]> git.proxmox.com Git - qemu.git/blobdiff - hw/spapr.c
cadence_uart: Fix buffer overflow
[qemu.git] / hw / spapr.c
index b1189755d30ba3aeedddf54466994e9409b952d1..c34b767c6e910c01bf47d653839c8e6fb1a1d844 100644 (file)
@@ -29,6 +29,9 @@
 #include "elf.h"
 #include "net.h"
 #include "blockdev.h"
+#include "cpus.h"
+#include "kvm.h"
+#include "kvm_ppc.h"
 
 #include "hw/boards.h"
 #include "hw/ppc.h"
 
 #include "hw/spapr.h"
 #include "hw/spapr_vio.h"
+#include "hw/spapr_pci.h"
 #include "hw/xics.h"
+#include "hw/msi.h"
 
 #include "kvm.h"
 #include "kvm_ppc.h"
+#include "pci.h"
+#include "vga-pci.h"
+
+#include "exec-memory.h"
+#include "hw/usb.h"
 
 #include <libfdt.h>
 
-#define KERNEL_LOAD_ADDR        0x00000000
-#define INITRD_LOAD_ADDR        0x02800000
+/* 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            0x10000
 #define RTAS_MAX_SIZE           0x10000
 #define FW_MAX_SIZE             0x400000
 #define FW_FILE_NAME            "slof.bin"
+#define FW_OVERHEAD             0x2800000
+#define KERNEL_LOAD_ADDR        FW_MAX_SIZE
 
-#define MIN_RAM_SLOF           512UL
+#define MIN_RMA_SLOF            128UL
 
 #define TIMEBASE_FREQ           512000000ULL
 
 #define MAX_CPUS                256
-#define XICS_IRQS              1024
+#define XICS_IRQS               1024
+
+#define SPAPR_PCI_BUID          0x800000020000001ULL
+#define SPAPR_PCI_MEM_WIN_ADDR  (0x10000000000ULL + 0xA0000000)
+#define SPAPR_PCI_MEM_WIN_SIZE  0x20000000
+#define SPAPR_PCI_IO_WIN_ADDR   (0x10000000000ULL + 0x80000000)
+#define SPAPR_PCI_MSI_WIN_ADDR  (0x10000000000ULL + 0x90000000)
 
 #define PHANDLE_XICP            0x00001111
 
 sPAPREnvironment *spapr;
 
-qemu_irq spapr_allocate_irq(uint32_t hint, uint32_t *irq_num)
+int spapr_allocate_irq(int hint, enum xics_irq_type type)
 {
-    uint32_t irq;
-    qemu_irq qirq;
+    int irq;
 
     if (hint) {
         irq = hint;
@@ -73,36 +98,143 @@ qemu_irq spapr_allocate_irq(uint32_t hint, uint32_t *irq_num)
         irq = spapr->next_irq++;
     }
 
-    qirq = xics_find_qirq(spapr->icp, irq);
-    if (!qirq) {
-        return NULL;
+    /* Configure irq type */
+    if (!xics_get_qirq(spapr->icp, irq)) {
+        return 0;
     }
 
-    if (irq_num) {
-        *irq_num = irq;
+    xics_set_irq_type(spapr->icp, irq, type);
+
+    return irq;
+}
+
+/* Allocate block of consequtive IRQs, returns a number of the first */
+int spapr_allocate_irq_block(int num, enum xics_irq_type type)
+{
+    int first = -1;
+    int i;
+
+    for (i = 0; i < num; ++i) {
+        int irq;
+
+        irq = spapr_allocate_irq(0, type);
+        if (!irq) {
+            return -1;
+        }
+
+        if (0 == i) {
+            first = irq;
+        }
+
+        /* If the above doesn't create a consecutive block then that's
+         * an internal bug */
+        assert(irq == (first + i));
     }
 
