qemu/hw/intc/xive.c
<<
>>
Prefs
   1/*
   2 * QEMU PowerPC XIVE interrupt controller model
   3 *
   4 * Copyright (c) 2017-2018, IBM Corporation.
   5 *
   6 * This code is licensed under the GPL version 2 or later. See the
   7 * COPYING file in the top-level directory.
   8 */
   9
  10#include "qemu/osdep.h"
  11#include "qemu/log.h"
  12#include "qemu/module.h"
  13#include "qapi/error.h"
  14#include "target/ppc/cpu.h"
  15#include "sysemu/cpus.h"
  16#include "sysemu/dma.h"
  17#include "sysemu/reset.h"
  18#include "hw/qdev-properties.h"
  19#include "migration/vmstate.h"
  20#include "monitor/monitor.h"
  21#include "hw/irq.h"
  22#include "hw/ppc/xive.h"
  23#include "hw/ppc/xive_regs.h"
  24#include "trace.h"
  25
  26/*
  27 * XIVE Thread Interrupt Management context
  28 */
  29
  30/*
  31 * Convert an Interrupt Pending Buffer (IPB) register to a Pending
  32 * Interrupt Priority Register (PIPR), which contains the priority of
  33 * the most favored pending notification.
  34 */
  35static uint8_t ipb_to_pipr(uint8_t ibp)
  36{
  37    return ibp ? clz32((uint32_t)ibp << 24) : 0xff;
  38}
  39
  40static uint8_t exception_mask(uint8_t ring)
  41{
  42    switch (ring) {
  43    case TM_QW1_OS:
  44        return TM_QW1_NSR_EO;
  45    case TM_QW3_HV_PHYS:
  46        return TM_QW3_NSR_HE;
  47    default:
  48        g_assert_not_reached();
  49    }
  50}
  51
  52static qemu_irq xive_tctx_output(XiveTCTX *tctx, uint8_t ring)
  53{
  54        switch (ring) {
  55        case TM_QW0_USER:
  56                return 0; /* Not supported */
  57        case TM_QW1_OS:
  58                return tctx->os_output;
  59        case TM_QW2_HV_POOL:
  60        case TM_QW3_HV_PHYS:
  61                return tctx->hv_output;
  62        default:
  63                return 0;
  64        }
  65}
  66
  67static uint64_t xive_tctx_accept(XiveTCTX *tctx, uint8_t ring)
  68{
  69    uint8_t *regs = &tctx->regs[ring];
  70    uint8_t nsr = regs[TM_NSR];
  71    uint8_t mask = exception_mask(ring);
  72
  73    qemu_irq_lower(xive_tctx_output(tctx, ring));
  74
  75    if (regs[TM_NSR] & mask) {
  76        uint8_t cppr = regs[TM_PIPR];
  77
  78        regs[TM_CPPR] = cppr;
  79
  80        /* Reset the pending buffer bit */
  81        regs[TM_IPB] &= ~xive_priority_to_ipb(cppr);
  82        regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
  83
  84        /* Drop Exception bit */
  85        regs[TM_NSR] &= ~mask;
  86
  87        trace_xive_tctx_accept(tctx->cs->cpu_index, ring,
  88                               regs[TM_IPB], regs[TM_PIPR],
  89                               regs[TM_CPPR], regs[TM_NSR]);
  90    }
  91
  92    return (nsr << 8) | regs[TM_CPPR];
  93}
  94
  95static void xive_tctx_notify(XiveTCTX *tctx, uint8_t ring)
  96{
  97    uint8_t *regs = &tctx->regs[ring];
  98
  99    if (regs[TM_PIPR] < regs[TM_CPPR]) {
 100        switch (ring) {
 101        case TM_QW1_OS:
 102            regs[TM_NSR] |= TM_QW1_NSR_EO;
 103            break;
 104        case TM_QW3_HV_PHYS:
 105            regs[TM_NSR] |= (TM_QW3_NSR_HE_PHYS << 6);
 106            break;
 107        default:
 108            g_assert_not_reached();
 109        }
 110        trace_xive_tctx_notify(tctx->cs->cpu_index, ring,
 111                               regs[TM_IPB], regs[TM_PIPR],
 112                               regs[TM_CPPR], regs[TM_NSR]);
 113        qemu_irq_raise(xive_tctx_output(tctx, ring));
 114    }
 115}
 116
 117void xive_tctx_reset_os_signal(XiveTCTX *tctx)
 118{
 119    /*
 120     * Lower the External interrupt. Used when pulling an OS
 121     * context. It is necessary to avoid catching it in the hypervisor
 122     * context. It should be raised again when re-pushing the OS
 123     * context.
 124     */
 125    qemu_irq_lower(xive_tctx_output(tctx, TM_QW1_OS));
 126}
 127
 128static void xive_tctx_set_cppr(XiveTCTX *tctx, uint8_t ring, uint8_t cppr)
 129{
 130    uint8_t *regs = &tctx->regs[ring];
 131
 132    trace_xive_tctx_set_cppr(tctx->cs->cpu_index, ring,
 133                             regs[TM_IPB], regs[TM_PIPR],
 134                             cppr, regs[TM_NSR]);
 135
 136    if (cppr > XIVE_PRIORITY_MAX) {
 137        cppr = 0xff;
 138    }
 139
 140    tctx->regs[ring + TM_CPPR] = cppr;
 141
 142    /* CPPR has changed, check if we need to raise a pending exception */
 143    xive_tctx_notify(tctx, ring);
 144}
 145
 146void xive_tctx_ipb_update(XiveTCTX *tctx, uint8_t ring, uint8_t ipb)
 147{
 148    uint8_t *regs = &tctx->regs[ring];
 149
 150    regs[TM_IPB] |= ipb;
 151    regs[TM_PIPR] = ipb_to_pipr(regs[TM_IPB]);
 152    xive_tctx_notify(tctx, ring);
 153}
 154
 155/*
 156 * XIVE Thread Interrupt Management Area (TIMA)
 157 */
 158
 159static void xive_tm_set_hv_cppr(XivePresenter *xptr, XiveTCTX *tctx,
 160                                hwaddr offset, uint64_t value, unsigned size)
 161{
 162    xive_tctx_set_cppr(tctx, TM_QW3_HV_PHYS, value & 0xff);
 163}
 164
 165static uint64_t xive_tm_ack_hv_reg(XivePresenter *xptr, XiveTCTX *tctx,
 166                                   hwaddr offset, unsigned size)
 167{
 168    return xive_tctx_accept(tctx, TM_QW3_HV_PHYS);
 169}
 170
 171static uint64_t xive_tm_pull_pool_ctx(XivePresenter *xptr, XiveTCTX *tctx,
 172                                      hwaddr offset, unsigned size)
 173{
 174    uint32_t qw2w2_prev = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
 175    uint32_t qw2w2;
 176
 177    qw2w2 = xive_set_field32(TM_QW2W2_VP, qw2w2_prev, 0);
 178    memcpy(&tctx->regs[TM_QW2_HV_POOL + TM_WORD2], &qw2w2, 4);
 179    return qw2w2;
 180}
 181
 182static void xive_tm_vt_push(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
 183                            uint64_t value, unsigned size)
 184{
 185    tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] = value & 0xff;
 186}
 187
 188static uint64_t xive_tm_vt_poll(XivePresenter *xptr, XiveTCTX *tctx,
 189                                hwaddr offset, unsigned size)
 190{
 191    return tctx->regs[TM_QW3_HV_PHYS + TM_WORD2] & 0xff;
 192}
 193
 194/*
 195 * Define an access map for each page of the TIMA that we will use in
 196 * the memory region ops to filter values when doing loads and stores
 197 * of raw registers values
 198 *
 199 * Registers accessibility bits :
 200 *
 201 *    0x0 - no access
 202 *    0x1 - write only
 203 *    0x2 - read only
 204 *    0x3 - read/write
 205 */
 206
 207static const uint8_t xive_tm_hw_view[] = {
 208    3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
 209    3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
 210    0, 0, 3, 3,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
 211    3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   3, 3, 3, 0, /* QW-3 PHYS */
 212};
 213
 214static const uint8_t xive_tm_hv_view[] = {
 215    3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
 216    3, 3, 3, 3,   3, 3, 0, 2,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-1 OS   */
 217    0, 0, 3, 3,   0, 0, 0, 0,   0, 3, 3, 3,   0, 0, 0, 0, /* QW-2 POOL */
 218    3, 3, 3, 3,   0, 3, 0, 2,   3, 0, 0, 3,   0, 0, 0, 0, /* QW-3 PHYS */
 219};
 220
 221static const uint8_t xive_tm_os_view[] = {
 222    3, 0, 0, 0,   0, 0, 0, 0,   3, 3, 3, 3,   0, 0, 0, 0, /* QW-0 User */
 223    2, 3, 2, 2,   2, 2, 0, 2,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
 224    0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
 225    0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
 226};
 227
 228static const uint8_t xive_tm_user_view[] = {
 229    3, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-0 User */
 230    0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-1 OS   */
 231    0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-2 POOL */
 232    0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0,   0, 0, 0, 0, /* QW-3 PHYS */
 233};
 234
