Re: [PATCH 2/3] pmdomain: Add support for system-suspend-only states

[Sudeep Holla <sudeep.holla@kernel.org>]

On Mon, Nov 09, 2026 at 04:05:56PM +0100, Ulf Hansson wrote:
> On Wed, Oct 28, 2026 at 12:48 PM Sudeep Holla <sudeep.holla@kernel.org> wrote:
> >

[...]

> > Now I understood the requirement much better. The core issue with OSI is
> > that it fails unless the platform downgrades special system states to
> > non-system states during routine CPUIdle paths.
> >
> > This leaves us with two directions for OSI:
> > 1. Firmware Resolution: Handle state transitions entirely in firmware.
> > 2. OS Intervention: Enforce the handling via the OS.
> >
> > Though I personally prefer (1), I don't have strong arguments against (2)
> > now that you made me understand the requirement. Sorry for taking time,
> > but mentally I was assuming (1) even for OSI until your above response.
> >
> > Since PC mode relies on the firmware to manage these states natively, this
> > binding isn't required for it. Therefore, Could we restrict this specific
> > binding exclusively to OSI implementations ?
> 
> I guess we could, but that seems rather messy. According to earlier
> discussions, if I recall correctly, we deliberately avoided OSI
> specific DT bindings/properties, since support for OSI mode should be
> detected through the FW.
>

Fair enough, we should at the least warn if used on the systems in
PC CPU suspend mode.

> Also note that the domain-idle-state DT bindings
> (Documentation/devicetree/bindings/power/domain-idle-state.yaml) are
> generic and used by RISC-V SBI as well.
> 

Agreed.

-- 
Regards,
Sudeep