Re: [PATCH v15 00/35] Introduce SCMI Telemetry support
From: Jonathan Cameron
Date: Wed Nov 18 2026 - 12:32:31 EST
On Wed, 18 Nov 2026 09:33:44 +0000
Sudeep Holla <sudeep.holla@xxxxxxxxxx> wrote:
> On Tue, Nov 17, 2026 at 05:07:09PM -0800, Jonathan Cameron wrote:
> > On Mon, 9 Nov 2026 09:11:46 +0000
> > Cristian Marussi <cristian.marussi@xxxxxxx> wrote:
> >
> > > Hi all,
> > >
> > > [TLDR Summary]
> > > [V15 highlights]
> > > - general cleanup and hardening driven by reviews
> > > - improved TDCF access routines to minimize lost samples under load
> > > - completed ABI docs
> > >
> > > [V16 TODO]
> > > - address any remaining V15 outstanding Sashiko/reviewers issues
> > >
> > > The upcoming SCMI v4.0 specification [0] introduces a new SCMI protocol
> > > dedicated to System Telemetry.
> >
> > This is one from the family of issues you find when reviewing ones own
> > (well LLM generated) emulation, realizing it doesn't match the spec,
> > and yet works.
> >
> > This works fine without the required SCMI protocol property in ACPI.
> > 2.2.2 of ACPI for System Control and Management Interface 1.0
> > Platform Design Document. 111115A 1.0
> >
> > Now, in here is an informative statement...
> >
> > Protocols not listed using a property may not be enumerate by the operating system.
> >
> > But they are...
> >
>
> Correct, I intentionally left that out. I do have a patch that restricts it
> strictly to the listed property. However, I held off on pushing it because of
> an issue we saw on an internal test platform.
>
> On that platform, the firmware advertised the protocol to the OSPM via the
> base protocol, even though it wasn't functional at all. Once they realized
> this, they simply dropped the entry from the ACPI SCMI device as a workaround
> instead of properly fixing the base protocol discovery. While understandable,
> relying on that workaround delays proper firmware fixes.
>
> That is why I went with the non-mandatory "may not be" wording. I am happy to
> merge that restriction patch if a real production platform actually needs it,
> but until then, leaving it open helps catch and fix broken firmware
> implementations 😄.
You can't merge it safely as at a later date because chances are high
someone is relying on the current behavior and not presenting that
(currently pointless) package in ACPI for some protocol or other.
I'm fine with a permanent decision not to check this but I don't see it
as a thing that can ever change (except maybe now whilst there are not
that many platforms out there using it!)
Jonathan
>
> > So, obviously we can ignore informative spec lines but that is a pretty strong
> > sentence. I can't immediately think of what would go wrong if we ignored it
> > - maybe we'd not pick an exclusive PCC and fallback to a common one the firmware
> > doesn't want us to use? But if it doesn't want us to use it why make Telemetry
> > discoverable on it.
> >
>
> Indeed, falling back to the common channel lets us uncover any underlying
> issues in the firmware implementation. That is the exact scenario we hit on
> the test platform, which convinced me to hold back the patch.
>
> > Should that be wired through from the ACPI parsing code?
> >
>
>
> As mentioned, we certainly can, though I'm not sure we should. The "may not
> be" phrasing simply leaves the choice up to the OSPM rather than mandating a
> strict restriction.
>
> > Anyhow, not something I care that much about but thought I'd call it out
> > as an oddity.
> >
>
> Thanks and well spotted. Much appreciated as it brought the information out
> which otherwise would have just got lost.
>