141    } else if (isudp) {
1142        *status_flags |= E1000_RXD_STAT_TCPCS |
1143                         E1000_RXD_STAT_UDPCS |
1144                         csum_error;
1145    }
1146}
1147
1148static inline bool
1149e1000e_is_tcp_ack(E1000ECore *core, struct NetRxPkt *rx_pkt)
1150{
1151    if (!net_rx_pkt_is_tcp_ack(rx_pkt)) {
1152        return false;
1153    }
1154
1155    if (core->mac[RFCTL] & E1000_RFCTL_ACK_DATA_DIS) {
1156        return !net_rx_pkt_has_tcp_data(rx_pkt);
1157    }
1158
1159    return true;
1160}
1161
1162static void
1163e1000e_build_rx_metadata(E1000ECore *core,
1164                         struct NetRxPkt *pkt,
1165                         bool is_eop,
1166                         const E1000E_RSSInfo *rss_info,
1167                         uint32_t *rss, uint32_t *mrq,
1168                         uint32_t *status_flags,
1169                         uint16_t *ip_id,
1170                         uint16_t *vlan_tag)
1171{
1172    struct virtio_net_hdr *vhdr;
1173    bool isip4, isip6, istcp, isudp;
1174    uint32_t pkt_type;
1175
1176    *status_flags = E1000_RXD_STAT_DD;
1177
1178    /* No additional metadata needed for non-EOP descriptors */
1179    if (!is_eop) {
1180        goto func_exit;
1181    }
1182
1183    *status_flags |= E1000_RXD_STAT_EOP;
1184
1185    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1186    trace_e1000e_rx_metadata_protocols(isip4, isip6, isudp, istcp);
1187
1188    /* VLAN state */
1189    if (net_rx_pkt_is_vlan_stripped(pkt)) {
1190        *status_flags |= E1000_RXD_STAT_VP;
1191        *vlan_tag = cpu_to_le16(net_rx_pkt_get_vlan_tag(pkt));
1192        trace_e1000e_rx_metadata_vlan(*vlan_tag);
1193    }
1194
1195    /* Packet parsing results */
1196    if ((core->mac[RXCSUM] & E1000_RXCSUM_PCSD) != 0) {
1197        if (rss_info->enabled) {
1198            *rss = cpu_to_le32(rss_info->hash);
1199            *mrq = cpu_to_le32(rss_info->type | (rss_info->queue << 8));
1200            trace_e1000e_rx_metadata_rss(*rss, *mrq);
1201        }
1202    } else if (isip4) {
1203            *status_flags |= E1000_RXD_STAT_IPIDV;
1204            *ip_id = cpu_to_le16(net_rx_pkt_get_ip_id(pkt));
1205            trace_e1000e_rx_metadata_ip_id(*ip_id);
1206    }
1207
1208    if (istcp && e1000e_is_tcp_ack(core, pkt)) {
1209        *status_flags |= E1000_RXD_STAT_ACK;
1210        trace_e1000e_rx_metadata_ack();
1211    }
1212
1213    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_DIS)) {
1214        trace_e1000e_rx_metadata_ipv6_filtering_disabled();
1215        pkt_type = E1000_RXD_PKT_MAC;
1216    } else if (istcp || isudp) {
1217        pkt_type = isip4 ? E1000_RXD_PKT_IP4_XDP : E1000_RXD_PKT_IP6_XDP;
1218    } else if (isip4 || isip6) {
1219        pkt_type = isip4 ? E1000_RXD_PKT_IP4 : E1000_RXD_PKT_IP6;
1220    } else {
1221        pkt_type = E1000_RXD_PKT_MAC;
1222    }
1223
1224    *status_flags |= E1000_RXD_PKT_TYPE(pkt_type);
1225    trace_e1000e_rx_metadata_pkt_type(pkt_type);
1226
1227    /* RX CSO information */
1228    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_XSUM_DIS)) {
1229        trace_e1000e_rx_metadata_ipv6_sum_disabled();
1230        goto func_exit;
1231    }
1232
1233    if (!net_rx_pkt_has_virt_hdr(pkt)) {
1234        trace_e1000e_rx_metadata_no_virthdr();
1235        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1236        goto func_exit;
1237    }
1238
1239    vhdr = net_rx_pkt_get_vhdr(pkt);
1240
1241    if (!(vhdr->flags & VIRTIO_NET_HDR_F_DATA_VALID) &&
1242        !(vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
1243        trace_e1000e_rx_metadata_virthdr_no_csum_info();
1244        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1245        goto func_exit;
1246    }
1247
1248    if (e1000e_rx_l3_cso_enabled(core)) {
1249        *status_flags |= isip4 ? E1000_RXD_STAT_IPCS : 0;
1250    } else {
1251        trace_e1000e_rx_metadata_l3_cso_disabled();
1252    }
1253
1254    if (e1000e_rx_l4_cso_enabled(core)) {
1255        if (istcp) {
1256            *status_flags |= E1000_RXD_STAT_TCPCS;
1257        } else if (isudp) {
1258            *status_flags |= E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS;
1259        }
1260    } else {
1261        trace_e1000e_rx_metadata_l4_cso_disabled();
1262    }
1263
1264    trace_e1000e_rx_metadata_status_flags(*status_flags);
1265
1266func_exit:
1267    *status_flags = cpu_to_le32(*status_flags);
1268}
1269
1270static inline void
1271e1000e_write_lgcy_rx_descr(E1000ECore *core, uint8_t *desc,
1272                           struct NetRxPkt *pkt,
1273                           const E1000E_RSSInfo *rss_info,
1274                           uint16_t length)
1275{
1276    uint32_t status_flags, rss, mrq;
1277    uint16_t ip_id;
1278
1279    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1280
1281    memset(d, 0, sizeof(*d));
1282
1283    assert(!rss_info->enabled);
1284
1285    d->length = cpu_to_le16(length);
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}
1295
1296static inline void
1297e1000e_write_ext_rx_descr(E1000ECore *core, uint8_t *desc,
1298                          struct NetRxPkt *pkt,
1299                          const E1000E_RSSInfo *rss_info,
1300                          uint16_t length)
1301{
1302    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1303
1304    memset(d, 0, sizeof(*d));
1305
1306    d->wb.upper.length = cpu_to_le16(length);
1307
1308    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1309                             rss_info,
1310                             &d->wb.lower.hi_dword.rss,
1311                             &d->wb.lower.mrq,
1312                             &d->wb.upper.status_error,
1313                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1314                             &d->wb.upper.vlan);
1315}
1316
1317static inline void
1318e1000e_write_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1319                         struct NetRxPkt *pkt,
1320                         const E1000E_RSSInfo *rss_info,
1321                         size_t ps_hdr_len,
1322                         uint16_t(*written)[MAX_PS_BUFFERS])
1323{
1324    int i;
1325    union e1000_rx_desc_packet_split *d =
1326        (union e1000_rx_desc_packet_split *) desc;
1327
1328    memset(d, 0, sizeof(*d));
1329
1330    d->wb.middle.length0 = cpu_to_le16((*written)[0]);
1331
1332    for (i = 0; i < PS_PAGE_BUFFERS; i++) {
1333        d->wb.upper.length[i] = cpu_to_le16((*written)[i + 1]);
1334    }
1335
1336    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1337                             rss_info,
1338                             &d->wb.lower.hi_dword.rss,
1339                             &d->wb.lower.mrq,
1340                             &d->wb.middle.status_error,
1341                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1342                             &d->wb.middle.vlan);
1343
1344    d->wb.upper.header_status =
1345        cpu_to_le16(ps_hdr_len | (ps_hdr_len ? E1000_RXDPS_HDRSTAT_HDRSP : 0));
1346
1347    trace_e1000e_rx_desc_ps_write((*written)[0], (*written)[1],
1348                                  (*written)[2], (*written)[3]);
1349}
1350
1351static inline void
1352e1000e_write_rx_descr(E1000ECore *core, uint8_t *desc,
1353struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info,
1354    size_t ps_hdr_len, uint16_t(*written)[MAX_PS_BUFFERS])
1355{
1356    if (e1000e_rx_use_legacy_descriptor(core)) {
1357        assert(ps_hdr_len == 0);
1358        e1000e_write_lgcy_rx_descr(core, desc, pkt, rss_info, (*written)[0]);
1359    } else {
1360        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1361            e1000e_write_ps_rx_descr(core, desc, pkt, rss_info,
1362                                      ps_hdr_len, written);
1363        } else {
1364            assert(ps_hdr_len == 0);
1365            e1000e_write_ext_rx_descr(core, desc, pkt, rss_info,
1366                                       (*written)[0]);
1367        }
1368    }
1369}
1370
1371typedef struct e1000e_ba_state_st {
1372    uint16_t written[MAX_PS_BUFFERS];
1373    uint8_t cur_idx;
1374} e1000e_ba_state;
1375
1376static inline void
1377e1000e_write_hdr_to_rx_buffers(E1000ECore *core,
1378                               hwaddr (*ba)[MAX_PS_BUFFERS],
1379                               e1000e_ba_state *bastate,
1380                               const char *data,
1381                               dma_addr_t data_len)
1382{
1383    assert(data_len <= core->rxbuf_sizes[0] - bastate->written[0]);
1384
1385    pci_dma_write(core->owner, (*ba)[0] + bastate->written[0], data, data_len);
1386    bastate->written[0] += data_len;
1387
1388    bastate->cur_idx = 1;
1389}
1390
1391static void
1392e1000e_write_to_rx_buffers(E1000ECore *core,
1393                           hwaddr (*ba)[MAX_PS_BUFFERS],
1394                           e1000e_ba_state *bastate,
1395                           const char *data,
1396                           dma_addr_t data_len)
1397{
1398    while (data_len > 0) {
1399        uint32_t cur_buf_len = core->rxbuf_sizes[bastate->cur_idx];
1400        uint32_t cur_buf_bytes_left = cur_buf_len -
1401                                      bastate->written[bastate->cur_idx];
1402        uint32_t bytes_to_write = MIN(data_len, cur_buf_bytes_left);
1403
1404        trace_e1000e_rx_desc_buff_write(bastate->cur_idx,
1405                                        (*ba)[bastate->cur_idx],
1406                                        bastate->written[bastate->cur_idx],
1407                                        data,
1408                                        bytes_to_write);
1409
1410        pci_dma_write(core->owner,
1411            (*ba)[bastate->cur_idx] + bastate->written[bastate->cur_idx],
1412            data, bytes_to_write);
1413
1414        bastate->written[bastate->cur_idx] += bytes_to_write;
1415        data += bytes_to_write;
1416        data_len -= bytes_to_write;
1417
1418        if (bastate->written[bastate->cur_idx] == cur_buf_len) {
1419            bastate->cur_idx++;
1420        }
1421
1422        assert(bastate->cur_idx < MAX_PS_BUFFERS);
1423    }
1424}
1425
1426static void
1427e1000e_update_rx_stats(E1000ECore *core,
1428                       size_t data_size,
1429                       size_t data_fcs_size)
1430{
1431    e1000x_update_rx_total_stats(core->mac, data_size, data_fcs_size);
1432
1433    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1434    case ETH_PKT_BCAST:
1435        e1000x_inc_reg_if_not_full(core->mac, BPRC);
1436        break;
1437
1438    case ETH_PKT_MCAST:
1439        e1000x_inc_reg_if_not_full(core->mac, MPRC);
1440        break;
1441
1442    default:
1443        break;
1444    }
1445}
1446
1447static inline bool
1448e1000e_rx_descr_threshold_hit(E1000ECore *core, const E1000E_RingInfo *rxi)
1449{
1450    return e1000e_ring_free_descr_num(core, rxi) ==
1451           e1000e_ring_len(core, rxi) >> core->rxbuf_min_shift;
1452}
1453
1454static bool
1455e1000e_do_ps(E1000ECore *core, struct NetRxPkt *pkt, size_t *hdr_len)
1456{
1457    bool isip4, isip6, isudp, istcp;
1458    bool fragment;
1459
1460    if (!e1000e_rx_use_ps_descriptor(core)) {
1461        return false;
1462    }
1463
1464    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1465
1466    if (isip4) {
1467        fragment = net_rx_pkt_get_ip4_info(pkt)->fragment;
1468    } else if (isip6) {