-    return qirq;
+    return first;
+}
+
+static int spapr_set_associativity(void *fdt, sPAPREnvironment *spapr)
+{
+    int ret = 0, offset;
+    CPUPPCState *env;
+    char cpu_model[32];
+    int smt = kvmppc_smt_threads();
+
+    assert(spapr->cpu_model);
+
+    for (env = first_cpu; env != NULL; env = env->next_cpu) {
+        uint32_t associativity[] = {cpu_to_be32(0x5),
+                                    cpu_to_be32(0x0),
+                                    cpu_to_be32(0x0),
+                                    cpu_to_be32(0x0),
+                                    cpu_to_be32(env->numa_node),
+                                    cpu_to_be32(env->cpu_index)};
+
+        if ((env->cpu_index % smt) != 0) {
+            continue;
+        }
+
+        snprintf(cpu_model, 32, "/cpus/%s@%x", spapr->cpu_model,
+                 env->cpu_index);
+
+        offset = fdt_path_offset(fdt, cpu_model);
+        if (offset < 0) {
+            return offset;
+        }
+
+        ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
+                          sizeof(associativity));
+        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 void *spapr_create_fdt_skel(const char *cpu_model,
+                                   target_phys_addr_t rma_size,
                                    target_phys_addr_t initrd_base,
                                    target_phys_addr_t initrd_size,
+                                   target_phys_addr_t kernel_size,
                                    const char *boot_device,
                                    const char *kernel_cmdline,
                                    long hash_shift)
 {
     void *fdt;
-    CPUState *env;
-    uint64_t mem_reg_property[] = { 0, cpu_to_be64(ram_size) };
+    CPUPPCState *env;
+    uint64_t mem_reg_property[2];
     uint32_t start_prop = cpu_to_be32(initrd_base);
     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
     uint32_t pft_size_prop[] = {0, cpu_to_be32(hash_shift)};
     char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
         "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk";
+    char qemu_hypertas_prop[] = "hcall-memop1";
     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
     int i;
     char *modelname;
+    int smt = kvmppc_smt_threads();
+    unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
+    uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
+    uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
+                                cpu_to_be32(0x0), cpu_to_be32(0x0),
+                                cpu_to_be32(0x0)};
+    char mem_name[32];
+    target_phys_addr_t node0_size, mem_start;
 
 #define _FDT(exp) \
     do { \
@@ -117,6 +249,12 @@ static void *spapr_create_fdt_skel(const char *cpu_model,
     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 */
@@ -130,24 +268,76 @@ static void *spapr_create_fdt_skel(const char *cpu_model,
     /* /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))));
+    }
     _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
+    _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)));
 
-    /* memory node */
-    _FDT((fdt_begin_node(fdt, "memory@0")));
+    /* memory node(s) */
+    node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
+    if (rma_size > node0_size) {
+        rma_size = node0_size;
+    }
 
+    /* RMA */
+    mem_reg_property[0] = 0;
+    mem_reg_property[1] = cpu_to_be64(rma_size);
+    _FDT((fdt_begin_node(fdt, "memory@0")));
     _FDT((fdt_property_string(fdt, "device_type", "memory")));
-    _FDT((fdt_property(fdt, "reg",
-                       mem_reg_property, sizeof(mem_reg_property))));
-
+    _FDT((fdt_property(fdt, "reg", mem_reg_property,
+        sizeof(mem_reg_property))));
+    _FDT((fdt_property(fdt, "ibm,associativity", associativity,
+        sizeof(associativity))));
     _FDT((fdt_end_node(fdt)));
 
