Add callback to register with energy model
Subsystems (6)
Viresh Kumar
maintainerv3 cover Add callback to register with energy model
Many cpufreq drivers register with the energy model for each policy and do exactly the same thing. Follow the footsteps of thermal-cooling, to get it done from the cpufreq core itself.
Provide a new callback, which will be called, if present, by the cpufreq core at the right moment (more on that in the code's comment). Also provide a generic implementation that uses dev_pm_opp_of_register_em().
This also allows us to register with the EM at a later point of time, compared to ->init(), from where the EM core can access cpufreq policy directly using cpufreq_cpu_get() type of helpers and perform other work, like marking few frequencies inefficient, this will be done separately.
This is build/boot tested by the bot for a couple of boards.
https://gitlab.com/vireshk/pmko/-/pipelines/351965580
FWIW, I have queued up the series for linux-next to get more testing, but your reviews are welcome. Thanks.
V2->V3:
- Drop the auto-register part from all logs, this isn't called auto registration anymore.
- Call register_em() only for new policies.
- Update scmi driver to register with EM only when required.
- Add Rby tags.
V1->V2:
- Add a callback instead of flag.
- Register before governor is initialized.
- Update scmi driver as well.
- Don't unregister from the EM core.
--
Viresh
Viresh Kumar (9):
cpufreq: Add callback to register with energy model
cpufreq: dt: Use .register_em() to register with energy model
cpufreq: imx6q: Use .register_em() to register with energy model
cpufreq: mediatek: Use .register_em() to register with energy model
cpufreq: omap: Use .register_em() to register with energy model
cpufreq: qcom-cpufreq-hw: Use .register_em() to register with energy
model
cpufreq: scpi: Use .register_em() to register with energy model
cpufreq: vexpress: Use .register_em() to register with energy model
cpufreq: scmi: Use .register_em() to register with energy model
drivers/cpufreq/cpufreq-dt.c | 3 +- drivers/cpufreq/cpufreq.c | 13 ++++++ drivers/cpufreq/imx6q-cpufreq.c | 2 +- drivers/cpufreq/mediatek-cpufreq.c | 3 +- drivers/cpufreq/omap-cpufreq.c | 2 +- drivers/cpufreq/qcom-cpufreq-hw.c | 3 +- drivers/cpufreq/scmi-cpufreq.c | 65 +++++++++++++++++--------- drivers/cpufreq/scpi-cpufreq.c | 3 +- drivers/cpufreq/vexpress-spc-cpufreq.c | 3 +- include/linux/cpufreq.h | 14 ++++++ 10 files changed, 76 insertions(+), 35 deletions(-)
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 1/9 cpufreq: Add callback to register with energy model
Many cpufreq drivers register with the energy model for each policy and do exactly the same thing. Follow the footsteps of thermal-cooling, to get it done from the cpufreq core itself.
Provide a new callback, which will be called, if present, by the cpufreq core at the right moment (more on that in the code's comment). Also provide a generic implementation that uses dev_pm_opp_of_register_em().
This also allows us to register with the EM at a later point of time, compared to ->init(), from where the EM core can access cpufreq policy directly using cpufreq_cpu_get() type of helpers and perform other work, like marking few frequencies inefficient, this will be done separately.
