1.1 --- a/xen/arch/x86/acpi/power.c	Wed Jul 11 15:47:14 2007 +0100
     1.2 +++ b/xen/arch/x86/acpi/power.c	Wed Jul 11 17:23:09 2007 +0100
     1.3 @@ -82,10 +82,27 @@ static void device_power_up(void)
     1.4      console_resume();
     1.5  }
     1.6  
     1.7 +static void freeze_domains(void)
     1.8 +{
     1.9 +    struct domain *d;
    1.10 +
    1.11 +    for_each_domain(d)
    1.12 +        if (d->domain_id != 0)
    1.13 +            domain_pause(d);
    1.14 +}
    1.15 +
    1.16 +static void thaw_domains(void)
    1.17 +{
    1.18 +    struct domain *d;
    1.19 +
    1.20 +    for_each_domain(d)
    1.21 +        if (d->domain_id != 0)
    1.22 +            domain_unpause(d);
    1.23 +}
    1.24 +
    1.25  /* Main interface to do xen specific suspend/resume */
    1.26  int enter_state(u32 state)
    1.27  {
    1.28 -    struct domain *d;
    1.29      unsigned long flags;
    1.30      int error;
    1.31  
    1.32 @@ -99,9 +116,9 @@ int enter_state(u32 state)
    1.33      if (!spin_trylock(&pm_lock))
    1.34          return -EBUSY;
    1.35      
    1.36 -    for_each_domain(d)
    1.37 -        if (d->domain_id != 0)
    1.38 -            domain_pause(d);
    1.39 +    freeze_domains();
    1.40 +
    1.41 +    hvm_suspend_cpu();
    1.42  
    1.43      pmprintk(XENLOG_INFO, "PM: Preparing system for %s sleep\n",
    1.44          acpi_states[state]);
    1.45 @@ -135,13 +152,11 @@ int enter_state(u32 state)
    1.46   Done:
    1.47      local_irq_restore(flags);
    1.48  
    1.49 -    for_each_domain(d)
    1.50 -       if (d->domain_id!=0)
    1.51 -           domain_unpause(d);
    1.52 +    hvm_resume_cpu();
    1.53  
    1.54 +    thaw_domains();
    1.55      spin_unlock(&pm_lock);
    1.56      return error;
    1.57 -
    1.58  }
    1.59  
    1.60  /*

     2.1 --- a/xen/arch/x86/hvm/hvm.c	Wed Jul 11 15:47:14 2007 +0100
     2.2 +++ b/xen/arch/x86/hvm/hvm.c	Wed Jul 11 17:23:09 2007 +0100
     2.3 @@ -78,8 +78,7 @@ void hvm_enable(struct hvm_function_tabl
     2.4  
     2.5  void hvm_disable(void)
     2.6  {
     2.7 -    if ( hvm_enabled )
     2.8 -        hvm_funcs.disable();
     2.9 +    hvm_suspend_cpu();
    2.10  }
    2.11  
    2.12  void hvm_stts(struct vcpu *v)

     3.1 --- a/xen/arch/x86/hvm/svm/svm.c	Wed Jul 11 15:47:14 2007 +0100
     3.2 +++ b/xen/arch/x86/hvm/svm/svm.c	Wed Jul 11 17:23:09 2007 +0100
     3.3 @@ -94,9 +94,8 @@ static void svm_inject_exception(struct 
     3.4      vmcb->eventinj = event;
     3.5  }
     3.6  
     3.7 -static void stop_svm(void)
     3.8 +static void svm_suspend_cpu(void)
     3.9  {
    3.10 -    /* We turn off the EFER_SVME bit. */
    3.11      write_efer(read_efer() & ~EFER_SVME);
    3.12  }
    3.13  
    3.14 @@ -974,7 +973,7 @@ static int svm_event_injection_faulted(s
    3.15  
    3.16  static struct hvm_function_table svm_function_table = {
    3.17      .name                 = "SVM",
    3.18 -    .disable              = stop_svm,
    3.19 +    .suspend_cpu          = svm_suspend_cpu,
    3.20      .domain_initialise    = svm_domain_initialise,
    3.21      .domain_destroy       = svm_domain_destroy,