+    /* RAM: Node 0 */
+    if (node0_size > rma_size) {
+        mem_reg_property[0] = cpu_to_be64(rma_size);
+        mem_reg_property[1] = cpu_to_be64(node0_size - rma_size);
+
+        sprintf(mem_name, "memory@" TARGET_FMT_lx, rma_size);
+        _FDT((fdt_begin_node(fdt, mem_name)));
+        _FDT((fdt_property_string(fdt, "device_type", "memory")));
+        _FDT((fdt_property(fdt, "reg", mem_reg_property,
+                           sizeof(mem_reg_property))));
+        _FDT((fdt_property(fdt, "ibm,associativity", associativity,
+                           sizeof(associativity))));
+        _FDT((fdt_end_node(fdt)));
+    }
+
+    /* RAM: Node 1 and beyond */
+    mem_start = node0_size;
+    for (i = 1; i < nb_numa_nodes; i++) {
+        mem_reg_property[0] = cpu_to_be64(mem_start);
+        mem_reg_property[1] = cpu_to_be64(node_mem[i]);
+        associativity[3] = associativity[4] = cpu_to_be32(i);
+        sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
+        _FDT((fdt_begin_node(fdt, mem_name)));
+        _FDT((fdt_property_string(fdt, "device_type", "memory")));
+        _FDT((fdt_property(fdt, "reg", mem_reg_property,
+            sizeof(mem_reg_property))));
+        _FDT((fdt_property(fdt, "ibm,associativity", associativity,
+            sizeof(associativity))));
+        _FDT((fdt_end_node(fdt)));
+        mem_start += node_mem[i];
+    }
+
     /* cpus */
     _FDT((fdt_begin_node(fdt, "cpus")));
 
@@ -160,14 +350,24 @@ static void *spapr_create_fdt_skel(const char *cpu_model,
         modelname[i] = toupper(modelname[i]);
     }
 
+    /* This is needed during FDT finalization */
+    spapr->cpu_model = g_strdup(modelname);
+
     for (env = first_cpu; env != NULL; env = env->next_cpu) {
         int index = env->cpu_index;
-        uint32_t gserver_prop[] = {cpu_to_be32(index), 0}; /* HACK! */
+        uint32_t servers_prop[smp_threads];
+        uint32_t gservers_prop[smp_threads * 2];
         char *nodename;
         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;
+
+        if ((index % smt) != 0) {
+            continue;
+        }
 
         if (asprintf(&nodename, "%s@%x", modelname, index) < 0) {
             fprintf(stderr, "Allocation failure\n");
@@ -193,15 +393,48 @@ static void *spapr_create_fdt_skel(const char *cpu_model,
                            pft_size_prop, sizeof(pft_size_prop))));
         _FDT((fdt_property_string(fdt, "status", "okay")));
         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
-        _FDT((fdt_property_cell(fdt, "ibm,ppc-interrupt-server#s", index)));
+
+        /* Build interrupt servers and gservers properties */
+        for (i = 0; i < smp_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;
+        }
+        _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
+                           servers_prop, sizeof(servers_prop))));
         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
-                           gserver_prop, sizeof(gserver_prop))));
+                           gservers_prop, sizeof(gservers_prop))));
 
         if (env->mmu_model & POWERPC_MMU_1TSEG) {
             _FDT((fdt_property(fdt, "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_property_cell(fdt, "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_property_cell(fdt, "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_property(fdt, "ibm,segment-page-sizes",
+                               page_sizes_prop, page_sizes_prop_size)));
+        }
+
         _FDT((fdt_end_node(fdt)));
     }
 
@@ -214,6 +447,11 @@ static void *spapr_create_fdt_skel(const char *cpu_model,
 
     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
                        sizeof(hypertas_prop))));
+    _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
+                       sizeof(qemu_hypertas_prop))));
+
+    _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
+        refpoints, sizeof(refpoints))));
 
     _FDT((fdt_end_node(fdt)));
 
@@ -258,6 +496,7 @@ static void spapr_finalize_fdt(sPAPREnvironment *spapr,
 {
     int ret;
     void *fdt;
+    sPAPRPHBState *phb;
 
     fdt = g_malloc(FDT_MAX_SIZE);
 