 235/*
 236 * Overall TIMA access map for the thread interrupt management context
 237 * registers
 238 */
 239static const uint8_t *xive_tm_views[] = {
 240    [XIVE_TM_HW_PAGE]   = xive_tm_hw_view,
 241    [XIVE_TM_HV_PAGE]   = xive_tm_hv_view,
 242    [XIVE_TM_OS_PAGE]   = xive_tm_os_view,
 243    [XIVE_TM_USER_PAGE] = xive_tm_user_view,
 244};
 245
 246/*
 247 * Computes a register access mask for a given offset in the TIMA
 248 */
 249static uint64_t xive_tm_mask(hwaddr offset, unsigned size, bool write)
 250{
 251    uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
 252    uint8_t reg_offset = offset & 0x3F;
 253    uint8_t reg_mask = write ? 0x1 : 0x2;
 254    uint64_t mask = 0x0;
 255    int i;
 256
 257    for (i = 0; i < size; i++) {
 258        if (xive_tm_views[page_offset][reg_offset + i] & reg_mask) {
 259            mask |= (uint64_t) 0xff << (8 * (size - i - 1));
 260        }
 261    }
 262
 263    return mask;
 264}
 265
 266static void xive_tm_raw_write(XiveTCTX *tctx, hwaddr offset, uint64_t value,
 267                              unsigned size)
 268{
 269    uint8_t ring_offset = offset & 0x30;
 270    uint8_t reg_offset = offset & 0x3F;
 271    uint64_t mask = xive_tm_mask(offset, size, true);
 272    int i;
 273
 274    /*
 275     * Only 4 or 8 bytes stores are allowed and the User ring is
 276     * excluded
 277     */
 278    if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
 279        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA @%"
 280                      HWADDR_PRIx"\n", offset);
 281        return;
 282    }
 283
 284    /*
 285     * Use the register offset for the raw values and filter out
 286     * reserved values
 287     */
 288    for (i = 0; i < size; i++) {
 289        uint8_t byte_mask = (mask >> (8 * (size - i - 1)));
 290        if (byte_mask) {
 291            tctx->regs[reg_offset + i] = (value >> (8 * (size - i - 1))) &
 292                byte_mask;
 293        }
 294    }
 295}
 296
 297static uint64_t xive_tm_raw_read(XiveTCTX *tctx, hwaddr offset, unsigned size)
 298{
 299    uint8_t ring_offset = offset & 0x30;
 300    uint8_t reg_offset = offset & 0x3F;
 301    uint64_t mask = xive_tm_mask(offset, size, false);
 302    uint64_t ret;
 303    int i;
 304
 305    /*
 306     * Only 4 or 8 bytes loads are allowed and the User ring is
 307     * excluded
 308     */
 309    if (size < 4 || !mask || ring_offset == TM_QW0_USER) {
 310        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access at TIMA @%"
 311                      HWADDR_PRIx"\n", offset);
 312        return -1;
 313    }
 314
 315    /* Use the register offset for the raw values */
 316    ret = 0;
 317    for (i = 0; i < size; i++) {
 318        ret |= (uint64_t) tctx->regs[reg_offset + i] << (8 * (size - i - 1));
 319    }
 320
 321    /* filter out reserved values */
 322    return ret & mask;
 323}
 324
 325/*
 326 * The TM context is mapped twice within each page. Stores and loads
 327 * to the first mapping below 2K write and read the specified values
 328 * without modification. The second mapping above 2K performs specific
 329 * state changes (side effects) in addition to setting/returning the
 330 * interrupt management area context of the processor thread.
 331 */
 332static uint64_t xive_tm_ack_os_reg(XivePresenter *xptr, XiveTCTX *tctx,
 333                                   hwaddr offset, unsigned size)
 334{
 335    return xive_tctx_accept(tctx, TM_QW1_OS);
 336}
 337
 338static void xive_tm_set_os_cppr(XivePresenter *xptr, XiveTCTX *tctx,
 339                                hwaddr offset, uint64_t value, unsigned size)
 340{
 341    xive_tctx_set_cppr(tctx, TM_QW1_OS, value & 0xff);
 342}
 343
 344/*
 345 * Adjust the IPB to allow a CPU to process event queues of other
 346 * priorities during one physical interrupt cycle.
 347 */
 348static void xive_tm_set_os_pending(XivePresenter *xptr, XiveTCTX *tctx,
 349                                   hwaddr offset, uint64_t value, unsigned size)
 350{
 351    xive_tctx_ipb_update(tctx, TM_QW1_OS, xive_priority_to_ipb(value & 0xff));
 352}
 353
 354static void xive_os_cam_decode(uint32_t cam, uint8_t *nvt_blk,
 355                               uint32_t *nvt_idx, bool *vo)
 356{
 357    if (nvt_blk) {
 358        *nvt_blk = xive_nvt_blk(cam);
 359    }
 360    if (nvt_idx) {
 361        *nvt_idx = xive_nvt_idx(cam);
 362    }
 363    if (vo) {
 364        *vo = !!(cam & TM_QW1W2_VO);
 365    }
 366}
 367
 368static uint32_t xive_tctx_get_os_cam(XiveTCTX *tctx, uint8_t *nvt_blk,
 369                                     uint32_t *nvt_idx, bool *vo)
 370{
 371    uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
 372    uint32_t cam = be32_to_cpu(qw1w2);
 373
 374    xive_os_cam_decode(cam, nvt_blk, nvt_idx, vo);
 375    return qw1w2;
 376}
 377
 378static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t qw1w2)
 379{
 380    memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
 381}
 382
 383static uint64_t xive_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
 384                                    hwaddr offset, unsigned size)
 385{
 386    uint32_t qw1w2;
 387    uint32_t qw1w2_new;
 388    uint8_t nvt_blk;
 389    uint32_t nvt_idx;
 390    bool vo;
 391
 392    qw1w2 = xive_tctx_get_os_cam(tctx, &nvt_blk, &nvt_idx, &vo);
 393
 394    if (!vo) {
 395        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: pulling invalid NVT %x/%x !?\n",
 396                      nvt_blk, nvt_idx);
 397    }
 398
 399    /* Invalidate CAM line */
 400    qw1w2_new = xive_set_field32(TM_QW1W2_VO, qw1w2, 0);
 401    xive_tctx_set_os_cam(tctx, qw1w2_new);
 402
 403    xive_tctx_reset_os_signal(tctx);
 404    return qw1w2;
 405}
 406
 407static void xive_tctx_need_resend(XiveRouter *xrtr, XiveTCTX *tctx,
 408                                  uint8_t nvt_blk, uint32_t nvt_idx)
 409{
 410    XiveNVT nvt;
 411    uint8_t ipb;
 412
 413    /*
 414     * Grab the associated NVT to pull the pending bits, and merge
 415     * them with the IPB of the thread interrupt context registers
 416     */
 417    if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
 418        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid NVT %x/%x\n",
 419                          nvt_blk, nvt_idx);
 420        return;
 421    }
 422
 423    ipb = xive_get_field32(NVT_W4_IPB, nvt.w4);
 424
 425    if (ipb) {
 426        /* Reset the NVT value */
 427        nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, 0);
 428        xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
 429    }
 430    /*
 431     * Always call xive_tctx_ipb_update(). Even if there were no
 432     * escalation triggered, there could be a pending interrupt which
 433     * was saved when the context was pulled and that we need to take
 434     * into account by recalculating the PIPR (which is not
 435     * saved/restored).
 436     * It will also raise the External interrupt signal if needed.
 437     */
 438    xive_tctx_ipb_update(tctx, TM_QW1_OS, ipb);
 439}
 440
 441/*
 442 * Updating the OS CAM line can trigger a resend of interrupt
 443 */
 444static void xive_tm_push_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
 445                                hwaddr offset, uint64_t value, unsigned size)
 446{
 447    uint32_t cam = value;
 448    uint32_t qw1w2 = cpu_to_be32(cam);
 449    uint8_t nvt_blk;
 450    uint32_t nvt_idx;
 451    bool vo;
 452
 453    xive_os_cam_decode(cam, &nvt_blk, &nvt_idx, &vo);
 454
 455    /* First update the registers */
 456    xive_tctx_set_os_cam(tctx, qw1w2);
 457
 458    /* Check the interrupt pending bits */
 459    if (vo) {
 460        xive_tctx_need_resend(XIVE_ROUTER(xptr), tctx, nvt_blk, nvt_idx);
 461    }
 462}
 463
 464/*
 465 * Define a mapping of "special" operations depending on the TIMA page
 466 * offset and the size of the operation.