--- drivers/cpufreq/cpufreq.c | 13 +++++++++++++ include/linux/cpufreq.h | 14 ++++++++++++++ 2 files changed, 27 insertions(+)
drivers/cpufreq/cpufreq.c +13 -0
@@ -1491,6 +1491,19 @@ static int cpufreq_online(unsigned int cpu)write_lock_irqsave (&cpufreq_driver_lock, flags);list_add (&policy->policy_list, &cpufreq_policy_list);write_unlock_irqrestore (&cpufreq_driver_lock, flags); + +/* +* Register with the energy model before +* sched_cpufreq_governor_change() is called, which will result +* in rebuilding of the sched domains, which should only be done +* once the energy model is properly initialized for the policy +* first. +* +* Also, this should be called before the policy is registered +* with cooling framework. +*/ +if (cpufreq_driver->register_em) + cpufreq_driver->register_em (policy); } ret =cpufreq_init_policy (policy);
include/linux/cpufreq.h +14 -0
@@ -9,10 +9,12 @@#define _LINUX_CPUFREQ_H#include <linux/clk.h> +#include <linux/cpu.h> #include <linux/cpumask.h> #include <linux/completion.h> #include <linux/kobject.h> #include <linux/notifier.h> +#include <linux/pm_opp.h> #include <linux/pm_qos.h> #include <linux/spinlock.h> #include <linux/sysfs.h>
@@ -373,6 +375,12 @@ struct cpufreq_driver {/* platform specific boost support code */ bool boost_enabled;int (*set_boost)(struct cpufreq_policy *policy,int state); + +/* +* Set by drivers that want to register with the energy model after the +* policy is properly initialized, but before the governor is started. +*/ +void (*register_em)(struct cpufreq_policy *policy); };/* flags */
@@ -1046,4 +1054,10 @@ unsigned int cpufreq_generic_get(unsigned int cpu);void cpufreq_generic_init (struct cpufreq_policy *policy,struct cpufreq_frequency_table *table,unsigned int transition_latency); + +static inline void cpufreq_register_em_with_opp (struct cpufreq_policy *policy) +{ +dev_pm_opp_of_register_em (get_cpu_device (policy->cpu), + policy->related_cpus); +}#endif /* _LINUX_CPUFREQ_H */
Signature · 1 line
2.31.1.272.g89b43f80a514
Re: [PATCH V3 1/9] cpufreq: Add callback to register with energy model
Quoted 89 lines · Viresh Kumar
Many cpufreq drivers register with the energy model for each policy and do exactly the same thing. Follow the footsteps of thermal-cooling, to get it done from the cpufreq core itself.
Provide a new callback, which will be called, if present, by the cpufreq core at the right moment (more on that in the code's comment). Also provide a generic implementation that uses
dev_pm_opp_of_register_em().This also allows us to register with the EM at a later point of time, compared to ->init(), from where the EM core can access cpufreq policy directly using
cpufreq_cpu_get()type of helpers and perform other work, like marking few frequencies inefficient, this will be done separately.Reviewed-by Quentin Perret <[email protected]>Reviewed-by Lukasz Luba <[email protected]>Signed-off-by Viresh Kumar <[email protected]>--- drivers/cpufreq/cpufreq.c | 13 +++++++++++++ include/linux/cpufreq.h | 14 ++++++++++++++ 2 files changed, 27 insertions(+)drivers/cpufreq/cpufreq.c +13
-0@@ -1491,6 +1491,19 @@ static int cpufreq_online(unsigned int cpu)write_lock_irqsave (&cpufreq_driver_lock, flags);list_add (&policy->policy_list, &cpufreq_policy_list);write_unlock_irqrestore (&cpufreq_driver_lock, flags); + +/* +* Register with the energy model before +* sched_cpufreq_governor_change() is called, which will result +* in rebuilding of the sched domains, which should only be done +* once the energy model is properly initialized for the policy +* first. +* +* Also, this should be called before the policy is registered +* with cooling framework. +*/ +if (cpufreq_driver->register_em) + cpufreq_driver->register_em (policy); } ret =cpufreq_init_policy (policy);include/linux/cpufreq.h +14
-0@@ -9,10 +9,12 @@#define _LINUX_CPUFREQ_H#include <linux/clk.h> +#include <linux/cpu.h> #include <linux/cpumask.h> #include <linux/completion.h> #include <linux/kobject.h> #include <linux/notifier.h> +#include <linux/pm_opp.h> #include <linux/pm_qos.h> #include <linux/spinlock.h> #include <linux/sysfs.h> @@ -373,6 +375,12 @@ struct cpufreq_driver {/* platform specific boost support code */ bool boost_enabled;int (*set_boost)(struct cpufreq_policy *policy,int state); + +/* +* Set by drivers that want to register with the energy model after the +* policy is properly initialized, but before the governor is started. +*/ +void (*register_em)(struct cpufreq_policy *policy); };/* flags */ @@ -1046,4 +1054,10 @@ unsigned int cpufreq_generic_get(unsigned int cpu);void cpufreq_generic_init (struct cpufreq_policy *policy,struct cpufreq_frequency_table *table,unsigned int transition_latency); + +static inline void cpufreq_register_em_with_opp (struct cpufreq_policy *policy) +{ +dev_pm_opp_of_register_em (get_cpu_device (policy->cpu), + policy->related_cpus); +}#endif /* _LINUX_CPUFREQ_H */ Signature · 1 line
2.31.1.272.g89b43f80a514
Tested on Hikey960 while working on the inefficient OPPs support.
