On Tue, 2009-03-03 at 12:34 -0800, Cynthia McGuire wrote: > Tom Pothier wrote: > > Hi Cindi, > > > > This is using the SMBIOS and/or ACPI for fru topology information. The > > "PRI like structure" only becomes intermediate data that then can be > > walked by a common back end walker. All FRU information (serial number, > > part number, etc...) is retrieved from (mostly) SMBIOS and then placed > > into a PRI like structure for the topology to be generated. > > > > There is no "firmware" generated PRI like entity, and we're not > > proposing one. We're generating the x86 PRI only so we could reuse the > > sun4pi walker (to generate the FMA topology). Otherwise we'd have to > > come up with an entirely x86 specific data format and walker... > > > > Does this make more sense? > > No. libtopo should be getting its SMBIOS information directly from > libsmbios. I see no motivation in your spec for an intermediate layer.
Several months ago when the project was first conceived, there was motivation for reusing and leveraging code from sun4v. As the project scoping has progressed, the amount of leverage has decreased. I had a quick chat with Tom just now, and at this point there's no driving need for an interim structure between libsmbios and a topo enumerator. The implementation approach aside, is there a general disagreement with the project concept - enumerate from an industry standard source? > What limitations in libsmbios or libtopo are forcing you to model this > information using the PRI. None to my knowledge. I do believe that the scoping effort determined some extensions to libsmbios (for contained objects, IIRC) > >> While I'm happy to see some convergence between x86 and SPARC, this > >> project moves in the wrong direction. If I understand correctly, you > >> are adopting the SPARC PRI for modeling the FMA topology on x86. This > >> has been discussed before and it was decided that a better approach > >> for all Sun platforms is to move toward adopting SMBIOS. SMBIOS (plus > >> whatever you glean from ACPI) then becomes the source of FRU and other > >> FMA topology information for things like libtopo and fmd(1M). With respect to SPARC, the problem of enumeration has already been addressed for sun4v. The sun4v PRI provides all the details needed to accurately enumerate, and we source from there. If and when SPARC platforms begin providing SMBIOS information, the enumeration strategy on sun4v can be revisited. > >> Further, Sun needs to draft a SMBIOS standard for platform vendors > >> wishing to participate in fault management and diagnosis for their > >> platforms. Included in the standard should be the required records for > >> the Sun FMA. This is definitely in scope. Drafts have already been shared with some vendors, and thus far there have been When the project goes on opensolaris.org, I'd expect the current draft of the Sun FMA SMBIOS requirements will be on the project page. Thanks, --scott --http://blogs.sun.com/sdaven > If a vendor conforms (including Sun SPARC), diagnosis > >> and FRU location and replacement is then guaranteed. > >> > >> Cindi > >> > > _______________________________________________ fm-discuss mailing list fm-discuss@opensolaris.org
