[PATCH 00/15] cpufreq: simplify boost handling

[PATCH 00/15] cpufreq: simplify boost handling

[Viresh Kumar]

Hello,

The boost feature can be controlled at two levels currently, driver
level (applies to all policies) and per-policy.

Currently most of the drivers enables driver level boost support from the
per-policy ->init() callback, which isn't really efficient as that gets called
for each policy and then there is online/offline path too where this gets done
unnecessarily.

Also it is possible to have a scenario where not all cpufreq policies support
boost frequencies. And letting sysfs (or other parts of the kernel) enable boost
feature for that policy isn't correct.

Simplify and cleanup handling of boost to solve these issues.

Pushed here:

git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm.git cpufreq/boost

Rebased over few dependencies from PM tree, will push to the arm-cpufreq tree
after merge window is closed.

Viresh Kumar (15):
  cpufreq: staticize cpufreq_boost_trigger_state()
  cpufreq: Export cpufreq_boost_set_sw()
  cpufreq: Introduce policy->boost_supported flag
  cpufreq: acpi: Set policy->boost_supported
  cpufreq: amd: Set policy->boost_supported
  cpufreq: cppc: Set policy->boost_supported
  cpufreq: Restrict enabling boost on policies with no boost frequencies
  cpufreq: apple: Set .set_boost directly
  cpufreq: loongson: Set .set_boost directly
  cpufreq: powernv: Set .set_boost directly
  cpufreq: scmi: Set .set_boost directly
  cpufreq: dt: Set .set_boost directly
  cpufreq: qcom: Set .set_boost directly
  cpufreq: staticize policy_has_boost_freq()
  cpufreq: Remove cpufreq_enable_boost_support()

 drivers/cpufreq/acpi-cpufreq.c      |  3 +++
 drivers/cpufreq/amd-pstate.c        |  4 ++--
 drivers/cpufreq/apple-soc-cpufreq.c | 10 +---------
 drivers/cpufreq/cppc_cpufreq.c      |  9 +--------
 drivers/cpufreq/cpufreq-dt.c        | 14 +-------------
 drivers/cpufreq/cpufreq.c           | 30 ++++++++++++-----------------
 drivers/cpufreq/freq_table.c        |  7 +++++--
 drivers/cpufreq/loongson3_cpufreq.c | 10 +---------
 drivers/cpufreq/powernv-cpufreq.c   |  5 +----
 drivers/cpufreq/qcom-cpufreq-hw.c   |  7 +------
 drivers/cpufreq/scmi-cpufreq.c      | 11 +----------
 include/linux/cpufreq.h             | 20 ++++++-------------
 12 files changed, 35 insertions(+), 95 deletions(-)

-- 
2.31.1.272.g89b43f80a514

[PATCH 08/15] cpufreq: apple: Set .set_boost directly

[Viresh Kumar]

The boost feature can be controlled at two levels currently, driver
level (applies to all policies) and per-policy.

Currently the driver enables driver level boost support from the
per-policy ->init() callback, which isn't really efficient as that gets
called for each policy and then there is online/offline path too where
this gets done unnecessarily.

Instead set the .set_boost field directly and always enable the boost
support. If a policy doesn't support boost feature, the core will not
enable it for that policy.

Keep the initial state of driver level boost to disabled and let the
user enable it if required as ideally the boost frequencies must be used
only when really required.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/apple-soc-cpufreq.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)

diff --git a/drivers/cpufreq/apple-soc-cpufreq.c b/drivers/cpufreq/apple-soc-cpufreq.c
index 6ff604a0fe79..4994c86feb57 100644
--- a/drivers/cpufreq/apple-soc-cpufreq.c
+++ b/drivers/cpufreq/apple-soc-cpufreq.c
@@ -310,15 +310,6 @@ static int apple_soc_cpufreq_init(struct cpufreq_policy *policy)
 	policy->fast_switch_possible = true;
 	policy->suspend_freq = freq_table[0].frequency;
 
