[PATCH v8 1/2] dt-bindings: arm: Add OP-TEE transport for SCMI

[Etienne Carriere]

Hello Ahmad,

On Tue, 8 Mar 2022 at 10:51, Ahmad Fatoum <a.fatoum at pengutronix.de> wrote:
>
> Hello Sudeep,
>
> On 01.03.22 16:12, Sudeep Holla wrote:
> >
> > Hi Ahmad,
> >
> > On Mon, Feb 28, 2022 at 05:01:39PM +0100, Ahmad Fatoum wrote:
> >> Hello Etienne,
> >>
> >> On 28.10.21 16:00, Etienne Carriere wrote:
> >>> Introduce compatible "linaro,scmi-optee" for SCMI transport channel
> >>> based on an OP-TEE service invocation. The compatible mandates a
> >>> channel ID defined with property "linaro,optee-channel-id".
> >>
> >
> > Not sure if Etienne's reply addressed your queries/concerns correctly.
> > I thought I will add my view anyways.
> >
> >> I just found this thread via the compatible in the STM32MP131 patch set:
> >> https://lore.kernel.org/all/20220225133137.813919-1-gabriel.fernandez@foss.st.com/
> >>
> >> Linux doesn't care whether PSCI is provided by TF-A, OP-TEE or something
> >> else, so there is just the arm,psci* compatible.
> >>
> >
> > Correct, the interface to the kernel is fixed and hence we must be able
> > to manage with the standard and fixed sole set of bindings for the same.
> >
> >> What's different about SCMI that this is not possible? Why couldn't the
> >> existing binding and driver be used to communicate with OP-TEE as secure
> >> monitor as well?
> >>
> >
> > However with SCMI, the spec concentrates and standardises all the aspects
> > of the protocol used for the communication while it allows the transport
> > used for such a communication to be implementation specific. It does
> > address some standard transports like mailbox and PCC(ACPI). However,
> > because of the flexibility and also depending on the hardware(or VM),
> > different transports have been added to the list. SMC/HVC was the one,
> > followed by the virtio and OPTEE. While I agree SMC/HVC and OPTEE seem
> > to have lot of common and may have avoided separate bindings.
> >
> > However the FIDs for SMC/HVC is vendor defined(the spec doesn't cover this
> > and hence we utilised/exploited DT). Some vendors wanted interrupt support
> > too which got added. OPTEE eliminates the need for FID and can also provide
> > dynamic shared memory info. In short, it does differ in a way that the driver
> > needs to understand the difference and act differently with each of the
> > unique transports defined in the binding.
> >
> > Hope that explains and addresses your concern.
>
> Thanks for the elaborate answer. I see now why it's beneficial to have
> an OP-TEE transport in general. I don't yet see the benefit to use it
> in the STM32MP13x instead of SMCs like with STM32MP15x, but that a discussion
> that I need to have in the aforementioned thread.

Some SCMI operations in OP-TEE need to execute in a threaded context
(preemptible, ...).
There is no SMC function ID defined for an SCMI thread entry in
OP-TEE. We rather use standard invocation of a TEE service: opening a
session and invoking commands.
Invoked commands are executed in an OP-TEE native threaded context.
The service accessed is referred to as the OP-TEE SCMI PTA.

As for STM32MP15x, one willing to extend resources assigned to secure
world may also need to move mp15 SCMI from SMC transport to optee
transport.

Regards,
Etienne

>
> Thanks again!
> Ahmad
>
> --
> Pengutronix e.K.                           |                             |
> Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
> 31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |