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 2.1 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 "qemu/log.h"
  38#include "net/net.h"
  39#include "net/tap.h"
  40#include "hw/pci/msi.h"
  41#include "hw/pci/msix.h"
  42#include "sysemu/runstate.h"
  43
  44#include "net_tx_pkt.h"
  45#include "net_rx_pkt.h"
  46
  47#include "e1000x_common.h"
  48#include "e1000e_core.h"
  49
  50#include "trace.h"
  51
  52#define E1000E_MIN_XITR     (500) /* No more then 7813 interrupts per
  53                                     second according to spec 10.2.4.2 */
  54#define E1000E_MAX_TX_FRAGS (64)
  55
  56static inline void
  57e1000e_set_interrupt_cause(E1000ECore *core, uint32_t val);
  58
  59static inline void
  60e1000e_process_ts_option(E1000ECore *core, struct e1000_tx_desc *dp)
  61{
  62    if (le32_to_cpu(dp->upper.data) & E1000_TXD_EXTCMD_TSTAMP) {
  63        trace_e1000e_wrn_no_ts_support();
  64    }
  65}
  66
  67static inline void
  68e1000e_process_snap_option(E1000ECore *core, uint32_t cmd_and_length)
  69{
  70    if (cmd_and_length & E1000_TXD_CMD_SNAP) {
  71        trace_e1000e_wrn_no_snap_support();
  72    }
  73}
  74
  75static inline void
  76e1000e_raise_legacy_irq(E1000ECore *core)
  77{
  78    trace_e1000e_irq_legacy_notify(true);
  79    e1000x_inc_reg_if_not_full(core->mac, IAC);
  80    pci_set_irq(core->owner, 1);
  81}
  82
  83static inline void
  84e1000e_lower_legacy_irq(E1000ECore *core)
  85{
  86    trace_e1000e_irq_legacy_notify(false);
  87    pci_set_irq(core->owner, 0);
  88}
  89
  90static inline void
  91e1000e_intrmgr_rearm_timer(E1000IntrDelayTimer *timer)
  92{
  93    int64_t delay_ns = (int64_t) timer->core->mac[timer->delay_reg] *
  94                                 timer->delay_resolution_ns;
  95
  96    trace_e1000e_irq_rearm_timer(timer->delay_reg << 2, delay_ns);
  97
  98    timer_mod(timer->timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + delay_ns);
  99
 100    timer->running = true;
 101}
 102
 103static void
 104e1000e_intmgr_timer_resume(E1000IntrDelayTimer *timer)
 105{
 106    if (timer->running) {
 107        e1000e_intrmgr_rearm_timer(timer);
 108    }
 109}
 110
 111static void
 112e1000e_intmgr_timer_pause(E1000IntrDelayTimer *timer)
 113{
 114    if (timer->running) {
 115        timer_del(timer->timer);
 116    }
 117}
 118
 119static inline void
 120e1000e_intrmgr_stop_timer(E1000IntrDelayTimer *timer)
 121{
 122    if (timer->running) {
 123        timer_del(timer->timer);
 124        timer->running = false;
 125    }
 126}
 127
 128static inline void
 129e1000e_intrmgr_fire_delayed_interrupts(E1000ECore *core)
 130{
 131    trace_e1000e_irq_fire_delayed_interrupts();
 132    e1000e_set_interrupt_cause(core, 0);
 133}
 134
 135static void
 136e1000e_intrmgr_on_timer(void *opaque)
 137{
 138    E1000IntrDelayTimer *timer = opaque;
 139
 140    trace_e1000e_irq_throttling_timer(timer->delay_reg << 2);
 141
 142    timer->running = false;
 143    e1000e_intrmgr_fire_delayed_interrupts(timer->core);
 144}
 145
 146static void
 147e1000e_intrmgr_on_throttling_timer(void *opaque)
 148{
 149    E1000IntrDelayTimer *timer = opaque;
 150
 151    assert(!msix_enabled(timer->core->owner));
 152
 153    timer->running = false;
 154
 155    if (!timer->core->itr_intr_pending) {
 156        trace_e1000e_irq_throttling_no_pending_interrupts();
 157        return;
 158    }
 159
 160    if (msi_enabled(timer->core->owner)) {
 161        trace_e1000e_irq_msi_notify_postponed();
 162        e1000e_set_interrupt_cause(timer->core, 0);
 163    } else {
 164        trace_e1000e_irq_legacy_notify_postponed();
 165        e1000e_set_interrupt_cause(timer->core, 0);
 166    }
 167}
 168
 169static void
 170e1000e_intrmgr_on_msix_throttling_timer(void *opaque)
 171{
 172    E1000IntrDelayTimer *timer = opaque;
 173    int idx = timer - &timer->core->eitr[0];
 174
 175    assert(msix_enabled(timer->core->owner));
 176
 177    timer->running = false;
 178
 179    if (!timer->core->eitr_intr_pending[idx]) {
 180        trace_e1000e_irq_throttling_no_pending_vec(idx);
 181        return;
 182    }
 183
 184    trace_e1000e_irq_msix_notify_postponed_vec(idx);
 185    msix_notify(timer->core->owner, idx);
 186}
 187
 188static void
 189e1000e_intrmgr_initialize_all_timers(E1000ECore *core, bool create)
 190{
 191    int i;
 192
 193    core->radv.delay_reg = RADV;
 194    core->rdtr.delay_reg = RDTR;
 195    core->raid.delay_reg = RAID;
 196    core->tadv.delay_reg = TADV;
 197    core->tidv.delay_reg = TIDV;
 198
 199    core->radv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 200    core->rdtr.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 201    core->raid.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 202    core->tadv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 203    core->tidv.delay_resolution_ns = E1000_INTR_DELAY_NS_RES;
 204
 205    core->radv.core = core;
 206    core->rdtr.core = core;
 207    core->raid.core = core;
 208    core->tadv.core = core;
 209    core->tidv.core = core;
 210
 211    core->itr.core = core;
 212    core->itr.delay_reg = ITR;
 213    core->itr.delay_resolution_ns = E1000_INTR_THROTTLING_NS_RES;
 214
 215    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 216        core->eitr[i].core = core;
 217        core->eitr[i].delay_reg = EITR + i;
 218        core->eitr[i].delay_resolution_ns = E1000_INTR_THROTTLING_NS_RES;
 219    }
 220
 221    if (!create) {
 222        return;
 223    }
 224
 225    core->radv.timer =
 226        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->radv);
 227    core->rdtr.timer =
 228        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->rdtr);
 229    core->raid.timer =
 230        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->raid);
 231
 232    core->tadv.timer =
 233        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->tadv);
 234    core->tidv.timer =
 235        timer_new_ns(QEMU_CLOCK_VIRTUAL, e1000e_intrmgr_on_timer, &core->tidv);
 236
 237    core->itr.timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
 238                                   e1000e_intrmgr_on_throttling_timer,
 239                                   &core->itr);
 240
 241    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 242        core->eitr[i].timer =
 243            timer_new_ns(QEMU_CLOCK_VIRTUAL,
 244                         e1000e_intrmgr_on_msix_throttling_timer,
 245                         &core->eitr[i]);
 246    }
 247}
 248
 249static inline void
 250e1000e_intrmgr_stop_delay_timers(E1000ECore *core)
 251{
 252    e1000e_intrmgr_stop_timer(&core->radv);
 253    e1000e_intrmgr_stop_timer(&core->rdtr);
 254    e1000e_intrmgr_stop_timer(&core->raid);
 255    e1000e_intrmgr_stop_timer(&core->tidv);
 256    e1000e_intrmgr_stop_timer(&core->tadv);
 257}
 258
 259static bool
 260e1000e_intrmgr_delay_rx_causes(E1000ECore *core, uint32_t *causes)
 261{
 262    uint32_t delayable_causes;
 263    uint32_t rdtr = core->mac[RDTR];
 264    uint32_t radv = core->mac[RADV];
 265    uint32_t raid = core->mac[RAID];
 266
 267    if (msix_enabled(core->owner)) {
 268        return false;
 269    }
 270
 271    delayable_causes = E1000_ICR_RXQ0 |
 272                       E1000_ICR_RXQ1 |
 273                       E1000_ICR_RXT0;
 274
 275    if (!(core->mac[RFCTL] & E1000_RFCTL_ACK_DIS)) {
 276        delayable_causes |= E1000_ICR_ACK;
 277    }
 278
 279    /* Clean up all causes that may be delayed */
 280    core->delayed_causes |= *causes & delayable_causes;
 281    *causes &= ~delayable_causes;
 282
 283    /* Check if delayed RX interrupts disabled by client
 284       or if there are causes that cannot be delayed */
 285    if ((rdtr == 0) || (*causes != 0)) {
 286        return false;
 287    }
 288
 289    /* Check if delayed RX ACK interrupts disabled by client
 290       and there is an ACK packet received */
 291    if ((raid == 0) && (core->delayed_causes & E1000_ICR_ACK)) {
 292        return false;
 293    }
 294
 295    /* All causes delayed */
 296    e1000e_intrmgr_rearm_timer(&core->rdtr);
 297
 298    if (!core->radv.running && (radv != 0)) {
 299        e1000e_intrmgr_rearm_timer(&core->radv);
 300    }
 301
 302    if (!core->raid.running && (core->delayed_causes & E1000_ICR_ACK)) {
 303        e1000e_intrmgr_rearm_timer(&core->raid);
 304    }
 305
 306    return true;
 307}
 308
 309static bool
 310e1000e_intrmgr_delay_tx_causes(E1000ECore *core, uint32_t *causes)
 311{
 312    static const uint32_t delayable_causes = E1000_ICR_TXQ0 |
 313                                             E1000_ICR_TXQ1 |
 314                                             E1000_ICR_TXQE |
 315                                             E1000_ICR_TXDW;
 316
 317    if (msix_enabled(core->owner)) {
 318        return false;
 319    }
 320
 321    /* Clean up all causes that may be delayed */
 322    core->delayed_causes |= *causes & delayable_causes;
 323    *causes &= ~delayable_causes;
 324
 325    /* If there are causes that cannot be delayed */
 326    if (*causes != 0) {
 327        return false;
 328    }
 329
 330    /* All causes delayed */
 331    e1000e_intrmgr_rearm_timer(&core->tidv);
 332
 333    if (!core->tadv.running && (core->mac[TADV] != 0)) {
 334        e1000e_intrmgr_rearm_timer(&core->tadv);
 335    }
 336
 337    return true;
 338}
 339
 340static uint32_t
 341e1000e_intmgr_collect_delayed_causes(E1000ECore *core)
 342{
 343    uint32_t res;
 344
 345    if (msix_enabled(core->owner)) {
 346        assert(core->delayed_causes == 0);
 347        return 0;
 348    }
 349
 350    res = core->delayed_causes;
 351    core->delayed_causes = 0;
 352
 353    e1000e_intrmgr_stop_delay_timers(core);
 354
 355    return res;
 356}
 357
 358static void
 359e1000e_intrmgr_fire_all_timers(E1000ECore *core)
 360{
 361    int i;
 362    uint32_t val = e1000e_intmgr_collect_delayed_causes(core);
 363
 364    trace_e1000e_irq_adding_delayed_causes(val, core->mac[ICR]);
 365    core->mac[ICR] |= val;
 366
 367    if (core->itr.running) {
 368        timer_del(core->itr.timer);
 369        e1000e_intrmgr_on_throttling_timer(&core->itr);
 370    }
 371
 372    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 373        if (core->eitr[i].running) {
 374            timer_del(core->eitr[i].timer);
 375            e1000e_intrmgr_on_msix_throttling_timer(&core->eitr[i]);
 376        }
 377    }
 378}
 379
 380static void
 381e1000e_intrmgr_resume(E1000ECore *core)
 382{
 383    int i;
 384
 385    e1000e_intmgr_timer_resume(&core->radv);
 386    e1000e_intmgr_timer_resume(&core->rdtr);
 387    e1000e_intmgr_timer_resume(&core->raid);
 388    e1000e_intmgr_timer_resume(&core->tidv);
 389    e1000e_intmgr_timer_resume(&core->tadv);
 390
 391    e1000e_intmgr_timer_resume(&core->itr);
 392
 393    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 394        e1000e_intmgr_timer_resume(&core->eitr[i]);
 395    }
 396}
 397
 398static void
 399e1000e_intrmgr_pause(E1000ECore *core)
 400{
 401    int i;
 402
 403    e1000e_intmgr_timer_pause(&core->radv);
 404    e1000e_intmgr_timer_pause(&core->rdtr);
 405    e1000e_intmgr_timer_pause(&core->raid);
 406    e1000e_intmgr_timer_pause(&core->tidv);
 407    e1000e_intmgr_timer_pause(&core->tadv);
 408
 409    e1000e_intmgr_timer_pause(&core->itr);
 410
 411    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 412        e1000e_intmgr_timer_pause(&core->eitr[i]);
 413    }
 414}
 415
 416static void
 417e1000e_intrmgr_reset(E1000ECore *core)
 418{
 419    int i;
 420
 421    core->delayed_causes = 0;
 422
 423    e1000e_intrmgr_stop_delay_timers(core);
 424
 425    e1000e_intrmgr_stop_timer(&core->itr);
 426
 427    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 428        e1000e_intrmgr_stop_timer(&core->eitr[i]);
 429    }
 430}
 431
 432static void
 433e1000e_intrmgr_pci_unint(E1000ECore *core)
 434{
 435    int i;
 436
 437    timer_free(core->radv.timer);
 438    timer_free(core->rdtr.timer);
 439    timer_free(core->raid.timer);
 440
 441    timer_free(core->tadv.timer);
 442    timer_free(core->tidv.timer);
 443
 444    timer_free(core->itr.timer);
 445
 446    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
 447        timer_free(core->eitr[i].timer);
 448    }
 449}
 450
 451static void
 452e1000e_intrmgr_pci_realize(E1000ECore *core)
 453{
 454    e1000e_intrmgr_initialize_all_timers(core, true);
 455}
 456
 457static inline bool
 458e1000e_rx_csum_enabled(E1000ECore *core)
 459{
 460    return (core->mac[RXCSUM] & E1000_RXCSUM_PCSD) ? false : true;
 461}
 462
 463static inline bool
 464e1000e_rx_use_legacy_descriptor(E1000ECore *core)
 465{
 466    return (core->mac[RFCTL] & E1000_RFCTL_EXTEN) ? false : true;
 467}
 468
 469static inline bool
 470e1000e_rx_use_ps_descriptor(E1000ECore *core)
 471{
 472    return !e1000e_rx_use_legacy_descriptor(core) &&
 473           (core->mac[RCTL] & E1000_RCTL_DTYP_PS);
 474}
 475
 476static inline bool
 477e1000e_rss_enabled(E1000ECore *core)
 478{
 479    return E1000_MRQC_ENABLED(core->mac[MRQC]) &&
 480           !e1000e_rx_csum_enabled(core) &&
 481           !e1000e_rx_use_legacy_descriptor(core);
 482}
 483
 484typedef struct E1000E_RSSInfo_st {
 485    bool enabled;
 486    uint32_t hash;
 487    uint32_t queue;
 488    uint32_t type;
 489} E1000E_RSSInfo;
 490
 491static uint32_t
 492e1000e_rss_get_hash_type(E1000ECore *core, struct NetRxPkt *pkt)
 493{
 494    bool isip4, isip6, isudp, istcp;
 495
 496    assert(e1000e_rss_enabled(core));
 497
 498    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
 499
 500    if (isip4) {
 501        bool fragment = net_rx_pkt_get_ip4_info(pkt)->fragment;
 502
 503        trace_e1000e_rx_rss_ip4(fragment, istcp, core->mac[MRQC],
 504                                E1000_MRQC_EN_TCPIPV4(core->mac[MRQC]),
 505                                E1000_MRQC_EN_IPV4(core->mac[MRQC]));
 506
 507        if (!fragment && istcp && E1000_MRQC_EN_TCPIPV4(core->mac[MRQC])) {
 508            return E1000_MRQ_RSS_TYPE_IPV4TCP;
 509        }
 510
 511        if (E1000_MRQC_EN_IPV4(core->mac[MRQC])) {
 512            return E1000_MRQ_RSS_TYPE_IPV4;
 513        }
 514    } else if (isip6) {
 515        eth_ip6_hdr_info *ip6info = net_rx_pkt_get_ip6_info(pkt);
 516
 517        bool ex_dis = core->mac[RFCTL] & E1000_RFCTL_IPV6_EX_DIS;
 518        bool new_ex_dis = core->mac[RFCTL] & E1000_RFCTL_NEW_IPV6_EXT_DIS;
 519
 520        /*
 521         * Following two traces must not be combined because resulting
 522         * event will have 11 arguments totally and some trace backends
 523         * (at least "ust") have limitation of maximum 10 arguments per
 524         * event. Events with more arguments fail to compile for
 525         * backends like these.