 467 */
 468typedef struct XiveTmOp {
 469    uint8_t  page_offset;
 470    uint32_t op_offset;
 471    unsigned size;
 472    void     (*write_handler)(XivePresenter *xptr, XiveTCTX *tctx,
 473                              hwaddr offset,
 474                              uint64_t value, unsigned size);
 475    uint64_t (*read_handler)(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
 476                             unsigned size);
 477} XiveTmOp;
 478
 479static const XiveTmOp xive_tm_operations[] = {
 480    /*
 481     * MMIOs below 2K : raw values and special operations without side
 482     * effects
 483     */
 484    { XIVE_TM_OS_PAGE, TM_QW1_OS + TM_CPPR,   1, xive_tm_set_os_cppr, NULL },
 485    { XIVE_TM_HV_PAGE, TM_QW1_OS + TM_WORD2,     4, xive_tm_push_os_ctx, NULL },
 486    { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_CPPR, 1, xive_tm_set_hv_cppr, NULL },
 487    { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, xive_tm_vt_push, NULL },
 488    { XIVE_TM_HV_PAGE, TM_QW3_HV_PHYS + TM_WORD2, 1, NULL, xive_tm_vt_poll },
 489
 490    /* MMIOs above 2K : special operations with side effects */
 491    { XIVE_TM_OS_PAGE, TM_SPC_ACK_OS_REG,     2, NULL, xive_tm_ack_os_reg },
 492    { XIVE_TM_OS_PAGE, TM_SPC_SET_OS_PENDING, 1, xive_tm_set_os_pending, NULL },
 493    { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    4, NULL, xive_tm_pull_os_ctx },
 494    { XIVE_TM_HV_PAGE, TM_SPC_PULL_OS_CTX,    8, NULL, xive_tm_pull_os_ctx },
 495    { XIVE_TM_HV_PAGE, TM_SPC_ACK_HV_REG,     2, NULL, xive_tm_ack_hv_reg },
 496    { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  4, NULL, xive_tm_pull_pool_ctx },
 497    { XIVE_TM_HV_PAGE, TM_SPC_PULL_POOL_CTX,  8, NULL, xive_tm_pull_pool_ctx },
 498};
 499
 500static const XiveTmOp *xive_tm_find_op(hwaddr offset, unsigned size, bool write)
 501{
 502    uint8_t page_offset = (offset >> TM_SHIFT) & 0x3;
 503    uint32_t op_offset = offset & 0xFFF;
 504    int i;
 505
 506    for (i = 0; i < ARRAY_SIZE(xive_tm_operations); i++) {
 507        const XiveTmOp *xto = &xive_tm_operations[i];
 508
 509        /* Accesses done from a more privileged TIMA page is allowed */
 510        if (xto->page_offset >= page_offset &&
 511            xto->op_offset == op_offset &&
 512            xto->size == size &&
 513            ((write && xto->write_handler) || (!write && xto->read_handler))) {
 514            return xto;
 515        }
 516    }
 517    return NULL;
 518}
 519
 520/*
 521 * TIMA MMIO handlers
 522 */
 523void xive_tctx_tm_write(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
 524                        uint64_t value, unsigned size)
 525{
 526    const XiveTmOp *xto;
 527
 528    trace_xive_tctx_tm_write(offset, size, value);
 529
 530    /*
 531     * TODO: check V bit in Q[0-3]W2
 532     */
 533
 534    /*
 535     * First, check for special operations in the 2K region
 536     */
 537    if (offset & 0x800) {
 538        xto = xive_tm_find_op(offset, size, true);
 539        if (!xto) {
 540            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid write access at TIMA "
 541                          "@%"HWADDR_PRIx"\n", offset);
 542        } else {
 543            xto->write_handler(xptr, tctx, offset, value, size);
 544        }
 545        return;
 546    }
 547
 548    /*
 549     * Then, for special operations in the region below 2K.
 550     */
 551    xto = xive_tm_find_op(offset, size, true);
 552    if (xto) {
 553        xto->write_handler(xptr, tctx, offset, value, size);
 554        return;
 555    }
 556
 557    /*
 558     * Finish with raw access to the register values
 559     */
 560    xive_tm_raw_write(tctx, offset, value, size);
 561}
 562
 563uint64_t xive_tctx_tm_read(XivePresenter *xptr, XiveTCTX *tctx, hwaddr offset,
 564                           unsigned size)
 565{
 566    const XiveTmOp *xto;
 567    uint64_t ret;
 568
 569    /*
 570     * TODO: check V bit in Q[0-3]W2
 571     */
 572
 573    /*
 574     * First, check for special operations in the 2K region
 575     */
 576    if (offset & 0x800) {
 577        xto = xive_tm_find_op(offset, size, false);
 578        if (!xto) {
 579            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid read access to TIMA"
 580                          "@%"HWADDR_PRIx"\n", offset);
 581            return -1;
 582        }
 583        ret = xto->read_handler(xptr, tctx, offset, size);
 584        goto out;
 585    }
 586
 587    /*
 588     * Then, for special operations in the region below 2K.
 589     */
 590    xto = xive_tm_find_op(offset, size, false);
 591    if (xto) {
 592        ret = xto->read_handler(xptr, tctx, offset, size);
 593        goto out;
 594    }
 595
 596    /*
 597     * Finish with raw access to the register values
 598     */
 599    ret = xive_tm_raw_read(tctx, offset, size);
 600out:
 601    trace_xive_tctx_tm_read(offset, size, ret);
 602    return ret;
 603}
 604
 605static char *xive_tctx_ring_print(uint8_t *ring)
 606{
 607    uint32_t w2 = xive_tctx_word2(ring);
 608
 609    return g_strdup_printf("%02x   %02x  %02x    %02x   %02x  "
 610                   "%02x  %02x   %02x  %08x",
 611                   ring[TM_NSR], ring[TM_CPPR], ring[TM_IPB], ring[TM_LSMFB],
 612                   ring[TM_ACK_CNT], ring[TM_INC], ring[TM_AGE], ring[TM_PIPR],
 613                   be32_to_cpu(w2));
 614}
 615
 616static const char * const xive_tctx_ring_names[] = {
 617    "USER", "OS", "POOL", "PHYS",
 618};
 619
 620/*
 621 * kvm_irqchip_in_kernel() will cause the compiler to turn this
 622 * info a nop if CONFIG_KVM isn't defined.
 623 */
 624#define xive_in_kernel(xptr)                                            \
 625    (kvm_irqchip_in_kernel() &&                                         \
 626     ({                                                                 \
 627         XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);      \
 628         xpc->in_kernel ? xpc->in_kernel(xptr) : false;                 \
 629     }))
 630
 631void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
 632{
 633    int cpu_index;
 634    int i;
 635
 636    /* Skip partially initialized vCPUs. This can happen on sPAPR when vCPUs
 637     * are hot plugged or unplugged.
 638     */
 639    if (!tctx) {
 640        return;
 641    }
 642
 643    cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
 644
 645    if (xive_in_kernel(tctx->xptr)) {
 646        Error *local_err = NULL;
 647
 648        kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
 649        if (local_err) {
 650            error_report_err(local_err);
 651            return;
 652        }
 653    }
 654
 655    monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
 656                   "  W2\n", cpu_index);
 657
 658    for (i = 0; i < XIVE_TM_RING_COUNT; i++) {
 659        char *s = xive_tctx_ring_print(&tctx->regs[i * XIVE_TM_RING_SIZE]);
 660        monitor_printf(mon, "CPU[%04x]: %4s    %s\n", cpu_index,
 661                       xive_tctx_ring_names[i], s);
 662        g_free(s);
 663    }
 664}
 665
 666void xive_tctx_reset(XiveTCTX *tctx)
 667{
 668    memset(tctx->regs, 0, sizeof(tctx->regs));
 669
 670    /* Set some defaults */
 671    tctx->regs[TM_QW1_OS + TM_LSMFB] = 0xFF;
 672    tctx->regs[TM_QW1_OS + TM_ACK_CNT] = 0xFF;
 673    tctx->regs[TM_QW1_OS + TM_AGE] = 0xFF;
 674
 675    /*
 676     * Initialize PIPR to 0xFF to avoid phantom interrupts when the
 677     * CPPR is first set.
 678     */
 679    tctx->regs[TM_QW1_OS + TM_PIPR] =
 680        ipb_to_pipr(tctx->regs[TM_QW1_OS + TM_IPB]);
 681    tctx->regs[TM_QW3_HV_PHYS + TM_PIPR] =
 682        ipb_to_pipr(tctx->regs[TM_QW3_HV_PHYS + TM_IPB]);
 683}
 684
 685static void xive_tctx_realize(DeviceState *dev, Error **errp)
 686{
 687    XiveTCTX *tctx = XIVE_TCTX(dev);
 688    PowerPCCPU *cpu;
 689    CPUPPCState *env;
 690
 691    assert(tctx->cs);
 692    assert(tctx->xptr);
 693
 694    cpu = POWERPC_CPU(tctx->cs);
 695    env = &cpu->env;
 696    switch (PPC_INPUT(env)) {
 697    case PPC_FLAGS_INPUT_POWER9:
 698        tctx->hv_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_HINT);
 699        tctx->os_output = qdev_get_gpio_in(DEVICE(cpu), POWER9_INPUT_INT);
 700        break;
 701
 702    default:
 703        error_setg(errp, "XIVE interrupt controller does not support "
 704                   "this CPU bus model");
 705        return;
 706    }
 707
 708    /* Connect the presenter to the VCPU (required for CPU hotplug) */
 709    if (xive_in_kernel(tctx->xptr)) {
 710        if (kvmppc_xive_cpu_connect(tctx, errp) < 0) {
 711            return;
 712        }
 713    }
 714}
 715
 716static int vmstate_xive_tctx_pre_save(void *opaque)
 717{
 718    XiveTCTX *tctx = XIVE_TCTX(opaque);
 719    Error *local_err = NULL;
 720    int ret;
 721
 722    if (xive_in_kernel(tctx->xptr)) {
 723        ret = kvmppc_xive_cpu_get_state(tctx, &local_err);
 724        if (ret < 0) {
 725            error_report_err(local_err);
 726            return ret;
 727        }
 728    }
 729
 730    return 0;
 731}
 732
 733static int vmstate_xive_tctx_post_load(void *opaque, int version_id)
 734{
 735    XiveTCTX *tctx = XIVE_TCTX(opaque);
 736    Error *local_err = NULL;
 737    int ret;
 738
 739    if (xive_in_kernel(tctx->xptr)) {
 740        /*
 741         * Required for hotplugged CPU, for which the state comes
 742         * after all states of the machine.
