Add qemu 2.4.0
[kvmfornfv.git] / qemu / hw / ppc / spapr.c
diff --git a/qemu/hw/ppc/spapr.c b/qemu/hw/ppc/spapr.c
new file mode 100644 (file)
index 0000000..a6f1947
--- /dev/null
@@ -0,0 +1,1998 @@
+/*
+ * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
+ *
+ * Copyright (c) 2004-2007 Fabrice Bellard
+ * Copyright (c) 2007 Jocelyn Mayer
+ * Copyright (c) 2010 David Gibson, IBM Corporation.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ */
+#include "sysemu/sysemu.h"
+#include "sysemu/numa.h"
+#include "hw/hw.h"
+#include "hw/fw-path-provider.h"
+#include "elf.h"
+#include "net/net.h"
+#include "sysemu/block-backend.h"
+#include "sysemu/cpus.h"
+#include "sysemu/kvm.h"
+#include "kvm_ppc.h"
+#include "migration/migration.h"
+#include "mmu-hash64.h"
+#include "qom/cpu.h"
+
+#include "hw/boards.h"
+#include "hw/ppc/ppc.h"
+#include "hw/loader.h"
+
+#include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_vio.h"
+#include "hw/pci-host/spapr.h"
+#include "hw/ppc/xics.h"
+#include "hw/pci/msi.h"
+
+#include "hw/pci/pci.h"
+#include "hw/scsi/scsi.h"
+#include "hw/virtio/virtio-scsi.h"
+
+#include "exec/address-spaces.h"
+#include "hw/usb.h"
+#include "qemu/config-file.h"
+#include "qemu/error-report.h"
+#include "trace.h"
+#include "hw/nmi.h"
+
+#include "hw/compat.h"
+
+#include <libfdt.h>
+
+/* SLOF memory layout:
+ *
+ * SLOF raw image loaded at 0, copies its romfs right below the flat
+ * device-tree, then position SLOF itself 31M below that
+ *
+ * So we set FW_OVERHEAD to 40MB which should account for all of that
+ * and more
+ *
+ * We load our kernel at 4M, leaving space for SLOF initial image
+ */
+#define FDT_MAX_SIZE            0x40000
+#define RTAS_MAX_SIZE           0x10000
+#define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
+#define FW_MAX_SIZE             0x400000
+#define FW_FILE_NAME            "slof.bin"
+#define FW_OVERHEAD             0x2800000
+#define KERNEL_LOAD_ADDR        FW_MAX_SIZE
+
+#define MIN_RMA_SLOF            128UL
+
+#define TIMEBASE_FREQ           512000000ULL
+
+#define MAX_CPUS                255
+
+#define PHANDLE_XICP            0x00001111
+
+#define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
+
+static XICSState *try_create_xics(const char *type, int nr_servers,
+                                  int nr_irqs, Error **errp)
+{
+    Error *err = NULL;
+    DeviceState *dev;
+
+    dev = qdev_create(NULL, type);
+    qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
+    qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
+    object_property_set_bool(OBJECT(dev), true, "realized", &err);
+    if (err) {
+        error_propagate(errp, err);
+        object_unparent(OBJECT(dev));
+        return NULL;
+    }
+    return XICS_COMMON(dev);
+}
+
+static XICSState *xics_system_init(MachineState *machine,
+                                   int nr_servers, int nr_irqs)
+{
+    XICSState *icp = NULL;
+
+    if (kvm_enabled()) {
+        Error *err = NULL;
+
+        if (machine_kernel_irqchip_allowed(machine)) {
+            icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs, &err);
+        }
+        if (machine_kernel_irqchip_required(machine) && !icp) {
+            error_report("kernel_irqchip requested but unavailable: %s",
+                         error_get_pretty(err));
+        }
+    }
+
+    if (!icp) {
+        icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs, &error_abort);
+    }
+
+    return icp;
+}
+
+static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
+                                  int smt_threads)
+{
+    int i, ret = 0;
+    uint32_t servers_prop[smt_threads];
+    uint32_t gservers_prop[smt_threads * 2];
+    int index = ppc_get_vcpu_dt_id(cpu);
+
+    if (cpu->cpu_version) {
+        ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
+        if (ret < 0) {
+            return ret;
+        }
+    }
+
+    /* Build interrupt servers and gservers properties */
+    for (i = 0; i < smt_threads; i++) {
+        servers_prop[i] = cpu_to_be32(index + i);
+        /* Hack, direct the group queues back to cpu 0 */
+        gservers_prop[i*2] = cpu_to_be32(index + i);
+        gservers_prop[i*2 + 1] = 0;
+    }
+    ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
+                      servers_prop, sizeof(servers_prop));
+    if (ret < 0) {
+        return ret;
+    }
+    ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
+                      gservers_prop, sizeof(gservers_prop));
+
+    return ret;
+}
+
+static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
+{
+    int ret = 0;
+    PowerPCCPU *cpu = POWERPC_CPU(cs);
+    int index = ppc_get_vcpu_dt_id(cpu);
+    uint32_t associativity[] = {cpu_to_be32(0x5),
+                                cpu_to_be32(0x0),
+                                cpu_to_be32(0x0),
+                                cpu_to_be32(0x0),
+                                cpu_to_be32(cs->numa_node),
+                                cpu_to_be32(index)};
+
+    /* Advertise NUMA via ibm,associativity */
+    if (nb_numa_nodes > 1) {
+        ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
+                          sizeof(associativity));
+    }
+
+    return ret;
+}
+
+static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
+{
+    int ret = 0, offset, cpus_offset;
+    CPUState *cs;
+    char cpu_model[32];
+    int smt = kvmppc_smt_threads();
+    uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
+
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+        DeviceClass *dc = DEVICE_GET_CLASS(cs);
+        int index = ppc_get_vcpu_dt_id(cpu);
+
+        if ((index % smt) != 0) {
+            continue;
+        }
+
+        snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
+
+        cpus_offset = fdt_path_offset(fdt, "/cpus");
+        if (cpus_offset < 0) {
+            cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
+                                          "cpus");
+            if (cpus_offset < 0) {
+                return cpus_offset;
+            }
+        }
+        offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
+        if (offset < 0) {
+            offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
+            if (offset < 0) {
+                return offset;
+            }
+        }
+
+        ret = fdt_setprop(fdt, offset, "ibm,pft-size",
+                          pft_size_prop, sizeof(pft_size_prop));
+        if (ret < 0) {
+            return ret;
+        }
+
+        ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
+        if (ret < 0) {
+            return ret;
+        }
+
+        ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
+                                     ppc_get_compat_smt_threads(cpu));
+        if (ret < 0) {
+            return ret;
+        }
+    }
+    return ret;
+}
+
+
+static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
+                                     size_t maxsize)
+{
+    size_t maxcells = maxsize / sizeof(uint32_t);
+    int i, j, count;
+    uint32_t *p = prop;
+
+    for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
+        struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
+
+        if (!sps->page_shift) {
+            break;
+        }
+        for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
+            if (sps->enc[count].page_shift == 0) {
+                break;
+            }
+        }
+        if ((p - prop) >= (maxcells - 3 - count * 2)) {
+            break;
+        }
+        *(p++) = cpu_to_be32(sps->page_shift);
+        *(p++) = cpu_to_be32(sps->slb_enc);
+        *(p++) = cpu_to_be32(count);
+        for (j = 0; j < count; j++) {
+            *(p++) = cpu_to_be32(sps->enc[j].page_shift);
+            *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
