Re: POWER_DOMAIN_ATTRIBUTES in SCMI

[Kevin Hilman <khilman@kernel.org>]

Souvik Chakravarty <souvik.chakravarty@arm.com> writes:

> Some quick thoughts below to keep this architecturally consistent.

[...]

> There are 2 ways to wakeup a device:
> a) The device is powered by one or more PDs, one or more of which needs
> to be kept ON to trigger the wakeup.
> b) The device is powered by one or more PDs, but none of them are
> required to trigger a wakeup. The device wake is derived from an out of
> band logic which is kept powered ON (either transparently to OS, or via
> another PD which does NOT belong to the device).

In the case of TI SoCs, it's (b).  In fact, most of the IO pads on the
SoC can be enabled as wakeup capable when the SoC hits certain low-power
states.  The bits to control this are part of the pinctrl settings for
the each driver on these SoCs.  So technically it's neither the device
nor its power domain that cause the wakeup, but the external pad, aided
by some dedicated, always-on hardware, called "IO daisy chain" in the TI
documentation[1].

Kevin

[1] c.f Section 6.2.3.11 I/O Power Management and Daisy Chaining
    in the AM62L TRM for b bit more details: https://www.ti.com/lit/pdf/sprujb4