 526         */
 527        trace_e1000e_rx_rss_ip6_rfctl(core->mac[RFCTL]);
 528        trace_e1000e_rx_rss_ip6(ex_dis, new_ex_dis, istcp,
 529                                ip6info->has_ext_hdrs,
 530                                ip6info->rss_ex_dst_valid,
 531                                ip6info->rss_ex_src_valid,
 532                                core->mac[MRQC],
 533                                E1000_MRQC_EN_TCPIPV6(core->mac[MRQC]),
 534                                E1000_MRQC_EN_IPV6EX(core->mac[MRQC]),
 535                                E1000_MRQC_EN_IPV6(core->mac[MRQC]));
 536
 537        if ((!ex_dis || !ip6info->has_ext_hdrs) &&
 538            (!new_ex_dis || !(ip6info->rss_ex_dst_valid ||
 539                              ip6info->rss_ex_src_valid))) {
 540
 541            if (istcp && !ip6info->fragment &&
 542                E1000_MRQC_EN_TCPIPV6(core->mac[MRQC])) {
 543                return E1000_MRQ_RSS_TYPE_IPV6TCP;
 544            }
 545
 546            if (E1000_MRQC_EN_IPV6EX(core->mac[MRQC])) {
 547                return E1000_MRQ_RSS_TYPE_IPV6EX;
 548            }
 549
 550        }
 551
 552        if (E1000_MRQC_EN_IPV6(core->mac[MRQC])) {
 553            return E1000_MRQ_RSS_TYPE_IPV6;
 554        }
 555
 556    }
 557
 558    return E1000_MRQ_RSS_TYPE_NONE;
 559}
 560
 561static uint32_t
 562e1000e_rss_calc_hash(E1000ECore *core,
 563                     struct NetRxPkt *pkt,
 564                     E1000E_RSSInfo *info)
 565{
 566    NetRxPktRssType type;
 567
 568    assert(e1000e_rss_enabled(core));
 569
 570    switch (info->type) {
 571    case E1000_MRQ_RSS_TYPE_IPV4:
 572        type = NetPktRssIpV4;
 573        break;
 574    case E1000_MRQ_RSS_TYPE_IPV4TCP:
 575        type = NetPktRssIpV4Tcp;
 576        break;
 577    case E1000_MRQ_RSS_TYPE_IPV6TCP:
 578        type = NetPktRssIpV6TcpEx;
 579        break;
 580    case E1000_MRQ_RSS_TYPE_IPV6:
 581        type = NetPktRssIpV6;
 582        break;
 583    case E1000_MRQ_RSS_TYPE_IPV6EX:
 584        type = NetPktRssIpV6Ex;
 585        break;
 586    default:
 587        assert(false);
 588        return 0;
 589    }
 590
 591    return net_rx_pkt_calc_rss_hash(pkt, type, (uint8_t *) &core->mac[RSSRK]);
 592}
 593
 594static void
 595e1000e_rss_parse_packet(E1000ECore *core,
 596                        struct NetRxPkt *pkt,
 597                        E1000E_RSSInfo *info)
 598{
 599    trace_e1000e_rx_rss_started();
 600
 601    if (!e1000e_rss_enabled(core)) {
 602        info->enabled = false;
 603        info->hash = 0;
 604        info->queue = 0;
 605        info->type = 0;
 606        trace_e1000e_rx_rss_disabled();
 607        return;
 608    }
 609
 610    info->enabled = true;
 611
 612    info->type = e1000e_rss_get_hash_type(core, pkt);
 613
 614    trace_e1000e_rx_rss_type(info->type);
 615
 616    if (info->type == E1000_MRQ_RSS_TYPE_NONE) {
 617        info->hash = 0;
 618        info->queue = 0;
 619        return;
 620    }
 621
 622    info->hash = e1000e_rss_calc_hash(core, pkt, info);
 623    info->queue = E1000_RSS_QUEUE(&core->mac[RETA], info->hash);
 624}
 625
 626static void
 627e1000e_setup_tx_offloads(E1000ECore *core, struct e1000e_tx *tx)
 628{
 629    if (tx->props.tse && tx->cptse) {
 630        net_tx_pkt_build_vheader(tx->tx_pkt, true, true, tx->props.mss);
 631        net_tx_pkt_update_ip_checksums(tx->tx_pkt);
 632        e1000x_inc_reg_if_not_full(core->mac, TSCTC);
 633        return;
 634    }
 635
 636    if (tx->sum_needed & E1000_TXD_POPTS_TXSM) {
 637        net_tx_pkt_build_vheader(tx->tx_pkt, false, true, 0);
 638    }
 639
 640    if (tx->sum_needed & E1000_TXD_POPTS_IXSM) {
 641        net_tx_pkt_update_ip_hdr_checksum(tx->tx_pkt);
 642    }
 643}
 644
 645static bool
 646e1000e_tx_pkt_send(E1000ECore *core, struct e1000e_tx *tx, int queue_index)
 647{
 648    int target_queue = MIN(core->max_queue_num, queue_index);
 649    NetClientState *queue = qemu_get_subqueue(core->owner_nic, target_queue);
 650
 651    e1000e_setup_tx_offloads(core, tx);
 652
 653    net_tx_pkt_dump(tx->tx_pkt);
 654
 655    if ((core->phy[0][PHY_CTRL] & MII_CR_LOOPBACK) ||
 656        ((core->mac[RCTL] & E1000_RCTL_LBM_MAC) == E1000_RCTL_LBM_MAC)) {
 657        return net_tx_pkt_send_loopback(tx->tx_pkt, queue);
 658    } else {
 659        return net_tx_pkt_send(tx->tx_pkt, queue);
 660    }
 661}
 662
 663static void
 664e1000e_on_tx_done_update_stats(E1000ECore *core, struct NetTxPkt *tx_pkt)
 665{
 666    static const int PTCregs[6] = { PTC64, PTC127, PTC255, PTC511,
 667                                    PTC1023, PTC1522 };
 668
 669    size_t tot_len = net_tx_pkt_get_total_len(tx_pkt);
 670
 671    e1000x_increase_size_stats(core->mac, PTCregs, tot_len);
 672    e1000x_inc_reg_if_not_full(core->mac, TPT);
 673    e1000x_grow_8reg_if_not_full(core->mac, TOTL, tot_len);
 674
 675    switch (net_tx_pkt_get_packet_type(tx_pkt)) {
 676    case ETH_PKT_BCAST:
 677        e1000x_inc_reg_if_not_full(core->mac, BPTC);
 678        break;
 679    case ETH_PKT_MCAST:
 680        e1000x_inc_reg_if_not_full(core->mac, MPTC);
 681        break;
 682    case ETH_PKT_UCAST:
 683        break;
 684    default:
 685        g_assert_not_reached();
 686    }
 687
 688    core->mac[GPTC] = core->mac[TPT];
 689    core->mac[GOTCL] = core->mac[TOTL];
 690    core->mac[GOTCH] = core->mac[TOTH];
 691}
 692
 693static void
 694e1000e_process_tx_desc(E1000ECore *core,
 695                       struct e1000e_tx *tx,
 696                       struct e1000_tx_desc *dp,
 697                       int queue_index)
 698{
 699    uint32_t txd_lower = le32_to_cpu(dp->lower.data);
 700    uint32_t dtype = txd_lower & (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D);
 701    unsigned int split_size = txd_lower & 0xffff;
 702    uint64_t addr;
 703    struct e1000_context_desc *xp = (struct e1000_context_desc *)dp;
 704    bool eop = txd_lower & E1000_TXD_CMD_EOP;
 705
 706    if (dtype == E1000_TXD_CMD_DEXT) { /* context descriptor */
 707        e1000x_read_tx_ctx_descr(xp, &tx->props);
 708        e1000e_process_snap_option(core, le32_to_cpu(xp->cmd_and_length));
 709        return;
 710    } else if (dtype == (E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D)) {
 711        /* data descriptor */
 712        tx->sum_needed = le32_to_cpu(dp->upper.data) >> 8;
 713        tx->cptse = (txd_lower & E1000_TXD_CMD_TSE) ? 1 : 0;
 714        e1000e_process_ts_option(core, dp);
 715    } else {
 716        /* legacy descriptor */
 717        e1000e_process_ts_option(core, dp);
 718        tx->cptse = 0;
 719    }
 720
 721    addr = le64_to_cpu(dp->buffer_addr);
 722
 723    if (!tx->skip_cp) {
 724        if (!net_tx_pkt_add_raw_fragment(tx->tx_pkt, addr, split_size)) {
 725            tx->skip_cp = true;
 726        }
 727    }
 728
 729    if (eop) {
 730        if (!tx->skip_cp && net_tx_pkt_parse(tx->tx_pkt)) {
 731            if (e1000x_vlan_enabled(core->mac) &&
 732                e1000x_is_vlan_txd(txd_lower)) {
 733                net_tx_pkt_setup_vlan_header_ex(tx->tx_pkt,
 734                    le16_to_cpu(dp->upper.fields.special), core->mac[VET]);
 735            }
 736            if (e1000e_tx_pkt_send(core, tx, queue_index)) {
 737                e1000e_on_tx_done_update_stats(core, tx->tx_pkt);
 738            }
 739        }
 740
 741        tx->skip_cp = false;
 742        net_tx_pkt_reset(tx->tx_pkt);
 743
 744        tx->sum_needed = 0;
 745        tx->cptse = 0;
 746    }
 747}
 748
 749static inline uint32_t
 750e1000e_tx_wb_interrupt_cause(E1000ECore *core, int queue_idx)
 751{
 752    if (!msix_enabled(core->owner)) {
 753        return E1000_ICR_TXDW;
 754    }
 755
 756    return (queue_idx == 0) ? E1000_ICR_TXQ0 : E1000_ICR_TXQ1;
 757}
 758
 759static inline uint32_t
 760e1000e_rx_wb_interrupt_cause(E1000ECore *core, int queue_idx,
 761                             bool min_threshold_hit)
 762{
 763    if (!msix_enabled(core->owner)) {
 764        return E1000_ICS_RXT0 | (min_threshold_hit ? E1000_ICS_RXDMT0 : 0);
 765    }
 766
 767    return (queue_idx == 0) ? E1000_ICR_RXQ0 : E1000_ICR_RXQ1;
 768}
 769
 770static uint32_t
 771e1000e_txdesc_writeback(E1000ECore *core, dma_addr_t base,
 772                        struct e1000_tx_desc *dp, bool *ide, int queue_idx)
 773{
 774    uint32_t txd_upper, txd_lower = le32_to_cpu(dp->lower.data);
 775
 776    if (!(txd_lower & E1000_TXD_CMD_RS) &&
 777        !(core->mac[IVAR] & E1000_IVAR_TX_INT_EVERY_WB)) {
 778        return 0;
 779    }
 780
 781    *ide = (txd_lower & E1000_TXD_CMD_IDE) ? true : false;
 782
 783    txd_upper = le32_to_cpu(dp->upper.data) | E1000_TXD_STAT_DD;
 784
 785    dp->upper.data = cpu_to_le32(txd_upper);
 786    pci_dma_write(core->owner, base + ((char *)&dp->upper - (char *)dp),
 787                  &dp->upper, sizeof(dp->upper));
 788    return e1000e_tx_wb_interrupt_cause(core, queue_idx);
 789}
 790
 791typedef struct E1000E_RingInfo_st {
 792    int dbah;
 793    int dbal;
 794    int dlen;
 795    int dh;
 796    int dt;
 797    int idx;
 798} E1000E_RingInfo;
 799
 800static inline bool
 801e1000e_ring_empty(E1000ECore *core, const E1000E_RingInfo *r)
 802{
 803    return core->mac[r->dh] == core->mac[r->dt] ||
 804                core->mac[r->dt] >= core->mac[r->dlen] / E1000_RING_DESC_LEN;
 805}
 806
 807static inline uint64_t
 808e1000e_ring_base(E1000ECore *core, const E1000E_RingInfo *r)
 809{
 810    uint64_t bah = core->mac[r->dbah];
 811    uint64_t bal = core->mac[r->dbal];
 812
 813    return (bah << 32) + bal;
 814}
 815
 816static inline uint64_t
 817e1000e_ring_head_descr(E1000ECore *core, const E1000E_RingInfo *r)
 818{
 819    return e1000e_ring_base(core, r) + E1000_RING_DESC_LEN * core->mac[r->dh];
 820}
 821
 822static inline void
 823e1000e_ring_advance(E1000ECore *core, const E1000E_RingInfo *r, uint32_t count)
 824{
 825    core->mac[r->dh] += count;
 826
 827    if (core->mac[r->dh] * E1000_RING_DESC_LEN >= core->mac[r->dlen]) {
 828        core->mac[r->dh] = 0;
 829    }
 830}
 831
 832static inline uint32_t
 833e1000e_ring_free_descr_num(E1000ECore *core, const E1000E_RingInfo *r)
 834{
 835    trace_e1000e_ring_free_space(r->idx, core->mac[r->dlen],
 836                                 core->mac[r->dh],  core->mac[r->dt]);
 837
 838    if (core->mac[r->dh] <= core->mac[r->dt]) {
 839        return core->mac[r->dt] - core->mac[r->dh];
 840    }
 841
 842    if (core->mac[r->dh] > core->mac[r->dt]) {
 843        return core->mac[r->dlen] / E1000_RING_DESC_LEN +
 844               core->mac[r->dt] - core->mac[r->dh];
 845    }
 846
 847    g_assert_not_reached();
 848    return 0;
 849}
 850
 851static inline bool
 852e1000e_ring_enabled(E1000ECore *core, const E1000E_RingInfo *r)
 853{
 854    return core->mac[r->dlen] > 0;
 855}
 856
 857static inline uint32_t
 858e1000e_ring_len(E1000ECore *core, const E1000E_RingInfo *r)
 859{
 860    return core->mac[r->dlen];
 861}
 862
 863typedef struct E1000E_TxRing_st {
 864    const E1000E_RingInfo *i;
 865    struct e1000e_tx *tx;
 866} E1000E_TxRing;
 867
 868static inline int
 869e1000e_mq_queue_idx(int base_reg_idx, int reg_idx)
 870{
 871    return (reg_idx - base_reg_idx) / (0x100 >> 2);
 872}
 873
 874static inline void
 875e1000e_tx_ring_init(E1000ECore *core, E1000E_TxRing *txr, int idx)
 876{
 877    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 878        { TDBAH,  TDBAL,  TDLEN,  TDH,  TDT, 0 },
 879        { TDBAH1, TDBAL1, TDLEN1, TDH1, TDT1, 1 }
 880    };
 881
 882    assert(idx < ARRAY_SIZE(i));
 883
 884    txr->i     = &i[idx];
 885    txr->tx    = &core->tx[idx];
 886}
 887
 888typedef struct E1000E_RxRing_st {
 889    const E1000E_RingInfo *i;
 890} E1000E_RxRing;
 891
 892static inline void
 893e1000e_rx_ring_init(E1000ECore *core, E1000E_RxRing *rxr, int idx)
 894{
 895    static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = {
 896        { RDBAH0, RDBAL0, RDLEN0, RDH0, RDT0, 0 },
 897        { RDBAH1, RDBAL1, RDLEN1, RDH1, RDT1, 1 }
 898    };
 899
 900    assert(idx < ARRAY_SIZE(i));
 901
 902    rxr->i      = &i[idx];
 903}
 904
 905static void
 906e1000e_start_xmit(E1000ECore *core, const E1000E_TxRing *txr)
 907{
 908    dma_addr_t base;
 909    struct e1000_tx_desc desc;
 910    bool ide = false;
 911    const E1000E_RingInfo *txi = txr->i;
 912    uint32_t cause = E1000_ICS_TXQE;
 913
 914    if (!(core->mac[TCTL] & E1000_TCTL_EN)) {
 915        trace_e1000e_tx_disabled();
 916        return;
 917    }
 918
 919    while (!e1000e_ring_empty(core, txi)) {
 920        base = e1000e_ring_head_descr(core, txi);
 921
 922        pci_dma_read(core->owner, base, &desc, sizeof(desc));
 923
 924        trace_e1000e_tx_descr((void *)(intptr_t)desc.buffer_addr,
 925                              desc.lower.data, desc.upper.data);
 926
 927        e1000e_process_tx_desc(core, txr->tx, &desc, txi->idx);
 928        cause |= e1000e_txdesc_writeback(core, base, &desc, &ide, txi->idx);
 929
 930        e1000e_ring_advance(core, txi, 1);
 931    }
 932
 933    if (!ide || !e1000e_intrmgr_delay_tx_causes(core, &cause)) {
 934        e1000e_set_interrupt_cause(core, cause);
 935    }
 936}
 937
 938static bool
 939e1000e_has_rxbufs(E1000ECore *core, const E1000E_RingInfo *r,
 940                  size_t total_size)
 941{
 942    uint32_t bufs = e1000e_ring_free_descr_num(core, r);
 943
 944    trace_e1000e_rx_has_buffers(r->idx, bufs, total_size,
 945                                core->rx_desc_buf_size);
 946
 947    return total_size <= bufs / (core->rx_desc_len / E1000_MIN_RX_DESC_LEN) *
 948                         core->rx_desc_buf_size;
 949}
 950
 951void
 952e1000e_start_recv(E1000ECore *core)
 953{
 954    int i;
 955
 956    trace_e1000e_rx_start_recv();
 957
 958    for (i = 0; i <= core->max_queue_num; i++) {
 959        qemu_flush_queued_packets(qemu_get_subqueue(core->owner_nic, i));
 960    }
 961}
 962
 963bool
 964e1000e_can_receive(E1000ECore *core)
 965{
 966    int i;
 967
 968    if (!e1000x_rx_ready(core->owner, core->mac)) {
 969        return false;
 970    }
 971
 972    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
 973        E1000E_RxRing rxr;
 974
 975        e1000e_rx_ring_init(core, &rxr, i);
 976        if (e1000e_ring_enabled(core, rxr.i) &&
 977            e1000e_has_rxbufs(core, rxr.i, 1)) {
 978            trace_e1000e_rx_can_recv();
 979            return true;
 980        }
 981    }
 982
 983    trace_e1000e_rx_can_recv_rings_full();
 984    return false;
