On Fri, 30 Jun 2017, Dou Liyang wrote: > static inline int apic_force_enable(unsigned long addr) > diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c > index 0601054..9bf7e95 100644 > --- a/arch/x86/kernel/apic/apic.c > +++ b/arch/x86/kernel/apic/apic.c > @@ -1198,6 +1198,10 @@ static int __init apic_intr_mode_select(int *upmode) > } > #endif > > +#ifdef CONFIG_UP_LATE_INIT > + *upmode = true; > +#endif
This is really wrong. The upmode decision, which is required for calling apic_bsp_setup() should not happen here, really. As I told you in the previous patch, use the return code and then you can make further decisions in apic_intr_mode_init(). And you do it there w/o any ifdeffery: static void apic_intr_mode_init(void) { bool upmode = IS_ENABLED(CONFIG_UP_LATE_INIT); switch (....) { case XXXX: upmode = true; .... } apic_bsp_setup(upmode); } Thanks, tglx