-	if (policy_has_boost_freq(policy)) {
-		ret = cpufreq_enable_boost_support();
-		if (ret) {
-			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
-		} else {
-			apple_soc_cpufreq_driver.boost_enabled = true;
-		}
-	}
-
 	return 0;
 
 out_free_cpufreq_table:
@@ -353,6 +344,7 @@ static struct cpufreq_driver apple_soc_cpufreq_driver = {
 	.target_index	= apple_soc_cpufreq_set_target,
 	.fast_switch	= apple_soc_cpufreq_fast_switch,
 	.register_em	= cpufreq_register_em_with_opp,
+	.set_boost	= cpufreq_boost_set_sw,
 	.suspend	= cpufreq_generic_suspend,
 };
 
-- 
2.31.1.272.g89b43f80a514

[PATCH 11/15] cpufreq: scmi: Set .set_boost directly

[Viresh Kumar]

The boost feature can be controlled at two levels currently, driver
level (applies to all policies) and per-policy.

Currently the driver enables driver level boost support from the
per-policy ->init() callback, which isn't really efficient as that gets
called for each policy and then there is online/offline path too where
this gets done unnecessarily.

Instead set the .set_boost field directly and always enable the boost
support. If a policy doesn't support boost feature, the core will not
enable it for that policy.

Keep the initial state of driver level boost to disabled and let the
user enable it if required as ideally the boost frequencies must be used
only when really required.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/cpufreq/scmi-cpufreq.c | 11 +----------
 1 file changed, 1 insertion(+), 10 deletions(-)

diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
index 4a3ee59cb771..ff2897789797 100644
--- a/drivers/cpufreq/scmi-cpufreq.c
+++ b/drivers/cpufreq/scmi-cpufreq.c
@@ -297,16 +297,6 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy)
 	policy->transition_delay_us =
 		scmi_get_rate_limit(domain, policy->fast_switch_possible);
 
-	if (policy_has_boost_freq(policy)) {
-		ret = cpufreq_enable_boost_support();
-		if (ret) {
-			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
-			goto out_free_table;
-		} else {
-			scmi_cpufreq_driver.boost_enabled = true;
-		}
-	}
-
 	ret = freq_qos_add_request(&policy->constraints, &priv->limits_freq_req, FREQ_QOS_MAX,
 				   FREQ_QOS_MAX_DEFAULT_VALUE);
 	if (ret < 0) {
@@ -394,6 +384,7 @@ static struct cpufreq_driver scmi_cpufreq_driver = {
 	.init	= scmi_cpufreq_init,
 	.exit	= scmi_cpufreq_exit,
 	.register_em	= scmi_cpufreq_register_em,
+	.set_boost	= cpufreq_boost_set_sw,
 };
 
 static int scmi_cpufreq_probe(struct scmi_device *sdev)
-- 
2.31.1.272.g89b43f80a514

Re: [PATCH 00/15] cpufreq: simplify boost handling

[Rafael J. Wysocki]

On Fri, Jan 24, 2025 at 9:58 AM Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> Hello,
>
> The boost feature can be controlled at two levels currently, driver
> level (applies to all policies) and per-policy.
>
> Currently most of the drivers enables driver level boost support from the
> per-policy ->init() callback, which isn't really efficient as that gets called
> for each policy and then there is online/offline path too where this gets done
> unnecessarily.
>
> Also it is possible to have a scenario where not all cpufreq policies support
> boost frequencies. And letting sysfs (or other parts of the kernel) enable boost
> feature for that policy isn't correct.
>
> Simplify and cleanup handling of boost to solve these issues.

I guess this depends on the previous series?

