[PATCH v4 01/26] dt-bindings: interrupt-controller: Add Arm GICv5

[Lorenzo Pieralisi]

On Wed, Jun 04, 2025 at 04:56:02PM +0100, Marc Zyngier wrote:
> On Wed, 04 Jun 2025 08:24:38 +0100,
> Lorenzo Pieralisi <lpieralisi at kernel.org> wrote:
> > 
> > On Tue, Jun 03, 2025 at 02:11:34PM -0500, Rob Herring wrote:
> > > On Tue, Jun 3, 2025 at 10:37 AM Peter Maydell <peter.maydell at linaro.org> wrote:
> > > >
> > > > On Tue, 3 Jun 2025 at 16:15, Rob Herring <robh at kernel.org> wrote:
> > > > >
> > > > > On Tue, Jun 3, 2025 at 2:48 AM Lorenzo Pieralisi <lpieralisi at kernel.org> wrote:
> > > > > >
> > > > > > On Thu, May 29, 2025 at 02:17:26PM +0100, Peter Maydell wrote:
> > > > > > > secure.txt says:
> > > > > > > # The general principle of the naming scheme for Secure world bindings
> > > > > > > # is that any property that needs a different value in the Secure world
> > > > > > > # can be supported by prefixing the property name with "secure-". So for
> > > > > > > # instance "secure-foo" would override "foo".
> > > > >
> > > > > Today I would say a 'secure-' prefix is a mistake. To my knowledge,
> > > > > it's never been used anyways. But I don't have much visibility into
> > > > > what secure world firmware is doing.
> > > >
> > > > QEMU uses it for communicating with the secure firmware if
> > > > you run secure firmware on the virt board. It's done that
> > > > since we introduced that binding. Indeed that use case is *why*
> > > > the binding is there. It works fine for the intended purpose,
> > > > which is "most devices are visible in both S and NS, but a few
> > > > things are S only (UART, a bit of RAM, secure-only flash").
> > > 
> > > I meant "secure-" as a prefix allowed on *any* property, not
> > > "secure-status" specifically, which is the only thing QEMU uses
> > > AFAICT. IOW, I don't think we should be creating secure-reg,
> > > secure-interrupts, secure-clocks, etc.
> > 
> > Reading secure.txt, what does it mean "device present and usable in
> > the secure world" ?
> > 
> > So:
> > 
> > status = "disabled"
> > secure-status = "okay"
> > 
> > basically means that the device in question allows secure-only MMIO
> > access, is that what it says ?
> > 
> > If that's the case and we really want to have all config frames
> > in a single DT, would it be reasonable to have an IRS/ITS DT node
> > per-frame ?
> > 
> > Then yes, the secure- tag is not enough any longer (because we have to
> > cope with 4 interrupt domains) but that's a separate problem - again,
> > this would leave the current reviewed bindings unchanged.
> 
> No, this is the same problem, and we need a way to address it.
> "secure-*" doesn't cut it in a system with FEAT_RME, where resources
> are only available to a single Physical Address Space (PAS). So we
> need a way to qualify these resources with a PAS.

Can I ask again what:

status = "disabled"
secure-status = "okay"

for a device means in practice in the current bindings ?

When I said "a separate problem", I meant that, extending secure- tag
(that applies to the "status" property only) to cover other PASes is
independent from the GICv5 binding *if* we define, for a single DT an eg
IRS device per-PAS (with realm-status, root-status, describing what the
reg property represents. Is that what secure-status does today ? Does
it say "this device MMIO space is secure-only" ?).

It does not look like there is much appetite for tagging the reg
property either and making it GICv5 specific is a shortcut IMO.

> Either that, or we have to restrict DT to describe the view of a
> single PAS. Which Peter will understandably be unhappy about.

Well, I listed a couple of options in this thread, let's try
to converge.

Thanks,
Lorenzo