qemu/hw/intc/spapr_xive.c
<<
>>
Prefs
   1/*
   2 * QEMU PowerPC sPAPR 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 "qemu/error-report.h"
  15#include "target/ppc/cpu.h"
  16#include "sysemu/cpus.h"
  17#include "sysemu/reset.h"
  18#include "migration/vmstate.h"
  19#include "monitor/monitor.h"
  20#include "hw/ppc/fdt.h"
  21#include "hw/ppc/spapr.h"
  22#include "hw/ppc/spapr_cpu_core.h"
  23#include "hw/ppc/spapr_xive.h"
  24#include "hw/ppc/xive.h"
  25#include "hw/ppc/xive_regs.h"
  26#include "hw/qdev-properties.h"
  27
  28/*
  29 * XIVE Virtualization Controller BAR and Thread Managment BAR that we
  30 * use for the ESB pages and the TIMA pages
  31 */
  32#define SPAPR_XIVE_VC_BASE   0x0006010000000000ull
  33#define SPAPR_XIVE_TM_BASE   0x0006030203180000ull
  34
  35/*
  36 * The allocation of VP blocks is a complex operation in OPAL and the
  37 * VP identifiers have a relation with the number of HW chips, the
  38 * size of the VP blocks, VP grouping, etc. The QEMU sPAPR XIVE
  39 * controller model does not have the same constraints and can use a
  40 * simple mapping scheme of the CPU vcpu_id
  41 *
  42 * These identifiers are never returned to the OS.
  43 */
  44
  45#define SPAPR_XIVE_NVT_BASE 0x400
  46
  47/*
  48 * sPAPR NVT and END indexing helpers
  49 */
  50static uint32_t spapr_xive_nvt_to_target(uint8_t nvt_blk, uint32_t nvt_idx)
  51{
  52    return nvt_idx - SPAPR_XIVE_NVT_BASE;
  53}
  54
  55static void spapr_xive_cpu_to_nvt(PowerPCCPU *cpu,
  56                                  uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
  57{
  58    assert(cpu);
  59
  60    if (out_nvt_blk) {
  61        *out_nvt_blk = SPAPR_XIVE_BLOCK_ID;
  62    }
  63
  64    if (out_nvt_blk) {
  65        *out_nvt_idx = SPAPR_XIVE_NVT_BASE + cpu->vcpu_id;
  66    }
  67}
  68
  69static int spapr_xive_target_to_nvt(uint32_t target,
  70                                    uint8_t *out_nvt_blk, uint32_t *out_nvt_idx)
  71{
  72    PowerPCCPU *cpu = spapr_find_cpu(target);
  73
  74    if (!cpu) {
  75        return -1;
  76    }
  77
  78    spapr_xive_cpu_to_nvt(cpu, out_nvt_blk, out_nvt_idx);
  79    return 0;
  80}
  81
  82/*
  83 * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
  84 * priorities per CPU
  85 */
  86int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
  87                             uint32_t *out_server, uint8_t *out_prio)
  88{
  89
  90    assert(end_blk == SPAPR_XIVE_BLOCK_ID);
  91
  92    if (out_server) {
  93        *out_server = end_idx >> 3;
  94    }
  95
  96    if (out_prio) {
  97        *out_prio = end_idx & 0x7;
  98    }
  99    return 0;
 100}
 101
 102static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
 103                                  uint8_t *out_end_blk, uint32_t *out_end_idx)
 104{
 105    assert(cpu);
 106
 107    if (out_end_blk) {
 108        *out_end_blk = SPAPR_XIVE_BLOCK_ID;
 109    }
 110
 111    if (out_end_idx) {
 112        *out_end_idx = (cpu->vcpu_id << 3) + prio;
 113    }
 114}
 115
 116static int spapr_xive_target_to_end(uint32_t target, uint8_t prio,
 117                                    uint8_t *out_end_blk, uint32_t *out_end_idx)
 118{
 119    PowerPCCPU *cpu = spapr_find_cpu(target);
 120
 121    if (!cpu) {
 122        return -1;
 123    }
 124
 125    spapr_xive_cpu_to_end(cpu, prio, out_end_blk, out_end_idx);
 126    return 0;
 127}
 128
 129/*
 130 * On sPAPR machines, use a simplified output for the XIVE END
 131 * structure dumping only the information related to the OS EQ.
 132 */
 133static void spapr_xive_end_pic_print_info(SpaprXive *xive, XiveEND *end,
 134                                          Monitor *mon)
 135{
 136    uint64_t qaddr_base = xive_end_qaddr(end);
 137    uint32_t qindex = xive_get_field32(END_W1_PAGE_OFF, end->w1);
 138    uint32_t qgen = xive_get_field32(END_W1_GENERATION, end->w1);
 139    uint32_t qsize = xive_get_field32(END_W0_QSIZE, end->w0);
 140    uint32_t qentries = 1 << (qsize + 10);
 141    uint32_t nvt = xive_get_field32(END_W6_NVT_INDEX, end->w6);
 142    uint8_t priority = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
 143
 144    monitor_printf(mon, "%3d/%d % 6d/%5d @%"PRIx64" ^%d",
 145                   spapr_xive_nvt_to_target(0, nvt),
 146                   priority, qindex, qentries, qaddr_base, qgen);
 147
 148    xive_end_queue_pic_print_info(end, 6, mon);
 149}
 150
 151void spapr_xive_pic_print_info(SpaprXive *xive, Monitor *mon)
 152{
 153    XiveSource *xsrc = &xive->source;
 154    int i;
 155
 156    if (kvm_irqchip_in_kernel()) {
 157        Error *local_err = NULL;
 158
 159        kvmppc_xive_synchronize_state(xive, &local_err);
 160        if (local_err) {
 161            error_report_err(local_err);
 162            return;
 163        }
 164    }
 165
 166    monitor_printf(mon, "  LISN         PQ    EISN     CPU/PRIO EQ\n");
 167
 168    for (i = 0; i < xive->nr_irqs; i++) {
 169        uint8_t pq = xive_source_esb_get(xsrc, i);
 170        XiveEAS *eas = &xive->eat[i];
 171
 172        if (!xive_eas_is_valid(eas)) {
 173            continue;
 174        }
 175
 176        monitor_printf(mon, "  %08x %s %c%c%c %s %08x ", i,
 177                       xive_source_irq_is_lsi(xsrc, i) ? "LSI" : "MSI",
 178                       pq & XIVE_ESB_VAL_P ? 'P' : '-',
 179                       pq & XIVE_ESB_VAL_Q ? 'Q' : '-',
 180                       xsrc->status[i] & XIVE_STATUS_ASSERTED ? 'A' : ' ',
 181                       xive_eas_is_masked(eas) ? "M" : " ",
 182                       (int) xive_get_field64(EAS_END_DATA, eas->w));
 183
 184        if (!xive_eas_is_masked(eas)) {
 185            uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
 186            XiveEND *end;
 187
 188            assert(end_idx < xive->nr_ends);
 189            end = &xive->endt[end_idx];
 190
 191            if (xive_end_is_valid(end)) {
 192                spapr_xive_end_pic_print_info(xive, end, mon);
 193            }
 194        }
 195        monitor_printf(mon, "\n");
 196    }
 197}
 198
 199void spapr_xive_mmio_set_enabled(SpaprXive *xive, bool enable)
 200{
 201    memory_region_set_enabled(&xive->source.esb_mmio, enable);
 202    memory_region_set_enabled(&xive->tm_mmio, enable);
 203
 204    /* Disable the END ESBs until a guest OS makes use of them */
 205    memory_region_set_enabled(&xive->end_source.esb_mmio, false);
 206}
 207
 208static void spapr_xive_end_reset(XiveEND *end)
 209{
 210    memset(end, 0, sizeof(*end));
 211
 212    /* switch off the escalation and notification ESBs */
 213    end->w1 = cpu_to_be32(END_W1_ESe_Q | END_W1_ESn_Q);
 214}
 215
 216static void spapr_xive_reset(void *dev)
 217{
 218    SpaprXive *xive = SPAPR_XIVE(dev);
 219    int i;
 220
 221    /*
 222     * The XiveSource has its own reset handler, which mask off all
 223     * IRQs (!P|Q)
 224     */
 225
 226    /* Mask all valid EASs in the IRQ number space. */
 227    for (i = 0; i < xive->nr_irqs; i++) {
 228        XiveEAS *eas = &xive->eat[i];
 229        if (xive_eas_is_valid(eas)) {
 230            eas->w = cpu_to_be64(EAS_VALID | EAS_MASKED);
 231        } else {
 232            eas->w = 0;
 233        }
 234    }
 235
 236    /* Clear all ENDs */
 237    for (i = 0; i < xive->nr_ends; i++) {
 238        spapr_xive_end_reset(&xive->endt[i]);
 239    }
 240}
 241
 242static void spapr_xive_instance_init(Object *obj)
 243{
 244    SpaprXive *xive = SPAPR_XIVE(obj);
 245
 246    object_initialize_child(obj, "source", &xive->source, sizeof(xive->source),
 247                            TYPE_XIVE_SOURCE, &error_abort, NULL);
 248
 249    object_initialize_child(obj, "end_source", &xive->end_source,
 250                            sizeof(xive->end_source), TYPE_XIVE_END_SOURCE,
 251                            &error_abort, NULL);
 252
 253    /* Not connected to the KVM XIVE device */
 254    xive->fd = -1;
 255}
 256
 257static void spapr_xive_realize(DeviceState *dev, Error **errp)
 258{
 259    SpaprXive *xive = SPAPR_XIVE(dev);
 260    XiveSource *xsrc = &xive->source;
 261    XiveENDSource *end_xsrc = &xive->end_source;
 262    Error *local_err = NULL;
 263
 264    if (!xive->nr_irqs) {
 265        error_setg(errp, "Number of interrupt needs to be greater 0");
 266        return;
 267    }
 268
 269    if (!xive->nr_ends) {
 270        error_setg(errp, "Number of interrupt needs to be greater 0");
 271        return;
 272    }
 273
 274    /*
 275     * Initialize the internal sources, for IPIs and virtual devices.
