Re: smci: how do reboots work in scmi?

[Cristian Marussi <cristian.marussi@arm.com>]

On Mon, Feb 03, 2025 at 08:05:30AM +0100, Vincent Guittot wrote:
> On Mon, 3 Feb 2025 at 07:25, Dan Carpenter <dan.carpenter@linaro.org> wrote:
> >
> > On Sun, Feb 02, 2025 at 10:12:17PM +0000, Khaled Ali Ahmed wrote:
> > > Hello Dan,
> > > the following message still has not been implemented yet:
> > > - PINCTRL_REQUEST.
> > > - PINCTRL_RELEASE.
> > > - PINCTRL_SET_PERMISSIONS.
> > > - NEGOTIATE_PROTOCOL_VERSION.
> > >
> > > thank you so much for your attention and participation.
> > >
> >
> > Ah right, I was thinking of PINCTRL_SETTINGS_CONFIGURE.
> >
> > When the system reboots, it would be nice to have a hook in SCP to just
> > reset everything.
> 
> Maybe BASE_RESET_AGENT_CONFIGURATION :
> " This command is used to reset platform resource settings that were
> previously configured by an agent.
> Platform resource settings refer to power domain, performance domain,
> clock, sensors and other
> settings associated with a device that the agent has access to. This
> command can also be used to reset
> agent-specific permission configurations to access devices and protocols."

You mean Linux OSPM Agent invoking this commands for itself upon reboot ?
(and the platform acting accordignly..)

First I think we should clarify here what we mean with "system reboot".

- a single Agent rebooting by its own decision
    BASE_RESET_AGENT_CONFIGURATION(self) could be an option

- some Trusted agent causing other agents to shutdown/reboot
    in this case maybe there could be some interactions via SystemPower
    protocol, with the server asking upfront for a graceful shutdown
    which would trigger similarly a BASE_RESET_AGENT_CONFIGURATION(self).
    This would give the server a notion of an "agent resetting" which is
    something that currently is not present in SCMI because the server
    cannot know really if an agent is alive (and I am NOT saying that
    this would be a bad thing.. :P)

    ...BUT, on the other side, if a not-so-graceful shutdown is required
    the victim-agent wont have a chance for a clean reboot/shutdown so
    somehow the TrustedAgent could have to issue a BASE_RESET_AGENT_CONFIGURATION
    on behalf of that agent, OR maybe the server could just cleanup the
    config on its own for the disappearing agent...

- the whole system at the platform level, including the server itself:
  does it matter in this case to reset the agent configs ?
  ...IOW is there a scenario in which an SCMI server could reset, needing
  all its agents' states to be reset too, WHILE the platform that is running
  on somehow survives and retains some state ?

I think we never discussed these scenarios, especially in a virtualized
environement...

Thanks,
Cristian