 985}
 986
 987ssize_t
 988e1000e_receive(E1000ECore *core, const uint8_t *buf, size_t size)
 989{
 990    const struct iovec iov = {
 991        .iov_base = (uint8_t *)buf,
 992        .iov_len = size
 993    };
 994
 995    return e1000e_receive_iov(core, &iov, 1);
 996}
 997
 998static inline bool
 999e1000e_rx_l3_cso_enabled(E1000ECore *core)
1000{
1001    return !!(core->mac[RXCSUM] & E1000_RXCSUM_IPOFLD);
1002}
1003
1004static inline bool
1005e1000e_rx_l4_cso_enabled(E1000ECore *core)
1006{
1007    return !!(core->mac[RXCSUM] & E1000_RXCSUM_TUOFLD);
1008}
1009
1010static bool
1011e1000e_receive_filter(E1000ECore *core, const uint8_t *buf, int size)
1012{
1013    uint32_t rctl = core->mac[RCTL];
1014
1015    if (e1000x_is_vlan_packet(buf, core->mac[VET]) &&
1016        e1000x_vlan_rx_filter_enabled(core->mac)) {
1017        uint16_t vid = lduw_be_p(buf + 14);
1018        uint32_t vfta = ldl_le_p((uint32_t *)(core->mac + VFTA) +
1019                                 ((vid >> 5) & 0x7f));
1020        if ((vfta & (1 << (vid & 0x1f))) == 0) {
1021            trace_e1000e_rx_flt_vlan_mismatch(vid);
1022            return false;
1023        } else {
1024            trace_e1000e_rx_flt_vlan_match(vid);
1025        }
1026    }
1027
1028    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1029    case ETH_PKT_UCAST:
1030        if (rctl & E1000_RCTL_UPE) {
1031            return true; /* promiscuous ucast */
1032        }
1033        break;
1034
1035    case ETH_PKT_BCAST:
1036        if (rctl & E1000_RCTL_BAM) {
1037            return true; /* broadcast enabled */
1038        }
1039        break;
1040
1041    case ETH_PKT_MCAST:
1042        if (rctl & E1000_RCTL_MPE) {
1043            return true; /* promiscuous mcast */
1044        }
1045        break;
1046
1047    default:
1048        g_assert_not_reached();
1049    }
1050
1051    return e1000x_rx_group_filter(core->mac, buf);
1052}
1053
1054static inline void
1055e1000e_read_lgcy_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1056{
1057    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1058    *buff_addr = le64_to_cpu(d->buffer_addr);
1059}
1060
1061static inline void
1062e1000e_read_ext_rx_descr(E1000ECore *core, uint8_t *desc, hwaddr *buff_addr)
1063{
1064    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1065    *buff_addr = le64_to_cpu(d->read.buffer_addr);
1066}
1067
1068static inline void
1069e1000e_read_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1070                        hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1071{
1072    int i;
1073    union e1000_rx_desc_packet_split *d =
1074        (union e1000_rx_desc_packet_split *) desc;
1075
1076    for (i = 0; i < MAX_PS_BUFFERS; i++) {
1077        (*buff_addr)[i] = le64_to_cpu(d->read.buffer_addr[i]);
1078    }
1079
1080    trace_e1000e_rx_desc_ps_read((*buff_addr)[0], (*buff_addr)[1],
1081                                 (*buff_addr)[2], (*buff_addr)[3]);
1082}
1083
1084static inline void
1085e1000e_read_rx_descr(E1000ECore *core, uint8_t *desc,
1086                     hwaddr (*buff_addr)[MAX_PS_BUFFERS])
1087{
1088    if (e1000e_rx_use_legacy_descriptor(core)) {
1089        e1000e_read_lgcy_rx_descr(core, desc, &(*buff_addr)[0]);
1090        (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1091    } else {
1092        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1093            e1000e_read_ps_rx_descr(core, desc, buff_addr);
1094        } else {
1095            e1000e_read_ext_rx_descr(core, desc, &(*buff_addr)[0]);
1096            (*buff_addr)[1] = (*buff_addr)[2] = (*buff_addr)[3] = 0;
1097        }
1098    }
1099}
1100
1101static void
1102e1000e_verify_csum_in_sw(E1000ECore *core,
1103                         struct NetRxPkt *pkt,
1104                         uint32_t *status_flags,
1105                         bool istcp, bool isudp)
1106{
1107    bool csum_valid;
1108    uint32_t csum_error;
1109
1110    if (e1000e_rx_l3_cso_enabled(core)) {
1111        if (!net_rx_pkt_validate_l3_csum(pkt, &csum_valid)) {
1112            trace_e1000e_rx_metadata_l3_csum_validation_failed();
1113        } else {
1114            csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_IPE;
1115            *status_flags |= E1000_RXD_STAT_IPCS | csum_error;
1116        }
1117    } else {
1118        trace_e1000e_rx_metadata_l3_cso_disabled();
1119    }
1120
1121    if (!e1000e_rx_l4_cso_enabled(core)) {
1122        trace_e1000e_rx_metadata_l4_cso_disabled();
1123        return;
1124    }
1125
1126    if (!net_rx_pkt_validate_l4_csum(pkt, &csum_valid)) {
1127        trace_e1000e_rx_metadata_l4_csum_validation_failed();
1128        return;
1129    }
1130
1131    csum_error = csum_valid ? 0 : E1000_RXDEXT_STATERR_TCPE;
1132
1133    if (istcp) {
1134        *status_flags |= E1000_RXD_STAT_TCPCS |
1135                         csum_error;
1136    } else if (isudp) {
1137        *status_flags |= E1000_RXD_STAT_TCPCS |
1138                         E1000_RXD_STAT_UDPCS |
1139                         csum_error;
1140    }
1141}
1142
1143static inline bool
1144e1000e_is_tcp_ack(E1000ECore *core, struct NetRxPkt *rx_pkt)
1145{
1146    if (!net_rx_pkt_is_tcp_ack(rx_pkt)) {
1147        return false;
1148    }
1149
1150    if (core->mac[RFCTL] & E1000_RFCTL_ACK_DATA_DIS) {
1151        return !net_rx_pkt_has_tcp_data(rx_pkt);
1152    }
1153
1154    return true;
1155}
1156
1157static void
1158e1000e_build_rx_metadata(E1000ECore *core,
1159                         struct NetRxPkt *pkt,
1160                         bool is_eop,
1161                         const E1000E_RSSInfo *rss_info,
1162                         uint32_t *rss, uint32_t *mrq,
1163                         uint32_t *status_flags,
1164                         uint16_t *ip_id,
1165                         uint16_t *vlan_tag)
1166{
1167    struct virtio_net_hdr *vhdr;
1168    bool isip4, isip6, istcp, isudp;
1169    uint32_t pkt_type;
1170
1171    *status_flags = E1000_RXD_STAT_DD;
1172
1173    /* No additional metadata needed for non-EOP descriptors */
1174    if (!is_eop) {
1175        goto func_exit;
1176    }
1177
1178    *status_flags |= E1000_RXD_STAT_EOP;
1179
1180    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1181    trace_e1000e_rx_metadata_protocols(isip4, isip6, isudp, istcp);
1182
1183    /* VLAN state */
1184    if (net_rx_pkt_is_vlan_stripped(pkt)) {
1185        *status_flags |= E1000_RXD_STAT_VP;
1186        *vlan_tag = cpu_to_le16(net_rx_pkt_get_vlan_tag(pkt));
1187        trace_e1000e_rx_metadata_vlan(*vlan_tag);
1188    }
1189
1190    /* Packet parsing results */
1191    if ((core->mac[RXCSUM] & E1000_RXCSUM_PCSD) != 0) {
1192        if (rss_info->enabled) {
1193            *rss = cpu_to_le32(rss_info->hash);
1194            *mrq = cpu_to_le32(rss_info->type | (rss_info->queue << 8));
1195            trace_e1000e_rx_metadata_rss(*rss, *mrq);
1196        }
1197    } else if (isip4) {
1198            *status_flags |= E1000_RXD_STAT_IPIDV;
1199            *ip_id = cpu_to_le16(net_rx_pkt_get_ip_id(pkt));
1200            trace_e1000e_rx_metadata_ip_id(*ip_id);
1201    }
1202
1203    if (istcp && e1000e_is_tcp_ack(core, pkt)) {
1204        *status_flags |= E1000_RXD_STAT_ACK;
1205        trace_e1000e_rx_metadata_ack();
1206    }
1207
1208    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_DIS)) {
1209        trace_e1000e_rx_metadata_ipv6_filtering_disabled();
1210        pkt_type = E1000_RXD_PKT_MAC;
1211    } else if (istcp || isudp) {
1212        pkt_type = isip4 ? E1000_RXD_PKT_IP4_XDP : E1000_RXD_PKT_IP6_XDP;
1213    } else if (isip4 || isip6) {
1214        pkt_type = isip4 ? E1000_RXD_PKT_IP4 : E1000_RXD_PKT_IP6;
1215    } else {
1216        pkt_type = E1000_RXD_PKT_MAC;
1217    }
1218
1219    *status_flags |= E1000_RXD_PKT_TYPE(pkt_type);
1220    trace_e1000e_rx_metadata_pkt_type(pkt_type);
1221
1222    /* RX CSO information */
1223    if (isip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_XSUM_DIS)) {
1224        trace_e1000e_rx_metadata_ipv6_sum_disabled();
1225        goto func_exit;
1226    }
1227
1228    if (!net_rx_pkt_has_virt_hdr(pkt)) {
1229        trace_e1000e_rx_metadata_no_virthdr();
1230        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1231        goto func_exit;
1232    }
1233
1234    vhdr = net_rx_pkt_get_vhdr(pkt);
1235
1236    if (!(vhdr->flags & VIRTIO_NET_HDR_F_DATA_VALID) &&
1237        !(vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) {
1238        trace_e1000e_rx_metadata_virthdr_no_csum_info();
1239        e1000e_verify_csum_in_sw(core, pkt, status_flags, istcp, isudp);
1240        goto func_exit;
1241    }
1242
1243    if (e1000e_rx_l3_cso_enabled(core)) {
1244        *status_flags |= isip4 ? E1000_RXD_STAT_IPCS : 0;
1245    } else {
1246        trace_e1000e_rx_metadata_l3_cso_disabled();
1247    }
1248
1249    if (e1000e_rx_l4_cso_enabled(core)) {
1250        if (istcp) {
1251            *status_flags |= E1000_RXD_STAT_TCPCS;
1252        } else if (isudp) {
1253            *status_flags |= E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS;
1254        }
1255    } else {
1256        trace_e1000e_rx_metadata_l4_cso_disabled();
1257    }
1258
1259    trace_e1000e_rx_metadata_status_flags(*status_flags);
1260
1261func_exit:
1262    *status_flags = cpu_to_le32(*status_flags);
1263}
1264
1265static inline void
1266e1000e_write_lgcy_rx_descr(E1000ECore *core, uint8_t *desc,
1267                           struct NetRxPkt *pkt,
1268                           const E1000E_RSSInfo *rss_info,
1269                           uint16_t length)
1270{
1271    uint32_t status_flags, rss, mrq;
1272    uint16_t ip_id;
1273
1274    struct e1000_rx_desc *d = (struct e1000_rx_desc *) desc;
1275
1276    assert(!rss_info->enabled);
1277
1278    d->length = cpu_to_le16(length);
1279    d->csum = 0;
1280
1281    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1282                             rss_info,
1283                             &rss, &mrq,
1284                             &status_flags, &ip_id,
1285                             &d->special);
1286    d->errors = (uint8_t) (le32_to_cpu(status_flags) >> 24);
1287    d->status = (uint8_t) le32_to_cpu(status_flags);
1288}
1289
1290static inline void
1291e1000e_write_ext_rx_descr(E1000ECore *core, uint8_t *desc,
1292                          struct NetRxPkt *pkt,
1293                          const E1000E_RSSInfo *rss_info,
1294                          uint16_t length)
1295{
1296    union e1000_rx_desc_extended *d = (union e1000_rx_desc_extended *) desc;
1297
1298    memset(&d->wb, 0, sizeof(d->wb));
1299
1300    d->wb.upper.length = cpu_to_le16(length);
1301
1302    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1303                             rss_info,
1304                             &d->wb.lower.hi_dword.rss,
1305                             &d->wb.lower.mrq,
1306                             &d->wb.upper.status_error,
1307                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1308                             &d->wb.upper.vlan);
1309}
1310
1311static inline void
1312e1000e_write_ps_rx_descr(E1000ECore *core, uint8_t *desc,
1313                         struct NetRxPkt *pkt,
1314                         const E1000E_RSSInfo *rss_info,
1315                         size_t ps_hdr_len,
1316                         uint16_t(*written)[MAX_PS_BUFFERS])
1317{
1318    int i;
1319    union e1000_rx_desc_packet_split *d =
1320        (union e1000_rx_desc_packet_split *) desc;
1321
1322    memset(&d->wb, 0, sizeof(d->wb));
1323
1324    d->wb.middle.length0 = cpu_to_le16((*written)[0]);
1325
1326    for (i = 0; i < PS_PAGE_BUFFERS; i++) {
1327        d->wb.upper.length[i] = cpu_to_le16((*written)[i + 1]);
1328    }
1329
1330    e1000e_build_rx_metadata(core, pkt, pkt != NULL,
1331                             rss_info,
1332                             &d->wb.lower.hi_dword.rss,
1333                             &d->wb.lower.mrq,
1334                             &d->wb.middle.status_error,
1335                             &d->wb.lower.hi_dword.csum_ip.ip_id,
1336                             &d->wb.middle.vlan);
1337
1338    d->wb.upper.header_status =
1339        cpu_to_le16(ps_hdr_len | (ps_hdr_len ? E1000_RXDPS_HDRSTAT_HDRSP : 0));
1340
1341    trace_e1000e_rx_desc_ps_write((*written)[0], (*written)[1],
1342                                  (*written)[2], (*written)[3]);
1343}
1344
1345static inline void
1346e1000e_write_rx_descr(E1000ECore *core, uint8_t *desc,
1347struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info,
1348    size_t ps_hdr_len, uint16_t(*written)[MAX_PS_BUFFERS])
1349{
1350    if (e1000e_rx_use_legacy_descriptor(core)) {
1351        assert(ps_hdr_len == 0);
1352        e1000e_write_lgcy_rx_descr(core, desc, pkt, rss_info, (*written)[0]);
1353    } else {
1354        if (core->mac[RCTL] & E1000_RCTL_DTYP_PS) {
1355            e1000e_write_ps_rx_descr(core, desc, pkt, rss_info,
1356                                      ps_hdr_len, written);
1357        } else {
1358            assert(ps_hdr_len == 0);
1359            e1000e_write_ext_rx_descr(core, desc, pkt, rss_info,
1360                                       (*written)[0]);
1361        }
1362    }
1363}
1364
1365typedef struct e1000e_ba_state_st {
1366    uint16_t written[MAX_PS_BUFFERS];
1367    uint8_t cur_idx;
1368} e1000e_ba_state;
1369
1370static inline void
1371e1000e_write_hdr_to_rx_buffers(E1000ECore *core,
1372                               hwaddr (*ba)[MAX_PS_BUFFERS],
1373                               e1000e_ba_state *bastate,
1374                               const char *data,
1375                               dma_addr_t data_len)
1376{
1377    assert(data_len <= core->rxbuf_sizes[0] - bastate->written[0]);
1378
1379    pci_dma_write(core->owner, (*ba)[0] + bastate->written[0], data, data_len);
1380    bastate->written[0] += data_len;
1381
1382    bastate->cur_idx = 1;
1383}
1384
1385static void
1386e1000e_write_to_rx_buffers(E1000ECore *core,
1387                           hwaddr (*ba)[MAX_PS_BUFFERS],
1388                           e1000e_ba_state *bastate,
1389                           const char *data,
1390                           dma_addr_t data_len)
1391{
1392    while (data_len > 0) {
1393        uint32_t cur_buf_len = core->rxbuf_sizes[bastate->cur_idx];
1394        uint32_t cur_buf_bytes_left = cur_buf_len -
1395                                      bastate->written[bastate->cur_idx];
1396        uint32_t bytes_to_write = MIN(data_len, cur_buf_bytes_left);
1397
