qemu/hw/arm/virt.c
<<
>>
Prefs
   1/*
   2 * ARM mach-virt emulation
   3 *
   4 * Copyright (c) 2013 Linaro Limited
   5 *
   6 * This program is free software; you can redistribute it and/or modify it
   7 * under the terms and conditions of the GNU General Public License,
   8 * version 2 or later, as published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope it will be useful, but WITHOUT
  11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13 * more details.
  14 *
  15 * You should have received a copy of the GNU General Public License along with
  16 * this program.  If not, see <http://www.gnu.org/licenses/>.
  17 *
  18 * Emulate a virtual board which works by passing Linux all the information
  19 * it needs about what devices are present via the device tree.
  20 * There are some restrictions about what we can do here:
  21 *  + we can only present devices whose Linux drivers will work based
  22 *    purely on the device tree with no platform data at all
  23 *  + we want to present a very stripped-down minimalist platform,
  24 *    both because this reduces the security attack surface from the guest
  25 *    and also because it reduces our exposure to being broken when
  26 *    the kernel updates its device tree bindings and requires further
  27 *    information in a device binding that we aren't providing.
  28 * This is essentially the same approach kvmtool uses.
  29 */
  30
  31#include "qemu/osdep.h"
  32#include "qemu-common.h"
  33#include "qemu/datadir.h"
  34#include "qemu/units.h"
  35#include "qemu/option.h"
  36#include "monitor/qdev.h"
  37#include "qapi/error.h"
  38#include "hw/sysbus.h"
  39#include "hw/arm/boot.h"
  40#include "hw/arm/primecell.h"
  41#include "hw/arm/virt.h"
  42#include "hw/block/flash.h"
  43#include "hw/vfio/vfio-calxeda-xgmac.h"
  44#include "hw/vfio/vfio-amd-xgbe.h"
  45#include "hw/display/ramfb.h"
  46#include "net/net.h"
  47#include "sysemu/device_tree.h"
  48#include "sysemu/numa.h"
  49#include "sysemu/runstate.h"
  50#include "sysemu/tpm.h"
  51#include "sysemu/kvm.h"
  52#include "sysemu/hvf.h"
  53#include "hw/loader.h"
  54#include "qapi/error.h"
  55#include "qemu/bitops.h"
  56#include "qemu/error-report.h"
  57#include "qemu/module.h"
  58#include "hw/pci-host/gpex.h"
  59#include "hw/virtio/virtio-pci.h"
  60#include "hw/arm/sysbus-fdt.h"
  61#include "hw/platform-bus.h"
  62#include "hw/qdev-properties.h"
  63#include "hw/arm/fdt.h"
  64#include "hw/intc/arm_gic.h"
  65#include "hw/intc/arm_gicv3_common.h"
  66#include "hw/irq.h"
  67#include "kvm_arm.h"
  68#include "hw/firmware/smbios.h"
  69#include "qapi/visitor.h"
  70#include "qapi/qapi-visit-common.h"
  71#include "standard-headers/linux/input.h"
  72#include "hw/arm/smmuv3.h"
  73#include "hw/acpi/acpi.h"
  74#include "target/arm/internals.h"
  75#include "hw/mem/pc-dimm.h"
  76#include "hw/mem/nvdimm.h"
  77#include "hw/acpi/generic_event_device.h"
  78#include "hw/virtio/virtio-iommu.h"
  79#include "hw/char/pl011.h"
  80#include "qemu/guest-random.h"
  81
  82#define DEFINE_VIRT_MACHINE_LATEST(major, minor, latest) \
  83    static void virt_##major##_##minor##_class_init(ObjectClass *oc, \
  84                                                    void *data) \
  85    { \
  86        MachineClass *mc = MACHINE_CLASS(oc); \
  87        virt_machine_##major##_##minor##_options(mc); \
  88        mc->desc = "QEMU " # major "." # minor " ARM Virtual Machine"; \
  89        if (latest) { \
  90            mc->alias = "virt"; \
  91        } \
  92    } \
  93    static const TypeInfo machvirt_##major##_##minor##_info = { \
  94        .name = MACHINE_TYPE_NAME("virt-" # major "." # minor), \
  95        .parent = TYPE_VIRT_MACHINE, \
  96        .class_init = virt_##major##_##minor##_class_init, \
  97    }; \
  98    static void machvirt_machine_##major##_##minor##_init(void) \
  99    { \
 100        type_register_static(&machvirt_##major##_##minor##_info); \
 101    } \
 102    type_init(machvirt_machine_##major##_##minor##_init);
 103
 104#define DEFINE_VIRT_MACHINE_AS_LATEST(major, minor) \
 105    DEFINE_VIRT_MACHINE_LATEST(major, minor, true)
 106#define DEFINE_VIRT_MACHINE(major, minor) \
 107    DEFINE_VIRT_MACHINE_LATEST(major, minor, false)
 108
 109
 110/* Number of external interrupt lines to configure the GIC with */
 111#define NUM_IRQS 256
 112
 113#define PLATFORM_BUS_NUM_IRQS 64
 114
 115/* Legacy RAM limit in GB (< version 4.0) */
 116#define LEGACY_RAMLIMIT_GB 255
 117#define LEGACY_RAMLIMIT_BYTES (LEGACY_RAMLIMIT_GB * GiB)
 118
 119/* Addresses and sizes of our components.
 120 * 0..128MB is space for a flash device so we can run bootrom code such as UEFI.
 121 * 128MB..256MB is used for miscellaneous device I/O.
 122 * 256MB..1GB is reserved for possible future PCI support (ie where the
 123 * PCI memory window will go if we add a PCI host controller).
 124 * 1GB and up is RAM (which may happily spill over into the
 125 * high memory region beyond 4GB).
 126 * This represents a compromise between how much RAM can be given to
 127 * a 32 bit VM and leaving space for expansion and in particular for PCI.
 128 * Note that devices should generally be placed at multiples of 0x10000,
 129 * to accommodate guests using 64K pages.
 130 */
 131static const MemMapEntry base_memmap[] = {
 132    /* Space up to 0x8000000 is reserved for a boot ROM */
 133    [VIRT_FLASH] =              {          0, 0x08000000 },
 134    [VIRT_CPUPERIPHS] =         { 0x08000000, 0x00020000 },
 135    /* GIC distributor and CPU interfaces sit inside the CPU peripheral space */
 136    [VIRT_GIC_DIST] =           { 0x08000000, 0x00010000 },
 137    [VIRT_GIC_CPU] =            { 0x08010000, 0x00010000 },
 138    [VIRT_GIC_V2M] =            { 0x08020000, 0x00001000 },
 139    [VIRT_GIC_HYP] =            { 0x08030000, 0x00010000 },
 140    [VIRT_GIC_VCPU] =           { 0x08040000, 0x00010000 },
 141    /* The space in between here is reserved for GICv3 CPU/vCPU/HYP */
 142    [VIRT_GIC_ITS] =            { 0x08080000, 0x00020000 },
 143    /* This redistributor space allows up to 2*64kB*123 CPUs */
 144    [VIRT_GIC_REDIST] =         { 0x080A0000, 0x00F60000 },
 145    [VIRT_UART] =               { 0x09000000, 0x00001000 },
 146    [VIRT_RTC] =                { 0x09010000, 0x00001000 },
 147    [VIRT_FW_CFG] =             { 0x09020000, 0x00000018 },
 148    [VIRT_GPIO] =               { 0x09030000, 0x00001000 },
 149    [VIRT_SECURE_UART] =        { 0x09040000, 0x00001000 },
 150    [VIRT_SMMU] =               { 0x09050000, 0x00020000 },
 151    [VIRT_PCDIMM_ACPI] =        { 0x09070000, MEMORY_HOTPLUG_IO_LEN },
 152    [VIRT_ACPI_GED] =           { 0x09080000, ACPI_GED_EVT_SEL_LEN },
 153    [VIRT_NVDIMM_ACPI] =        { 0x09090000, NVDIMM_ACPI_IO_LEN},
 154    [VIRT_PVTIME] =             { 0x090a0000, 0x00010000 },
 155    [VIRT_SECURE_GPIO] =        { 0x090b0000, 0x00001000 },
 156    [VIRT_MMIO] =               { 0x0a000000, 0x00000200 },
 157    /* ...repeating for a total of NUM_VIRTIO_TRANSPORTS, each of that size */
 158    [VIRT_PLATFORM_BUS] =       { 0x0c000000, 0x02000000 },
 159    [VIRT_SECURE_MEM] =         { 0x0e000000, 0x01000000 },
 160    [VIRT_PCIE_MMIO] =          { 0x10000000, 0x2eff0000 },
 161    [VIRT_PCIE_PIO] =           { 0x3eff0000, 0x00010000 },
 162    [VIRT_PCIE_ECAM] =          { 0x3f000000, 0x01000000 },
 163    /* Actual RAM size depends on initial RAM and device memory settings */
 164    [VIRT_MEM] =                { GiB, LEGACY_RAMLIMIT_BYTES },
 165};
 166
 167/*
 168 * Highmem IO Regions: This memory map is floating, located after the RAM.
 169 * Each MemMapEntry base (GPA) will be dynamically computed, depending on the
 170 * top of the RAM, so that its base get the same alignment as the size,
 171 * ie. a 512GiB entry will be aligned on a 512GiB boundary. If there is
 172 * less than 256GiB of RAM, the floating area starts at the 256GiB mark.
 173 * Note the extended_memmap is sized so that it eventually also includes the
 174 * base_memmap entries (VIRT_HIGH_GIC_REDIST2 index is greater than the last
 175 * index of base_memmap).
 176 */
 177static MemMapEntry extended_memmap[] = {
 178    /* Additional 64 MB redist region (can contain up to 512 redistributors) */
 179    [VIRT_HIGH_GIC_REDIST2] =   { 0x0, 64 * MiB },
 180    [VIRT_HIGH_PCIE_ECAM] =     { 0x0, 256 * MiB },
 181    /* Second PCIe window */
 182    [VIRT_HIGH_PCIE_MMIO] =     { 0x0, 512 * GiB },
 183};
 184
 185static const int a15irqmap[] = {
 186    [VIRT_UART] = 1,
 187    [VIRT_RTC] = 2,
 188    [VIRT_PCIE] = 3, /* ... to 6 */
 189    [VIRT_GPIO] = 7,
 190    [VIRT_SECURE_UART] = 8,
 191    [VIRT_ACPI_GED] = 9,
 192    [VIRT_MMIO] = 16, /* ...to 16 + NUM_VIRTIO_TRANSPORTS - 1 */
 193    [VIRT_GIC_V2M] = 48, /* ...to 48 + NUM_GICV2M_SPIS - 1 */
 194    [VIRT_SMMU] = 74,    /* ...to 74 + NUM_SMMU_IRQS - 1 */
 195    [VIRT_PLATFORM_BUS] = 112, /* ...to 112 + PLATFORM_BUS_NUM_IRQS -1 */
 196};
 197
 198static const char *valid_cpus[] = {
 199    ARM_CPU_TYPE_NAME("cortex-a7"),
 200    ARM_CPU_TYPE_NAME("cortex-a15"),
 201    ARM_CPU_TYPE_NAME("cortex-a53"),
 202    ARM_CPU_TYPE_NAME("cortex-a57"),
 203    ARM_CPU_TYPE_NAME("cortex-a72"),
 204    ARM_CPU_TYPE_NAME("a64fx"),
 205    ARM_CPU_TYPE_NAME("host"),
 206    ARM_CPU_TYPE_NAME("max"),
 207};
 208
 209static bool cpu_type_valid(const char *cpu)
 210{
 211    int i;
 212
 213    for (i = 0; i < ARRAY_SIZE(valid_cpus); i++) {
 214        if (strcmp(cpu, valid_cpus[i]) == 0) {
 215            return true;
 216        }
 217    }
 218    return false;
 219}
 220
 221static void create_kaslr_seed(MachineState *ms, const char *node)
 222{
 223    uint64_t seed;
 224
 225    if (qemu_guest_getrandom(&seed, sizeof(seed), NULL)) {
 226        return;
 227    }
 228    qemu_fdt_setprop_u64(ms->fdt, node, "kaslr-seed", seed);
 229}
 230
 231static void create_fdt(VirtMachineState *vms)
 232{
 233    MachineState *ms = MACHINE(vms);
 234    int nb_numa_nodes = ms->numa_state->num_nodes;
 235    void *fdt = create_device_tree(&vms->fdt_size);
 236
 237    if (!fdt) {
 238        error_report("create_device_tree() failed");
 239        exit(1);
 240    }
 241
 242    ms->fdt = fdt;
 243
 244    /* Header */
 245    qemu_fdt_setprop_string(fdt, "/", "compatible", "linux,dummy-virt");
 246    qemu_fdt_setprop_cell(fdt, "/", "#address-cells", 0x2);
 247    qemu_fdt_setprop_cell(fdt, "/", "#size-cells", 0x2);
 248
 249    /* /chosen must exist for load_dtb to fill in necessary properties later */
 250    qemu_fdt_add_subnode(fdt, "/chosen");
 251    create_kaslr_seed(ms, "/chosen");
 252
 253    if (vms->secure) {
 254        qemu_fdt_add_subnode(fdt, "/secure-chosen");
 255        create_kaslr_seed(ms, "/secure-chosen");
 256    }
 257
 258    /* Clock node, for the benefit of the UART. The kernel device tree
 259     * binding documentation claims the PL011 node clock properties are
 260     * optional but in practice if you omit them the kernel refuses to
 261     * probe for the device.
 262     */
 263    vms->clock_phandle = qemu_fdt_alloc_phandle(fdt);
 264    qemu_fdt_add_subnode(fdt, "/apb-pclk");
 265    qemu_fdt_setprop_string(fdt, "/apb-pclk", "compatible", "fixed-clock");
 266    qemu_fdt_setprop_cell(fdt, "/apb-pclk", "#clock-cells", 0x0);
 267    qemu_fdt_setprop_cell(fdt, "/apb-pclk", "clock-frequency", 24000000);
 268    qemu_fdt_setprop_string(fdt, "/apb-pclk", "clock-output-names",
 269                                "clk24mhz");
 270    qemu_fdt_setprop_cell(fdt, "/apb-pclk", "phandle", vms->clock_phandle);
 271
 272    if (nb_numa_nodes > 0 && ms->numa_state->have_numa_distance) {
 273        int size = nb_numa_nodes * nb_numa_nodes * 3 * sizeof(uint32_t);
 274        uint32_t *matrix = g_malloc0(size);
 275        int idx, i, j;
 276
 277        for (i = 0; i < nb_numa_nodes; i++) {
 278            for (j = 0; j < nb_numa_nodes; j++) {
 279                idx = (i * nb_numa_nodes + j) * 3;
 280                matrix[idx + 0] = cpu_to_be32(i);
 281                matrix[idx + 1] = cpu_to_be32(j);
 282                matrix[idx + 2] =
 283                    cpu_to_be32(ms->numa_state->nodes[i].distance[j]);
 284            }
 285        }
 286
 287        qemu_fdt_add_subnode(fdt, "/distance-map");
 288        qemu_fdt_setprop_string(fdt, "/distance-map", "compatible",
 289                                "numa-distance-map-v1");
 290        qemu_fdt_setprop(fdt, "/distance-map", "distance-matrix",
 291                         matrix, size);
 292        g_free(matrix);
 293    }
 294}
 295
 296static void fdt_add_timer_nodes(const VirtMachineState *vms)
 297{
 298    /* On real hardware these interrupts are level-triggered.
 299     * On KVM they were edge-triggered before host kernel version 4.4,
 300     * and level-triggered afterwards.
 301     * On emulated QEMU they are level-triggered.
 302     *
 303     * Getting the DTB info about them wrong is awkward for some
 304     * guest kernels:
 305     *  pre-4.8 ignore the DT and leave the interrupt configured
 306     *   with whatever the GIC reset value (or the bootloader) left it at
 307     *  4.8 before rc6 honour the incorrect data by programming it back
 308     *   into the GIC, causing problems
 309     *  4.8rc6 and later ignore the DT and always write "level triggered"
 310     *   into the GIC
 311     *
 312     * For backwards-compatibility, virt-2.8 and earlier will continue
 313     * to say these are edge-triggered, but later machines will report
 314     * the correct information.