1469        fragment = net_rx_pkt_get_ip6_info(pkt)->fragment;
1470    } else {
1471        return false;
1472    }
1473
1474    if (fragment && (core->mac[RFCTL] & E1000_RFCTL_IPFRSP_DIS)) {
1475        return false;
1476    }
1477
1478    if (!fragment && (isudp || istcp)) {
1479        *hdr_len = net_rx_pkt_get_l5_hdr_offset(pkt);
1480    } else {
1481        *hdr_len = net_rx_pkt_get_l4_hdr_offset(pkt);
1482    }
1483
1484    if ((*hdr_len > core->rxbuf_sizes[0]) ||
1485        (*hdr_len > net_rx_pkt_get_total_len(pkt))) {
1486        return false;
1487    }
1488
1489    return true;
1490}
1491
1492static void
1493e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt,
1494                             const E1000E_RxRing *rxr,
1495                             const E1000E_RSSInfo *rss_info)
1496{
1497    PCIDevice *d = core->owner;
1498    dma_addr_t base;
1499    uint8_t desc[E1000_MAX_RX_DESC_LEN];
1500    size_t desc_size;
1501    size_t desc_offset = 0;
1502    size_t iov_ofs = 0;
1503
1504    struct iovec *iov = net_rx_pkt_get_iovec(pkt);
1505    size_t size = net_rx_pkt_get_total_len(pkt);
1506    size_t total_size = size + e1000x_fcs_len(core->mac);
1507    const E1000E_RingInfo *rxi;
1508    size_t ps_hdr_len = 0;
1509    bool do_ps = e1000e_do_ps(core, pkt, &ps_hdr_len);
1510
1511    rxi = rxr->i;
1512
1513    do {
1514        hwaddr ba[MAX_PS_BUFFERS];
1515        e1000e_ba_state bastate = { { 0 } };
1516        bool is_last = false;
1517        bool is_first = true;
1518
1519        desc_size = total_size - desc_offset;
1520
1521        if (desc_size > core->rx_desc_buf_size) {
1522            desc_size = core->rx_desc_buf_size;
1523        }
1524
1525        base = e1000e_ring_head_descr(core, rxi);
1526
1527        pci_dma_read(d, base, &desc, core->rx_desc_len);
1528
1529        trace_e1000e_rx_descr(rxi->idx, base, core->rx_desc_len);
1530
1531        e1000e_read_rx_descr(core, desc, &ba);
1532
1533        if (ba[0]) {
1534            if (desc_offset < size) {
1535                static const uint32_t fcs_pad;
1536                size_t iov_copy;
1537                size_t copy_size = size - desc_offset;
1538                if (copy_size > core->rx_desc_buf_size) {
1539                    copy_size = core->rx_desc_buf_size;
1540                }
1541
1542                /* For PS mode copy the packet header first */
1543                if (do_ps) {
1544                    if (is_first) {
1545                        size_t ps_hdr_copied = 0;
1546                        do {
1547                            iov_copy = MIN(ps_hdr_len - ps_hdr_copied,
1548                                           iov->iov_len - iov_ofs);
1549
1550                            e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1551                                                      iov->iov_base, iov_copy);
1552
1553                            copy_size -= iov_copy;
1554                            ps_hdr_copied += iov_copy;
1555
1556                            iov_ofs += iov_copy;
1557                            if (iov_ofs == iov->iov_len) {
1558                                iov++;
1559                                iov_ofs = 0;
1560                            }
1561                        } while (ps_hdr_copied < ps_hdr_len);
1562
1563                        is_first = false;
1564                    } else {
1565                        /* Leave buffer 0 of each descriptor except first */
1566                        /* empty as per spec 7.1.5.1                      */
1567                        e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1568                                                       NULL, 0);
1569                    }
1570                }
1571
1572                /* Copy packet payload */
1573                while (copy_size) {
1574                    iov_copy = MIN(copy_size, iov->iov_len - iov_ofs);
1575
1576                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1577                                            iov->iov_base + iov_ofs, iov_copy);
1578
1579                    copy_size -= iov_copy;
1580                    iov_ofs += iov_copy;
1581                    if (iov_ofs == iov->iov_len) {
1582                        iov++;
1583                        iov_ofs = 0;
1584                    }
1585                }
1586
1587                if (desc_offset + desc_size >= total_size) {
1588                    /* Simulate FCS checksum presence in the last descriptor */
1589                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1590                          (const char *) &fcs_pad, e1000x_fcs_len(core->mac));
1591                }
1592            }
1593            desc_offset += desc_size;
1594            if (desc_offset >= total_size) {
1595                is_last = true;
1596            }
1597        } else { /* as per intel docs; skip descriptors with null buf addr */
1598            trace_e1000e_rx_null_descriptor();
1599        }
1600
1601        e1000e_write_rx_descr(core, desc, is_last ? core->rx_pkt : NULL,
1602                           rss_info, do_ps ? ps_hdr_len : 0, &bastate.written);
1603        pci_dma_write(d, base, &desc, core->rx_desc_len);
1604
1605        e1000e_ring_advance(core, rxi,
1606                            core->rx_desc_len / E1000_MIN_RX_DESC_LEN);
1607
1608    } while (desc_offset < total_size);
1609
1610    e1000e_update_rx_stats(core, size, total_size);
1611}
1612
1613static inline void
1614e1000e_rx_fix_l4_csum(E1000ECore *core, struct NetRxPkt *pkt)
1615{
1616    if (net_rx_pkt_has_virt_hdr(pkt)) {
1617        struct virtio_net_hdr *vhdr = net_rx_pkt_get_vhdr(pkt);
1618
1619        if (vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1620            net_rx_pkt_fix_l4_csum(pkt);
1621        }
1622    }
1623}
1624
1625ssize_t
1626e1000e_receive_iov(E1000ECore *core, const struct iovec *iov, int iovcnt)
1627{
1628    static const int maximum_ethernet_hdr_len = (14 + 4);
1629    /* Min. octets in an ethernet frame sans FCS */
1630    static const int min_buf_size = 60;
1631
1632    uint32_t n = 0;
1633    uint8_t min_buf[min_buf_size];
1634    struct iovec min_iov;
1635    uint8_t *filter_buf;
1636    size_t size, orig_size;
1637    size_t iov_ofs = 0;
1638    E1000E_RxRing rxr;
1639    E1000E_RSSInfo rss_info;
1640    size_t total_size;
1641    ssize_t retval;
1642    bool rdmts_hit;
1643
1644    trace_e1000e_rx_receive_iov(iovcnt);
1645
1646    if (!e1000x_hw_rx_enabled(core->mac)) {
1647        return -1;
1648    }
1649
1650    /* Pull virtio header in */
1651    if (core->has_vnet) {
1652        net_rx_pkt_set_vhdr_iovec(core->rx_pkt, iov, iovcnt);
1653        iov_ofs = sizeof(struct virtio_net_hdr);
1654    }
1655
1656    filter_buf = iov->iov_base + iov_ofs;
1657    orig_size = iov_size(iov, iovcnt);
1658    size = orig_size - iov_ofs;
1659
1660    /* Pad to minimum Ethernet frame length */
1661    if (size < sizeof(min_buf)) {
1662        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, size);
1663        memset(&min_buf[size], 0, sizeof(min_buf) - size);
1664        e1000x_inc_reg_if_not_full(core->mac, RUC);
1665        min_iov.iov_base = filter_buf = min_buf;
1666        min_iov.iov_len = size = sizeof(min_buf);
1667        iovcnt = 1;
1668        iov = &min_iov;
1669        iov_ofs = 0;
1670    } else if (iov->iov_len < maximum_ethernet_hdr_len) {
1671        /* This is very unlikely, but may happen. */
1672        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, maximum_ethernet_hdr_len);
1673        filter_buf = min_buf;
1674    }
1675
1676    /* Discard oversized packets if !LPE and !SBP. */
1677    if (e1000x_is_oversized(core->mac, size)) {
1678        return orig_size;
1679    }
1680
1681    net_rx_pkt_set_packet_type(core->rx_pkt,
1682        get_eth_packet_type(PKT_GET_ETH_HDR(filter_buf)));
1683
1684    if (!e1000e_receive_filter(core, filter_buf, size)) {
1685        trace_e1000e_rx_flt_dropped();
1686        return orig_size;
1687    }
1688
1689    net_rx_pkt_attach_iovec_ex(core->rx_pkt, iov, iovcnt, iov_ofs,
1690                               e1000x_vlan_enabled(core->mac), core->vet);
1691
1692    e1000e_rss_parse_packet(core, core->rx_pkt, &rss_info);
1693    e1000e_rx_ring_init(core, &rxr, rss_info.queue);
1694
1695    trace_e1000e_rx_rss_dispatched_to_queue(rxr.i->idx);
1696
1697    total_size = net_rx_pkt_get_total_len(core->rx_pkt) +
1698        e1000x_fcs_len(core->mac);
1699
1700    if (e1000e_has_rxbufs(core, rxr.i, total_size)) {
1701        e1000e_rx_fix_l4_csum(core, core->rx_pkt);
1702
1703        e1000e_write_packet_to_guest(core, core->rx_pkt, &rxr, &rss_info);
1704
1705        retval = orig_size;
1706
1707        /* Perform small receive detection (RSRPD) */
1708        if (total_size < core->mac[RSRPD]) {
1709            n |= E1000_ICS_SRPD;
1710        }
1711
1712        /* Perform ACK receive detection */
1713        if (e1000e_is_tcp_ack(core, core->rx_pkt)) {
1714            n |= E1000_ICS_ACK;
1715        }
1716
1717        /* Check if receive descriptor minimum threshold hit */
1718        rdmts_hit = e1000e_rx_descr_threshold_hit(core, rxr.i);
1719        n |= e1000e_rx_wb_interrupt_cause(core, rxr.i->idx, rdmts_hit);
1720
1721        trace_e1000e_rx_written_to_guest(n);
1722    } else {
1723        n |= E1000_ICS_RXO;
1724        retval = 0;
1725
1726        trace_e1000e_rx_not_written_to_guest(n);
1727    }
1728
1729    if (!e1000e_intrmgr_delay_rx_causes(core, &n)) {
1730        trace_e1000e_rx_interrupt_set(n);
1731        e1000e_set_interrupt_cause(core, n);
1732    } else {
1733        trace_e1000e_rx_interrupt_delayed(n);
1734    }
1735
1736    return retval;
1737}
1738
1739static inline bool
1740e1000e_have_autoneg(E1000ECore *core)
1741{
1742    return core->phy[0][PHY_CTRL] & MII_CR_AUTO_NEG_EN;
1743}
1744
1745static void e1000e_update_flowctl_status(E1000ECore *core)
1746{
1747    if (e1000e_have_autoneg(core) &&
1748        core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE) {
1749        trace_e1000e_link_autoneg_flowctl(true);
1750        core->mac[CTRL] |= E1000_CTRL_TFCE | E1000_CTRL_RFCE;
1751    } else {
1752        trace_e1000e_link_autoneg_flowctl(false);
1753    }
1754}
1755
1756static inline void
1757e1000e_link_down(E1000ECore *core)
1758{
1759    e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1760    e1000e_update_flowctl_status(core);
1761}
1762
1763static inline void
1764e1000e_set_phy_ctrl(E1000ECore *core, int index, uint16_t val)
1765{
1766    /* bits 0-5 reserved; MII_CR_[RESTART_AUTO_NEG,RESET] are self clearing */
1767    core->phy[0][PHY_CTRL] = val & ~(0x3f |
1768                                     MII_CR_RESET |
1769                                     MII_CR_RESTART_AUTO_NEG);
1770
1771    if ((val & MII_CR_RESTART_AUTO_NEG) &&
1772        e1000e_have_autoneg(core)) {
1773        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1774    }
1775}
1776
1777static void
1778e1000e_set_phy_oem_bits(E1000ECore *core, int index, uint16_t val)
1779{
1780    core->phy[0][PHY_OEM_BITS] = val & ~BIT(10);
1781
1782    if (val & BIT(10)) {
1783        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1784    }
1785}
1786
1787static void
1788e1000e_set_phy_page(E1000ECore *core, int index, uint16_t val)
1789{
1790    core->phy[0][PHY_PAGE] = val & PHY_PAGE_RW_MASK;
1791}
1792
1793void
1794e1000e_core_set_link_status(E1000ECore *core)
1795{