 276     */
 277    object_property_set_int(OBJECT(xsrc), xive->nr_irqs, "nr-irqs",
 278                            &error_fatal);
 279    object_property_add_const_link(OBJECT(xsrc), "xive", OBJECT(xive),
 280                                   &error_fatal);
 281    object_property_set_bool(OBJECT(xsrc), true, "realized", &local_err);
 282    if (local_err) {
 283        error_propagate(errp, local_err);
 284        return;
 285    }
 286    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
 287
 288    /*
 289     * Initialize the END ESB source
 290     */
 291    object_property_set_int(OBJECT(end_xsrc), xive->nr_irqs, "nr-ends",
 292                            &error_fatal);
 293    object_property_add_const_link(OBJECT(end_xsrc), "xive", OBJECT(xive),
 294                                   &error_fatal);
 295    object_property_set_bool(OBJECT(end_xsrc), true, "realized", &local_err);
 296    if (local_err) {
 297        error_propagate(errp, local_err);
 298        return;
 299    }
 300    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
 301
 302    /* Set the mapping address of the END ESB pages after the source ESBs */
 303    xive->end_base = xive->vc_base + (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
 304
 305    /*
 306     * Allocate the routing tables
 307     */
 308    xive->eat = g_new0(XiveEAS, xive->nr_irqs);
 309    xive->endt = g_new0(XiveEND, xive->nr_ends);
 310
 311    xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
 312                           xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
 313
 314    qemu_register_reset(spapr_xive_reset, dev);
 315
 316    /* TIMA initialization */
 317    memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
 318                          "xive.tima", 4ull << TM_SHIFT);
 319    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
 320
 321    /*
 322     * Map all regions. These will be enabled or disabled at reset and
 323     * can also be overridden by KVM memory regions if active
 324     */
 325    sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
 326    sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
 327    sysbus_mmio_map(SYS_BUS_DEVICE(xive), 2, xive->tm_base);
 328}
 329
 330static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
 331                              uint32_t eas_idx, XiveEAS *eas)
 332{
 333    SpaprXive *xive = SPAPR_XIVE(xrtr);
 334
 335    if (eas_idx >= xive->nr_irqs) {
 336        return -1;
 337    }
 338
 339    *eas = xive->eat[eas_idx];
 340    return 0;
 341}
 342
 343static int spapr_xive_get_end(XiveRouter *xrtr,
 344                              uint8_t end_blk, uint32_t end_idx, XiveEND *end)
 345{
 346    SpaprXive *xive = SPAPR_XIVE(xrtr);
 347
 348    if (end_idx >= xive->nr_ends) {
 349        return -1;
 350    }
 351
 352    memcpy(end, &xive->endt[end_idx], sizeof(XiveEND));
 353    return 0;
 354}
 355
 356static int spapr_xive_write_end(XiveRouter *xrtr, uint8_t end_blk,
 357                                uint32_t end_idx, XiveEND *end,
 358                                uint8_t word_number)
 359{
 360    SpaprXive *xive = SPAPR_XIVE(xrtr);
 361
 362    if (end_idx >= xive->nr_ends) {
 363        return -1;
 364    }
 365
 366    memcpy(&xive->endt[end_idx], end, sizeof(XiveEND));
 367    return 0;
 368}
 369
 370static int spapr_xive_get_nvt(XiveRouter *xrtr,
 371                              uint8_t nvt_blk, uint32_t nvt_idx, XiveNVT *nvt)
 372{
 373    uint32_t vcpu_id = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
 374    PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
 375
 376    if (!cpu) {
 377        /* TODO: should we assert() if we can find a NVT ? */
 378        return -1;
 379    }
 380
 381    /*
 382     * sPAPR does not maintain a NVT table. Return that the NVT is
 383     * valid if we have found a matching CPU
 384     */
 385    nvt->w0 = cpu_to_be32(NVT_W0_VALID);
 386    return 0;
 387}
 388
 389static int spapr_xive_write_nvt(XiveRouter *xrtr, uint8_t nvt_blk,
 390                                uint32_t nvt_idx, XiveNVT *nvt,
 391                                uint8_t word_number)
 392{
 393    /*
 394     * We don't need to write back to the NVTs because the sPAPR
 395     * machine should never hit a non-scheduled NVT. It should never
 396     * get called.
 397     */
 398    g_assert_not_reached();
 399}
 400
 401static XiveTCTX *spapr_xive_get_tctx(XiveRouter *xrtr, CPUState *cs)
 402{
 403    PowerPCCPU *cpu = POWERPC_CPU(cs);
 404
 405    return spapr_cpu_state(cpu)->tctx;
 406}
 407
 408static const VMStateDescription vmstate_spapr_xive_end = {
 409    .name = TYPE_SPAPR_XIVE "/end",
 410    .version_id = 1,
 411    .minimum_version_id = 1,
 412    .fields = (VMStateField []) {
 413        VMSTATE_UINT32(w0, XiveEND),
 414        VMSTATE_UINT32(w1, XiveEND),
 415        VMSTATE_UINT32(w2, XiveEND),
 416        VMSTATE_UINT32(w3, XiveEND),
 417        VMSTATE_UINT32(w4, XiveEND),
 418        VMSTATE_UINT32(w5, XiveEND),
 419        VMSTATE_UINT32(w6, XiveEND),
 420        VMSTATE_UINT32(w7, XiveEND),
 421        VMSTATE_END_OF_LIST()
 422    },
 423};
 424
 425static const VMStateDescription vmstate_spapr_xive_eas = {
 426    .name = TYPE_SPAPR_XIVE "/eas",
 427    .version_id = 1,
 428    .minimum_version_id = 1,
 429    .fields = (VMStateField []) {
 430        VMSTATE_UINT64(w, XiveEAS),
 431        VMSTATE_END_OF_LIST()
 432    },
 433};
 434
 435static int vmstate_spapr_xive_pre_save(void *opaque)
 436{
 437    if (kvm_irqchip_in_kernel()) {
 438        return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
 439    }
 440
 441    return 0;
 442}
 443
 444/*
 445 * Called by the sPAPR IRQ backend 'post_load' method at the machine
 446 * level.
