Re: [PATCH] Move APIC ID validity check into platform APIC code

[Steffen Persvold <sp@numascale.com>]

On 3/14/2012 18:58, Yinghai Lu wrote:
> On Wed, Mar 14, 2012 at 12:17 AM, Daniel J Blueman
[]
>
> can you check if you can update
>
> !cpu_has_x2apic&&  (apic_id>= 0xff)&&  enabled
>
> in arch/x86/kernel/acpi/boot.c::acpi_parse_x2apic()
>
> to use some kind of apic_id_valid()
>
> so you could avoid setting that feature bit.
>
> the checking in SRAT could be removed.
>

Yinghai/Team,

One question (as I don't really know *why* this was added to the 
acpi/srat parsing code). In arch/x86/kernel/smpboot.c the check was 
originally :

!x2apic_mode && apicid >= 255

However in arch/x86/kernel/acpi/boot.c and arch/x86/mm/srat.c these 
tests are used :

!cpu_has_x2apic && (apic_id >= 0xff)

Clearly, "cpu_has_x2apic" and "x2apic_mode" are two different things.

Since we can force "cpu_has_x2apic", when Daniel crafted this patch he 
made the following "default" function :

static inline int default_apic_id_valid(int apicid)
{
	return x2apic_mode || (apicid < 255);
}

which, as you can see, checks against "x2apic_mode".

My question is; Is checking for "x2apic_mode" going to do the trick in 
the arch/x86/kernel/acpi/boot.c::acpi_parse_x2apic() ?

If the answer is yes, the patch is going to be very simple. But we can't 
verify that the code in arch/x86/kernel/acpi/boot.c::acpi_parse_x2apic() 
actually triggers for the case you wanted it to trigger for because then 
it will check against "x2apic_mode" and not "cpu_has_x2apic".

Cheers,
-- 
Steffen Persvold, Chief Architect NumaChip
Numascale AS - www.numascale.com
Tel: +47 92 49 25 54 Skype: spersvold