    3.22      .vcpu_initialise      = svm_vcpu_initialise,

     4.1 --- a/xen/arch/x86/hvm/vmx/vmcs.c	Wed Jul 11 15:47:14 2007 +0100
     4.2 +++ b/xen/arch/x86/hvm/vmx/vmcs.c	Wed Jul 11 17:23:09 2007 +0100
     4.3 @@ -45,7 +45,9 @@ u32 vmx_vmexit_control __read_mostly;
     4.4  u32 vmx_vmentry_control __read_mostly;
     4.5  bool_t cpu_has_vmx_ins_outs_instr_info __read_mostly;
     4.6  
     4.7 +static DEFINE_PER_CPU(struct vmcs_struct *, host_vmcs);
     4.8  static DEFINE_PER_CPU(struct vmcs_struct *, current_vmcs);
     4.9 +static DEFINE_PER_CPU(struct list_head, active_vmcs_list);
    4.10  
    4.11  static u32 vmcs_revision_id __read_mostly;
    4.12  
    4.13 @@ -185,34 +187,81 @@ static void vmx_free_vmcs(struct vmcs_st
    4.14  static void __vmx_clear_vmcs(void *info)
    4.15  {
    4.16      struct vcpu *v = info;
    4.17 +    struct arch_vmx_struct *arch_vmx = &v->arch.hvm_vmx;
    4.18  
    4.19 -    __vmpclear(virt_to_maddr(v->arch.hvm_vmx.vmcs));
    4.20 +    /* Otherwise we can nest (vmx_suspend_cpu() vs. vmx_clear_vmcs()). */
    4.21 +    ASSERT(!local_irq_is_enabled());
    4.22 +
    4.23 +    if ( arch_vmx->active_cpu == smp_processor_id() )
    4.24 +    {
    4.25 +        __vmpclear(virt_to_maddr(arch_vmx->vmcs));
    4.26  
    4.27 -    v->arch.hvm_vmx.active_cpu = -1;
    4.28 -    v->arch.hvm_vmx.launched   = 0;
    4.29 +        arch_vmx->active_cpu = -1;
    4.30 +        arch_vmx->launched   = 0;
    4.31  
    4.32 -    if ( v->arch.hvm_vmx.vmcs == this_cpu(current_vmcs) )
    4.33 -        this_cpu(current_vmcs) = NULL;
    4.34 +        list_del(&arch_vmx->active_list);
    4.35 +
    4.36 +        if ( arch_vmx->vmcs == this_cpu(current_vmcs) )
    4.37 +            this_cpu(current_vmcs) = NULL;
    4.38 +    }
    4.39  }
    4.40  
    4.41  static void vmx_clear_vmcs(struct vcpu *v)
    4.42  {
    4.43      int cpu = v->arch.hvm_vmx.active_cpu;
    4.44  
    4.45 -    if ( cpu == -1 )
    4.46 -        return;
    4.47 -
    4.48 -    if ( cpu == smp_processor_id() )
    4.49 -        return __vmx_clear_vmcs(v);
    4.50 -
    4.51 -    on_selected_cpus(cpumask_of_cpu(cpu), __vmx_clear_vmcs, v, 1, 1);
    4.52 +    if ( cpu != -1 )
    4.53 +        on_selected_cpus(cpumask_of_cpu(cpu), __vmx_clear_vmcs, v, 1, 1);
    4.54  }
    4.55  
    4.56  static void vmx_load_vmcs(struct vcpu *v)
    4.57  {
    4.58 +    unsigned long flags;
    4.59 +
    4.60 +    local_irq_save(flags);
    4.61 +
    4.62 +    if ( v->arch.hvm_vmx.active_cpu == -1 )
    4.63 +    {
    4.64 +        list_add(&v->arch.hvm_vmx.active_list, &this_cpu(active_vmcs_list));
    4.65 +        v->arch.hvm_vmx.active_cpu = smp_processor_id();
    4.66 +    }
    4.67 +
    4.68 +    ASSERT(v->arch.hvm_vmx.active_cpu == smp_processor_id());
    4.69 +
    4.70      __vmptrld(virt_to_maddr(v->arch.hvm_vmx.vmcs));
    4.71 -    v->arch.hvm_vmx.active_cpu = smp_processor_id();
    4.72      this_cpu(current_vmcs) = v->arch.hvm_vmx.vmcs;
    4.73 +
    4.74 +    local_irq_restore(flags);
    4.75 +}
    4.76 +
    4.77 +void vmx_suspend_cpu(void)
    4.78 +{
    4.79 +    struct list_head *active_vmcs_list = &this_cpu(active_vmcs_list);
