On 2/23/22 11:27, Viresh Kumar wrote:
On 23-02-22, 11:22, Lukasz Luba wrote:
On 2/23/22 10:43, Viresh Kumar wrote:
On 23-02-22, 10:52, Daniel Lezcano wrote:
why not extend the energy model to any kind of devices?
FWIW, the OPP core supports a wide range of devices now, not just CPUs.
There are many other devices which still use Freq.
Is that the "opp-level" thing which would allow that?
For power supplies/regulators, we don't have freq and they use level, right.
Also for interconnect we use bandwidth, in a similar way.
I can see some DT files with regulators(?) using it e.g. [1].
It looks flexible, the opp-hz is not hard requirement,
the opp-level can be used instead IIUC.
Right.
Looks good. It also doesn't collide with this patch set.
We could have an opp entry like:
opp_1: opp-1 {
opp-level = <1>;
opp-microwatt = <200000>;
};
Daniel would that design make sense to you?
If yes, we could discuss this further after this first
step for fixing GPU in merged. I would need to re-think
the EM em_perf_state and maybe the new ::level there.