 315     */
 316    ARMCPU *armcpu;
 317    VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
 318    uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
 319    MachineState *ms = MACHINE(vms);
 320
 321    if (vmc->claim_edge_triggered_timers) {
 322        irqflags = GIC_FDT_IRQ_FLAGS_EDGE_LO_HI;
 323    }
 324
 325    if (vms->gic_version == VIRT_GIC_VERSION_2) {
 326        irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
 327                             GIC_FDT_IRQ_PPI_CPU_WIDTH,
 328                             (1 << MACHINE(vms)->smp.cpus) - 1);
 329    }
 330
 331    qemu_fdt_add_subnode(ms->fdt, "/timer");
 332
 333    armcpu = ARM_CPU(qemu_get_cpu(0));
 334    if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
 335        const char compat[] = "arm,armv8-timer\0arm,armv7-timer";
 336        qemu_fdt_setprop(ms->fdt, "/timer", "compatible",
 337                         compat, sizeof(compat));
 338    } else {
 339        qemu_fdt_setprop_string(ms->fdt, "/timer", "compatible",
 340                                "arm,armv7-timer");
 341    }
 342    qemu_fdt_setprop(ms->fdt, "/timer", "always-on", NULL, 0);
 343    qemu_fdt_setprop_cells(ms->fdt, "/timer", "interrupts",
 344                       GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_S_EL1_IRQ, irqflags,
 345                       GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL1_IRQ, irqflags,
 346                       GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_VIRT_IRQ, irqflags,
 347                       GIC_FDT_IRQ_TYPE_PPI, ARCH_TIMER_NS_EL2_IRQ, irqflags);
 348}
 349
 350static void fdt_add_cpu_nodes(const VirtMachineState *vms)
 351{
 352    int cpu;
 353    int addr_cells = 1;
 354    const MachineState *ms = MACHINE(vms);
 355    const VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
 356    int smp_cpus = ms->smp.cpus;
 357
 358    /*
 359     * See Linux Documentation/devicetree/bindings/arm/cpus.yaml
 360     * On ARM v8 64-bit systems value should be set to 2,
 361     * that corresponds to the MPIDR_EL1 register size.
 362     * If MPIDR_EL1[63:32] value is equal to 0 on all CPUs
 363     * in the system, #address-cells can be set to 1, since
 364     * MPIDR_EL1[63:32] bits are not used for CPUs
 365     * identification.
 366     *
 367     * Here we actually don't know whether our system is 32- or 64-bit one.
 368     * The simplest way to go is to examine affinity IDs of all our CPUs. If
 369     * at least one of them has Aff3 populated, we set #address-cells to 2.
 370     */
 371    for (cpu = 0; cpu < smp_cpus; cpu++) {
 372        ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
 373
 374        if (armcpu->mp_affinity & ARM_AFF3_MASK) {
 375            addr_cells = 2;
 376            break;
 377        }
 378    }
 379
 380    qemu_fdt_add_subnode(ms->fdt, "/cpus");
 381    qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#address-cells", addr_cells);
 382    qemu_fdt_setprop_cell(ms->fdt, "/cpus", "#size-cells", 0x0);
 383
 384    for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
 385        char *nodename = g_strdup_printf("/cpus/cpu@%d", cpu);
 386        ARMCPU *armcpu = ARM_CPU(qemu_get_cpu(cpu));
 387        CPUState *cs = CPU(armcpu);
 388
 389        qemu_fdt_add_subnode(ms->fdt, nodename);
 390        qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "cpu");
 391        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
 392                                    armcpu->dtb_compatible);
 393
 394        if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED && smp_cpus > 1) {
 395            qemu_fdt_setprop_string(ms->fdt, nodename,
 396                                        "enable-method", "psci");
 397        }
 398
 399        if (addr_cells == 2) {
 400            qemu_fdt_setprop_u64(ms->fdt, nodename, "reg",
 401                                 armcpu->mp_affinity);
 402        } else {
 403            qemu_fdt_setprop_cell(ms->fdt, nodename, "reg",
 404                                  armcpu->mp_affinity);
 405        }
 406
 407        if (ms->possible_cpus->cpus[cs->cpu_index].props.has_node_id) {
 408            qemu_fdt_setprop_cell(ms->fdt, nodename, "numa-node-id",
 409                ms->possible_cpus->cpus[cs->cpu_index].props.node_id);
 410        }
 411
 412        if (!vmc->no_cpu_topology) {
 413            qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle",
 414                                  qemu_fdt_alloc_phandle(ms->fdt));
 415        }
 416
 417        g_free(nodename);
 418    }
 419
 420    if (!vmc->no_cpu_topology) {
 421        /*
 422         * Add vCPU topology description through fdt node cpu-map.
 423         *
 424         * See Linux Documentation/devicetree/bindings/cpu/cpu-topology.txt
 425         * In a SMP system, the hierarchy of CPUs can be defined through
 426         * four entities that are used to describe the layout of CPUs in
 427         * the system: socket/cluster/core/thread.
 428         *
 429         * A socket node represents the boundary of system physical package
 430         * and its child nodes must be one or more cluster nodes. A system
 431         * can contain several layers of clustering within a single physical
 432         * package and cluster nodes can be contained in parent cluster nodes.
 433         *
 434         * Given that cluster is not yet supported in the vCPU topology,
 435         * we currently generate one cluster node within each socket node
 436         * by default.
 437         */
 438        qemu_fdt_add_subnode(ms->fdt, "/cpus/cpu-map");
 439
 440        for (cpu = smp_cpus - 1; cpu >= 0; cpu--) {
 441            char *cpu_path = g_strdup_printf("/cpus/cpu@%d", cpu);
 442            char *map_path;
 443
 444            if (ms->smp.threads > 1) {
 445                map_path = g_strdup_printf(
 446                    "/cpus/cpu-map/socket%d/cluster0/core%d/thread%d",
 447                    cpu / (ms->smp.cores * ms->smp.threads),
 448                    (cpu / ms->smp.threads) % ms->smp.cores,
 449                    cpu % ms->smp.threads);
 450            } else {
 451                map_path = g_strdup_printf(
 452                    "/cpus/cpu-map/socket%d/cluster0/core%d",
 453                    cpu / ms->smp.cores,
 454                    cpu % ms->smp.cores);
 455            }
 456            qemu_fdt_add_path(ms->fdt, map_path);
 457            qemu_fdt_setprop_phandle(ms->fdt, map_path, "cpu", cpu_path);
 458
 459            g_free(map_path);
 460            g_free(cpu_path);
 461        }
 462    }
 463}
 464
 465static void fdt_add_its_gic_node(VirtMachineState *vms)
 466{
 467    char *nodename;
 468    MachineState *ms = MACHINE(vms);
 469
 470    vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
 471    nodename = g_strdup_printf("/intc/its@%" PRIx64,
 472                               vms->memmap[VIRT_GIC_ITS].base);
 473    qemu_fdt_add_subnode(ms->fdt, nodename);
 474    qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
 475                            "arm,gic-v3-its");
 476    qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
 477    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 478                                 2, vms->memmap[VIRT_GIC_ITS].base,
 479                                 2, vms->memmap[VIRT_GIC_ITS].size);
 480    qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
 481    g_free(nodename);
 482}
 483
 484static void fdt_add_v2m_gic_node(VirtMachineState *vms)
 485{
 486    MachineState *ms = MACHINE(vms);
 487    char *nodename;
 488
 489    nodename = g_strdup_printf("/intc/v2m@%" PRIx64,
 490                               vms->memmap[VIRT_GIC_V2M].base);
 491    vms->msi_phandle = qemu_fdt_alloc_phandle(ms->fdt);
 492    qemu_fdt_add_subnode(ms->fdt, nodename);
 493    qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
 494                            "arm,gic-v2m-frame");
 495    qemu_fdt_setprop(ms->fdt, nodename, "msi-controller", NULL, 0);
 496    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 497                                 2, vms->memmap[VIRT_GIC_V2M].base,
 498                                 2, vms->memmap[VIRT_GIC_V2M].size);
 499    qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->msi_phandle);
 500    g_free(nodename);
 501}
 502
 503static void fdt_add_gic_node(VirtMachineState *vms)
 504{
 505    MachineState *ms = MACHINE(vms);
 506    char *nodename;
 507
 508    vms->gic_phandle = qemu_fdt_alloc_phandle(ms->fdt);
 509    qemu_fdt_setprop_cell(ms->fdt, "/", "interrupt-parent", vms->gic_phandle);
 510
 511    nodename = g_strdup_printf("/intc@%" PRIx64,
 512                               vms->memmap[VIRT_GIC_DIST].base);
 513    qemu_fdt_add_subnode(ms->fdt, nodename);
 514    qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 3);
 515    qemu_fdt_setprop(ms->fdt, nodename, "interrupt-controller", NULL, 0);
 516    qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 0x2);
 517    qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 0x2);
 518    qemu_fdt_setprop(ms->fdt, nodename, "ranges", NULL, 0);
 519    if (vms->gic_version == VIRT_GIC_VERSION_3) {
 520        int nb_redist_regions = virt_gicv3_redist_region_count(vms);
 521
 522        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
 523                                "arm,gic-v3");
 524
 525        qemu_fdt_setprop_cell(ms->fdt, nodename,
 526                              "#redistributor-regions", nb_redist_regions);
 527
 528        if (nb_redist_regions == 1) {
 529            qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 530                                         2, vms->memmap[VIRT_GIC_DIST].base,
 531                                         2, vms->memmap[VIRT_GIC_DIST].size,
 532                                         2, vms->memmap[VIRT_GIC_REDIST].base,
 533                                         2, vms->memmap[VIRT_GIC_REDIST].size);
 534        } else {
 535            qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 536                                 2, vms->memmap[VIRT_GIC_DIST].base,
 537                                 2, vms->memmap[VIRT_GIC_DIST].size,
 538                                 2, vms->memmap[VIRT_GIC_REDIST].base,
 539                                 2, vms->memmap[VIRT_GIC_REDIST].size,
 540                                 2, vms->memmap[VIRT_HIGH_GIC_REDIST2].base,
 541                                 2, vms->memmap[VIRT_HIGH_GIC_REDIST2].size);
 542        }
 543
 544        if (vms->virt) {
 545            qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
 546                                   GIC_FDT_IRQ_TYPE_PPI, ARCH_GIC_MAINT_IRQ,
 547                                   GIC_FDT_IRQ_FLAGS_LEVEL_HI);
 548        }
 549    } else {
 550        /* 'cortex-a15-gic' means 'GIC v2' */
 551        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible",
 552                                "arm,cortex-a15-gic");
 553        if (!vms->virt) {
 554            qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 555                                         2, vms->memmap[VIRT_GIC_DIST].base,
 556                                         2, vms->memmap[VIRT_GIC_DIST].size,
 557                                         2, vms->memmap[VIRT_GIC_CPU].base,
 558                                         2, vms->memmap[VIRT_GIC_CPU].size);
 559        } else {
 560            qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 561                                         2, vms->memmap[VIRT_GIC_DIST].base,
 562                                         2, vms->memmap[VIRT_GIC_DIST].size,
 563                                         2, vms->memmap[VIRT_GIC_CPU].base,
 564                                         2, vms->memmap[VIRT_GIC_CPU].size,
 565                                         2, vms->memmap[VIRT_GIC_HYP].base,
 566                                         2, vms->memmap[VIRT_GIC_HYP].size,
 567                                         2, vms->memmap[VIRT_GIC_VCPU].base,
 568                                         2, vms->memmap[VIRT_GIC_VCPU].size);
 569            qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
 570                                   GIC_FDT_IRQ_TYPE_PPI, ARCH_GIC_MAINT_IRQ,
 571                                   GIC_FDT_IRQ_FLAGS_LEVEL_HI);
 572        }
 573    }
 574
 575    qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", vms->gic_phandle);
 576    g_free(nodename);
 577}
 578
 579static void fdt_add_pmu_nodes(const VirtMachineState *vms)
 580{
 581    ARMCPU *armcpu = ARM_CPU(first_cpu);
 582    uint32_t irqflags = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
 583    MachineState *ms = MACHINE(vms);
 584
 585    if (!arm_feature(&armcpu->env, ARM_FEATURE_PMU)) {
 586        assert(!object_property_get_bool(OBJECT(armcpu), "pmu", NULL));
 587        return;
 588    }
 589
 590    if (vms->gic_version == VIRT_GIC_VERSION_2) {
 591        irqflags = deposit32(irqflags, GIC_FDT_IRQ_PPI_CPU_START,
 592                             GIC_FDT_IRQ_PPI_CPU_WIDTH,
 593                             (1 << MACHINE(vms)->smp.cpus) - 1);
 594    }
 595
 596    qemu_fdt_add_subnode(ms->fdt, "/pmu");
 597    if (arm_feature(&armcpu->env, ARM_FEATURE_V8)) {
 598        const char compat[] = "arm,armv8-pmuv3";
 599        qemu_fdt_setprop(ms->fdt, "/pmu", "compatible",
 600                         compat, sizeof(compat));
 601        qemu_fdt_setprop_cells(ms->fdt, "/pmu", "interrupts",
 602                               GIC_FDT_IRQ_TYPE_PPI, VIRTUAL_PMU_IRQ, irqflags);
 603    }
 604}
 605
 606static inline DeviceState *create_acpi_ged(VirtMachineState *vms)
 607{
 608    DeviceState *dev;
 609    MachineState *ms = MACHINE(vms);
 610    int irq = vms->irqmap[VIRT_ACPI_GED];
 611    uint32_t event = ACPI_GED_PWR_DOWN_EVT;
 612
 613    if (ms->ram_slots) {
 614        event |= ACPI_GED_MEM_HOTPLUG_EVT;
 615    }
 616
 617    if (ms->nvdimms_state->is_enabled) {
 618        event |= ACPI_GED_NVDIMM_HOTPLUG_EVT;
 619    }
 620
 621    dev = qdev_new(TYPE_ACPI_GED);
 622    qdev_prop_set_uint32(dev, "ged-event", event);
 623
 624    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_ACPI_GED].base);
 625    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 1, vms->memmap[VIRT_PCDIMM_ACPI].base);
 626    sysbus_connect_irq(SYS_BUS_DEVICE(dev), 0, qdev_get_gpio_in(vms->gic, irq));
 627
 628    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
 629
 630    return dev;
 631}
 632
 633static void create_its(VirtMachineState *vms)
 634{
 635    const char *itsclass = its_class_name();
 636    DeviceState *dev;
 637
 638    if (!strcmp(itsclass, "arm-gicv3-its")) {
 639        if (!vms->tcg_its) {
 640            itsclass = NULL;
 641        }
 642    }
 643
 644    if (!itsclass) {
 645        /* Do nothing if not supported */
 646        return;
 647    }
 648
 649    dev = qdev_new(itsclass);
 650
 651    object_property_set_link(OBJECT(dev), "parent-gicv3", OBJECT(vms->gic),
 652                             &error_abort);
 653    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
 654    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_ITS].base);
 655
 656    fdt_add_its_gic_node(vms);
 657    vms->msi_controller = VIRT_MSI_CTRL_ITS;
 658}
 659
 660static void create_v2m(VirtMachineState *vms)
 661{
 662    int i;
 663    int irq = vms->irqmap[VIRT_GIC_V2M];
 664    DeviceState *dev;
 665
 666    dev = qdev_new("arm-gicv2m");
 667    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, vms->memmap[VIRT_GIC_V2M].base);
 668    qdev_prop_set_uint32(dev, "base-spi", irq);
 669    qdev_prop_set_uint32(dev, "num-spi", NUM_GICV2M_SPIS);
 670    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
 671
 672    for (i = 0; i < NUM_GICV2M_SPIS; i++) {
 673        sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
 674                           qdev_get_gpio_in(vms->gic, irq + i));
 675    }
 676
 677    fdt_add_v2m_gic_node(vms);
 678    vms->msi_controller = VIRT_MSI_CTRL_GICV2M;
 679}
 680
 681static void create_gic(VirtMachineState *vms, MemoryRegion *mem)
 682{
 683    MachineState *ms = MACHINE(vms);
 684    /* We create a standalone GIC */
 685    SysBusDevice *gicbusdev;
 686    const char *gictype;
 687    int type = vms->gic_version, i;
 688    unsigned int smp_cpus = ms->smp.cpus;
 689    uint32_t nb_redist_regions = 0;
 690
 691    gictype = (type == 3) ? gicv3_class_name() : gic_class_name();
 692
 693    vms->gic = qdev_new(gictype);
 694    qdev_prop_set_uint32(vms->gic, "revision", type);
 695    qdev_prop_set_uint32(vms->gic, "num-cpu", smp_cpus);
 696    /* Note that the num-irq property counts both internal and external
 697     * interrupts; there are always 32 of the former (mandated by GIC spec).