 447 */
 448static int spapr_xive_post_load(SpaprInterruptController *intc, int version_id)
 449{
 450    if (kvm_irqchip_in_kernel()) {
 451        return kvmppc_xive_post_load(SPAPR_XIVE(intc), version_id);
 452    }
 453
 454    return 0;
 455}
 456
 457static const VMStateDescription vmstate_spapr_xive = {
 458    .name = TYPE_SPAPR_XIVE,
 459    .version_id = 1,
 460    .minimum_version_id = 1,
 461    .pre_save = vmstate_spapr_xive_pre_save,
 462    .post_load = NULL, /* handled at the machine level */
 463    .fields = (VMStateField[]) {
 464        VMSTATE_UINT32_EQUAL(nr_irqs, SpaprXive, NULL),
 465        VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, SpaprXive, nr_irqs,
 466                                     vmstate_spapr_xive_eas, XiveEAS),
 467        VMSTATE_STRUCT_VARRAY_POINTER_UINT32(endt, SpaprXive, nr_ends,
 468                                             vmstate_spapr_xive_end, XiveEND),
 469        VMSTATE_END_OF_LIST()
 470    },
 471};
 472
 473static int spapr_xive_claim_irq(SpaprInterruptController *intc, int lisn,
 474                                bool lsi, Error **errp)
 475{
 476    SpaprXive *xive = SPAPR_XIVE(intc);
 477    XiveSource *xsrc = &xive->source;
 478
 479    assert(lisn < xive->nr_irqs);
 480
 481    if (xive_eas_is_valid(&xive->eat[lisn])) {
 482        error_setg(errp, "IRQ %d is not free", lisn);
 483        return -EBUSY;
 484    }
 485
 486    /*
 487     * Set default values when allocating an IRQ number
 488     */
 489    xive->eat[lisn].w |= cpu_to_be64(EAS_VALID | EAS_MASKED);
 490    if (lsi) {
 491        xive_source_irq_set_lsi(xsrc, lisn);
 492    }
 493
 494    if (kvm_irqchip_in_kernel()) {
 495        return kvmppc_xive_source_reset_one(xsrc, lisn, errp);
 496    }
 497
 498    return 0;
 499}
 500
 501static void spapr_xive_free_irq(SpaprInterruptController *intc, int lisn)
 502{
 503    SpaprXive *xive = SPAPR_XIVE(intc);
 504    assert(lisn < xive->nr_irqs);
 505
 506    xive->eat[lisn].w &= cpu_to_be64(~EAS_VALID);
 507}
 508
 509static Property spapr_xive_properties[] = {
 510    DEFINE_PROP_UINT32("nr-irqs", SpaprXive, nr_irqs, 0),
 511    DEFINE_PROP_UINT32("nr-ends", SpaprXive, nr_ends, 0),
 512    DEFINE_PROP_UINT64("vc-base", SpaprXive, vc_base, SPAPR_XIVE_VC_BASE),
 513    DEFINE_PROP_UINT64("tm-base", SpaprXive, tm_base, SPAPR_XIVE_TM_BASE),
 514    DEFINE_PROP_END_OF_LIST(),
 515};
 516
 517static int spapr_xive_cpu_intc_create(SpaprInterruptController *intc,
 518                                      PowerPCCPU *cpu, Error **errp)
 519{
 520    SpaprXive *xive = SPAPR_XIVE(intc);
 521    Object *obj;
 522    SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
 523
 524    obj = xive_tctx_create(OBJECT(cpu), XIVE_ROUTER(xive), errp);
 525    if (!obj) {
 526        return -1;
 527    }
 528
 529    spapr_cpu->tctx = XIVE_TCTX(obj);
 530    return 0;
 531}
 532
 533static void xive_tctx_set_os_cam(XiveTCTX *tctx, uint32_t os_cam)
 534{
 535    uint32_t qw1w2 = cpu_to_be32(TM_QW1W2_VO | os_cam);
 536    memcpy(&tctx->regs[TM_QW1_OS + TM_WORD2], &qw1w2, 4);
 537}
 538
 539static void spapr_xive_cpu_intc_reset(SpaprInterruptController *intc,
 540                                     PowerPCCPU *cpu)
 541{
 542    XiveTCTX *tctx = spapr_cpu_state(cpu)->tctx;
 543    uint8_t  nvt_blk;
 544    uint32_t nvt_idx;
 545
 546    xive_tctx_reset(tctx);
 547
 548    /*
 549     * When a Virtual Processor is scheduled to run on a HW thread,
 550     * the hypervisor pushes its identifier in the OS CAM line.
 551     * Emulate the same behavior under QEMU.
 552     */
 553    spapr_xive_cpu_to_nvt(cpu, &nvt_blk, &nvt_idx);
 554
 555    xive_tctx_set_os_cam(tctx, xive_nvt_cam_line(nvt_blk, nvt_idx));
 556}
 557
 558static void spapr_xive_cpu_intc_destroy(SpaprInterruptController *intc,
 559                                        PowerPCCPU *cpu)
 560{
 561    SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
 562
 563    xive_tctx_destroy(spapr_cpu->tctx);
 564    spapr_cpu->tctx = NULL;
 565}
 566
 567static void spapr_xive_set_irq(SpaprInterruptController *intc, int irq, int val)
 568{
 569    SpaprXive *xive = SPAPR_XIVE(intc);
 570
 571    if (kvm_irqchip_in_kernel()) {
 572        kvmppc_xive_source_set_irq(&xive->source, irq, val);
 573    } else {
 574        xive_source_set_irq(&xive->source, irq, val);
 575    }
 576}
 577
 578static void spapr_xive_print_info(SpaprInterruptController *intc, Monitor *mon)
 579{
 580    SpaprXive *xive = SPAPR_XIVE(intc);
 581    CPUState *cs;
 582
 583    CPU_FOREACH(cs) {
 584        PowerPCCPU *cpu = POWERPC_CPU(cs);
 585
 586        xive_tctx_pic_print_info(spapr_cpu_state(cpu)->tctx, mon);
 587    }
 588
 589    spapr_xive_pic_print_info(xive, mon);
 590}
 591
 592static void spapr_xive_dt(SpaprInterruptController *intc, uint32_t nr_servers,
 593                          void *fdt, uint32_t phandle)
 594{
 595    SpaprXive *xive = SPAPR_XIVE(intc);
 596    int node;
 597    uint64_t timas[2 * 2];
 598    /* Interrupt number ranges for the IPIs */
 599    uint32_t lisn_ranges[] = {
 600        cpu_to_be32(0),
 601        cpu_to_be32(nr_servers),
 602    };
 603    /*
 604     * EQ size - the sizes of pages supported by the system 4K, 64K,
 605     * 2M, 16M. We only advertise 64K for the moment.
 606     */
 607    uint32_t eq_sizes[] = {
 608        cpu_to_be32(16), /* 64K */
 609    };
 610    /*
 611     * The following array is in sync with the reserved priorities
 612     * defined by the 'spapr_xive_priority_is_reserved' routine.