1796    NetClientState *nc = qemu_get_queue(core->owner_nic);
1797    uint32_t old_status = core->mac[STATUS];
1798
1799    trace_e1000e_link_status_changed(nc->link_down ? false : true);
1800
1801    if (nc->link_down) {
1802        e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1803    } else {
1804        if (e1000e_have_autoneg(core) &&
1805            !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
1806            e1000x_restart_autoneg(core->mac, core->phy[0],
1807                                   core->autoneg_timer);
1808        } else {
1809            e1000x_update_regs_on_link_up(core->mac, core->phy[0]);
1810        }
1811    }
1812
1813    if (core->mac[STATUS] != old_status) {
1814        e1000e_set_interrupt_cause(core, E1000_ICR_LSC);
1815    }
1816}
1817
1818static void
1819e1000e_set_ctrl(E1000ECore *core, int index, uint32_t val)
1820{
1821    trace_e1000e_core_ctrl_write(index, val);
1822
1823    /* RST is self clearing */
1824    core->mac[CTRL] = val & ~E1000_CTRL_RST;
1825    core->mac[CTRL_DUP] = core->mac[CTRL];
1826
1827    trace_e1000e_link_set_params(
1828        !!(val & E1000_CTRL_ASDE),
1829        (val & E1000_CTRL_SPD_SEL) >> E1000_CTRL_SPD_SHIFT,
1830        !!(val & E1000_CTRL_FRCSPD),
1831        !!(val & E1000_CTRL_FRCDPX),
1832        !!(val & E1000_CTRL_RFCE),
1833        !!(val & E1000_CTRL_TFCE));
1834
1835    if (val & E1000_CTRL_RST) {
1836        trace_e1000e_core_ctrl_sw_reset();
1837        e1000x_reset_mac_addr(core->owner_nic, core->mac, core->permanent_mac);
1838    }
1839
1840    if (val & E1000_CTRL_PHY_RST) {
1841        trace_e1000e_core_ctrl_phy_reset();
1842        core->mac[STATUS] |= E1000_STATUS_PHYRA;
1843    }
1844}
1845
1846static void
1847e1000e_set_rfctl(E1000ECore *core, int index, uint32_t val)
1848{
1849    trace_e1000e_rx_set_rfctl(val);
1850
1851    if (!(val & E1000_RFCTL_ISCSI_DIS)) {
1852        trace_e1000e_wrn_iscsi_filtering_not_supported();
1853    }
1854
1855    if (!(val & E1000_RFCTL_NFSW_DIS)) {
1856        trace_e1000e_wrn_nfsw_filtering_not_supported();
1857    }
1858
1859    if (!(val & E1000_RFCTL_NFSR_DIS)) {
1860        trace_e1000e_wrn_nfsr_filtering_not_supported();
1861    }
1862
1863    core->mac[RFCTL] = val;
1864}
1865
1866static void
1867e1000e_calc_per_desc_buf_size(E1000ECore *core)
1868{
1869    int i;
1870    core->rx_desc_buf_size = 0;
1871
1872    for (i = 0; i < ARRAY_SIZE(core->rxbuf_sizes); i++) {
1873        core->rx_desc_buf_size += core->rxbuf_sizes[i];
1874    }
1875}
1876
1877static void
1878e1000e_parse_rxbufsize(E1000ECore *core)
1879{
1880    uint32_t rctl = core->mac[RCTL];
1881
1882    memset(core->rxbuf_sizes, 0, sizeof(core->rxbuf_sizes));
1883
1884    if (rctl & E1000_RCTL_DTYP_MASK) {
1885        uint32_t bsize;
1886
1887        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE0_MASK;
1888        core->rxbuf_sizes[0] = (bsize >> E1000_PSRCTL_BSIZE0_SHIFT) * 128;
1889
1890        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE1_MASK;
1891        core->rxbuf_sizes[1] = (bsize >> E1000_PSRCTL_BSIZE1_SHIFT) * 1024;
1892
1893        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE2_MASK;
1894        core->rxbuf_sizes[2] = (bsize >> E1000_PSRCTL_BSIZE2_SHIFT) * 1024;
1895
1896        bsize = core->mac[PSRCTL] & E1000_PSRCTL_BSIZE3_MASK;
1897        core->rxbuf_sizes[3] = (bsize >> E1000_PSRCTL_BSIZE3_SHIFT) * 1024;
1898    } else if (rctl & E1000_RCTL_FLXBUF_MASK) {
1899        int flxbuf = rctl & E1000_RCTL_FLXBUF_MASK;
1900        core->rxbuf_sizes[0] = (flxbuf >> E1000_RCTL_FLXBUF_SHIFT) * 1024;
1901    } else {
1902        core->rxbuf_sizes[0] = e1000x_rxbufsize(rctl);
1903    }
1904
1905    trace_e1000e_rx_desc_buff_sizes(core->rxbuf_sizes[0], core->rxbuf_sizes[1],
1906                                    core->rxbuf_sizes[2], core->rxbuf_sizes[3]);
1907
1908    e1000e_calc_per_desc_buf_size(core);
1909}
1910
1911static void
1912e1000e_calc_rxdesclen(E1000ECore *core)
1913{
1914    if (e1000e_rx_use_legacy_descriptor(core)) {
1915        core->rx_desc_len = sizeof(struct e1000_rx_desc);
1916    } else {
1917        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1918            core->rx_desc_len = sizeof(union e1000_rx_desc_packet_split);
1919        } else {
1920            core->rx_desc_len = sizeof(union e1000_rx_desc_extended);
1921        }
1922    }
1923    trace_e1000e_rx_desc_len(core->rx_desc_len);
1924}
1925
1926static void
1927e1000e_set_rx_control(E1000ECore *core, int index, uint32_t val)
1928{
1929    core->mac[RCTL] = val;
1930    trace_e1000e_rx_set_rctl(core->mac[RCTL]);
1931
1932    if (val & E1000_RCTL_EN) {
1933        e1000e_parse_rxbufsize(core);
1934        e1000e_calc_rxdesclen(core);
1935        core->rxbuf_min_shift = ((val / E1000_RCTL_RDMTS_QUAT) & 3) + 1 +
1936                                E1000_RING_DESC_LEN_SHIFT;
1937
1938        e1000e_start_recv(core);
1939    }
1940}
1941
1942static
1943void(*e1000e_phyreg_writeops[E1000E_PHY_PAGES][E1000E_PHY_PAGE_SIZE])
1944(E1000ECore *, int, uint16_t) = {
1945    [0] = {
1946        [PHY_CTRL]     = e1000e_set_phy_ctrl,
1947        [PHY_PAGE]     = e1000e_set_phy_page,
1948        [PHY_OEM_BITS] = e1000e_set_phy_oem_bits
1949    }
1950};
1951
1952static inline void
1953e1000e_clear_ims_bits(E1000ECore *core, uint32_t bits)
1954{
1955    trace_e1000e_irq_clear_ims(bits, core->mac[IMS], core->mac[IMS] & ~bits);
1956    core->mac[IMS] &= ~bits;
1957}
1958
1959static inline bool
1960e1000e_postpone_interrupt(bool *interrupt_pending,
1961                           E1000IntrDelayTimer *timer)
1962{
1963    if (timer->running) {
1964        trace_e1000e_irq_postponed_by_xitr(timer->delay_reg << 2);
1965
1966        *interrupt_pending = true;
1967        return true;
1968    }
1969
1970    if (timer->core->mac[timer->delay_reg] != 0) {
1971        e1000e_intrmgr_rearm_timer(timer);
1972    }
1973
1974    return false;
1975}
1976
1977static inline bool
1978e1000e_itr_should_postpone(E1000ECore *core)
1979{
1980    return e1000e_postpone_interrupt(&core->itr_intr_pending, &core->itr);
1981}
1982
1983static inline bool
1984e1000e_eitr_should_postpone(E1000ECore *core, int idx)
1985{
1986    return e1000e_postpone_interrupt(&core->eitr_intr_pending[idx],
1987                                     &core->eitr[idx]);
1988}
1989
1990static void
1991e1000e_msix_notify_one(E1000ECore *core, uint32_t cause, uint32_t int_cfg)
1992{
1993    uint32_t effective_eiac;
1994
1995    if (E1000_IVAR_ENTRY_VALID(int_cfg)) {
1996        uint32_t vec = E1000_IVAR_ENTRY_VEC(int_cfg);
1997        if (vec < E1000E_MSIX_VEC_NUM) {
1998            if (!e1000e_eitr_should_postpone(core, vec)) {
1999                trace_e1000e_irq_msix_notify_vec(vec);
2000                msix_notify(core->owner, vec);
2001            }
2002        } else {
2003            trace_e1000e_wrn_msix_vec_wrong(cause, int_cfg);
2004        }
2005    } else {
2006        trace_e1000e_wrn_msix_invalid(cause, int_cfg);
2007    }
2008
2009    if (core->mac[CTRL_EXT] & E1000_CTRL_EXT_EIAME) {
2010        trace_e1000e_irq_ims_clear_eiame(core->mac[IAM], cause);
2011        e1000e_clear_ims_bits(core, core->mac[IAM] & cause);
2012    }
2013
2014    trace_e1000e_irq_icr_clear_eiac(core->mac[ICR], core->mac[EIAC]);
2015
2016    if (core->mac[EIAC] & E1000_ICR_OTHER) {
2017        effective_eiac = (core->mac[EIAC] & E1000_EIAC_MASK) |
2018                         E1000_ICR_OTHER_CAUSES;
2019    } else {
2020        effective_eiac = core->mac[EIAC] & E1000_EIAC_MASK;
2021    }
2022    core->mac[ICR] &= ~effective_eiac;
2023}
2024
2025static void
2026e1000e_msix_notify(E1000ECore *core, uint32_t causes)
2027{
2028    if (causes & E1000_ICR_RXQ0) {
2029        e1000e_msix_notify_one(core, E1000_ICR_RXQ0,
2030                               E1000_IVAR_RXQ0(core->mac[IVAR]));
2031    }
2032
2033    if (causes & E1000_ICR_RXQ1) {
2034        e1000e_msix_notify_one(core, E1000_ICR_RXQ1,
2035                               E1000_IVAR_RXQ1(core->mac[IVAR]));
2036    }
2037
2038    if (causes & E1000_ICR_TXQ0) {
2039        e1000e_msix_notify_one(core, E1000_ICR_TXQ0,
2040                               E1000_IVAR_TXQ0(core->mac[IVAR]));
2041    }
2042
2043    if (causes & E1000_ICR_TXQ1) {
2044        e1000e_msix_notify_one(core, E1000_ICR_TXQ1,
2045                               E1000_IVAR_TXQ1(core->mac[IVAR]));
2046    }
2047
2048    if (causes & E1000_ICR_OTHER) {
2049        e1000e_msix_notify_one(core, E1000_ICR_OTHER,
2050                               E1000_IVAR_OTHER(core->mac[IVAR]));
2051    }
2052}
2053
2054static void
2055e1000e_msix_clear_one(E1000ECore *core, uint32_t cause, uint32_t int_cfg)
2056{
2057    if (E1000_IVAR_ENTRY_VALID(int_cfg)) {
2058        uint32_t vec = E1000_IVAR_ENTRY_VEC(int_cfg);
2059        if (vec < E1000E_MSIX_VEC_NUM) {
2060            trace_e1000e_irq_msix_pending_clearing(cause, int_cfg, vec);
2061            msix_clr_pending(core->owner, vec);
2062        } else {
2063            trace_e1000e_wrn_msix_vec_wrong(cause, int_cfg);
2064        }
2065    } else {
2066        trace_e1000e_wrn_msix_invalid(cause, int_cfg);
2067    }
2068}
2069
2070static void
2071e1000e_msix_clear(E1000ECore *core, uint32_t causes)
2072{
2073    if (causes & E1000_ICR_RXQ0) {
2074        e1000e_msix_clear_one(core, E1000_ICR_RXQ0,
2075                              E1000_IVAR_RXQ0(core->mac[IVAR]));
2076    }
2077
2078    if (causes & E1000_ICR_RXQ1) {
2079        e1000e_msix_clear_one(core, E1000_ICR_RXQ1,
2080                              E1000_IVAR_RXQ1(core->mac[IVAR]));
2081    }
2082
2083    if (causes & E1000_ICR_TXQ0) {
2084        e1000e_msix_clear_one(core, E1000_ICR_TXQ0,
2085                              E1000_IVAR_TXQ0(core->mac[IVAR]));
2086    }
2087
2088    if (causes & E1000_ICR_TXQ1) {
2089        e1000e_msix_clear_one(core, E1000_ICR_TXQ1,
2090                              E1000_IVAR_TXQ1(core->mac[IVAR]));
2091    }
2092
2093    if (causes & E1000_ICR_OTHER) {
2094        e1000e_msix_clear_one(core, E1000_ICR_OTHER,
2095                              E1000_IVAR_OTHER(core->mac[IVAR]));
2096    }
2097}
2098
2099static inline void
2100e1000e_fix_icr_asserted(E1000ECore *core)
2101{
2102    core->mac[ICR] &= ~E1000_ICR_ASSERTED;
2103    if (core->mac[ICR]) {
2104        core->mac[ICR] |= E1000_ICR_ASSERTED;
2105    }
2106
2107    trace_e1000e_irq_fix_icr_asserted(core->mac[ICR]);
2108}
2109
2110static void
2111e1000e_send_msi(E1000ECore *core, bool msix)
2112{
2113    uint32_t causes = core->mac[ICR] & core->mac[IMS] & ~E1000_ICR_ASSERTED;
2114
2115    if (msix) {
2116        e1000e_msix_notify(core, causes);
2117    } else {
2118        if (!e1000e_itr_should_postpone(core)) {
2119            trace_e1000e_irq_msi_notify(causes);
2120            msi_notify(core->owner, 0);
2121        }
2122    }
2123}
2124
2125static void
2126e1000e_update_interrupt_state(E1000ECore *core)
2127{
2128    bool interrupts_pending;
2129    bool is_msix = msix_enabled(core->owner);
2130
2131    /* Set ICR[OTHER] for MSI-X */
2132    if (is_msix) {
2133        if (core->mac[ICR] & core->mac[IMS] & E1000_ICR_OTHER_CAUSES) {
2134            core->mac[ICR] |= E1000_ICR_OTHER;
2135            trace_e1000e_irq_add_msi_other(core->mac[ICR]);
2136        }
2137    }
2138
2139    e1000e_fix_icr_asserted(core);
2140
2141    /*
2142     * Make sure ICR and ICS registers have the same value.