    4.80 +    unsigned long flags;
    4.81 +
    4.82 +    local_irq_save(flags);
    4.83 +
    4.84 +    while ( !list_empty(active_vmcs_list) )
    4.85 +        __vmx_clear_vmcs(list_entry(active_vmcs_list->next,
    4.86 +                                    struct vcpu, arch.hvm_vmx.active_list));
    4.87 +
    4.88 +    if ( read_cr4() & X86_CR4_VMXE )
    4.89 +    {
    4.90 +        __vmxoff();
    4.91 +        clear_in_cr4(X86_CR4_VMXE);
    4.92 +    }
    4.93 +
    4.94 +    local_irq_restore(flags);
    4.95 +}
    4.96 +
    4.97 +void vmx_resume_cpu(void)
    4.98 +{
    4.99 +    if ( !read_cr4() & X86_CR4_VMXE )
   4.100 +    {
   4.101 +        set_in_cr4(X86_CR4_VMXE);
   4.102 +        if ( __vmxon(virt_to_maddr(this_cpu(host_vmcs))) )
   4.103 +            BUG();
   4.104 +    }
   4.105  }
   4.106  
   4.107  void vmx_vmcs_enter(struct vcpu *v)
   4.108 @@ -247,12 +296,17 @@ void vmx_vmcs_exit(struct vcpu *v)
   4.109  
   4.110  struct vmcs_struct *vmx_alloc_host_vmcs(void)
   4.111  {
   4.112 -    return vmx_alloc_vmcs();
   4.113 +    ASSERT(this_cpu(host_vmcs) == NULL);
   4.114 +    this_cpu(host_vmcs) = vmx_alloc_vmcs();
   4.115 +    INIT_LIST_HEAD(&this_cpu(active_vmcs_list));
   4.116 +    return this_cpu(host_vmcs);
   4.117  }
   4.118  
   4.119  void vmx_free_host_vmcs(struct vmcs_struct *vmcs)
   4.120  {
   4.121 +    ASSERT(vmcs == this_cpu(host_vmcs));
   4.122      vmx_free_vmcs(vmcs);
   4.123 +    this_cpu(host_vmcs) = NULL;
   4.124  }
   4.125  
   4.126  struct xgt_desc {
   4.127 @@ -451,12 +505,17 @@ static void construct_vmcs(struct vcpu *
   4.128  
   4.129  int vmx_create_vmcs(struct vcpu *v)
   4.130  {
   4.131 -    if ( v->arch.hvm_vmx.vmcs == NULL )
   4.132 +    struct arch_vmx_struct *arch_vmx = &v->arch.hvm_vmx;
   4.133 +
   4.134 +    if ( arch_vmx->vmcs == NULL )
   4.135      {
   4.136 -        if ( (v->arch.hvm_vmx.vmcs = vmx_alloc_vmcs()) == NULL )
   4.137 +        if ( (arch_vmx->vmcs = vmx_alloc_vmcs()) == NULL )
   4.138              return -ENOMEM;
   4.139  
   4.140 -        __vmx_clear_vmcs(v);
   4.141 +        INIT_LIST_HEAD(&arch_vmx->active_list);
   4.142 +        __vmpclear(virt_to_maddr(arch_vmx->vmcs));
   4.143 +        arch_vmx->active_cpu = -1;
   4.144 +        arch_vmx->launched   = 0;
   4.145      }
   4.146  
   4.147      construct_vmcs(v);

     5.1 --- a/xen/arch/x86/hvm/vmx/vmx.c	Wed Jul 11 15:47:14 2007 +0100
     5.2 +++ b/xen/arch/x86/hvm/vmx/vmx.c	Wed Jul 11 17:23:09 2007 +0100
     5.3 @@ -907,15 +907,6 @@ static void vmx_ctxt_switch_to(struct vc
     5.4      vmx_restore_dr(v);
     5.5  }
     5.6  
     5.7 -static void stop_vmx(void)
     5.8 -{
     5.9 -    if ( !(read_cr4() & X86_CR4_VMXE) )
    5.10 -        return;
    5.11 -
    5.12 -    __vmxoff();
    5.13 -    clear_in_cr4(X86_CR4_VMXE);
    5.14 -}
    5.15 -
    5.16  static void vmx_store_cpu_guest_regs(
    5.17      struct vcpu *v, struct cpu_user_regs *regs, unsigned long *crs)
    5.18  {
    5.19 @@ -1244,7 +1235,6 @@ static void disable_intercept_for_msr(u3
    5.20  
    5.21  static struct hvm_function_table vmx_function_table = {
    5.22      .name                 = "VMX",