+        }
+    }
+
+    return (p - prop) * sizeof(uint32_t);
+}
+
+static hwaddr spapr_node0_size(void)
+{
+    MachineState *machine = MACHINE(qdev_get_machine());
+
+    if (nb_numa_nodes) {
+        int i;
+        for (i = 0; i < nb_numa_nodes; ++i) {
+            if (numa_info[i].node_mem) {
+                return MIN(pow2floor(numa_info[i].node_mem),
+                           machine->ram_size);
+            }
+        }
+    }
+    return machine->ram_size;
+}
+
+#define _FDT(exp) \
+    do { \
+        int ret = (exp);                                           \
+        if (ret < 0) {                                             \
+            fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
+                    #exp, fdt_strerror(ret));                      \
+            exit(1);                                               \
+        }                                                          \
+    } while (0)
+
+static void add_str(GString *s, const gchar *s1)
+{
+    g_string_append_len(s, s1, strlen(s1) + 1);
+}
+
+static void *spapr_create_fdt_skel(hwaddr initrd_base,
+                                   hwaddr initrd_size,
+                                   hwaddr kernel_size,
+                                   bool little_endian,
+                                   const char *kernel_cmdline,
+                                   uint32_t epow_irq)
+{
+    void *fdt;
+    uint32_t start_prop = cpu_to_be32(initrd_base);
+    uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
+    GString *hypertas = g_string_sized_new(256);
+    GString *qemu_hypertas = g_string_sized_new(256);
+    uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
+    uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(max_cpus)};
+    unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
+    char *buf;
+
+    add_str(hypertas, "hcall-pft");
+    add_str(hypertas, "hcall-term");
+    add_str(hypertas, "hcall-dabr");
+    add_str(hypertas, "hcall-interrupt");
+    add_str(hypertas, "hcall-tce");
+    add_str(hypertas, "hcall-vio");
+    add_str(hypertas, "hcall-splpar");
+    add_str(hypertas, "hcall-bulk");
+    add_str(hypertas, "hcall-set-mode");
+    add_str(qemu_hypertas, "hcall-memop1");
+
+    fdt = g_malloc0(FDT_MAX_SIZE);
+    _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
+
+    if (kernel_size) {
+        _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
+    }
+    if (initrd_size) {
+        _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
+    }
+    _FDT((fdt_finish_reservemap(fdt)));
+
+    /* Root node */
+    _FDT((fdt_begin_node(fdt, "")));
+    _FDT((fdt_property_string(fdt, "device_type", "chrp")));
+    _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
+    _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
+
+    /*
+     * Add info to guest to indentify which host is it being run on
+     * and what is the uuid of the guest
+     */
+    if (kvmppc_get_host_model(&buf)) {
+        _FDT((fdt_property_string(fdt, "host-model", buf)));
+        g_free(buf);
+    }
+    if (kvmppc_get_host_serial(&buf)) {
+        _FDT((fdt_property_string(fdt, "host-serial", buf)));
+        g_free(buf);
+    }
+
+    buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
+                          qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
+                          qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
+                          qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
+                          qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
+                          qemu_uuid[14], qemu_uuid[15]);
+
+    _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
+    g_free(buf);
+
+    _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
+    _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
+
+    /* /chosen */
+    _FDT((fdt_begin_node(fdt, "chosen")));
+
+    /* Set Form1_affinity */
+    _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
+
+    _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
+    _FDT((fdt_property(fdt, "linux,initrd-start",
+                       &start_prop, sizeof(start_prop))));
+    _FDT((fdt_property(fdt, "linux,initrd-end",
+                       &end_prop, sizeof(end_prop))));
+    if (kernel_size) {
+        uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
+                              cpu_to_be64(kernel_size) };
+
+        _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
+        if (little_endian) {
+            _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
+        }
+    }
+    if (boot_menu) {
+        _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
+    }
+    _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
+    _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
+    _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
+
+    _FDT((fdt_end_node(fdt)));
+
+    /* RTAS */
+    _FDT((fdt_begin_node(fdt, "rtas")));
+
+    if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
+        add_str(hypertas, "hcall-multi-tce");
+    }
+    _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
+                       hypertas->len)));
+    g_string_free(hypertas, TRUE);
+    _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
+                       qemu_hypertas->len)));
+    g_string_free(qemu_hypertas, TRUE);
+
+    _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
+        refpoints, sizeof(refpoints))));
+
+    _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
+    _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
+                            RTAS_EVENT_SCAN_RATE)));
+
+    /*
+     * According to PAPR, rtas ibm,os-term does not guarantee a return
+     * back to the guest cpu.
+     *
+     * While an additional ibm,extended-os-term property indicates that
+     * rtas call return will always occur. Set this property.
+     */
+    _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
+
+    _FDT((fdt_end_node(fdt)));
+
+    /* interrupt controller */
+    _FDT((fdt_begin_node(fdt, "interrupt-controller")));
+
+    _FDT((fdt_property_string(fdt, "device_type",
+                              "PowerPC-External-Interrupt-Presentation")));
+    _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
+    _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
+    _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
+                       interrupt_server_ranges_prop,
+                       sizeof(interrupt_server_ranges_prop))));
+    _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
+    _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
+    _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
+
+    _FDT((fdt_end_node(fdt)));
+
+    /* vdevice */
+    _FDT((fdt_begin_node(fdt, "vdevice")));
+
+    _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
+    _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
+    _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
+    _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
+    _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
+    _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
+
+    _FDT((fdt_end_node(fdt)));
+
+    /* event-sources */
+    spapr_events_fdt_skel(fdt, epow_irq);
+
+    /* /hypervisor node */
+    if (kvm_enabled()) {
+        uint8_t hypercall[16];
+
+        /* indicate KVM hypercall interface */
+        _FDT((fdt_begin_node(fdt, "hypervisor")));
+        _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
+        if (kvmppc_has_cap_fixup_hcalls()) {
+            /*
+             * Older KVM versions with older guest kernels were broken with the
+             * magic page, don't allow the guest to map it.