 613     */
 614    uint32_t plat_res_int_priorities[] = {
 615        cpu_to_be32(7),    /* start */
 616        cpu_to_be32(0xf8), /* count */
 617    };
 618
 619    /* Thread Interrupt Management Area : User (ring 3) and OS (ring 2) */
 620    timas[0] = cpu_to_be64(xive->tm_base +
 621                           XIVE_TM_USER_PAGE * (1ull << TM_SHIFT));
 622    timas[1] = cpu_to_be64(1ull << TM_SHIFT);
 623    timas[2] = cpu_to_be64(xive->tm_base +
 624                           XIVE_TM_OS_PAGE * (1ull << TM_SHIFT));
 625    timas[3] = cpu_to_be64(1ull << TM_SHIFT);
 626
 627    _FDT(node = fdt_add_subnode(fdt, 0, xive->nodename));
 628
 629    _FDT(fdt_setprop_string(fdt, node, "device_type", "power-ivpe"));
 630    _FDT(fdt_setprop(fdt, node, "reg", timas, sizeof(timas)));
 631
 632    _FDT(fdt_setprop_string(fdt, node, "compatible", "ibm,power-ivpe"));
 633    _FDT(fdt_setprop(fdt, node, "ibm,xive-eq-sizes", eq_sizes,
 634                     sizeof(eq_sizes)));
 635    _FDT(fdt_setprop(fdt, node, "ibm,xive-lisn-ranges", lisn_ranges,
 636                     sizeof(lisn_ranges)));
 637
 638    /* For Linux to link the LSIs to the interrupt controller. */
 639    _FDT(fdt_setprop(fdt, node, "interrupt-controller", NULL, 0));
 640    _FDT(fdt_setprop_cell(fdt, node, "#interrupt-cells", 2));
 641
 642    /* For SLOF */
 643    _FDT(fdt_setprop_cell(fdt, node, "linux,phandle", phandle));
 644    _FDT(fdt_setprop_cell(fdt, node, "phandle", phandle));
 645
 646    /*
 647     * The "ibm,plat-res-int-priorities" property defines the priority
 648     * ranges reserved by the hypervisor
 649     */
 650    _FDT(fdt_setprop(fdt, 0, "ibm,plat-res-int-priorities",
 651                     plat_res_int_priorities, sizeof(plat_res_int_priorities)));
 652}
 653
 654static int spapr_xive_activate(SpaprInterruptController *intc, Error **errp)
 655{
 656    SpaprXive *xive = SPAPR_XIVE(intc);
 657
 658    if (kvm_enabled()) {
 659        int rc = spapr_irq_init_kvm(kvmppc_xive_connect, intc, errp);
 660        if (rc < 0) {
 661            return rc;
 662        }
 663    }
 664
 665    /* Activate the XIVE MMIOs */
 666    spapr_xive_mmio_set_enabled(xive, true);
 667
 668    return 0;
 669}
 670
 671static void spapr_xive_deactivate(SpaprInterruptController *intc)
 672{
 673    SpaprXive *xive = SPAPR_XIVE(intc);
 674
 675    spapr_xive_mmio_set_enabled(xive, false);
 676
 677    if (kvm_irqchip_in_kernel()) {
 678        kvmppc_xive_disconnect(intc);
 679    }
 680}
 681
 682static void spapr_xive_class_init(ObjectClass *klass, void *data)
 683{
 684    DeviceClass *dc = DEVICE_CLASS(klass);
 685    XiveRouterClass *xrc = XIVE_ROUTER_CLASS(klass);
 686    SpaprInterruptControllerClass *sicc = SPAPR_INTC_CLASS(klass);
 687
 688    dc->desc    = "sPAPR XIVE Interrupt Controller";
 689    dc->props   = spapr_xive_properties;
 690    dc->realize = spapr_xive_realize;
 691    dc->vmsd    = &vmstate_spapr_xive;
 692
 693    xrc->get_eas = spapr_xive_get_eas;
 694    xrc->get_end = spapr_xive_get_end;
 695    xrc->write_end = spapr_xive_write_end;
 696    xrc->get_nvt = spapr_xive_get_nvt;
 697    xrc->write_nvt = spapr_xive_write_nvt;
 698    xrc->get_tctx = spapr_xive_get_tctx;
 699
 700    sicc->activate = spapr_xive_activate;
 701    sicc->deactivate = spapr_xive_deactivate;
 702    sicc->cpu_intc_create = spapr_xive_cpu_intc_create;
 703    sicc->cpu_intc_reset = spapr_xive_cpu_intc_reset;
 704    sicc->cpu_intc_destroy = spapr_xive_cpu_intc_destroy;
 705    sicc->claim_irq = spapr_xive_claim_irq;
 706    sicc->free_irq = spapr_xive_free_irq;
 707    sicc->set_irq = spapr_xive_set_irq;
 708    sicc->print_info = spapr_xive_print_info;
 709    sicc->dt = spapr_xive_dt;
 710    sicc->post_load = spapr_xive_post_load;
 711}
 712
 713static const TypeInfo spapr_xive_info = {
 714    .name = TYPE_SPAPR_XIVE,
 715    .parent = TYPE_XIVE_ROUTER,
 716    .instance_init = spapr_xive_instance_init,
 717    .instance_size = sizeof(SpaprXive),
 718    .class_init = spapr_xive_class_init,
 719    .interfaces = (InterfaceInfo[]) {
 720        { TYPE_SPAPR_INTC },
 721        { }
 722    },
 723};
 724
 725static void spapr_xive_register_types(void)
 726{
 727    type_register_static(&spapr_xive_info);
 728}
 729
 730type_init(spapr_xive_register_types)
 731
 732/*
 733 * XIVE hcalls
 734 *
 735 * The terminology used by the XIVE hcalls is the following :
 736 *
 737 *   TARGET vCPU number
 738 *   EQ     Event Queue assigned by OS to receive event data
 739 *   ESB    page for source interrupt management
 740 *   LISN   Logical Interrupt Source Number identifying a source in the
 741 *          machine
 742 *   EISN   Effective Interrupt Source Number used by guest OS to
 743 *          identify source in the guest
 744 *
 745 * The EAS, END, NVT structures are not exposed.
 746 */
 747
 748/*
 749 * Linux hosts under OPAL reserve priority 7 for their own escalation
 750 * interrupts (DD2.X POWER9). So we only allow the guest to use
 751 * priorities [0..6].
 752 */
 753static bool spapr_xive_priority_is_reserved(uint8_t priority)
 754{
 755    switch (priority) {
 756    case 0 ... 6:
 757        return false;
 758    case 7: /* OPAL escalation queue */
 759    default:
 760        return true;
 761    }
 762}
 763
 764/*
 765 * The H_INT_GET_SOURCE_INFO hcall() is used to obtain the logical
 766 * real address of the MMIO page through which the Event State Buffer
 767 * entry associated with the value of the "lisn" parameter is managed.
 768 *
 769 * Parameters:
 770 * Input
 771 * - R4: "flags"
 772 *         Bits 0-63 reserved
 773 * - R5: "lisn" is per "interrupts", "interrupt-map", or
 774 *       "ibm,xive-lisn-ranges" properties, or as returned by the
 775 *       ibm,query-interrupt-source-number RTAS call, or as returned
 776 *       by the H_ALLOCATE_VAS_WINDOW hcall
 777 *
 778 * Output
 779 * - R4: "flags"
 780 *         Bits 0-59: Reserved
 781 *         Bit 60: H_INT_ESB must be used for Event State Buffer
 782 *                 management
 783 *         Bit 61: 1 == LSI  0 == MSI
 784 *         Bit 62: the full function page supports trigger
 785 *         Bit 63: Store EOI Supported
 786 * - R5: Logical Real address of full function Event State Buffer
 787 *       management page, -1 if H_INT_ESB hcall flag is set to 1.
 788 * - R6: Logical Real Address of trigger only Event State Buffer
 789 *       management page or -1.
 790 * - R7: Power of 2 page size for the ESB management pages returned in
 791 *       R5 and R6.
 792 */
 793
 794#define SPAPR_XIVE_SRC_H_INT_ESB     PPC_BIT(60) /* ESB manage with H_INT_ESB */
 795#define SPAPR_XIVE_SRC_LSI           PPC_BIT(61) /* Virtual LSI type */
 796#define SPAPR_XIVE_SRC_TRIGGER       PPC_BIT(62) /* Trigger and management
 797                                                    on same page */
 798#define SPAPR_XIVE_SRC_STORE_EOI     PPC_BIT(63) /* Store EOI support */
 799
 800static target_ulong h_int_get_source_info(PowerPCCPU *cpu,
 801                                          SpaprMachineState *spapr,
 802                                          target_ulong opcode,
 803                                          target_ulong *args)
 804{
 805    SpaprXive *xive = spapr->xive;
 806    XiveSource *xsrc = &xive->source;
 807    target_ulong flags  = args[0];
 808    target_ulong lisn   = args[1];
 809
 810    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
 811        return H_FUNCTION;
 812    }
 813
 814    if (flags) {
 815        return H_PARAMETER;
 816    }
 817
 818    if (lisn >= xive->nr_irqs) {
 819        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
 820                      lisn);
 821        return H_P2;
 822    }
 823
 824    if (!xive_eas_is_valid(&xive->eat[lisn])) {
 825        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
 826                      lisn);
 827        return H_P2;
 828    }
 829
 830    /*
 831     * All sources are emulated under the main XIVE object and share
 832     * the same characteristics.