    5.23 -    .disable              = stop_vmx,
    5.24      .domain_initialise    = vmx_domain_initialise,
    5.25      .domain_destroy       = vmx_domain_destroy,
    5.26      .vcpu_initialise      = vmx_vcpu_initialise,
    5.27 @@ -1271,7 +1261,9 @@ static struct hvm_function_table vmx_fun
    5.28      .inject_exception     = vmx_inject_exception,
    5.29      .init_ap_context      = vmx_init_ap_context,
    5.30      .init_hypercall_page  = vmx_init_hypercall_page,
    5.31 -    .event_injection_faulted = vmx_event_injection_faulted
    5.32 +    .event_injection_faulted = vmx_event_injection_faulted,
    5.33 +    .suspend_cpu          = vmx_suspend_cpu,
    5.34 +    .resume_cpu           = vmx_resume_cpu,
    5.35  };
    5.36  
    5.37  int start_vmx(void)

     6.1 --- a/xen/include/asm-x86/hvm/hvm.h	Wed Jul 11 15:47:14 2007 +0100
     6.2 +++ b/xen/include/asm-x86/hvm/hvm.h	Wed Jul 11 17:23:09 2007 +0100
     6.3 @@ -72,11 +72,6 @@ struct hvm_function_table {
     6.4      char *name;
     6.5  
     6.6      /*
     6.7 -     *  Disable HVM functionality
     6.8 -     */
     6.9 -    void (*disable)(void);
    6.10 -
    6.11 -    /*
    6.12       * Initialise/destroy HVM domain/vcpu resources
    6.13       */
    6.14      int  (*domain_initialise)(struct domain *d);
    6.15 @@ -160,6 +155,9 @@ struct hvm_function_table {
    6.16      void (*init_hypercall_page)(struct domain *d, void *hypercall_page);
    6.17  
    6.18      int  (*event_injection_faulted)(struct vcpu *v);
    6.19 +
    6.20 +    void (*suspend_cpu)(void);
    6.21 +    void (*resume_cpu)(void);
    6.22  };
    6.23  
    6.24  extern struct hvm_function_table hvm_funcs;
    6.25 @@ -316,4 +314,16 @@ static inline int hvm_event_injection_fa
    6.26  /* These exceptions must always be intercepted. */
    6.27  #define HVM_TRAP_MASK (1U << TRAP_machine_check)
    6.28  
    6.29 +static inline void hvm_suspend_cpu(void)
    6.30 +{
    6.31 +    if ( hvm_funcs.suspend_cpu )
    6.32 +        hvm_funcs.suspend_cpu();
    6.33 +}
    6.34 +
    6.35 +static inline void hvm_resume_cpu(void)
    6.36 +{
    6.37 +    if ( hvm_funcs.resume_cpu )
    6.38 +        hvm_funcs.resume_cpu();
    6.39 +}
    6.40 +
    6.41  #endif /* __ASM_X86_HVM_HVM_H__ */

     7.1 --- a/xen/include/asm-x86/hvm/vmx/vmcs.h	Wed Jul 11 15:47:14 2007 +0100
     7.2 +++ b/xen/include/asm-x86/hvm/vmx/vmcs.h	Wed Jul 11 17:23:09 2007 +0100
     7.3 @@ -28,6 +28,8 @@ extern int start_vmx(void);
     7.4  extern void vmcs_dump_vcpu(void);
     7.5  extern void vmx_init_vmcs_config(void);
     7.6  extern void setup_vmcs_dump(void);
     7.7 +extern void vmx_suspend_cpu(void);
     7.8 +extern void vmx_resume_cpu(void);
     7.9  
    7.10  struct vmcs_struct {
    7.11      u32 vmcs_revision_id;
    7.12 @@ -59,6 +61,7 @@ struct arch_vmx_struct {
    7.13       *  - Activated on a CPU by VMPTRLD. Deactivated by VMCLEAR.
    7.14       *  - Launched on active CPU by VMLAUNCH when current VMCS.
    7.15       */
    7.16 +    struct list_head     active_list;
    7.17      int                  active_cpu;
    7.18      int                  launched;
    7.19