1398        trace_e1000e_rx_desc_buff_write(bastate->cur_idx,
1399                                        (*ba)[bastate->cur_idx],
1400                                        bastate->written[bastate->cur_idx],
1401                                        data,
1402                                        bytes_to_write);
1403
1404        pci_dma_write(core->owner,
1405            (*ba)[bastate->cur_idx] + bastate->written[bastate->cur_idx],
1406            data, bytes_to_write);
1407
1408        bastate->written[bastate->cur_idx] += bytes_to_write;
1409        data += bytes_to_write;
1410        data_len -= bytes_to_write;
1411
1412        if (bastate->written[bastate->cur_idx] == cur_buf_len) {
1413            bastate->cur_idx++;
1414        }
1415
1416        assert(bastate->cur_idx < MAX_PS_BUFFERS);
1417    }
1418}
1419
1420static void
1421e1000e_update_rx_stats(E1000ECore *core,
1422                       size_t data_size,
1423                       size_t data_fcs_size)
1424{
1425    e1000x_update_rx_total_stats(core->mac, data_size, data_fcs_size);
1426
1427    switch (net_rx_pkt_get_packet_type(core->rx_pkt)) {
1428    case ETH_PKT_BCAST:
1429        e1000x_inc_reg_if_not_full(core->mac, BPRC);
1430        break;
1431
1432    case ETH_PKT_MCAST:
1433        e1000x_inc_reg_if_not_full(core->mac, MPRC);
1434        break;
1435
1436    default:
1437        break;
1438    }
1439}
1440
1441static inline bool
1442e1000e_rx_descr_threshold_hit(E1000ECore *core, const E1000E_RingInfo *rxi)
1443{
1444    return e1000e_ring_free_descr_num(core, rxi) ==
1445           e1000e_ring_len(core, rxi) >> core->rxbuf_min_shift;
1446}
1447
1448static bool
1449e1000e_do_ps(E1000ECore *core, struct NetRxPkt *pkt, size_t *hdr_len)
1450{
1451    bool isip4, isip6, isudp, istcp;
1452    bool fragment;
1453
1454    if (!e1000e_rx_use_ps_descriptor(core)) {
1455        return false;
1456    }
1457
1458    net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp);
1459
1460    if (isip4) {
1461        fragment = net_rx_pkt_get_ip4_info(pkt)->fragment;
1462    } else if (isip6) {
1463        fragment = net_rx_pkt_get_ip6_info(pkt)->fragment;
1464    } else {
1465        return false;
1466    }
1467
1468    if (fragment && (core->mac[RFCTL] & E1000_RFCTL_IPFRSP_DIS)) {
1469        return false;
1470    }
1471
1472    if (!fragment && (isudp || istcp)) {
1473        *hdr_len = net_rx_pkt_get_l5_hdr_offset(pkt);
1474    } else {
1475        *hdr_len = net_rx_pkt_get_l4_hdr_offset(pkt);
1476    }
1477
1478    if ((*hdr_len > core->rxbuf_sizes[0]) ||
1479        (*hdr_len > net_rx_pkt_get_total_len(pkt))) {
1480        return false;
1481    }
1482
1483    return true;
1484}
1485
1486static void
1487e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt,
1488                             const E1000E_RxRing *rxr,
1489                             const E1000E_RSSInfo *rss_info)
1490{
1491    PCIDevice *d = core->owner;
1492    dma_addr_t base;
1493    uint8_t desc[E1000_MAX_RX_DESC_LEN];
1494    size_t desc_size;
1495    size_t desc_offset = 0;
1496    size_t iov_ofs = 0;
1497
1498    struct iovec *iov = net_rx_pkt_get_iovec(pkt);
1499    size_t size = net_rx_pkt_get_total_len(pkt);
1500    size_t total_size = size + e1000x_fcs_len(core->mac);
1501    const E1000E_RingInfo *rxi;
1502    size_t ps_hdr_len = 0;
1503    bool do_ps = e1000e_do_ps(core, pkt, &ps_hdr_len);
1504    bool is_first = true;
1505
1506    rxi = rxr->i;
1507
1508    do {
1509        hwaddr ba[MAX_PS_BUFFERS];
1510        e1000e_ba_state bastate = { { 0 } };
1511        bool is_last = false;
1512
1513        desc_size = total_size - desc_offset;
1514
1515        if (desc_size > core->rx_desc_buf_size) {
1516            desc_size = core->rx_desc_buf_size;
1517        }
1518
1519        if (e1000e_ring_empty(core, rxi)) {
1520            return;
1521        }
1522
1523        base = e1000e_ring_head_descr(core, rxi);
1524
1525        pci_dma_read(d, base, &desc, core->rx_desc_len);
1526
1527        trace_e1000e_rx_descr(rxi->idx, base, core->rx_desc_len);
1528
1529        e1000e_read_rx_descr(core, desc, &ba);
1530
1531        if (ba[0]) {
1532            if (desc_offset < size) {
1533                static const uint32_t fcs_pad;
1534                size_t iov_copy;
1535                size_t copy_size = size - desc_offset;
1536                if (copy_size > core->rx_desc_buf_size) {
1537                    copy_size = core->rx_desc_buf_size;
1538                }
1539
1540                /* For PS mode copy the packet header first */
1541                if (do_ps) {
1542                    if (is_first) {
1543                        size_t ps_hdr_copied = 0;
1544                        do {
1545                            iov_copy = MIN(ps_hdr_len - ps_hdr_copied,
1546                                           iov->iov_len - iov_ofs);
1547
1548                            e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1549                                                      iov->iov_base, iov_copy);
1550
1551                            copy_size -= iov_copy;
1552                            ps_hdr_copied += iov_copy;
1553
1554                            iov_ofs += iov_copy;
1555                            if (iov_ofs == iov->iov_len) {
1556                                iov++;
1557                                iov_ofs = 0;
1558                            }
1559                        } while (ps_hdr_copied < ps_hdr_len);
1560
1561                        is_first = false;
1562                    } else {
1563                        /* Leave buffer 0 of each descriptor except first */
1564                        /* empty as per spec 7.1.5.1                      */
1565                        e1000e_write_hdr_to_rx_buffers(core, &ba, &bastate,
1566                                                       NULL, 0);
1567                    }
1568                }
1569
1570                /* Copy packet payload */
1571                while (copy_size) {
1572                    iov_copy = MIN(copy_size, iov->iov_len - iov_ofs);
1573
1574                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1575                                            iov->iov_base + iov_ofs, iov_copy);
1576
1577                    copy_size -= iov_copy;
1578                    iov_ofs += iov_copy;
1579                    if (iov_ofs == iov->iov_len) {
1580                        iov++;
1581                        iov_ofs = 0;
1582                    }
1583                }
1584
1585                if (desc_offset + desc_size >= total_size) {
1586                    /* Simulate FCS checksum presence in the last descriptor */
1587                    e1000e_write_to_rx_buffers(core, &ba, &bastate,
1588                          (const char *) &fcs_pad, e1000x_fcs_len(core->mac));
1589                }
1590            }
1591        } else { /* as per intel docs; skip descriptors with null buf addr */
1592            trace_e1000e_rx_null_descriptor();
1593        }
1594        desc_offset += desc_size;
1595        if (desc_offset >= total_size) {
1596            is_last = true;
1597        }
1598
1599        e1000e_write_rx_descr(core, desc, is_last ? core->rx_pkt : NULL,
1600                           rss_info, do_ps ? ps_hdr_len : 0, &bastate.written);
1601        pci_dma_write(d, base, &desc, core->rx_desc_len);
1602
1603        e1000e_ring_advance(core, rxi,
1604                            core->rx_desc_len / E1000_MIN_RX_DESC_LEN);
1605
1606    } while (desc_offset < total_size);
1607
1608    e1000e_update_rx_stats(core, size, total_size);
1609}
1610
1611static inline void
1612e1000e_rx_fix_l4_csum(E1000ECore *core, struct NetRxPkt *pkt)
1613{
1614    if (net_rx_pkt_has_virt_hdr(pkt)) {
1615        struct virtio_net_hdr *vhdr = net_rx_pkt_get_vhdr(pkt);
1616
1617        if (vhdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1618            net_rx_pkt_fix_l4_csum(pkt);
1619        }
1620    }
1621}
1622
1623ssize_t
1624e1000e_receive_iov(E1000ECore *core, const struct iovec *iov, int iovcnt)
1625{
1626    static const int maximum_ethernet_hdr_len = (14 + 4);
1627    /* Min. octets in an ethernet frame sans FCS */
1628    static const int min_buf_size = 60;
1629
1630    uint32_t n = 0;
1631    uint8_t min_buf[min_buf_size];
1632    struct iovec min_iov;
1633    uint8_t *filter_buf;
1634    size_t size, orig_size;
1635    size_t iov_ofs = 0;
1636    E1000E_RxRing rxr;
1637    E1000E_RSSInfo rss_info;
1638    size_t total_size;
1639    ssize_t retval;
1640    bool rdmts_hit;
1641
1642    trace_e1000e_rx_receive_iov(iovcnt);
1643
1644    if (!e1000x_hw_rx_enabled(core->mac)) {
1645        return -1;
1646    }
1647
1648    /* Pull virtio header in */
1649    if (core->has_vnet) {
1650        net_rx_pkt_set_vhdr_iovec(core->rx_pkt, iov, iovcnt);
1651        iov_ofs = sizeof(struct virtio_net_hdr);
1652    }
1653
1654    filter_buf = iov->iov_base + iov_ofs;
1655    orig_size = iov_size(iov, iovcnt);
1656    size = orig_size - iov_ofs;
1657
1658    /* Pad to minimum Ethernet frame length */
1659    if (size < sizeof(min_buf)) {
1660        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, size);
1661        memset(&min_buf[size], 0, sizeof(min_buf) - size);
1662        e1000x_inc_reg_if_not_full(core->mac, RUC);
1663        min_iov.iov_base = filter_buf = min_buf;
1664        min_iov.iov_len = size = sizeof(min_buf);
1665        iovcnt = 1;
1666        iov = &min_iov;
1667        iov_ofs = 0;
1668    } else if (iov->iov_len < maximum_ethernet_hdr_len) {
1669        /* This is very unlikely, but may happen. */
1670        iov_to_buf(iov, iovcnt, iov_ofs, min_buf, maximum_ethernet_hdr_len);
1671        filter_buf = min_buf;
1672    }
1673
1674    /* Discard oversized packets if !LPE and !SBP. */
1675    if (e1000x_is_oversized(core->mac, size)) {
1676        return orig_size;
1677    }
1678
1679    net_rx_pkt_set_packet_type(core->rx_pkt,
1680        get_eth_packet_type(PKT_GET_ETH_HDR(filter_buf)));
1681
1682    if (!e1000e_receive_filter(core, filter_buf, size)) {
1683        trace_e1000e_rx_flt_dropped();
1684        return orig_size;
1685    }
1686
1687    net_rx_pkt_attach_iovec_ex(core->rx_pkt, iov, iovcnt, iov_ofs,
1688                               e1000x_vlan_enabled(core->mac), core->mac[VET]);
1689
1690    e1000e_rss_parse_packet(core, core->rx_pkt, &rss_info);
1691    e1000e_rx_ring_init(core, &rxr, rss_info.queue);
1692
1693    trace_e1000e_rx_rss_dispatched_to_queue(rxr.i->idx);
1694
1695    total_size = net_rx_pkt_get_total_len(core->rx_pkt) +
1696        e1000x_fcs_len(core->mac);
1697
1698    if (e1000e_has_rxbufs(core, rxr.i, total_size)) {
1699        e1000e_rx_fix_l4_csum(core, core->rx_pkt);
1700
1701        e1000e_write_packet_to_guest(core, core->rx_pkt, &rxr, &rss_info);
1702
1703        retval = orig_size;
1704
1705        /* Perform small receive detection (RSRPD) */
1706        if (total_size < core->mac[RSRPD]) {
1707            n |= E1000_ICS_SRPD;
1708        }
1709
1710        /* Perform ACK receive detection */
1711        if  (!(core->mac[RFCTL] & E1000_RFCTL_ACK_DIS) &&
1712             (e1000e_is_tcp_ack(core, core->rx_pkt))) {
1713            n |= E1000_ICS_ACK;
1714        }
1715
1716        /* Check if receive descriptor minimum threshold hit */
1717        rdmts_hit = e1000e_rx_descr_threshold_hit(core, rxr.i);
1718        n |= e1000e_rx_wb_interrupt_cause(core, rxr.i->idx, rdmts_hit);
1719
1720        trace_e1000e_rx_written_to_guest(n);
1721    } else {
1722        n |= E1000_ICS_RXO;
1723        retval = 0;
1724
1725        trace_e1000e_rx_not_written_to_guest(n);
1726    }
1727
1728    if (!e1000e_intrmgr_delay_rx_causes(core, &n)) {
1729        trace_e1000e_rx_interrupt_set(n);
1730        e1000e_set_interrupt_cause(core, n);
1731    } else {
1732        trace_e1000e_rx_interrupt_delayed(n);
1733    }
1734
1735    return retval;
1736}
1737
1738static inline bool
1739e1000e_have_autoneg(E1000ECore *core)
1740{
1741    return core->phy[0][PHY_CTRL] & MII_CR_AUTO_NEG_EN;
1742}
1743
1744static void e1000e_update_flowctl_status(E1000ECore *core)
1745{
1746    if (e1000e_have_autoneg(core) &&
1747        core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE) {
1748        trace_e1000e_link_autoneg_flowctl(true);
1749        core->mac[CTRL] |= E1000_CTRL_TFCE | E1000_CTRL_RFCE;
1750    } else {
1751        trace_e1000e_link_autoneg_flowctl(false);
1752    }
1753}
1754
1755static inline void
1756e1000e_link_down(E1000ECore *core)
1757{
1758    e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1759    e1000e_update_flowctl_status(core);
1760}
1761
1762static inline void
1763e1000e_set_phy_ctrl(E1000ECore *core, int index, uint16_t val)
1764{
1765    /* bits 0-5 reserved; MII_CR_[RESTART_AUTO_NEG,RESET] are self clearing */
1766    core->phy[0][PHY_CTRL] = val & ~(0x3f |
1767                                     MII_CR_RESET |
1768                                     MII_CR_RESTART_AUTO_NEG);
1769
1770    if ((val & MII_CR_RESTART_AUTO_NEG) &&
1771        e1000e_have_autoneg(core)) {
1772        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1773    }
1774}
1775
1776static void
1777e1000e_set_phy_oem_bits(E1000ECore *core, int index, uint16_t val)
1778{
1779    core->phy[0][PHY_OEM_BITS] = val & ~BIT(10);
1780
1781    if (val & BIT(10)) {
1782        e1000x_restart_autoneg(core->mac, core->phy[0], core->autoneg_timer);
1783    }
1784}
1785
1786static void
1787e1000e_set_phy_page(E1000ECore *core, int index, uint16_t val)
1788{
1789    core->phy[0][PHY_PAGE] = val & PHY_PAGE_RW_MASK;
1790}
1791
1792void
1793e1000e_core_set_link_status(E1000ECore *core)
1794{
1795    NetClientState *nc = qemu_get_queue(core->owner_nic);
1796    uint32_t old_status = core->mac[STATUS];
1797
1798    trace_e1000e_link_status_changed(nc->link_down ? false : true);
1799
1800    if (nc->link_down) {
1801        e1000x_update_regs_on_link_down(core->mac, core->phy[0]);
1802    } else {
1803        if (e1000e_have_autoneg(core) &&
1804            !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
1805            e1000x_restart_autoneg(core->mac, core->phy[0],
1806                                   core->autoneg_timer);
1807        } else {
1808            e1000x_update_regs_on_link_up(core->mac, core->phy[0]);
1809            e1000e_start_recv(core);
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_iam_clear_eiame(core->mac[IAM], cause);
2011        core->mac[IAM] &= ~cause;
2012    }
2013
2014    trace_e1000e_irq_icr_clear_eiac(core->mac[ICR], core->mac[EIAC]);