Viresh Kumar
maintainerPATCH v3 2/9 cpufreq: dt: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/cpufreq-dt.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/cpufreq/cpufreq-dt.c +1 -2
@@ -143,8 +143,6 @@ static int cpufreq_init(struct cpufreq_policy *policy) cpufreq_dt_attr[1 ] = &cpufreq_freq_attr_scaling_boost_freqs; } -dev_pm_opp_of_register_em (cpu_dev, policy->cpus); -return 0 ; out_clk_put:
@@ -184,6 +182,7 @@ static struct cpufreq_driver dt_cpufreq_driver = { .exit = cpufreq_exit, .online = cpufreq_online, .offline = cpufreq_offline, + .register_em = cpufreq_register_em_with_opp, .name = "cpufreq-dt", .attr = cpufreq_dt_attr, .suspend = cpufreq_generic_suspend,
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 3/9 cpufreq: imx6q: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/imx6q-cpufreq.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cpufreq/imx6q-cpufreq.c +1 -1
@@ -192,7 +192,6 @@ static int imx6q_cpufreq_init(struct cpufreq_policy *policy) policy->clk = clks[ARM ].clk;cpufreq_generic_init (policy, freq_table, transition_latency); policy->suspend_freq = max_freq; -dev_pm_opp_of_register_em (cpu_dev, policy->cpus);return 0 ; }
@@ -204,6 +203,7 @@ static struct cpufreq_driver imx6q_cpufreq_driver = { .target_index = imx6q_set_target, .get = cpufreq_generic_get, .init = imx6q_cpufreq_init, + .register_em = cpufreq_register_em_with_opp, .name = "imx6q-cpufreq", .attr = cpufreq_generic_attr, .suspend = cpufreq_generic_suspend,
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 4/9 cpufreq: mediatek: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/mediatek-cpufreq.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/cpufreq/mediatek-cpufreq.c +1 -2
@@ -448,8 +448,6 @@ static int mtk_cpufreq_init(struct cpufreq_policy *policy) policy->driver_data = info; policy->clk = info->cpu_clk; -dev_pm_opp_of_register_em (info->cpu_dev, policy->cpus); -return 0 ; }
@@ -471,6 +469,7 @@ static struct cpufreq_driver mtk_cpufreq_driver = { .get = cpufreq_generic_get, .init = mtk_cpufreq_init, .exit = mtk_cpufreq_exit, + .register_em = cpufreq_register_em_with_opp, .name = "mtk-cpufreq", .attr = cpufreq_generic_attr, };
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 5/9 cpufreq: omap: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/omap-cpufreq.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cpufreq/omap-cpufreq.c +1 -1
@@ -131,7 +131,6 @@ static int omap_cpu_init(struct cpufreq_policy *policy)/* FIXME: what's the actual transition time? */ cpufreq_generic_init (policy, freq_table,300 *1000 ); -dev_pm_opp_of_register_em (mpu_dev, policy->cpus);return 0 ; }
@@ -150,6 +149,7 @@ static struct cpufreq_driver omap_driver = { .get = cpufreq_generic_get, .init = omap_cpu_init, .exit = omap_cpu_exit, + .register_em = cpufreq_register_em_with_opp, .name = "omap", .attr = cpufreq_generic_attr, };
