Re: [HW PROBLEM] Intel I7 MCE. Erratum or not?

["Giangiacomo Mariotti" <gg.mariotti@gmail.com>]

On Sun, Dec 7, 2008 at 4:25 AM, Robert Hancock <hancockr@shaw.ca> wrote:
> Giangiacomo Mariotti wrote:
>>
>> On Sat, Dec 6, 2008 at 10:47 PM, Robert Hancock <hancockr@shaw.ca> wrote:
>>>
>>> Giangiacomo Mariotti wrote:
>>>>
>>>> On Sat, Dec 6, 2008 at 9:58 PM, Robert Hancock <hancockr@shaw.ca> wrote:
>>>>>
>>>>> Giangiacomo Mariotti wrote:
>>>>>>
>>>>>> Hi everyone,
>>>>>> Mcelog just logged on my new Intel I7 920 (on Linux 2.6.27.8) this :
>>>>>> MCE 0
>>>>>> HARDWARE ERROR. This is *NOT* a software problem!
>>>>>> Please contact your hardware vendor
>>>>>> CPU 0 BANK 6 MISC 202d ADDR ffeef740
>>>>>> MCG status:
>>>>>> MCi status:
>>>>>> Error overflow
>>>>>> Uncorrected error
>>>>>> MCi_MISC register valid
>>>>>> MCi_ADDR register valid
>>>>>> Processor context corrupt
>>>>>> MCA: Generic CACHE Level-2 Data-Write Error
>>>>>> STATUS ee0000000100014a MCGSTATUS 0
>>>>>>
>>>>>> I'm reporting this here, because I found in the Intel I7 Technical
>>>>>> Specification November 2008 update that something which seems very
>>>>>> similar is in fact an erratum. So my question is : Is there any way
>>>>>> for me to verify that my problem is due to one of those errata,instead
>>>>>> of a broken hardware(if we don't want to consider all those errata as
>>>>>> broken hardware)? I'm also reporting this because I thought it may be
>>>>>> useful to signal that(if actually due to those errata) these problems
>>>>>> actually occur, so it may be useful to find workarounds in the kernel
>>>>>> to not scare to death poor Linux users!
>>>>>
>>>>> Which erratum are you talking about? I don't see one in that document
>>>>> that
>>>>> would match this case..
>>>>>
>>>> Well, the first one seems very similar, even if it talks about a dtlb
>>>> error instead of cache error. But sure,being similar doesn't mean too
>>>> much. Number 52 seems similar too. I guess I should just give up and
>>>> admit that my hardware is broken!
>>>>
>>> The first one is just indicating that if a DTLB error occurs the overflow
>>> bit may be set incorrectly. It's not a false error though. The AAJ52
>>> erratum
>>> would only occur immediately after powerup or wake from sleep states.
>>>
>> The mce actually got logged once immediately after powerup and never
>> more. Is that reasonable? A cache error which happens just once after
>> boot?
>
> The erratum refers to an internal parity error, not an L2 cache write error.
>
> If it only happened once then who knows, could be a cosmic ray or
> something.. but if it happens again it sounds like you likely have a bad
> CPU.
>
It happens once every time I boot kernel 2.6.27.8, right after the
boot. If I boot kernel 2.6.26 in debian/unstable(based on 2.6.26.8)
though, I never get the mce log message. Also now I got another really
bad problem with 2.6.27.8 which corrupted most of my partitions. I'm
gonna post about it now.