Re: 2.6.39-rc5-git2 boot crashs

["H. Peter Anvin" <hpa@zytor.com>]

On 05/03/2011 02:45 PM, Linus Torvalds wrote:
> On Tue, May 3, 2011 at 1:50 PM, Ingo Molnar <mingo@elte.hu> wrote:
>>
>> the ELAN .config option influences the following details:
>>
>>  - sets X86_L1_CACHE_SHIFT to 4 (16 bytes) instead of the typical 6 (64 bytes)
>>  - sets X86_ALIGNMENT_16
>>  - sets the -march=i486 compiler flag
> 
> It also does this to the config diff:
> 
>   306,307c328,332
>   < CONFIG_X86_ALIGNMENT_16=y
>   < CONFIG_X86_MINIMUM_CPU_FAMILY=4
>   ---
>   > CONFIG_X86_USE_PPRO_CHECKSUM=y
>   > CONFIG_X86_TSC=y
>   > CONFIG_X86_CMPXCHG64=y
>   > CONFIG_X86_CMOV=y
>   > CONFIG_X86_MINIMUM_CPU_FAMILY=5
> 
> because of all the indirect changes it causes.
> 
> Now, Werner is actually _running_ on an AMD Opteron (or whatever
> family 15 is), I think. And his kernel is SMP-enabled. And that whole
> thin looks really really iffy.
> 
> How/why do we even allow that combination of SMP and (for example)
> arch/x86/lib/atomic64_386_32.S to be picked?
> 
> I don't think he actually runs SMP, but the fact that we even allow
> that combination looks really odd/iffy. Am I missing something?
> 

We would end up in those paths before alternatives are run, but
alternatives should be run before we start the second processor.

	-hpa