Re: [PATCH v2 2/5] x86/mce: dump error msg from severities

[Shuai Xue <xueshuai@linux.alibaba.com>]

在 2025/3/2 15:37, Borislav Petkov 写道:
> On Sun, Mar 02, 2025 at 03:14:52PM +0800, Shuai Xue wrote:
>>>>       "mce: Uncorrected hardware memory error in user-access at 3b116c400"
>>
>> It is the current message in kill_me_maybe(), not added by me.
> 
> Doesn't change the fact that it is not really helpful when it comes to logging
> all errors properly.
> 
>    [ Properly means using a structured log format with the tracepoint and not
>      dumping it into dmesg. ]
> 
> And figuring out what hw is failing so that it can be replaced. No one has
> come with a real need for making it better, more useful.
> 
> You're coming with what I think is such a need and I'm trying to explain to
> you what needs to be done. But you want to feed your AI with dmesg and solve
> it this way.
> 
> If you wanna do it right, we can talk. Otherwise, have fun.

I see. So I am just curious why we define `msg` in `severities`?

I perfer to use structured log format with the tracepoint, and we do use it in
production, but it lacks of process context.

AMD folks add error message for panic errors[1] to help debugging
in which the EDAC driver is not able to decode.

For non-fatal errors, is it reasonable to assume that all users are using
tracepoint-based tools like Rasdaemon?

[1]https://lore.kernel.org/all/20220405183212.354606-1-carlos.bilbao@amd.com/

> 
>> 3. We need to identify and implement potential improvements.
>>
>> "mce: Uncorrected hardware memory error in user-access at 3b116c400"
>>
>> is *nothing* but
>>
>> "mce: Action required: data load in error recoverable area of kernel"
>>
>> helps.
> 
> I don't think you've read what I wrote but that's ok. If you think it helps,
> you can keep it in your kernels.
> 

Fine, I could drop patch 1 and 2 in next version.

Thanks.
Shuai