Hi all,
This patch set introduces support for devices in the Energy Model (EM) framework. It will unify the power model for thermal subsystem and make it simpler. The 1st patch refactors EM framework and adds support for devices. The 2nd patch changes dev_pm_opp_of_register_em() in OPP/OF which now should take as an argument struct device pointer. It touches a few trees (OMAP, NXP, Mediatek) updating their CPUfreq drivers to the new interface. Patch 3/4 changes thermal devfreq cooling removing old code for calculating local power table. It simplifies the code and uses EM for requested power calculation. Last patch 4/4 adds EM to Panfrost driver.
The patch set is based on linux-next tag next-20200221.
Changes: v3: - added back the cpumask 'cpus' in the em_perf_domain due potential cache misses - removed _is_cpu_em() since there is no need for it - changed function name from em_pd_energy() to em_cpu_energy(), which is optimized for usage from the scheduler making some assumptions and not validating arguments to speed-up, there is a comment stressing that it should be used only for CPUs em_perf_domain - changed em_get_pd() to em_pd_get() which is now aligned with em_cpu_get() naming - extended comment in em_cpu_get() describing the need for this function - fixed build warning reported on x86 by kbuild test robot in devfreq_cooling.c - updated documentation in the energy-model.rst - changed print messages from 'energy_model' to 'EM' - changed dev_warn to dev_dbg, should calm down test scripts in case the platform has OPPs less efficient in the OPP table (some of them are there for cooling reasons, we shouldn't warn in this case, debug info is enough)
v2 [2]: - changed EM API em_register_perf_domain() adding cpumask_t pointer as last argument (which was discussed with Dietmar and Quentin) - removed dependency on PM_OPP, thanks to the cpumask_t argument - removed enum em_type and em->type dependent code - em_get_pd() can handle CPU device as well as devfreq device - updated EM documentation - in devfreq cooling added code which prevents from race condition with devfreq governors which are trying to use OPPs while thermal is in the middle of disabling them. - in devfreq cooling added code which updates state of the devfreq device to avoid working on stale data when governor has not updated it for a long time - in devfreq cooling added backward compatibility frequency table for drivers which did not provide EM - added Steven's Reviewed-by to trace code in thermal - added another CPUFreq driver which needs to be updated to the new API
The v1 can be found here [1].
Regards, Lukasz Luba
[1] https://lkml.org/lkml/2020/1/16/619 [2] https://lkml.org/lkml/2020/2/6/377
Lukasz Luba (4): PM / EM: add devices to Energy Model OPP: change parameter to device pointer in dev_pm_opp_of_register_em() thermal: devfreq_cooling: Refactor code and switch to use Energy Model drm/panfrost: Register to the Energy Model with devfreq device
Documentation/power/energy-model.rst | 133 +++--- Documentation/scheduler/sched-energy.rst | 2 +- drivers/cpufreq/cpufreq-dt.c | 2 +- drivers/cpufreq/imx6q-cpufreq.c | 2 +- drivers/cpufreq/mediatek-cpufreq.c | 2 +- drivers/cpufreq/omap-cpufreq.c | 2 +- drivers/cpufreq/qcom-cpufreq-hw.c | 2 +- drivers/cpufreq/scmi-cpufreq.c | 11 +- drivers/cpufreq/scpi-cpufreq.c | 2 +- drivers/cpufreq/vexpress-spc-cpufreq.c | 2 +- drivers/gpu/drm/panfrost/panfrost_devfreq.c | 3 + drivers/opp/of.c | 45 +-- drivers/thermal/cpufreq_cooling.c | 10 +- drivers/thermal/devfreq_cooling.c | 425 +++++++++----------- include/linux/devfreq_cooling.h | 17 - include/linux/energy_model.h | 107 +++-- include/linux/pm_opp.h | 14 +- include/trace/events/thermal.h | 19 +- kernel/power/energy_model.c | 400 ++++++++++++++---- kernel/sched/fair.c | 2 +- kernel/sched/topology.c | 4 +- 21 files changed, 711 insertions(+), 495 deletions(-)
Add support of other devices into the Energy Model framework not only the CPUs. Change the interface to be more unified which can handle other devices as well.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com --- Documentation/power/energy-model.rst | 133 ++++---- Documentation/scheduler/sched-energy.rst | 2 +- drivers/cpufreq/scmi-cpufreq.c | 11 +- drivers/opp/of.c | 9 +- drivers/thermal/cpufreq_cooling.c | 10 +- include/linux/energy_model.h | 107 +++--- kernel/power/energy_model.c | 400 ++++++++++++++++++----- kernel/sched/fair.c | 2 +- kernel/sched/topology.c | 4 +- 9 files changed, 469 insertions(+), 209 deletions(-)
diff --git a/Documentation/power/energy-model.rst b/Documentation/power/energy-model.rst index 90a345d57ae9..7576820664e5 100644 --- a/Documentation/power/energy-model.rst +++ b/Documentation/power/energy-model.rst @@ -1,15 +1,17 @@ -==================== -Energy Model of CPUs -==================== +.. SPDX-License-Identifier: GPL-2.0 + +======================= +Energy Model of devices +=======================
1. Overview -----------
The Energy Model (EM) framework serves as an interface between drivers knowing -the power consumed by CPUs at various performance levels, and the kernel +the power consumed by devices at various performance levels, and the kernel subsystems willing to use that information to make energy-aware decisions.