2015
2016    effective_eiac = core->mac[EIAC] & cause;
2017
2018    core->mac[ICR] &= ~effective_eiac;
2019    core->msi_causes_pending &= ~effective_eiac;
2020
2021    if (!(core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2022        core->mac[IMS] &= ~effective_eiac;
2023    }
2024}
2025
2026static void
2027e1000e_msix_notify(E1000ECore *core, uint32_t causes)
2028{
2029    if (causes & E1000_ICR_RXQ0) {
2030        e1000e_msix_notify_one(core, E1000_ICR_RXQ0,
2031                               E1000_IVAR_RXQ0(core->mac[IVAR]));
2032    }
2033
2034    if (causes & E1000_ICR_RXQ1) {
2035        e1000e_msix_notify_one(core, E1000_ICR_RXQ1,
2036                               E1000_IVAR_RXQ1(core->mac[IVAR]));
2037    }
2038
2039    if (causes & E1000_ICR_TXQ0) {
2040        e1000e_msix_notify_one(core, E1000_ICR_TXQ0,
2041                               E1000_IVAR_TXQ0(core->mac[IVAR]));
2042    }
2043
2044    if (causes & E1000_ICR_TXQ1) {
2045        e1000e_msix_notify_one(core, E1000_ICR_TXQ1,
2046                               E1000_IVAR_TXQ1(core->mac[IVAR]));
2047    }
2048
2049    if (causes & E1000_ICR_OTHER) {
2050        e1000e_msix_notify_one(core, E1000_ICR_OTHER,
2051                               E1000_IVAR_OTHER(core->mac[IVAR]));
2052    }
2053}
2054
2055static void
2056e1000e_msix_clear_one(E1000ECore *core, uint32_t cause, uint32_t int_cfg)
2057{
2058    if (E1000_IVAR_ENTRY_VALID(int_cfg)) {
2059        uint32_t vec = E1000_IVAR_ENTRY_VEC(int_cfg);
2060        if (vec < E1000E_MSIX_VEC_NUM) {
2061            trace_e1000e_irq_msix_pending_clearing(cause, int_cfg, vec);
2062            msix_clr_pending(core->owner, vec);
2063        } else {
2064            trace_e1000e_wrn_msix_vec_wrong(cause, int_cfg);
2065        }
2066    } else {
2067        trace_e1000e_wrn_msix_invalid(cause, int_cfg);
2068    }
2069}
2070
2071static void
2072e1000e_msix_clear(E1000ECore *core, uint32_t causes)
2073{
2074    if (causes & E1000_ICR_RXQ0) {
2075        e1000e_msix_clear_one(core, E1000_ICR_RXQ0,
2076                              E1000_IVAR_RXQ0(core->mac[IVAR]));
2077    }
2078
2079    if (causes & E1000_ICR_RXQ1) {
2080        e1000e_msix_clear_one(core, E1000_ICR_RXQ1,
2081                              E1000_IVAR_RXQ1(core->mac[IVAR]));
2082    }
2083
2084    if (causes & E1000_ICR_TXQ0) {
2085        e1000e_msix_clear_one(core, E1000_ICR_TXQ0,
2086                              E1000_IVAR_TXQ0(core->mac[IVAR]));
2087    }
2088
2089    if (causes & E1000_ICR_TXQ1) {
2090        e1000e_msix_clear_one(core, E1000_ICR_TXQ1,
2091                              E1000_IVAR_TXQ1(core->mac[IVAR]));
2092    }
2093
2094    if (causes & E1000_ICR_OTHER) {
2095        e1000e_msix_clear_one(core, E1000_ICR_OTHER,
2096                              E1000_IVAR_OTHER(core->mac[IVAR]));
2097    }
2098}
2099
2100static inline void
2101e1000e_fix_icr_asserted(E1000ECore *core)
2102{
2103    core->mac[ICR] &= ~E1000_ICR_ASSERTED;
2104    if (core->mac[ICR]) {
2105        core->mac[ICR] |= E1000_ICR_ASSERTED;
2106    }
2107
2108    trace_e1000e_irq_fix_icr_asserted(core->mac[ICR]);
2109}
2110
2111static void
2112e1000e_send_msi(E1000ECore *core, bool msix)
2113{
2114    uint32_t causes = core->mac[ICR] & core->mac[IMS] & ~E1000_ICR_ASSERTED;
2115
2116    core->msi_causes_pending &= causes;
2117    causes ^= core->msi_causes_pending;
2118    if (causes == 0) {
2119        return;
2120    }
2121    core->msi_causes_pending |= causes;
2122
2123    if (msix) {
2124        e1000e_msix_notify(core, causes);
2125    } else {
2126        if (!e1000e_itr_should_postpone(core)) {
2127            trace_e1000e_irq_msi_notify(causes);
2128            msi_notify(core->owner, 0);
2129        }
2130    }
2131}
2132
2133static void
2134e1000e_update_interrupt_state(E1000ECore *core)
2135{
2136    bool interrupts_pending;
2137    bool is_msix = msix_enabled(core->owner);
2138
2139    /* Set ICR[OTHER] for MSI-X */
2140    if (is_msix) {
2141        if (core->mac[ICR] & E1000_ICR_OTHER_CAUSES) {
2142            core->mac[ICR] |= E1000_ICR_OTHER;
2143            trace_e1000e_irq_add_msi_other(core->mac[ICR]);
2144        }
2145    }
2146
2147    e1000e_fix_icr_asserted(core);
2148
2149    /*
2150     * Make sure ICR and ICS registers have the same value.
2151     * The spec says that the ICS register is write-only.  However in practice,
2152     * on real hardware ICS is readable, and for reads it has the same value as
2153     * ICR (except that ICS does not have the clear on read behaviour of ICR).
2154     *
2155     * The VxWorks PRO/1000 driver uses this behaviour.
2156     */
2157    core->mac[ICS] = core->mac[ICR];
2158
2159    interrupts_pending = (core->mac[IMS] & core->mac[ICR]) ? true : false;
2160    if (!interrupts_pending) {
2161        core->msi_causes_pending = 0;
2162    }
2163
2164    trace_e1000e_irq_pending_interrupts(core->mac[ICR] & core->mac[IMS],
2165                                        core->mac[ICR], core->mac[IMS]);
2166
2167    if (is_msix || msi_enabled(core->owner)) {
2168        if (interrupts_pending) {
2169            e1000e_send_msi(core, is_msix);
2170        }
2171    } else {
2172        if (interrupts_pending) {
2173            if (!e1000e_itr_should_postpone(core)) {
2174                e1000e_raise_legacy_irq(core);
2175            }
2176        } else {
2177            e1000e_lower_legacy_irq(core);
2178        }
2179    }
2180}
2181
2182static void
2183e1000e_set_interrupt_cause(E1000ECore *core, uint32_t val)
2184{
2185    trace_e1000e_irq_set_cause_entry(val, core->mac[ICR]);
2186
2187    val |= e1000e_intmgr_collect_delayed_causes(core);
2188    core->mac[ICR] |= val;
2189
2190    trace_e1000e_irq_set_cause_exit(val, core->mac[ICR]);
2191
2192    e1000e_update_interrupt_state(core);
2193}
2194
2195static inline void
2196e1000e_autoneg_timer(void *opaque)
2197{
2198    E1000ECore *core = opaque;
2199    if (!qemu_get_queue(core->owner_nic)->link_down) {
2200        e1000x_update_regs_on_autoneg_done(core->mac, core->phy[0]);
2201        e1000e_start_recv(core);
2202
2203        e1000e_update_flowctl_status(core);
2204        /* signal link status change to the guest */
2205        e1000e_set_interrupt_cause(core, E1000_ICR_LSC);
2206    }
2207}
2208
2209static inline uint16_t
2210e1000e_get_reg_index_with_offset(const uint16_t *mac_reg_access, hwaddr addr)
2211{
2212    uint16_t index = (addr & 0x1ffff) >> 2;
2213    return index + (mac_reg_access[index] & 0xfffe);
2214}
2215
2216static const char e1000e_phy_regcap[E1000E_PHY_PAGES][0x20] = {
2217    [0] = {
2218        [PHY_CTRL]          = PHY_ANYPAGE | PHY_RW,
2219        [PHY_STATUS]        = PHY_ANYPAGE | PHY_R,
2220        [PHY_ID1]           = PHY_ANYPAGE | PHY_R,
2221        [PHY_ID2]           = PHY_ANYPAGE | PHY_R,
2222        [PHY_AUTONEG_ADV]   = PHY_ANYPAGE | PHY_RW,
2223        [PHY_LP_ABILITY]    = PHY_ANYPAGE | PHY_R,
2224        [PHY_AUTONEG_EXP]   = PHY_ANYPAGE | PHY_R,
2225        [PHY_NEXT_PAGE_TX]  = PHY_ANYPAGE | PHY_RW,
2226        [PHY_LP_NEXT_PAGE]  = PHY_ANYPAGE | PHY_R,
2227        [PHY_1000T_CTRL]    = PHY_ANYPAGE | PHY_RW,
2228        [PHY_1000T_STATUS]  = PHY_ANYPAGE | PHY_R,
2229        [PHY_EXT_STATUS]    = PHY_ANYPAGE | PHY_R,
2230        [PHY_PAGE]          = PHY_ANYPAGE | PHY_RW,
2231
2232        [PHY_COPPER_CTRL1]      = PHY_RW,
2233        [PHY_COPPER_STAT1]      = PHY_R,
2234        [PHY_COPPER_CTRL3]      = PHY_RW,
2235        [PHY_RX_ERR_CNTR]       = PHY_R,
2236        [PHY_OEM_BITS]          = PHY_RW,
2237        [PHY_BIAS_1]            = PHY_RW,
2238        [PHY_BIAS_2]            = PHY_RW,
2239        [PHY_COPPER_INT_ENABLE] = PHY_RW,
2240        [PHY_COPPER_STAT2]      = PHY_R,
2241        [PHY_COPPER_CTRL2]      = PHY_RW
2242    },
2243    [2] = {
2244        [PHY_MAC_CTRL1]         = PHY_RW,
2245        [PHY_MAC_INT_ENABLE]    = PHY_RW,
2246        [PHY_MAC_STAT]          = PHY_R,
2247        [PHY_MAC_CTRL2]         = PHY_RW
2248    },
2249    [3] = {
2250        [PHY_LED_03_FUNC_CTRL1] = PHY_RW,
2251        [PHY_LED_03_POL_CTRL]   = PHY_RW,
2252        [PHY_LED_TIMER_CTRL]    = PHY_RW,
2253        [PHY_LED_45_CTRL]       = PHY_RW
2254    },
2255    [5] = {
2256        [PHY_1000T_SKEW]        = PHY_R,
2257        [PHY_1000T_SWAP]        = PHY_R
2258    },
2259    [6] = {
2260        [PHY_CRC_COUNTERS]      = PHY_R
2261    }
2262};
2263
2264static bool
2265e1000e_phy_reg_check_cap(E1000ECore *core, uint32_t addr,
2266                         char cap, uint8_t *page)
2267{
2268    *page =
2269        (e1000e_phy_regcap[0][addr] & PHY_ANYPAGE) ? 0
2270                                                    : core->phy[0][PHY_PAGE];
2271
2272    if (*page >= E1000E_PHY_PAGES) {
2273        return false;
2274    }
2275
2276    return e1000e_phy_regcap[*page][addr] & cap;
2277}
2278
2279static void
2280e1000e_phy_reg_write(E1000ECore *core, uint8_t page,
2281                     uint32_t addr, uint16_t data)
2282{
2283    assert(page < E1000E_PHY_PAGES);
2284    assert(addr < E1000E_PHY_PAGE_SIZE);
2285
2286    if (e1000e_phyreg_writeops[page][addr]) {
2287        e1000e_phyreg_writeops[page][addr](core, addr, data);
2288    } else {
2289        core->phy[page][addr] = data;
2290    }
2291}
2292
2293static void
2294e1000e_set_mdic(E1000ECore *core, int index, uint32_t val)
2295{
2296    uint32_t data = val & E1000_MDIC_DATA_MASK;
2297    uint32_t addr = ((val & E1000_MDIC_REG_MASK) >> E1000_MDIC_REG_SHIFT);
2298    uint8_t page;
2299
2300    if ((val & E1000_MDIC_PHY_MASK) >> E1000_MDIC_PHY_SHIFT != 1) { /* phy # */
2301        val = core->mac[MDIC] | E1000_MDIC_ERROR;
2302    } else if (val & E1000_MDIC_OP_READ) {
2303        if (!e1000e_phy_reg_check_cap(core, addr, PHY_R, &page)) {
2304            trace_e1000e_core_mdic_read_unhandled(page, addr);
2305            val |= E1000_MDIC_ERROR;
2306        } else {
2307            val = (val ^ data) | core->phy[page][addr];
2308            trace_e1000e_core_mdic_read(page, addr, val);
2309        }
2310    } else if (val & E1000_MDIC_OP_WRITE) {
2311        if (!e1000e_phy_reg_check_cap(core, addr, PHY_W, &page)) {
2312            trace_e1000e_core_mdic_write_unhandled(page, addr);
2313            val |= E1000_MDIC_ERROR;
2314        } else {
2315            trace_e1000e_core_mdic_write(page, addr, data);
2316            e1000e_phy_reg_write(core, page, addr, data);
2317        }
2318    }
2319    core->mac[MDIC] = val | E1000_MDIC_READY;
2320
2321    if (val & E1000_MDIC_INT_EN) {
2322        e1000e_set_interrupt_cause(core, E1000_ICR_MDAC);
2323    }
2324}
2325
2326static void
2327e1000e_set_rdt(E1000ECore *core, int index, uint32_t val)
2328{
2329    core->mac[index] = val & 0xffff;
2330    trace_e1000e_rx_set_rdt(e1000e_mq_queue_idx(RDT0, index), val);
2331    e1000e_start_recv(core);
2332}
2333
2334static void
2335e1000e_set_status(E1000ECore *core, int index, uint32_t val)
2336{
2337    if ((val & E1000_STATUS_PHYRA) == 0) {
2338        core->mac[index] &= ~E1000_STATUS_PHYRA;
2339    }
2340}
2341
2342static void
2343e1000e_set_ctrlext(E1000ECore *core, int index, uint32_t val)
2344{
2345    trace_e1000e_link_set_ext_params(!!(val & E1000_CTRL_EXT_ASDCHK),
2346                                     !!(val & E1000_CTRL_EXT_SPD_BYPS));
2347
2348    /* Zero self-clearing bits */
2349    val &= ~(E1000_CTRL_EXT_ASDCHK | E1000_CTRL_EXT_EE_RST);
2350    core->mac[CTRL_EXT] = val;
2351}
2352
2353static void
2354e1000e_set_pbaclr(E1000ECore *core, int index, uint32_t val)
2355{
2356    int i;
2357
2358    core->mac[PBACLR] = val & E1000_PBACLR_VALID_MASK;
2359
2360    if (!msix_enabled(core->owner)) {
2361        return;
2362    }
2363
2364    for (i = 0; i < E1000E_MSIX_VEC_NUM; i++) {
2365        if (core->mac[PBACLR] & BIT(i)) {
2366            msix_clr_pending(core->owner, i);
2367        }
2368    }
2369}
2370
2371static void
2372e1000e_set_fcrth(E1000ECore *core, int index, uint32_t val)
2373{
2374    core->mac[FCRTH] = val & 0xFFF8;
2375}
2376
2377static void
2378e1000e_set_fcrtl(E1000ECore *core, int index, uint32_t val)
2379{
2380    core->mac[FCRTL] = val & 0x8000FFF8;
2381}
2382
2383static inline void
2384e1000e_set_16bit(E1000ECore *core, int index, uint32_t val)
2385{
2386    core->mac[index] = val & 0xffff;
2387}
2388
2389static void
2390e1000e_set_12bit(E1000ECore *core, int index, uint32_t val)
2391{
2392    core->mac[index] = val & 0xfff;
2393}
2394
2395static void
2396e1000e_set_vet(E1000ECore *core, int index, uint32_t val)
2397{
2398    core->mac[VET] = val & 0xffff;
2399    trace_e1000e_vlan_vet(core->mac[VET]);
2400}
2401
2402static void
2403e1000e_set_dlen(E1000ECore *core, int index, uint32_t val)
2404{
2405    core->mac[index] = val & E1000_XDLEN_MASK;
2406}
2407
2408static void
2409e1000e_set_dbal(E1000ECore *core, int index, uint32_t val)
2410{
2411    core->mac[index] = val & E1000_XDBAL_MASK;
2412}
2413
2414static void
2415e1000e_set_tctl(E1000ECore *core, int index, uint32_t val)
2416{
2417    E1000E_TxRing txr;
2418    core->mac[index] = val;
2419
2420    if (core->mac[TARC0] & E1000_TARC_ENABLE) {
2421        e1000e_tx_ring_init(core, &txr, 0);
2422        e1000e_start_xmit(core, &txr);
2423    }
2424
2425    if (core->mac[TARC1] & E1000_TARC_ENABLE) {
2426        e1000e_tx_ring_init(core, &txr, 1);
2427        e1000e_start_xmit(core, &txr);
2428    }
2429}
2430
2431static void
2432e1000e_set_tdt(E1000ECore *core, int index, uint32_t val)
2433{
2434    E1000E_TxRing txr;
2435    int qidx = e1000e_mq_queue_idx(TDT, index);
2436    uint32_t tarc_reg = (qidx == 0) ? TARC0 : TARC1;
2437
2438    core->mac[index] = val & 0xffff;
2439
2440    if (core->mac[tarc_reg] & E1000_TARC_ENABLE) {
2441        e1000e_tx_ring_init(core, &txr, qidx);
2442        e1000e_start_xmit(core, &txr);
2443    }
2444}
2445
2446static void
2447e1000e_set_ics(E1000ECore *core, int index, uint32_t val)
2448{
2449    trace_e1000e_irq_write_ics(val);
2450    e1000e_set_interrupt_cause(core, val);
2451}
2452
2453static void
2454e1000e_set_icr(E1000ECore *core, int index, uint32_t val)
2455{
2456    uint32_t icr = 0;
2457    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2458        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2459        trace_e1000e_irq_icr_process_iame();
2460        e1000e_clear_ims_bits(core, core->mac[IAM]);
2461    }
2462
2463    icr = core->mac[ICR] & ~val;
2464    /* Windows driver expects that the "receive overrun" bit and other
2465     * ones to be cleared when the "Other" bit (#24) is cleared.
