Re: [PATCH 4/9 UPDATED-1] x86: Initialize 32bit logical apicid mapping early during boot

[Ingo Molnar <mingo@elte.hu>]

* Tejun Heo <tj@kernel.org> wrote:

> On x86_32, non-standard logical apicid mapping can be used by
> different NUMA setups and the mapping is queried while bringing up
> each CPU using apic->cpu_to_logical_apicid() to build
> cpu_2_logical_apicid[] array.  The logical apicid is then used to
> deliver IPIs and determine NUMA configuration.
> 
> Unfortunately, initializing at SMP bring up is too late for percpu
> setup making static percpu variables setup w/o considering NUMA.  This
> also is different from how x86_64 is configured making the code
> difficult to follow and maintain.
> 
> This patch updates logical apicid mapping handling such that,
> 
> * early_percpu variable x86_cpu_to_logical_apicid replaces
>   cpu_2_logical_apicid[].
> 
> * apic->cpu_to_logical_apicid() is called once during get_smp_config()
>   and the output is recorded in x86_cpu_to_logical_apicid.
> 
> * apic->cpu_to_logical_apicid() is allowed to return BAD_APICID if it
>   can't determine the value that early during boot.  In this case, the
>   mapping will be initialized during SMP bring up by reading APIC LDR
>   as before.
> 
> - Brian Gerst spotted that setup_per_cpu_areas() was not copying the
>   early x86_cpu_to_logical_apicid to the permanent percpu area and
>   es7000_32 is using per_cpu() instead of early_per_cpu(), which in
>   itself is not incorrect as they're never used before setup_per_cpu()
>   but still confusing.  Both updated.
> 
> - Using local variable @cpu to cache smp_processor_id() in
>   setup_local_APIC() separated out into a separate patch as suggested
>   by Yinghai Lu.

This patch is still _WAY_ too large.

Also, these:

> +#ifdef CONFIG_X86_32
> +#endif

> +#ifdef CONFIG_X86_32
> +#endif

> +#ifdef CONFIG_X86_32
> +#endif

> +#ifdef CONFIG_X86_32
> +#endif

Are rather ugly.

	Ingo