RE: [PATCH v3 4/6] x86/gsseg: move local_irq_save/restore() into asm_load_gs_index()

[Li, Xin3]

> On October 18, 2022 10:25:31 AM PDT, "Li, Xin3" <xin3.li@xxxxxxxxx> wrote:
> >> >  static inline void native_load_gs_index(unsigned int selector)  {
> >> > -	unsigned long flags;
> >> > -
> >> > -	local_irq_save(flags);
> >> >  	asm_load_gs_index(selector);
> >> > -	local_irq_restore(flags);
> >> >  }
> >>
> >> static inline void native_load_gs_index(unsigned int selector) {
> >> 	unsigned long flags;
> >>
> >>         if (cpu_feature_enabled(LKGS)) {
> >>               native_lkgs(selector);
> >>         } else {
> >> 	      local_irq_save(flags);
> >>               asm_load_gs_index(selector);
> >> 	      local_irq_restore(flags);
> >>        }
> >> }
> >>
> >> For paravirt enabled kernels we want during feature detection:
> >>
> >>         if (cpu_feature_enabled(LKGS)))
> >>         	pv_ops.cpu.load_gs_index = native_lkgs;
> >
> >If we use static_cpu_has in native_load_gs_index
> >       if (static_cpu_has(X86_FEATURE_LKGS)) {
> >               native_lkgs(selector);
> >       }
> >
> >We don't have to change pv_ops.cpu.load_gs_index, right?
> >
> >Thanks!
> >Xin
> >
> >>
> >> No?
> >>
> >> Thanks,
> >>
> >>         tglx
> >
> 
> You don't *have* to, but it would mean a branch to a branch, so it would be
> more efficient. It would strictly be an optimization.

Right, the generated object file shows that static_cpu_has is less efficient than ALTERNATIVE.