2466     */
2467    icr = (val & E1000_ICR_OTHER) ? (icr & ~E1000_ICR_OTHER_CAUSES) : icr;
2468    trace_e1000e_irq_icr_write(val, core->mac[ICR], icr);
2469    core->mac[ICR] = icr;
2470    e1000e_update_interrupt_state(core);
2471}
2472
2473static void
2474e1000e_set_imc(E1000ECore *core, int index, uint32_t val)
2475{
2476    trace_e1000e_irq_ims_clear_set_imc(val);
2477    e1000e_clear_ims_bits(core, val);
2478    e1000e_update_interrupt_state(core);
2479}
2480
2481static void
2482e1000e_set_ims(E1000ECore *core, int index, uint32_t val)
2483{
2484    static const uint32_t ims_ext_mask =
2485        E1000_IMS_RXQ0 | E1000_IMS_RXQ1 |
2486        E1000_IMS_TXQ0 | E1000_IMS_TXQ1 |
2487        E1000_IMS_OTHER;
2488
2489    static const uint32_t ims_valid_mask =
2490        E1000_IMS_TXDW      | E1000_IMS_TXQE    | E1000_IMS_LSC  |
2491        E1000_IMS_RXDMT0    | E1000_IMS_RXO     | E1000_IMS_RXT0 |
2492        E1000_IMS_MDAC      | E1000_IMS_TXD_LOW | E1000_IMS_SRPD |
2493        E1000_IMS_ACK       | E1000_IMS_MNG     | E1000_IMS_RXQ0 |
2494        E1000_IMS_RXQ1      | E1000_IMS_TXQ0    | E1000_IMS_TXQ1 |
2495        E1000_IMS_OTHER;
2496
2497    uint32_t valid_val = val & ims_valid_mask;
2498
2499    trace_e1000e_irq_set_ims(val, core->mac[IMS], core->mac[IMS] | valid_val);
2500    core->mac[IMS] |= valid_val;
2501
2502    if ((valid_val & ims_ext_mask) &&
2503        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_PBA_CLR) &&
2504        msix_enabled(core->owner)) {
2505        e1000e_msix_clear(core, valid_val);
2506    }
2507
2508    if ((valid_val == ims_valid_mask) &&
2509        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_INT_TIMERS_CLEAR_ENA)) {
2510        trace_e1000e_irq_fire_all_timers(val);
2511        e1000e_intrmgr_fire_all_timers(core);
2512    }
2513
2514    e1000e_update_interrupt_state(core);
2515}
2516
2517static void
2518e1000e_set_rdtr(E1000ECore *core, int index, uint32_t val)
2519{
2520    e1000e_set_16bit(core, index, val);
2521
2522    if ((val & E1000_RDTR_FPD) && (core->rdtr.running)) {
2523        trace_e1000e_irq_rdtr_fpd_running();
2524        e1000e_intrmgr_fire_delayed_interrupts(core);
2525    } else {
2526        trace_e1000e_irq_rdtr_fpd_not_running();
2527    }
2528}
2529
2530static void
2531e1000e_set_tidv(E1000ECore *core, int index, uint32_t val)
2532{
2533    e1000e_set_16bit(core, index, val);
2534
2535    if ((val & E1000_TIDV_FPD) && (core->tidv.running)) {
2536        trace_e1000e_irq_tidv_fpd_running();
2537        e1000e_intrmgr_fire_delayed_interrupts(core);
2538    } else {
2539        trace_e1000e_irq_tidv_fpd_not_running();
2540    }
2541}
2542
2543static uint32_t
2544e1000e_mac_readreg(E1000ECore *core, int index)
2545{
2546    return core->mac[index];
2547}
2548
2549static uint32_t
2550e1000e_mac_ics_read(E1000ECore *core, int index)
2551{
2552    trace_e1000e_irq_read_ics(core->mac[ICS]);
2553    return core->mac[ICS];
2554}
2555
2556static uint32_t
2557e1000e_mac_ims_read(E1000ECore *core, int index)
2558{
2559    trace_e1000e_irq_read_ims(core->mac[IMS]);
2560    return core->mac[IMS];
2561}
2562
2563#define E1000E_LOW_BITS_READ_FUNC(num)                      \
2564    static uint32_t                                         \
2565    e1000e_mac_low##num##_read(E1000ECore *core, int index) \
2566    {                                                       \
2567        return core->mac[index] & (BIT(num) - 1);           \
2568    }                                                       \
2569
2570#define E1000E_LOW_BITS_READ(num)                           \
2571    e1000e_mac_low##num##_read
2572
2573E1000E_LOW_BITS_READ_FUNC(4);
2574E1000E_LOW_BITS_READ_FUNC(6);
2575E1000E_LOW_BITS_READ_FUNC(11);
2576E1000E_LOW_BITS_READ_FUNC(13);
2577E1000E_LOW_BITS_READ_FUNC(16);
2578
2579static uint32_t
2580e1000e_mac_swsm_read(E1000ECore *core, int index)
2581{
2582    uint32_t val = core->mac[SWSM];
2583    core->mac[SWSM] = val | 1;
2584    return val;
2585}
2586
2587static uint32_t
2588e1000e_mac_itr_read(E1000ECore *core, int index)
2589{
2590    return core->itr_guest_value;
2591}
2592
2593static uint32_t
2594e1000e_mac_eitr_read(E1000ECore *core, int index)
2595{
2596    return core->eitr_guest_value[index - EITR];
2597}
2598
2599static uint32_t
2600e1000e_mac_icr_read(E1000ECore *core, int index)
2601{
2602    uint32_t ret = core->mac[ICR];
2603    trace_e1000e_irq_icr_read_entry(ret);
2604
2605    if (core->mac[IMS] == 0) {
2606        trace_e1000e_irq_icr_clear_zero_ims();
2607        core->mac[ICR] = 0;
2608    }
2609
2610    if ((core->mac[ICR] & E1000_ICR_ASSERTED) &&
2611        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_IAME)) {
2612        trace_e1000e_irq_icr_clear_iame();
2613        core->mac[ICR] = 0;
2614        trace_e1000e_irq_icr_process_iame();
2615        e1000e_clear_ims_bits(core, core->mac[IAM]);
2616    }
2617
2618    trace_e1000e_irq_icr_read_exit(core->mac[ICR]);
2619    e1000e_update_interrupt_state(core);
2620    return ret;
2621}
2622
2623static uint32_t
2624e1000e_mac_read_clr4(E1000ECore *core, int index)
2625{
2626    uint32_t ret = core->mac[index];
2627
2628    core->mac[index] = 0;
2629    return ret;
2630}
2631
2632static uint32_t
2633e1000e_mac_read_clr8(E1000ECore *core, int index)
2634{
2635    uint32_t ret = core->mac[index];
2636
2637    core->mac[index] = 0;
2638    core->mac[index - 1] = 0;
2639    return ret;
2640}
2641
2642static uint32_t
2643e1000e_get_ctrl(E1000ECore *core, int index)
2644{
2645    uint32_t val = core->mac[CTRL];
2646
2647    trace_e1000e_link_read_params(
2648        !!(val & E1000_CTRL_ASDE),
2649        (val & E1000_CTRL_SPD_SEL) >> E1000_CTRL_SPD_SHIFT,
2650        !!(val & E1000_CTRL_FRCSPD),
2651        !!(val & E1000_CTRL_FRCDPX),
2652        !!(val & E1000_CTRL_RFCE),
2653        !!(val & E1000_CTRL_TFCE));
2654
2655    return val;
2656}
2657
2658static uint32_t
2659e1000e_get_status(E1000ECore *core, int index)
2660{
2661    uint32_t res = core->mac[STATUS];
2662
2663    if (!(core->mac[CTRL] & E1000_CTRL_GIO_MASTER_DISABLE)) {
2664        res |= E1000_STATUS_GIO_MASTER_ENABLE;
2665    }
2666
2667    if (core->mac[CTRL] & E1000_CTRL_FRCDPX) {
2668        res |= (core->mac[CTRL] & E1000_CTRL_FD) ? E1000_STATUS_FD : 0;
2669    } else {
2670        res |= E1000_STATUS_FD;
2671    }
2672
2673    if ((core->mac[CTRL] & E1000_CTRL_FRCSPD) ||
2674        (core->mac[CTRL_EXT] & E1000_CTRL_EXT_SPD_BYPS)) {
2675        switch (core->mac[CTRL] & E1000_CTRL_SPD_SEL) {
2676        case E1000_CTRL_SPD_10:
2677            res |= E1000_STATUS_SPEED_10;
2678            break;
2679        case E1000_CTRL_SPD_100:
2680            res |= E1000_STATUS_SPEED_100;
2681            break;
2682        case E1000_CTRL_SPD_1000:
2683        default:
2684            res |= E1000_STATUS_SPEED_1000;
2685            break;
2686        }
2687    } else {
2688        res |= E1000_STATUS_SPEED_1000;
2689    }
2690
2691    trace_e1000e_link_status(
2692        !!(res & E1000_STATUS_LU),
2693        !!(res & E1000_STATUS_FD),
2694        (res & E1000_STATUS_SPEED_MASK) >> E1000_STATUS_SPEED_SHIFT,
2695        (res & E1000_STATUS_ASDV) >> E1000_STATUS_ASDV_SHIFT);
2696
2697    return res;
2698}
2699
2700static uint32_t
2701e1000e_get_tarc(E1000ECore *core, int index)
2702{
2703    return core->mac[index] & ((BIT(11) - 1) |
2704                                BIT(27)      |
2705                                BIT(28)      |
2706                                BIT(29)      |
2707                                BIT(30));
2708}
2709
2710static void
2711e1000e_mac_writereg(E1000ECore *core, int index, uint32_t val)
2712{
2713    core->mac[index] = val;
2714}
2715
2716static void
2717e1000e_mac_setmacaddr(E1000ECore *core, int index, uint32_t val)
2718{
2719    uint32_t macaddr[2];
2720
2721    core->mac[index] = val;
2722
2723    macaddr[0] = cpu_to_le32(core->mac[RA]);
2724    macaddr[1] = cpu_to_le32(core->mac[RA + 1]);
2725    qemu_format_nic_info_str(qemu_get_queue(core->owner_nic),
2726        (uint8_t *) macaddr);
2727
2728    trace_e1000e_mac_set_sw(MAC_ARG(macaddr));
2729}
2730
2731static void
2732e1000e_set_eecd(E1000ECore *core, int index, uint32_t val)
2733{
2734    static const uint32_t ro_bits = E1000_EECD_PRES          |
2735                                    E1000_EECD_AUTO_RD       |
2736                                    E1000_EECD_SIZE_EX_MASK;
2737
2738    core->mac[EECD] = (core->mac[EECD] & ro_bits) | (val & ~ro_bits);
2739}
2740
2741static void
2742e1000e_set_eerd(E1000ECore *core, int index, uint32_t val)
2743{
2744    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2745    uint32_t flags = 0;
2746    uint32_t data = 0;
2747
2748    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2749        data = core->eeprom[addr];
2750        flags = E1000_EERW_DONE;
2751    }
2752
2753    core->mac[EERD] = flags                           |
2754                      (addr << E1000_EERW_ADDR_SHIFT) |
2755                      (data << E1000_EERW_DATA_SHIFT);
2756}
2757
2758static void
2759e1000e_set_eewr(E1000ECore *core, int index, uint32_t val)
2760{
2761    uint32_t addr = (val >> E1000_EERW_ADDR_SHIFT) & E1000_EERW_ADDR_MASK;
2762    uint32_t data = (val >> E1000_EERW_DATA_SHIFT) & E1000_EERW_DATA_MASK;
2763    uint32_t flags = 0;
2764
2765    if ((addr < E1000E_EEPROM_SIZE) && (val & E1000_EERW_START)) {
2766        core->eeprom[addr] = data;
2767        flags = E1000_EERW_DONE;
2768    }
2769
2770    core->mac[EERD] = flags                           |
2771                      (addr << E1000_EERW_ADDR_SHIFT) |
2772                      (data << E1000_EERW_DATA_SHIFT);
2773}
2774
2775static void
2776e1000e_set_rxdctl(E1000ECore *core, int index, uint32_t val)
2777{
2778    core->mac[RXDCTL] = core->mac[RXDCTL1] = val;
2779}
2780
2781static void
2782e1000e_set_itr(E1000ECore *core, int index, uint32_t val)
2783{
2784    uint32_t interval = val & 0xffff;
2785
2786    trace_e1000e_irq_itr_set(val);
2787
2788    core->itr_guest_value = interval;
2789    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2790}
2791
2792static void
2793e1000e_set_eitr(E1000ECore *core, int index, uint32_t val)
2794{
2795    uint32_t interval = val & 0xffff;
2796    uint32_t eitr_num = index - EITR;
2797
2798    trace_e1000e_irq_eitr_set(eitr_num, val);
2799
2800    core->eitr_guest_value[eitr_num] = interval;
2801    core->mac[index] = MAX(interval, E1000E_MIN_XITR);
2802}
2803
2804static void
2805e1000e_set_psrctl(E1000ECore *core, int index, uint32_t val)
2806{
2807    if (core->mac[RCTL] & E1000_RCTL_DTYP_MASK) {
2808
2809        if ((val & E1000_PSRCTL_BSIZE0_MASK) == 0) {
2810            qemu_log_mask(LOG_GUEST_ERROR,
2811                          "e1000e: PSRCTL.BSIZE0 cannot be zero");
2812            return;
2813        }
2814
2815        if ((val & E1000_PSRCTL_BSIZE1_MASK) == 0) {
2816            qemu_log_mask(LOG_GUEST_ERROR,
2817                          "e1000e: PSRCTL.BSIZE1 cannot be zero");
2818            return;
2819        }
2820    }
2821
2822    core->mac[PSRCTL] = val;
2823}
2824
2825static void
2826e1000e_update_rx_offloads(E1000ECore *core)
2827{
2828    int cso_state = e1000e_rx_l4_cso_enabled(core);
2829
2830    trace_e1000e_rx_set_cso(cso_state);
2831
2832    if (core->has_vnet) {
2833        qemu_set_offload(qemu_get_queue(core->owner_nic)->peer,
2834                         cso_state, 0, 0, 0, 0);
2835    }
2836}
2837
2838static void
2839e1000e_set_rxcsum(E1000ECore *core, int index, uint32_t val)
2840{
2841    core->mac[RXCSUM] = val;
2842    e1000e_update_rx_offloads(core);
2843}
2844
2845static void
2846e1000e_set_gcr(E1000ECore *core, int index, uint32_t val)
2847{
2848    uint32_t ro_bits = core->mac[GCR] & E1000_GCR_RO_BITS;
2849    core->mac[GCR] = (val & ~E1000_GCR_RO_BITS) | ro_bits;
2850}
2851
2852#define e1000e_getreg(x)    [x] = e1000e_mac_readreg
2853typedef uint32_t (*readops)(E1000ECore *, int);
2854static const readops e1000e_macreg_readops[] = {
