Re: [PATCH] x86: remove update_apic from x86_quirks

[Ingo Molnar <mingo@elte.hu>]

* Yinghai Lu <yinghai@kernel.org> wrote:

> Ingo Molnar wrote:
> > except that it doesnt build:
> > 
> > arch/x86/kernel/built-in.o:(.data+0x8c08): undefined reference to `wakeup_secondary_cpu_via_init'
> > arch/x86/kernel/built-in.o:(.data+0x8d68): undefined reference to `wakeup_secondary_cpu_via_init'
> > arch/x86/kernel/built-in.o:(.data+0x8ec8): undefined reference to `wakeup_secondary_cpu_via_init'
> > arch/x86/kernel/built-in.o:(.data+0x9038): undefined reference to `wakeup_secondary_cpu_via_init'
> > 
> 
> .config?

There's no need for a config :)

When introducing a widely used symbol as 
wakeup_secondary_cpu_via_init(), the first thing to check is to 
make sure it's built in widely enough:

$ git grep wakeup_secondary_cpu_via_init

arch/x86/include/asm/apic.h:extern int wakeup_secondary_cpu_via_init(int apicid, unsigned long start_eip);
arch/x86/kernel/apic/apic_flat_64.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/apic_flat_64.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/bigsmp_32.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/es7000_32.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/probe_32.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/summit_32.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/x2apic_cluster.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/apic/x2apic_phys.c:	.wakeup_cpu			= wakeup_secondary_cpu_via_init,
arch/x86/kernel/smpboot.c:wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)

as you can see it's defined in arch/x86/kernel/smpboot.c - 
which, as its name already tells us, is only built on SMP. 

There's other usage though, like 
arch/x86/kernel/apic/probe_32.c, which is built on 
CONFIG_X86_LOCAL_APIC=y, which can be enabled on !SMP as well.

Since on SMP we are not likely to boot other CPUs via 
apic->wakeup_cpu(), i suspect we need a !SMP inline for 
wakeup_secondary_cpu_via_init()?

	Ingo