+             */
+            kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
+                                 sizeof(hypercall));
+            _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
+                              sizeof(hypercall))));
+        }
+        _FDT((fdt_end_node(fdt)));
+    }
+
+    _FDT((fdt_end_node(fdt))); /* close root node */
+    _FDT((fdt_finish(fdt)));
+
+    return fdt;
+}
+
+int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
+                                 target_ulong addr, target_ulong size)
+{
+    void *fdt, *fdt_skel;
+    sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
+
+    size -= sizeof(hdr);
+
+    /* Create sceleton */
+    fdt_skel = g_malloc0(size);
+    _FDT((fdt_create(fdt_skel, size)));
+    _FDT((fdt_begin_node(fdt_skel, "")));
+    _FDT((fdt_end_node(fdt_skel)));
+    _FDT((fdt_finish(fdt_skel)));
+    fdt = g_malloc0(size);
+    _FDT((fdt_open_into(fdt_skel, fdt, size)));
+    g_free(fdt_skel);
+
+    /* Fix skeleton up */
+    _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
+
+    /* Pack resulting tree */
+    _FDT((fdt_pack(fdt)));
+
+    if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
+        trace_spapr_cas_failed(size);
+        return -1;
+    }
+
+    cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
+    cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
+    trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
+    g_free(fdt);
+
+    return 0;
+}
+
+static void spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
+                                       hwaddr size)
+{
+    uint32_t associativity[] = {
+        cpu_to_be32(0x4), /* length */
+        cpu_to_be32(0x0), cpu_to_be32(0x0),
+        cpu_to_be32(0x0), cpu_to_be32(nodeid)
+    };
+    char mem_name[32];
+    uint64_t mem_reg_property[2];
+    int off;
+
+    mem_reg_property[0] = cpu_to_be64(start);
+    mem_reg_property[1] = cpu_to_be64(size);
+
+    sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
+    off = fdt_add_subnode(fdt, 0, mem_name);
+    _FDT(off);
+    _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
+    _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
+                      sizeof(mem_reg_property))));
+    _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
+                      sizeof(associativity))));
+}
+
+static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
+{
+    MachineState *machine = MACHINE(spapr);
+    hwaddr mem_start, node_size;
+    int i, nb_nodes = nb_numa_nodes;
+    NodeInfo *nodes = numa_info;
+    NodeInfo ramnode;
+
+    /* No NUMA nodes, assume there is just one node with whole RAM */
+    if (!nb_numa_nodes) {
+        nb_nodes = 1;
+        ramnode.node_mem = machine->ram_size;
+        nodes = &ramnode;
+    }
+
+    for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
+        if (!nodes[i].node_mem) {
+            continue;
+        }
+        if (mem_start >= machine->ram_size) {
+            node_size = 0;
+        } else {
+            node_size = nodes[i].node_mem;
+            if (node_size > machine->ram_size - mem_start) {
+                node_size = machine->ram_size - mem_start;
+            }
+        }
+        if (!mem_start) {
+            /* ppc_spapr_init() checks for rma_size <= node0_size already */
+            spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
+            mem_start += spapr->rma_size;
+            node_size -= spapr->rma_size;
+        }
+        for ( ; node_size; ) {
+            hwaddr sizetmp = pow2floor(node_size);
+
+            /* mem_start != 0 here */
+            if (ctzl(mem_start) < ctzl(sizetmp)) {
+                sizetmp = 1ULL << ctzl(mem_start);
+            }
+
+            spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
+            node_size -= sizetmp;
+            mem_start += sizetmp;
+        }
+    }
+
+    return 0;
+}
+
+static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
+                                  sPAPRMachineState *spapr)
+{
+    PowerPCCPU *cpu = POWERPC_CPU(cs);
+    CPUPPCState *env = &cpu->env;
+    PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
+    int index = ppc_get_vcpu_dt_id(cpu);
+    uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
+                       0xffffffff, 0xffffffff};
+    uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
+    uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
+    uint32_t page_sizes_prop[64];
+    size_t page_sizes_prop_size;
+    QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
+    unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
+    uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
+    uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
+
+    _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
+    _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
+
+    _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
+    _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
+                           env->dcache_line_size)));
+    _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
+                           env->dcache_line_size)));
+    _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
+                           env->icache_line_size)));
+    _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
+                           env->icache_line_size)));
+
+    if (pcc->l1_dcache_size) {
+        _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
+                               pcc->l1_dcache_size)));
+    } else {
+        fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
+    }
+    if (pcc->l1_icache_size) {
+        _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
+                               pcc->l1_icache_size)));
+    } else {
+        fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
+    }
+
+    _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
+    _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
+    _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
+    _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
+    _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
+
+    if (env->spr_cb[SPR_PURR].oea_read) {
+        _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
+    }
+
+    if (env->mmu_model & POWERPC_MMU_1TSEG) {
+        _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
+                          segs, sizeof(segs))));
+    }
+
+    /* Advertise VMX/VSX (vector extensions) if available
+     *   0 / no property == no vector extensions
+     *   1               == VMX / Altivec available
+     *   2               == VSX available */
+    if (env->insns_flags & PPC_ALTIVEC) {
+        uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
+
+        _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
+    }
+
+    /* Advertise DFP (Decimal Floating Point) if available
+     *   0 / no property == no DFP
+     *   1               == DFP available */
+    if (env->insns_flags2 & PPC2_DFP) {
+        _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
+    }
+
+    page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
+                                                  sizeof(page_sizes_prop));
+    if (page_sizes_prop_size) {
+        _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
+                          page_sizes_prop, page_sizes_prop_size)));
+    }
+
+    _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
+                           cs->cpu_index / cpus_per_socket)));
+
+    _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
+                      pft_size_prop, sizeof(pft_size_prop))));
+
+    _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
+
+    _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
+                                ppc_get_compat_smt_threads(cpu)));
+}
+
+static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
+{
+    CPUState *cs;
+    int cpus_offset;
+    char *nodename;
+    int smt = kvmppc_smt_threads();
+
+    cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
+    _FDT(cpus_offset);
+    _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
+    _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
+
+    /*
+     * We walk the CPUs in reverse order to ensure that CPU DT nodes
+     * created by fdt_add_subnode() end up in the right order in FDT
+     * for the guest kernel the enumerate the CPUs correctly.
