policy's fields definitions:
`transition_delay_us`:
The minimum amount of time between two consecutive freq. requests
for one policy.
`transition_latency`:
Delta between freq. change request and effective freq. change on
the hardware.

cpufreq_policy_transition_delay_us() uses the `transition_delay_us`
value if available. Otherwise a value is induced from the policy's
`transition_latency`.

The scmi-cpufreq driver doesn't populate the `transition_delay_us`.
Values matching the definition are available through the SCMI
specification.
Add support to fetch these values and use them in the scmi-cpufreq
driver.

Pierre Gondois (3):
  firmware: arm_scmi: Populate perf commands rate_limit
  firmware: arm_scmi: Populate fast channel rate_limit
  cpufreq: scmi: Set transition_delay_us

 drivers/cpufreq/scmi-cpufreq.c        | 26 +++++++++++++
 drivers/firmware/arm_scmi/driver.c    |  5 ++-
 drivers/firmware/arm_scmi/perf.c      | 53 +++++++++++++++++++++++++--
 drivers/firmware/arm_scmi/powercap.c  | 12 ++++--
 drivers/firmware/arm_scmi/protocols.h |  4 +-
 include/linux/scmi_protocol.h         |  8 ++++
 6 files changed, 98 insertions(+), 10 deletions(-)

-- 
2.25.1

On 22-02-24, 14:56, Pierre Gondois wrote:
> policy's fields definitions:
> `transition_delay_us`:
> The minimum amount of time between two consecutive freq. requests
> for one policy.
> `transition_latency`:
> Delta between freq. change request and effective freq. change on
> the hardware.
> 
> cpufreq_policy_transition_delay_us() uses the `transition_delay_us`
> value if available. Otherwise a value is induced from the policy's
> `transition_latency`.
> 
> The scmi-cpufreq driver doesn't populate the `transition_delay_us`.
> Values matching the definition are available through the SCMI
> specification.
> Add support to fetch these values and use them in the scmi-cpufreq
> driver.

How do we merge this series ? I can only pick the last commit.

-- 
viresh

On Mon, Mar 04, 2024 at 12:30:58PM +0530, Viresh Kumar wrote:
> On 22-02-24, 14:56, Pierre Gondois wrote:
> > policy's fields definitions:
> > `transition_delay_us`:
> > The minimum amount of time between two consecutive freq. requests
> > for one policy.
> > `transition_latency`:
> > Delta between freq. change request and effective freq. change on
> > the hardware.
> >
> > cpufreq_policy_transition_delay_us() uses the `transition_delay_us`
> > value if available. Otherwise a value is induced from the policy's
> > `transition_latency`.
> >
> > The scmi-cpufreq driver doesn't populate the `transition_delay_us`.
> > Values matching the definition are available through the SCMI
> > specification.
> > Add support to fetch these values and use them in the scmi-cpufreq
> > driver.
>
> How do we merge this series ? I can only pick the last commit.

I have sent my PR for v6.9 already and was deferring this to v6.10
The changes look good to me. If it doesn't conflict much with -next
SCMI content, then I am happy to ack and you can take all of them
together. Otherwise we can revisit strategy at -rc1. Thoughts ?

--
Regards,
Sudeep

On 04-03-24, 11:42, Sudeep Holla wrote:
> On Mon, Mar 04, 2024 at 12:30:58PM +0530, Viresh Kumar wrote:
> > On 22-02-24, 14:56, Pierre Gondois wrote:
> > > policy's fields definitions:
> > > `transition_delay_us`:
> > > The minimum amount of time between two consecutive freq. requests
> > > for one policy.
> > > `transition_latency`:
> > > Delta between freq. change request and effective freq. change on
> > > the hardware.
> > >
> > > cpufreq_policy_transition_delay_us() uses the `transition_delay_us`
> > > value if available. Otherwise a value is induced from the policy's
> > > `transition_latency`.
> > >
> > > The scmi-cpufreq driver doesn't populate the `transition_delay_us`.
> > > Values matching the definition are available through the SCMI
> > > specification.
> > > Add support to fetch these values and use them in the scmi-cpufreq
> > > driver.
> >
> > How do we merge this series ? I can only pick the last commit.
> 
> I have sent my PR for v6.9 already and was deferring this to v6.10
> The changes look good to me. If it doesn't conflict much with -next
> SCMI content, then I am happy to ack and you can take all of them
> together. Otherwise we can revisit strategy at -rc1. Thoughts ?

Applied. Thanks.

-- 
viresh