2143     * The spec says that the ICS register is write-only.  However in practice,
2144     * on real hardware ICS is readable, and for reads it has the same value as
2145     * ICR (except that ICS does not have the clear on read behaviour of ICR).
2146     *
2147     * The VxWorks PRO/1000 driver uses this behaviour.
2148     */
2149    core->mac[ICS] = core->mac[ICR];
2150
2151    interrupts_pending = (core->mac[IMS] & core->mac[ICR]) ? true : false;
2152
2153    trace_e1000e_irq_pending_interrupts(core->mac[ICR] & core->mac[IMS],
2154                                        core->mac[ICR], core->mac[IMS]);
2155
2156    if (is_msix || msi_enabled(core->owner)) {
2157        if (interrupts_pending) {
2158            e1000e_send_msi(core, is_msix);
2159        }
2160    } else {
2161        if (interrupts_pending) {
2162            if (!e1000e_itr_should_postpone(core)) {
2163                e1000e_raise_legacy_irq(core);
2164            }
2165        } else {
2166            e1000e_lower_legacy_irq(core);
2167        }
2168    }
2169}
2170
2171static inline void
2172e1000e_set_interrupt_cause(E1000ECore *core, uint32_t val)
2173{
2174    trace_e1000e_irq_set_cause_entry(val, core->mac[ICR]);
2175
2176    val |= e1000e_intmgr_collect_delayed_causes(core);
2177    core->mac[ICR] |= val;
2178
2179    trace_e1000e_irq_set_cause_exit(val, core->mac[ICR]);
2180
2181    e1000e_update_interrupt_state(core);
2182}
2183
2184static inline void
2185e1000e_autoneg_timer(void *opaque)
2186{
2187    E1000ECore *core = opaque;
2188    if (!qemu_get_queue(core->owner_nic)->link_down) {
2189        e1000x_update_regs_on_autoneg_done(core->mac, core->phy[0]);
2190        e1000e_update_flowctl_status(core);
2191        /* signal link status change to the guest */
2192        e1000e_set_interrupt_cause(core, E1000_ICR_LSC);
2193    }
2194}
2195
2196static inline uint16_t
2197e1000e_get_reg_index_with_offset(const uint16_t *mac_reg_access, hwaddr addr)
2198{
2199    uint16_t index = (addr & 0x1ffff) >> 2;
2200    return index + (mac_reg_access[index] & 0xfffe);
2201}
2202
2203static const char e1000e_phy_regcap[E1000E_PHY_PAGES][0x20] = {
2204    [0] = {
2205        [PHY_CTRL]          = PHY_ANYPAGE | PHY_RW,
2206        [PHY_STATUS]        = PHY_ANYPAGE | PHY_R,
2207        [PHY_ID1]           = PHY_ANYPAGE | PHY_R,
2208        [PHY_ID2]           = PHY_ANYPAGE | PHY_R,
2209        [PHY_AUTONEG_ADV]   = PHY_ANYPAGE | PHY_RW,
2210        [PHY_LP_ABILITY]    = PHY_ANYPAGE | PHY_R,
2211        [PHY_AUTONEG_EXP]   = PHY_ANYPAGE | PHY_R,
2212        [PHY_NEXT_PAGE_TX]  = PHY_ANYPAGE | PHY_RW,
2213        [PHY_LP_NEXT_PAGE]  = PHY_ANYPAGE | PHY_R,
2214        [PHY_1000T_CTRL]    = PHY_ANYPAGE | PHY_RW,
2215        [PHY_1000T_STATUS]  = PHY_ANYPAGE | PHY_R,
2216        [PHY_EXT_STATUS]    = PHY_ANYPAGE | PHY_R,
2217        [PHY_PAGE]          = PHY_ANYPAGE | PHY_RW,
2218
2219        [PHY_COPPER_CTRL1]      = PHY_RW,
2220        [PHY_COPPER_STAT1]      = PHY_R,
2221        [PHY_COPPER_CTRL3]      = PHY_RW,
2222        [PHY_RX_ERR_CNTR]       = PHY_R,
2223        [PHY_OEM_BITS]          = PHY_RW,
2224        [PHY_BIAS_1]            = PHY_RW,
2225        [PHY_BIAS_2]            = PHY_RW,
2226        [PHY_COPPER_INT_ENABLE] = PHY_RW,
2227        [PHY_COPPER_STAT2]      = PHY_R,
2228        [PHY_COPPER_CTRL2]      = PHY_RW
2229    },
2230    [2] = {
2231        [PHY_MAC_CTRL1]         = PHY_RW,
2232        [PHY_MAC_INT_ENABLE]    = PHY_RW,
2233        [PHY_MAC_STAT]          = PHY_R,
2234        [PHY_MAC_CTRL2]         = PHY_RW
2235    },
2236    [3] = {
2237        [PHY_LED_03_FUNC_CTRL1] = PHY_RW,
2238        [PHY_LED_03_POL_CTRL]   = PHY_RW,
2239        [PHY_LED_TIMER_CTRL]    = PHY_RW,
2240        [PHY_LED_45_CTRL]       = PHY_RW
2241    },
2242    [5] = {
2243        [PHY_1000T_SKEW]        = PHY_R,
2244        [PHY_1000T_SWAP]        = PHY_R
2245    },
2246    [6] = {
2247        [PHY_CRC_COUNTERS]      = PHY_R
2248    }
2249};
2250
2251static bool
2252e1000e_phy_reg_check_cap(E1000ECore *core, uint32_t addr,
2253                         char cap, uint8_t *page)
2254{
2255    *page =
2256        (e1000e_phy_regcap[0][addr] & PHY_ANYPAGE) ? 0
2257                                                    : core->phy[0][PHY_PAGE];
2258
2259    if (*page >= E1000E_PHY_PAGES) {
2260        return false;
2261    }
2262
2263    return e1000e_phy_regcap[*page][addr] & cap;
2264}
2265
2266static void
2267e1000e_phy_reg_write(E1000ECore *core, uint8_t page,
2268                     uint32_t addr, uint16_t data)
2269{
2270    assert(page < E1000E_PHY_PAGES);
2271    assert(addr < E1000E_PHY_PAGE_SIZE);
2272
2273    if (e1000e_phyreg_writeops[page][addr]) {
2274        e1000e_phyreg_writeops[page][addr](core, addr, data);
2275    } else {
2276        core->phy[page][addr] = data;
2277    }
2278}
2279
2280static void
2281e1000e_set_mdic(E1000ECore *core, int index, uint32_t val)
2282{
2283    uint32_t data = val & E1000_MDIC_DATA_MASK;
2284    uint32_t addr = ((val & E1000_MDIC_REG_MASK) >> E1000_MDIC_REG_SHIFT);
2285    uint8_t page;
2286
2287    if ((val & E1000_MDIC_PHY_MASK) >> E1000_MDIC_PHY_SHIFT != 1) { /* phy # */
2288        val = core->mac[MDIC] | E1000_MDIC_ERROR;
2289    } else if (val & E1000_MDIC_OP_READ) {
2290        if (!e1000e_phy_reg_check_cap(core, addr, PHY_R, &page)) {
2291            trace_e1000e_core_mdic_read_unhandled(page, addr);
2292            val |= E1000_MDIC_ERROR;
2293        } else {
2294            val = (val ^ data) | core->phy[page][addr];
2295            trace_e1000e_core_mdic_read(page, addr, val);
2296        }
2297    } else if (val & E1000_MDIC_OP_WRITE) {
2298        if (!e1000e_phy_reg_check_cap(core, addr, PHY_W, &page)) {
2299            trace_e1000e_core_mdic_write_unhandled(page, addr);
2300            val |= E1000_MDIC_ERROR;
2301        } else {
2302            trace_e1000e_core_mdic_write(page, addr, data);
2303            e1000e_phy_reg_write(core, page, addr, data);
2304        }
2305    }
2306    core->mac[MDIC] = val | E1000_MDIC_READY;
2307
2308    if (val & E1000_MDIC_INT_EN) {
2309        e1000e_set_interrupt_cause(core, E1000_ICR_MDAC);
2310    }
2311}
2312
2313static void
2314e1000e_set_rdt(E1000ECore *core, int index, uint32_t val)
2315{
2316    core->mac[index] = val & 0xffff;
2317    trace_e1000e_rx_set_rdt(e1000e_mq_queue_idx(RDT0, index), val);
2318    e1000e_start_recv(core);
2319}
2320
2321static void
2322e1000e_set_status(E1000ECore *core, int index, uint32_t val)
2323{
2324    if ((val & E1000_STATUS_PHYRA) == 0) {
2325        core->mac[index] &= ~E1000_STATUS_PHYRA;
2326    }
2327}
2328
2329static void
2330e1000e_set_ctrlext(E1000ECore *core, int index, uint32_t val)
2331{
2332    trace_e1000e_link_set_ext_params(!!(val & E1000_CTRL_EXT_ASDCHK),
2333                                     !!(val & E1000_CTRL_EXT_SPD_BYPS));
2334
2335    /* Zero self-clearing bits */
2336    val &= ~(E1000_CTRL_EXT_ASDCHK | E1000_CTRL_EXT_EE_RST);
2337    core->mac[CTRL_EXT] = val;
2338}
2339
2340static void
2341e1000e_set_pbaclr(E1000ECore *core, int index, uint32_t val)
2342{
2343    int i;
2344
2345    core->mac[PBACLR] = val & E1000_PBACLR_VALID_MASK;
2346
2347    if (msix_enabled(core->owner)) {
2348        return;
2349    }
2350
2351    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
2352        if (core->mac[PBACLR] & BIT(i)) {
2353            msix_clr_pending(core->owner, i);
2354        }
2355    }
2356}
2357
2358static void
2359e1000e_set_fcrth(E1000ECore *core, int index, uint32_t val)
2360{
2361    core->mac[FCRTH] = val & 0xFFF8;
2362}
2363
2364static void
2365e1000e_set_fcrtl(E1000ECore *core, int index, uint32_t val)
2366{
2367    core->mac[FCRTL] = val & 0x8000FFF8;
2368}
2369
2370static inline void
2371e1000e_set_16bit(E1000ECore *core, int index, uint32_t val)
2372{
2373    core->mac[index] = val & 0xffff;
2374}
2375
2376static void
2377e1000e_set_12bit(E1000ECore *core, int index, uint32_t val)
2378{
2379    core->mac[index] = val & 0xfff;
2380}
2381
2382static void
2383e1000e_set_vet(E1000ECore *core, int index, uint32_t val)
2384{
2385    core->mac[VET] = val & 0xffff;
2386    core->vet = le16_to_cpu(core->mac[VET]);
2387    trace_e1000e_vlan_vet(core->vet);
2388}
2389
2390static void
2391e1000e_set_dlen(E1000ECore *core, int index, uint32_t val)
2392{
2393    core->mac[index] = val & E1000_XDLEN_MASK;
2394}
2395
2396static void
2397e1000e_set_dbal(E1000ECore *core, int index, uint32_t val)
2398{
2399    core->mac[index] = val & E1000_XDBAL_MASK;
2400}
2401
2402static void
2403e1000e_set_tctl(E1000ECore *core, int index, uint32_t val)
2404{
2405    E1000E_TxRing txr;
2406    core->mac[index] = val;
2407
2408    if (core->mac[TARC0] & E1000_TARC_ENABLE) {
2409        e1000e_tx_ring_init(core, &txr, 0);
2410        e1000e_start_xmit(core, &txr);
2411    }
2412
2413    if (core->mac[TARC1] & E1000_TARC_ENABLE) {
2414        e1000e_tx_ring_init(core, &txr, 1);
2415        e1000e_start_xmit(core, &txr);
2416    }
2417}
2418
2419static void
2420e1000e_set_tdt(E1000ECore *core, int index, uint32_t val)
2421{
2422    E1000E_TxRing txr;