 698     */
 699    qdev_prop_set_uint32(vms->gic, "num-irq", NUM_IRQS + 32);
 700    if (!kvm_irqchip_in_kernel()) {
 701        qdev_prop_set_bit(vms->gic, "has-security-extensions", vms->secure);
 702    }
 703
 704    if (type == 3) {
 705        uint32_t redist0_capacity =
 706                    vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE;
 707        uint32_t redist0_count = MIN(smp_cpus, redist0_capacity);
 708
 709        nb_redist_regions = virt_gicv3_redist_region_count(vms);
 710
 711        qdev_prop_set_uint32(vms->gic, "len-redist-region-count",
 712                             nb_redist_regions);
 713        qdev_prop_set_uint32(vms->gic, "redist-region-count[0]", redist0_count);
 714
 715        if (!kvm_irqchip_in_kernel()) {
 716            if (vms->tcg_its) {
 717                object_property_set_link(OBJECT(vms->gic), "sysmem",
 718                                         OBJECT(mem), &error_fatal);
 719                qdev_prop_set_bit(vms->gic, "has-lpi", true);
 720            }
 721        }
 722
 723        if (nb_redist_regions == 2) {
 724            uint32_t redist1_capacity =
 725                    vms->memmap[VIRT_HIGH_GIC_REDIST2].size / GICV3_REDIST_SIZE;
 726
 727            qdev_prop_set_uint32(vms->gic, "redist-region-count[1]",
 728                MIN(smp_cpus - redist0_count, redist1_capacity));
 729        }
 730    } else {
 731        if (!kvm_irqchip_in_kernel()) {
 732            qdev_prop_set_bit(vms->gic, "has-virtualization-extensions",
 733                              vms->virt);
 734        }
 735    }
 736    gicbusdev = SYS_BUS_DEVICE(vms->gic);
 737    sysbus_realize_and_unref(gicbusdev, &error_fatal);
 738    sysbus_mmio_map(gicbusdev, 0, vms->memmap[VIRT_GIC_DIST].base);
 739    if (type == 3) {
 740        sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_REDIST].base);
 741        if (nb_redist_regions == 2) {
 742            sysbus_mmio_map(gicbusdev, 2,
 743                            vms->memmap[VIRT_HIGH_GIC_REDIST2].base);
 744        }
 745    } else {
 746        sysbus_mmio_map(gicbusdev, 1, vms->memmap[VIRT_GIC_CPU].base);
 747        if (vms->virt) {
 748            sysbus_mmio_map(gicbusdev, 2, vms->memmap[VIRT_GIC_HYP].base);
 749            sysbus_mmio_map(gicbusdev, 3, vms->memmap[VIRT_GIC_VCPU].base);
 750        }
 751    }
 752
 753    /* Wire the outputs from each CPU's generic timer and the GICv3
 754     * maintenance interrupt signal to the appropriate GIC PPI inputs,
 755     * and the GIC's IRQ/FIQ/VIRQ/VFIQ interrupt outputs to the CPU's inputs.
 756     */
 757    for (i = 0; i < smp_cpus; i++) {
 758        DeviceState *cpudev = DEVICE(qemu_get_cpu(i));
 759        int ppibase = NUM_IRQS + i * GIC_INTERNAL + GIC_NR_SGIS;
 760        int irq;
 761        /* Mapping from the output timer irq lines from the CPU to the
 762         * GIC PPI inputs we use for the virt board.
 763         */
 764        const int timer_irq[] = {
 765            [GTIMER_PHYS] = ARCH_TIMER_NS_EL1_IRQ,
 766            [GTIMER_VIRT] = ARCH_TIMER_VIRT_IRQ,
 767            [GTIMER_HYP]  = ARCH_TIMER_NS_EL2_IRQ,
 768            [GTIMER_SEC]  = ARCH_TIMER_S_EL1_IRQ,
 769        };
 770
 771        for (irq = 0; irq < ARRAY_SIZE(timer_irq); irq++) {
 772            qdev_connect_gpio_out(cpudev, irq,
 773                                  qdev_get_gpio_in(vms->gic,
 774                                                   ppibase + timer_irq[irq]));
 775        }
 776
 777        if (type == 3) {
 778            qemu_irq irq = qdev_get_gpio_in(vms->gic,
 779                                            ppibase + ARCH_GIC_MAINT_IRQ);
 780            qdev_connect_gpio_out_named(cpudev, "gicv3-maintenance-interrupt",
 781                                        0, irq);
 782        } else if (vms->virt) {
 783            qemu_irq irq = qdev_get_gpio_in(vms->gic,
 784                                            ppibase + ARCH_GIC_MAINT_IRQ);
 785            sysbus_connect_irq(gicbusdev, i + 4 * smp_cpus, irq);
 786        }
 787
 788        qdev_connect_gpio_out_named(cpudev, "pmu-interrupt", 0,
 789                                    qdev_get_gpio_in(vms->gic, ppibase
 790                                                     + VIRTUAL_PMU_IRQ));
 791
 792        sysbus_connect_irq(gicbusdev, i, qdev_get_gpio_in(cpudev, ARM_CPU_IRQ));
 793        sysbus_connect_irq(gicbusdev, i + smp_cpus,
 794                           qdev_get_gpio_in(cpudev, ARM_CPU_FIQ));
 795        sysbus_connect_irq(gicbusdev, i + 2 * smp_cpus,
 796                           qdev_get_gpio_in(cpudev, ARM_CPU_VIRQ));
 797        sysbus_connect_irq(gicbusdev, i + 3 * smp_cpus,
 798                           qdev_get_gpio_in(cpudev, ARM_CPU_VFIQ));
 799    }
 800
 801    fdt_add_gic_node(vms);
 802
 803    if (type == 3 && vms->its) {
 804        create_its(vms);
 805    } else if (type == 2) {
 806        create_v2m(vms);
 807    }
 808}
 809
 810static void create_uart(const VirtMachineState *vms, int uart,
 811                        MemoryRegion *mem, Chardev *chr)
 812{
 813    char *nodename;
 814    hwaddr base = vms->memmap[uart].base;
 815    hwaddr size = vms->memmap[uart].size;
 816    int irq = vms->irqmap[uart];
 817    const char compat[] = "arm,pl011\0arm,primecell";
 818    const char clocknames[] = "uartclk\0apb_pclk";
 819    DeviceState *dev = qdev_new(TYPE_PL011);
 820    SysBusDevice *s = SYS_BUS_DEVICE(dev);
 821    MachineState *ms = MACHINE(vms);
 822
 823    qdev_prop_set_chr(dev, "chardev", chr);
 824    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
 825    memory_region_add_subregion(mem, base,
 826                                sysbus_mmio_get_region(s, 0));
 827    sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
 828
 829    nodename = g_strdup_printf("/pl011@%" PRIx64, base);
 830    qemu_fdt_add_subnode(ms->fdt, nodename);
 831    /* Note that we can't use setprop_string because of the embedded NUL */
 832    qemu_fdt_setprop(ms->fdt, nodename, "compatible",
 833                         compat, sizeof(compat));
 834    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 835                                     2, base, 2, size);
 836    qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
 837                               GIC_FDT_IRQ_TYPE_SPI, irq,
 838                               GIC_FDT_IRQ_FLAGS_LEVEL_HI);
 839    qemu_fdt_setprop_cells(ms->fdt, nodename, "clocks",
 840                               vms->clock_phandle, vms->clock_phandle);
 841    qemu_fdt_setprop(ms->fdt, nodename, "clock-names",
 842                         clocknames, sizeof(clocknames));
 843
 844    if (uart == VIRT_UART) {
 845        qemu_fdt_setprop_string(ms->fdt, "/chosen", "stdout-path", nodename);
 846    } else {
 847        /* Mark as not usable by the normal world */
 848        qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
 849        qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
 850
 851        qemu_fdt_setprop_string(ms->fdt, "/secure-chosen", "stdout-path",
 852                                nodename);
 853    }
 854
 855    g_free(nodename);
 856}
 857
 858static void create_rtc(const VirtMachineState *vms)
 859{
 860    char *nodename;
 861    hwaddr base = vms->memmap[VIRT_RTC].base;
 862    hwaddr size = vms->memmap[VIRT_RTC].size;
 863    int irq = vms->irqmap[VIRT_RTC];
 864    const char compat[] = "arm,pl031\0arm,primecell";
 865    MachineState *ms = MACHINE(vms);
 866
 867    sysbus_create_simple("pl031", base, qdev_get_gpio_in(vms->gic, irq));
 868
 869    nodename = g_strdup_printf("/pl031@%" PRIx64, base);
 870    qemu_fdt_add_subnode(ms->fdt, nodename);
 871    qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
 872    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 873                                 2, base, 2, size);
 874    qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
 875                           GIC_FDT_IRQ_TYPE_SPI, irq,
 876                           GIC_FDT_IRQ_FLAGS_LEVEL_HI);
 877    qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
 878    qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
 879    g_free(nodename);
 880}
 881
 882static DeviceState *gpio_key_dev;
 883static void virt_powerdown_req(Notifier *n, void *opaque)
 884{
 885    VirtMachineState *s = container_of(n, VirtMachineState, powerdown_notifier);
 886
 887    if (s->acpi_dev) {
 888        acpi_send_event(s->acpi_dev, ACPI_POWER_DOWN_STATUS);
 889    } else {
 890        /* use gpio Pin 3 for power button event */
 891        qemu_set_irq(qdev_get_gpio_in(gpio_key_dev, 0), 1);
 892    }
 893}
 894
 895static void create_gpio_keys(char *fdt, DeviceState *pl061_dev,
 896                             uint32_t phandle)
 897{
 898    gpio_key_dev = sysbus_create_simple("gpio-key", -1,
 899                                        qdev_get_gpio_in(pl061_dev, 3));
 900
 901    qemu_fdt_add_subnode(fdt, "/gpio-keys");
 902    qemu_fdt_setprop_string(fdt, "/gpio-keys", "compatible", "gpio-keys");
 903    qemu_fdt_setprop_cell(fdt, "/gpio-keys", "#size-cells", 0);
 904    qemu_fdt_setprop_cell(fdt, "/gpio-keys", "#address-cells", 1);
 905
 906    qemu_fdt_add_subnode(fdt, "/gpio-keys/poweroff");
 907    qemu_fdt_setprop_string(fdt, "/gpio-keys/poweroff",
 908                            "label", "GPIO Key Poweroff");
 909    qemu_fdt_setprop_cell(fdt, "/gpio-keys/poweroff", "linux,code",
 910                          KEY_POWER);
 911    qemu_fdt_setprop_cells(fdt, "/gpio-keys/poweroff",
 912                           "gpios", phandle, 3, 0);
 913}
 914
 915#define SECURE_GPIO_POWEROFF 0
 916#define SECURE_GPIO_RESET    1
 917
 918static void create_secure_gpio_pwr(char *fdt, DeviceState *pl061_dev,
 919                                   uint32_t phandle)
 920{
 921    DeviceState *gpio_pwr_dev;
 922
 923    /* gpio-pwr */
 924    gpio_pwr_dev = sysbus_create_simple("gpio-pwr", -1, NULL);
 925
 926    /* connect secure pl061 to gpio-pwr */
 927    qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_RESET,
 928                          qdev_get_gpio_in_named(gpio_pwr_dev, "reset", 0));
 929    qdev_connect_gpio_out(pl061_dev, SECURE_GPIO_POWEROFF,
 930                          qdev_get_gpio_in_named(gpio_pwr_dev, "shutdown", 0));
 931
 932    qemu_fdt_add_subnode(fdt, "/gpio-poweroff");
 933    qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "compatible",
 934                            "gpio-poweroff");
 935    qemu_fdt_setprop_cells(fdt, "/gpio-poweroff",
 936                           "gpios", phandle, SECURE_GPIO_POWEROFF, 0);
 937    qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "status", "disabled");
 938    qemu_fdt_setprop_string(fdt, "/gpio-poweroff", "secure-status",
 939                            "okay");
 940
 941    qemu_fdt_add_subnode(fdt, "/gpio-restart");
 942    qemu_fdt_setprop_string(fdt, "/gpio-restart", "compatible",
 943                            "gpio-restart");
 944    qemu_fdt_setprop_cells(fdt, "/gpio-restart",
 945                           "gpios", phandle, SECURE_GPIO_RESET, 0);
 946    qemu_fdt_setprop_string(fdt, "/gpio-restart", "status", "disabled");
 947    qemu_fdt_setprop_string(fdt, "/gpio-restart", "secure-status",
 948                            "okay");
 949}
 950
 951static void create_gpio_devices(const VirtMachineState *vms, int gpio,
 952                                MemoryRegion *mem)
 953{
 954    char *nodename;
 955    DeviceState *pl061_dev;
 956    hwaddr base = vms->memmap[gpio].base;
 957    hwaddr size = vms->memmap[gpio].size;
 958    int irq = vms->irqmap[gpio];
 959    const char compat[] = "arm,pl061\0arm,primecell";
 960    SysBusDevice *s;
 961    MachineState *ms = MACHINE(vms);
 962
 963    pl061_dev = qdev_new("pl061");
 964    /* Pull lines down to 0 if not driven by the PL061 */
 965    qdev_prop_set_uint32(pl061_dev, "pullups", 0);
 966    qdev_prop_set_uint32(pl061_dev, "pulldowns", 0xff);
 967    s = SYS_BUS_DEVICE(pl061_dev);
 968    sysbus_realize_and_unref(s, &error_fatal);
 969    memory_region_add_subregion(mem, base, sysbus_mmio_get_region(s, 0));
 970    sysbus_connect_irq(s, 0, qdev_get_gpio_in(vms->gic, irq));
 971
 972    uint32_t phandle = qemu_fdt_alloc_phandle(ms->fdt);
 973    nodename = g_strdup_printf("/pl061@%" PRIx64, base);
 974    qemu_fdt_add_subnode(ms->fdt, nodename);
 975    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
 976                                 2, base, 2, size);
 977    qemu_fdt_setprop(ms->fdt, nodename, "compatible", compat, sizeof(compat));
 978    qemu_fdt_setprop_cell(ms->fdt, nodename, "#gpio-cells", 2);
 979    qemu_fdt_setprop(ms->fdt, nodename, "gpio-controller", NULL, 0);
 980    qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
 981                           GIC_FDT_IRQ_TYPE_SPI, irq,
 982                           GIC_FDT_IRQ_FLAGS_LEVEL_HI);
 983    qemu_fdt_setprop_cell(ms->fdt, nodename, "clocks", vms->clock_phandle);
 984    qemu_fdt_setprop_string(ms->fdt, nodename, "clock-names", "apb_pclk");
 985    qemu_fdt_setprop_cell(ms->fdt, nodename, "phandle", phandle);
 986
 987    if (gpio != VIRT_GPIO) {
 988        /* Mark as not usable by the normal world */
 989        qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
 990        qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
 991    }
 992    g_free(nodename);
 993
 994    /* Child gpio devices */
 995    if (gpio == VIRT_GPIO) {
 996        create_gpio_keys(ms->fdt, pl061_dev, phandle);
 997    } else {
 998        create_secure_gpio_pwr(ms->fdt, pl061_dev, phandle);
 999    }
1000}
1001
1002static void create_virtio_devices(const VirtMachineState *vms)
1003{
1004    int i;
1005    hwaddr size = vms->memmap[VIRT_MMIO].size;
1006    MachineState *ms = MACHINE(vms);
1007
1008    /* We create the transports in forwards order. Since qbus_realize()
1009     * prepends (not appends) new child buses, the incrementing loop below will
1010     * create a list of virtio-mmio buses with decreasing base addresses.
1011     *
1012     * When a -device option is processed from the command line,
1013     * qbus_find_recursive() picks the next free virtio-mmio bus in forwards
1014     * order. The upshot is that -device options in increasing command line
1015     * order are mapped to virtio-mmio buses with decreasing base addresses.
1016     *
1017     * When this code was originally written, that arrangement ensured that the
1018     * guest Linux kernel would give the lowest "name" (/dev/vda, eth0, etc) to
1019     * the first -device on the command line. (The end-to-end order is a
1020     * function of this loop, qbus_realize(), qbus_find_recursive(), and the
1021     * guest kernel's name-to-address assignment strategy.)