 833     */
 834    args[0] = 0;
 835    if (!xive_source_esb_has_2page(xsrc)) {
 836        args[0] |= SPAPR_XIVE_SRC_TRIGGER;
 837    }
 838    if (xsrc->esb_flags & XIVE_SRC_STORE_EOI) {
 839        args[0] |= SPAPR_XIVE_SRC_STORE_EOI;
 840    }
 841
 842    /*
 843     * Force the use of the H_INT_ESB hcall in case of an LSI
 844     * interrupt. This is necessary under KVM to re-trigger the
 845     * interrupt if the level is still asserted
 846     */
 847    if (xive_source_irq_is_lsi(xsrc, lisn)) {
 848        args[0] |= SPAPR_XIVE_SRC_H_INT_ESB | SPAPR_XIVE_SRC_LSI;
 849    }
 850
 851    if (!(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
 852        args[1] = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn);
 853    } else {
 854        args[1] = -1;
 855    }
 856
 857    if (xive_source_esb_has_2page(xsrc) &&
 858        !(args[0] & SPAPR_XIVE_SRC_H_INT_ESB)) {
 859        args[2] = xive->vc_base + xive_source_esb_page(xsrc, lisn);
 860    } else {
 861        args[2] = -1;
 862    }
 863
 864    if (xive_source_esb_has_2page(xsrc)) {
 865        args[3] = xsrc->esb_shift - 1;
 866    } else {
 867        args[3] = xsrc->esb_shift;
 868    }
 869
 870    return H_SUCCESS;
 871}
 872
 873/*
 874 * The H_INT_SET_SOURCE_CONFIG hcall() is used to assign a Logical
 875 * Interrupt Source to a target. The Logical Interrupt Source is
 876 * designated with the "lisn" parameter and the target is designated
 877 * with the "target" and "priority" parameters.  Upon return from the
 878 * hcall(), no additional interrupts will be directed to the old EQ.
 879 *
 880 * Parameters:
 881 * Input:
 882 * - R4: "flags"
 883 *         Bits 0-61: Reserved
 884 *         Bit 62: set the "eisn" in the EAS
 885 *         Bit 63: masks the interrupt source in the hardware interrupt
 886 *       control structure. An interrupt masked by this mechanism will
 887 *       be dropped, but it's source state bits will still be
 888 *       set. There is no race-free way of unmasking and restoring the
 889 *       source. Thus this should only be used in interrupts that are
 890 *       also masked at the source, and only in cases where the
 891 *       interrupt is not meant to be used for a large amount of time
 892 *       because no valid target exists for it for example
 893 * - R5: "lisn" is per "interrupts", "interrupt-map", or
 894 *       "ibm,xive-lisn-ranges" properties, or as returned by the
 895 *       ibm,query-interrupt-source-number RTAS call, or as returned by
 896 *       the H_ALLOCATE_VAS_WINDOW hcall
 897 * - R6: "target" is per "ibm,ppc-interrupt-server#s" or
 898 *       "ibm,ppc-interrupt-gserver#s"
 899 * - R7: "priority" is a valid priority not in
 900 *       "ibm,plat-res-int-priorities"
 901 * - R8: "eisn" is the guest EISN associated with the "lisn"
 902 *
 903 * Output:
 904 * - None
 905 */
 906
 907#define SPAPR_XIVE_SRC_SET_EISN PPC_BIT(62)
 908#define SPAPR_XIVE_SRC_MASK     PPC_BIT(63)
 909
 910static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
 911                                            SpaprMachineState *spapr,
 912                                            target_ulong opcode,
 913                                            target_ulong *args)
 914{
 915    SpaprXive *xive = spapr->xive;
 916    XiveEAS eas, new_eas;
 917    target_ulong flags    = args[0];
 918    target_ulong lisn     = args[1];
 919    target_ulong target   = args[2];
 920    target_ulong priority = args[3];
 921    target_ulong eisn     = args[4];
 922    uint8_t end_blk;
 923    uint32_t end_idx;
 924
 925    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
 926        return H_FUNCTION;
 927    }
 928
 929    if (flags & ~(SPAPR_XIVE_SRC_SET_EISN | SPAPR_XIVE_SRC_MASK)) {
 930        return H_PARAMETER;
 931    }
 932
 933    if (lisn >= xive->nr_irqs) {
 934        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
 935                      lisn);
 936        return H_P2;
 937    }
 938
 939    eas = xive->eat[lisn];
 940    if (!xive_eas_is_valid(&eas)) {
 941        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
 942                      lisn);
 943        return H_P2;
 944    }
 945
 946    /* priority 0xff is used to reset the EAS */
 947    if (priority == 0xff) {
 948        new_eas.w = cpu_to_be64(EAS_VALID | EAS_MASKED);
 949        goto out;
 950    }
 951
 952    if (flags & SPAPR_XIVE_SRC_MASK) {
 953        new_eas.w = eas.w | cpu_to_be64(EAS_MASKED);
 954    } else {
 955        new_eas.w = eas.w & cpu_to_be64(~EAS_MASKED);
 956    }
 957
 958    if (spapr_xive_priority_is_reserved(priority)) {
 959        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
 960                      " is reserved\n", priority);
 961        return H_P4;
 962    }
 963
 964    /*
 965     * Validate that "target" is part of the list of threads allocated
 966     * to the partition. For that, find the END corresponding to the
 967     * target.
 968     */
 969    if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
 970        return H_P3;
 971    }
 972
 973    new_eas.w = xive_set_field64(EAS_END_BLOCK, new_eas.w, end_blk);
 974    new_eas.w = xive_set_field64(EAS_END_INDEX, new_eas.w, end_idx);
 975
 976    if (flags & SPAPR_XIVE_SRC_SET_EISN) {
 977        new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
 978    }
 979
 980    if (kvm_irqchip_in_kernel()) {
 981        Error *local_err = NULL;
 982
 983        kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
 984        if (local_err) {
 985            error_report_err(local_err);
 986            return H_HARDWARE;
 987        }
 988    }
 989
 990out:
 991    xive->eat[lisn] = new_eas;
 992    return H_SUCCESS;
 993}
 994
 995/*
 996 * The H_INT_GET_SOURCE_CONFIG hcall() is used to determine to which
 997 * target/priority pair is assigned to the specified Logical Interrupt
 998 * Source.