2423    int qidx = e1000e_mq_queue_idx(TDT, index);
2424    uint32_t tarc_reg = (qidx == 0) ? TARC0 : TARC1;
2425
2426    core->mac[index] = val & 0xffff;
2427
2428    if (core->mac[tarc_reg] & E1000_TARC_ENABLE) {
2429        e1000e_tx_ring_init(core, &txr, qidx);
2430        e1000e_start_xmit(core, &txr);
2431    }
2432}
2433
2434static void
2435e1000e_set_ics(E1000ECore *core, int index, uint32_t val)
2436{
2437    trace_e1000e_irq_write_ics(val);
2438    e1000e_set_interrupt_cause(core, val);
2439}
2440
2441static void
2442e1000e_set_icr(E1000ECore *core, int index, uint32_t val)
2443{
2444    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2445        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2446        trace_e1000e_irq_icr_process_iame();
2447        e1000e_clear_ims_bits(core, core->mac[IAM]);
2448    }
2449
2450    trace_e1000e_irq_icr_write(val, core->mac[ICR], core->mac[ICR] & ~val);
2451    core->mac[ICR] &= ~val;
2452    e1000e_update_interrupt_state(core);
2453}
2454
2455static void
2456e1000e_set_imc(E1000ECore *core, int index, uint32_t val)
2457{
2458    trace_e1000e_irq_ims_clear_set_imc(val);
2459    e1000e_clear_ims_bits(core, val);
2460    e1000e_update_interrupt_state(core);
2461}
2462
2463static void
2464e1000e_set_ims(E1000ECore *core, int index, uint32_t val)
2465{
2466    static const uint32_t ims_ext_mask =
2467        E1000_IMS_RXQ0 | E1000_IMS_RXQ1 |
2468        E1000_IMS_TXQ0 | E1000_IMS_TXQ1 |
2469        E1000_IMS_OTHER;
2470
2471    static const uint32_t ims_valid_mask =
2472        E1000_IMS_TXDW      | E1000_IMS_TXQE    | E1000_IMS_LSC  |
2473        E1000_IMS_RXDMT0    | E1000_IMS_RXO     | E1000_IMS_RXT0 |
2474        E1000_IMS_MDAC      | E1000_IMS_TXD_LOW | E1000_IMS_SRPD |
2475        E1000_IMS_ACK       | E1000_IMS_MNG     | E1000_IMS_RXQ0 |
2476        E1000_IMS_RXQ1      | E1000_IMS_TXQ0    | E1000_IMS_TXQ1 |
2477        E1000_IMS_OTHER;
2478
2479    uint32_t valid_val = val & ims_valid_mask;
2480
2481    trace_e1000e_irq_set_ims(val, core->mac[IMS], core->mac[IMS] | valid_val);
2482    core->mac[IMS] |= valid_val;
2483
2484    if ((valid_val & ims_ext_mask) &&
2485        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_PBA_CLR) &&
2486        msix_enabled(core->owner)) {
2487        e1000e_msix_clear(core, valid_val);
2488    }
2489
2490    if ((valid_val == ims_valid_mask) &&
2491        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_INT_TIMERS_CLEAR_ENA)) {
2492        trace_e1000e_irq_fire_all_timers(val);
2493        e1000e_intrmgr_fire_all_timers(core);
2494    }
2495
2496    e1000e_update_interrupt_state(core);
2497}
2498
2499static void
2500e1000e_set_rdtr(E1000ECore *core, int index, uint32_t val)
2501{
2502    e1000e_set_16bit(core, index, val);
2503
2504    if ((val & E1000_RDTR_FPD) && (core->rdtr.running)) {
2505        trace_e1000e_irq_rdtr_fpd_running();
2506        e1000e_intrmgr_fire_delayed_interrupts(core);
2507    } else {
2508        trace_e1000e_irq_rdtr_fpd_not_running();
2509    }
2510}
2511
2512static void
2513e1000e_set_tidv(E1000ECore *core, int index, uint32_t val)
2514{
2515    e1000e_set_16bit(core, index, val);
2516
2517    if ((val & E1000_TIDV_FPD) && (core->tidv.running)) {
2518        trace_e1000e_irq_tidv_fpd_running();
2519        e1000e_intrmgr_fire_delayed_interrupts(core);
2520    } else {
2521        trace_e1000e_irq_tidv_fpd_not_running();
2522    }
2523}
2524
2525static uint32_t
2526e1000e_mac_readreg(E1000ECore *core, int index)
2527{
2528    return core->mac[index];
2529}
2530
2531static uint32_t
2532e1000e_mac_ics_read(E1000ECore *core, int index)
2533{
2534    trace_e1000e_irq_read_ics(core->mac[ICS]);
2535    return core->mac[ICS];
2536}
2537
2538static uint32_t
2539e1000e_mac_ims_read(E1000ECore *core, int index)
2540{
2541    trace_e1000e_irq_read_ims(core->mac[IMS]);
2542    return core->mac[IMS];
2543}
2544
2545#define E1000E_LOW_BITS_READ_FUNC(num)                      \
2546    static uint32_t                                         \
2547    e1000e_mac_low##num##_read(E1000ECore *core, int index) \
2548    {                                                       \
2549        return core->mac[index] & (BIT(num) - 1);           \
2550    }                                                       \
2551
2552#define E1000E_LOW_BITS_READ(num)                           \
2553    e1000e_mac_low##num##_read
2554
2555E1000E_LOW_BITS_READ_FUNC(4);
2556E1000E_LOW_BITS_READ_FUNC(6);
2557E1000E_LOW_BITS_READ_FUNC(11);
2558E1000E_LOW_BITS_READ_FUNC(13);
2559E1000E_LOW_BITS_READ_FUNC(16);
2560
2561static uint32_t
2562e1000e_mac_swsm_read(E1000ECore *core, int index)
2563{
2564    uint32_t val = core->mac[SWSM];
2565    core->mac[SWSM] = val | 1;
2566    return val;
2567}
2568
2569static uint32_t
2570e1000e_mac_itr_read(E1000ECore *core, int index)
2571{
2572    return core->itr_guest_value;
2573}
2574
2575static uint32_t
2576e1000e_mac_eitr_read(E1000ECore *core, int index)
2577{
2578    return core->eitr_guest_value[index - EITR];
2579}
2580
2581static uint32_t
2582e1000e_mac_icr_read(E1000ECore *core, int index)
2583{
2584    uint32_t ret = core->mac[ICR];
2585    trace_e1000e_irq_icr_read_entry(ret);
2586
2587    if (core->mac[IMS] == 0) {
2588        trace_e1000e_irq_icr_clear_zero_ims();
2589        core->mac[ICR] = 0;
2590    }
2591
2592    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2593        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2594        trace_e1000e_irq_icr_clear_iame();
2595        core->mac[ICR] = 0;
2596        trace_e1000e_irq_icr_process_iame();
2597        e1000e_clear_ims_bits(core, core->mac[IAM]);
2598    }
2599
2600    trace_e1000e_irq_icr_read_exit(core->mac[ICR]);
2601    e1000e_update_interrupt_state(core);
2602    return ret;
2603}
2604
2605static uint32_t
2606e1000e_mac_read_clr4(E1000ECore *core, int index)
2607{
2608    uint32_t ret = core->mac[index];
2609
2610    core->mac[index] = 0;
2611    return ret;
2612}
2613
2614static uint32_t
2615e1000e_mac_read_clr8(E1000ECore *core, int index)
2616{
2617    uint32_t ret = core->mac[index];
2618
2619    core->mac[index] = 0;
2620    core->mac[index - 1] = 0;
2621    return ret;
2622}
2623
2624static uint32_t
2625e1000e_get_ctrl(E1000ECore *core, int index)
2626{
2627    uint32_t val = core->mac[CTRL];
2628
2629    trace_e1000e_link_read_params(
2630        !!(val & E1000_CTRL_ASDE),
2631        (val & E1000_CTRL_SPD_SEL) >> E1000_CTRL_SPD_SHIFT,
2632        !!(val & E1000_CTRL_FRCSPD),
2633        !!(val & E1000_CTRL_FRCDPX),
2634        !!(val & E1000_CTRL_RFCE),
2635        !!(val & E1000_CTRL_TFCE));
2636
2637    return val;
2638}
2639
2640static uint32_t
2641e1000e_get_status(E1000ECore *core, int index)
2642{
2643    uint32_t res = core->mac[STATUS];
2644
2645    if (!(core->mac[CTRL] & E1000_CTRL_GIO_MASTER_DISABLE)) {
2646        res |= E1000_STATUS_GIO_MASTER_ENABLE;
2647    }
2648
2649    if (core->mac[CTRL] & E1000_CTRL_FRCDPX) {
2650        res |= (core->mac[CTRL] & E1000_CTRL_FD) ? E1000_STATUS_FD : 0;
2651    } else {
2652        res |= E1000_STATUS_FD;
2653    }
2654
2655    if ((core->mac[CTRL] & E1000_CTRL_FRCSPD) ||
2656        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_SPD_BYPS)) {
2657        switch (core->mac[CTRL] & E1000_CTRL_SPD_SEL) {
2658        case E1000_CTRL_SPD_10:
2659            res |= E1000_STATUS_SPEED_10;
2660            break;
2661        case E1000_CTRL_SPD_100:
2662            res |= E1000_STATUS_SPEED_100;
2663            break;
2664        case E1000_CTRL_SPD_1000:
2665        default:
2666            res |= E1000_STATUS_SPEED_1000;
2667            break;
2668        }
2669    } else {
2670        res |= E1000_STATUS_SPEED_1000;
2671    }
2672
2673    trace_e1000e_link_status(
2674        !!(res & E1000_STATUS_LU),
2675        !!(res & E1000_STATUS_FD),
2676        (res & E1000_STATUS_SPEED_MASK) >> E1000_STATUS_SPEED_SHIFT,
2677        (res & E1000_STATUS_ASDV) >> E1000_STATUS_ASDV_SHIFT);
2678
2679    return res;
2680}
2681
2682static uint32_t
2683e1000e_get_tarc(E1000ECore *core, int index)
2684{
2685    return core->mac[index] & ((BIT(11) - 1) |
2686                                BIT(27)      |
2687                                BIT(28)      |
2688                                BIT(29)      |
2689                                BIT(30));
2690}
2691
2692static void
2693e1000e_mac_writereg(E1000ECore *core, int index, uint32_t val)
2694{
2695    core->mac[index] = val;
2696}
2697
2698static void
2699e1000e_mac_setmacaddr(E1000ECore *core, int index, uint32_t val)
2700{
2701    uint32_t macaddr[2];
2702
2703    core->mac[index] = val;
2704
2705    macaddr[0] = cpu_to_le32(core->mac[RA]);
2706    macaddr[1] = cpu_to_le32(core->mac[RA + 1]);
2707    qemu_format_nic_info_str(qemu_get_queue(core->owner_nic),
2708        (uint8_t *) macaddr);
2709
2710    trace_e1000e_mac_set_sw(MAC_ARG(macaddr));
2711}
2712
2713static void
2714e1000e_set_eecd(E1000ECore *core, int index, uint32_t val)
2715{
2716    static const uint32_t ro_bits = E1000_EECD_PRES          |
2717                                    E1000_EECD_AUTO_RD       |