1022     *
1023     * Meanwhile, the kernel's traversal seems to have been reversed; see eg.
1024     * the message, if not necessarily the code, of commit 70161ff336.
1025     * Therefore the loop now establishes the inverse of the original intent.
1026     *
1027     * Unfortunately, we can't counteract the kernel change by reversing the
1028     * loop; it would break existing command lines.
1029     *
1030     * In any case, the kernel makes no guarantee about the stability of
1031     * enumeration order of virtio devices (as demonstrated by it changing
1032     * between kernel versions). For reliable and stable identification
1033     * of disks users must use UUIDs or similar mechanisms.
1034     */
1035    for (i = 0; i < NUM_VIRTIO_TRANSPORTS; i++) {
1036        int irq = vms->irqmap[VIRT_MMIO] + i;
1037        hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1038
1039        sysbus_create_simple("virtio-mmio", base,
1040                             qdev_get_gpio_in(vms->gic, irq));
1041    }
1042
1043    /* We add dtb nodes in reverse order so that they appear in the finished
1044     * device tree lowest address first.
1045     *
1046     * Note that this mapping is independent of the loop above. The previous
1047     * loop influences virtio device to virtio transport assignment, whereas
1048     * this loop controls how virtio transports are laid out in the dtb.
1049     */
1050    for (i = NUM_VIRTIO_TRANSPORTS - 1; i >= 0; i--) {
1051        char *nodename;
1052        int irq = vms->irqmap[VIRT_MMIO] + i;
1053        hwaddr base = vms->memmap[VIRT_MMIO].base + i * size;
1054
1055        nodename = g_strdup_printf("/virtio_mmio@%" PRIx64, base);
1056        qemu_fdt_add_subnode(ms->fdt, nodename);
1057        qemu_fdt_setprop_string(ms->fdt, nodename,
1058                                "compatible", "virtio,mmio");
1059        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1060                                     2, base, 2, size);
1061        qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupts",
1062                               GIC_FDT_IRQ_TYPE_SPI, irq,
1063                               GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1064        qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1065        g_free(nodename);
1066    }
1067}
1068
1069#define VIRT_FLASH_SECTOR_SIZE (256 * KiB)
1070
1071static PFlashCFI01 *virt_flash_create1(VirtMachineState *vms,
1072                                        const char *name,
1073                                        const char *alias_prop_name)
1074{
1075    /*
1076     * Create a single flash device.  We use the same parameters as
1077     * the flash devices on the Versatile Express board.
1078     */
1079    DeviceState *dev = qdev_new(TYPE_PFLASH_CFI01);
1080
1081    qdev_prop_set_uint64(dev, "sector-length", VIRT_FLASH_SECTOR_SIZE);
1082    qdev_prop_set_uint8(dev, "width", 4);
1083    qdev_prop_set_uint8(dev, "device-width", 2);
1084    qdev_prop_set_bit(dev, "big-endian", false);
1085    qdev_prop_set_uint16(dev, "id0", 0x89);
1086    qdev_prop_set_uint16(dev, "id1", 0x18);
1087    qdev_prop_set_uint16(dev, "id2", 0x00);
1088    qdev_prop_set_uint16(dev, "id3", 0x00);
1089    qdev_prop_set_string(dev, "name", name);
1090    object_property_add_child(OBJECT(vms), name, OBJECT(dev));
1091    object_property_add_alias(OBJECT(vms), alias_prop_name,
1092                              OBJECT(dev), "drive");
1093    return PFLASH_CFI01(dev);
1094}
1095
1096static void virt_flash_create(VirtMachineState *vms)
1097{
1098    vms->flash[0] = virt_flash_create1(vms, "virt.flash0", "pflash0");
1099    vms->flash[1] = virt_flash_create1(vms, "virt.flash1", "pflash1");
1100}
1101
1102static void virt_flash_map1(PFlashCFI01 *flash,
1103                            hwaddr base, hwaddr size,
1104                            MemoryRegion *sysmem)
1105{
1106    DeviceState *dev = DEVICE(flash);
1107
1108    assert(QEMU_IS_ALIGNED(size, VIRT_FLASH_SECTOR_SIZE));
1109    assert(size / VIRT_FLASH_SECTOR_SIZE <= UINT32_MAX);
1110    qdev_prop_set_uint32(dev, "num-blocks", size / VIRT_FLASH_SECTOR_SIZE);
1111    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1112
1113    memory_region_add_subregion(sysmem, base,
1114                                sysbus_mmio_get_region(SYS_BUS_DEVICE(dev),
1115                                                       0));
1116}
1117
1118static void virt_flash_map(VirtMachineState *vms,
1119                           MemoryRegion *sysmem,
1120                           MemoryRegion *secure_sysmem)
1121{
1122    /*
1123     * Map two flash devices to fill the VIRT_FLASH space in the memmap.
1124     * sysmem is the system memory space. secure_sysmem is the secure view
1125     * of the system, and the first flash device should be made visible only
1126     * there. The second flash device is visible to both secure and nonsecure.
1127     * If sysmem == secure_sysmem this means there is no separate Secure
1128     * address space and both flash devices are generally visible.
1129     */
1130    hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1131    hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1132
1133    virt_flash_map1(vms->flash[0], flashbase, flashsize,
1134                    secure_sysmem);
1135    virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
1136                    sysmem);
1137}
1138
1139static void virt_flash_fdt(VirtMachineState *vms,
1140                           MemoryRegion *sysmem,
1141                           MemoryRegion *secure_sysmem)
1142{
1143    hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
1144    hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
1145    MachineState *ms = MACHINE(vms);
1146    char *nodename;
1147
1148    if (sysmem == secure_sysmem) {
1149        /* Report both flash devices as a single node in the DT */
1150        nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
1151        qemu_fdt_add_subnode(ms->fdt, nodename);
1152        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1153        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1154                                     2, flashbase, 2, flashsize,
1155                                     2, flashbase + flashsize, 2, flashsize);
1156        qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1157        g_free(nodename);
1158    } else {
1159        /*
1160         * Report the devices as separate nodes so we can mark one as
1161         * only visible to the secure world.
1162         */
1163        nodename = g_strdup_printf("/secflash@%" PRIx64, flashbase);
1164        qemu_fdt_add_subnode(ms->fdt, nodename);
1165        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1166        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1167                                     2, flashbase, 2, flashsize);
1168        qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1169        qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1170        qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1171        g_free(nodename);
1172
1173        nodename = g_strdup_printf("/flash@%" PRIx64, flashbase);
1174        qemu_fdt_add_subnode(ms->fdt, nodename);
1175        qemu_fdt_setprop_string(ms->fdt, nodename, "compatible", "cfi-flash");
1176        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1177                                     2, flashbase + flashsize, 2, flashsize);
1178        qemu_fdt_setprop_cell(ms->fdt, nodename, "bank-width", 4);
1179        g_free(nodename);
1180    }
1181}
1182
1183static bool virt_firmware_init(VirtMachineState *vms,
1184                               MemoryRegion *sysmem,
1185                               MemoryRegion *secure_sysmem)
1186{
1187    int i;
1188    const char *bios_name;
1189    BlockBackend *pflash_blk0;
1190
1191    /* Map legacy -drive if=pflash to machine properties */
1192    for (i = 0; i < ARRAY_SIZE(vms->flash); i++) {
1193        pflash_cfi01_legacy_drive(vms->flash[i],
1194                                  drive_get(IF_PFLASH, 0, i));
1195    }
1196
1197    virt_flash_map(vms, sysmem, secure_sysmem);
1198
1199    pflash_blk0 = pflash_cfi01_get_blk(vms->flash[0]);
1200
1201    bios_name = MACHINE(vms)->firmware;
1202    if (bios_name) {
1203        char *fname;
1204        MemoryRegion *mr;
1205        int image_size;
1206
1207        if (pflash_blk0) {
1208            error_report("The contents of the first flash device may be "
1209                         "specified with -bios or with -drive if=pflash... "
1210                         "but you cannot use both options at once");
1211            exit(1);
1212        }
1213
1214        /* Fall back to -bios */
1215
1216        fname = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1217        if (!fname) {
1218            error_report("Could not find ROM image '%s'", bios_name);
1219            exit(1);
1220        }
1221        mr = sysbus_mmio_get_region(SYS_BUS_DEVICE(vms->flash[0]), 0);
1222        image_size = load_image_mr(fname, mr);
1223        g_free(fname);
1224        if (image_size < 0) {
1225            error_report("Could not load ROM image '%s'", bios_name);
1226            exit(1);
1227        }
1228    }
1229
1230    return pflash_blk0 || bios_name;
1231}
1232
1233static FWCfgState *create_fw_cfg(const VirtMachineState *vms, AddressSpace *as)
1234{
1235    MachineState *ms = MACHINE(vms);
1236    hwaddr base = vms->memmap[VIRT_FW_CFG].base;
1237    hwaddr size = vms->memmap[VIRT_FW_CFG].size;
1238    FWCfgState *fw_cfg;
1239    char *nodename;
1240
1241    fw_cfg = fw_cfg_init_mem_wide(base + 8, base, 8, base + 16, as);
1242    fw_cfg_add_i16(fw_cfg, FW_CFG_NB_CPUS, (uint16_t)ms->smp.cpus);
1243
1244    nodename = g_strdup_printf("/fw-cfg@%" PRIx64, base);
1245    qemu_fdt_add_subnode(ms->fdt, nodename);
1246    qemu_fdt_setprop_string(ms->fdt, nodename,
1247                            "compatible", "qemu,fw-cfg-mmio");
1248    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1249                                 2, base, 2, size);
1250    qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1251    g_free(nodename);
1252    return fw_cfg;
1253}
1254
1255static void create_pcie_irq_map(const MachineState *ms,
1256                                uint32_t gic_phandle,
1257                                int first_irq, const char *nodename)
1258{
1259    int devfn, pin;
1260    uint32_t full_irq_map[4 * 4 * 10] = { 0 };
1261    uint32_t *irq_map = full_irq_map;
1262
1263    for (devfn = 0; devfn <= 0x18; devfn += 0x8) {
1264        for (pin = 0; pin < 4; pin++) {
1265            int irq_type = GIC_FDT_IRQ_TYPE_SPI;
1266            int irq_nr = first_irq + ((pin + PCI_SLOT(devfn)) % PCI_NUM_PINS);
1267            int irq_level = GIC_FDT_IRQ_FLAGS_LEVEL_HI;
1268            int i;
1269
1270            uint32_t map[] = {
1271                devfn << 8, 0, 0,                           /* devfn */
1272                pin + 1,                                    /* PCI pin */
1273                gic_phandle, 0, 0, irq_type, irq_nr, irq_level }; /* GIC irq */
1274
1275            /* Convert map to big endian */
1276            for (i = 0; i < 10; i++) {
1277                irq_map[i] = cpu_to_be32(map[i]);
1278            }
1279            irq_map += 10;
1280        }
1281    }
1282
1283    qemu_fdt_setprop(ms->fdt, nodename, "interrupt-map",
1284                     full_irq_map, sizeof(full_irq_map));
1285
1286    qemu_fdt_setprop_cells(ms->fdt, nodename, "interrupt-map-mask",
1287                           cpu_to_be16(PCI_DEVFN(3, 0)), /* Slot 3 */
1288                           0, 0,
1289                           0x7           /* PCI irq */);
1290}
1291
1292static void create_smmu(const VirtMachineState *vms,
1293                        PCIBus *bus)
1294{
1295    char *node;
1296    const char compat[] = "arm,smmu-v3";
1297    int irq =  vms->irqmap[VIRT_SMMU];
1298    int i;
1299    hwaddr base = vms->memmap[VIRT_SMMU].base;
1300    hwaddr size = vms->memmap[VIRT_SMMU].size;
1301    const char irq_names[] = "eventq\0priq\0cmdq-sync\0gerror";
1302    DeviceState *dev;
1303    MachineState *ms = MACHINE(vms);
1304
1305    if (vms->iommu != VIRT_IOMMU_SMMUV3 || !vms->iommu_phandle) {
1306        return;
1307    }
1308
1309    dev = qdev_new("arm-smmuv3");
1310
1311    object_property_set_link(OBJECT(dev), "primary-bus", OBJECT(bus),
1312                             &error_abort);
1313    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1314    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 0, base);
1315    for (i = 0; i < NUM_SMMU_IRQS; i++) {
1316        sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1317                           qdev_get_gpio_in(vms->gic, irq + i));
1318    }
1319
1320    node = g_strdup_printf("/smmuv3@%" PRIx64, base);
1321    qemu_fdt_add_subnode(ms->fdt, node);
1322    qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1323    qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg", 2, base, 2, size);
1324
1325    qemu_fdt_setprop_cells(ms->fdt, node, "interrupts",
1326            GIC_FDT_IRQ_TYPE_SPI, irq    , GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1327            GIC_FDT_IRQ_TYPE_SPI, irq + 1, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1328            GIC_FDT_IRQ_TYPE_SPI, irq + 2, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI,
1329            GIC_FDT_IRQ_TYPE_SPI, irq + 3, GIC_FDT_IRQ_FLAGS_EDGE_LO_HI);
1330
1331    qemu_fdt_setprop(ms->fdt, node, "interrupt-names", irq_names,
1332                     sizeof(irq_names));
1333
1334    qemu_fdt_setprop_cell(ms->fdt, node, "clocks", vms->clock_phandle);
1335    qemu_fdt_setprop_string(ms->fdt, node, "clock-names", "apb_pclk");
1336    qemu_fdt_setprop(ms->fdt, node, "dma-coherent", NULL, 0);
1337
1338    qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1339
1340    qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1341    g_free(node);
1342}
1343
1344static void create_virtio_iommu_dt_bindings(VirtMachineState *vms)
1345{
1346    const char compat[] = "virtio,pci-iommu";
1347    uint16_t bdf = vms->virtio_iommu_bdf;
1348    MachineState *ms = MACHINE(vms);
1349    char *node;
1350
1351    vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1352
1353    node = g_strdup_printf("%s/virtio_iommu@%d", vms->pciehb_nodename, bdf);
1354    qemu_fdt_add_subnode(ms->fdt, node);
1355    qemu_fdt_setprop(ms->fdt, node, "compatible", compat, sizeof(compat));
1356    qemu_fdt_setprop_sized_cells(ms->fdt, node, "reg",
1357                                 1, bdf << 8, 1, 0, 1, 0,
1358                                 1, 0, 1, 0);
1359
1360    qemu_fdt_setprop_cell(ms->fdt, node, "#iommu-cells", 1);
1361    qemu_fdt_setprop_cell(ms->fdt, node, "phandle", vms->iommu_phandle);
1362    g_free(node);
1363
1364    qemu_fdt_setprop_cells(ms->fdt, vms->pciehb_nodename, "iommu-map",
1365                           0x0, vms->iommu_phandle, 0x0, bdf,
1366                           bdf + 1, vms->iommu_phandle, bdf + 1, 0xffff - bdf);
1367}
1368
1369static void create_pcie(VirtMachineState *vms)
1370{
1371    hwaddr base_mmio = vms->memmap[VIRT_PCIE_MMIO].base;
1372    hwaddr size_mmio = vms->memmap[VIRT_PCIE_MMIO].size;
1373    hwaddr base_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].base;
1374    hwaddr size_mmio_high = vms->memmap[VIRT_HIGH_PCIE_MMIO].size;
1375    hwaddr base_pio = vms->memmap[VIRT_PCIE_PIO].base;
1376    hwaddr size_pio = vms->memmap[VIRT_PCIE_PIO].size;
1377    hwaddr base_ecam, size_ecam;
1378    hwaddr base = base_mmio;
1379    int nr_pcie_buses;
1380    int irq = vms->irqmap[VIRT_PCIE];
1381    MemoryRegion *mmio_alias;
1382    MemoryRegion *mmio_reg;
1383    MemoryRegion *ecam_alias;
1384    MemoryRegion *ecam_reg;
1385    DeviceState *dev;
1386    char *nodename;
1387    int i, ecam_id;
1388    PCIHostState *pci;
1389    MachineState *ms = MACHINE(vms);
1390
1391    dev = qdev_new(TYPE_GPEX_HOST);
1392    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1393
1394    ecam_id = VIRT_ECAM_ID(vms->highmem_ecam);
1395    base_ecam = vms->memmap[ecam_id].base;
1396    size_ecam = vms->memmap[ecam_id].size;
1397    nr_pcie_buses = size_ecam / PCIE_MMCFG_SIZE_MIN;
1398    /* Map only the first size_ecam bytes of ECAM space */
1399    ecam_alias = g_new0(MemoryRegion, 1);
1400    ecam_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 0);
1401    memory_region_init_alias(ecam_alias, OBJECT(dev), "pcie-ecam",
1402                             ecam_reg, 0, size_ecam);
1403    memory_region_add_subregion(get_system_memory(), base_ecam, ecam_alias);
1404
1405    /* Map the MMIO window into system address space so as to expose
1406     * the section of PCI MMIO space which starts at the same base address
1407     * (ie 1:1 mapping for that part of PCI MMIO space visible through
1408     * the window).