+     */
+    CPU_FOREACH_REVERSE(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+        int index = ppc_get_vcpu_dt_id(cpu);
+        DeviceClass *dc = DEVICE_GET_CLASS(cs);
+        int offset;
+
+        if ((index % smt) != 0) {
+            continue;
+        }
+
+        nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
+        offset = fdt_add_subnode(fdt, cpus_offset, nodename);
+        g_free(nodename);
+        _FDT(offset);
+        spapr_populate_cpu_dt(cs, fdt, offset, spapr);
+    }
+
+}
+
+static void spapr_finalize_fdt(sPAPRMachineState *spapr,
+                               hwaddr fdt_addr,
+                               hwaddr rtas_addr,
+                               hwaddr rtas_size)
+{
+    MachineState *machine = MACHINE(qdev_get_machine());
+    const char *boot_device = machine->boot_order;
+    int ret, i;
+    size_t cb = 0;
+    char *bootlist;
+    void *fdt;
+    sPAPRPHBState *phb;
+
+    fdt = g_malloc(FDT_MAX_SIZE);
+
+    /* open out the base tree into a temp buffer for the final tweaks */
+    _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
+
+    ret = spapr_populate_memory(spapr, fdt);
+    if (ret < 0) {
+        fprintf(stderr, "couldn't setup memory nodes in fdt\n");
+        exit(1);
+    }
+
+    ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
+    if (ret < 0) {
+        fprintf(stderr, "couldn't setup vio devices in fdt\n");
+        exit(1);
+    }
+
+    QLIST_FOREACH(phb, &spapr->phbs, list) {
+        ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
+    }
+
+    if (ret < 0) {
+        fprintf(stderr, "couldn't setup PCI devices in fdt\n");
+        exit(1);
+    }
+
+    /* RTAS */
+    ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
+    if (ret < 0) {
+        fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
+    }
+
+    /* cpus */
+    spapr_populate_cpus_dt_node(fdt, spapr);
+
+    bootlist = get_boot_devices_list(&cb, true);
+    if (cb && bootlist) {
+        int offset = fdt_path_offset(fdt, "/chosen");
+        if (offset < 0) {
+            exit(1);
+        }
+        for (i = 0; i < cb; i++) {
+            if (bootlist[i] == '\n') {
+                bootlist[i] = ' ';
+            }
+
+        }
+        ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
+    }
+
+    if (boot_device && strlen(boot_device)) {
+        int offset = fdt_path_offset(fdt, "/chosen");
+
+        if (offset < 0) {
+            exit(1);
+        }
+        fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
+    }
+
+    if (!spapr->has_graphics) {
+        spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
+    }
+
+    _FDT((fdt_pack(fdt)));
+
+    if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
+        error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
+                     fdt_totalsize(fdt), FDT_MAX_SIZE);
+        exit(1);
+    }
+
+    cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
+
+    g_free(bootlist);
+    g_free(fdt);
+}
+
+static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
+{
+    return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
+}
+
+static void emulate_spapr_hypercall(PowerPCCPU *cpu)
+{
+    CPUPPCState *env = &cpu->env;
+
+    if (msr_pr) {
+        hcall_dprintf("Hypercall made with MSR[PR]=1\n");
+        env->gpr[3] = H_PRIVILEGE;
+    } else {
+        env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
+    }
+}
+
+#define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
+#define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
+#define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
+#define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
+#define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
+
+static void spapr_reset_htab(sPAPRMachineState *spapr)
+{
+    long shift;
+    int index;
+
+    /* allocate hash page table.  For now we always make this 16mb,
+     * later we should probably make it scale to the size of guest
+     * RAM */
+
+    shift = kvmppc_reset_htab(spapr->htab_shift);
+
+    if (shift > 0) {
+        /* Kernel handles htab, we don't need to allocate one */
+        spapr->htab_shift = shift;
+        kvmppc_kern_htab = true;
+
+        /* Tell readers to update their file descriptor */
+        if (spapr->htab_fd >= 0) {
+            spapr->htab_fd_stale = true;
+        }
+    } else {
+        if (!spapr->htab) {
+            /* Allocate an htab if we don't yet have one */
+            spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
+        }
+
+        /* And clear it */
+        memset(spapr->htab, 0, HTAB_SIZE(spapr));
+
+        for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
+            DIRTY_HPTE(HPTE(spapr->htab, index));
+        }
+    }
+
+    /* Update the RMA size if necessary */
+    if (spapr->vrma_adjust) {
+        spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
+                                          spapr->htab_shift);
+    }
+}
+
+static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
+{
+    bool matched = false;
+
+    if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
+        matched = true;
+    }
+
+    if (!matched) {
+        error_report("Device %s is not supported by this machine yet.",
+                     qdev_fw_name(DEVICE(sbdev)));
+        exit(1);
+    }
+
+    return 0;
+}
+
+/*
+ * A guest reset will cause spapr->htab_fd to become stale if being used.
+ * Reopen the file descriptor to make sure the whole HTAB is properly read.
+ */
+static int spapr_check_htab_fd(sPAPRMachineState *spapr)
+{
+    int rc = 0;
+
+    if (spapr->htab_fd_stale) {
+        close(spapr->htab_fd);
+        spapr->htab_fd = kvmppc_get_htab_fd(false);
+        if (spapr->htab_fd < 0) {
+            error_report("Unable to open fd for reading hash table from KVM: "
+                         "%s", strerror(errno));
+            rc = -1;
+        }
+        spapr->htab_fd_stale = false;
+    }
+
+    return rc;
+}
+
+static void ppc_spapr_reset(void)
+{
+    sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
+    PowerPCCPU *first_ppc_cpu;
+    uint32_t rtas_limit;
+
+    /* Check for unknown sysbus devices */
+    foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
+
+    /* Reset the hash table & recalc the RMA */
+    spapr_reset_htab(spapr);
+
+    qemu_devices_reset();
+
+    /*
+     * We place the device tree and RTAS just below either the top of the RMA,
+     * or just below 2GB, whichever is lowere, so that it can be
+     * processed with 32-bit real mode code if necessary
+     */
+    rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
+    spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
+    spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
+
+    /* Load the fdt */
+    spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
+                       spapr->rtas_size);
+
+    /* Copy RTAS over */
+    cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
+                              spapr->rtas_size);
+
+    /* Set up the entry state */
+    first_ppc_cpu = POWERPC_CPU(first_cpu);
+    first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
+    first_ppc_cpu->env.gpr[5] = 0;
+    first_cpu->halted = 0;
+    first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
+
+}
+
+static void spapr_cpu_reset(void *opaque)
+{
+    sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
+    PowerPCCPU *cpu = opaque;
+    CPUState *cs = CPU(cpu);
+    CPUPPCState *env = &cpu->env;
+
+    cpu_reset(cs);
+
+    /* All CPUs start halted.  CPU0 is unhalted from the machine level
+     * reset code and the rest are explicitly started up by the guest
+     * using an RTAS call */
+    cs->halted = 1;
+
+    env->spr[SPR_HIOR] = 0;
+
+    env->external_htab = (uint8_t *)spapr->htab;
+    if (kvm_enabled() && !env->external_htab) {
+        /*
+         * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
+         * functions do the right thing.
+         */
+        env->external_htab = (void *)1;
+    }
+    env->htab_base = -1;
+    /*
+     * htab_mask is the mask used to normalize hash value to PTEG index.
+     * htab_shift is log2 of hash table size.
+     * We have 8 hpte per group, and each hpte is 16 bytes.
+     * ie have 128 bytes per hpte entry.
+     */
+    env->htab_mask = (1ULL << (spapr->htab_shift - 7)) - 1;
+    env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
+        (spapr->htab_shift - 18);
+}
+
+static void spapr_create_nvram(sPAPRMachineState *spapr)
+{
+    DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
+    DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
+
+    if (dinfo) {
+        qdev_prop_set_drive_nofail(dev, "drive", blk_by_legacy_dinfo(dinfo));
+    }
+
+    qdev_init_nofail(dev);
+
+    spapr->nvram = (struct sPAPRNVRAM *)dev;
+}
+
+static void spapr_rtc_create(sPAPRMachineState *spapr)
+{
+    DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
+
+    qdev_init_nofail(dev);
+    spapr->rtc = dev;
+
+    object_property_add_alias(qdev_get_machine(), "rtc-time",
+                              OBJECT(spapr->rtc), "date", NULL);
+}
+
+/* Returns whether we want to use VGA or not */
+static int spapr_vga_init(PCIBus *pci_bus)
+{
+    switch (vga_interface_type) {
+    case VGA_NONE:
+        return false;
+    case VGA_DEVICE:
+        return true;
+    case VGA_STD:
+        return pci_vga_init(pci_bus) != NULL;
+    default:
+        fprintf(stderr, "This vga model is not supported,"
+                "currently it only supports -vga std\n");
+        exit(0);
+    }
+}
+
+static int spapr_post_load(void *opaque, int version_id)
+{
+    sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
+    int err = 0;
+
+    /* In earlier versions, there was no separate qdev for the PAPR
+     * RTC, so the RTC offset was stored directly in sPAPREnvironment.