2718                                    E1000_EECD_SIZE_EX_MASK;
2719
2720    core->mac[EECD] = (core->mac[EECD] & ro_bits) | (val & ~ro_bits);
2721}
2722
2723static void
2724e1000e_set_eerd(E1000ECore *core, int index, uint32_t val)
2725{
2726    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2727    uint32_t flags = 0;
2728    uint32_t data = 0;
2729
2730    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2731        data = core->eeprom[addr];
2732        flags = E1000_EERW_DONE;
2733    }
2734
2735    core->mac[EERD] = flags                           |
2736                      (addr << E1000_EERW_ADDR_SHIFT) |
2737                      (data << E1000_EERW_DATA_SHIFT);
2738}
2739
2740static void
2741e1000e_set_eewr(E1000ECore *core, int index, uint32_t val)
2742{
2743    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2744    uint32_t data = (val >> E1000_EERW_DATA_SHIFT) & E1000_EERW_DATA_MASK;
2745    uint32_t flags = 0;
2746
2747    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2748        core->eeprom[addr] = data;
2749        flags = E1000_EERW_DONE;
2750    }
2751
2752    core->mac[EERD] = flags                           |
2753                      (addr << E1000_EERW_ADDR_SHIFT) |
2754                      (data << E1000_EERW_DATA_SHIFT);
2755}
2756
2757static void
2758e1000e_set_rxdctl(E1000ECore *core, int index, uint32_t val)
2759{
2760    core->mac[RXDCTL] = core->mac[RXDCTL1] = val;
2761}
2762
2763static void
2764e1000e_set_itr(E1000ECore *core, int index, uint32_t val)
2765{
2766    uint32_t interval = val & 0xffff;
2767
2768    trace_e1000e_irq_itr_set(val);
2769
2770    core->itr_guest_value = interval;
2771    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2772}
2773
2774static void
2775e1000e_set_eitr(E1000ECore *core, int index, uint32_t val)
2776{
2777    uint32_t interval = val & 0xffff;
2778    uint32_t eitr_num = index - EITR;
2779
2780    trace_e1000e_irq_eitr_set(eitr_num, val);
2781
2782    core->eitr_guest_value[eitr_num] = interval;
2783    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2784}
2785
2786static void
2787e1000e_set_psrctl(E1000ECore *core, int index, uint32_t val)
2788{
2789    if ((val & E1000_PSRCTL_BSIZE0_MASK) == 0) {
2790        hw_error("e1000e: PSRCTL.BSIZE0 cannot be zero");
2791    }
2792
2793    if ((val & E1000_PSRCTL_BSIZE1_MASK) == 0) {
2794        hw_error("e1000e: PSRCTL.BSIZE1 cannot be zero");
2795    }
2796
2797    core->mac[PSRCTL] = val;
2798}
2799
2800static void
2801e1000e_update_rx_offloads(E1000ECore *core)
2802{
2803    int cso_state = e1000e_rx_l4_cso_enabled(core);
2804
2805    trace_e1000e_rx_set_cso(cso_state);
2806
2807    if (core->has_vnet) {
2808        qemu_set_offload(qemu_get_queue(core->owner_nic)->peer,
2809                         cso_state, 0, 0, 0, 0);
2810    }
2811}
2812
2813static void
2814e1000e_set_rxcsum(E1000ECore *core, int index, uint32_t val)
2815{
2816    core->mac[RXCSUM] = val;
2817    e1000e_update_rx_offloads(core);
2818}
2819
2820static void
2821e1000e_set_gcr(E1000ECore *core, int index, uint32_t val)
2822{
2823    uint32_t ro_bits = core->mac[GCR] & E1000_GCR_RO_BITS;
2824    core->mac[GCR] = (val & ~E1000_GCR_RO_BITS) | ro_bits;
2825}
2826
2827#define e1000e_getreg(x)    [x] = e1000e_mac_readreg
2828static uint32_t (*e1000e_macreg_readops[])(E1000ECore *, int) = {
2829    e1000e_getreg(PBA),
2830    e1000e_getreg(WUFC),
2831    e1000e_getreg(MANC),
2832    e1000e_getreg(TOTL),
2833    e1000e_getreg(RDT0),
2834    e1000e_getreg(RDBAH0),
2835    e1000e_getreg(TDBAL1),
2836    e1000e_getreg(RDLEN0),
2837    e1000e_getreg(RDH1),
2838    e1000e_getreg(LATECOL),
2839    e1000e_getreg(SEC),
2840    e1000e_getreg(XONTXC),
2841    e1000e_getreg(WUS),
2842    e1000e_getreg(GORCL),
2843    e1000e_getreg(MGTPRC),
2844    e1000e_getreg(EERD),
2845    e1000e_getreg(EIAC),
2846    e1000e_getreg(PSRCTL),
2847    e1000e_getreg(MANC2H),
2848    e1000e_getreg(RXCSUM),
2849    e1000e_getreg(GSCL_3),
2850    e1000e_getreg(GSCN_2),
2851    e1000e_getreg(RSRPD),
2852    e1000e_getreg(RDBAL1),
2853    e1000e_getreg(FCAH),
2854    e1000e_getreg(FCRTH),
2855    e1000e_getreg(FLOP),
2856    e1000e_getreg(FLASHT),
2857    e1000e_getreg(RXSTMPH),
2858    e1000e_getreg(TXSTMPL),
2859    e1000e_getreg(TIMADJL),
2860    e1000e_getreg(TXDCTL),
2861    e1000e_getreg(RDH0),
2862    e1000e_getreg(TDT1),
2863    e1000e_getreg(TNCRS),
2864    e1000e_getreg(RJC),
2865    e1000e_getreg(IAM),
2866    e1000e_getreg(GSCL_2),
2867    e1000e_getreg(RDBAH1),
2868    e1000e_getreg(FLSWDATA),
2869    e1000e_getreg(RXSATRH),
2870    e1000e_getreg(TIPG),
2871    e1000e_getreg(FLMNGCTL),
2872    e1000e_getreg(FLMNGCNT),
2873    e1000e_getreg(TSYNCTXCTL),
2874    e1000e_getreg(EXTCNF_SIZE),
2875    e1000e_getreg(EXTCNF_CTRL),
2876    e1000e_getreg(EEMNGDATA),
2877    e1000e_getreg(CTRL_EXT),
2878    e1000e_getreg(SYSTIMH),
2879    e1000e_getreg(EEMNGCTL),
2880    e1000e_getreg(FLMNGDATA),
2881    e1000e_getreg(TSYNCRXCTL),
2882    e1000e_getreg(TDH),
2883    e1000e_getreg(LEDCTL),
2884    e1000e_getreg(STATUS),
2885    e1000e_getreg(TCTL),
2886    e1000e_getreg(TDBAL),
2887    e1000e_getreg(TDLEN),
2888    e1000e_getreg(TDH1),
2889    e1000e_getreg(RADV),
2890    e1000e_getreg(ECOL),
2891    e1000e_getreg(DC),
2892    e1000e_getreg(RLEC),
2893    e1000e_getreg(XOFFTXC),
2894    e1000e_getreg(RFC),
2895    e1000e_getreg(RNBC),
2896    e1000e_getreg(MGTPTC),
2897    e1000e_getreg(TIMINCA),
2898    e1000e_getreg(RXCFGL),
2899    e1000e_getreg(MFUTP01),
2900    e1000e_getreg(FACTPS),
2901    e1000e_getreg(GSCL_1),
2902    e1000e_getreg(GSCN_0),
2903    e1000e_getreg(GCR2),
2904    e1000e_getreg(RDT1),
2905    e1000e_getreg(PBACLR),
2906    e1000e_getreg(FCTTV),
2907    e1000e_getreg(EEWR),
2908    e1000e_getreg(FLSWCTL),
2909    e1000e_getreg(RXDCTL1),
2910    e1000e_getreg(RXSATRL),
2911    e1000e_getreg(SYSTIML),
2912    e1000e_getreg(RXUDP),
2913    e1000e_getreg(TORL),
2914    e1000e_getreg(TDLEN1),
2915    e1000e_getreg(MCC),
2916    e1000e_getreg(WUC),
2917    e1000e_getreg(EECD),
2918    e1000e_getreg(MFUTP23),
2919    e1000e_getreg(RAID),
2920    e1000e_getreg(FCRTV),
2921    e1000e_getreg(TXDCTL1),
2922    e1000e_getreg(RCTL),
2923    e1000e_getreg(TDT),
2924    e1000e_getreg(MDIC),
2925    e1000e_getreg(FCRUC),
2926    e1000e_getreg(VET),
2927    e1000e_getreg(RDBAL0),
2928    e1000e_getreg(TDBAH1),
2929    e1000e_getreg(RDTR),
2930    e1000e_getreg(SCC),
2931    e1000e_getreg(COLC),
2932    e1000e_getreg(CEXTERR),
2933    e1000e_getreg(XOFFRXC),
2934    e1000e_getreg(IPAV),
2935    e1000e_getreg(GOTCL),
2936    e1000e_getreg(MGTPDC),
2937    e1000e_getreg(GCR),
2938    e1000e_getreg(IVAR),
2939    e1000e_getreg(POEMB),
2940    e1000e_getreg(MFVAL),
2941    e1000e_getreg(FUNCTAG),
2942    e1000e_getreg(GSCL_4),
2943    e1000e_getreg(GSCN_3),
2944    e1000e_getreg(MRQC),
2945    e1000e_getreg(RDLEN1),
2946    e1000e_getreg(FCT),
2947    e1000e_getreg(FLA),
2948    e1000e_getreg(FLOL),
2949    e1000e_getreg(RXDCTL),
2950    e1000e_getreg(RXSTMPL),
2951    e1000e_getreg(TXSTMPH),
2952    e1000e_getreg(TIMADJH),
2953    e1000e_getreg(FCRTL),
2954    e1000e_getreg(TDBAH),
2955    e1000e_getreg(TADV),
2956    e1000e_getreg(XONRXC),
2957    e1000e_getreg(TSCTFC),
2958    e1000e_getreg(RFCTL),
2959    e1000e_getreg(GSCN_1),
2960    e1000e_getreg(FCAL),
2961    e1000e_getreg(FLSWCNT),
2962
2963    [TOTH]    = e1000e_mac_read_clr8,
2964    [GOTCH]   = e1000e_mac_read_clr8,
2965    [PRC64]   = e1000e_mac_read_clr4,
2966    [PRC255]  = e1000e_mac_read_clr4,
2967    [PRC1023] = e1000e_mac_read_clr4,
2968    [PTC64]   = e1000e_mac_read_clr4,
2969    [PTC255]  = e1000e_mac_read_clr4,
2970    [PTC1023] = e1000e_mac_read_clr4,
2971    [GPRC]    = e1000e_mac_read_clr4,
2972    [TPT]     = e1000e_mac_read_clr4,
2973    [RUC]     = e1000e_mac_read_clr4,
2974    [BPRC]    = e1000e_mac_read_clr4,
2975    [MPTC]    = e1000e_mac_read_clr4,
2976    [IAC]     = e1000e_mac_read_clr4,
2977    [ICR]     = e1000e_mac_icr_read,
2978    [RDFH]    = E1000E_LOW_BITS_READ(13),
2979    [RDFHS]   = E1000E_LOW_BITS_READ(13),
2980    [RDFPC]   = E1000E_LOW_BITS_READ(13),
2981    [TDFH]    = E1000E_LOW_BITS_READ(13),
2982    [TDFHS]   = E1000E_LOW_BITS_READ(13),
2983    [STATUS]  = e1000e_get_status,
2984    [TARC0]   = e1000e_get_tarc,
2985    [PBS]     = E1000E_LOW_BITS_READ(6),
2986    [ICS]     = e1000e_mac_ics_read,
2987    [AIT]     = E1000E_LOW_BITS_READ(16),
2988    [TORH]    = e1000e_mac_read_clr8,
2989    [GORCH]   = e1000e_mac_read_clr8,
2990    [PRC127]  = e1000e_mac_read_clr4,
2991    [PRC511]  = e1000e_mac_read_clr4,
2992    [PRC1522] = e1000e_mac_read_clr4,
2993    [PTC127]  = e1000e_mac_read_clr4,
2994    [PTC511]  = e1000e_mac_read_clr4,
2995    [PTC1522] = e1000e_mac_read_clr4,
2996    [GPTC]    = e1000e_mac_read_clr4,
2997    [TPR]     = e1000e_mac_read_clr4,
2998    [ROC]     = e1000e_mac_read_clr4,
2999    [MPRC]    = e1000e_mac_read_clr4,
3000    [BPTC]    = e1000e_mac_read_clr4,
3001    [TSCTC]   = e1000e_mac_read_clr4,
3002    [ITR]     = e1000e_mac_itr_read,
3003    [RDFT]    = E1000E_LOW_BITS_READ(13),
3004    [RDFTS]   = E1000E_LOW_BITS_READ(13),