1409     */
1410    mmio_alias = g_new0(MemoryRegion, 1);
1411    mmio_reg = sysbus_mmio_get_region(SYS_BUS_DEVICE(dev), 1);
1412    memory_region_init_alias(mmio_alias, OBJECT(dev), "pcie-mmio",
1413                             mmio_reg, base_mmio, size_mmio);
1414    memory_region_add_subregion(get_system_memory(), base_mmio, mmio_alias);
1415
1416    if (vms->highmem) {
1417        /* Map high MMIO space */
1418        MemoryRegion *high_mmio_alias = g_new0(MemoryRegion, 1);
1419
1420        memory_region_init_alias(high_mmio_alias, OBJECT(dev), "pcie-mmio-high",
1421                                 mmio_reg, base_mmio_high, size_mmio_high);
1422        memory_region_add_subregion(get_system_memory(), base_mmio_high,
1423                                    high_mmio_alias);
1424    }
1425
1426    /* Map IO port space */
1427    sysbus_mmio_map(SYS_BUS_DEVICE(dev), 2, base_pio);
1428
1429    for (i = 0; i < GPEX_NUM_IRQS; i++) {
1430        sysbus_connect_irq(SYS_BUS_DEVICE(dev), i,
1431                           qdev_get_gpio_in(vms->gic, irq + i));
1432        gpex_set_irq_num(GPEX_HOST(dev), i, irq + i);
1433    }
1434
1435    pci = PCI_HOST_BRIDGE(dev);
1436    pci->bypass_iommu = vms->default_bus_bypass_iommu;
1437    vms->bus = pci->bus;
1438    if (vms->bus) {
1439        for (i = 0; i < nb_nics; i++) {
1440            NICInfo *nd = &nd_table[i];
1441
1442            if (!nd->model) {
1443                nd->model = g_strdup("virtio");
1444            }
1445
1446            pci_nic_init_nofail(nd, pci->bus, nd->model, NULL);
1447        }
1448    }
1449
1450    nodename = vms->pciehb_nodename = g_strdup_printf("/pcie@%" PRIx64, base);
1451    qemu_fdt_add_subnode(ms->fdt, nodename);
1452    qemu_fdt_setprop_string(ms->fdt, nodename,
1453                            "compatible", "pci-host-ecam-generic");
1454    qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "pci");
1455    qemu_fdt_setprop_cell(ms->fdt, nodename, "#address-cells", 3);
1456    qemu_fdt_setprop_cell(ms->fdt, nodename, "#size-cells", 2);
1457    qemu_fdt_setprop_cell(ms->fdt, nodename, "linux,pci-domain", 0);
1458    qemu_fdt_setprop_cells(ms->fdt, nodename, "bus-range", 0,
1459                           nr_pcie_buses - 1);
1460    qemu_fdt_setprop(ms->fdt, nodename, "dma-coherent", NULL, 0);
1461
1462    if (vms->msi_phandle) {
1463        qemu_fdt_setprop_cells(ms->fdt, nodename, "msi-parent",
1464                               vms->msi_phandle);
1465    }
1466
1467    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg",
1468                                 2, base_ecam, 2, size_ecam);
1469
1470    if (vms->highmem) {
1471        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1472                                     1, FDT_PCI_RANGE_IOPORT, 2, 0,
1473                                     2, base_pio, 2, size_pio,
1474                                     1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1475                                     2, base_mmio, 2, size_mmio,
1476                                     1, FDT_PCI_RANGE_MMIO_64BIT,
1477                                     2, base_mmio_high,
1478                                     2, base_mmio_high, 2, size_mmio_high);
1479    } else {
1480        qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "ranges",
1481                                     1, FDT_PCI_RANGE_IOPORT, 2, 0,
1482                                     2, base_pio, 2, size_pio,
1483                                     1, FDT_PCI_RANGE_MMIO, 2, base_mmio,
1484                                     2, base_mmio, 2, size_mmio);
1485    }
1486
1487    qemu_fdt_setprop_cell(ms->fdt, nodename, "#interrupt-cells", 1);
1488    create_pcie_irq_map(ms, vms->gic_phandle, irq, nodename);
1489
1490    if (vms->iommu) {
1491        vms->iommu_phandle = qemu_fdt_alloc_phandle(ms->fdt);
1492
1493        switch (vms->iommu) {
1494        case VIRT_IOMMU_SMMUV3:
1495            create_smmu(vms, vms->bus);
1496            qemu_fdt_setprop_cells(ms->fdt, nodename, "iommu-map",
1497                                   0x0, vms->iommu_phandle, 0x0, 0x10000);
1498            break;
1499        default:
1500            g_assert_not_reached();
1501        }
1502    }
1503}
1504
1505static void create_platform_bus(VirtMachineState *vms)
1506{
1507    DeviceState *dev;
1508    SysBusDevice *s;
1509    int i;
1510    MemoryRegion *sysmem = get_system_memory();
1511
1512    dev = qdev_new(TYPE_PLATFORM_BUS_DEVICE);
1513    dev->id = g_strdup(TYPE_PLATFORM_BUS_DEVICE);
1514    qdev_prop_set_uint32(dev, "num_irqs", PLATFORM_BUS_NUM_IRQS);
1515    qdev_prop_set_uint32(dev, "mmio_size", vms->memmap[VIRT_PLATFORM_BUS].size);
1516    sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
1517    vms->platform_bus_dev = dev;
1518
1519    s = SYS_BUS_DEVICE(dev);
1520    for (i = 0; i < PLATFORM_BUS_NUM_IRQS; i++) {
1521        int irq = vms->irqmap[VIRT_PLATFORM_BUS] + i;
1522        sysbus_connect_irq(s, i, qdev_get_gpio_in(vms->gic, irq));
1523    }
1524
1525    memory_region_add_subregion(sysmem,
1526                                vms->memmap[VIRT_PLATFORM_BUS].base,
1527                                sysbus_mmio_get_region(s, 0));
1528}
1529
1530static void create_tag_ram(MemoryRegion *tag_sysmem,
1531                           hwaddr base, hwaddr size,
1532                           const char *name)
1533{
1534    MemoryRegion *tagram = g_new(MemoryRegion, 1);
1535
1536    memory_region_init_ram(tagram, NULL, name, size / 32, &error_fatal);
1537    memory_region_add_subregion(tag_sysmem, base / 32, tagram);
1538}
1539
1540static void create_secure_ram(VirtMachineState *vms,
1541                              MemoryRegion *secure_sysmem,
1542                              MemoryRegion *secure_tag_sysmem)
1543{
1544    MemoryRegion *secram = g_new(MemoryRegion, 1);
1545    char *nodename;
1546    hwaddr base = vms->memmap[VIRT_SECURE_MEM].base;
1547    hwaddr size = vms->memmap[VIRT_SECURE_MEM].size;
1548    MachineState *ms = MACHINE(vms);
1549
1550    memory_region_init_ram(secram, NULL, "virt.secure-ram", size,
1551                           &error_fatal);
1552    memory_region_add_subregion(secure_sysmem, base, secram);
1553
1554    nodename = g_strdup_printf("/secram@%" PRIx64, base);
1555    qemu_fdt_add_subnode(ms->fdt, nodename);
1556    qemu_fdt_setprop_string(ms->fdt, nodename, "device_type", "memory");
1557    qemu_fdt_setprop_sized_cells(ms->fdt, nodename, "reg", 2, base, 2, size);
1558    qemu_fdt_setprop_string(ms->fdt, nodename, "status", "disabled");
1559    qemu_fdt_setprop_string(ms->fdt, nodename, "secure-status", "okay");
1560
1561    if (secure_tag_sysmem) {
1562        create_tag_ram(secure_tag_sysmem, base, size, "mach-virt.secure-tag");
1563    }
1564
1565    g_free(nodename);
1566}
1567
1568static void *machvirt_dtb(const struct arm_boot_info *binfo, int *fdt_size)
1569{
1570    const VirtMachineState *board = container_of(binfo, VirtMachineState,
1571                                                 bootinfo);
1572    MachineState *ms = MACHINE(board);
1573
1574
1575    *fdt_size = board->fdt_size;
1576    return ms->fdt;
1577}
1578
1579static void virt_build_smbios(VirtMachineState *vms)
1580{
1581    MachineClass *mc = MACHINE_GET_CLASS(vms);
1582    VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1583    uint8_t *smbios_tables, *smbios_anchor;
1584    size_t smbios_tables_len, smbios_anchor_len;
1585    const char *product = "QEMU Virtual Machine";
1586
1587    if (kvm_enabled()) {
1588        product = "KVM Virtual Machine";
1589    }
1590
1591    smbios_set_defaults("QEMU", product,
1592                        vmc->smbios_old_sys_ver ? "1.0" : mc->name, false,
1593                        true, SMBIOS_ENTRY_POINT_30);
1594
1595    smbios_get_tables(MACHINE(vms), NULL, 0,
1596                      &smbios_tables, &smbios_tables_len,
1597                      &smbios_anchor, &smbios_anchor_len,
1598                      &error_fatal);
1599
1600    if (smbios_anchor) {
1601        fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-tables",
1602                        smbios_tables, smbios_tables_len);
1603        fw_cfg_add_file(vms->fw_cfg, "etc/smbios/smbios-anchor",
1604                        smbios_anchor, smbios_anchor_len);
1605    }
1606}
1607
1608static
1609void virt_machine_done(Notifier *notifier, void *data)
1610{
1611    VirtMachineState *vms = container_of(notifier, VirtMachineState,
1612                                         machine_done);
1613    MachineState *ms = MACHINE(vms);
1614    ARMCPU *cpu = ARM_CPU(first_cpu);
1615    struct arm_boot_info *info = &vms->bootinfo;
1616    AddressSpace *as = arm_boot_address_space(cpu, info);
1617
1618    /*
1619     * If the user provided a dtb, we assume the dynamic sysbus nodes
1620     * already are integrated there. This corresponds to a use case where
1621     * the dynamic sysbus nodes are complex and their generation is not yet
1622     * supported. In that case the user can take charge of the guest dt
1623     * while qemu takes charge of the qom stuff.
1624     */
1625    if (info->dtb_filename == NULL) {
1626        platform_bus_add_all_fdt_nodes(ms->fdt, "/intc",
1627                                       vms->memmap[VIRT_PLATFORM_BUS].base,
1628                                       vms->memmap[VIRT_PLATFORM_BUS].size,
1629                                       vms->irqmap[VIRT_PLATFORM_BUS]);
1630    }
1631    if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as, ms) < 0) {
1632        exit(1);
1633    }
1634
1635    fw_cfg_add_extra_pci_roots(vms->bus, vms->fw_cfg);
1636
1637    virt_acpi_setup(vms);
1638    virt_build_smbios(vms);
1639}
1640
1641static uint64_t virt_cpu_mp_affinity(VirtMachineState *vms, int idx)
1642{
1643    uint8_t clustersz = ARM_DEFAULT_CPUS_PER_CLUSTER;
1644    VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
1645
1646    if (!vmc->disallow_affinity_adjustment) {
1647        /* Adjust MPIDR like 64-bit KVM hosts, which incorporate the
1648         * GIC's target-list limitations. 32-bit KVM hosts currently
1649         * always create clusters of 4 CPUs, but that is expected to
1650         * change when they gain support for gicv3. When KVM is enabled
1651         * it will override the changes we make here, therefore our
1652         * purposes are to make TCG consistent (with 64-bit KVM hosts)
1653         * and to improve SGI efficiency.
1654         */
1655        if (vms->gic_version == VIRT_GIC_VERSION_3) {
1656            clustersz = GICV3_TARGETLIST_BITS;
1657        } else {
1658            clustersz = GIC_TARGETLIST_BITS;
1659        }
1660    }
1661    return arm_cpu_mp_affinity(idx, clustersz);
1662}
1663
1664static void virt_set_memmap(VirtMachineState *vms)
1665{
1666    MachineState *ms = MACHINE(vms);
1667    hwaddr base, device_memory_base, device_memory_size;
1668    int i;
1669
1670    vms->memmap = extended_memmap;
1671
1672    for (i = 0; i < ARRAY_SIZE(base_memmap); i++) {
1673        vms->memmap[i] = base_memmap[i];
1674    }
1675
1676    if (ms->ram_slots > ACPI_MAX_RAM_SLOTS) {
1677        error_report("unsupported number of memory slots: %"PRIu64,
1678                     ms->ram_slots);
1679        exit(EXIT_FAILURE);
1680    }
1681
1682    /*
1683     * We compute the base of the high IO region depending on the
1684     * amount of initial and device memory. The device memory start/size
1685     * is aligned on 1GiB. We never put the high IO region below 256GiB
1686     * so that if maxram_size is < 255GiB we keep the legacy memory map.
1687     * The device region size assumes 1GiB page max alignment per slot.
1688     */
1689    device_memory_base =
1690        ROUND_UP(vms->memmap[VIRT_MEM].base + ms->ram_size, GiB);
1691    device_memory_size = ms->maxram_size - ms->ram_size + ms->ram_slots * GiB;
1692
1693    /* Base address of the high IO region */
1694    base = device_memory_base + ROUND_UP(device_memory_size, GiB);
1695    if (base < device_memory_base) {
1696        error_report("maxmem/slots too huge");
1697        exit(EXIT_FAILURE);
1698    }
1699    if (base < vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES) {
1700        base = vms->memmap[VIRT_MEM].base + LEGACY_RAMLIMIT_BYTES;
1701    }
1702
1703    for (i = VIRT_LOWMEMMAP_LAST; i < ARRAY_SIZE(extended_memmap); i++) {
1704        hwaddr size = extended_memmap[i].size;
1705
1706        base = ROUND_UP(base, size);
1707        vms->memmap[i].base = base;
1708        vms->memmap[i].size = size;
1709        base += size;
1710    }
1711    vms->highest_gpa = base - 1;
1712    if (device_memory_size > 0) {
1713        ms->device_memory = g_malloc0(sizeof(*ms->device_memory));
1714        ms->device_memory->base = device_memory_base;
1715        memory_region_init(&ms->device_memory->mr, OBJECT(vms),
1716                           "device-memory", device_memory_size);
1717    }
1718}
1719
1720/*
1721 * finalize_gic_version - Determines the final gic_version
1722 * according to the gic-version property
1723 *
1724 * Default GIC type is v2
1725 */
1726static void finalize_gic_version(VirtMachineState *vms)
1727{
1728    unsigned int max_cpus = MACHINE(vms)->smp.max_cpus;
1729
1730    if (kvm_enabled()) {
1731        int probe_bitmap;
1732
1733        if (!kvm_irqchip_in_kernel()) {
1734            switch (vms->gic_version) {
1735            case VIRT_GIC_VERSION_HOST:
1736                warn_report(
1737                    "gic-version=host not relevant with kernel-irqchip=off "
1738                     "as only userspace GICv2 is supported. Using v2 ...");
1739                return;
1740            case VIRT_GIC_VERSION_MAX:
1741            case VIRT_GIC_VERSION_NOSEL:
1742                vms->gic_version = VIRT_GIC_VERSION_2;
1743                return;
1744            case VIRT_GIC_VERSION_2:
1745                return;
1746            case VIRT_GIC_VERSION_3:
1747                error_report(
1748                    "gic-version=3 is not supported with kernel-irqchip=off");
1749                exit(1);
1750            }
1751        }
1752
1753        probe_bitmap = kvm_arm_vgic_probe();
1754        if (!probe_bitmap) {
1755            error_report("Unable to determine GIC version supported by host");
1756            exit(1);
1757        }
1758
1759        switch (vms->gic_version) {
1760        case VIRT_GIC_VERSION_HOST:
1761        case VIRT_GIC_VERSION_MAX:
1762            if (probe_bitmap & KVM_ARM_VGIC_V3) {
1763                vms->gic_version = VIRT_GIC_VERSION_3;
1764            } else {
1765                vms->gic_version = VIRT_GIC_VERSION_2;
1766            }
1767            return;
1768        case VIRT_GIC_VERSION_NOSEL:
1769            if ((probe_bitmap & KVM_ARM_VGIC_V2) && max_cpus <= GIC_NCPU) {
1770                vms->gic_version = VIRT_GIC_VERSION_2;
1771            } else if (probe_bitmap & KVM_ARM_VGIC_V3) {
1772                /*
1773                 * in case the host does not support v2 in-kernel emulation or
1774                 * the end-user requested more than 8 VCPUs we now default
1775                 * to v3. In any case defaulting to v2 would be broken.