+     * So when migrating from those versions, poke the incoming offset
+     * value into the RTC device */
+    if (version_id < 3) {
+        err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
+    }
+
+    return err;
+}
+
+static bool version_before_3(void *opaque, int version_id)
+{
+    return version_id < 3;
+}
+
+static const VMStateDescription vmstate_spapr = {
+    .name = "spapr",
+    .version_id = 3,
+    .minimum_version_id = 1,
+    .post_load = spapr_post_load,
+    .fields = (VMStateField[]) {
+        /* used to be @next_irq */
+        VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
+
+        /* RTC offset */
+        VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
+
+        VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
+        VMSTATE_END_OF_LIST()
+    },
+};
+
+static int htab_save_setup(QEMUFile *f, void *opaque)
+{
+    sPAPRMachineState *spapr = opaque;
+
+    /* "Iteration" header */
+    qemu_put_be32(f, spapr->htab_shift);
+
+    if (spapr->htab) {
+        spapr->htab_save_index = 0;
+        spapr->htab_first_pass = true;
+    } else {
+        assert(kvm_enabled());
+
+        spapr->htab_fd = kvmppc_get_htab_fd(false);
+        spapr->htab_fd_stale = false;
+        if (spapr->htab_fd < 0) {
+            fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
+                    strerror(errno));
+            return -1;
+        }
+    }
+
+
+    return 0;
+}
+
+static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
+                                 int64_t max_ns)
+{
+    int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
+    int index = spapr->htab_save_index;
+    int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+
+    assert(spapr->htab_first_pass);
+
+    do {
+        int chunkstart;
+
+        /* Consume invalid HPTEs */
+        while ((index < htabslots)
+               && !HPTE_VALID(HPTE(spapr->htab, index))) {
+            index++;
+            CLEAN_HPTE(HPTE(spapr->htab, index));
+        }
+
+        /* Consume valid HPTEs */
+        chunkstart = index;
+        while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
+               && HPTE_VALID(HPTE(spapr->htab, index))) {
+            index++;
+            CLEAN_HPTE(HPTE(spapr->htab, index));
+        }
+
+        if (index > chunkstart) {
+            int n_valid = index - chunkstart;
+
+            qemu_put_be32(f, chunkstart);
+            qemu_put_be16(f, n_valid);
+            qemu_put_be16(f, 0);
+            qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
+                            HASH_PTE_SIZE_64 * n_valid);
+
+            if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
+                break;
+            }
+        }
+    } while ((index < htabslots) && !qemu_file_rate_limit(f));
+
+    if (index >= htabslots) {
+        assert(index == htabslots);
+        index = 0;
+        spapr->htab_first_pass = false;
+    }
+    spapr->htab_save_index = index;
+}
+
+static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
+                                int64_t max_ns)
+{
+    bool final = max_ns < 0;
+    int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
+    int examined = 0, sent = 0;
+    int index = spapr->htab_save_index;
+    int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
+
+    assert(!spapr->htab_first_pass);
+
+    do {
+        int chunkstart, invalidstart;
+
+        /* Consume non-dirty HPTEs */
+        while ((index < htabslots)
+               && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
+            index++;
+            examined++;
+        }
+
+        chunkstart = index;
+        /* Consume valid dirty HPTEs */
+        while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
+               && HPTE_DIRTY(HPTE(spapr->htab, index))
+               && HPTE_VALID(HPTE(spapr->htab, index))) {
+            CLEAN_HPTE(HPTE(spapr->htab, index));
+            index++;
+            examined++;
+        }
+
+        invalidstart = index;
+        /* Consume invalid dirty HPTEs */
+        while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
+               && HPTE_DIRTY(HPTE(spapr->htab, index))
+               && !HPTE_VALID(HPTE(spapr->htab, index))) {
+            CLEAN_HPTE(HPTE(spapr->htab, index));
+            index++;
+            examined++;
+        }
+
+        if (index > chunkstart) {
+            int n_valid = invalidstart - chunkstart;
+            int n_invalid = index - invalidstart;
+
+            qemu_put_be32(f, chunkstart);
+            qemu_put_be16(f, n_valid);
+            qemu_put_be16(f, n_invalid);
+            qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
+                            HASH_PTE_SIZE_64 * n_valid);
+            sent += index - chunkstart;
+
+            if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
+                break;
+            }
+        }
+
+        if (examined >= htabslots) {
+            break;
+        }
+
+        if (index >= htabslots) {
+            assert(index == htabslots);
+            index = 0;
+        }
+    } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
+
+    if (index >= htabslots) {
+        assert(index == htabslots);
+        index = 0;
+    }
+
+    spapr->htab_save_index = index;
+
+    return (examined >= htabslots) && (sent == 0) ? 1 : 0;
+}
+
+#define MAX_ITERATION_NS    5000000 /* 5 ms */
+#define MAX_KVM_BUF_SIZE    2048
+
+static int htab_save_iterate(QEMUFile *f, void *opaque)
+{
+    sPAPRMachineState *spapr = opaque;
+    int rc = 0;
+
+    /* Iteration header */
+    qemu_put_be32(f, 0);
+
+    if (!spapr->htab) {
+        assert(kvm_enabled());
+
+        rc = spapr_check_htab_fd(spapr);
+        if (rc < 0) {
+            return rc;
+        }
+
+        rc = kvmppc_save_htab(f, spapr->htab_fd,
+                              MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
+        if (rc < 0) {
+            return rc;
+        }
+    } else  if (spapr->htab_first_pass) {
