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

[Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>]

Andi Kleen wrote:
> Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com> writes:
> 
>> Andi Kleen wrote:
>>> From: Huang Ying <ying.huang@intel.com>
>>>
>>> Return rip/cs if MCG_STATUS_EIPV is set in mce_get_rip(). Remain m->cs
>>> if RIP is read from rip_msr.
>> It means we use "Error IP" as "Return IP" if RIPV=0 but EIPV=1 ...?
>> Sounds strange.
> 
> It's not return IP, but "reported IP" in this case.

Wait, I'm ashamed to say, it seems we missed the name of instruction
pointer register: The 64bit one is RIP, and the 32bit one is EIP.

Anyway we have proved a major point - It is confusing expression.

>> I guess it would make much sense if we stop mixing RIP and EIP and rename
>> the mce_get_rip() to mce_get_eip(), and the rip_msr to eip_msr too.
> 
> Ok fair enough.  I admit the code was always a bit dubious.
> 
>> And then it would be acceptable if we print RIP with "!INEXACT!" annotation
>> instead of printing precise EIP in case of RIPV=0 but EIPV=1.
> 
> Please send a patch to do all that.

I will.

A trivial question is if you unified 32/64bit mce codes, would you
like to print only "IP %02x:<%016Lx>", or "RIP ..." and "EIP ..." ?


Thanks,
H.Seto