1776                 */
1777                vms->gic_version = VIRT_GIC_VERSION_3;
1778            } else if (max_cpus > GIC_NCPU) {
1779                error_report("host only supports in-kernel GICv2 emulation "
1780                             "but more than 8 vcpus are requested");
1781                exit(1);
1782            }
1783            break;
1784        case VIRT_GIC_VERSION_2:
1785        case VIRT_GIC_VERSION_3:
1786            break;
1787        }
1788
1789        /* Check chosen version is effectively supported by the host */
1790        if (vms->gic_version == VIRT_GIC_VERSION_2 &&
1791            !(probe_bitmap & KVM_ARM_VGIC_V2)) {
1792            error_report("host does not support in-kernel GICv2 emulation");
1793            exit(1);
1794        } else if (vms->gic_version == VIRT_GIC_VERSION_3 &&
1795                   !(probe_bitmap & KVM_ARM_VGIC_V3)) {
1796            error_report("host does not support in-kernel GICv3 emulation");
1797            exit(1);
1798        }
1799        return;
1800    }
1801
1802    /* TCG mode */
1803    switch (vms->gic_version) {
1804    case VIRT_GIC_VERSION_NOSEL:
1805        vms->gic_version = VIRT_GIC_VERSION_2;
1806        break;
1807    case VIRT_GIC_VERSION_MAX:
1808        vms->gic_version = VIRT_GIC_VERSION_3;
1809        break;
1810    case VIRT_GIC_VERSION_HOST:
1811        error_report("gic-version=host requires KVM");
1812        exit(1);
1813    case VIRT_GIC_VERSION_2:
1814    case VIRT_GIC_VERSION_3:
1815        break;
1816    }
1817}
1818
1819/*
1820 * virt_cpu_post_init() must be called after the CPUs have
1821 * been realized and the GIC has been created.
1822 */
1823static void virt_cpu_post_init(VirtMachineState *vms, MemoryRegion *sysmem)
1824{
1825    int max_cpus = MACHINE(vms)->smp.max_cpus;
1826    bool aarch64, pmu, steal_time;
1827    CPUState *cpu;
1828
1829    aarch64 = object_property_get_bool(OBJECT(first_cpu), "aarch64", NULL);
1830    pmu = object_property_get_bool(OBJECT(first_cpu), "pmu", NULL);
1831    steal_time = object_property_get_bool(OBJECT(first_cpu),
1832                                          "kvm-steal-time", NULL);
1833
1834    if (kvm_enabled()) {
1835        hwaddr pvtime_reg_base = vms->memmap[VIRT_PVTIME].base;
1836        hwaddr pvtime_reg_size = vms->memmap[VIRT_PVTIME].size;
1837
1838        if (steal_time) {
1839            MemoryRegion *pvtime = g_new(MemoryRegion, 1);
1840            hwaddr pvtime_size = max_cpus * PVTIME_SIZE_PER_CPU;
1841
1842            /* The memory region size must be a multiple of host page size. */
1843            pvtime_size = REAL_HOST_PAGE_ALIGN(pvtime_size);
1844
1845            if (pvtime_size > pvtime_reg_size) {
1846                error_report("pvtime requires a %" HWADDR_PRId
1847                             " byte memory region for %d CPUs,"
1848                             " but only %" HWADDR_PRId " has been reserved",
1849                             pvtime_size, max_cpus, pvtime_reg_size);
1850                exit(1);
1851            }
1852
1853            memory_region_init_ram(pvtime, NULL, "pvtime", pvtime_size, NULL);
1854            memory_region_add_subregion(sysmem, pvtime_reg_base, pvtime);
1855        }
1856
1857        CPU_FOREACH(cpu) {
1858            if (pmu) {
1859                assert(arm_feature(&ARM_CPU(cpu)->env, ARM_FEATURE_PMU));
1860                if (kvm_irqchip_in_kernel()) {
1861                    kvm_arm_pmu_set_irq(cpu, PPI(VIRTUAL_PMU_IRQ));
1862                }
1863                kvm_arm_pmu_init(cpu);
1864            }
1865            if (steal_time) {
1866                kvm_arm_pvtime_init(cpu, pvtime_reg_base +
1867                                         cpu->cpu_index * PVTIME_SIZE_PER_CPU);
1868            }
1869        }
1870    } else {
1871        if (aarch64 && vms->highmem) {
1872            int requested_pa_size = 64 - clz64(vms->highest_gpa);
1873            int pamax = arm_pamax(ARM_CPU(first_cpu));
1874
1875            if (pamax < requested_pa_size) {
1876                error_report("VCPU supports less PA bits (%d) than "
1877                             "requested by the memory map (%d)",
1878                             pamax, requested_pa_size);
1879                exit(1);
1880            }
1881        }
1882    }
1883}
1884
1885static void machvirt_init(MachineState *machine)
1886{
1887    VirtMachineState *vms = VIRT_MACHINE(machine);
1888    VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(machine);
1889    MachineClass *mc = MACHINE_GET_CLASS(machine);
1890    const CPUArchIdList *possible_cpus;
1891    MemoryRegion *sysmem = get_system_memory();
1892    MemoryRegion *secure_sysmem = NULL;
1893    MemoryRegion *tag_sysmem = NULL;
1894    MemoryRegion *secure_tag_sysmem = NULL;
1895    int n, virt_max_cpus;
1896    bool firmware_loaded;
1897    bool aarch64 = true;
1898    bool has_ged = !vmc->no_ged;
1899    unsigned int smp_cpus = machine->smp.cpus;
1900    unsigned int max_cpus = machine->smp.max_cpus;
1901
1902    /*
1903     * In accelerated mode, the memory map is computed earlier in kvm_type()
1904     * to create a VM with the right number of IPA bits.
1905     */
1906    if (!vms->memmap) {
1907        virt_set_memmap(vms);
1908    }
1909
1910    /* We can probe only here because during property set
1911     * KVM is not available yet
1912     */
1913    finalize_gic_version(vms);
1914
1915    if (!cpu_type_valid(machine->cpu_type)) {
1916        error_report("mach-virt: CPU type %s not supported", machine->cpu_type);
1917        exit(1);
1918    }
1919
1920    if (vms->secure) {
1921        /*
1922         * The Secure view of the world is the same as the NonSecure,
1923         * but with a few extra devices. Create it as a container region
1924         * containing the system memory at low priority; any secure-only
1925         * devices go in at higher priority and take precedence.
1926         */
1927        secure_sysmem = g_new(MemoryRegion, 1);
1928        memory_region_init(secure_sysmem, OBJECT(machine), "secure-memory",
1929                           UINT64_MAX);
1930        memory_region_add_subregion_overlap(secure_sysmem, 0, sysmem, -1);
1931    }
1932
1933    firmware_loaded = virt_firmware_init(vms, sysmem,
1934                                         secure_sysmem ?: sysmem);
1935
1936    /* If we have an EL3 boot ROM then the assumption is that it will
1937     * implement PSCI itself, so disable QEMU's internal implementation
1938     * so it doesn't get in the way. Instead of starting secondary
1939     * CPUs in PSCI powerdown state we will start them all running and
1940     * let the boot ROM sort them out.
1941     * The usual case is that we do use QEMU's PSCI implementation;
1942     * if the guest has EL2 then we will use SMC as the conduit,
1943     * and otherwise we will use HVC (for backwards compatibility and
1944     * because if we're using KVM then we must use HVC).
1945     */
1946    if (vms->secure && firmware_loaded) {
1947        vms->psci_conduit = QEMU_PSCI_CONDUIT_DISABLED;
1948    } else if (vms->virt) {
1949        vms->psci_conduit = QEMU_PSCI_CONDUIT_SMC;
1950    } else {
1951        vms->psci_conduit = QEMU_PSCI_CONDUIT_HVC;
1952    }
1953
1954    /* The maximum number of CPUs depends on the GIC version, or on how
1955     * many redistributors we can fit into the memory map.
1956     */
1957    if (vms->gic_version == VIRT_GIC_VERSION_3) {
1958        virt_max_cpus =
1959            vms->memmap[VIRT_GIC_REDIST].size / GICV3_REDIST_SIZE;
1960        virt_max_cpus +=
1961            vms->memmap[VIRT_HIGH_GIC_REDIST2].size / GICV3_REDIST_SIZE;
1962    } else {
1963        virt_max_cpus = GIC_NCPU;
1964    }
1965
1966    if (max_cpus > virt_max_cpus) {
1967        error_report("Number of SMP CPUs requested (%d) exceeds max CPUs "
1968                     "supported by machine 'mach-virt' (%d)",
1969                     max_cpus, virt_max_cpus);
1970        exit(1);
1971    }
1972
1973    if (vms->virt && (kvm_enabled() || hvf_enabled())) {
1974        error_report("mach-virt: %s does not support providing "
1975                     "Virtualization extensions to the guest CPU",
1976                     kvm_enabled() ? "KVM" : "HVF");
1977        exit(1);
1978    }
1979
1980    if (vms->mte && (kvm_enabled() || hvf_enabled())) {
1981        error_report("mach-virt: %s does not support providing "
1982                     "MTE to the guest CPU",
1983                     kvm_enabled() ? "KVM" : "HVF");
1984        exit(1);
1985    }
1986
1987    create_fdt(vms);
1988
1989    possible_cpus = mc->possible_cpu_arch_ids(machine);
1990    assert(possible_cpus->len == max_cpus);
1991    for (n = 0; n < possible_cpus->len; n++) {
1992        Object *cpuobj;
1993        CPUState *cs;
1994
1995        if (n >= smp_cpus) {
1996            break;
1997        }
1998
1999        cpuobj = object_new(possible_cpus->cpus[n].type);
2000        object_property_set_int(cpuobj, "mp-affinity",
2001                                possible_cpus->cpus[n].arch_id, NULL);
2002
2003        cs = CPU(cpuobj);
2004        cs->cpu_index = n;
2005
2006        numa_cpu_pre_plug(&possible_cpus->cpus[cs->cpu_index], DEVICE(cpuobj),
2007                          &error_fatal);
2008
2009        aarch64 &= object_property_get_bool(cpuobj, "aarch64", NULL);
2010
2011        if (!vms->secure) {
2012            object_property_set_bool(cpuobj, "has_el3", false, NULL);
2013        }
2014
2015        if (!vms->virt && object_property_find(cpuobj, "has_el2")) {
2016            object_property_set_bool(cpuobj, "has_el2", false, NULL);
2017        }
2018
2019        if (vms->psci_conduit != QEMU_PSCI_CONDUIT_DISABLED) {
2020            object_property_set_int(cpuobj, "psci-conduit", vms->psci_conduit,
2021                                    NULL);
2022
2023            /* Secondary CPUs start in PSCI powered-down state */
2024            if (n > 0) {
2025                object_property_set_bool(cpuobj, "start-powered-off", true,
2026                                         NULL);
2027            }
2028        }
2029
2030        if (vmc->kvm_no_adjvtime &&
2031            object_property_find(cpuobj, "kvm-no-adjvtime")) {
2032            object_property_set_bool(cpuobj, "kvm-no-adjvtime", true, NULL);
2033        }
2034
2035        if (vmc->no_kvm_steal_time &&
2036            object_property_find(cpuobj, "kvm-steal-time")) {
2037            object_property_set_bool(cpuobj, "kvm-steal-time", false, NULL);
2038        }
2039
2040        if (vmc->no_pmu && object_property_find(cpuobj, "pmu")) {
2041            object_property_set_bool(cpuobj, "pmu", false, NULL);
2042        }
2043
2044        if (object_property_find(cpuobj, "reset-cbar")) {
2045            object_property_set_int(cpuobj, "reset-cbar",
2046                                    vms->memmap[VIRT_CPUPERIPHS].base,
2047                                    &error_abort);
2048        }
2049
2050        object_property_set_link(cpuobj, "memory", OBJECT(sysmem),
2051                                 &error_abort);
2052        if (vms->secure) {
2053            object_property_set_link(cpuobj, "secure-memory",
2054                                     OBJECT(secure_sysmem), &error_abort);
2055        }
2056
2057        if (vms->mte) {
2058            /* Create the memory region only once, but link to all cpus. */
2059            if (!tag_sysmem) {
2060                /*
2061                 * The property exists only if MemTag is supported.
2062                 * If it is, we must allocate the ram to back that up.
2063                 */
2064                if (!object_property_find(cpuobj, "tag-memory")) {
2065                    error_report("MTE requested, but not supported "
2066                                 "by the guest CPU");
2067                    exit(1);
2068                }
2069
2070                tag_sysmem = g_new(MemoryRegion, 1);
2071                memory_region_init(tag_sysmem, OBJECT(machine),
2072                                   "tag-memory", UINT64_MAX / 32);
2073
2074                if (vms->secure) {
2075                    secure_tag_sysmem = g_new(MemoryRegion, 1);
2076                    memory_region_init(secure_tag_sysmem, OBJECT(machine),
2077                                       "secure-tag-memory", UINT64_MAX / 32);
2078
2079                    /* As with ram, secure-tag takes precedence over tag.  */
2080                    memory_region_add_subregion_overlap(secure_tag_sysmem, 0,
2081                                                        tag_sysmem, -1);
2082                }
2083            }
2084
2085            object_property_set_link(cpuobj, "tag-memory", OBJECT(tag_sysmem),
2086                                     &error_abort);
2087            if (vms->secure) {
2088                object_property_set_link(cpuobj, "secure-tag-memory",
2089                                         OBJECT(secure_tag_sysmem),
2090                                         &error_abort);
2091            }
2092        }
2093
2094        qdev_realize(DEVICE(cpuobj), NULL, &error_fatal);
2095        object_unref(cpuobj);
2096    }
2097    fdt_add_timer_nodes(vms);
2098    fdt_add_cpu_nodes(vms);
2099
2100    memory_region_add_subregion(sysmem, vms->memmap[VIRT_MEM].base,
2101                                machine->ram);
2102    if (machine->device_memory) {
2103        memory_region_add_subregion(sysmem, machine->device_memory->base,
2104                                    &machine->device_memory->mr);
2105    }
2106
2107    virt_flash_fdt(vms, sysmem, secure_sysmem ?: sysmem);
2108
2109    create_gic(vms, sysmem);
2110
2111    virt_cpu_post_init(vms, sysmem);
2112
2113    fdt_add_pmu_nodes(vms);
2114
2115    create_uart(vms, VIRT_UART, sysmem, serial_hd(0));
2116
2117    if (vms->secure) {
2118        create_secure_ram(vms, secure_sysmem, secure_tag_sysmem);
2119        create_uart(vms, VIRT_SECURE_UART, secure_sysmem, serial_hd(1));
2120    }
2121
2122    if (tag_sysmem) {
2123        create_tag_ram(tag_sysmem, vms->memmap[VIRT_MEM].base,
2124                       machine->ram_size, "mach-virt.tag");
2125    }
2126
2127    vms->highmem_ecam &= vms->highmem && (!firmware_loaded || aarch64);
2128
2129    create_rtc(vms);
2130
2131    create_pcie(vms);
2132
2133    if (has_ged && aarch64 && firmware_loaded && virt_is_acpi_enabled(vms)) {
2134        vms->acpi_dev = create_acpi_ged(vms);
2135    } else {
2136        create_gpio_devices(vms, VIRT_GPIO, sysmem);
2137    }
2138
2139    if (vms->secure && !vmc->no_secure_gpio) {
2140        create_gpio_devices(vms, VIRT_SECURE_GPIO, secure_sysmem);
2141    }
2142
2143     /* connect powerdown request */
2144     vms->powerdown_notifier.notify = virt_powerdown_req;
2145     qemu_register_powerdown_notifier(&vms->powerdown_notifier);
2146
2147    /* Create mmio transports, so the user can create virtio backends
2148     * (which will be automatically plugged in to the transports). If
2149     * no backend is created the transport will just sit harmlessly idle.
