Re: [PATCH 1/3] efi/cper, cxl: Decode CXL Component Events General Media Event Record

[Smita Koralahalli]

Hi Dan,

On 10/12/2023 4:26 PM, Dan Williams wrote:
> Smita Koralahalli wrote:
> > Add support for decoding CXL Component Events General Media Event Record
> > as defined in CXL rev 3.0 section 8.2.9.2.1.1.
> >
> > All the event records as defined in Event Record Identifier field of the
> > Common Event Record Format in CXL rev 3.0 section 8.2.9.2.1 follow the
> > CPER format for representing the hardware errors if reported by a
> > platform.
> >
> > According to the CPER format, each event record including the General
> > Media is logged as a CXL Component Event as defined in UEFI 2.10
> > Section N.2.14 and is identified by a UUID as defined by Event Record
> > Identifier field in Common Event Record Format of CXL rev 3.0 section
> > 8.2.9.2.1. CXL Component Event Log field in Component Events Section
> > corresponds to the component/event specified by the section type UUID.
> >
> > Add support for decoding CXL Component Events as defined in UEFI 2.10
> > Section N.2.14 and decoding Common Event Record as defined in CXL rev 3.0
> > section 8.2.9.2.1.
>
> So I think this is missing the rationale for the code duplication.
>
> Specifically, who is the consumer of this parsing relative to who is the
> consumer of the event records emitted by the CXL subsystem. Given the
> CXL subsystem event parsing already exists, and unlike this
> implementation can support DPA-to-HPA translation, why should Linux
> carry 2 emitters for what is effectively the same data?
>
> What I would like to see is the CPER code post these record payloads on
> a notifier chain for the CXL subsystem to parse, annotate with extra CXL
> subsystem information, and emit from one control point.
>
> Otherwise if folks would like to see this printk() version of the error
> records then they also need to answer why the CXL subsystem should not
> also emit decoder error logs to printk?

I'm fine with your suggestion. I don't have any strong argument which 
proves the other way round. This was just a starting point in adding 
support for Component error logging in FW-First. I will take a look at 
Ira's patches which adds notifier chain support.

Thanks,
Smita