+        htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
+    } else {
+        rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
+    }
+
+    /* End marker */
+    qemu_put_be32(f, 0);
+    qemu_put_be16(f, 0);
+    qemu_put_be16(f, 0);
+
+    return rc;
+}
+
+static int htab_save_complete(QEMUFile *f, void *opaque)
+{
+    sPAPRMachineState *spapr = opaque;
+
+    /* Iteration header */
+    qemu_put_be32(f, 0);
+
+    if (!spapr->htab) {
+        int rc;
+
+        assert(kvm_enabled());
+
+        rc = spapr_check_htab_fd(spapr);
+        if (rc < 0) {
+            return rc;
+        }
+
+        rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
+        if (rc < 0) {
+            return rc;
+        }
+        close(spapr->htab_fd);
+        spapr->htab_fd = -1;
+    } else {
+        htab_save_later_pass(f, spapr, -1);
+    }
+
+    /* End marker */
+    qemu_put_be32(f, 0);
+    qemu_put_be16(f, 0);
+    qemu_put_be16(f, 0);
+
+    return 0;
+}
+
+static int htab_load(QEMUFile *f, void *opaque, int version_id)
+{
+    sPAPRMachineState *spapr = opaque;
+    uint32_t section_hdr;
+    int fd = -1;
+
+    if (version_id < 1 || version_id > 1) {
+        fprintf(stderr, "htab_load() bad version\n");
+        return -EINVAL;
+    }
+
+    section_hdr = qemu_get_be32(f);
+
+    if (section_hdr) {
+        /* First section, just the hash shift */
+        if (spapr->htab_shift != section_hdr) {
+            return -EINVAL;
+        }
+        return 0;
+    }
+
+    if (!spapr->htab) {
+        assert(kvm_enabled());
+
+        fd = kvmppc_get_htab_fd(true);
+        if (fd < 0) {
+            fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
+                    strerror(errno));
+        }
+    }
+
+    while (true) {
+        uint32_t index;
+        uint16_t n_valid, n_invalid;
+
+        index = qemu_get_be32(f);
+        n_valid = qemu_get_be16(f);
+        n_invalid = qemu_get_be16(f);
+
+        if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
+            /* End of Stream */
+            break;
+        }
+
+        if ((index + n_valid + n_invalid) >
+            (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
+            /* Bad index in stream */
+            fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
+                    "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
+                    spapr->htab_shift);
+            return -EINVAL;
+        }
+
+        if (spapr->htab) {
+            if (n_valid) {
+                qemu_get_buffer(f, HPTE(spapr->htab, index),
+                                HASH_PTE_SIZE_64 * n_valid);
+            }
+            if (n_invalid) {
+                memset(HPTE(spapr->htab, index + n_valid), 0,
+                       HASH_PTE_SIZE_64 * n_invalid);
+            }
+        } else {
+            int rc;
+
+            assert(fd >= 0);
+
+            rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
+            if (rc < 0) {
+                return rc;
+            }
+        }
+    }
+
+    if (!spapr->htab) {
+        assert(fd >= 0);
+        close(fd);
+    }
+
+    return 0;
+}
+
+static SaveVMHandlers savevm_htab_handlers = {
+    .save_live_setup = htab_save_setup,
+    .save_live_iterate = htab_save_iterate,
+    .save_live_complete = htab_save_complete,
+    .load_state = htab_load,
+};
+
+static void spapr_boot_set(void *opaque, const char *boot_device,
+                           Error **errp)
+{
+    MachineState *machine = MACHINE(qdev_get_machine());
+    machine->boot_order = g_strdup(boot_device);
+}
+
+static void spapr_cpu_init(sPAPRMachineState *spapr, PowerPCCPU *cpu)
+{
+    CPUPPCState *env = &cpu->env;
+
+    /* Set time-base frequency to 512 MHz */
+    cpu_ppc_tb_init(env, TIMEBASE_FREQ);
+
+    /* PAPR always has exception vectors in RAM not ROM. To ensure this,
+     * MSR[IP] should never be set.
+     */
+    env->msr_mask &= ~(1 << 6);
+
+    /* Tell KVM that we're in PAPR mode */
+    if (kvm_enabled()) {
+        kvmppc_set_papr(cpu);
+    }
+
+    if (cpu->max_compat) {
+        if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
+            exit(1);
+        }
+    }
+
+    xics_cpu_setup(spapr->icp, cpu);
+
+    qemu_register_reset(spapr_cpu_reset, cpu);
+}
+
+/* pSeries LPAR / sPAPR hardware init */
+static void ppc_spapr_init(MachineState *machine)
+{
+    sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
+    const char *kernel_filename = machine->kernel_filename;
+    const char *kernel_cmdline = machine->kernel_cmdline;
+    const char *initrd_filename = machine->initrd_filename;
+    PowerPCCPU *cpu;
+    PCIHostState *phb;
+    int i;
+    MemoryRegion *sysmem = get_system_memory();
+    MemoryRegion *ram = g_new(MemoryRegion, 1);
+    MemoryRegion *rma_region;
+    void *rma = NULL;
+    hwaddr rma_alloc_size;
+    hwaddr node0_size = spapr_node0_size();
+    uint32_t initrd_base = 0;
+    long kernel_size = 0, initrd_size = 0;
+    long load_limit, fw_size;
+    bool kernel_le = false;
+    char *filename;
+
+    msi_supported = true;
+
+    QLIST_INIT(&spapr->phbs);
+
+    cpu_ppc_hypercall = emulate_spapr_hypercall;
+
+    /* Allocate RMA if necessary */
+    rma_alloc_size = kvmppc_alloc_rma(&rma);
+
+    if (rma_alloc_size == -1) {
+        error_report("Unable to create RMA");
+        exit(1);
+    }
+
+    if (rma_alloc_size && (rma_alloc_size < node0_size)) {
+        spapr->rma_size = rma_alloc_size;
+    } else {
+        spapr->rma_size = node0_size;
+
+        /* With KVM, we don't actually know whether KVM supports an
+         * unbounded RMA (PR KVM) or is limited by the hash table size
+         * (HV KVM using VRMA), so we always assume the latter
+         *
+         * In that case, we also limit the initial allocations for RTAS
+         * etc... to 256M since we have no way to know what the VRMA size
+         * is going to be as it depends on the size of the hash table
+         * isn't determined yet.
