Re: [PATCH] [v2] x86: xen: add missing prototypes

[Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>]

On 5/23/23 4:37 PM, Arnd Bergmann wrote:
> On Sat, May 20, 2023, at 00:24, Boris Ostrovsky wrote:
> > On 5/19/23 5:28 AM, Arnd Bergmann wrote:
> >
> > > diff --git a/arch/x86/xen/smp.h b/arch/x86/xen/smp.h
> > > index 22fb982ff971..81a7821dd07f 100644
> > > --- a/arch/x86/xen/smp.h
> > > +++ b/arch/x86/xen/smp.h
> > > @@ -1,7 +1,11 @@
> > >    /* SPDX-License-Identifier: GPL-2.0 */
> > >    #ifndef _XEN_SMP_H
> > >    
> > > +void asm_cpu_bringup_and_idle(void);
> > > +asmlinkage void cpu_bringup_and_idle(void);
> > These can go under CONFIG_SMP
> Not sure if there is much point for the second one, since
> it's only called from assembler, so the #else path is
> never seen, but I can do that for consistency if you
> like.
>
> I generally prefer to avoid the extra #if checks
> when there is no strict need for an empty stub.
Do we need the empty stubs? Neither of these should be referenced if !SMP (or 
more precisely if !CONFIG_XEN_PV_SMP)

> I guess you also want me to add the empty stubs for the
> other functions that only have a bit in #ifdef but not
> #else then?
>
> > > +void xen_force_evtchn_callback(void);
> > These ...
> >
> > > +pteval_t xen_pte_val(pte_t pte);
> > > +pgdval_t xen_pgd_val(pgd_t pgd);
> > > +pte_t xen_make_pte(pteval_t pte);
> > > +pgd_t xen_make_pgd(pgdval_t pgd);
> > > +pmdval_t xen_pmd_val(pmd_t pmd);
> > > +pmd_t xen_make_pmd(pmdval_t pmd);
> > > +pudval_t xen_pud_val(pud_t pud);
> > > +pud_t xen_make_pud(pudval_t pud);
> > > +p4dval_t xen_p4d_val(p4d_t p4d);
> > > +p4d_t xen_make_p4d(p4dval_t p4d);
> > > +pte_t xen_make_pte_init(pteval_t pte);
> > > +void xen_start_kernel(struct start_info *si);
> >
> > ... should go under '#ifdef CONFIG_XEN_PV'
> What should the stubs do here? I guess just return the
> unmodified value?

Same here -- these should only be referenced if CONFIG_XEN_PV is defined which 
is why I suggested moving them under ifdef.


-boris