Re: trace, RAS: New eMCA trace event interface

[Mauro Carvalho Chehab <m.chehab@samsung.com>]

Em Mon, 10 Mar 2014 11:55:24 -0700
Tony Luck <tony.luck@gmail.com> escreveu:

> On Fri, Mar 7, 2014 at 1:10 AM, Mauro Carvalho Chehab
> <m.chehab@samsung.com> wrote:
> > As I just pointed, we'll also need some code at the rasdaemon that would
> > associate the CPER error location data into the corresponding DIMM label.
> 
> In most cases the kernel will provide a label (that it got from
> SMBIOS). I assume
> this is for the case that the SMBIOS table is bad?

Well, except for reference boards, I'm yet to see such info
properly filled at the SMBIOS table that matches the names
printed at the motherboard.

> I haven't looked too hard at how EDAC does this - the "--register-labels" option
> to edac-ctl?

Right now, it uses the EDAC sysfs nodes to store the labels at
the Kernel. The advantage is that a fatal error dmesg will also
print the proper labels.

Moving it to userspace will only solve for non-fatal errors where
the rasdaemon will handle the traces.

> Since we don't have a sysfs interface for extlog - perhaps any
> translation can be handled entirely in rasdaemon(8).

That could be done, but some additional logic will be needed to
teach the rasdaemon that the information should be stored on some
database. The rasdaemon would also need to detect at runtime if
the stored data matches the machine where it is running.

Regards,
Mauro