@@ -270,14 +509,41 @@ static void spapr_finalize_fdt(sPAPREnvironment *spapr,
         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");
     }
 
+    /* Advertise NUMA via ibm,associativity */
+    if (nb_numa_nodes > 1) {
+        ret = spapr_set_associativity(fdt, spapr);
+        if (ret < 0) {
+            fprintf(stderr, "Couldn't set up NUMA device tree properties\n");
+        }
+    }
+
+    if (!spapr->has_graphics) {
+        spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
+    }
+
     _FDT((fdt_pack(fdt)));
 
+    if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
+        hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
+                 fdt_totalsize(fdt), FDT_MAX_SIZE);
+        exit(1);
+    }
+
     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
 
     g_free(fdt);
@@ -288,7 +554,7 @@ static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
 }
 
-static void emulate_spapr_hypercall(CPUState *env)
+static void emulate_spapr_hypercall(CPUPPCState *env)
 {
     env->gpr[3] = spapr_hypercall(env, env->gpr[3], &env->gpr[4]);
 }
@@ -297,8 +563,6 @@ static void spapr_reset(void *opaque)
 {
     sPAPREnvironment *spapr = (sPAPREnvironment *)opaque;
 
-    fprintf(stderr, "sPAPR reset\n");
-
     /* flush out the hash table */
     memset(spapr->htab, 0, spapr->htab_size);
 
@@ -314,6 +578,30 @@ static void spapr_reset(void *opaque)
 
 }
 
+static void spapr_cpu_reset(void *opaque)
+{
+    PowerPCCPU *cpu = opaque;
+
+    cpu_reset(CPU(cpu));
+}
+
+/* Returns whether we want to use VGA or not */
+static int spapr_vga_init(PCIBus *pci_bus)
+{
+    switch (vga_interface_type) {
+    case VGA_STD:
+        pci_vga_init(pci_bus);
+        return 1;
+    case VGA_NONE:
+        return 0;
+    default:
+        fprintf(stderr, "This vga model is not supported,"
+                "currently it only supports -vga std\n");
+        exit(0);
+        break;
+    }
+}
+
 /* pSeries LPAR / sPAPR hardware init */
 static void ppc_spapr_init(ram_addr_t ram_size,
                            const char *boot_device,
@@ -322,37 +610,62 @@ static void ppc_spapr_init(ram_addr_t ram_size,
                            const char *initrd_filename,
                            const char *cpu_model)
 {
-    CPUState *env;
+    PowerPCCPU *cpu;
+    CPUPPCState *env;
+    PCIHostState *phb;
     int i;
-    ram_addr_t ram_offset;
-    uint32_t initrd_base;
-    long kernel_size, initrd_size, fw_size;
+    MemoryRegion *sysmem = get_system_memory();
+    MemoryRegion *ram = g_new(MemoryRegion, 1);
+    target_phys_addr_t rma_alloc_size, rma_size;
+    uint32_t initrd_base = 0;
+    long kernel_size = 0, initrd_size = 0;
+    long load_limit, rtas_limit, fw_size;
     long pteg_shift = 17;
     char *filename;
 
-    spapr = g_malloc(sizeof(*spapr));
+    msi_supported = true;
+
+    spapr = g_malloc0(sizeof(*spapr));
+    QLIST_INIT(&spapr->phbs);
+
     cpu_ppc_hypercall = emulate_spapr_hypercall;
 
-    /* We place the device tree just below either the top of RAM, or
-     * 2GB, so that it can be processed with 32-bit code if
-     * necessary */
-    spapr->fdt_addr = MIN(ram_size, 0x80000000) - FDT_MAX_SIZE;
-    spapr->rtas_addr = spapr->fdt_addr - RTAS_MAX_SIZE;
+    /* Allocate RMA if necessary */
+    rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
+
+    if (rma_alloc_size == -1) {
+        hw_error("qemu: Unable to create RMA\n");
+        exit(1);
+    }
+    if (rma_alloc_size && (rma_alloc_size < ram_size)) {
+        rma_size = rma_alloc_size;
+    } else {
+        rma_size = ram_size;
+    }
+
+    /* 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(rma_size, 0x80000000);
+    spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
+    spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
+    load_limit = spapr->fdt_addr - FW_OVERHEAD;
 