+         */
+        if (kvm_enabled()) {
+            spapr->vrma_adjust = 1;
+            spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
+        }
+    }
+
+    if (spapr->rma_size > node0_size) {
+        fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
+                spapr->rma_size);
+        exit(1);
+    }
+
+    /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
+    load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
+
+    /* We aim for a hash table of size 1/128 the size of RAM.  The
+     * normal rule of thumb is 1/64 the size of RAM, but that's much
+     * more than needed for the Linux guests we support. */
+    spapr->htab_shift = 18; /* Minimum architected size */
+    while (spapr->htab_shift <= 46) {
+        if ((1ULL << (spapr->htab_shift + 7)) >= machine->ram_size) {
+            break;
+        }
+        spapr->htab_shift++;
+    }
+
+    /* Set up Interrupt Controller before we create the VCPUs */
+    spapr->icp = xics_system_init(machine,
+                                  DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(),
+                                               smp_threads),
+                                  XICS_IRQS);
+
+    /* init CPUs */
+    if (machine->cpu_model == NULL) {
+        machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
+    }
+    for (i = 0; i < smp_cpus; i++) {
+        cpu = cpu_ppc_init(machine->cpu_model);
+        if (cpu == NULL) {
+            fprintf(stderr, "Unable to find PowerPC CPU definition\n");
+            exit(1);
+        }
+        spapr_cpu_init(spapr, cpu);
+    }
+
+    if (kvm_enabled()) {
+        /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
+        kvmppc_enable_logical_ci_hcalls();
+    }
+
+    /* allocate RAM */
+    memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
+                                         machine->ram_size);
+    memory_region_add_subregion(sysmem, 0, ram);
+
+    if (rma_alloc_size && rma) {
+        rma_region = g_new(MemoryRegion, 1);
+        memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
+                                   rma_alloc_size, rma);
+        vmstate_register_ram_global(rma_region);
+        memory_region_add_subregion(sysmem, 0, rma_region);
+    }
+
+    filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
+    if (!filename) {
+        error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
+        exit(1);
+    }
+    spapr->rtas_size = get_image_size(filename);
+    spapr->rtas_blob = g_malloc(spapr->rtas_size);
+    if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
+        error_report("Could not load LPAR rtas '%s'", filename);
+        exit(1);
+    }
+    if (spapr->rtas_size > RTAS_MAX_SIZE) {
+        error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
+                     (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
+        exit(1);
+    }
+    g_free(filename);
+
+    /* Set up EPOW events infrastructure */
+    spapr_events_init(spapr);
+
+    /* Set up the RTC RTAS interfaces */
+    spapr_rtc_create(spapr);
+
+    /* Set up VIO bus */
+    spapr->vio_bus = spapr_vio_bus_init();
+
+    for (i = 0; i < MAX_SERIAL_PORTS; i++) {
+        if (serial_hds[i]) {
+            spapr_vty_create(spapr->vio_bus, serial_hds[i]);
+        }
+    }
+
+    /* We always have at least the nvram device on VIO */
+    spapr_create_nvram(spapr);
+
+    /* Set up PCI */
+    spapr_pci_rtas_init();
+
+    phb = spapr_create_phb(spapr, 0);
+
+    for (i = 0; i < nb_nics; i++) {
+        NICInfo *nd = &nd_table[i];
+
+        if (!nd->model) {
+            nd->model = g_strdup("ibmveth");
+        }
+
+        if (strcmp(nd->model, "ibmveth") == 0) {
+            spapr_vlan_create(spapr->vio_bus, nd);
+        } else {
+            pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
+        }
+    }
+
+    for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
+        spapr_vscsi_create(spapr->vio_bus);
+    }
+
+    /* Graphics */
+    if (spapr_vga_init(phb->bus)) {
+        spapr->has_graphics = true;
+        machine->usb |= defaults_enabled() && !machine->usb_disabled;
+    }
+
+    if (machine->usb) {
+        pci_create_simple(phb->bus, -1, "pci-ohci");
+
+        if (spapr->has_graphics) {
+            USBBus *usb_bus = usb_bus_find(-1);
+
+            usb_create_simple(usb_bus, "usb-kbd");
+            usb_create_simple(usb_bus, "usb-mouse");
+        }
+    }
+
+    if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
+        fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
+                "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
+        exit(1);
+    }
+
+    if (kernel_filename) {
+        uint64_t lowaddr = 0;
+
+        kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
+                               NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
+        if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
+            kernel_size = load_elf(kernel_filename,
+                                   translate_kernel_address, NULL,
+                                   NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
+            kernel_le = kernel_size > 0;
+        }
+        if (kernel_size < 0) {
+            fprintf(stderr, "qemu: error loading %s: %s\n",
+                    kernel_filename, load_elf_strerror(kernel_size));
+            exit(1);
+        }
+
+        /* load initrd */
+        if (initrd_filename) {
+            /* Try to locate the initrd in the gap between the kernel
+             * and the firmware. Add a bit of space just in case
+             */
+            initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
+            initrd_size = load_image_targphys(initrd_filename, initrd_base,
+                                              load_limit - initrd_base);
+            if (initrd_size < 0) {
+                fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
+                        initrd_filename);
+                exit(1);
+            }
+        } else {
+            initrd_base = 0;
+            initrd_size = 0;
+        }
+    }
+
+    if (bios_name == NULL) {
+        bios_name = FW_FILE_NAME;
+    }
+    filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
+    if (!filename) {
+        error_report("Could not find LPAR firmware '%s'", bios_name);
+        exit(1);
+    }
+    fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
+    if (fw_size <= 0) {
+        error_report("Could not load LPAR firmware '%s'", filename);
+        exit(1);
+    }
+    g_free(filename);
+
+    /* FIXME: Should register things through the MachineState's qdev
+     * interface, this is a legacy from the sPAPREnvironment structure
+     * which predated MachineState but had a similar function */
+    vmstate_register(NULL, 0, &vmstate_spapr, spapr);
+    register_savevm_live(NULL, "spapr/htab", -1, 1,
+                         &savevm_htab_handlers, spapr);
+
+    /* Prepare the device tree */
+    spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
+                                            kernel_size, kernel_le,
+                                            kernel_cmdline,
+                                            spapr->check_exception_irq);
+    assert(spapr->fdt_skel != NULL);
+
+    /* used by RTAS */
+    QTAILQ_INIT(&spapr->ccs_list);
+    qemu_register_reset(spapr_ccs_reset_hook, spapr);
+
+    qemu_register_boot_set(spapr_boot_set, spapr);
+}
+
+static int spapr_kvm_type(const char *vm_type)
+{
+    if (!vm_type) {
+        return 0;
+    }
+
+    if (!strcmp(vm_type, "HV")) {
+        return 1;
+    }
+
+    if (!strcmp(vm_type, "PR")) {
+        return 2;
+    }
+
+    error_report("Unknown kvm-type specified '%s'", vm_type);
+    exit(1);
+}
+
+/*
+ * Implementation of an interface to adjust firmware path
+ * for the bootindex property handling.
