int cpu = smp_processor_id();
 
        cpu_init();
+       preempt_disable();
+
        xen_enable_sysenter();
 
-       preempt_disable();
-       per_cpu(cpu_state, cpu) = CPU_ONLINE;
+       cpu = smp_processor_id();
+       smp_store_cpu_info(cpu);
+       cpu_data(cpu).x86_max_cores = 1;
+       set_cpu_sibling_map(cpu);
 
        xen_setup_cpu_clockevents();
 
+       cpu_set(cpu, cpu_online_map);
+       x86_write_percpu(cpu_state, CPU_ONLINE);
+       wmb();
+
        /* We can take interrupts now: we're officially "up". */
        local_irq_enable();
 
        return rc;
 }
 
-void __init xen_fill_possible_map(void)
+static void __init xen_fill_possible_map(void)
 {
        int i, rc;
 
 
 static void __init xen_smp_prepare_boot_cpu(void)
 {
-       int cpu;
-
        BUG_ON(smp_processor_id() != 0);
        native_smp_prepare_boot_cpu();
 
        /* We've switched to the "real" per-cpu gdt, so make sure the
           old memory can be recycled */
-       make_lowmem_page_readwrite(&per_cpu__gdt_page);
-
-       for_each_possible_cpu(cpu) {
-               cpus_clear(per_cpu(cpu_sibling_map, cpu));
-               /*
-                * cpu_core_map lives in a per cpu area that is cleared
-                * when the per cpu array is allocated.
-                *
-                * cpus_clear(per_cpu(cpu_core_map, cpu));
-                */
-       }
+       make_lowmem_page_readwrite(&per_cpu_var(gdt_page));
 
        xen_setup_vcpu_info_placement();
 }
 {
        unsigned cpu;
 
-       for_each_possible_cpu(cpu) {
-               cpus_clear(per_cpu(cpu_sibling_map, cpu));
-               /*
-                * cpu_core_ map will be zeroed when the per
-                * cpu area is allocated.
-                *
-                * cpus_clear(per_cpu(cpu_core_map, cpu));
-                */
-       }
-
        smp_store_cpu_info(0);
+       cpu_data(0).x86_max_cores = 1;
        set_cpu_sibling_map(0);
 
        if (xen_smp_intr_init(0))
 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
 {
        struct vcpu_guest_context *ctxt;
-       struct gdt_page *gdt = &per_cpu(gdt_page, cpu);
+       struct desc_struct *gdt;
 
        if (cpu_test_and_set(cpu, xen_cpu_initialized_map))
                return 0;
        if (ctxt == NULL)
                return -ENOMEM;
 
+       gdt = get_cpu_gdt_table(cpu);
+
        ctxt->flags = VGCF_IN_KERNEL;
        ctxt->user_regs.ds = __USER_DS;
        ctxt->user_regs.es = __USER_DS;
-       ctxt->user_regs.fs = __KERNEL_PERCPU;
-       ctxt->user_regs.gs = 0;
        ctxt->user_regs.ss = __KERNEL_DS;
+#ifdef CONFIG_X86_32
+       ctxt->user_regs.fs = __KERNEL_PERCPU;
+#endif
        ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
        ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
 
 
        ctxt->ldt_ents = 0;
 
-       BUG_ON((unsigned long)gdt->gdt & ~PAGE_MASK);
-       make_lowmem_page_readonly(gdt->gdt);
+       BUG_ON((unsigned long)gdt & ~PAGE_MASK);
+       make_lowmem_page_readonly(gdt);
 
-       ctxt->gdt_frames[0] = virt_to_mfn(gdt->gdt);
-       ctxt->gdt_ents      = ARRAY_SIZE(gdt->gdt);
+       ctxt->gdt_frames[0] = virt_to_mfn(gdt);
+       ctxt->gdt_ents      = GDT_ENTRIES;
 
        ctxt->user_regs.cs = __KERNEL_CS;
        ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
        ctxt->kernel_ss = __KERNEL_DS;
        ctxt->kernel_sp = idle->thread.sp0;
 
+#ifdef CONFIG_X86_32
        ctxt->event_callback_cs     = __KERNEL_CS;
-       ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
        ctxt->failsafe_callback_cs  = __KERNEL_CS;
+#endif
+       ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
        ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
 
        per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
                return rc;
 #endif
 
+#ifdef CONFIG_X86_64
+       /* Allocate node local memory for AP pdas */
+       WARN_ON(cpu == 0);
+       if (cpu > 0) {
+               rc = get_local_pda(cpu);
+               if (rc)
+                       return rc;
+       }
+#endif
+
+#ifdef CONFIG_X86_32
        init_gdt(cpu);
        per_cpu(current_task, cpu) = idle;
        irq_ctx_init(cpu);
+#else
+       cpu_pda(cpu)->pcurrent = idle;
+       clear_tsk_thread_flag(idle, TIF_FORK);
+#endif
        xen_setup_timer(cpu);
 
+       per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
+
        /* make sure interrupts start blocked */
        per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
 
        if (rc)
                return rc;
 
-       smp_store_cpu_info(cpu);
-       set_cpu_sibling_map(cpu);
-       /* This must be done before setting cpu_online_map */
-       wmb();
-
-       cpu_set(cpu, cpu_online_map);
-
        rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
        BUG_ON(rc);
 
+       while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
+               HYPERVISOR_sched_op(SCHEDOP_yield, 0);
+               barrier();
+       }
+
        return 0;
 }
 
 {
        irq_enter();
        generic_smp_call_function_interrupt();
+#ifdef CONFIG_X86_32
        __get_cpu_var(irq_stat).irq_call_count++;
+#else
+       add_pda(irq_call_count, 1);
+#endif
        irq_exit();
 
        return IRQ_HANDLED;
 {
        irq_enter();
        generic_smp_call_function_single_interrupt();
+#ifdef CONFIG_X86_32
        __get_cpu_var(irq_stat).irq_call_count++;
+#else
+       add_pda(irq_call_count, 1);
+#endif
        irq_exit();
 
        return IRQ_HANDLED;
 void __init xen_smp_init(void)
 {
        smp_ops = xen_smp_ops;
+       xen_fill_possible_map();
 }