2855    e1000e_getreg(PBA),
2856    e1000e_getreg(WUFC),
2857    e1000e_getreg(MANC),
2858    e1000e_getreg(TOTL),
2859    e1000e_getreg(RDT0),
2860    e1000e_getreg(RDBAH0),
2861    e1000e_getreg(TDBAL1),
2862    e1000e_getreg(RDLEN0),
2863    e1000e_getreg(RDH1),
2864    e1000e_getreg(LATECOL),
2865    e1000e_getreg(SEQEC),
2866    e1000e_getreg(XONTXC),
2867    e1000e_getreg(WUS),
2868    e1000e_getreg(GORCL),
2869    e1000e_getreg(MGTPRC),
2870    e1000e_getreg(EERD),
2871    e1000e_getreg(EIAC),
2872    e1000e_getreg(PSRCTL),
2873    e1000e_getreg(MANC2H),
2874    e1000e_getreg(RXCSUM),
2875    e1000e_getreg(GSCL_3),
2876    e1000e_getreg(GSCN_2),
2877    e1000e_getreg(RSRPD),
2878    e1000e_getreg(RDBAL1),
2879    e1000e_getreg(FCAH),
2880    e1000e_getreg(FCRTH),
2881    e1000e_getreg(FLOP),
2882    e1000e_getreg(FLASHT),
2883    e1000e_getreg(RXSTMPH),
2884    e1000e_getreg(TXSTMPL),
2885    e1000e_getreg(TIMADJL),
2886    e1000e_getreg(TXDCTL),
2887    e1000e_getreg(RDH0),
2888    e1000e_getreg(TDT1),
2889    e1000e_getreg(TNCRS),
2890    e1000e_getreg(RJC),
2891    e1000e_getreg(IAM),
2892    e1000e_getreg(GSCL_2),
2893    e1000e_getreg(RDBAH1),
2894    e1000e_getreg(FLSWDATA),
2895    e1000e_getreg(RXSATRH),
2896    e1000e_getreg(TIPG),
2897    e1000e_getreg(FLMNGCTL),
2898    e1000e_getreg(FLMNGCNT),
2899    e1000e_getreg(TSYNCTXCTL),
2900    e1000e_getreg(EXTCNF_SIZE),
2901    e1000e_getreg(EXTCNF_CTRL),
2902    e1000e_getreg(EEMNGDATA),
2903    e1000e_getreg(CTRL_EXT),
2904    e1000e_getreg(SYSTIMH),
2905    e1000e_getreg(EEMNGCTL),
2906    e1000e_getreg(FLMNGDATA),
2907    e1000e_getreg(TSYNCRXCTL),
2908    e1000e_getreg(TDH),
2909    e1000e_getreg(LEDCTL),
2910    e1000e_getreg(TCTL),
2911    e1000e_getreg(TDBAL),
2912    e1000e_getreg(TDLEN),
2913    e1000e_getreg(TDH1),
2914    e1000e_getreg(RADV),
2915    e1000e_getreg(ECOL),
2916    e1000e_getreg(DC),
2917    e1000e_getreg(RLEC),
2918    e1000e_getreg(XOFFTXC),
2919    e1000e_getreg(RFC),
2920    e1000e_getreg(RNBC),
2921    e1000e_getreg(MGTPTC),
2922    e1000e_getreg(TIMINCA),
2923    e1000e_getreg(RXCFGL),
2924    e1000e_getreg(MFUTP01),
2925    e1000e_getreg(FACTPS),
2926    e1000e_getreg(GSCL_1),
2927    e1000e_getreg(GSCN_0),
2928    e1000e_getreg(GCR2),
2929    e1000e_getreg(RDT1),
2930    e1000e_getreg(PBACLR),
2931    e1000e_getreg(FCTTV),
2932    e1000e_getreg(EEWR),
2933    e1000e_getreg(FLSWCTL),
2934    e1000e_getreg(RXDCTL1),
2935    e1000e_getreg(RXSATRL),
2936    e1000e_getreg(SYSTIML),
2937    e1000e_getreg(RXUDP),
2938    e1000e_getreg(TORL),
2939    e1000e_getreg(TDLEN1),
2940    e1000e_getreg(MCC),
2941    e1000e_getreg(WUC),
2942    e1000e_getreg(EECD),
2943    e1000e_getreg(MFUTP23),
2944    e1000e_getreg(RAID),
2945    e1000e_getreg(FCRTV),
2946    e1000e_getreg(TXDCTL1),
2947    e1000e_getreg(RCTL),
2948    e1000e_getreg(TDT),
2949    e1000e_getreg(MDIC),
2950    e1000e_getreg(FCRUC),
2951    e1000e_getreg(VET),
2952    e1000e_getreg(RDBAL0),
2953    e1000e_getreg(TDBAH1),
2954    e1000e_getreg(RDTR),
2955    e1000e_getreg(SCC),
2956    e1000e_getreg(COLC),
2957    e1000e_getreg(CEXTERR),
2958    e1000e_getreg(XOFFRXC),
2959    e1000e_getreg(IPAV),
2960    e1000e_getreg(GOTCL),
2961    e1000e_getreg(MGTPDC),
2962    e1000e_getreg(GCR),
2963    e1000e_getreg(IVAR),
2964    e1000e_getreg(POEMB),
2965    e1000e_getreg(MFVAL),
2966    e1000e_getreg(FUNCTAG),
2967    e1000e_getreg(GSCL_4),
2968    e1000e_getreg(GSCN_3),
2969    e1000e_getreg(MRQC),
2970    e1000e_getreg(RDLEN1),
2971    e1000e_getreg(FCT),
2972    e1000e_getreg(FLA),
2973    e1000e_getreg(FLOL),
2974    e1000e_getreg(RXDCTL),
2975    e1000e_getreg(RXSTMPL),
2976    e1000e_getreg(TXSTMPH),
2977    e1000e_getreg(TIMADJH),
2978    e1000e_getreg(FCRTL),
2979    e1000e_getreg(TDBAH),
2980    e1000e_getreg(TADV),
2981    e1000e_getreg(XONRXC),
2982    e1000e_getreg(TSCTFC),
2983    e1000e_getreg(RFCTL),
2984    e1000e_getreg(GSCN_1),
2985    e1000e_getreg(FCAL),
2986    e1000e_getreg(FLSWCNT),
2987
2988    [TOTH]    = e1000e_mac_read_clr8,
2989    [GOTCH]   = e1000e_mac_read_clr8,
2990    [PRC64]   = e1000e_mac_read_clr4,
2991    [PRC255]  = e1000e_mac_read_clr4,
2992    [PRC1023] = e1000e_mac_read_clr4,
2993    [PTC64]   = e1000e_mac_read_clr4,
2994    [PTC255]  = e1000e_mac_read_clr4,
2995    [PTC1023] = e1000e_mac_read_clr4,
2996    [GPRC]    = e1000e_mac_read_clr4,
2997    [TPT]     = e1000e_mac_read_clr4,
2998    [RUC]     = e1000e_mac_read_clr4,
2999    [BPRC]    = e1000e_mac_read_clr4,
3000    [MPTC]    = e1000e_mac_read_clr4,
3001    [IAC]     = e1000e_mac_read_clr4,
3002    [ICR]     = e1000e_mac_icr_read,
3003    [RDFH]    = E1000E_LOW_BITS_READ(13),
3004    [RDFHS]   = E1000E_LOW_BITS_READ(13),
3005    [RDFPC]   = E1000E_LOW_BITS_READ(13),
3006    [TDFH]    = E1000E_LOW_BITS_READ(13),
3007    [TDFHS]   = E1000E_LOW_BITS_READ(13),
3008    [STATUS]  = e1000e_get_status,
3009    [TARC0]   = e1000e_get_tarc,
3010    [PBS]     = E1000E_LOW_BITS_READ(6),
3011    [ICS]     = e1000e_mac_ics_read,
3012    [AIT]     = E1000E_LOW_BITS_READ(16),
3013    [TORH]    = e1000e_mac_read_clr8,
3014    [GORCH]   = e1000e_mac_read_clr8,
3015    [PRC127]  = e1000e_mac_read_clr4,
3016    [PRC511]  = e1000e_mac_read_clr4,
3017    [PRC1522] = e1000e_mac_read_clr4,
3018    [PTC127]  = e1000e_mac_read_clr4,
3019    [PTC511]  = e1000e_mac_read_clr4,
3020    [PTC1522] = e1000e_mac_read_clr4,
3021    [GPTC]    = e1000e_mac_read_clr4,
3022    [TPR]     = e1000e_mac_read_clr4,
3023    [ROC]     = e1000e_mac_read_clr4,
3024    [MPRC]    = e1000e_mac_read_clr4,
3025    [BPTC]    = e1000e_mac_read_clr4,
3026    [TSCTC]   = e1000e_mac_read_clr4,
3027    [ITR]     = e1000e_mac_itr_read,
3028    [RDFT]    = E1000E_LOW_BITS_READ(13),
3029    [RDFTS]   = E1000E_LOW_BITS_READ(13),
3030    [TDFPC]   = E1000E_LOW_BITS_READ(13),
3031    [TDFT]    = E1000E_LOW_BITS_READ(13),
3032    [TDFTS]   = E1000E_LOW_BITS_READ(13),
3033    [CTRL]    = e1000e_get_ctrl,
3034    [TARC1]   = e1000e_get_tarc,
3035    [SWSM]    = e1000e_mac_swsm_read,
3036    [IMS]     = e1000e_mac_ims_read,
3037
3038    [CRCERRS ... MPC]      = e1000e_mac_readreg,
3039    [IP6AT ... IP6AT + 3]  = e1000e_mac_readreg,
3040    [IP4AT ... IP4AT + 6]  = e1000e_mac_readreg,
3041    [RA ... RA + 31]       = e1000e_mac_readreg,
3042    [WUPM ... WUPM + 31]   = e1000e_mac_readreg,
3043    [MTA ... MTA + 127]    = e1000e_mac_readreg,
3044    [VFTA ... VFTA + 127]  = e1000e_mac_readreg,
3045    [FFMT ... FFMT + 254]  = E1000E_LOW_BITS_READ(4),
3046    [FFVT ... FFVT + 254]  = e1000e_mac_readreg,
3047    [MDEF ... MDEF + 7]    = e1000e_mac_readreg,
3048    [FFLT ... FFLT + 10]   = E1000E_LOW_BITS_READ(11),
3049    [FTFT ... FTFT + 254]  = e1000e_mac_readreg,
3050    [PBM ... PBM + 10239]  = e1000e_mac_readreg,
3051    [RETA ... RETA + 31]   = e1000e_mac_readreg,
3052    [RSSRK ... RSSRK + 31] = e1000e_mac_readreg,
3053    [MAVTV0 ... MAVTV3]    = e1000e_mac_readreg,
3054    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_mac_eitr_read
3055};
3056enum { E1000E_NREADOPS = ARRAY_SIZE(e1000e_macreg_readops) };
3057
3058#define e1000e_putreg(x)    [x] = e1000e_mac_writereg
3059typedef void (*writeops)(E1000ECore *, int, uint32_t);
3060static const writeops e1000e_macreg_writeops[] = {
3061    e1000e_putreg(PBA),
3062    e1000e_putreg(SWSM),
3063    e1000e_putreg(WUFC),
3064    e1000e_putreg(RDBAH1),
3065    e1000e_putreg(TDBAH),
3066    e1000e_putreg(TXDCTL),
3067    e1000e_putreg(RDBAH0),
3068    e1000e_putreg(LEDCTL),
3069    e1000e_putreg(FCAL),
3070    e1000e_putreg(FCRUC),
3071    e1000e_putreg(AIT),
3072    e1000e_putreg(TDFH),
3073    e1000e_putreg(TDFT),
3074    e1000e_putreg(TDFHS),
3075    e1000e_putreg(TDFTS),
3076    e1000e_putreg(TDFPC),
3077    e1000e_putreg(WUC),
3078    e1000e_putreg(WUS),
3079    e1000e_putreg(RDFH),
3080    e1000e_putreg(RDFT),
3081    e1000e_putreg(RDFHS),
3082    e1000e_putreg(RDFTS),
3083    e1000e_putreg(RDFPC),
3084    e1000e_putreg(IPAV),
3085    e1000e_putreg(TDBAH1),
3086    e1000e_putreg(TIMINCA),
3087    e1000e_putreg(IAM),
3088    e1000e_putreg(EIAC),
3089    e1000e_putreg(IVAR),
3090    e1000e_putreg(TARC0),
3091    e1000e_putreg(TARC1),
3092    e1000e_putreg(FLSWDATA),
3093    e1000e_putreg(POEMB),
3094    e1000e_putreg(PBS),
3095    e1000e_putreg(MFUTP01),
3096    e1000e_putreg(MFUTP23),
3097    e1000e_putreg(MANC),
3098    e1000e_putreg(MANC2H),
3099    e1000e_putreg(MFVAL),
3100    e1000e_putreg(EXTCNF_CTRL),
3101    e1000e_putreg(FACTPS),
3102    e1000e_putreg(FUNCTAG),
3103    e1000e_putreg(GSCL_1),
3104    e1000e_putreg(GSCL_2),
3105    e1000e_putreg(GSCL_3),
3106    e1000e_putreg(GSCL_4),
3107    e1000e_putreg(GSCN_0),
3108    e1000e_putreg(GSCN_1),
3109    e1000e_putreg(GSCN_2),
3110    e1000e_putreg(GSCN_3),
3111    e1000e_putreg(GCR2),
3112    e1000e_putreg(MRQC),
3113    e1000e_putreg(FLOP),
3114    e1000e_putreg(FLOL),
3115    e1000e_putreg(FLSWCTL),
3116    e1000e_putreg(FLSWCNT),
3117    e1000e_putreg(FLA),
3118    e1000e_putreg(RXDCTL1),
3119    e1000e_putreg(TXDCTL1),
3120    e1000e_putreg(TIPG),
3121    e1000e_putreg(RXSTMPH),
3122    e1000e_putreg(RXSTMPL),
3123    e1000e_putreg(RXSATRL),
3124    e1000e_putreg(RXSATRH),
3125    e1000e_putreg(TXSTMPL),
3126    e1000e_putreg(TXSTMPH),
3127    e1000e_putreg(SYSTIML),
3128    e1000e_putreg(SYSTIMH),
3129    e1000e_putreg(TIMADJL),
3130    e1000e_putreg(TIMADJH),
3131    e1000e_putreg(RXUDP),
3132    e1000e_putreg(RXCFGL),
3133    e1000e_putreg(TSYNCRXCTL),
3134    e1000e_putreg(TSYNCTXCTL),
3135    e1000e_putreg(EXTCNF_SIZE),
3136    e1000e_putreg(EEMNGCTL),
3137    e1000e_putreg(RA),
3138
3139    [TDH1]     = e1000e_set_16bit,
3140    [TDT1]     = e1000e_set_tdt,
3141    [TCTL]     = e1000e_set_tctl,
3142    [TDT]      = e1000e_set_tdt,
3143    [MDIC]     = e1000e_set_mdic,
3144    [ICS]      = e1000e_set_ics,
3145    [TDH]      = e1000e_set_16bit,
3146    [RDH0]     = e1000e_set_16bit,
3147    [RDT0]     = e1000e_set_rdt,
3148    [IMC]      = e1000e_set_imc,
3149    [IMS]      = e1000e_set_ims,
3150    [ICR]      = e1000e_set_icr,
3151    [EECD]     = e1000e_set_eecd,
3152    [RCTL]     = e1000e_set_rx_control,
3153    [CTRL]     = e1000e_set_ctrl,
3154    [RDTR]     = e1000e_set_rdtr,
3155    [RADV]     = e1000e_set_16bit,
3156    [TADV]     = e1000e_set_16bit,
3157    [ITR]      = e1000e_set_itr,
3158    [EERD]     = e1000e_set_eerd,
3159    [GCR]      = e1000e_set_gcr,
3160    [PSRCTL]   = e1000e_set_psrctl,
3161    [RXCSUM]   = e1000e_set_rxcsum,
3162    [RAID]     = e1000e_set_16bit,
3163    [RSRPD]    = e1000e_set_12bit,
3164    [TIDV]     = e1000e_set_tidv,
3165    [TDLEN1]   = e1000e_set_dlen,
3166    [TDLEN]    = e1000e_set_dlen,
3167    [RDLEN0]   = e1000e_set_dlen,
3168    [RDLEN1]   = e1000e_set_dlen,
3169    [TDBAL]    = e1000e_set_dbal,
3170    [TDBAL1]   = e1000e_set_dbal,
3171    [RDBAL0]   = e1000e_set_dbal,
3172    [RDBAL1]   = e1000e_set_dbal,
3173    [RDH1]     = e1000e_set_16bit,
3174    [RDT1]     = e1000e_set_rdt,
3175    [STATUS]   = e1000e_set_status,
3176    [PBACLR]   = e1000e_set_pbaclr,
3177    [CTRL_EXT] = e1000e_set_ctrlext,
3178    [FCAH]     = e1000e_set_16bit,
3179    [FCT]      = e1000e_set_16bit,
3180    [FCTTV]    = e1000e_set_16bit,
3181    [FCRTV]    = e1000e_set_16bit,
3182    [FCRTH]    = e1000e_set_fcrth,
3183    [FCRTL]    = e1000e_set_fcrtl,
3184    [VET]      = e1000e_set_vet,
3185    [RXDCTL]   = e1000e_set_rxdctl,
3186    [FLASHT]   = e1000e_set_16bit,
3187    [EEWR]     = e1000e_set_eewr,
3188    [CTRL_DUP] = e1000e_set_ctrl,
3189    [RFCTL]    = e1000e_set_rfctl,
3190    [RA + 1]   = e1000e_mac_setmacaddr,
3191
3192    [IP6AT ... IP6AT + 3]    = e1000e_mac_writereg,
3193    [IP4AT ... IP4AT + 6]    = e1000e_mac_writereg,
3194    [RA + 2 ... RA + 31]     = e1000e_mac_writereg,
3195    [WUPM ... WUPM + 31]     = e1000e_mac_writereg,
3196    [MTA ... MTA + 127]      = e1000e_mac_writereg,
3197    [VFTA ... VFTA + 127]    = e1000e_mac_writereg,
3198    [FFMT ... FFMT + 254]    = e1000e_mac_writereg,