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 6/9 cpufreq: qcom-cpufreq-hw: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/qcom-cpufreq-hw.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/cpufreq/qcom-cpufreq-hw.c +1 -2
@@ -362,8 +362,6 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy) goto error; } -dev_pm_opp_of_register_em (cpu_dev, policy->cpus); -if (policy_has_boost_freq (policy)) { ret =cpufreq_enable_boost_support ();if (ret)
@@ -412,6 +410,7 @@ static struct cpufreq_driver cpufreq_qcom_hw_driver = { .get = qcom_cpufreq_hw_get, .init = qcom_cpufreq_hw_cpu_init, .exit = qcom_cpufreq_hw_cpu_exit, + .register_em = cpufreq_register_em_with_opp, .fast_switch = qcom_cpufreq_hw_fast_switch, .name = "qcom-cpufreq-hw", .attr = qcom_cpufreq_hw_attr,
Signature · 1 line
2.31.1.272.g89b43f80a514
Viresh Kumar
maintainerPATCH v3 7/9 cpufreq: scpi: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/scpi-cpufreq.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/cpufreq/scpi-cpufreq.c +1 -2
@@ -163,8 +163,6 @@ static int scpi_cpufreq_init(struct cpufreq_policy *policy) policy->fast_switch_possible = false; -dev_pm_opp_of_register_em (cpu_dev, policy->cpus); -return 0 ; out_free_cpufreq_table:
@@ -200,6 +198,7 @@ static struct cpufreq_driver scpi_cpufreq_driver = { .init = scpi_cpufreq_init, .exit = scpi_cpufreq_exit, .target_index = scpi_cpufreq_set_target, + .register_em = cpufreq_register_em_with_opp, };static int scpi_cpufreq_probe (struct platform_device *pdev)
Signature · 1 line
2.31.1.272.g89b43f80a514
Re: [PATCH V3 7/9] cpufreq: scpi: Use .register_em() to register with energy model
Quoted 3 lines · Viresh Kumar
Set the newly added .register_em() callback with
cpufreq_register_em_with_opp()to register with the EM core.
If not too late,
Signature · 2 lines
Regards, Sudeep
Viresh Kumar
maintainerPATCH v3 8/9 cpufreq: vexpress: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback with cpufreq_register_em_with_opp() to register with the EM core.
--- drivers/cpufreq/vexpress-spc-cpufreq.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/cpufreq/vexpress-spc-cpufreq.c +1 -2
@@ -440,8 +440,6 @@ static int ve_spc_cpufreq_init(struct cpufreq_policy *policy) policy->freq_table = freq_table[cur_cluster]; policy->cpuinfo.transition_latency =1000000 ;/* 1 ms */ -dev_pm_opp_of_register_em (cpu_dev, policy->cpus); -if (is_bL_switching_enabled ())per_cpu (cpu_last_req_freq, policy->cpu) =clk_get_cpu_rate (policy->cpu);
@@ -475,6 +473,7 @@ static struct cpufreq_driver ve_spc_cpufreq_driver = { .get = ve_spc_cpufreq_get_rate, .init = ve_spc_cpufreq_init, .exit = ve_spc_cpufreq_exit, + .register_em = cpufreq_register_em_with_opp, .attr = cpufreq_generic_attr, };
Signature · 1 line
2.31.1.272.g89b43f80a514
Re: [PATCH V3 8/9] cpufreq: vexpress: Use .register_em() to register with energy model
Quoted 3 lines · Viresh Kumar
Set the newly added .register_em() callback with
cpufreq_register_em_with_opp()to register with the EM core.
