[MODERATED] Re: [patch V2 02/12] x86/smp: Provide topology_is_primary_thread()

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 12/06/18 09:05, speck for Thomas Gleixner wrote:
> On Mon, 11 Jun 2018, speck for Konrad Rzeszutek Wilk wrote:
>> On Wed, Jun 06, 2018 at 09:27:16PM +0200, speck for Thomas Gleixner wrote:
>> Anyhow if we did have four siblings, the mask would end up with 11b
>> (see below):
>>
>> smp_num_siblings	fls	-1	1<<	 -1
>> 	1 (001b)	1	0	1	0
>> 	2 (010b)	2	1	2	01b
>> 	4 (100b)	3	2	4	11b
> That's what I intended :)
>
>>> +	return !(apicid & mask);
>> So with an normal 2-sibling, say APICID 0x76 and 0x77 with mask 1b:
>>
>> 	0x76	01110110b & 01 -> !0 -> return 1
>> 	0x77	01110111b & 01 -> !1 -> return 0
>>
>> that looks good.
>>
>> If this is a 4 thread. APICID 0x4, 0x5, and 0x6
>>
>> 	0x4	0100 & 11b -> 00 -> !0 -> return 1
>> 	0x5	0101 & 11b -> 01 -> !1 -> return 0 
>> 	0x6	0110 & 11b -> 10 -> !1111..1101 -> return 0xffff..which cast to bool is 'true' [BAD]
>>
>> That is the 3rd sibling would be incorrectly identified as primary thread.
>>
>> I think? I would welcome somebody checking my logic over, but then
>> perhaps this is non-issue-b/c-we-don-do more than 2 so go away.
> You are somehow confusing '!' and '~'

The original logic looks correct to me, and matches my equivalent in Xen.

~Andrew