Re: [PATCH] Use x2apic_supported() in the default_apic_id_valid() function.

[Steffen Persvold <sp@numascale.com>]

On 3/16/2012 00:04, Suresh Siddha wrote:
> On Thu, 2012-03-15 at 23:34 +0100, Steffen Persvold wrote:
>> Is my understanding of your suggestion correct that in
>> x2apic_phys/cluster.c we add the following apic_id_valid() function :
>>
>> static int x2apic_apic_id_valid(int apicid)
>> {
>>    	return x2apic_mode || (apicid<  255);
>> }
>
> Steffen, We can have something like:
>
> static int x2apic_apic_id_valid(int apicid)
> {
>     	return 1;
> }
>
> and
>
> static int xapic_apic_id_valid(int apicid)
> {
>     	return apicid<  255;
> }
>
> If we have selected x2apic driver, then we know we are already in x2apic
> mode. And also x2apic_uv_x need to use the x2apic version above.
>
>> Considering that this function (apic->apic_id_valid()) is called already
>> in the acpi/boot.c::acpi_parse_x2apic() function is it sufficient enough
>> to test for x2apic_mode ? Yinghai indicated that x2apic_mode was not set
>> at this point, thus it was testing cpu_has_x2apic instead ?
>
> If the bios has handed over to us in x2apic mode (or if it is a numachip
> platform), then by this point apic driver is already set to the
> corresponding x2apic/numachip driver etc. so we should be fine.
>
> When we are in xapic mode, typically there should be no x2apic MADT
> entries. And even if there are any (bios not following x2apic spec), the
> above xapic_apic_id_valid() check will consider only those x2apic MADT
> entries whose id's are less than 255. xapic mode can go into x2apic mode
> later but that flow is not supposed to bring up any cpu with apic id>
> 255. So parsing only entries with apic id<  255 here should be fine.
>
> Hope this clarifies.
>

Yes, this is now my understanding aswell. Thank you.

I will resend a reviced patch shortly.

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