     /* init CPUs */
     if (cpu_model == NULL) {
-        cpu_model = "POWER7";
+        cpu_model = kvm_enabled() ? "host" : "POWER7";
     }
     for (i = 0; i < smp_cpus; i++) {
-        env = cpu_init(cpu_model);
-
-        if (!env) {
+        cpu = cpu_ppc_init(cpu_model);
+        if (cpu == NULL) {
             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
             exit(1);
         }
+        env = &cpu->env;
+
         /* Set time-base frequency to 512 MHz */
         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
-        qemu_register_reset((QEMUResetHandler *)&cpu_reset, env);
+        qemu_register_reset(spapr_cpu_reset, cpu);
 
         env->hreset_vector = 0x60;
         env->hreset_excp_prefix = 0;
@@ -361,8 +674,14 @@ static void ppc_spapr_init(ram_addr_t ram_size,
 
     /* allocate RAM */
     spapr->ram_limit = ram_size;
-    ram_offset = qemu_ram_alloc(NULL, "ppc_spapr.ram", spapr->ram_limit);
-    cpu_register_physical_memory(0, ram_size, ram_offset);
+    if (spapr->ram_limit > rma_alloc_size) {
+        ram_addr_t nonrma_base = rma_alloc_size;
+        ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
+
+        memory_region_init_ram(ram, "ppc_spapr.ram", nonrma_size);
+        vmstate_register_ram_global(ram);
+        memory_region_add_subregion(sysmem, nonrma_base, ram);
+    }
 
     /* allocate hash page table.  For now we always make this 16mb,
      * later we should probably make it scale to the size of guest
@@ -387,27 +706,45 @@ static void ppc_spapr_init(ram_addr_t ram_size,
 
     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
-                                           ram_size - spapr->rtas_addr);
+                                           rtas_limit - spapr->rtas_addr);
     if (spapr->rtas_size < 0) {
         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
         exit(1);
     }
+    if (spapr->rtas_size > RTAS_MAX_SIZE) {
+        hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
+                 spapr->rtas_size, RTAS_MAX_SIZE);
+        exit(1);
+    }
     g_free(filename);
 
+
     /* Set up Interrupt Controller */
     spapr->icp = xics_system_init(XICS_IRQS);
     spapr->next_irq = 16;
 
+    /* Set up IOMMU */
+    spapr_iommu_init();
+
     /* 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, SPAPR_VTY_BASE_ADDRESS + i,
-                             serial_hds[i]);
+            spapr_vty_create(spapr->vio_bus, serial_hds[i]);
         }
     }
 
+    /* Set up PCI */
+    spapr_pci_rtas_init();
+
+    spapr_create_phb(spapr, "pci", SPAPR_PCI_BUID,
+                     SPAPR_PCI_MEM_WIN_ADDR,
+                     SPAPR_PCI_MEM_WIN_SIZE,
+                     SPAPR_PCI_IO_WIN_ADDR,
+                     SPAPR_PCI_MSI_WIN_ADDR);
+    phb = PCI_HOST_BRIDGE(QLIST_FIRST(&spapr->phbs));
+
     for (i = 0; i < nb_nics; i++) {
         NICInfo *nd = &nd_table[i];
 