3199    [FFVT ... FFVT + 254]    = e1000e_mac_writereg,
3200    [PBM ... PBM + 10239]    = e1000e_mac_writereg,
3201    [MDEF ... MDEF + 7]      = e1000e_mac_writereg,
3202    [FFLT ... FFLT + 10]     = e1000e_mac_writereg,
3203    [FTFT ... FTFT + 254]    = e1000e_mac_writereg,
3204    [RETA ... RETA + 31]     = e1000e_mac_writereg,
3205    [RSSRK ... RSSRK + 31]   = e1000e_mac_writereg,
3206    [MAVTV0 ... MAVTV3]      = e1000e_mac_writereg,
3207    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = e1000e_set_eitr
3208};
3209enum { E1000E_NWRITEOPS = ARRAY_SIZE(e1000e_macreg_writeops) };
3210
3211enum { MAC_ACCESS_PARTIAL = 1 };
3212
3213/* The array below combines alias offsets of the index values for the
3214 * MAC registers that have aliases, with the indication of not fully
3215 * implemented registers (lowest bit). This combination is possible
3216 * because all of the offsets are even. */
3217static const uint16_t mac_reg_access[E1000E_MAC_SIZE] = {
3218    /* Alias index offsets */
3219    [FCRTL_A] = 0x07fe, [FCRTH_A] = 0x0802,
3220    [RDH0_A]  = 0x09bc, [RDT0_A]  = 0x09bc, [RDTR_A] = 0x09c6,
3221    [RDFH_A]  = 0xe904, [RDFT_A]  = 0xe904,
3222    [TDH_A]   = 0x0cf8, [TDT_A]   = 0x0cf8, [TIDV_A] = 0x0cf8,
3223    [TDFH_A]  = 0xed00, [TDFT_A]  = 0xed00,
3224    [RA_A ... RA_A + 31]      = 0x14f0,
3225    [VFTA_A ... VFTA_A + 127] = 0x1400,
3226    [RDBAL0_A ... RDLEN0_A] = 0x09bc,
3227    [TDBAL_A ... TDLEN_A]   = 0x0cf8,
3228    /* Access options */
3229    [RDFH]  = MAC_ACCESS_PARTIAL,    [RDFT]  = MAC_ACCESS_PARTIAL,
3230    [RDFHS] = MAC_ACCESS_PARTIAL,    [RDFTS] = MAC_ACCESS_PARTIAL,
3231    [RDFPC] = MAC_ACCESS_PARTIAL,
3232    [TDFH]  = MAC_ACCESS_PARTIAL,    [TDFT]  = MAC_ACCESS_PARTIAL,
3233    [TDFHS] = MAC_ACCESS_PARTIAL,    [TDFTS] = MAC_ACCESS_PARTIAL,
3234    [TDFPC] = MAC_ACCESS_PARTIAL,    [EECD]  = MAC_ACCESS_PARTIAL,
3235    [PBM]   = MAC_ACCESS_PARTIAL,    [FLA]   = MAC_ACCESS_PARTIAL,
3236    [FCAL]  = MAC_ACCESS_PARTIAL,    [FCAH]  = MAC_ACCESS_PARTIAL,
3237    [FCT]   = MAC_ACCESS_PARTIAL,    [FCTTV] = MAC_ACCESS_PARTIAL,
3238    [FCRTV] = MAC_ACCESS_PARTIAL,    [FCRTL] = MAC_ACCESS_PARTIAL,
3239    [FCRTH] = MAC_ACCESS_PARTIAL,    [TXDCTL] = MAC_ACCESS_PARTIAL,
3240    [TXDCTL1] = MAC_ACCESS_PARTIAL,
3241    [MAVTV0 ... MAVTV3] = MAC_ACCESS_PARTIAL
3242};
3243
3244void
3245e1000e_core_write(E1000ECore *core, hwaddr addr, uint64_t val, unsigned size)
3246{
3247    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3248
3249    if (index < E1000E_NWRITEOPS && e1000e_macreg_writeops[index]) {
3250        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3251            trace_e1000e_wrn_regs_write_trivial(index << 2);
3252        }
3253        trace_e1000e_core_write(index << 2, size, val);
3254        e1000e_macreg_writeops[index](core, index, val);
3255    } else if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3256        trace_e1000e_wrn_regs_write_ro(index << 2, size, val);
3257    } else {
3258        trace_e1000e_wrn_regs_write_unknown(index << 2, size, val);
3259    }
3260}
3261
3262uint64_t
3263e1000e_core_read(E1000ECore *core, hwaddr addr, unsigned size)
3264{
3265    uint64_t val;
3266    uint16_t index = e1000e_get_reg_index_with_offset(mac_reg_access, addr);
3267
3268    if (index < E1000E_NREADOPS && e1000e_macreg_readops[index]) {
3269        if (mac_reg_access[index] & MAC_ACCESS_PARTIAL) {
3270            trace_e1000e_wrn_regs_read_trivial(index << 2);
3271        }
3272        val = e1000e_macreg_readops[index](core, index);
3273        trace_e1000e_core_read(index << 2, size, val);
3274        return val;
3275    } else {
3276        trace_e1000e_wrn_regs_read_unknown(index << 2, size);
3277    }
3278    return 0;
3279}
3280
3281static inline void
3282e1000e_autoneg_pause(E1000ECore *core)
3283{
3284    timer_del(core->autoneg_timer);
3285}
3286
3287static void
3288e1000e_autoneg_resume(E1000ECore *core)
3289{
3290    if (e1000e_have_autoneg(core) &&
3291        !(core->phy[0][PHY_STATUS] & MII_SR_AUTONEG_COMPLETE)) {
3292        qemu_get_queue(core->owner_nic)->link_down = false;
3293        timer_mod(core->autoneg_timer,
3294                  qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 500);
3295    }
3296}
3297
3298static void
3299e1000e_vm_state_change(void *opaque, bool running, RunState state)
3300{
3301    E1000ECore *core = opaque;
3302
3303    if (running) {
3304        trace_e1000e_vm_state_running();
3305        e1000e_intrmgr_resume(core);
3306        e1000e_autoneg_resume(core);
3307    } else {
3308        trace_e1000e_vm_state_stopped();
3309        e1000e_autoneg_pause(core);
3310        e1000e_intrmgr_pause(core);
3311    }
3312}
3313
3314void
3315e1000e_core_pci_realize(E1000ECore     *core,
3316                        const uint16_t *eeprom_templ,
3317                        uint32_t        eeprom_size,
3318                        const uint8_t  *macaddr)
3319{
3320    int i;
3321
3322    core->autoneg_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL,
3323                                       e1000e_autoneg_timer, core);
3324    e1000e_intrmgr_pci_realize(core);
3325
3326    core->vmstate =
3327        qemu_add_vm_change_state_handler(e1000e_vm_state_change, core);
3328
3329    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3330        net_tx_pkt_init(&core->tx[i].tx_pkt, core->owner,
3331                        E1000E_MAX_TX_FRAGS, core->has_vnet);
3332    }
3333
3334    net_rx_pkt_init(&core->rx_pkt, core->has_vnet);
3335
3336    e1000x_core_prepare_eeprom(core->eeprom,
3337                               eeprom_templ,
3338                               eeprom_size,
3339                               PCI_DEVICE_GET_CLASS(core->owner)->device_id,
3340                               macaddr);
3341    e1000e_update_rx_offloads(core);
3342}
3343
3344void
3345e1000e_core_pci_uninit(E1000ECore *core)
3346{
3347    int i;
3348
3349    timer_free(core->autoneg_timer);
3350
3351    e1000e_intrmgr_pci_unint(core);
3352
3353    qemu_del_vm_change_state_handler(core->vmstate);
3354
3355    for (i = 0; i < E1000E_NUM_QUEUES; i++) {
3356        net_tx_pkt_reset(core->tx[i].tx_pkt);
3357        net_tx_pkt_uninit(core->tx[i].tx_pkt);
3358    }
3359
3360    net_rx_pkt_uninit(core->rx_pkt);
3361}
3362
3363static const uint16_t
3364e1000e_phy_reg_init[E1000E_PHY_PAGES][E1000E_PHY_PAGE_SIZE] = {
3365    [0] = {
3366        [PHY_CTRL] =   MII_CR_SPEED_SELECT_MSB  |
3367                       MII_CR_FULL_DUPLEX       |
3368                       MII_CR_AUTO_NEG_EN,
3369
3370        [PHY_STATUS] = MII_SR_EXTENDED_CAPS     |
3371                       MII_SR_LINK_STATUS       |
3372                       MII_SR_AUTONEG_CAPS      |
3373                       MII_SR_PREAMBLE_SUPPRESS |
3374                       MII_SR_EXTENDED_STATUS   |
3375                       MII_SR_10T_HD_CAPS       |
3376                       MII_SR_10T_FD_CAPS       |
3377                       MII_SR_100X_HD_CAPS      |
3378                       MII_SR_100X_FD_CAPS,
3379
3380        [PHY_ID1]               = 0x141,
3381        [PHY_ID2]               = E1000_PHY_ID2_82574x,
3382        [PHY_AUTONEG_ADV]       = 0xde1,
3383        [PHY_LP_ABILITY]        = 0x7e0,
3384        [PHY_AUTONEG_EXP]       = BIT(2),
3385        [PHY_NEXT_PAGE_TX]      = BIT(0) | BIT(13),
3386        [PHY_1000T_CTRL]        = BIT(8) | BIT(9) | BIT(10) | BIT(11),
3387        [PHY_1000T_STATUS]      = 0x3c00,
3388        [PHY_EXT_STATUS]        = BIT(12) | BIT(13),
3389
3390        [PHY_COPPER_CTRL1]      = BIT(5) | BIT(6) | BIT(8) | BIT(9) |
3391                                  BIT(12) | BIT(13),
3392        [PHY_COPPER_STAT1]      = BIT(3) | BIT(10) | BIT(11) | BIT(13) | BIT(15)
3393    },
3394    [2] = {
3395        [PHY_MAC_CTRL1]         = BIT(3) | BIT(7),
3396        [PHY_MAC_CTRL2]         = BIT(1) | BIT(2) | BIT(6) | BIT(12)
3397    },
3398    [3] = {
3399        [PHY_LED_TIMER_CTRL]    = BIT(0) | BIT(2) | BIT(14)
3400    }
3401};
3402
3403static const uint32_t e1000e_mac_reg_init[] = {
3404    [PBA]           =     0x00140014,
3405    [LEDCTL]        =  BIT(1) | BIT(8) | BIT(9) | BIT(15) | BIT(17) | BIT(18),
3406    [EXTCNF_CTRL]   = BIT(3),
3407    [EEMNGCTL]      = BIT(31),
3408    [FLASHT]        = 0x2,
3409    [FLSWCTL]       = BIT(30) | BIT(31),
3410    [FLOL]          = BIT(0),
3411    [RXDCTL]        = BIT(16),
3412    [RXDCTL1]       = BIT(16),
3413    [TIPG]          = 0x8 | (0x8 << 10) | (0x6 << 20),
3414    [RXCFGL]        = 0x88F7,
3415    [RXUDP]         = 0x319,
3416    [CTRL]          = E1000_CTRL_FD | E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN0 |
3417                      E1000_CTRL_SPD_1000 | E1000_CTRL_SLU |
3418                      E1000_CTRL_ADVD3WUC,
3419    [STATUS]        =  E1000_STATUS_ASDV_1000 | E1000_STATUS_LU,
3420    [PSRCTL]        = (2 << E1000_PSRCTL_BSIZE0_SHIFT) |
3421                      (4 << E1000_PSRCTL_BSIZE1_SHIFT) |
3422                      (4 << E1000_PSRCTL_BSIZE2_SHIFT),
3423    [TARC0]         = 0x3 | E1000_TARC_ENABLE,
3424    [TARC1]         = 0x3 | E1000_TARC_ENABLE,
3425    [EECD]          = E1000_EECD_AUTO_RD | E1000_EECD_PRES,
3426    [EERD]          = E1000_EERW_DONE,
3427    [EEWR]          = E1000_EERW_DONE,
3428    [GCR]           = E1000_L0S_ADJUST |
3429                      E1000_L1_ENTRY_LATENCY_MSB |
3430                      E1000_L1_ENTRY_LATENCY_LSB,
3431    [TDFH]          = 0x600,
3432    [TDFT]          = 0x600,
3433    [TDFHS]         = 0x600,
3434    [TDFTS]         = 0x600,
3435    [POEMB]         = 0x30D,
3436    [PBS]           = 0x028,
3437    [MANC]          = E1000_MANC_DIS_IP_CHK_ARP,
3438    [FACTPS]        = E1000_FACTPS_LAN0_ON | 0x20000000,
3439    [SWSM]          = 1,
3440    [RXCSUM]        = E1000_RXCSUM_IPOFLD | E1000_RXCSUM_TUOFLD,
3441    [ITR]           = E1000E_MIN_XITR,
3442    [EITR...EITR + E1000E_MSIX_VEC_NUM - 1] = E1000E_MIN_XITR,
3443};
3444
3445void
3446e1000e_core_reset(E1000ECore *core)
3447{
3448    int i;
3449
3450    timer_del(core->autoneg_timer);
3451
3452    e1000e_intrmgr_reset(core);
3453
3454    memset(core->phy, 0, sizeof core->phy);
3455    memmove(core->phy, e1000e_phy_reg_init, sizeof e1000e_phy_reg_init);
3456    memset(core->mac, 0, sizeof core->mac);
3457    memmove(core->mac, e1000e_mac_reg_init, sizeof e1000e_mac_reg_init);
3458
3459    core->rxbuf_min_shift = 1 + E1000_RING_DESC_LEN_SHIFT;
3460
3461    if (qemu_get_queue(core->owner_nic)->link_down) {
3462        e1000e_link_down(core);
3463    }
3464
3465    e1000x_reset_mac_addr(core->owner_nic, core->mac, core->permanent_mac);
3466
3467    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3468        net_tx_pkt_reset(core->tx[i].tx_pkt);
3469        memset(&core->tx[i].props, 0, sizeof(core->tx[i].props));
3470        core->tx[i].skip_cp = false;
3471    }
3472}
3473
3474void e1000e_core_pre_save(E1000ECore *core)
3475{
3476    int i;
3477    NetClientState *nc = qemu_get_queue(core->owner_nic);
3478
3479    /*
3480    * If link is down and auto-negotiation is supported and ongoing,
3481    * complete auto-negotiation immediately. This allows us to look
3482    * at MII_SR_AUTONEG_COMPLETE to infer link status on load.
3483    */
3484    if (nc->link_down && e1000e_have_autoneg(core)) {
3485        core->phy[0][PHY_STATUS] |= MII_SR_AUTONEG_COMPLETE;
3486        e1000e_update_flowctl_status(core);
3487    }
3488
3489    for (i = 0; i < ARRAY_SIZE(core->tx); i++) {
3490        if (net_tx_pkt_has_fragments(core->tx[i].tx_pkt)) {
3491            core->tx[i].skip_cp = true;
3492        }
3493    }
3494}
3495
3496int
3497e1000e_core_post_load(E1000ECore *core)
3498{
3499    NetClientState *nc = qemu_get_queue(core->owner_nic);
3500
3501    /* nc.link_down can't be migrated, so infer link_down according
3502     * to link status bit in core.mac[STATUS].
3503     */
3504    nc->link_down = (core->mac[STATUS] & E1000_STATUS_LU) == 0;
3505
3506    return 0;
3507}
3508