3005    [TDFPC]   = E1000E_LOW_BITS_READ(13),
3006    [TDFT]    = E1000E_LOW_BITS_READ(13),
3007    [TDFTS]   = E1000E_LOW_BITS_READ(13),
3008    [CTRL]    = e1000e_get_ctrl,
3009    [TARC1]   = e1000e_get_tarc,
3010    [SWSM]    = e1000e_mac_swsm_read,
3011    [IMS]     = e1000e_mac_ims_read,
3012
3013    [CRCERRS ... MPC]      = e1000e_mac_readreg,
3014    [IP6AT ... IP6AT + 3]  = e1000e_mac_readreg,
3015    [IP4AT ... IP4AT + 6]  = e1000e_mac_readreg,
3016    [RA ... RA + 31]       = e1000e_mac_readreg,
3017    [WUPM ... WUPM + 31]   = e1000e_mac_readreg,
3018    [MTA ... MTA + 127]    = e1000e_mac_readreg,
3019    [VFTA ... VFTA + 127]  = e1000e_mac_readreg,
3020    [FFMT ... FFMT + 254]  = E1000E_LOW_BITS_READ(4),
3021    [FFVT ... FFVT + 254]  = e1000e_mac_readreg,
3022    [MDEF ... MDEF + 7]    = e1000e_mac_readreg,
3023    [FFLT ... FFLT + 10]   = E1000E_LOW_BITS_READ(11),
3024    [FTFT ... FTFT + 254]  = e1000e_mac_readreg,
3025    [PBM ... PBM + 10239]  = e1000e_mac_readreg,
3026    [RETA ... RETA + 31]   = e1000e_mac_readreg,
3027    [RSSRK ... RSSRK + 31] = e1000e_mac_readreg,
3028    [MAVTV0 ... MAVTV3]    = e1000e_mac_readreg,
3029    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_mac_eitr_read
3030};
3031enum { E1000E_NREADOPS = ARRAY_SIZE(e1000e_macreg_readops) };
3032
3033#define e1000e_putreg(x)    [x] = e1000e_mac_writereg
3034static void (*e1000e_macreg_writeops[])(E1000ECore *, int, uint32_t) = {
3035    e1000e_putreg(PBA),
3036    e1000e_putreg(SWSM),
3037    e1000e_putreg(WUFC),
3038    e1000e_putreg(RDBAH1),
3039    e1000e_putreg(TDBAH),
3040    e1000e_putreg(TXDCTL),
3041    e1000e_putreg(RDBAH0),
3042    e1000e_putreg(LEDCTL),
3043    e1000e_putreg(FCAL),
3044    e1000e_putreg(FCRUC),
3045    e1000e_putreg(AIT),
3046    e1000e_putreg(TDFH),
3047    e1000e_putreg(TDFT),
3048    e1000e_putreg(TDFHS),
3049    e1000e_putreg(TDFTS),
3050    e1000e_putreg(TDFPC),
3051    e1000e_putreg(WUC),
3052    e1000e_putreg(WUS),
3053    e1000e_putreg(RDFH),
3054    e1000e_putreg(RDFT),
3055    e1000e_putreg(RDFHS),
3056    e1000e_putreg(RDFTS),
3057    e1000e_putreg(RDFPC),
3058    e1000e_putreg(IPAV),
3059    e1000e_putreg(TDBAH1),
3060    e1000e_putreg(TIMINCA),
3061    e1000e_putreg(IAM),
3062    e1000e_putreg(EIAC),
3063    e1000e_putreg(IVAR),
3064    e1000e_putreg(TARC0),
3065    e1000e_putreg(TARC1),
3066    e1000e_putreg(FLSWDATA),
3067    e1000e_putreg(POEMB),
3068    e1000e_putreg(PBS),
3069    e1000e_putreg(MFUTP01),
3070    e1000e_putreg(MFUTP23),
3071    e1000e_putreg(MANC),
3072    e1000e_putreg(MANC2H),
3073    e1000e_putreg(MFVAL),
3074    e1000e_putreg(EXTCNF_CTRL),
3075    e1000e_putreg(FACTPS),
3076    e1000e_putreg(FUNCTAG),
3077    e1000e_putreg(GSCL_1),
3078    e1000e_putreg(GSCL_2),
3079    e1000e_putreg(GSCL_3),
3080    e1000e_putreg(GSCL_4),
3081    e1000e_putreg(GSCN_0),
3082    e1000e_putreg(GSCN_1),
3083    e1000e_putreg(GSCN_2),
3084    e1000e_putreg(GSCN_3),
3085    e1000e_putreg(GCR2),
3086    e1000e_putreg(MRQC),
3087    e1000e_putreg(FLOP),
3088    e1000e_putreg(FLOL),
3089    e1000e_putreg(FLSWCTL),
3090    e1000e_putreg(FLSWCNT),
3091    e1000e_putreg(FLA),
3092    e1000e_putreg(RXDCTL1),
3093    e1000e_putreg(TXDCTL1),
3094    e1000e_putreg(TIPG),
3095    e1000e_putreg(RXSTMPH),
3096    e1000e_putreg(RXSTMPL),
3097    e1000e_putreg(RXSATRL),
3098    e1000e_putreg(RXSATRH),
3099    e1000e_putreg(TXSTMPL),
3100    e1000e_putreg(TXSTMPH),
3101    e1000e_putreg(SYSTIML),
3102    e1000e_putreg(SYSTIMH),
3103    e1000e_putreg(TIMADJL),
3104    e1000e_putreg(TIMADJH),
3105    e1000e_putreg(RXUDP),
3106    e1000e_putreg(RXCFGL),
3107    e1000e_putreg(TSYNCRXCTL),
3108    e1000e_putreg(TSYNCTXCTL),
3109    e1000e_putreg(FLSWDATA),
3110    e1000e_putreg(EXTCNF_SIZE),
3111    e1000e_putreg(EEMNGCTL),
3112    e1000e_putreg(RA),
3113
3114    [TDH1]     = e1000e_set_16bit,
3115    [TDT1]     = e1000e_set_tdt,
3116    [TCTL]     = e1000e_set_tctl,
3117    [TDT]      = e1000e_set_tdt,
3118    [MDIC]     = e1000e_set_mdic,
3119    [ICS]      = e1000e_set_ics,
3120    [TDH]      = e1000e_set_16bit,
3121    [RDH0]     = e1000e_set_16bit,
3122    [RDT0]     = e1000e_set_rdt,
3123    [IMC]      = e1000e_set_imc,
3124    [IMS]      = e1000e_set_ims,
3125    [ICR]      = e1000e_set_icr,
3126    [EECD]     = e1000e_set_eecd,
3127    [RCTL]     = e1000e_set_rx_control,
3128    [CTRL]     = e1000e_set_ctrl,
3129    [RDTR]     = e1000e_set_rdtr,
3130    [RADV]     = e1000e_set_16bit,
3131    [TADV]     = e1000e_set_16bit,
3132    [ITR]      = e1000e_set_itr,
3133    [EERD]     = e1000e_set_eerd,
3134    [GCR]      = e1000e_set_gcr,
3135    [PSRCTL]   = e1000e_set_psrctl,
3136    [RXCSUM]   = e1000e_set_rxcsum,
3137    [RAID]     = e1000e_set_16bit,
3138    [RSRPD]    = e1000e_set_12bit,
3139    [TIDV]     = e1000e_set_tidv,
3140    [TDLEN1]   = e1000e_set_dlen,
3141    [TDLEN]    = e1000e_set_dlen,
3142    [RDLEN0]   = e1000e_set_dlen,
3143    [RDLEN1]   = e1000e_set_dlen,
3144    [TDBAL]    = e1000e_set_dbal,
3145    [TDBAL1]   = e1000e_set_dbal,
3146    [RDBAL0]   = e1000e_set_dbal,
3147    [RDBAL1]   = e1000e_set_dbal,
3148    [RDH1]     = e1000e_set_16bit,
3149    [RDT1]     = e1000e_set_rdt,
3150    [STATUS]   = e1000e_set_status,
3151    [PBACLR]   = e1000e_set_pbaclr,
3152    [CTRL_EXT] = e1000e_set_ctrlext,
3153    [FCAH]     = e1000e_set_16bit,
3154    [FCT]      = e1000e_set_16bit,
3155    [FCTTV]    = e1000e_set_16bit,
3156    [FCRTV]    = e1000e_set_16bit,
3157    [FCRTH]    = e1000e_set_fcrth,
3158    [FCRTL]    = e1000e_set_fcrtl,
3159    [VET]      = e1000e_set_vet,
3160    [RXDCTL]   = e1000e_set_rxdctl,
3161    [FLASHT]   = e1000e_set_16bit,
3162    [EEWR]     = e1000e_set_eewr,
3163    [CTRL_DUP] = e1000e_set_ctrl,
3164    [RFCTL]    = e1000e_set_rfctl,
3165    [RA + 1]   = e1000e_mac_setmacaddr,
3166
3167    [IP6AT ... IP6AT + 3]    = e1000e_mac_writereg,
3168    [IP4AT ... IP4AT + 6]    = e1000e_mac_writereg,
3169    [RA + 2 ... RA + 31]     = e1000e_mac_writereg,
3170    [WUPM ... WUPM + 31]     = e1000e_mac_writereg,
3171    [MTA ... MTA + 127]      = e1000e_mac_writereg,
3172    [VFTA ... VFTA + 127]    = e1000e_mac_writereg,
3173    [FFMT ... FFMT + 254]    = e1000e_mac_writereg,
3174    [FFVT ... FFVT + 254]    = e1000e_mac_writereg,
3175    [PBM ... PBM + 10239]    = e1000e_mac_writereg,
3176    [MDEF ... MDEF + 7]      = e1000e_mac_writereg,
3177    [FFLT ... FFLT + 10]     = e1000e_mac_writereg,
3178    [FTFT ... FTFT + 254]    = e1000e_mac_writereg,
3179    [RETA ... RETA + 31]     = e1000e_mac_writereg,
3180    [RSSRK ... RSSRK + 31]   = e1000e_mac_writereg,
3181    [MAVTV0 ... MAVTV3]      = e1000e_mac_writereg,
3182    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_set_eitr
3183};
3184enum { E1000E_NWRITEOPS = ARRAY_SIZE(e1000e_macreg_writeops) };
3185
3186enum { MAC_ACCESS_PARTIAL = 1 };
3187
3188/* The array below combines alias offsets of the index values for the
3189 * MAC registers that have aliases, with the indication of not fully
3190 * implemented registers (lowest bit). This combination is possible
3191 * because all of the offsets are even. */
3192static const uint16_t mac_reg_access[E1000E_MAC_SIZE] = {
3193    /* Alias index offsets */
3194    [FCRTL_A] = 0x07fe, [FCRTH_A] = 0x0802,
3195    [RDH0_A]  = 0x09bc, [RDT0_A]  = 0x09bc, [RDTR_A] = 0x09c6,
3196    [RDFH_A]  = 0xe904, [RDFT_A]  = 0xe904,
3197    [TDH_A]   = 0x0cf8, [TDT_A]   = 0x0cf8, [TIDV_A] = 0x0cf8,
3198    [TDFH_A]  = 0xed00, [TDFT_A]  = 0xed00,
3199    [RA_A ... RA_A + 31]      = 0x14f0,
3200    [VFTA_A ... VFTA_A + 127] = 0x1400,
3201    [RDBAL0_A ... RDLEN0_A] = 0x09bc,
3202    [TDBAL_A ... TDLEN_A]   = 0x0cf8,
3203    /* Access options */
3204    [RDFH]  = MAC_ACCESS_PARTIAL,    [RDFT]  = MAC_ACCESS_PARTIAL,
3205    [RDFHS] = MAC_ACCESS_PARTIAL,    [RDFTS] = MAC_ACCESS_PARTIAL,
3206    [RDFPC] = MAC_ACCESS_PARTIAL,
3207    [TDFH]  = MAC_ACCESS_PARTIAL,    [TDFT]  = MAC_ACCESS_PARTIAL,
3208    [TDFHS] = MAC_ACCESS_PARTIAL,    [TDFTS] = MAC_ACCESS_PARTIAL,
3209    [TDFPC] = MAC_ACCESS_PARTIAL,    [EECD]  = MAC_ACCESS_PARTIAL,
3210    [PBM]   = MAC_ACCESS_PARTIAL,    [FLA]   = MAC_ACCESS_PARTIAL,
3211    [FCAL]  = MAC_ACCESS_PARTIAL,    [FCAH]  = MAC_ACCESS_PARTIAL,
3212    [FCT]   = MAC_ACCESS_PARTIAL,    [FCTTV] = MAC_ACCESS_PARTIAL,
3213    [FCRTV] = MAC_ACCESS_PARTIAL,    [FCRTL] = MAC_ACCESS_PARTIAL,
3214    [FCRTH] = MAC_ACCESS_PARTIAL,    [TXDCTL] = MAC_ACCESS_PARTIAL,
3215    [TXDCTL1] = MAC_ACCESS_PARTIAL,
3216    [MAVTV0 ... MAVTV3] = MAC_ACCESS_PARTIAL
3217};
3218
3219void
3220e1000e_core_write(E1000ECore *core, hwaddr addr, uint64_t val, unsigned size)
3221{
3222    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3223
3224    if (index < E1000E_NWRITEOPS && e1000e_macreg_writeops[index]) {
3225        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3226            trace_e1000e_wrn_regs_write_trivial(index << 2);
3227        }
3228        trace_e1000e_core_write(index << 2, size, val);
3229        e1000e_macreg_writeops[index](core, index, val);
