Re: [PATCH 7/8] Implement smp_processor_id() with the PDA.

[Jeremy Fitzhardinge <jeremy@goop.org>]

Ian Campbell wrote:
> This doesn't compile for me if CONFIG_SMP=n
>   

Ah, good point.

>           LD      .tmp_vmlinux1
>         arch/i386/kernel/built-in.o: In function `cpu_init':
>         (.init.text+0x1eda): undefined reference to `early_smp_processor_id'
>         arch/i386/kernel/built-in.o: In function `cpu_init':
>         (.init.text+0x1f11): undefined reference to `early_smp_processor_id'
>         
> smp_processor_id() is defined for !SMP in include/linux/smp.h, I don't
> know if it would be appropriate to add early_smp_processor_id() there
> since it seems i386 specific. asm/smp.h isn't included by linux/smp.h
> when !SMP but you could add an explicit include to common.c I suppose.
>   

I'll have a look.

I think my preferred solution would be to get rid of all the early* 
stuff, and try to arrange to have the PDA set up before C code gets 
run.  For the boot CPU, it really could be done statically (I'm not 
quite sure why the boot CPU's GDT is allocated, given that it already 
has a static one; I think this might have been a Xen-related change?).  
The secondary CPUs could have their GDT+PDA completely allocated and 
initialized in advance, making secondary CPU PDA setup just a matter of 
doing lgdt and setting %gs in head.S, even before hitting C code.

    J