Comments
Patch
@@ -1733,6 +1733,8 @@ int cpu_x86_register(X86CPU *cpu, const char *cpu_model)
QDict *features = NULL;
char *name = NULL;
+ cpu->env.cpuid_apic_id = env->cpu_index;
+
/* for CPU subclasses should go into cpu_x86_init() before object_new() */
compat_normalize_cpu_model(cpu_model, &name, &features, &error);
if (error_is_set(&error)) {
@@ -2361,8 +2363,6 @@ static void x86_cpu_initfn(Object *obj)
x86_register_cpuid_properties(obj, svm_feature_name);
x86_register_cpuid_properties(obj, cpuid_7_0_ebx_feature_name);
- env->cpuid_apic_id = env->cpu_index;
-
/* init various static tables used in TCG mode */
if (tcg_enabled() && !inited) {
inited = 1;
The problem here is: - The CPU object can't define its APIC ID itself, it has to be defined by the creator of the CPU object - The object instance_init() function can't get extra arguments - I don't want to add a APIC ID argument to x86_cpu_realize(), as it should eventually become a ObjectClass::realize() method So, the only place where the CPU object can get an additional x86-specific initialization argument (the APIC ID) is cpu_x86_register(). This patch just moves the initialization code there, the apic_id argument to cpu_x86_register() will be added later. The cpuid_apic_id field is used only at: - pc_new_cpu(), after the cpu_x86_register() call - kvm_init_vcpu(), that is called from the VCPU thread created by qemu_init_vcpu(), called by x86_cpu_realize() (called after cpu_x86_register()) - helper_cpuid(), called only when the VCPU is already running - kvm_arch_init_vcpu(), that's called by kvm_init_vcpu() So it's safe to initialize it later. Signed-off-by: Eduardo Habkost <ehabkost@redhat.com> --- target-i386/cpu.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)