[PATCH 0/6] introduce Arm FF-A support

[Simon Glass]

Hi,

On Wed, 13 Apr 2022 at 10:46, Tom Rini <trini at konsulko.com> wrote:
>
> On Wed, Apr 13, 2022 at 03:20:23PM +0100, Abdellatif El Khlifi wrote:
> > On Tue, Apr 12, 2022 at 08:28:42AM -0500, Rob Herring wrote:
> > > On Tue, Apr 12, 2022 at 7:01 AM Tom Rini <trini at konsulko.com> wrote:
> > > >
> > > > On Tue, Apr 12, 2022 at 12:43:15PM +0100, Abdellatif El Khlifi wrote:
> > > > > On Thu, Apr 07, 2022 at 08:58:11AM -0400, Tom Rini wrote:
> > > > > > On Thu, Apr 07, 2022 at 01:54:24PM +0100, Abdellatif El Khlifi wrote:
> > > > > > > On Wed, Apr 06, 2022 at 03:47:11PM -0400, Tom Rini wrote:
> > > > > > > > On Tue, Mar 29, 2022 at 04:16:53PM +0100, abdellatif.elkhlifi at arm.com wrote:
> > > > > > > > > From: Abdellatif El Khlifi <abdellatif.elkhlifi at arm.com>
> > > > > > > > >
> > > > > > > > > This patchset adds support for Arm FF-A (Arm Firmware Framework for Armv8-A v1.0).
> > > > > > > > >
> > > > > > > > > FF-A support is generic by design and can be used by any Arm platform.
> > > > > > > > >
> > > > > > > > > The features added are as follows:
> > > > > > > > >
> > > > > > > > > 1/ FF-A device driver
> > > > > > > > > 2/ armffa command
> > > > > > > > > 3/ FF-A Sandbox driver
> > > > > > > > > 4/ FF-A Sandbox test cases
> > > > > > > > > 5/ FF-A MM communication
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > The suggested design sees FF-A as a data bus allowing data exchange with the firmware
> > > > > > > > > running under TrustZone HW (such as Optee). The same approach was followed in the
> > > > > > > > > FF-A driver in Linux kernel (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/firmware/arm_ffa/bus.c?h=v5.15#n211))
> > > > > > > > >
> > > > > > > > > u-boot boards using FF-A can provide a device tree node in a <board>-u-boot.dtsi file.
> > > > > > > > > Since the node can not be hosted in Linux device tree, we suggest using u-boot device tree.
> > > > > > > >
> > > > > > > > Why can't the node be in the upstream tree?  It should be, so that it
> > > > > > > > can be shared between all users.  Especially since there's in-Linux
> > > > > > > > users?
> > > > > > > >
> > > > > > > > --
> > > > > > > > Tom
> > > > > > >
> > > > > > > Linux already has an FF-A bus driver and doesn't use a device tree node for FF-A.
> > > > > > >
> > > > > > > The Linux driver registers FF-A as a bus:
> > > > > > >
> > > > > > > int arm_ffa_bus_init(void)
> > > > > > > {
> > > > > > >   return bus_register(&ffa_bus_type);
> > > > > > > }
> > > > > > >
> > > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/firmware/arm_ffa/bus.c?h=v5.15#n211
> > > > > > >
> > > > > > > So, there is no user for the node in Linux. That's why we suggest hosting the node in the u-boot device tree (a u-boot.dtsi file)
> > > > > >
> > > > > > OK, but you can still push it upstream as it's not required to have an
> > > > > > in tree user.
> > > > >
> > > > > During the review of Corstone-1000 patchset, Rui Silva had a discussion with the Linux device tree maintainer
> > > > > (Rob Herring). Rob is not in favour of an FFA node in the kernel device tree. This is why we are including the FFA node
> > > > > in u-boot device tree (u-boot.dtsi files).
> > >
> > > Sigh. There is not a 'kernel device tree' and a 'u-boot device tree'.
> > > There is only 1. For SystemReadyIR compliance, that is a hard
> > > requirement.
> > >
> > > > I'm a bit confused now, can you please link to the kernel thread?  Or
> > > > Rob, can you chime in here please?
> > >
> > > The FFA DT binding was rejected in favor of making FFA discoverable.
> > > The FFA spec was amended to address that. DT is only for what we
> > > failed to make discoverable. For hardware, we're stuck with it. We
> > > shouldn't repeat that for software interfaces.
> > >
> > > Rob
> >
> > Guys,
> >
> > Since we can not add an FFA node in the device tree, we will make FFA a discoverable bus.
> > So, we will manually create the udevice, binding it to the driver and probing it.
> > Manually means directly calling device_bind and device_probe APIs.
> >
> > Any thoughts about this approach ?
>
> How is it both discoverable and doesn't have a device tree node, in the
> kernel?

Also, if it is discoverable, we can still use U-Boot to discover it
and then pass the info to Linux in the DT.

I am seeing several series which don't have 'proper' DT bindings in
Linux. First I heard it was for legacy reasons, but now I am hearing
something different. For U-Boot, we really do need to have DT bindings
for devices. All this ad-hoc creation of stuff makes things hard to
discover, adds to code size and makes things like of-platdata
impossible.

Furthermore, if the bindings affect U-Boot, then the U-Boot project
should have a say in what is being done there, not just be downstream
of all such changes.

Regards,
Simon