 743         */
 744        ret = kvmppc_xive_cpu_set_state(tctx, &local_err);
 745        if (ret < 0) {
 746            error_report_err(local_err);
 747            return ret;
 748        }
 749    }
 750
 751    return 0;
 752}
 753
 754static const VMStateDescription vmstate_xive_tctx = {
 755    .name = TYPE_XIVE_TCTX,
 756    .version_id = 1,
 757    .minimum_version_id = 1,
 758    .pre_save = vmstate_xive_tctx_pre_save,
 759    .post_load = vmstate_xive_tctx_post_load,
 760    .fields = (VMStateField[]) {
 761        VMSTATE_BUFFER(regs, XiveTCTX),
 762        VMSTATE_END_OF_LIST()
 763    },
 764};
 765
 766static Property xive_tctx_properties[] = {
 767    DEFINE_PROP_LINK("cpu", XiveTCTX, cs, TYPE_CPU, CPUState *),
 768    DEFINE_PROP_LINK("presenter", XiveTCTX, xptr, TYPE_XIVE_PRESENTER,
 769                     XivePresenter *),
 770    DEFINE_PROP_END_OF_LIST(),
 771};
 772
 773static void xive_tctx_class_init(ObjectClass *klass, void *data)
 774{
 775    DeviceClass *dc = DEVICE_CLASS(klass);
 776
 777    dc->desc = "XIVE Interrupt Thread Context";
 778    dc->realize = xive_tctx_realize;
 779    dc->vmsd = &vmstate_xive_tctx;
 780    device_class_set_props(dc, xive_tctx_properties);
 781    /*
 782     * Reason: part of XIVE interrupt controller, needs to be wired up
 783     * by xive_tctx_create().
 784     */
 785    dc->user_creatable = false;
 786}
 787
 788static const TypeInfo xive_tctx_info = {
 789    .name          = TYPE_XIVE_TCTX,
 790    .parent        = TYPE_DEVICE,
 791    .instance_size = sizeof(XiveTCTX),
 792    .class_init    = xive_tctx_class_init,
 793};
 794
 795Object *xive_tctx_create(Object *cpu, XivePresenter *xptr, Error **errp)
 796{
 797    Object *obj;
 798
 799    obj = object_new(TYPE_XIVE_TCTX);
 800    object_property_add_child(cpu, TYPE_XIVE_TCTX, obj);
 801    object_unref(obj);
 802    object_property_set_link(obj, "cpu", cpu, &error_abort);
 803    object_property_set_link(obj, "presenter", OBJECT(xptr), &error_abort);
 804    if (!qdev_realize(DEVICE(obj), NULL, errp)) {
 805        object_unparent(obj);
 806        return NULL;
 807    }
 808    return obj;
 809}
 810
 811void xive_tctx_destroy(XiveTCTX *tctx)
 812{
 813    Object *obj = OBJECT(tctx);
 814
 815    object_unparent(obj);
 816}
 817
 818/*
 819 * XIVE ESB helpers
 820 */
 821
 822uint8_t xive_esb_set(uint8_t *pq, uint8_t value)
 823{
 824    uint8_t old_pq = *pq & 0x3;
 825
 826    *pq &= ~0x3;
 827    *pq |= value & 0x3;
 828
 829    return old_pq;
 830}
 831
 832bool xive_esb_trigger(uint8_t *pq)
 833{
 834    uint8_t old_pq = *pq & 0x3;
 835
 836    switch (old_pq) {
 837    case XIVE_ESB_RESET:
 838        xive_esb_set(pq, XIVE_ESB_PENDING);
 839        return true;
 840    case XIVE_ESB_PENDING:
 841    case XIVE_ESB_QUEUED:
 842        xive_esb_set(pq, XIVE_ESB_QUEUED);
 843        return false;
 844    case XIVE_ESB_OFF:
 845        xive_esb_set(pq, XIVE_ESB_OFF);
 846        return false;
 847    default:
 848         g_assert_not_reached();
 849    }
 850}
 851
 852bool xive_esb_eoi(uint8_t *pq)
 853{
 854    uint8_t old_pq = *pq & 0x3;
 855
 856    switch (old_pq) {
 857    case XIVE_ESB_RESET:
 858    case XIVE_ESB_PENDING:
 859        xive_esb_set(pq, XIVE_ESB_RESET);
 860        return false;
 861    case XIVE_ESB_QUEUED:
 862        xive_esb_set(pq, XIVE_ESB_PENDING);
 863        return true;
 864    case XIVE_ESB_OFF:
 865        xive_esb_set(pq, XIVE_ESB_OFF);
 866        return false;
 867    default:
 868         g_assert_not_reached();
 869    }
 870}
 871
 872/*
 873 * XIVE Interrupt Source (or IVSE)
 874 */
 875
 876uint8_t xive_source_esb_get(XiveSource *xsrc, uint32_t srcno)
 877{
 878    assert(srcno < xsrc->nr_irqs);
 879
 880    return xsrc->status[srcno] & 0x3;
 881}
 882
 883uint8_t xive_source_esb_set(XiveSource *xsrc, uint32_t srcno, uint8_t pq)
 884{
 885    assert(srcno < xsrc->nr_irqs);
 886
 887    return xive_esb_set(&xsrc->status[srcno], pq);
 888}
 889
 890/*
 891 * Returns whether the event notification should be forwarded.
 892 */
 893static bool xive_source_lsi_trigger(XiveSource *xsrc, uint32_t srcno)
 894{
 895    uint8_t old_pq = xive_source_esb_get(xsrc, srcno);
 896
 897    xive_source_set_asserted(xsrc, srcno, true);
 898
 899    switch (old_pq) {
 900    case XIVE_ESB_RESET:
 901        xive_source_esb_set(xsrc, srcno, XIVE_ESB_PENDING);
 902        return true;
 903    default:
 904        return false;
 905    }
 906}
 907
 908/*
 909 * Sources can be configured with PQ offloading in which case the check
 910 * on the PQ state bits of MSIs is disabled
 911 */
 912static bool xive_source_esb_disabled(XiveSource *xsrc, uint32_t srcno)
 913{
 914    return (xsrc->esb_flags & XIVE_SRC_PQ_DISABLE) &&
 915        !xive_source_irq_is_lsi(xsrc, srcno);
 916}
 917
 918/*
 919 * Returns whether the event notification should be forwarded.
 920 */
 921static bool xive_source_esb_trigger(XiveSource *xsrc, uint32_t srcno)
 922{
 923    bool ret;
 924
 925    assert(srcno < xsrc->nr_irqs);
 926
 927    if (xive_source_esb_disabled(xsrc, srcno)) {
 928        return true;
 929    }
 930
 931    ret = xive_esb_trigger(&xsrc->status[srcno]);
 932
 933    if (xive_source_irq_is_lsi(xsrc, srcno) &&
 934        xive_source_esb_get(xsrc, srcno) == XIVE_ESB_QUEUED) {
 935        qemu_log_mask(LOG_GUEST_ERROR,
 936                      "XIVE: queued an event on LSI IRQ %d\n", srcno);
 937    }
 938
 939    return ret;
 940}
 941
 942/*
 943 * Returns whether the event notification should be forwarded.
 944 */
 945static bool xive_source_esb_eoi(XiveSource *xsrc, uint32_t srcno)
 946{
 947    bool ret;
 948
 949    assert(srcno < xsrc->nr_irqs);
 950
 951    if (xive_source_esb_disabled(xsrc, srcno)) {
 952        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EOI for IRQ %d\n", srcno);
 953        return false;
 954    }
 955
 956    ret = xive_esb_eoi(&xsrc->status[srcno]);
 957
 958    /*
 959     * LSI sources do not set the Q bit but they can still be
 960     * asserted, in which case we should forward a new event
 961     * notification
 962     */
 963    if (xive_source_irq_is_lsi(xsrc, srcno) &&
 964        xive_source_is_asserted(xsrc, srcno)) {
 965        ret = xive_source_lsi_trigger(xsrc, srcno);
 966    }
 967
 968    return ret;
 969}
 970
 971/*
 972 * Forward the source event notification to the Router
 973 */
 974static void xive_source_notify(XiveSource *xsrc, int srcno)
 975{
 976    XiveNotifierClass *xnc = XIVE_NOTIFIER_GET_CLASS(xsrc->xive);
 977    bool pq_checked = !xive_source_esb_disabled(xsrc, srcno);
 978
 979    if (xnc->notify) {
 980        xnc->notify(xsrc->xive, srcno, pq_checked);
 981    }
 982}
 983
 984/*
 985 * In a two pages ESB MMIO setting, even page is the trigger page, odd
 986 * page is for management
 987 */
 988static inline bool addr_is_even(hwaddr addr, uint32_t shift)
 989{
 990    return !((addr >> shift) & 1);
 991}
 992
 993static inline bool xive_source_is_trigger_page(XiveSource *xsrc, hwaddr addr)
 994{
 995    return xive_source_esb_has_2page(xsrc) &&
 996        addr_is_even(addr, xsrc->esb_shift - 1);
 997}
 998
 999/*
1000 * ESB MMIO loads
1001 *                      Trigger page    Management/EOI page
1002 *
1003 * ESB MMIO setting     2 pages         1 or 2 pages
1004 *
1005 * 0x000 .. 0x3FF       -1              EOI and return 0|1
1006 * 0x400 .. 0x7FF       -1              EOI and return 0|1
1007 * 0x800 .. 0xBFF       -1              return PQ
1008 * 0xC00 .. 0xCFF       -1              return PQ and atomically PQ=00
1009 * 0xD00 .. 0xDFF       -1              return PQ and atomically PQ=01
1010 * 0xE00 .. 0xDFF       -1              return PQ and atomically PQ=10
1011 * 0xF00 .. 0xDFF       -1              return PQ and atomically PQ=11
1012 */
1013static uint64_t xive_source_esb_read(void *opaque, hwaddr addr, unsigned size)
1014{
1015    XiveSource *xsrc = XIVE_SOURCE(opaque);
1016    uint32_t offset = addr & 0xFFF;
1017    uint32_t srcno = addr >> xsrc->esb_shift;
1018    uint64_t ret = -1;
1019
1020    /* In a two pages ESB MMIO setting, trigger page should not be read */
1021    if (xive_source_is_trigger_page(xsrc, addr)) {
1022        qemu_log_mask(LOG_GUEST_ERROR,
1023                      "XIVE: invalid load on IRQ %d trigger page at "
1024                      "0x%"HWADDR_PRIx"\n", srcno, addr);
1025        return -1;
1026    }
1027
1028    switch (offset) {
1029    case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1030        ret = xive_source_esb_eoi(xsrc, srcno);
1031
1032        /* Forward the source event notification for routing */
1033        if (ret) {
1034            xive_source_notify(xsrc, srcno);
1035        }
1036        break;
1037
1038    case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1039        ret = xive_source_esb_get(xsrc, srcno);
1040        break;
1041
1042    case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1043    case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1044    case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1045    case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1046        ret = xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1047        break;
1048    default:
1049        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB load addr %x\n",
1050                      offset);
1051    }
1052
1053    trace_xive_source_esb_read(addr, srcno, ret);
1054
1055    return ret;
1056}
1057
1058/*
1059 * ESB MMIO stores
1060 *                      Trigger page    Management/EOI page
1061 *
1062 * ESB MMIO setting     2 pages         1 or 2 pages
1063 *
1064 * 0x000 .. 0x3FF       Trigger         Trigger
1065 * 0x400 .. 0x7FF       Trigger         EOI
1066 * 0x800 .. 0xBFF       Trigger         undefined
1067 * 0xC00 .. 0xCFF       Trigger         PQ=00
1068 * 0xD00 .. 0xDFF       Trigger         PQ=01
1069 * 0xE00 .. 0xDFF       Trigger         PQ=10
1070 * 0xF00 .. 0xDFF       Trigger         PQ=11
1071 */
1072static void xive_source_esb_write(void *opaque, hwaddr addr,
1073                                  uint64_t value, unsigned size)
1074{
1075    XiveSource *xsrc = XIVE_SOURCE(opaque);
1076    uint32_t offset = addr & 0xFFF;
1077    uint32_t srcno = addr >> xsrc->esb_shift;
1078    bool notify = false;
1079
1080    trace_xive_source_esb_write(addr, srcno, value);
1081
1082    /* In a two pages ESB MMIO setting, trigger page only triggers */
1083    if (xive_source_is_trigger_page(xsrc, addr)) {
1084        notify = xive_source_esb_trigger(xsrc, srcno);
1085        goto out;
1086    }
1087
1088    switch (offset) {
1089    case 0 ... 0x3FF:
1090        notify = xive_source_esb_trigger(xsrc, srcno);
1091        break;
1092
1093    case XIVE_ESB_STORE_EOI ... XIVE_ESB_STORE_EOI + 0x3FF:
1094        if (!(xsrc->esb_flags & XIVE_SRC_STORE_EOI)) {
1095            qemu_log_mask(LOG_GUEST_ERROR,
1096                          "XIVE: invalid Store EOI for IRQ %d\n", srcno);
1097            return;
1098        }
1099
1100        notify = xive_source_esb_eoi(xsrc, srcno);
1101        break;
1102
1103    /*
1104     * This is an internal offset used to inject triggers when the PQ
1105     * state bits are not controlled locally. Such as for LSIs when
1106     * under ABT mode.
1107     */
1108    case XIVE_ESB_INJECT ... XIVE_ESB_INJECT + 0x3FF:
1109        notify = true;
1110        break;
1111
1112    case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1113    case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1114    case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1115    case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1116        xive_source_esb_set(xsrc, srcno, (offset >> 8) & 0x3);
1117        break;
1118
1119    default:
1120        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr %x\n",
1121                      offset);
1122        return;
1123    }
1124
1125out:
1126    /* Forward the source event notification for routing */
1127    if (notify) {
1128        xive_source_notify(xsrc, srcno);
1129    }
1130}
1131
1132static const MemoryRegionOps xive_source_esb_ops = {
1133    .read = xive_source_esb_read,
1134    .write = xive_source_esb_write,
1135    .endianness = DEVICE_BIG_ENDIAN,
1136    .valid = {
1137        .min_access_size = 8,
1138        .max_access_size = 8,
1139    },
1140    .impl = {
1141        .min_access_size = 8,
1142        .max_access_size = 8,
1143    },
1144};
1145
1146void xive_source_set_irq(void *opaque, int srcno, int val)
1147{
1148    XiveSource *xsrc = XIVE_SOURCE(opaque);
1149    bool notify = false;
1150
1151    if (xive_source_irq_is_lsi(xsrc, srcno)) {
1152        if (val) {
1153            notify = xive_source_lsi_trigger(xsrc, srcno);
1154        } else {
1155            xive_source_set_asserted(xsrc, srcno, false);
1156        }
1157    } else {
1158        if (val) {
1159            notify = xive_source_esb_trigger(xsrc, srcno);
1160        }
1161    }
1162
1163    /* Forward the source event notification for routing */
1164    if (notify) {
1165        xive_source_notify(xsrc, srcno);
1166    }
1167}
1168
1169void xive_source_pic_print_info(XiveSource *xsrc, uint32_t offset, Monitor *mon)
1170{
1171    int i;
1172
1173    for (i = 0; i < xsrc->nr_irqs; i++) {
1174        uint8_t pq = xive_source_esb_get(xsrc, i);
1175
1176        if (pq == XIVE_ESB_OFF) {
1177            continue;
1178        }
1179
1180        monitor_printf(mon, "  %08x %s %c%c%c\n", i + offset,
1181                       xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
1182                       pq & XIVE_ESB_VAL_P ? 'P' : '-',
1183                       pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1184                       xive_source_is_asserted(xsrc, i) ? 'A' : ' ');
1185    }
1186}
1187
1188static void xive_source_reset(void *dev)
1189{
1190    XiveSource *xsrc = XIVE_SOURCE(dev);
1191
1192    /* Do not clear the LSI bitmap */
1193
1194    /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
1195    memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
1196}
1197
1198static void xive_source_realize(DeviceState *dev, Error **errp)
1199{
1200    XiveSource *xsrc = XIVE_SOURCE(dev);
1201    size_t esb_len = xive_source_esb_len(xsrc);
1202
1203    assert(xsrc->xive);
1204
1205    if (!xsrc->nr_irqs) {
1206        error_setg(errp, "Number of interrupt needs to be greater than 0");
1207        return;
1208    }
1209
1210    if (xsrc->esb_shift != XIVE_ESB_4K &&
1211        xsrc->esb_shift != XIVE_ESB_4K_2PAGE &&
1212        xsrc->esb_shift != XIVE_ESB_64K &&
1213        xsrc->esb_shift != XIVE_ESB_64K_2PAGE) {
1214        error_setg(errp, "Invalid ESB shift setting");
1215        return;
1216    }
1217
1218    xsrc->status = g_malloc0(xsrc->nr_irqs);
1219    xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
1220
1221    memory_region_init(&xsrc->esb_mmio, OBJECT(xsrc), "xive.esb", esb_len);
1222    memory_region_init_io(&xsrc->esb_mmio_emulated, OBJECT(xsrc),
1223                          &xive_source_esb_ops, xsrc, "xive.esb-emulated",
1224                          esb_len);
1225    memory_region_add_subregion(&xsrc->esb_mmio, 0, &xsrc->esb_mmio_emulated);
1226
1227    qemu_register_reset(xive_source_reset, dev);
1228}
1229
1230static const VMStateDescription vmstate_xive_source = {
1231    .name = TYPE_XIVE_SOURCE,
1232    .version_id = 1,
1233    .minimum_version_id = 1,
1234    .fields = (VMStateField[]) {
1235        VMSTATE_UINT32_EQUAL(nr_irqs, XiveSource, NULL),
1236        VMSTATE_VBUFFER_UINT32(status, XiveSource, 1, NULL, nr_irqs),
1237        VMSTATE_END_OF_LIST()
1238    },
1239};
1240
1241/*
1242 * The default XIVE interrupt source setting for the ESB MMIOs is two
1243 * 64k pages without Store EOI, to be in sync with KVM.
1244 */
1245static Property xive_source_properties[] = {
1246    DEFINE_PROP_UINT64("flags", XiveSource, esb_flags, 0),
1247    DEFINE_PROP_UINT32("nr-irqs", XiveSource, nr_irqs, 0),
1248    DEFINE_PROP_UINT32("shift", XiveSource, esb_shift, XIVE_ESB_64K_2PAGE),
1249    DEFINE_PROP_LINK("xive", XiveSource, xive, TYPE_XIVE_NOTIFIER,
1250                     XiveNotifier *),
1251    DEFINE_PROP_END_OF_LIST(),
1252};
1253
1254static void xive_source_class_init(ObjectClass *klass, void *data)
1255{
1256    DeviceClass *dc = DEVICE_CLASS(klass);
1257
1258    dc->desc    = "XIVE Interrupt Source";
1259    device_class_set_props(dc, xive_source_properties);
1260    dc->realize = xive_source_realize;
1261    dc->vmsd    = &vmstate_xive_source;
1262    /*
1263     * Reason: part of XIVE interrupt controller, needs to be wired up,
1264     * e.g. by spapr_xive_instance_init().
1265     */
1266    dc->user_creatable = false;
1267}
1268
1269static const TypeInfo xive_source_info = {
1270    .name          = TYPE_XIVE_SOURCE,
1271    .parent        = TYPE_DEVICE,
1272    .instance_size = sizeof(XiveSource),
1273    .class_init    = xive_source_class_init,
1274};
1275
1276/*
1277 * XiveEND helpers
1278 */
1279
1280void xive_end_queue_pic_print_info(XiveEND *end, uint32_t width, Monitor *mon)
1281{
1282    uint64_t qaddr_base = xive_end_qaddr(end);
1283    uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1284    uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1285    uint32_t qentries = 1 << (qsize + 10);
1286    int i;
1287
1288    /*
1289     * print out the [ (qindex - (width - 1)) .. (qindex + 1)] window
1290     */
1291    monitor_printf(mon, " [ ");
1292    qindex = (qindex - (width - 1)) & (qentries - 1);
1293    for (i = 0; i < width; i++) {
1294        uint64_t qaddr = qaddr_base + (qindex << 2);
1295        uint32_t qdata = -1;
1296
1297        if (dma_memory_read(&address_space_memory, qaddr,
1298                            &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1299            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to read EQ @0x%"
1300                          HWADDR_PRIx "\n", qaddr);
1301            return;
1302        }
1303        monitor_printf(mon, "%s%08x ", i == width - 1 ? "^" : "",
1304                       be32_to_cpu(qdata));
1305        qindex = (qindex + 1) & (qentries - 1);
1306    }
1307    monitor_printf(mon, "]");
1308}
1309
1310void xive_end_pic_print_info(XiveEND *end, uint32_t end_idx, Monitor *mon)
1311{
1312    uint64_t qaddr_base = xive_end_qaddr(end);
1313    uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1314    uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1315    uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1316    uint32_t qentries = 1 << (qsize + 10);
1317
1318    uint32_t nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1319    uint32_t nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1320    uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1321    uint8_t pq;
1322
1323    if (!xive_end_is_valid(end)) {
1324        return;
1325    }
1326
1327    pq = xive_get_field32(END_W1_ESn, end->w1);
1328
1329    monitor_printf(mon, "  %08x %c%c %c%c%c%c%c%c%c%c prio:%d nvt:%02x/%04x",
1330                   end_idx,
1331                   pq & XIVE_ESB_VAL_P ? 'P' : '-',
1332                   pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1333                   xive_end_is_valid(end)    ? 'v' : '-',
1334                   xive_end_is_enqueue(end)  ? 'q' : '-',
1335                   xive_end_is_notify(end)   ? 'n' : '-',
1336                   xive_end_is_backlog(end)  ? 'b' : '-',
1337                   xive_end_is_escalate(end) ? 'e' : '-',
1338                   xive_end_is_uncond_escalation(end)   ? 'u' : '-',
1339                   xive_end_is_silent_escalation(end)   ? 's' : '-',
1340                   xive_end_is_firmware(end)   ? 'f' : '-',
1341                   priority, nvt_blk, nvt_idx);
1342
1343    if (qaddr_base) {
1344        monitor_printf(mon, " eq:@%08"PRIx64"% 6d/%5d ^%d",
1345                       qaddr_base, qindex, qentries, qgen);
1346        xive_end_queue_pic_print_info(end, 6, mon);
1347    }
1348    monitor_printf(mon, "\n");
1349}
1350
1351static void xive_end_enqueue(XiveEND *end, uint32_t data)
1352{
1353    uint64_t qaddr_base = xive_end_qaddr(end);
1354    uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
1355    uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1356    uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
1357
1358    uint64_t qaddr = qaddr_base + (qindex << 2);
1359    uint32_t qdata = cpu_to_be32((qgen << 31) | (data & 0x7fffffff));
1360    uint32_t qentries = 1 << (qsize + 10);
1361
1362    if (dma_memory_write(&address_space_memory, qaddr,
1363                         &qdata, sizeof(qdata), MEMTXATTRS_UNSPECIFIED)) {
1364        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to write END data @0x%"
1365                      HWADDR_PRIx "\n", qaddr);
1366        return;
1367    }
1368
1369    qindex = (qindex + 1) & (qentries - 1);
1370    if (qindex == 0) {
1371        qgen ^= 1;
1372        end->w1 = xive_set_field32(END_W1_GENERATION, end->w1, qgen);
1373    }
1374    end->w1 = xive_set_field32(END_W1_PAGE_OFF, end->w1, qindex);
1375}
1376
1377void xive_end_eas_pic_print_info(XiveEND *end, uint32_t end_idx,
1378                                   Monitor *mon)
1379{
1380    XiveEAS *eas = (XiveEAS *) &end->w4;
1381    uint8_t pq;
1382
1383    if (!xive_end_is_escalate(end)) {
1384        return;
1385    }
1386
1387    pq = xive_get_field32(END_W1_ESe, end->w1);
1388
1389    monitor_printf(mon, "  %08x %c%c %c%c end:%02x/%04x data:%08x\n",
1390                   end_idx,
1391                   pq & XIVE_ESB_VAL_P ? 'P' : '-',
1392                   pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
1393                   xive_eas_is_valid(eas) ? 'V' : ' ',
1394                   xive_eas_is_masked(eas) ? 'M' : ' ',
1395                   (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1396                   (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1397                   (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1398}
1399
1400/*
1401 * XIVE Router (aka. Virtualization Controller or IVRE)
1402 */
1403
1404int xive_router_get_eas(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1405                        XiveEAS *eas)
1406{
1407    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1408
1409    return xrc->get_eas(xrtr, eas_blk, eas_idx, eas);
1410}
1411
1412static
1413int xive_router_get_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1414                       uint8_t *pq)
1415{
1416    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1417
1418    return xrc->get_pq(xrtr, eas_blk, eas_idx, pq);
1419}
1420
1421static
1422int xive_router_set_pq(XiveRouter *xrtr, uint8_t eas_blk, uint32_t eas_idx,
1423                       uint8_t *pq)
1424{
1425    XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1426
1427    return xrc->set_pq(xrtr, eas_blk, eas_idx, pq);
1428}
1429
1430int xive_router_get_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1431                        XiveEND *end)
1432{
1433   XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1434
1435   return xrc->get_end(xrtr, end_blk, end_idx, end);
1436}
1437
1438int xive_router_write_end(XiveRouter *xrtr, uint8_t end_blk, uint32_t end_idx,
1439                          XiveEND *end, uint8_t word_number)
1440{
1441   XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1442
1443   return xrc->write_end(xrtr, end_blk, end_idx, end, word_number);
1444}
1445
1446int xive_router_get_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1447                        XiveNVT *nvt)
1448{
1449   XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1450
1451   return xrc->get_nvt(xrtr, nvt_blk, nvt_idx, nvt);
1452}
1453
1454int xive_router_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk, uint32_t nvt_idx,
1455                        XiveNVT *nvt, uint8_t word_number)
1456{
1457   XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1458
1459   return xrc->write_nvt(xrtr, nvt_blk, nvt_idx, nvt, word_number);
1460}
1461
1462static int xive_router_get_block_id(XiveRouter *xrtr)
1463{
1464   XiveRouterClass *xrc = XIVE_ROUTER_GET_CLASS(xrtr);
1465
1466   return xrc->get_block_id(xrtr);
1467}
1468
1469static void xive_router_realize(DeviceState *dev, Error **errp)
1470{
1471    XiveRouter *xrtr = XIVE_ROUTER(dev);
1472
1473    assert(xrtr->xfb);
1474}
1475
1476/*
1477 * Encode the HW CAM line in the block group mode format :
1478 *
1479 *   chip << 19 | 0000000 0 0001 thread (7Bit)
1480 */
1481static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
1482{
1483    CPUPPCState *env = &POWERPC_CPU(tctx->cs)->env;
1484    uint32_t pir = env->spr_cb[SPR_PIR].default_value;
1485    uint8_t blk = xive_router_get_block_id(XIVE_ROUTER(xptr));
1486
1487    return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
1488}
1489
1490/*
1491 * The thread context register words are in big-endian format.
1492 */
1493int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
1494                              uint8_t format,
1495                              uint8_t nvt_blk, uint32_t nvt_idx,
1496                              bool cam_ignore, uint32_t logic_serv)
1497{
1498    uint32_t cam = xive_nvt_cam_line(nvt_blk, nvt_idx);
1499    uint32_t qw3w2 = xive_tctx_word2(&tctx->regs[TM_QW3_HV_PHYS]);
1500    uint32_t qw2w2 = xive_tctx_word2(&tctx->regs[TM_QW2_HV_POOL]);
1501    uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
1502    uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
1503
1504    /*
1505     * TODO (PowerNV): ignore mode. The low order bits of the NVT
1506     * identifier are ignored in the "CAM" match.
1507     */
1508
1509    if (format == 0) {
1510        if (cam_ignore == true) {
1511            /*
1512             * F=0 & i=1: Logical server notification (bits ignored at
1513             * the end of the NVT identifier)
1514             */
1515            qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
1516                          nvt_blk, nvt_idx);
1517             return -1;
1518        }
1519
1520        /* F=0 & i=0: Specific NVT notification */
1521
1522        /* PHYS ring */
1523        if ((be32_to_cpu(qw3w2) & TM_QW3W2_VT) &&
1524            cam == xive_tctx_hw_cam_line(xptr, tctx)) {
1525            return TM_QW3_HV_PHYS;
1526        }
1527
1528        /* HV POOL ring */
1529        if ((be32_to_cpu(qw2w2) & TM_QW2W2_VP) &&
1530            cam == xive_get_field32(TM_QW2W2_POOL_CAM, qw2w2)) {
1531            return TM_QW2_HV_POOL;
1532        }
1533
1534        /* OS ring */
1535        if ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1536            cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) {
1537            return TM_QW1_OS;
1538        }
1539    } else {
1540        /* F=1 : User level Event-Based Branch (EBB) notification */
1541
1542        /* USER ring */
1543        if  ((be32_to_cpu(qw1w2) & TM_QW1W2_VO) &&
1544             (cam == xive_get_field32(TM_QW1W2_OS_CAM, qw1w2)) &&
1545             (be32_to_cpu(qw0w2) & TM_QW0W2_VU) &&
1546             (logic_serv == xive_get_field32(TM_QW0W2_LOGIC_SERV, qw0w2))) {
1547            return TM_QW0_USER;
1548        }
1549    }
1550    return -1;
1551}
1552
1553/*
1554 * This is our simple Xive Presenter Engine model. It is merged in the
1555 * Router as it does not require an extra object.
1556 *
1557 * It receives notification requests sent by the IVRE to find one
1558 * matching NVT (or more) dispatched on the processor threads. In case
1559 * of a single NVT notification, the process is abreviated and the
1560 * thread is signaled if a match is found. In case of a logical server
1561 * notification (bits ignored at the end of the NVT identifier), the
1562 * IVPE and IVRE select a winning thread using different filters. This
1563 * involves 2 or 3 exchanges on the PowerBus that the model does not
1564 * support.
1565 *
1566 * The parameters represent what is sent on the PowerBus
1567 */
1568bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
1569                           uint8_t nvt_blk, uint32_t nvt_idx,
1570                           bool cam_ignore, uint8_t priority,
1571                           uint32_t logic_serv)
1572{
1573    XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
1574    XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
1575    int count;
1576
1577    /*
1578     * Ask the machine to scan the interrupt controllers for a match
1579     */
1580    count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
1581                           priority, logic_serv, &match);
1582    if (count < 0) {
1583        return false;
1584    }
1585
1586    /* handle CPU exception delivery */
1587    if (count) {
1588        trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring);
1589        xive_tctx_ipb_update(match.tctx, match.ring,
1590                             xive_priority_to_ipb(priority));
1591    }
1592
1593    return !!count;
1594}
1595
1596/*
1597 * Notification using the END ESe/ESn bit (Event State Buffer for
1598 * escalation and notification). Provide further coalescing in the
1599 * Router.
1600 */
1601static bool xive_router_end_es_notify(XiveRouter *xrtr, uint8_t end_blk,
1602                                      uint32_t end_idx, XiveEND *end,
1603                                      uint32_t end_esmask)
1604{
1605    uint8_t pq = xive_get_field32(end_esmask, end->w1);
1606    bool notify = xive_esb_trigger(&pq);
1607
1608    if (pq != xive_get_field32(end_esmask, end->w1)) {
1609        end->w1 = xive_set_field32(end_esmask, end->w1, pq);
1610        xive_router_write_end(xrtr, end_blk, end_idx, end, 1);
1611    }
1612
1613    /* ESe/n[Q]=1 : end of notification */
1614    return notify;
1615}
1616
1617/*
1618 * An END trigger can come from an event trigger (IPI or HW) or from
1619 * another chip. We don't model the PowerBus but the END trigger
1620 * message has the same parameters than in the function below.
1621 */
1622static void xive_router_end_notify(XiveRouter *xrtr, uint8_t end_blk,
1623                                   uint32_t end_idx, uint32_t end_data)
1624{
1625    XiveEND end;
1626    uint8_t priority;
1627    uint8_t format;
1628    uint8_t nvt_blk;
1629    uint32_t nvt_idx;
1630    XiveNVT nvt;
1631    bool found;
1632
1633    /* END cache lookup */
1634    if (xive_router_get_end(xrtr, end_blk, end_idx, &end)) {
1635        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1636                      end_idx);
1637        return;
1638    }
1639
1640    if (!xive_end_is_valid(&end)) {
1641        trace_xive_router_end_notify(end_blk, end_idx, end_data);
1642        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1643                      end_blk, end_idx);
1644        return;
1645    }
1646
1647    if (xive_end_is_enqueue(&end)) {
1648        xive_end_enqueue(&end, end_data);
1649        /* Enqueuing event data modifies the EQ toggle and index */
1650        xive_router_write_end(xrtr, end_blk, end_idx, &end, 1);
1651    }
1652
1653    /*
1654     * When the END is silent, we skip the notification part.
1655     */
1656    if (xive_end_is_silent_escalation(&end)) {
1657        goto do_escalation;
1658    }
1659
1660    /*
1661     * The W7 format depends on the F bit in W6. It defines the type
1662     * of the notification :
1663     *
1664     *   F=0 : single or multiple NVT notification
1665     *   F=1 : User level Event-Based Branch (EBB) notification, no
1666     *         priority
1667     */
1668    format = xive_get_field32(END_W6_FORMAT_BIT, end.w6);
1669    priority = xive_get_field32(END_W7_F0_PRIORITY, end.w7);
1670
1671    /* The END is masked */
1672    if (format == 0 && priority == 0xff) {
1673        return;
1674    }
1675
1676    /*
1677     * Check the END ESn (Event State Buffer for notification) for
1678     * even further coalescing in the Router
1679     */
1680    if (!xive_end_is_notify(&end)) {
1681        /* ESn[Q]=1 : end of notification */
1682        if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1683                                       &end, END_W1_ESn)) {
1684            return;
1685        }
1686    }
1687
1688    /*
1689     * Follows IVPE notification
1690     */
1691    nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end.w6);
1692    nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end.w6);
1693
1694    /* NVT cache lookup */
1695    if (xive_router_get_nvt(xrtr, nvt_blk, nvt_idx, &nvt)) {
1696        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: no NVT %x/%x\n",
1697                      nvt_blk, nvt_idx);
1698        return;
1699    }
1700
1701    if (!xive_nvt_is_valid(&nvt)) {
1702        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is invalid\n",
1703                      nvt_blk, nvt_idx);
1704        return;
1705    }
1706
1707    found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
1708                          xive_get_field32(END_W7_F0_IGNORE, end.w7),
1709                          priority,
1710                          xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
1711
1712    /* TODO: Auto EOI. */
1713
1714    if (found) {
1715        return;
1716    }
1717
1718    /*
1719     * If no matching NVT is dispatched on a HW thread :
1720     * - specific VP: update the NVT structure if backlog is activated
1721     * - logical server : forward request to IVPE (not supported)
1722     */
1723    if (xive_end_is_backlog(&end)) {
1724        uint8_t ipb;
1725
1726        if (format == 1) {
1727            qemu_log_mask(LOG_GUEST_ERROR,
1728                          "XIVE: END %x/%x invalid config: F1 & backlog\n",
1729                          end_blk, end_idx);
1730            return;
1731        }
1732        /*
1733         * Record the IPB in the associated NVT structure for later
1734         * use. The presenter will resend the interrupt when the vCPU
1735         * is dispatched again on a HW thread.
1736         */
1737        ipb = xive_get_field32(NVT_W4_IPB, nvt.w4) |
1738            xive_priority_to_ipb(priority);
1739        nvt.w4 = xive_set_field32(NVT_W4_IPB, nvt.w4, ipb);
1740        xive_router_write_nvt(xrtr, nvt_blk, nvt_idx, &nvt, 4);
1741
1742        /*
1743         * On HW, follows a "Broadcast Backlog" to IVPEs
1744         */
1745    }
1746
1747do_escalation:
1748    /*
1749     * If activated, escalate notification using the ESe PQ bits and
1750     * the EAS in w4-5
1751     */
1752    if (!xive_end_is_escalate(&end)) {
1753        return;
1754    }
1755
1756    /*
1757     * Check the END ESe (Event State Buffer for escalation) for even
1758     * further coalescing in the Router
1759     */
1760    if (!xive_end_is_uncond_escalation(&end)) {
1761        /* ESe[Q]=1 : end of notification */
1762        if (!xive_router_end_es_notify(xrtr, end_blk, end_idx,
1763                                       &end, END_W1_ESe)) {
1764            return;
1765        }
1766    }
1767
1768    trace_xive_router_end_escalate(end_blk, end_idx,
1769           (uint8_t) xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1770           (uint32_t) xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1771           (uint32_t) xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1772    /*
1773     * The END trigger becomes an Escalation trigger
1774     */
1775    xive_router_end_notify(xrtr,
1776                           xive_get_field32(END_W4_ESC_END_BLOCK, end.w4),
1777                           xive_get_field32(END_W4_ESC_END_INDEX, end.w4),
1778                           xive_get_field32(END_W5_ESC_END_DATA,  end.w5));
1779}
1780
1781void xive_router_notify(XiveNotifier *xn, uint32_t lisn, bool pq_checked)
1782{
1783    XiveRouter *xrtr = XIVE_ROUTER(xn);
1784    uint8_t eas_blk = XIVE_EAS_BLOCK(lisn);
1785    uint32_t eas_idx = XIVE_EAS_INDEX(lisn);
1786    XiveEAS eas;
1787
1788    /* EAS cache lookup */
1789    if (xive_router_get_eas(xrtr, eas_blk, eas_idx, &eas)) {
1790        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN %x\n", lisn);
1791        return;
1792    }
1793
1794    if (!pq_checked) {
1795        bool notify;
1796        uint8_t pq;
1797
1798        /* PQ cache lookup */
1799        if (xive_router_get_pq(xrtr, eas_blk, eas_idx, &pq)) {
1800            /* Set FIR */
1801            g_assert_not_reached();
1802        }
1803
1804        notify = xive_esb_trigger(&pq);
1805
1806        if (xive_router_set_pq(xrtr, eas_blk, eas_idx, &pq)) {
1807            /* Set FIR */
1808            g_assert_not_reached();
1809        }
1810
1811        if (!notify) {
1812            return;
1813        }
1814    }
1815
1816    if (!xive_eas_is_valid(&eas)) {
1817        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid LISN %x\n", lisn);
1818        return;
1819    }
1820
1821    if (xive_eas_is_masked(&eas)) {
1822        /* Notification completed */
1823        return;
1824    }
1825
1826    /*
1827     * The event trigger becomes an END trigger
1828     */
1829    xive_router_end_notify(xrtr,
1830                           xive_get_field64(EAS_END_BLOCK, eas.w),
1831                           xive_get_field64(EAS_END_INDEX, eas.w),
1832                           xive_get_field64(EAS_END_DATA,  eas.w));
1833}
1834
1835static Property xive_router_properties[] = {
1836    DEFINE_PROP_LINK("xive-fabric", XiveRouter, xfb,
1837                     TYPE_XIVE_FABRIC, XiveFabric *),
1838    DEFINE_PROP_END_OF_LIST(),
1839};
1840
1841static void xive_router_class_init(ObjectClass *klass, void *data)
1842{
1843    DeviceClass *dc = DEVICE_CLASS(klass);
1844    XiveNotifierClass *xnc = XIVE_NOTIFIER_CLASS(klass);
1845
1846    dc->desc    = "XIVE Router Engine";
1847    device_class_set_props(dc, xive_router_properties);
1848    /* Parent is SysBusDeviceClass. No need to call its realize hook */
1849    dc->realize = xive_router_realize;
1850    xnc->notify = xive_router_notify;
1851}
1852
1853static const TypeInfo xive_router_info = {
1854    .name          = TYPE_XIVE_ROUTER,
1855    .parent        = TYPE_SYS_BUS_DEVICE,
1856    .abstract      = true,
1857    .instance_size = sizeof(XiveRouter),
1858    .class_size    = sizeof(XiveRouterClass),
1859    .class_init    = xive_router_class_init,
1860    .interfaces    = (InterfaceInfo[]) {
1861        { TYPE_XIVE_NOTIFIER },
1862        { TYPE_XIVE_PRESENTER },
1863        { }
1864    }
1865};
1866
1867void xive_eas_pic_print_info(XiveEAS *eas, uint32_t lisn, Monitor *mon)
1868{
1869    if (!xive_eas_is_valid(eas)) {
1870        return;
1871    }
1872
1873    monitor_printf(mon, "  %08x %s end:%02x/%04x data:%08x\n",
1874                   lisn, xive_eas_is_masked(eas) ? "M" : " ",
1875                   (uint8_t)  xive_get_field64(EAS_END_BLOCK, eas->w),
1876                   (uint32_t) xive_get_field64(EAS_END_INDEX, eas->w),
1877                   (uint32_t) xive_get_field64(EAS_END_DATA, eas->w));
1878}
1879
1880/*
1881 * END ESB MMIO loads
1882 */
1883static uint64_t xive_end_source_read(void *opaque, hwaddr addr, unsigned size)
1884{
1885    XiveENDSource *xsrc = XIVE_END_SOURCE(opaque);
1886    uint32_t offset = addr & 0xFFF;
1887    uint8_t end_blk;
1888    uint32_t end_idx;
1889    XiveEND end;
1890    uint32_t end_esmask;
1891    uint8_t pq;
1892    uint64_t ret = -1;
1893
1894    /*
1895     * The block id should be deduced from the load address on the END
1896     * ESB MMIO but our model only supports a single block per XIVE chip.
1897     */
1898    end_blk = xive_router_get_block_id(xsrc->xrtr);
1899    end_idx = addr >> (xsrc->esb_shift + 1);
1900
1901    trace_xive_end_source_read(end_blk, end_idx, addr);
1902
1903    if (xive_router_get_end(xsrc->xrtr, end_blk, end_idx, &end)) {
1904        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: No END %x/%x\n", end_blk,
1905                      end_idx);
1906        return -1;
1907    }
1908
1909    if (!xive_end_is_valid(&end)) {
1910        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: END %x/%x is invalid\n",
1911                      end_blk, end_idx);
1912        return -1;
1913    }
1914
1915    end_esmask = addr_is_even(addr, xsrc->esb_shift) ? END_W1_ESn : END_W1_ESe;
1916    pq = xive_get_field32(end_esmask, end.w1);
1917
1918    switch (offset) {
1919    case XIVE_ESB_LOAD_EOI ... XIVE_ESB_LOAD_EOI + 0x7FF:
1920        ret = xive_esb_eoi(&pq);
1921
1922        /* Forward the source event notification for routing ?? */
1923        break;
1924
1925    case XIVE_ESB_GET ... XIVE_ESB_GET + 0x3FF:
1926        ret = pq;
1927        break;
1928
1929    case XIVE_ESB_SET_PQ_00 ... XIVE_ESB_SET_PQ_00 + 0x0FF:
1930    case XIVE_ESB_SET_PQ_01 ... XIVE_ESB_SET_PQ_01 + 0x0FF:
1931    case XIVE_ESB_SET_PQ_10 ... XIVE_ESB_SET_PQ_10 + 0x0FF:
1932    case XIVE_ESB_SET_PQ_11 ... XIVE_ESB_SET_PQ_11 + 0x0FF:
1933        ret = xive_esb_set(&pq, (offset >> 8) & 0x3);
1934        break;
1935    default:
1936        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid END ESB load addr %d\n",
1937                      offset);
1938        return -1;
1939    }
1940
1941    if (pq != xive_get_field32(end_esmask, end.w1)) {
1942        end.w1 = xive_set_field32(end_esmask, end.w1, pq);
1943        xive_router_write_end(xsrc->xrtr, end_blk, end_idx, &end, 1);
1944    }
1945
1946    return ret;
1947}
1948
1949/*
1950 * END ESB MMIO stores are invalid
1951 */
1952static void xive_end_source_write(void *opaque, hwaddr addr,
1953                                  uint64_t value, unsigned size)
1954{
1955    qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid ESB write addr 0x%"
1956                  HWADDR_PRIx"\n", addr);
1957}
1958
1959static const MemoryRegionOps xive_end_source_ops = {
1960    .read = xive_end_source_read,
1961    .write = xive_end_source_write,
1962    .endianness = DEVICE_BIG_ENDIAN,
1963    .valid = {
1964        .min_access_size = 8,
1965        .max_access_size = 8,
1966    },
1967    .impl = {
1968        .min_access_size = 8,
1969        .max_access_size = 8,
1970    },
1971};
1972
1973static void xive_end_source_realize(DeviceState *dev, Error **errp)
1974{
1975    XiveENDSource *xsrc = XIVE_END_SOURCE(dev);
1976
1977    assert(xsrc->xrtr);
1978
1979    if (!xsrc->nr_ends) {
1980        error_setg(errp, "Number of interrupt needs to be greater than 0");
1981        return;
1982    }
1983
1984    if (xsrc->esb_shift != XIVE_ESB_4K &&
1985        xsrc->esb_shift != XIVE_ESB_64K) {
1986        error_setg(errp, "Invalid ESB shift setting");
1987        return;
1988    }
1989
1990    /*
1991     * Each END is assigned an even/odd pair of MMIO pages, the even page
1992     * manages the ESn field while the odd page manages the ESe field.
1993     */
1994    memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
1995                          &xive_end_source_ops, xsrc, "xive.end",
1996                          (1ull << (xsrc->esb_shift + 1)) * xsrc->nr_ends);
1997}
1998
1999static Property xive_end_source_properties[] = {
2000    DEFINE_PROP_UINT32("nr-ends", XiveENDSource, nr_ends, 0),
2001    DEFINE_PROP_UINT32("shift", XiveENDSource, esb_shift, XIVE_ESB_64K),
2002    DEFINE_PROP_LINK("xive", XiveENDSource, xrtr, TYPE_XIVE_ROUTER,
2003                     XiveRouter *),
2004    DEFINE_PROP_END_OF_LIST(),
2005};
2006
2007static void xive_end_source_class_init(ObjectClass *klass, void *data)
2008{
2009    DeviceClass *dc = DEVICE_CLASS(klass);
2010
2011    dc->desc    = "XIVE END Source";
2012    device_class_set_props(dc, xive_end_source_properties);
2013    dc->realize = xive_end_source_realize;
2014    /*
2015     * Reason: part of XIVE interrupt controller, needs to be wired up,
2016     * e.g. by spapr_xive_instance_init().
2017     */
2018    dc->user_creatable = false;
2019}
2020
2021static const TypeInfo xive_end_source_info = {
2022    .name          = TYPE_XIVE_END_SOURCE,
2023    .parent        = TYPE_DEVICE,
2024    .instance_size = sizeof(XiveENDSource),
2025    .class_init    = xive_end_source_class_init,
2026};
2027
2028/*
2029 * XIVE Notifier
2030 */
2031static const TypeInfo xive_notifier_info = {
2032    .name = TYPE_XIVE_NOTIFIER,
2033    .parent = TYPE_INTERFACE,
2034    .class_size = sizeof(XiveNotifierClass),
2035};
2036
2037/*
2038 * XIVE Presenter
2039 */
2040static const TypeInfo xive_presenter_info = {
2041    .name = TYPE_XIVE_PRESENTER,
2042    .parent = TYPE_INTERFACE,
2043    .class_size = sizeof(XivePresenterClass),
2044};
2045
2046/*
2047 * XIVE Fabric
2048 */
2049static const TypeInfo xive_fabric_info = {
2050    .name = TYPE_XIVE_FABRIC,
2051    .parent = TYPE_INTERFACE,
2052    .class_size = sizeof(XiveFabricClass),
2053};
2054
2055static void xive_register_types(void)
2056{
2057    type_register_static(&xive_fabric_info);
2058    type_register_static(&xive_source_info);
2059    type_register_static(&xive_notifier_info);
2060    type_register_static(&xive_presenter_info);
2061    type_register_static(&xive_router_info);
2062    type_register_static(&xive_end_source_info);
2063    type_register_static(&xive_tctx_info);
2064}
2065
2066type_init(xive_register_types)
2067