2150     */
2151    create_virtio_devices(vms);
2152
2153    vms->fw_cfg = create_fw_cfg(vms, &address_space_memory);
2154    rom_set_fw(vms->fw_cfg);
2155
2156    create_platform_bus(vms);
2157
2158    if (machine->nvdimms_state->is_enabled) {
2159        const struct AcpiGenericAddress arm_virt_nvdimm_acpi_dsmio = {
2160            .space_id = AML_AS_SYSTEM_MEMORY,
2161            .address = vms->memmap[VIRT_NVDIMM_ACPI].base,
2162            .bit_width = NVDIMM_ACPI_IO_LEN << 3
2163        };
2164
2165        nvdimm_init_acpi_state(machine->nvdimms_state, sysmem,
2166                               arm_virt_nvdimm_acpi_dsmio,
2167                               vms->fw_cfg, OBJECT(vms));
2168    }
2169
2170    vms->bootinfo.ram_size = machine->ram_size;
2171    vms->bootinfo.nb_cpus = smp_cpus;
2172    vms->bootinfo.board_id = -1;
2173    vms->bootinfo.loader_start = vms->memmap[VIRT_MEM].base;
2174    vms->bootinfo.get_dtb = machvirt_dtb;
2175    vms->bootinfo.skip_dtb_autoload = true;
2176    vms->bootinfo.firmware_loaded = firmware_loaded;
2177    arm_load_kernel(ARM_CPU(first_cpu), machine, &vms->bootinfo);
2178
2179    vms->machine_done.notify = virt_machine_done;
2180    qemu_add_machine_init_done_notifier(&vms->machine_done);
2181}
2182
2183static bool virt_get_secure(Object *obj, Error **errp)
2184{
2185    VirtMachineState *vms = VIRT_MACHINE(obj);
2186
2187    return vms->secure;
2188}
2189
2190static void virt_set_secure(Object *obj, bool value, Error **errp)
2191{
2192    VirtMachineState *vms = VIRT_MACHINE(obj);
2193
2194    vms->secure = value;
2195}
2196
2197static bool virt_get_virt(Object *obj, Error **errp)
2198{
2199    VirtMachineState *vms = VIRT_MACHINE(obj);
2200
2201    return vms->virt;
2202}
2203
2204static void virt_set_virt(Object *obj, bool value, Error **errp)
2205{
2206    VirtMachineState *vms = VIRT_MACHINE(obj);
2207
2208    vms->virt = value;
2209}
2210
2211static bool virt_get_highmem(Object *obj, Error **errp)
2212{
2213    VirtMachineState *vms = VIRT_MACHINE(obj);
2214
2215    return vms->highmem;
2216}
2217
2218static void virt_set_highmem(Object *obj, bool value, Error **errp)
2219{
2220    VirtMachineState *vms = VIRT_MACHINE(obj);
2221
2222    vms->highmem = value;
2223}
2224
2225static bool virt_get_its(Object *obj, Error **errp)
2226{
2227    VirtMachineState *vms = VIRT_MACHINE(obj);
2228
2229    return vms->its;
2230}
2231
2232static void virt_set_its(Object *obj, bool value, Error **errp)
2233{
2234    VirtMachineState *vms = VIRT_MACHINE(obj);
2235
2236    vms->its = value;
2237}
2238
2239static char *virt_get_oem_id(Object *obj, Error **errp)
2240{
2241    VirtMachineState *vms = VIRT_MACHINE(obj);
2242
2243    return g_strdup(vms->oem_id);
2244}
2245
2246static void virt_set_oem_id(Object *obj, const char *value, Error **errp)
2247{
2248    VirtMachineState *vms = VIRT_MACHINE(obj);
2249    size_t len = strlen(value);
2250
2251    if (len > 6) {
2252        error_setg(errp,
2253                   "User specified oem-id value is bigger than 6 bytes in size");
2254        return;
2255    }
2256
2257    strncpy(vms->oem_id, value, 6);
2258}
2259
2260static char *virt_get_oem_table_id(Object *obj, Error **errp)
2261{
2262    VirtMachineState *vms = VIRT_MACHINE(obj);
2263
2264    return g_strdup(vms->oem_table_id);
2265}
2266
2267static void virt_set_oem_table_id(Object *obj, const char *value,
2268                                  Error **errp)
2269{
2270    VirtMachineState *vms = VIRT_MACHINE(obj);
2271    size_t len = strlen(value);
2272
2273    if (len > 8) {
2274        error_setg(errp,
2275                   "User specified oem-table-id value is bigger than 8 bytes in size");
2276        return;
2277    }
2278    strncpy(vms->oem_table_id, value, 8);
2279}
2280
2281
2282bool virt_is_acpi_enabled(VirtMachineState *vms)
2283{
2284    if (vms->acpi == ON_OFF_AUTO_OFF) {
2285        return false;
2286    }
2287    return true;
2288}
2289
2290static void virt_get_acpi(Object *obj, Visitor *v, const char *name,
2291                          void *opaque, Error **errp)
2292{
2293    VirtMachineState *vms = VIRT_MACHINE(obj);
2294    OnOffAuto acpi = vms->acpi;
2295
2296    visit_type_OnOffAuto(v, name, &acpi, errp);
2297}
2298
2299static void virt_set_acpi(Object *obj, Visitor *v, const char *name,
2300                          void *opaque, Error **errp)
2301{
2302    VirtMachineState *vms = VIRT_MACHINE(obj);
2303
2304    visit_type_OnOffAuto(v, name, &vms->acpi, errp);
2305}
2306
2307static bool virt_get_ras(Object *obj, Error **errp)
2308{
2309    VirtMachineState *vms = VIRT_MACHINE(obj);
2310
2311    return vms->ras;
2312}
2313
2314static void virt_set_ras(Object *obj, bool value, Error **errp)
2315{
2316    VirtMachineState *vms = VIRT_MACHINE(obj);
2317
2318    vms->ras = value;
2319}
2320
2321static bool virt_get_mte(Object *obj, Error **errp)
2322{
2323    VirtMachineState *vms = VIRT_MACHINE(obj);
2324
2325    return vms->mte;
2326}
2327
2328static void virt_set_mte(Object *obj, bool value, Error **errp)
2329{
2330    VirtMachineState *vms = VIRT_MACHINE(obj);
2331
2332    vms->mte = value;
2333}
2334
2335static char *virt_get_gic_version(Object *obj, Error **errp)
2336{
2337    VirtMachineState *vms = VIRT_MACHINE(obj);
2338    const char *val = vms->gic_version == VIRT_GIC_VERSION_3 ? "3" : "2";
2339
2340    return g_strdup(val);
2341}
2342
2343static void virt_set_gic_version(Object *obj, const char *value, Error **errp)
2344{
2345    VirtMachineState *vms = VIRT_MACHINE(obj);
2346
2347    if (!strcmp(value, "3")) {
2348        vms->gic_version = VIRT_GIC_VERSION_3;
2349    } else if (!strcmp(value, "2")) {
2350        vms->gic_version = VIRT_GIC_VERSION_2;
2351    } else if (!strcmp(value, "host")) {
2352        vms->gic_version = VIRT_GIC_VERSION_HOST; /* Will probe later */
2353    } else if (!strcmp(value, "max")) {
2354        vms->gic_version = VIRT_GIC_VERSION_MAX; /* Will probe later */
2355    } else {
2356        error_setg(errp, "Invalid gic-version value");
2357        error_append_hint(errp, "Valid values are 3, 2, host, max.\n");
2358    }
2359}
2360
2361static char *virt_get_iommu(Object *obj, Error **errp)
2362{
2363    VirtMachineState *vms = VIRT_MACHINE(obj);
2364
2365    switch (vms->iommu) {
2366    case VIRT_IOMMU_NONE:
2367        return g_strdup("none");
2368    case VIRT_IOMMU_SMMUV3:
2369        return g_strdup("smmuv3");
2370    default:
2371        g_assert_not_reached();
2372    }
2373}
2374
2375static void virt_set_iommu(Object *obj, const char *value, Error **errp)
2376{
2377    VirtMachineState *vms = VIRT_MACHINE(obj);
2378
2379    if (!strcmp(value, "smmuv3")) {
2380        vms->iommu = VIRT_IOMMU_SMMUV3;
2381    } else if (!strcmp(value, "none")) {
2382        vms->iommu = VIRT_IOMMU_NONE;
2383    } else {
2384        error_setg(errp, "Invalid iommu value");
2385        error_append_hint(errp, "Valid values are none, smmuv3.\n");
2386    }
2387}
2388
2389static bool virt_get_default_bus_bypass_iommu(Object *obj, Error **errp)
2390{
2391    VirtMachineState *vms = VIRT_MACHINE(obj);
2392
2393    return vms->default_bus_bypass_iommu;
2394}
2395
2396static void virt_set_default_bus_bypass_iommu(Object *obj, bool value,
2397                                              Error **errp)
2398{
2399    VirtMachineState *vms = VIRT_MACHINE(obj);
2400
2401    vms->default_bus_bypass_iommu = value;
2402}
2403
2404static CpuInstanceProperties
2405virt_cpu_index_to_props(MachineState *ms, unsigned cpu_index)
2406{
2407    MachineClass *mc = MACHINE_GET_CLASS(ms);
2408    const CPUArchIdList *possible_cpus = mc->possible_cpu_arch_ids(ms);
2409
2410    assert(cpu_index < possible_cpus->len);
2411    return possible_cpus->cpus[cpu_index].props;
2412}
2413
2414static int64_t virt_get_default_cpu_node_id(const MachineState *ms, int idx)
2415{
2416    return idx % ms->numa_state->num_nodes;
2417}
2418
2419static const CPUArchIdList *virt_possible_cpu_arch_ids(MachineState *ms)
2420{
2421    int n;
2422    unsigned int max_cpus = ms->smp.max_cpus;
2423    VirtMachineState *vms = VIRT_MACHINE(ms);
2424
2425    if (ms->possible_cpus) {
2426        assert(ms->possible_cpus->len == max_cpus);
2427        return ms->possible_cpus;
2428    }
2429
2430    ms->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2431                                  sizeof(CPUArchId) * max_cpus);
2432    ms->possible_cpus->len = max_cpus;
2433    for (n = 0; n < ms->possible_cpus->len; n++) {
2434        ms->possible_cpus->cpus[n].type = ms->cpu_type;
2435        ms->possible_cpus->cpus[n].arch_id =
2436            virt_cpu_mp_affinity(vms, n);
2437        ms->possible_cpus->cpus[n].props.has_thread_id = true;
2438        ms->possible_cpus->cpus[n].props.thread_id = n;
2439    }
2440    return ms->possible_cpus;
2441}
2442
2443static void virt_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2444                                 Error **errp)
2445{
2446    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2447    const MachineState *ms = MACHINE(hotplug_dev);
2448    const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2449
2450    if (!vms->acpi_dev) {
2451        error_setg(errp,
2452                   "memory hotplug is not enabled: missing acpi-ged device");
2453        return;
2454    }
2455
2456    if (vms->mte) {
2457        error_setg(errp, "memory hotplug is not enabled: MTE is enabled");
2458        return;
2459    }
2460
2461    if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
2462        error_setg(errp, "nvdimm is not enabled: add 'nvdimm=on' to '-M'");
2463        return;
2464    }
2465
2466    pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev), NULL, errp);
2467}
2468
2469static void virt_memory_plug(HotplugHandler *hotplug_dev,
2470                             DeviceState *dev, Error **errp)
2471{
2472    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2473    MachineState *ms = MACHINE(hotplug_dev);
2474    bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
2475
2476    pc_dimm_plug(PC_DIMM(dev), MACHINE(vms));
2477
2478    if (is_nvdimm) {
2479        nvdimm_plug(ms->nvdimms_state);
2480    }
2481
2482    hotplug_handler_plug(HOTPLUG_HANDLER(vms->acpi_dev),
2483                         dev, &error_abort);
2484}
2485
2486static void virt_machine_device_pre_plug_cb(HotplugHandler *hotplug_dev,
2487                                            DeviceState *dev, Error **errp)
2488{
2489    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2490
2491    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2492        virt_memory_pre_plug(hotplug_dev, dev, errp);
2493    } else if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2494        hwaddr db_start = 0, db_end = 0;
2495        char *resv_prop_str;
2496
2497        switch (vms->msi_controller) {
2498        case VIRT_MSI_CTRL_NONE:
2499            return;
2500        case VIRT_MSI_CTRL_ITS:
2501            /* GITS_TRANSLATER page */
2502            db_start = base_memmap[VIRT_GIC_ITS].base + 0x10000;
2503            db_end = base_memmap[VIRT_GIC_ITS].base +
2504                     base_memmap[VIRT_GIC_ITS].size - 1;
2505            break;
2506        case VIRT_MSI_CTRL_GICV2M:
2507            /* MSI_SETSPI_NS page */
2508            db_start = base_memmap[VIRT_GIC_V2M].base;
2509            db_end = db_start + base_memmap[VIRT_GIC_V2M].size - 1;
2510            break;
2511        }
2512        resv_prop_str = g_strdup_printf("0x%"PRIx64":0x%"PRIx64":%u",
2513                                        db_start, db_end,
2514                                        VIRTIO_IOMMU_RESV_MEM_T_MSI);
2515
2516        qdev_prop_set_uint32(dev, "len-reserved-regions", 1);
2517        qdev_prop_set_string(dev, "reserved-regions[0]", resv_prop_str);
2518        g_free(resv_prop_str);
2519    }
2520}
2521
2522static void virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
2523                                        DeviceState *dev, Error **errp)
2524{
2525    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2526
2527    if (vms->platform_bus_dev) {
2528        MachineClass *mc = MACHINE_GET_CLASS(vms);
2529
2530        if (device_is_dynamic_sysbus(mc, dev)) {
2531            platform_bus_link_device(PLATFORM_BUS_DEVICE(vms->platform_bus_dev),
2532                                     SYS_BUS_DEVICE(dev));
2533        }
2534    }
2535    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2536        virt_memory_plug(hotplug_dev, dev, errp);
2537    }
2538    if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2539        PCIDevice *pdev = PCI_DEVICE(dev);
2540
2541        vms->iommu = VIRT_IOMMU_VIRTIO;
2542        vms->virtio_iommu_bdf = pci_get_bdf(pdev);
2543        create_virtio_iommu_dt_bindings(vms);
2544    }
2545}
2546
2547static void virt_dimm_unplug_request(HotplugHandler *hotplug_dev,
2548                                     DeviceState *dev, Error **errp)
2549{
2550    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2551    Error *local_err = NULL;
2552
2553    if (!vms->acpi_dev) {
2554        error_setg(&local_err,
2555                   "memory hotplug is not enabled: missing acpi-ged device");
2556        goto out;
2557    }
2558
2559    if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
2560        error_setg(&local_err,
2561                   "nvdimm device hot unplug is not supported yet.");
2562        goto out;
2563    }
2564
2565    hotplug_handler_unplug_request(HOTPLUG_HANDLER(vms->acpi_dev), dev,
2566                                   &local_err);
2567out:
2568    error_propagate(errp, local_err);
2569}
2570
2571static void virt_dimm_unplug(HotplugHandler *hotplug_dev,
2572                             DeviceState *dev, Error **errp)
2573{
2574    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
2575    Error *local_err = NULL;
2576
2577    hotplug_handler_unplug(HOTPLUG_HANDLER(vms->acpi_dev), dev, &local_err);
2578    if (local_err) {
2579        goto out;
2580    }
2581
2582    pc_dimm_unplug(PC_DIMM(dev), MACHINE(vms));
2583    qdev_unrealize(dev);
2584
2585out:
2586    error_propagate(errp, local_err);
2587}
2588
2589static void virt_machine_device_unplug_request_cb(HotplugHandler *hotplug_dev,
2590                                          DeviceState *dev, Error **errp)
2591{
2592    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2593        virt_dimm_unplug_request(hotplug_dev, dev, errp);
2594    } else {
2595        error_setg(errp, "device unplug request for unsupported device"
2596                   " type: %s", object_get_typename(OBJECT(dev)));
2597    }
2598}
2599
2600static void virt_machine_device_unplug_cb(HotplugHandler *hotplug_dev,
2601                                          DeviceState *dev, Error **errp)
2602{
2603    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2604        virt_dimm_unplug(hotplug_dev, dev, errp);
2605    } else {
2606        error_setg(errp, "virt: device unplug for unsupported device"
2607                   " type: %s", object_get_typename(OBJECT(dev)));
2608    }
2609}
2610
2611static HotplugHandler *virt_machine_get_hotplug_handler(MachineState *machine,
2612                                                        DeviceState *dev)
2613{
2614    MachineClass *mc = MACHINE_GET_CLASS(machine);
2615
2616    if (device_is_dynamic_sysbus(mc, dev) ||
2617       (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM))) {
2618        return HOTPLUG_HANDLER(machine);
2619    }
2620    if (object_dynamic_cast(OBJECT(dev), TYPE_VIRTIO_IOMMU_PCI)) {
2621        VirtMachineState *vms = VIRT_MACHINE(machine);
2622
2623        if (!vms->bootinfo.firmware_loaded || !virt_is_acpi_enabled(vms)) {
2624            return HOTPLUG_HANDLER(machine);
2625        }
2626    }
2627    return NULL;
2628}
2629
2630/*
2631 * for arm64 kvm_type [7-0] encodes the requested number of bits
2632 * in the IPA address space
2633 */
2634static int virt_kvm_type(MachineState *ms, const char *type_str)
2635{
2636    VirtMachineState *vms = VIRT_MACHINE(ms);
2637    int max_vm_pa_size, requested_pa_size;
2638    bool fixed_ipa;
2639
2640    max_vm_pa_size = kvm_arm_get_max_vm_ipa_size(ms, &fixed_ipa);
2641
2642    /* we freeze the memory map to compute the highest gpa */
2643    virt_set_memmap(vms);
2644
2645    requested_pa_size = 64 - clz64(vms->highest_gpa);
2646
2647    /*
2648     * KVM requires the IPA size to be at least 32 bits.