 999 *
1000 * Parameters:
1001 * Input:
1002 * - R4: "flags"
1003 *         Bits 0-63 Reserved
1004 * - R5: "lisn" is per "interrupts", "interrupt-map", or
1005 *       "ibm,xive-lisn-ranges" properties, or as returned by the
1006 *       ibm,query-interrupt-source-number RTAS call, or as
1007 *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1008 *
1009 * Output:
1010 * - R4: Target to which the specified Logical Interrupt Source is
1011 *       assigned
1012 * - R5: Priority to which the specified Logical Interrupt Source is
1013 *       assigned
1014 * - R6: EISN for the specified Logical Interrupt Source (this will be
1015 *       equivalent to the LISN if not changed by H_INT_SET_SOURCE_CONFIG)
1016 */
1017static target_ulong h_int_get_source_config(PowerPCCPU *cpu,
1018                                            SpaprMachineState *spapr,
1019                                            target_ulong opcode,
1020                                            target_ulong *args)
1021{
1022    SpaprXive *xive = spapr->xive;
1023    target_ulong flags = args[0];
1024    target_ulong lisn = args[1];
1025    XiveEAS eas;
1026    XiveEND *end;
1027    uint8_t nvt_blk;
1028    uint32_t end_idx, nvt_idx;
1029
1030    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1031        return H_FUNCTION;
1032    }
1033
1034    if (flags) {
1035        return H_PARAMETER;
1036    }
1037
1038    if (lisn >= xive->nr_irqs) {
1039        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1040                      lisn);
1041        return H_P2;
1042    }
1043
1044    eas = xive->eat[lisn];
1045    if (!xive_eas_is_valid(&eas)) {
1046        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1047                      lisn);
1048        return H_P2;
1049    }
1050
1051    /* EAS_END_BLOCK is unused on sPAPR */
1052    end_idx = xive_get_field64(EAS_END_INDEX, eas.w);
1053
1054    assert(end_idx < xive->nr_ends);
1055    end = &xive->endt[end_idx];
1056
1057    nvt_blk = xive_get_field32(END_W6_NVT_BLOCK, end->w6);
1058    nvt_idx = xive_get_field32(END_W6_NVT_INDEX, end->w6);
1059    args[0] = spapr_xive_nvt_to_target(nvt_blk, nvt_idx);
1060
1061    if (xive_eas_is_masked(&eas)) {
1062        args[1] = 0xff;
1063    } else {
1064        args[1] = xive_get_field32(END_W7_F0_PRIORITY, end->w7);
1065    }
1066
1067    args[2] = xive_get_field64(EAS_END_DATA, eas.w);
1068
1069    return H_SUCCESS;
1070}
1071
1072/*
1073 * The H_INT_GET_QUEUE_INFO hcall() is used to get the logical real
1074 * address of the notification management page associated with the
1075 * specified target and priority.
1076 *
1077 * Parameters:
1078 * Input:
1079 * - R4: "flags"
1080 *         Bits 0-63 Reserved
1081 * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1082 *       "ibm,ppc-interrupt-gserver#s"
1083 * - R6: "priority" is a valid priority not in
1084 *       "ibm,plat-res-int-priorities"
1085 *
1086 * Output:
1087 * - R4: Logical real address of notification page
1088 * - R5: Power of 2 page size of the notification page
1089 */
1090static target_ulong h_int_get_queue_info(PowerPCCPU *cpu,
1091                                         SpaprMachineState *spapr,
1092                                         target_ulong opcode,
1093                                         target_ulong *args)
1094{
1095    SpaprXive *xive = spapr->xive;
1096    XiveENDSource *end_xsrc = &xive->end_source;
1097    target_ulong flags = args[0];
1098    target_ulong target = args[1];
1099    target_ulong priority = args[2];
1100    XiveEND *end;
1101    uint8_t end_blk;
1102    uint32_t end_idx;
1103
1104    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1105        return H_FUNCTION;
1106    }
1107
1108    if (flags) {
1109        return H_PARAMETER;
1110    }
1111
1112    /*
1113     * H_STATE should be returned if a H_INT_RESET is in progress.
1114     * This is not needed when running the emulation under QEMU
1115     */
1116
1117    if (spapr_xive_priority_is_reserved(priority)) {
1118        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1119                      " is reserved\n", priority);
1120        return H_P3;
1121    }
1122
1123    /*
1124     * Validate that "target" is part of the list of threads allocated
1125     * to the partition. For that, find the END corresponding to the
1126     * target.
1127     */
1128    if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1129        return H_P2;
1130    }
1131
1132    assert(end_idx < xive->nr_ends);
1133    end = &xive->endt[end_idx];
1134
1135    args[0] = xive->end_base + (1ull << (end_xsrc->esb_shift + 1)) * end_idx;
1136    if (xive_end_is_enqueue(end)) {
1137        args[1] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1138    } else {
1139        args[1] = 0;
1140    }
1141
1142    return H_SUCCESS;
1143}
1144
1145/*
1146 * The H_INT_SET_QUEUE_CONFIG hcall() is used to set or reset a EQ for
1147 * a given "target" and "priority".  It is also used to set the
1148 * notification config associated with the EQ.  An EQ size of 0 is
1149 * used to reset the EQ config for a given target and priority. If
1150 * resetting the EQ config, the END associated with the given "target"
1151 * and "priority" will be changed to disable queueing.
1152 *
1153 * Upon return from the hcall(), no additional interrupts will be
1154 * directed to the old EQ (if one was set). The old EQ (if one was
1155 * set) should be investigated for interrupts that occurred prior to
1156 * or during the hcall().
1157 *
1158 * Parameters:
1159 * Input:
1160 * - R4: "flags"
1161 *         Bits 0-62: Reserved
1162 *         Bit 63: Unconditional Notify (n) per the XIVE spec
1163 * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1164 *       "ibm,ppc-interrupt-gserver#s"
1165 * - R6: "priority" is a valid priority not in
1166 *       "ibm,plat-res-int-priorities"
1167 * - R7: "eventQueue": The logical real address of the start of the EQ
1168 * - R8: "eventQueueSize": The power of 2 EQ size per "ibm,xive-eq-sizes"
1169 *
1170 * Output:
1171 * - None
1172 */
1173
1174#define SPAPR_XIVE_END_ALWAYS_NOTIFY PPC_BIT(63)
1175
1176static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
1177                                           SpaprMachineState *spapr,
1178                                           target_ulong opcode,
1179                                           target_ulong *args)
1180{
1181    SpaprXive *xive = spapr->xive;
1182    target_ulong flags = args[0];
1183    target_ulong target = args[1];
1184    target_ulong priority = args[2];
1185    target_ulong qpage = args[3];
1186    target_ulong qsize = args[4];
1187    XiveEND end;
1188    uint8_t end_blk, nvt_blk;
1189    uint32_t end_idx, nvt_idx;
1190
1191    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1192        return H_FUNCTION;
1193    }
1194
1195    if (flags & ~SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1196        return H_PARAMETER;
1197    }
1198
1199    /*
1200     * H_STATE should be returned if a H_INT_RESET is in progress.
1201     * This is not needed when running the emulation under QEMU
1202     */
1203
1204    if (spapr_xive_priority_is_reserved(priority)) {
1205        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1206                      " is reserved\n", priority);
1207        return H_P3;
1208    }
1209
1210    /*
1211     * Validate that "target" is part of the list of threads allocated
1212     * to the partition. For that, find the END corresponding to the
1213     * target.
1214     */
1215
1216    if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1217        return H_P2;
1218    }
1219
1220    assert(end_idx < xive->nr_ends);
1221    memcpy(&end, &xive->endt[end_idx], sizeof(XiveEND));
1222
1223    switch (qsize) {
1224    case 12:
1225    case 16:
1226    case 21:
1227    case 24:
1228        if (!QEMU_IS_ALIGNED(qpage, 1ul << qsize)) {
1229            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: EQ @0x%" HWADDR_PRIx
1230                          " is not naturally aligned with %" HWADDR_PRIx "\n",
1231                          qpage, (hwaddr)1 << qsize);
1232            return H_P4;
1233        }
1234        end.w2 = cpu_to_be32((qpage >> 32) & 0x0fffffff);
1235        end.w3 = cpu_to_be32(qpage & 0xffffffff);
1236        end.w0 |= cpu_to_be32(END_W0_ENQUEUE);
1237        end.w0 = xive_set_field32(END_W0_QSIZE, end.w0, qsize - 12);
1238        break;
1239    case 0:
1240        /* reset queue and disable queueing */
1241        spapr_xive_end_reset(&end);
1242        goto out;
1243
1244    default:
1245        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: invalid EQ size %"PRIx64"\n",
1246                      qsize);
1247        return H_P5;
1248    }
1249
1250    if (qsize) {
1251        hwaddr plen = 1 << qsize;
1252        void *eq;
1253
1254        /*
1255         * Validate the guest EQ. We should also check that the queue
1256         * has been zeroed by the OS.