3230    } else if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3231        trace_e1000e_wrn_regs_write_ro(index << 2, size, val);
3232    } else {
3233        trace_e1000e_wrn_regs_write_unknown(index << 2, size, val);
3234    }
3235}
3236
3237uint64_t
3238e1000e_core_read(E1000ECore *core, hwaddr addr, unsigned size)
3239{
3240    uint64_t val;
3241    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3242
3243    if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3244        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3245            trace_e1000e_wrn_regs_read_trivial(index << 2);
3246        }
3247        val = e1000e_macreg_readops[index](core, index);
3248        trace_e1000e_core_read(index << 2, size, val);
3249        return val;
3250    } else {
3251        trace_e1000e_wrn_regs_read_unknown(index << 2, size);
3252    }
3253    return 0;
3254}
3255
3256static inline void
3257e1000e_autoneg_pause(E1000ECore *core)
3258{
3259    timer_del(core->autoneg_timer);
3260}
3261
3262static void
3263e1000e_autoneg_resume(E1000ECore *core)
3264{
3265    if (e1000e_have_autoneg(core) &&
3266        !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
3267        qemu_get_queue(core->owner_nic)->link_down = false;
3268        timer_mod(core->autoneg_timer,
3269                  qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 500);
3270    }
3271}
3272
3273static void
3274e1000e_vm_state_change(void *opaque, int running, RunState state)
3275{
3276    E1000ECore *core = opaque;
3277
3278    if (running) {
3279        trace_e1000e_vm_state_running();
3280        e1000e_intrmgr_resume(core);
3281        e1000e_autoneg_resume(core);
3282    } else {
3283        trace_e1000e_vm_state_stopped();
3284        e1000e_autoneg_pause(core);
3285        e1000e_intrmgr_pause(core);
3286    }
3287}
3288
3289void
3290e1000e_core_pci_realize(E1000ECore     *core,
3291                        const uint16_t *eeprom_templ,
3292                        uint32_t        eeprom_size,
3293                        const uint8_t  *macaddr)
3294{
3295    int i;
3296
3297    core->autoneg_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
3298                                       e1000e_autoneg_timer, core);
3299    e1000e_intrmgr_pci_realize(core);
3300
3301    core->vmstate =
3302        qemu_add_vm_change_state_handler(e1000e_vm_state_change, core);
3303
3304    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3305        net_tx_pkt_init(&core->tx[i].tx_pkt, core->owner,
3306                        E1000E_MAX_TX_FRAGS, core->has_vnet);
3307    }
3308
3309    net_rx_pkt_init(&core->rx_pkt, core->has_vnet);
3310
3311    e1000x_core_prepare_eeprom(core->eeprom,
3312                               eeprom_templ,
3313                               eeprom_size,
3314                               PCI_DEVICE_GET_CLASS(core->owner)->device_id,
3315                               macaddr);
3316    e1000e_update_rx_offloads(core);
3317}
3318
3319void
3320e1000e_core_pci_uninit(E1000ECore *core)
3321{
3322    int i;
3323
3324    timer_del(core->autoneg_timer);
3325    timer_free(core->autoneg_timer);
3326
3327    e1000e_intrmgr_pci_unint(core);
3328
3329    qemu_del_vm_change_state_handler(core->vmstate);
3330
3331    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3332        net_tx_pkt_reset(core->tx[i].tx_pkt);
3333        net_tx_pkt_uninit(core->tx[i].tx_pkt);
3334    }
3335
3336    net_rx_pkt_uninit(core->rx_pkt);
3337}
3338
3339static const uint16_t
3340e1000e_phy_reg_init[E1000E_PHY_PAGES][E1000E_PHY_PAGE_SIZE] = {
3341    [0] = {
3342        [PHY_CTRL] =   MII_CR_SPEED_SELECT_MSB  |
3343                       MII_CR_FULL_DUPLEX       |
3344                       MII_CR_AUTO_NEG_EN,
3345
3346        [PHY_STATUS] = MII_SR_EXTENDED_CAPS     |
3347                       MII_SR_LINK_STATUS       |
3348                       MII_SR_AUTONEG_CAPS      |
3349                       MII_SR_PREAMBLE_SUPPRESS |
3350                       MII_SR_EXTENDED_STATUS   |
3351                       MII_SR_10T_HD_CAPS       |
3352                       MII_SR_10T_FD_CAPS       |
3353                       MII_SR_100X_HD_CAPS      |
3354                       MII_SR_100X_FD_CAPS,
3355
3356        [PHY_ID1]               = 0x141,
3357        [PHY_ID2]               = E1000_PHY_ID2_82574x,
3358        [PHY_AUTONEG_ADV]       = 0xde1,
3359        [PHY_LP_ABILITY]        = 0x7e0,
3360        [PHY_AUTONEG_EXP]       = BIT(2),
3361        [PHY_NEXT_PAGE_TX]      = BIT(0) | BIT(13),
3362        [PHY_1000T_CTRL]        = BIT(8) | BIT(9) | BIT(10) | BIT(11),
3363        [PHY_1000T_STATUS]      = 0x3c00,
3364        [PHY_EXT_STATUS]        = BIT(12) | BIT(13),
3365
3366        [PHY_COPPER_CTRL1]      = BIT(5) | BIT(6) | BIT(8) | BIT(9) |
3367                                  BIT(12) | BIT(13),
3368        [PHY_COPPER_STAT1]      = BIT(3) | BIT(10) | BIT(11) | BIT(13) | BIT(15)
3369    },
3370    [2] = {
3371        [PHY_MAC_CTRL1]         = BIT(3) | BIT(7),
3372        [PHY_MAC_CTRL2]         = BIT(1) | BIT(2) | BIT(6) | BIT(12)
3373    },
3374    [3] = {
3375        [PHY_LED_TIMER_CTRL]    = BIT(0) | BIT(2) | BIT(14)
3376    }
3377};
3378
3379static const uint32_t e1000e_mac_reg_init[] = {
3380    [PBA]           =     0x00140014,
3381    [LEDCTL]        =  BIT(1) | BIT(8) | BIT(9) | BIT(15) | BIT(17) | BIT(18),
3382    [EXTCNF_CTRL]   = BIT(3),
3383    [EEMNGCTL]      = BIT(31),
3384    [FLASHT]        = 0x2,
3385    [FLSWCTL]       = BIT(30) | BIT(31),
3386    [FLOL]          = BIT(0),
3387    [RXDCTL]        = BIT(16),
3388    [RXDCTL1]       = BIT(16),
3389    [TIPG]          = 0x8 | (0x8 << 10) | (0x6 << 20),
3390    [RXCFGL]        = 0x88F7,
3391    [RXUDP]         = 0x319,
3392    [CTRL]          = E1000_CTRL_FD | E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN0 |
3393                      E1000_CTRL_SPD_1000 | E1000_CTRL_SLU |
3394                      E1000_CTRL_ADVD3WUC,
3395    [STATUS]        =  E1000_STATUS_ASDV_1000 | E1000_STATUS_LU,
3396    [PSRCTL]        = (2 << E1000_PSRCTL_BSIZE0_SHIFT) |
3397                      (4 << E1000_PSRCTL_BSIZE1_SHIFT) |
3398                      (4 << E1000_PSRCTL_BSIZE2_SHIFT),
3399    [TARC0]         = 0x3 | E1000_TARC_ENABLE,
3400    [TARC1]         = 0x3 | E1000_TARC_ENABLE,
3401    [EECD]          = E1000_EECD_AUTO_RD | E1000_EECD_PRES,
3402    [EERD]          = E1000_EERW_DONE,
3403    [EEWR]          = E1000_EERW_DONE,
3404    [GCR]           = E1000_L0S_ADJUST |
3405                      E1000_L1_ENTRY_LATENCY_MSB |
3406                      E1000_L1_ENTRY_LATENCY_LSB,
3407    [TDFH]          = 0x600,
3408    [TDFT]          = 0x600,
3409    [TDFHS]         = 0x600,
3410    [TDFTS]         = 0x600,
3411    [POEMB]         = 0x30D,
3412    [PBS]           = 0x028,
3413    [MANC]          = E1000_MANC_DIS_IP_CHK_ARP,
3414    [FACTPS]        = E1000_FACTPS_LAN0_ON | 0x20000000,
3415    [SWSM]          = 1,
3416    [RXCSUM]        = E1000_RXCSUM_IPOFLD | E1000_RXCSUM_TUOFLD,
3417    [ITR]           = E1000E_MIN_XITR,
3418    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = E1000E_MIN_XITR,
3419};
3420
3421void
3422e1000e_core_reset(E1000ECore *core)
3423{
3424    int i;
3425
3426    timer_del(core->autoneg_timer);
3427
3428    e1000e_intrmgr_reset(core);
3429
3430    memset(core->phy, 0, sizeof core->phy);
3431    memmove(core->phy, e1000e_phy_reg_init, sizeof e1000e_phy_reg_init);
3432    memset(core->mac, 0, sizeof core->mac);
3433    memmove(core->mac, e1000e_mac_reg_init, sizeof e1000e_mac_reg_init);
3434
3435    core->rxbuf_min_shift = 1 + E1000_RING_DESC_LEN_SHIFT;
3436
3437    if (qemu_get_queue(core->owner_nic)->link_down) {
3438        e1000e_link_down(core);
3439    }
3440
3441    e1000x_reset_mac_addr(core->owner_nic, core->mac, core->permanent_mac);
3442
3443    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3444        net_tx_pkt_reset(core->tx[i].tx_pkt);
3445        memset(&core->tx[i].props, 0, sizeof(core->tx[i].props));
3446        core->tx[i].skip_cp = false;
3447    }
3448}
3449
3450void e1000e_core_pre_save(E1000ECore *core)
3451{
3452    int i;
3453    NetClientState *nc = qemu_get_queue(core->owner_nic);
3454
3455    /*
3456    * If link is down and auto-negotiation is supported and ongoing,
3457    * complete auto-negotiation immediately. This allows us to look
3458    * at MII_SR_AUTONEG_COMPLETE to infer link status on load.
3459    */
3460    if (nc->link_down && e1000e_have_autoneg(core)) {
3461        core->phy[0][PHY_STATUS] |= MII_SR_AUTONEG_COMPLETE;
3462        e1000e_update_flowctl_status(core);
3463    }
3464
3465    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3466        if (net_tx_pkt_has_fragments(core->tx[i].tx_pkt)) {
3467            core->tx[i].skip_cp = true;
3468        }
3469    }
3470}
3471
3472int
3473e1000e_core_post_load(E1000ECore *core)
3474{
3475    NetClientState *nc = qemu_get_queue(core->owner_nic);
3476
3477    /* nc.link_down can't be migrated, so infer link_down according
3478     * to link status bit in core.mac[STATUS].
3479     */
3480    nc->link_down = (core->mac[STATUS] & E1000_STATUS_LU) == 0;
3481
3482    return 0;
3483}
3484