2649     */
2650    if (requested_pa_size < 32) {
2651        requested_pa_size = 32;
2652    }
2653
2654    if (requested_pa_size > max_vm_pa_size) {
2655        error_report("-m and ,maxmem option values "
2656                     "require an IPA range (%d bits) larger than "
2657                     "the one supported by the host (%d bits)",
2658                     requested_pa_size, max_vm_pa_size);
2659        exit(1);
2660    }
2661    /*
2662     * We return the requested PA log size, unless KVM only supports
2663     * the implicit legacy 40b IPA setting, in which case the kvm_type
2664     * must be 0.
2665     */
2666    return fixed_ipa ? 0 : requested_pa_size;
2667}
2668
2669static void virt_machine_class_init(ObjectClass *oc, void *data)
2670{
2671    MachineClass *mc = MACHINE_CLASS(oc);
2672    HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2673
2674    mc->init = machvirt_init;
2675    /* Start with max_cpus set to 512, which is the maximum supported by KVM.
2676     * The value may be reduced later when we have more information about the
2677     * configuration of the particular instance.
2678     */
2679    mc->max_cpus = 512;
2680    machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_CALXEDA_XGMAC);
2681    machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_AMD_XGBE);
2682    machine_class_allow_dynamic_sysbus_dev(mc, TYPE_RAMFB_DEVICE);
2683    machine_class_allow_dynamic_sysbus_dev(mc, TYPE_VFIO_PLATFORM);
2684#ifdef CONFIG_TPM
2685    machine_class_allow_dynamic_sysbus_dev(mc, TYPE_TPM_TIS_SYSBUS);
2686#endif
2687    mc->block_default_type = IF_VIRTIO;
2688    mc->no_cdrom = 1;
2689    mc->pci_allow_0_address = true;
2690    /* We know we will never create a pre-ARMv7 CPU which needs 1K pages */
2691    mc->minimum_page_bits = 12;
2692    mc->possible_cpu_arch_ids = virt_possible_cpu_arch_ids;
2693    mc->cpu_index_to_instance_props = virt_cpu_index_to_props;
2694    mc->default_cpu_type = ARM_CPU_TYPE_NAME("cortex-a15");
2695    mc->get_default_cpu_node_id = virt_get_default_cpu_node_id;
2696    mc->kvm_type = virt_kvm_type;
2697    assert(!mc->get_hotplug_handler);
2698    mc->get_hotplug_handler = virt_machine_get_hotplug_handler;
2699    hc->pre_plug = virt_machine_device_pre_plug_cb;
2700    hc->plug = virt_machine_device_plug_cb;
2701    hc->unplug_request = virt_machine_device_unplug_request_cb;
2702    hc->unplug = virt_machine_device_unplug_cb;
2703    mc->nvdimm_supported = true;
2704    mc->auto_enable_numa_with_memhp = true;
2705    mc->auto_enable_numa_with_memdev = true;
2706    mc->default_ram_id = "mach-virt.ram";
2707
2708    object_class_property_add(oc, "acpi", "OnOffAuto",
2709        virt_get_acpi, virt_set_acpi,
2710        NULL, NULL);
2711    object_class_property_set_description(oc, "acpi",
2712        "Enable ACPI");
2713    object_class_property_add_bool(oc, "secure", virt_get_secure,
2714                                   virt_set_secure);
2715    object_class_property_set_description(oc, "secure",
2716                                                "Set on/off to enable/disable the ARM "
2717                                                "Security Extensions (TrustZone)");
2718
2719    object_class_property_add_bool(oc, "virtualization", virt_get_virt,
2720                                   virt_set_virt);
2721    object_class_property_set_description(oc, "virtualization",
2722                                          "Set on/off to enable/disable emulating a "
2723                                          "guest CPU which implements the ARM "
2724                                          "Virtualization Extensions");
2725
2726    object_class_property_add_bool(oc, "highmem", virt_get_highmem,
2727                                   virt_set_highmem);
2728    object_class_property_set_description(oc, "highmem",
2729                                          "Set on/off to enable/disable using "
2730                                          "physical address space above 32 bits");
2731
2732    object_class_property_add_str(oc, "gic-version", virt_get_gic_version,
2733                                  virt_set_gic_version);
2734    object_class_property_set_description(oc, "gic-version",
2735                                          "Set GIC version. "
2736                                          "Valid values are 2, 3, host and max");
2737
2738    object_class_property_add_str(oc, "iommu", virt_get_iommu, virt_set_iommu);
2739    object_class_property_set_description(oc, "iommu",
2740                                          "Set the IOMMU type. "
2741                                          "Valid values are none and smmuv3");
2742
2743    object_class_property_add_bool(oc, "default-bus-bypass-iommu",
2744                                   virt_get_default_bus_bypass_iommu,
2745                                   virt_set_default_bus_bypass_iommu);
2746    object_class_property_set_description(oc, "default-bus-bypass-iommu",
2747                                          "Set on/off to enable/disable "
2748                                          "bypass_iommu for default root bus");
2749
2750    object_class_property_add_bool(oc, "ras", virt_get_ras,
2751                                   virt_set_ras);
2752    object_class_property_set_description(oc, "ras",
2753                                          "Set on/off to enable/disable reporting host memory errors "
2754                                          "to a KVM guest using ACPI and guest external abort exceptions");
2755
2756    object_class_property_add_bool(oc, "mte", virt_get_mte, virt_set_mte);
2757    object_class_property_set_description(oc, "mte",
2758                                          "Set on/off to enable/disable emulating a "
2759                                          "guest CPU which implements the ARM "
2760                                          "Memory Tagging Extension");
2761
2762    object_class_property_add_bool(oc, "its", virt_get_its,
2763                                   virt_set_its);
2764    object_class_property_set_description(oc, "its",
2765                                          "Set on/off to enable/disable "
2766                                          "ITS instantiation");
2767
2768    object_class_property_add_str(oc, "x-oem-id",
2769                                  virt_get_oem_id,
2770                                  virt_set_oem_id);
2771    object_class_property_set_description(oc, "x-oem-id",
2772                                          "Override the default value of field OEMID "
2773                                          "in ACPI table header."
2774                                          "The string may be up to 6 bytes in size");
2775
2776
2777    object_class_property_add_str(oc, "x-oem-table-id",
2778                                  virt_get_oem_table_id,
2779                                  virt_set_oem_table_id);
2780    object_class_property_set_description(oc, "x-oem-table-id",
2781                                          "Override the default value of field OEM Table ID "
2782                                          "in ACPI table header."
2783                                          "The string may be up to 8 bytes in size");
2784
2785}
2786
2787static void virt_instance_init(Object *obj)
2788{
2789    VirtMachineState *vms = VIRT_MACHINE(obj);
2790    VirtMachineClass *vmc = VIRT_MACHINE_GET_CLASS(vms);
2791
2792    /* EL3 is disabled by default on virt: this makes us consistent
2793     * between KVM and TCG for this board, and it also allows us to
2794     * boot UEFI blobs which assume no TrustZone support.
2795     */
2796    vms->secure = false;
2797
2798    /* EL2 is also disabled by default, for similar reasons */
2799    vms->virt = false;
2800
2801    /* High memory is enabled by default */
2802    vms->highmem = true;
2803    vms->gic_version = VIRT_GIC_VERSION_NOSEL;
2804
2805    vms->highmem_ecam = !vmc->no_highmem_ecam;
2806
2807    if (vmc->no_its) {
2808        vms->its = false;
2809    } else {
2810        /* Default allows ITS instantiation */
2811        vms->its = true;
2812
2813        if (vmc->no_tcg_its) {
2814            vms->tcg_its = false;
2815        } else {
2816            vms->tcg_its = true;
2817        }
2818    }
2819
2820    /* Default disallows iommu instantiation */
2821    vms->iommu = VIRT_IOMMU_NONE;
2822
2823    /* The default root bus is attached to iommu by default */
2824    vms->default_bus_bypass_iommu = false;
2825
2826    /* Default disallows RAS instantiation */
2827    vms->ras = false;
2828
2829    /* MTE is disabled by default.  */
2830    vms->mte = false;
2831
2832    vms->irqmap = a15irqmap;
2833
2834    virt_flash_create(vms);
2835
2836    vms->oem_id = g_strndup(ACPI_BUILD_APPNAME6, 6);
2837    vms->oem_table_id = g_strndup(ACPI_BUILD_APPNAME8, 8);
2838}
2839
2840static const TypeInfo virt_machine_info = {
2841    .name          = TYPE_VIRT_MACHINE,
2842    .parent        = TYPE_MACHINE,
2843    .abstract      = true,
2844    .instance_size = sizeof(VirtMachineState),
2845    .class_size    = sizeof(VirtMachineClass),
2846    .class_init    = virt_machine_class_init,
2847    .instance_init = virt_instance_init,
2848    .interfaces = (InterfaceInfo[]) {
2849         { TYPE_HOTPLUG_HANDLER },
2850         { }
2851    },
2852};
2853
2854static void machvirt_machine_init(void)
2855{
2856    type_register_static(&virt_machine_info);
2857}
2858type_init(machvirt_machine_init);
2859
2860static void virt_machine_6_2_options(MachineClass *mc)
2861{
2862}
2863DEFINE_VIRT_MACHINE_AS_LATEST(6, 2)
2864
2865static void virt_machine_6_1_options(MachineClass *mc)
2866{
2867    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2868
2869    virt_machine_6_2_options(mc);
2870    compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
2871    mc->smp_props.prefer_sockets = true;
2872    vmc->no_cpu_topology = true;
2873
2874    /* qemu ITS was introduced with 6.2 */
2875    vmc->no_tcg_its = true;
2876}
2877DEFINE_VIRT_MACHINE(6, 1)
2878
2879static void virt_machine_6_0_options(MachineClass *mc)
2880{
2881    virt_machine_6_1_options(mc);
2882    compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
2883}
2884DEFINE_VIRT_MACHINE(6, 0)
2885
2886static void virt_machine_5_2_options(MachineClass *mc)
2887{
2888    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2889
2890    virt_machine_6_0_options(mc);
2891    compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
2892    vmc->no_secure_gpio = true;
2893}
2894DEFINE_VIRT_MACHINE(5, 2)
2895
2896static void virt_machine_5_1_options(MachineClass *mc)
2897{
2898    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2899
2900    virt_machine_5_2_options(mc);
2901    compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
2902    vmc->no_kvm_steal_time = true;
2903}
2904DEFINE_VIRT_MACHINE(5, 1)
2905
2906static void virt_machine_5_0_options(MachineClass *mc)
2907{
2908    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2909
2910    virt_machine_5_1_options(mc);
2911    compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
2912    mc->numa_mem_supported = true;
2913    vmc->acpi_expose_flash = true;
2914    mc->auto_enable_numa_with_memdev = false;
2915}
2916DEFINE_VIRT_MACHINE(5, 0)
2917
2918static void virt_machine_4_2_options(MachineClass *mc)
2919{
2920    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2921
2922    virt_machine_5_0_options(mc);
2923    compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
2924    vmc->kvm_no_adjvtime = true;
2925}
2926DEFINE_VIRT_MACHINE(4, 2)
2927
2928static void virt_machine_4_1_options(MachineClass *mc)
2929{
2930    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2931
2932    virt_machine_4_2_options(mc);
2933    compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
2934    vmc->no_ged = true;
2935    mc->auto_enable_numa_with_memhp = false;
2936}
2937DEFINE_VIRT_MACHINE(4, 1)
2938
2939static void virt_machine_4_0_options(MachineClass *mc)
2940{
2941    virt_machine_4_1_options(mc);
2942    compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
2943}
2944DEFINE_VIRT_MACHINE(4, 0)
2945
2946static void virt_machine_3_1_options(MachineClass *mc)
2947{
2948    virt_machine_4_0_options(mc);
2949    compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
2950}
2951DEFINE_VIRT_MACHINE(3, 1)
2952
2953static void virt_machine_3_0_options(MachineClass *mc)
2954{
2955    virt_machine_3_1_options(mc);
2956    compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
2957}
2958DEFINE_VIRT_MACHINE(3, 0)
2959
2960static void virt_machine_2_12_options(MachineClass *mc)
2961{
2962    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2963
2964    virt_machine_3_0_options(mc);
2965    compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
2966    vmc->no_highmem_ecam = true;
2967    mc->max_cpus = 255;
2968}
2969DEFINE_VIRT_MACHINE(2, 12)
2970
2971static void virt_machine_2_11_options(MachineClass *mc)
2972{
2973    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
2974
2975    virt_machine_2_12_options(mc);
2976    compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
2977    vmc->smbios_old_sys_ver = true;
2978}
2979DEFINE_VIRT_MACHINE(2, 11)
2980
2981static void virt_machine_2_10_options(MachineClass *mc)
2982{
2983    virt_machine_2_11_options(mc);
2984    compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
2985    /* before 2.11 we never faulted accesses to bad addresses */
2986    mc->ignore_memory_transaction_failures = true;
2987}
2988DEFINE_VIRT_MACHINE(2, 10)
2989
2990static void virt_machine_2_9_options(MachineClass *mc)
2991{
2992    virt_machine_2_10_options(mc);
2993    compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
2994}
2995DEFINE_VIRT_MACHINE(2, 9)
2996
2997static void virt_machine_2_8_options(MachineClass *mc)
2998{
2999    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3000
3001    virt_machine_2_9_options(mc);
3002    compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
3003    /* For 2.8 and earlier we falsely claimed in the DT that
3004     * our timers were edge-triggered, not level-triggered.
3005     */
3006    vmc->claim_edge_triggered_timers = true;
3007}
3008DEFINE_VIRT_MACHINE(2, 8)
3009
3010static void virt_machine_2_7_options(MachineClass *mc)
3011{
3012    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3013
3014    virt_machine_2_8_options(mc);
3015    compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
3016    /* ITS was introduced with 2.8 */
3017    vmc->no_its = true;
3018    /* Stick with 1K pages for migration compatibility */
3019    mc->minimum_page_bits = 0;
3020}
3021DEFINE_VIRT_MACHINE(2, 7)
3022
3023static void virt_machine_2_6_options(MachineClass *mc)
3024{
3025    VirtMachineClass *vmc = VIRT_MACHINE_CLASS(OBJECT_CLASS(mc));
3026
3027    virt_machine_2_7_options(mc);
3028    compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
3029    vmc->disallow_affinity_adjustment = true;
3030    /* Disable PMU for 2.6 as PMU support was first introduced in 2.7 */
3031    vmc->no_pmu = true;
3032}
3033DEFINE_VIRT_MACHINE(2, 6)
3034