1257         */
1258        eq = address_space_map(CPU(cpu)->as, qpage, &plen, true,
1259                               MEMTXATTRS_UNSPECIFIED);
1260        if (plen != 1 << qsize) {
1261            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to map EQ @0x%"
1262                          HWADDR_PRIx "\n", qpage);
1263            return H_P4;
1264        }
1265        address_space_unmap(CPU(cpu)->as, eq, plen, true, plen);
1266    }
1267
1268    /* "target" should have been validated above */
1269    if (spapr_xive_target_to_nvt(target, &nvt_blk, &nvt_idx)) {
1270        g_assert_not_reached();
1271    }
1272
1273    /*
1274     * Ensure the priority and target are correctly set (they will not
1275     * be right after allocation)
1276     */
1277    end.w6 = xive_set_field32(END_W6_NVT_BLOCK, 0ul, nvt_blk) |
1278        xive_set_field32(END_W6_NVT_INDEX, 0ul, nvt_idx);
1279    end.w7 = xive_set_field32(END_W7_F0_PRIORITY, 0ul, priority);
1280
1281    if (flags & SPAPR_XIVE_END_ALWAYS_NOTIFY) {
1282        end.w0 |= cpu_to_be32(END_W0_UCOND_NOTIFY);
1283    } else {
1284        end.w0 &= cpu_to_be32((uint32_t)~END_W0_UCOND_NOTIFY);
1285    }
1286
1287    /*
1288     * The generation bit for the END starts at 1 and The END page
1289     * offset counter starts at 0.
1290     */
1291    end.w1 = cpu_to_be32(END_W1_GENERATION) |
1292        xive_set_field32(END_W1_PAGE_OFF, 0ul, 0ul);
1293    end.w0 |= cpu_to_be32(END_W0_VALID);
1294
1295    /*
1296     * TODO: issue syncs required to ensure all in-flight interrupts
1297     * are complete on the old END
1298     */
1299
1300out:
1301    if (kvm_irqchip_in_kernel()) {
1302        Error *local_err = NULL;
1303
1304        kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
1305        if (local_err) {
1306            error_report_err(local_err);
1307            return H_HARDWARE;
1308        }
1309    }
1310
1311    /* Update END */
1312    memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
1313    return H_SUCCESS;
1314}
1315
1316/*
1317 * The H_INT_GET_QUEUE_CONFIG hcall() is used to get a EQ for a given
1318 * target and priority.
1319 *
1320 * Parameters:
1321 * Input:
1322 * - R4: "flags"
1323 *         Bits 0-62: Reserved
1324 *         Bit 63: Debug: Return debug data
1325 * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1326 *       "ibm,ppc-interrupt-gserver#s"
1327 * - R6: "priority" is a valid priority not in
1328 *       "ibm,plat-res-int-priorities"
1329 *
1330 * Output:
1331 * - R4: "flags":
1332 *       Bits 0-61: Reserved
1333 *       Bit 62: The value of Event Queue Generation Number (g) per
1334 *              the XIVE spec if "Debug" = 1
1335 *       Bit 63: The value of Unconditional Notify (n) per the XIVE spec
1336 * - R5: The logical real address of the start of the EQ
1337 * - R6: The power of 2 EQ size per "ibm,xive-eq-sizes"
1338 * - R7: The value of Event Queue Offset Counter per XIVE spec
1339 *       if "Debug" = 1, else 0
1340 *
1341 */
1342
1343#define SPAPR_XIVE_END_DEBUG     PPC_BIT(63)
1344
1345static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
1346                                           SpaprMachineState *spapr,
1347                                           target_ulong opcode,
1348                                           target_ulong *args)
1349{
1350    SpaprXive *xive = spapr->xive;
1351    target_ulong flags = args[0];
1352    target_ulong target = args[1];
1353    target_ulong priority = args[2];
1354    XiveEND *end;
1355    uint8_t end_blk;
1356    uint32_t end_idx;
1357
1358    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1359        return H_FUNCTION;
1360    }
1361
1362    if (flags & ~SPAPR_XIVE_END_DEBUG) {
1363        return H_PARAMETER;
1364    }
1365
1366    /*
1367     * H_STATE should be returned if a H_INT_RESET is in progress.
1368     * This is not needed when running the emulation under QEMU
1369     */
1370
1371    if (spapr_xive_priority_is_reserved(priority)) {
1372        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: priority " TARGET_FMT_ld
1373                      " is reserved\n", priority);
1374        return H_P3;
1375    }
1376
1377    /*
1378     * Validate that "target" is part of the list of threads allocated
1379     * to the partition. For that, find the END corresponding to the
1380     * target.
1381     */
1382    if (spapr_xive_target_to_end(target, priority, &end_blk, &end_idx)) {
1383        return H_P2;
1384    }
1385
1386    assert(end_idx < xive->nr_ends);
1387    end = &xive->endt[end_idx];
1388
1389    args[0] = 0;
1390    if (xive_end_is_notify(end)) {
1391        args[0] |= SPAPR_XIVE_END_ALWAYS_NOTIFY;
1392    }
1393
1394    if (xive_end_is_enqueue(end)) {
1395        args[1] = xive_end_qaddr(end);
1396        args[2] = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
1397    } else {
1398        args[1] = 0;
1399        args[2] = 0;
1400    }
1401
1402    if (kvm_irqchip_in_kernel()) {
1403        Error *local_err = NULL;
1404
1405        kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
1406        if (local_err) {
1407            error_report_err(local_err);
1408            return H_HARDWARE;
1409        }
1410    }
1411
1412    /* TODO: do we need any locking on the END ? */
1413    if (flags & SPAPR_XIVE_END_DEBUG) {
1414        /* Load the event queue generation number into the return flags */
1415        args[0] |= (uint64_t)xive_get_field32(END_W1_GENERATION, end->w1) << 62;
1416
1417        /* Load R7 with the event queue offset counter */
1418        args[3] = xive_get_field32(END_W1_PAGE_OFF, end->w1);
1419    } else {
1420        args[3] = 0;
1421    }
1422
1423    return H_SUCCESS;
1424}
1425
1426/*
1427 * The H_INT_SET_OS_REPORTING_LINE hcall() is used to set the
1428 * reporting cache line pair for the calling thread.  The reporting
1429 * cache lines will contain the OS interrupt context when the OS
1430 * issues a CI store byte to @TIMA+0xC10 to acknowledge the OS
1431 * interrupt. The reporting cache lines can be reset by inputting -1
1432 * in "reportingLine".  Issuing the CI store byte without reporting
1433 * cache lines registered will result in the data not being accessible
1434 * to the OS.
1435 *
1436 * Parameters:
1437 * Input:
1438 * - R4: "flags"
1439 *         Bits 0-63: Reserved
1440 * - R5: "reportingLine": The logical real address of the reporting cache
1441 *       line pair
1442 *
1443 * Output:
1444 * - None
1445 */
1446static target_ulong h_int_set_os_reporting_line(PowerPCCPU *cpu,
1447                                                SpaprMachineState *spapr,
1448                                                target_ulong opcode,
1449                                                target_ulong *args)
1450{
1451    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1452        return H_FUNCTION;
1453    }
1454
1455    /*
1456     * H_STATE should be returned if a H_INT_RESET is in progress.
1457     * This is not needed when running the emulation under QEMU
1458     */
1459
1460    /* TODO: H_INT_SET_OS_REPORTING_LINE */
1461    return H_FUNCTION;
1462}
1463
1464/*
1465 * The H_INT_GET_OS_REPORTING_LINE hcall() is used to get the logical
1466 * real address of the reporting cache line pair set for the input
1467 * "target".  If no reporting cache line pair has been set, -1 is
1468 * returned.
1469 *
1470 * Parameters:
1471 * Input:
1472 * - R4: "flags"
1473 *         Bits 0-63: Reserved
1474 * - R5: "target" is per "ibm,ppc-interrupt-server#s" or
1475 *       "ibm,ppc-interrupt-gserver#s"
1476 * - R6: "reportingLine": The logical real address of the reporting
1477 *        cache line pair
1478 *
1479 * Output:
1480 * - R4: The logical real address of the reporting line if set, else -1
1481 */
1482static target_ulong h_int_get_os_reporting_line(PowerPCCPU *cpu,
1483                                                SpaprMachineState *spapr,
1484                                                target_ulong opcode,
1485                                                target_ulong *args)
1486{
1487    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1488        return H_FUNCTION;
1489    }
1490
1491    /*
1492     * H_STATE should be returned if a H_INT_RESET is in progress.