Signature · 2 lines
Regards, Sudeep
Viresh Kumar
maintainerPATCH v3 9/9 cpufreq: scmi: Use .register_em() to register with energy model
In reply to Viresh KumarSet the newly added .register_em() callback to register with the EM after the cpufreq policy is properly initialized.
--- drivers/cpufreq/scmi-cpufreq.c | 65 ++++++++++++++++++++++------------ 1 file changed, 42 insertions(+), 23 deletions(-)
drivers/cpufreq/scmi-cpufreq.c +42 -23
@@ -22,7 +22,9 @@struct scmi_data {int domain_id; +int nr_opp;struct device *cpu_dev; +cpumask_var_t opp_shared_cpus; };static struct scmi_protocol_handle *ph;
@@ -123,9 +125,6 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy)struct device *cpu_dev;struct scmi_data *priv;struct cpufreq_frequency_table *freq_table; -struct em_data_callback em_cb =EM_DATA_CB (scmi_get_cpu_power); -cpumask_var_t opp_shared_cpus; -bool power_scale_mw; cpu_dev =get_cpu_device (policy->cpu);if (!cpu_dev) {
@@ -133,9 +132,15 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy)return -ENODEV ; } -if (!zalloc_cpumask_var (&opp_shared_cpus,GFP_KERNEL )) + priv =kzalloc (sizeof (*priv),GFP_KERNEL ); +if (!priv)return -ENOMEM ; +if (!zalloc_cpumask_var (&priv->opp_shared_cpus,GFP_KERNEL )) { + ret = -ENOMEM ; + goto out_free_priv; + } +/* Obtain CPUs that share SCMI performance controls */ ret =scmi_get_sharing_cpus (cpu_dev, policy->cpus);if (ret) {
@@ -148,14 +153,14 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy)* The OPP 'sharing cpus' info may come from DT through an empty opp * table and opp-shared. */ - ret =dev_pm_opp_of_get_sharing_cpus (cpu_dev, opp_shared_cpus); -if (ret || !cpumask_weight (opp_shared_cpus)) { + ret =dev_pm_opp_of_get_sharing_cpus (cpu_dev, priv->opp_shared_cpus); +if (ret || !cpumask_weight (priv->opp_shared_cpus)) {/* * Either opp-table is not set or no opp-shared was found. * Use the CPU mask from SCMI to designate CPUs sharing an OPP * table. */ -cpumask_copy (opp_shared_cpus, policy->cpus); +cpumask_copy (priv->opp_shared_cpus, policy->cpus); }/*
@@ -180,7 +185,7 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) goto out_free_opp; } - ret =dev_pm_opp_set_sharing_cpus (cpu_dev, opp_shared_cpus); + ret =dev_pm_opp_set_sharing_cpus (cpu_dev, priv->opp_shared_cpus);if (ret) {dev_err (cpu_dev,"%s: failed to mark OPPs as shared: %d\n" , __func__, ret);
@@ -188,21 +193,13 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) goto out_free_opp; } - power_scale_mw = perf_ops->power_scale_mw_get (ph); -em_dev_register_perf_domain (cpu_dev, nr_opp, &em_cb, - opp_shared_cpus, power_scale_mw); - } - - priv =kzalloc (sizeof (*priv),GFP_KERNEL ); -if (!priv) { - ret = -ENOMEM ; - goto out_free_opp; + priv->nr_opp = nr_opp; } ret =dev_pm_opp_init_cpufreq_table (cpu_dev, &freq_table);if (ret) {dev_err (cpu_dev,"failed to init cpufreq table: %d\n" , ret); - goto out_free_priv; + goto out_free_opp; } priv->cpu_dev = cpu_dev;
