Re: Fwd: [PATCH] x86, MCE, AMD: save IA32_MCi_STATUS before machine_check_poll() resets it

[Aravind Gopalakrishnan <aravind.gopalakrishnan@amd.com>]

On 10/29/2014 10:59 AM, Aravind Gopalakrishnan wrote:
> On 10/22/2014 4:30 AM, Borislav Petkov wrote:
>> Hi Aravind,
>>
>> question: what's the story with MC?_MISC[IntP], is that bit still there?
>> Because I don't see it in my BKDGs here.
>
> Yep, It exists.
> Maybe you are referring to Fam15h M0h BKDG? I think the bit was 
> introduced only from F15h M30h onwards.
> The bit does *not* exist for bank=4, But-
> if (bank ==4)
>   return true;
>
> takes care of that.
>
>> The background of the story is
>>
>> https://lkml.org/lkml/2014/10/7/84
>>
>> There's this thing we did at the time
>>
>> f227d4306cf3 ("x86, MCE, AMD: Make APIC LVT thresholding interrupt 
>> optional")
>>
>> which, AFAICR, is about some F15h versions having a counter but *not*
>> generating a thresholding interrupt. Can you confirm that is still
>> the case and we can have a counter but no interrupt gets generated on
>> overflow?
>>
>
> So yes, moving the assignment inside the if condition should work just 
> fine.
>
> I see the patch on your 'ras-for-3.19' branch does not have this, so 
> I'll make this modification
> to the branch before I test it.
>

Hi Boris,
I have tested the branch with this bit:

  if (b.interrupt_capable) {
             ... ...
             if (mce_threshold_vector != amd_threshold_interrupt)
                     mce_threshold_vector = amd_threshold_interrupt;
     }

and it works fine.

Thanks,
-Aravind.