RE: [PATCH v2 2/4] firmware: arm_scmi: Add machine_allowlist and machine_blocklist

[Peng Fan <peng.fan@xxxxxxx>]

> Subject: Re: [PATCH v2 2/4] firmware: arm_scmi: Add
> machine_allowlist and machine_blocklist
> 
> On Tue, Feb 11, 2025 at 03:46:36PM +0000, Sudeep Holla wrote:
> >On Mon, Feb 10, 2025 at 01:19:14PM +0000, Peng Fan wrote:
> >>
> >> I just have a prototype and tested on i.MX95.
> >
> >You didn't answer me @[1]. How can we disable it for perf/cpufreq if
> >there are users already. I will look at the code once I am convince we
> can do that.
> >For now, I am not. I am worried we may break some platform.
> 
> The only user in upstream kernel with using the dummy clock is juno-
> scmi.dtsi.
> 
> SCMI_PROTOCOL_PERF is used by two drivers cpufreq-scmi.c,
> scmi_perf_domain.c.
> 
> In cpufreq-scmi.c a dummy clock proviver is created, the gpu node in
> juno-scmi.dtsi takes "<&scmi_dvfs 2>" into clocks property. I think this
> is wrong.
> 
> Why not use scmi_clk node? cpufreq created clk provider should only
> be limited for cpu device which will not be impacted by fwdevlink.
> 
> If wanna to tune gpu performance, the power-domains property should
> be used, not clocks property.
> 
> It is the juno-scmi.dtsi should be fixed.
> 
> If juno-scmi.dtsi will keep as it is, please suggest possible solution on
> fixing the issue.

Ignore the upper which maybe wrong.

The dummy clock provider will always be ready before opp. So no issue.
But anyway if juno-scmi.dtsi using power-domains for perf, it should
be better.

I just replied in V1.

Regards,
Peng.

> 
> Regards,
> Peng
> 
> >
> >--
> >Regards,
> >Sudeep
> >
> >[1]
> https://lore.kernel.org/all/20241227151306.jh2oabc64xd54dms@bog
> us