@@ -416,17 +753,33 @@ static void ppc_spapr_init(ram_addr_t ram_size,
         }
 
         if (strcmp(nd->model, "ibmveth") == 0) {
-            spapr_vlan_create(spapr->vio_bus, 0x1000 + i, nd);
+            spapr_vlan_create(spapr->vio_bus, nd);
         } else {
-            fprintf(stderr, "pSeries (sPAPR) platform does not support "
-                    "NIC model '%s' (only ibmveth is supported)\n",
-                    nd->model);
-            exit(1);
+            pci_nic_init_nofail(&nd_table[i], nd->model, NULL);
         }
     }
 
     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
-        spapr_vscsi_create(spapr->vio_bus, 0x2000 + i);
+        spapr_vscsi_create(spapr->vio_bus);
+    }
+
+    /* Graphics */
+    if (spapr_vga_init(phb->bus)) {
+        spapr->has_graphics = true;
+    }
+
+    if (usb_enabled) {
+        pci_create_simple(phb->bus, -1, "pci-ohci");
+        if (spapr->has_graphics) {
+            usbdevice_create("keyboard");
+            usbdevice_create("mouse");
+        }
+    }
+
+    if (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) {
@@ -437,7 +790,7 @@ static void ppc_spapr_init(ram_addr_t ram_size,
         if (kernel_size < 0) {
             kernel_size = load_image_targphys(kernel_filename,
                                               KERNEL_LOAD_ADDR,
-                                              ram_size - KERNEL_LOAD_ADDR);
+                                              load_limit - KERNEL_LOAD_ADDR);
         }
         if (kernel_size < 0) {
             fprintf(stderr, "qemu: could not load kernel '%s'\n",
@@ -447,9 +800,12 @@ static void ppc_spapr_init(ram_addr_t ram_size,
 
         /* load initrd */
         if (initrd_filename) {
-            initrd_base = INITRD_LOAD_ADDR;
+            /* 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,
-                                              ram_size - initrd_base);
+                                              load_limit - initrd_base);
             if (initrd_size < 0) {
                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
                         initrd_filename);
@@ -459,35 +815,27 @@ static void ppc_spapr_init(ram_addr_t ram_size,
             initrd_base = 0;
             initrd_size = 0;
         }
+    }
 
-        spapr->entry_point = KERNEL_LOAD_ADDR;
-    } else {
-        if (ram_size < (MIN_RAM_SLOF << 20)) {
-            fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
-                    "%ldM guest RAM\n", MIN_RAM_SLOF);
-            exit(1);
-        }
-        filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, FW_FILE_NAME);
-        fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
-        if (fw_size < 0) {
-            hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
-            exit(1);
-        }
-        g_free(filename);
-        spapr->entry_point = 0x100;
-        initrd_base = 0;
-        initrd_size = 0;
-
-        /* SLOF will startup the secondary CPUs using RTAS,
-           rather than expecting a kexec() style entry */
-        for (env = first_cpu; env != NULL; env = env->next_cpu) {
-            env->halted = 1;
-        }
+    filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, FW_FILE_NAME);
+    fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
+    if (fw_size < 0) {
+        hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
+        exit(1);
+    }
+    g_free(filename);
+
+    spapr->entry_point = 0x100;
+
+    /* SLOF will startup the secondary CPUs using RTAS */
+    for (env = first_cpu; env != NULL; env = env->next_cpu) {
+        env->halted = 1;
     }
 
     /* Prepare the device tree */
-    spapr->fdt_skel = spapr_create_fdt_skel(cpu_model,
+    spapr->fdt_skel = spapr_create_fdt_skel(cpu_model, rma_size,
                                             initrd_base, initrd_size,
+                                            kernel_size,
                                             boot_device, kernel_cmdline,
                                             pteg_shift + 7);
     assert(spapr->fdt_skel != NULL);
@@ -500,7 +848,6 @@ static QEMUMachine spapr_machine = {
     .desc = "pSeries Logical Partition (PAPR compliant)",
     .init = ppc_spapr_init,
     .max_cpus = MAX_CPUS,
-    .no_vga = 1,
     .no_parallel = 1,
     .use_scsi = 1,
 };