[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [RFC PATCH v1 01/01] PCIe DOE for PCIe and CXL 2.0
From: |
Ben Widawsky |
Subject: |
Re: [RFC PATCH v1 01/01] PCIe DOE for PCIe and CXL 2.0 |
Date: |
Mon, 8 Feb 2021 09:51:53 -0800 |
On 21-02-08 10:55:51, Jonathan Cameron wrote:
> ...
>
> >
> > >
> > >>
> > >>>>
> > >>>> Just like you we feel what's most important is to have DOE supported
> > >>>> so that
> > >>>> UEFI and Linux kernel and drivers can progress. We're also
> > >>>> contributing to
> > >>>> writing compliance tests for the CXL Compliance Software Development
> > >>>> WG.
> > >>>
> > >>> Great.
> > >>
> > >> Is anyone doing the kernel enabling for it?
> > >
> > > Planning to look at this but plenty of other things on my todo list if
> > > someone
> > > else gets to it first.
> > >
> > > Generic DOE support should be straight forward (the infrastructure).
> > > Parsing CDAT also straight forward.
> > > Doing something with the results is hard unless we just provide an
> > > interface for
> > > userspace to query them for a given device - or dump the table
> > > (I think we do want to be able to that).
> > >
> > > What I'm really not sure on is how to handle NUMA domains that are
> > > created late
> > > in the kernel boot sequence. The ACPI flow is set up with the assumption
> > > that we can get them from SRAT very early in boot. Need to figure out how
> > > to
> > > work around that. (e.g. preallocate a bunch of spare nodes for example
> > > though that's
> > > ugly). Note IIRC the kernel doesn't do runtime update of any of the ACPI
> > > performance parameters yet (_SLI, _HMA) so there probably isn't any
> > > infrastructure
> > > that we can reuse.
> > >
> > > There is also the firmware based enumeration and description option (OS
> > > not necessarily
> > > aware of CXL) in which this is all up to EDK2 and the kernel gets it all
> > > presented
> > > as standard tables.
> >
> > Do we know who’s on this as part of the EDK2 development? It would be
> > great if they could
> > address the SRAT/HMAT generation from reading CDAT. EDK2 does address CXL
> > 1.1 now.
>
> No idea who, if anyone, is looking at this currently. Perhaps ask on the
> EDK2 list?
>
> Jonathan
>
I did ping the folks at #edk2 in OFTC a few months back and got basically no
response. Mailing list might be best though.
> >
> > >
> > > As you can perhaps tell from my half done reviews, this week disappeared
> > > in
> > > other things so bit of catch up for me to do next week.
> > >
> > > Thanks,
> > >
> > > Joanthan
> > >
> ...
>