@@ -223,17 +220,16 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) policy->fast_switch_possible = perf_ops->fast_switch_possible (ph, cpu_dev); -free_cpumask_var (opp_shared_cpus);return 0 ; -out_free_priv: -kfree (priv); - out_free_opp:dev_pm_opp_remove_all_dynamic (cpu_dev); out_free_cpumask: -free_cpumask_var (opp_shared_cpus); +free_cpumask_var (priv->opp_shared_cpus); + +out_free_priv: +kfree (priv);return ret; }
@@ -244,11 +240,33 @@ static int scmi_cpufreq_exit(struct cpufreq_policy *policy)dev_pm_opp_free_cpufreq_table (priv->cpu_dev, &policy->freq_table);dev_pm_opp_remove_all_dynamic (priv->cpu_dev); +free_cpumask_var (priv->opp_shared_cpus);kfree (priv);return 0 ; } +static void scmi_cpufreq_register_em (struct cpufreq_policy *policy) +{ +struct em_data_callback em_cb =EM_DATA_CB (scmi_get_cpu_power); +bool power_scale_mw = perf_ops->power_scale_mw_get (ph); +struct scmi_data *priv = policy->driver_data; + +/* +* This callback will be called for each policy, but we don't need to +* register with EM every time. Despite not being part of the same +* policy, some CPUs may still share their perf-domains, and a CPU from +* another policy may already have registered with EM on behalf of CPUs +* of this policy. +*/ +if (!priv->nr_opp) +return ; + +em_dev_register_perf_domain (get_cpu_device (policy->cpu), priv->nr_opp, + &em_cb, priv->opp_shared_cpus, + power_scale_mw); +} +static struct cpufreq_driver scmi_cpufreq_driver = { .name ="scmi" , .flags = CPUFREQ_HAVE_GOVERNOR_PER_POLICY |
@@ -261,6 +279,7 @@ static struct cpufreq_driver scmi_cpufreq_driver = { .get = scmi_cpufreq_get_rate, .init = scmi_cpufreq_init, .exit = scmi_cpufreq_exit, + .register_em = scmi_cpufreq_register_em, };static int scmi_cpufreq_probe (struct scmi_device *sdev)
Signature · 1 line
2.31.1.272.g89b43f80a514
Re: [PATCH V3 9/9] cpufreq: scmi: Use .register_em() to register with energy model
Quoted 3 lines · Viresh Kumar
Set the newly added .register_em() callback to register with the EM after the cpufreq policy is properly initialized.
Signature · 2 lines
Regards, Sudeep
Re: [PATCH V3 9/9] cpufreq: scmi: Use .register_em() to register with energy model
In reply to Viresh KumarHi Viresh,
My apologies for delay I was on holidays. I've seen the pull requests, so it's too late, but just for history...
I have tested this new callback with hacked juno FW which works in the per-cpu perf requests. There are no issues.
Some debug prints:
[ 3.110072] cpu cpu0: cpumask weight(opp_shared_cpus)=4 weight policy->cpus=1 [ 3.117666] cpu cpu0: Empty OPP table [ 3.121418] cpu cpu0: OPP table empty [ 3.131367] cpu cpu0: EM: created perf domain [ 3.137848] cpu cpu1: cpumask weight(opp_shared_cpus)=2 weight policy->cpus=1 [ 3.145220] cpu cpu1: Empty OPP table [ 3.148961] cpu cpu1: OPP table empty [ 3.158193] cpu cpu1: EM: created perf domain [ 3.164325] cpu cpu2: cpumask weight(opp_shared_cpus)=2 weight policy->cpus=1 [ 3.173430] cpu cpu3: cpumask weight(opp_shared_cpus)=4 weight policy->cpus=1 [ 3.181947] cpu cpu4: cpumask weight(opp_shared_cpus)=4 weight policy->cpus=1 [ 3.190620] cpu cpu5: cpumask weight(opp_shared_cpus)=4 weight policy->cpus=1
