Re: [RFC PATCH 00/11] soc: mediatek: add a DRAM ECC error handling driver

[Jonathan Cameron <jonathan.cameron@oss.qualcomm.com>]

On Wed, 4 Nov 2026 20:53:21 -0800
Borislav Petkov <bp@alien8.de> wrote:

> On Thu, Nov 05, 2026 at 04:19:27AM +0000, Santino Wang (王俊傑) wrote:
> > Which existing interfaces would you recommend for obtaining retirement
> > status from each memory owner? With that information available, can items
> > 5–7 be maintained in userspace, or would they require support from an MTK
> > driver?  
> 
> mm/memory-failure.c
> 
> Look how the others use it.
> 
> What you write above sounds like you need an EDAC driver which collects and
> reports errors. Look at examples in drivers/edac/ and there's some
> documentation in Documentation/edac/. Perhaps ask AI too.
> 
> If the stuff you need is not there, you can extend the existing interfaces by
> justifying the extensions properly.
> 
> But what you have - a completely separate RAS implementation, homegrown
> somewhere else in the kernel doesn't fly. I'm actually wondering how you even
> managed to do this driver without *finding* all the RAS/EDAC glue we have in
> the kernel. Googling for it would've told you that there's something...
> 

I would add to this that if the standards don't work for you, consider changing
them.  These days it isn't that hard to get ACPI changed and they have a nice
code first path so discussion can take place on kernel mailing lists to refine
the idea before going to the standards org.

There is some overlap with things we will probably be proposing for ACPI in
the next few weeks. I'll make sure to +CC everyone in this thread on that
discussion.  Inputs form Mediatek (and ideally Borislav, Tony etc!)
should help refine that and avoid pitfalls that have made previous RAS ACPI
stuff less than ideal.

Jonathan

> HTH.
>