1493     * This is not needed when running the emulation under QEMU
1494     */
1495
1496    /* TODO: H_INT_GET_OS_REPORTING_LINE */
1497    return H_FUNCTION;
1498}
1499
1500/*
1501 * The H_INT_ESB hcall() is used to issue a load or store to the ESB
1502 * page for the input "lisn".  This hcall is only supported for LISNs
1503 * that have the ESB hcall flag set to 1 when returned from hcall()
1504 * H_INT_GET_SOURCE_INFO.
1505 *
1506 * Parameters:
1507 * Input:
1508 * - R4: "flags"
1509 *         Bits 0-62: Reserved
1510 *         bit 63: Store: Store=1, store operation, else load operation
1511 * - R5: "lisn" is per "interrupts", "interrupt-map", or
1512 *       "ibm,xive-lisn-ranges" properties, or as returned by the
1513 *       ibm,query-interrupt-source-number RTAS call, or as
1514 *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1515 * - R6: "esbOffset" is the offset into the ESB page for the load or
1516 *       store operation
1517 * - R7: "storeData" is the data to write for a store operation
1518 *
1519 * Output:
1520 * - R4: The value of the load if load operation, else -1
1521 */
1522
1523#define SPAPR_XIVE_ESB_STORE PPC_BIT(63)
1524
1525static target_ulong h_int_esb(PowerPCCPU *cpu,
1526                              SpaprMachineState *spapr,
1527                              target_ulong opcode,
1528                              target_ulong *args)
1529{
1530    SpaprXive *xive = spapr->xive;
1531    XiveEAS eas;
1532    target_ulong flags  = args[0];
1533    target_ulong lisn   = args[1];
1534    target_ulong offset = args[2];
1535    target_ulong data   = args[3];
1536    hwaddr mmio_addr;
1537    XiveSource *xsrc = &xive->source;
1538
1539    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1540        return H_FUNCTION;
1541    }
1542
1543    if (flags & ~SPAPR_XIVE_ESB_STORE) {
1544        return H_PARAMETER;
1545    }
1546
1547    if (lisn >= xive->nr_irqs) {
1548        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1549                      lisn);
1550        return H_P2;
1551    }
1552
1553    eas = xive->eat[lisn];
1554    if (!xive_eas_is_valid(&eas)) {
1555        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1556                      lisn);
1557        return H_P2;
1558    }
1559
1560    if (offset > (1ull << xsrc->esb_shift)) {
1561        return H_P3;
1562    }
1563
1564    if (kvm_irqchip_in_kernel()) {
1565        args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
1566                                     flags & SPAPR_XIVE_ESB_STORE);
1567    } else {
1568        mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
1569
1570        if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
1571                          (flags & SPAPR_XIVE_ESB_STORE))) {
1572            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
1573                          HWADDR_PRIx "\n", mmio_addr);
1574            return H_HARDWARE;
1575        }
1576        args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
1577    }
1578    return H_SUCCESS;
1579}
1580
1581/*
1582 * The H_INT_SYNC hcall() is used to issue hardware syncs that will
1583 * ensure any in flight events for the input lisn are in the event
1584 * queue.
1585 *
1586 * Parameters:
1587 * Input:
1588 * - R4: "flags"
1589 *         Bits 0-63: Reserved
1590 * - R5: "lisn" is per "interrupts", "interrupt-map", or
1591 *       "ibm,xive-lisn-ranges" properties, or as returned by the
1592 *       ibm,query-interrupt-source-number RTAS call, or as
1593 *       returned by the H_ALLOCATE_VAS_WINDOW hcall
1594 *
1595 * Output:
1596 * - None
1597 */
1598static target_ulong h_int_sync(PowerPCCPU *cpu,
1599                               SpaprMachineState *spapr,
1600                               target_ulong opcode,
1601                               target_ulong *args)
1602{
1603    SpaprXive *xive = spapr->xive;
1604    XiveEAS eas;
1605    target_ulong flags = args[0];
1606    target_ulong lisn = args[1];
1607
1608    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1609        return H_FUNCTION;
1610    }
1611
1612    if (flags) {
1613        return H_PARAMETER;
1614    }
1615
1616    if (lisn >= xive->nr_irqs) {
1617        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Unknown LISN " TARGET_FMT_lx "\n",
1618                      lisn);
1619        return H_P2;
1620    }
1621
1622    eas = xive->eat[lisn];
1623    if (!xive_eas_is_valid(&eas)) {
1624        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: Invalid LISN " TARGET_FMT_lx "\n",
1625                      lisn);
1626        return H_P2;
1627    }
1628
1629    /*
1630     * H_STATE should be returned if a H_INT_RESET is in progress.
1631     * This is not needed when running the emulation under QEMU
1632     */
1633
1634    /*
1635     * This is not real hardware. Nothing to be done unless when
1636     * under KVM
1637     */
1638
1639    if (kvm_irqchip_in_kernel()) {
1640        Error *local_err = NULL;
1641
1642        kvmppc_xive_sync_source(xive, lisn, &local_err);
1643        if (local_err) {
1644            error_report_err(local_err);
1645            return H_HARDWARE;
1646        }
1647    }
1648    return H_SUCCESS;
1649}
1650
1651/*
1652 * The H_INT_RESET hcall() is used to reset all of the partition's
1653 * interrupt exploitation structures to their initial state.  This
1654 * means losing all previously set interrupt state set via
1655 * H_INT_SET_SOURCE_CONFIG and H_INT_SET_QUEUE_CONFIG.
1656 *
1657 * Parameters:
1658 * Input:
1659 * - R4: "flags"
1660 *         Bits 0-63: Reserved
1661 *
1662 * Output:
1663 * - None
1664 */
1665static target_ulong h_int_reset(PowerPCCPU *cpu,
1666                                SpaprMachineState *spapr,
1667                                target_ulong opcode,
1668                                target_ulong *args)
1669{
1670    SpaprXive *xive = spapr->xive;
1671    target_ulong flags   = args[0];
1672
1673    if (!spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
1674        return H_FUNCTION;
1675    }
1676
1677    if (flags) {
1678        return H_PARAMETER;
1679    }
1680
1681    device_reset(DEVICE(xive));
1682
1683    if (kvm_irqchip_in_kernel()) {
1684        Error *local_err = NULL;
1685
1686        kvmppc_xive_reset(xive, &local_err);
1687        if (local_err) {
1688            error_report_err(local_err);
1689            return H_HARDWARE;
1690        }
1691    }
1692    return H_SUCCESS;
1693}
1694
1695void spapr_xive_hcall_init(SpaprMachineState *spapr)
1696{
1697    spapr_register_hypercall(H_INT_GET_SOURCE_INFO, h_int_get_source_info);
1698    spapr_register_hypercall(H_INT_SET_SOURCE_CONFIG, h_int_set_source_config);
1699    spapr_register_hypercall(H_INT_GET_SOURCE_CONFIG, h_int_get_source_config);
1700    spapr_register_hypercall(H_INT_GET_QUEUE_INFO, h_int_get_queue_info);
1701    spapr_register_hypercall(H_INT_SET_QUEUE_CONFIG, h_int_set_queue_config);
1702    spapr_register_hypercall(H_INT_GET_QUEUE_CONFIG, h_int_get_queue_config);
1703    spapr_register_hypercall(H_INT_SET_OS_REPORTING_LINE,
1704                             h_int_set_os_reporting_line);
1705    spapr_register_hypercall(H_INT_GET_OS_REPORTING_LINE,
1706                             h_int_get_os_reporting_line);
1707    spapr_register_hypercall(H_INT_ESB, h_int_esb);
1708    spapr_register_hypercall(H_INT_SYNC, h_int_sync);
1709    spapr_register_hypercall(H_INT_RESET, h_int_reset);
1710}
1711