+ */
+static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
+                                   DeviceState *dev)
+{
+#define CAST(type, obj, name) \
+    ((type *)object_dynamic_cast(OBJECT(obj), (name)))
+    SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
+    sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
+
+    if (d) {
+        void *spapr = CAST(void, bus->parent, "spapr-vscsi");
+        VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
+        USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
+
+        if (spapr) {
+            /*
+             * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
+             * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
+             * in the top 16 bits of the 64-bit LUN
+             */
+            unsigned id = 0x8000 | (d->id << 8) | d->lun;
+            return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
+                                   (uint64_t)id << 48);
+        } else if (virtio) {
+            /*
+             * We use SRP luns of the form 01000000 | (target << 8) | lun
+             * in the top 32 bits of the 64-bit LUN
+             * Note: the quote above is from SLOF and it is wrong,
+             * the actual binding is:
+             * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
+             */
+            unsigned id = 0x1000000 | (d->id << 16) | d->lun;
+            return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
+                                   (uint64_t)id << 32);
+        } else if (usb) {
+            /*
+             * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
+             * in the top 32 bits of the 64-bit LUN
+             */
+            unsigned usb_port = atoi(usb->port->path);
+            unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
+            return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
+                                   (uint64_t)id << 32);
+        }
+    }
+
+    if (phb) {
+        /* Replace "pci" with "pci@800000020000000" */
+        return g_strdup_printf("pci@%"PRIX64, phb->buid);
+    }
+
+    return NULL;
+}
+
+static char *spapr_get_kvm_type(Object *obj, Error **errp)
+{
+    sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
+
+    return g_strdup(spapr->kvm_type);
+}
+
+static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
+{
+    sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
+
+    g_free(spapr->kvm_type);
+    spapr->kvm_type = g_strdup(value);
+}
+
+static void spapr_machine_initfn(Object *obj)
+{
+    object_property_add_str(obj, "kvm-type",
+                            spapr_get_kvm_type, spapr_set_kvm_type, NULL);
+    object_property_set_description(obj, "kvm-type",
+                                    "Specifies the KVM virtualization mode (HV, PR)",
+                                    NULL);
+}
+
+static void ppc_cpu_do_nmi_on_cpu(void *arg)
+{
+    CPUState *cs = arg;
+
+    cpu_synchronize_state(cs);
+    ppc_cpu_do_system_reset(cs);
+}
+
+static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
+{
+    CPUState *cs;
+
+    CPU_FOREACH(cs) {
+        async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
+    }
+}
+
+static void spapr_machine_class_init(ObjectClass *oc, void *data)
+{
+    MachineClass *mc = MACHINE_CLASS(oc);
+    FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
+    NMIClass *nc = NMI_CLASS(oc);
+
+    mc->init = ppc_spapr_init;
+    mc->reset = ppc_spapr_reset;
+    mc->block_default_type = IF_SCSI;
+    mc->max_cpus = MAX_CPUS;
+    mc->no_parallel = 1;
+    mc->default_boot_order = "";
+    mc->default_ram_size = 512 * M_BYTE;
+    mc->kvm_type = spapr_kvm_type;
+    mc->has_dynamic_sysbus = true;
+
+    fwc->get_dev_path = spapr_get_fw_dev_path;
+    nc->nmi_monitor_handler = spapr_nmi;
+}
+
+static const TypeInfo spapr_machine_info = {
+    .name          = TYPE_SPAPR_MACHINE,
+    .parent        = TYPE_MACHINE,
+    .abstract      = true,
+    .instance_size = sizeof(sPAPRMachineState),
+    .instance_init = spapr_machine_initfn,
+    .class_size    = sizeof(sPAPRMachineClass),
+    .class_init    = spapr_machine_class_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_FW_PATH_PROVIDER },
+        { TYPE_NMI },
+        { }
+    },
+};
+
+#define SPAPR_COMPAT_2_3 \
+        HW_COMPAT_2_3 \
+        {\
+            .driver   = "spapr-pci-host-bridge",\
+            .property = "dynamic-reconfiguration",\
+            .value    = "off",\
+        },
+
+#define SPAPR_COMPAT_2_2 \
+        SPAPR_COMPAT_2_3 \
+        HW_COMPAT_2_2 \
+        {\
+            .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
+            .property = "mem_win_size",\
+            .value    = "0x20000000",\
+        },
+
+#define SPAPR_COMPAT_2_1 \
+        SPAPR_COMPAT_2_2 \
+        HW_COMPAT_2_1
+
+static void spapr_compat_2_3(Object *obj)
+{
+    savevm_skip_section_footers();
+    global_state_set_optional();
+}
+
+static void spapr_compat_2_2(Object *obj)
+{
+    spapr_compat_2_3(obj);
+}
+
+static void spapr_compat_2_1(Object *obj)
+{
+    spapr_compat_2_2(obj);
+}
+
+static void spapr_machine_2_3_instance_init(Object *obj)
+{
+    spapr_compat_2_3(obj);
+    spapr_machine_initfn(obj);
+}
+
+static void spapr_machine_2_2_instance_init(Object *obj)
+{
+    spapr_compat_2_2(obj);
+    spapr_machine_initfn(obj);
+}
+
+static void spapr_machine_2_1_instance_init(Object *obj)
+{
+    spapr_compat_2_1(obj);
+    spapr_machine_initfn(obj);
+}
+
+static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
+{
+    MachineClass *mc = MACHINE_CLASS(oc);
+    static GlobalProperty compat_props[] = {
+        SPAPR_COMPAT_2_1
+        { /* end of list */ }
+    };
+
+    mc->name = "pseries-2.1";
+    mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
+    mc->compat_props = compat_props;
+}
+
+static const TypeInfo spapr_machine_2_1_info = {
+    .name          = TYPE_SPAPR_MACHINE "2.1",
+    .parent        = TYPE_SPAPR_MACHINE,
+    .class_init    = spapr_machine_2_1_class_init,
+    .instance_init = spapr_machine_2_1_instance_init,
+};
+
+static void spapr_machine_2_2_class_init(ObjectClass *oc, void *data)
+{
+    static GlobalProperty compat_props[] = {
+        SPAPR_COMPAT_2_2
+        { /* end of list */ }
+    };
+    MachineClass *mc = MACHINE_CLASS(oc);
+
+    mc->name = "pseries-2.2";
+    mc->desc = "pSeries Logical Partition (PAPR compliant) v2.2";
+    mc->compat_props = compat_props;
+}
+
+static const TypeInfo spapr_machine_2_2_info = {
+    .name          = TYPE_SPAPR_MACHINE "2.2",
+    .parent        = TYPE_SPAPR_MACHINE,
+    .class_init    = spapr_machine_2_2_class_init,
+    .instance_init = spapr_machine_2_2_instance_init,
+};
+
+static void spapr_machine_2_3_class_init(ObjectClass *oc, void *data)
+{
+    static GlobalProperty compat_props[] = {
+        SPAPR_COMPAT_2_3
+        { /* end of list */ }
+    };
+    MachineClass *mc = MACHINE_CLASS(oc);
+
+    mc->name = "pseries-2.3";
+    mc->desc = "pSeries Logical Partition (PAPR compliant) v2.3";
+    mc->compat_props = compat_props;
+}
+
+static const TypeInfo spapr_machine_2_3_info = {
+    .name          = TYPE_SPAPR_MACHINE "2.3",
+    .parent        = TYPE_SPAPR_MACHINE,
+    .class_init    = spapr_machine_2_3_class_init,
+    .instance_init = spapr_machine_2_3_instance_init,
+};
+
+static void spapr_machine_2_4_class_init(ObjectClass *oc, void *data)
+{
+    MachineClass *mc = MACHINE_CLASS(oc);
+
+    mc->name = "pseries-2.4";
+    mc->desc = "pSeries Logical Partition (PAPR compliant) v2.4";
+    mc->alias = "pseries";
+    mc->is_default = 1;
+}
+
+static const TypeInfo spapr_machine_2_4_info = {
+    .name          = TYPE_SPAPR_MACHINE "2.4",
+    .parent        = TYPE_SPAPR_MACHINE,
+    .class_init    = spapr_machine_2_4_class_init,
+};
+
+static void spapr_machine_register_types(void)
+{
+    type_register_static(&spapr_machine_info);
+    type_register_static(&spapr_machine_2_1_info);
+    type_register_static(&spapr_machine_2_2_info);
+    type_register_static(&spapr_machine_2_3_info);
+    type_register_static(&spapr_machine_2_4_info);
+}
+
+type_init(spapr_machine_register_types)