root@sqwt-ubuntu:~# grep . /sys/kernel/debug/energy_model/cpu0/ps\:*/* /sys/kernel/debug/energy_model/cpu0/ps:450000/cost:79 /sys/kernel/debug/energy_model/cpu0/ps:450000/frequency:450000 /sys/kernel/debug/energy_model/cpu0/ps:450000/power:42 /sys/kernel/debug/energy_model/cpu0/ps:575000/cost:85 /sys/kernel/debug/energy_model/cpu0/ps:575000/frequency:575000 /sys/kernel/debug/energy_model/cpu0/ps:575000/power:58 /sys/kernel/debug/energy_model/cpu0/ps:700000/cost:95 /sys/kernel/debug/energy_model/cpu0/ps:700000/frequency:700000 /sys/kernel/debug/energy_model/cpu0/ps:700000/power:79 /sys/kernel/debug/energy_model/cpu0/ps:775000/cost:106 /sys/kernel/debug/energy_model/cpu0/ps:775000/frequency:775000 /sys/kernel/debug/energy_model/cpu0/ps:775000/power:97 /sys/kernel/debug/energy_model/cpu0/ps:850000/cost:119 /sys/kernel/debug/energy_model/cpu0/ps:850000/frequency:850000 /sys/kernel/debug/energy_model/cpu0/ps:850000/power:119 root@sqwt-ubuntu:~# cat /sys/kernel/debug/energy_model/cpu0/cpus 0,3-5 root@sqwt-ubuntu:~# cat /sys/kernel/debug/energy_model/cpu1/cpus 1-2 root@sqwt-ubuntu:~# grep . /sys/kernel/debug/energy_model/cpu1/ps\:*/* /sys/kernel/debug/energy_model/cpu1/ps:1100000/cost:583 /sys/kernel/debug/energy_model/cpu1/ps:1100000/frequency:1100000 /sys/kernel/debug/energy_model/cpu1/ps:1100000/power:583 /sys/kernel/debug/energy_model/cpu1/ps:450000/cost:391 /sys/kernel/debug/energy_model/cpu1/ps:450000/frequency:450000 /sys/kernel/debug/energy_model/cpu1/ps:450000/power:160 /sys/kernel/debug/energy_model/cpu1/ps:625000/cost:420 /sys/kernel/debug/energy_model/cpu1/ps:625000/frequency:625000 /sys/kernel/debug/energy_model/cpu1/ps:625000/power:239 /sys/kernel/debug/energy_model/cpu1/ps:800000/cost:471 /sys/kernel/debug/energy_model/cpu1/ps:800000/frequency:800000 /sys/kernel/debug/energy_model/cpu1/ps:800000/power:343 /sys/kernel/debug/energy_model/cpu1/ps:950000/cost:525 /sys/kernel/debug/energy_model/cpu1/ps:950000/frequency:950000 /sys/kernel/debug/energy_model/cpu1/ps:950000/power:454 root@sqwt-ubuntu:~#
Regards,
Lukasz
Quoted 20 lines · Viresh Kumar
Many cpufreq drivers register with the energy model for each policy and do exactly the same thing. Follow the footsteps of thermal-cooling, to get it done from the cpufreq core itself.
Provide a new callback, which will be called, if present, by the cpufreq core at the right moment (more on that in the code's comment). Also provide a generic implementation that uses
dev_pm_opp_of_register_em().This also allows us to register with the EM at a later point of time, compared to ->init(), from where the EM core can access cpufreq policy directly using
cpufreq_cpu_get()type of helpers and perform other work, like marking few frequencies inefficient, this will be done separately.This is build/boot tested by the bot for a couple of boards.
https://gitlab.com/vireshk/pmko/-/pipelines/351965580
FWIW, I have queued up the series for linux-next to get more testing, but your reviews are welcome. Thanks.
Quoted 6 lines
V2->V3:
- Drop the auto-register part from all logs, this isn't called auto registration anymore.
- Callregister_em()only for new policies.
- Update scmi driver to register with EM only when required.
- Add Rby tags.
LGTM. Tested on TC2 (vexpress-spc), Juno-r0 (scpi-cpufreq and scmi).