On Fri, Jan 26, 2024 at 08:49:03PM +0000, Luck, Tony wrote:
> Every patch that adds new code or data structures adds to the kernel
> memory footprint. Each should be considered on its merits. The basic
> question being:
>
> "Is the new functionality worth the cost?"
>
> Where does it end? It would end if Linus declared:
>
> "Linux is now complete. Stop sending patches".
>
> I.e. it is never going to end.
No, it's not that - it is the merit thing which determines.
> 1) PPIN
> Cost = 8 bytes.
> Benefit: Emdeds a system identifier into the trace record so there
> can be no ambiguity about which machine generated this error.
> Also definitively indicates which socket on a multi-socket system.
>
> 2) MICROCODE
> Cost = 4 bytes
> Benefit: Certainty about the microcode version active on the core
> at the time the error was detected.
>
> RAS = Reliability, Availability, Serviceability
>
> These changes fall into the serviceability bucket. They make it
> easier to diagnose what went wrong.
So does dmesg. Let's add it to the tracepoint...
But no, that's not the right question to ask.
It is rather: which bits of information are very relevant to an error
record and which are transient enough so that they cannot be gathered
from a system by other means or only gathered in a difficult way, and
should be part of that record.
The PPIN is not transient but you have to go map ->extcpu to the PPIN so
adding it to the tracepoint is purely a convenience thing. More or less.
The microcode revision thing I still don't buy but it is already there
so whateva...
So we'd need a rule hammered out and put there in a prominent place so
that it is clear what goes into struct mce and what not.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
