Re: [PATCH] [3/4] x86: MCE: Improve mce_get_rip

[Hidetoshi Seto]

Huang Ying wrote:
> On Fri, 2009-04-24 at 15:28 +0800, Hidetoshi Seto wrote:
>> Huang Ying wrote:
>>> On Fri, 2009-04-24 at 14:16 +0800, Hidetoshi Seto wrote:
>>>> One question is: if (RIPV,EIPV) = (0,0), then is the IP on the stack
>>>> really invalid value, or is it still point IP when MCE is generated?
>>>> I suppose it is not invalid.  If a processor encounters MCE and if it
>>>> is not sure what happened, then it will store the IP on the stack,
>>>> indicating neither of flags.
>>>>
>>>> If this supposition is correct, the best way is pick the value on
>>>> the stack unconditionally, and record valid flags together.
>>> According to spec, the IP on stack can be not related to MCE if
>>> (RIPV,EIPV) = (0,0). So it is meaningless to report them. If you report
>>> them unconditionally, you just push the logic to user space or
>>> administrator.
>> Sorry, I could not find good page in the spec (Intel64 and IA-32 ASDM)...
>> Could you point one?
> 
> 14.3.1.2 IA32_MCG_STATUS MSR
> * EIPV

Quote:
 "EIPV (error IP valid) flag, bit 1 ― Indicates (when set) that the
  instruction pointed to by the instruction pointer pushed onto the
  stack when the machine-check exception is generated is directly
  associated with the error. When this flag is cleared, the instruction
  pointed to may not be associated with the error."

My understanding is:
 If EIPV is 1:
    IP value on the stack is one pushed when the MCE is generated,
    and the IP is associated with the error.
 If EIPV is 0:
    IP value on the stack is one pushed when the MCE is generated,
    but the IP is not associated with the error.

So I repeat my question again:
You stated in the description of this patch:
   "mce_get_rip() is used to get IP when MCE is generated, ..."
Is this right?

If right, I think EIPV is not matter.
If not, please rewrite the description.

>> I believe that the IP with (RIPV,EIPV) = (1,0) is "not associated with the
>> error" too, so is it meaningless to report the IP?
>> If you think so then correct fix is replacing RIPV check by EIPV check.
> 
> In theory, that is possible (not associated), but I think in practical,
> IP with (RIPV,EIPV) = (1,0) is still meaningful as Andi said.

Then, why IP with (0,0) is meaningless?
Why not use it with the !INEXACT! marker?

>> From another point of view, the reported IP will be one of followings:
>>   - IP that associated with error (= related to MCE)
>>   - IP that the interrupted program can restart from
>>   - IP that when MCE is generated
>> Are there no way to distinguish them in user space?
> 
> I think you just push same logic to user space.

No, I just want a logical explanation.

It seems we already can provide records with "inexact" value.
Why not expand such cases?


Thanks,
H.Seto

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/