> Pushed here:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm.git cpufreq/boost
>
> Rebased over few dependencies from PM tree, will push to the arm-cpufreq tree
> after merge window is closed.
>
> Viresh Kumar (15):
>   cpufreq: staticize cpufreq_boost_trigger_state()
>   cpufreq: Export cpufreq_boost_set_sw()
>   cpufreq: Introduce policy->boost_supported flag
>   cpufreq: acpi: Set policy->boost_supported
>   cpufreq: amd: Set policy->boost_supported
>   cpufreq: cppc: Set policy->boost_supported
>   cpufreq: Restrict enabling boost on policies with no boost frequencies
>   cpufreq: apple: Set .set_boost directly
>   cpufreq: loongson: Set .set_boost directly
>   cpufreq: powernv: Set .set_boost directly
>   cpufreq: scmi: Set .set_boost directly
>   cpufreq: dt: Set .set_boost directly
>   cpufreq: qcom: Set .set_boost directly
>   cpufreq: staticize policy_has_boost_freq()
>   cpufreq: Remove cpufreq_enable_boost_support()
>
>  drivers/cpufreq/acpi-cpufreq.c      |  3 +++
>  drivers/cpufreq/amd-pstate.c        |  4 ++--
>  drivers/cpufreq/apple-soc-cpufreq.c | 10 +---------
>  drivers/cpufreq/cppc_cpufreq.c      |  9 +--------
>  drivers/cpufreq/cpufreq-dt.c        | 14 +-------------
>  drivers/cpufreq/cpufreq.c           | 30 ++++++++++++-----------------
>  drivers/cpufreq/freq_table.c        |  7 +++++--
>  drivers/cpufreq/loongson3_cpufreq.c | 10 +---------
>  drivers/cpufreq/powernv-cpufreq.c   |  5 +----
>  drivers/cpufreq/qcom-cpufreq-hw.c   |  7 +------
>  drivers/cpufreq/scmi-cpufreq.c      | 11 +----------
>  include/linux/cpufreq.h             | 20 ++++++-------------
>  12 files changed, 35 insertions(+), 95 deletions(-)
>
> --
> 2.31.1.272.g89b43f80a514
>

Re: [PATCH 00/15] cpufreq: simplify boost handling

[Viresh Kumar]

On 24-01-25, 12:05, Rafael J. Wysocki wrote:
> On Fri, Jan 24, 2025 at 9:58 AM Viresh Kumar <viresh.kumar@linaro.org> wrote:
> >
> > Hello,
> >
> > The boost feature can be controlled at two levels currently, driver
> > level (applies to all policies) and per-policy.
> >
> > Currently most of the drivers enables driver level boost support from the
> > per-policy ->init() callback, which isn't really efficient as that gets called
> > for each policy and then there is online/offline path too where this gets done
> > unnecessarily.
> >
> > Also it is possible to have a scenario where not all cpufreq policies support
> > boost frequencies. And letting sysfs (or other parts of the kernel) enable boost
> > feature for that policy isn't correct.
> >
> > Simplify and cleanup handling of boost to solve these issues.
> 
> I guess this depends on the previous series?

Yes my series and the boost related patches in your tree.

-- 
viresh

Re: [PATCH 11/15] cpufreq: scmi: Set .set_boost directly

[Sudeep Holla]

On Fri, Jan 24, 2025 at 02:28:15PM +0530, Viresh Kumar wrote:
> The boost feature can be controlled at two levels currently, driver
> level (applies to all policies) and per-policy.
> 
> Currently the driver enables driver level boost support from the
> per-policy ->init() callback, which isn't really efficient as that gets
> called for each policy and then there is online/offline path too where
> this gets done unnecessarily.
> 
> Instead set the .set_boost field directly and always enable the boost
> support. If a policy doesn't support boost feature, the core will not
> enable it for that policy.
> 
> Keep the initial state of driver level boost to disabled and let the
> user enable it if required as ideally the boost frequencies must be used
> only when really required.
>

Acked-by: Sudeep Holla <sudeep.holla@arm.com>

-- 
Regards,
Sudeep