-The source of the information about the power consumed by CPUs can vary greatly +The source of the information about the power consumed by devices can vary greatly from one platform to another. These power costs can be estimated using devicetree data in some cases. In others, the firmware will know better. Alternatively, userspace might be best positioned. And so on. In order to avoid @@ -25,7 +27,7 @@ framework, and interested clients reading the data from it:: +---------------+ +-----------------+ +---------------+ | Thermal (IPA) | | Scheduler (EAS) | | Other | +---------------+ +-----------------+ +---------------+ - | | em_pd_energy() | + | | em_cpu_energy() | | | em_cpu_get() | +---------+ | +---------+ | | | @@ -47,12 +49,12 @@ framework, and interested clients reading the data from it:: | Device Tree | | Firmware | | ? | +--------------+ +---------------+ +--------------+
-The EM framework manages power cost tables per 'performance domain' in the -system. A performance domain is a group of CPUs whose performance is scaled -together. Performance domains generally have a 1-to-1 mapping with CPUFreq -policies. All CPUs in a performance domain are required to have the same -micro-architecture. CPUs in different performance domains can have different -micro-architectures. +In case of CPU devices the EM framework manages power cost tables per +'performance domain' in the system. A performance domain is a group of CPUs +whose performance is scaled together. Performance domains generally have a +1-to-1 mapping with CPUFreq policies. All CPUs in a performance domain are +required to have the same micro-architecture. CPUs in different performance +domains can have different micro-architectures.
2. Core APIs @@ -70,14 +72,16 @@ CONFIG_ENERGY_MODEL must be enabled to use the EM framework. Drivers are expected to register performance domains into the EM framework by calling the following API::
- int em_register_perf_domain(cpumask_t *span, unsigned int nr_states, - struct em_data_callback *cb); + int em_register_perf_domain(struct device *dev, unsigned int nr_states, + struct em_data_callback *cb, cpumask_t *cpus);
-Drivers must specify the CPUs of the performance domains using the cpumask -argument, and provide a callback function returning <frequency, power> tuples -for each capacity state. The callback function provided by the driver is free +Drivers must provide a callback function returning <frequency, power> tuples +for each performance state. The callback function provided by the driver is free to fetch data from any relevant location (DT, firmware, ...), and by any mean -deemed necessary. See Section 3. for an example of driver implementing this +deemed necessary. Only for CPU devices, drivers must specify the CPUs of the +performance domains using cpumask. For other devices than CPUs the last +argument must be set to NULL. +See Section 3. for an example of driver implementing this callback, and kernel/power/energy_model.c for further documentation on this API.
@@ -85,13 +89,20 @@ API. 2.3 Accessing performance domains ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+There is two API functions which provide the access to the energy model: +em_cpu_get() which takes CPU id as an argument and em_pd_get() with device +pointer as an argument. It depends on the subsystem which interface it is +going to use, but in case of CPU devices both functions return the same +performance domain. + Subsystems interested in the energy model of a CPU can retrieve it using the em_cpu_get() API. The energy model tables are allocated once upon creation of the performance domains, and kept in memory untouched.
The energy consumed by a performance domain can be estimated using the -em_pd_energy() API. The estimation is performed assuming that the schedutil -CPUfreq governor is in use. +em_cpu_energy() API. The estimation is performed assuming that the schedutil +CPUfreq governor is in use in case of CPU device. Currently this calculation is +not provided for other type of devices.
More details about the above APIs can be found in include/linux/energy_model.h.
@@ -106,42 +117,46 @@ EM framework::
-> drivers/cpufreq/foo_cpufreq.c
- 01 static int est_power(unsigned long *mW, unsigned long *KHz, int cpu) - 02 { - 03 long freq, power; - 04 - 05 /* Use the 'foo' protocol to ceil the frequency */ - 06 freq = foo_get_freq_ceil(cpu, *KHz); - 07 if (freq < 0); - 08 return freq; - 09 - 10 /* Estimate the power cost for the CPU at the relevant freq. */ - 11 power = foo_estimate_power(cpu, freq); - 12 if (power < 0); - 13 return power; - 14 - 15 /* Return the values to the EM framework */ - 16 *mW = power; - 17 *KHz = freq; - 18 - 19 return 0; - 20 } - 21 - 22 static int foo_cpufreq_init(struct cpufreq_policy *policy) - 23 { - 24 struct em_data_callback em_cb = EM_DATA_CB(est_power); - 25 int nr_opp, ret; - 26 - 27 /* Do the actual CPUFreq init work ... */ - 28 ret = do_foo_cpufreq_init(policy); - 29 if (ret) - 30 return ret; - 31 - 32 /* Find the number of OPPs for this policy */ - 33 nr_opp = foo_get_nr_opp(policy); - 34 - 35 /* And register the new performance domain */ - 36 em_register_perf_domain(policy->cpus, nr_opp, &em_cb); - 37 - 38 return 0; - 39 } + 01 static int est_power(unsigned long *mW, unsigned long *KHz, + 02 struct device *dev) + 03 { + 04 long freq, power; + 05 + 06 /* Use the 'foo' protocol to ceil the frequency */ + 07 freq = foo_get_freq_ceil(dev, *KHz); + 08 if (freq < 0); + 09 return freq; + 10 + 11 /* Estimate the power cost for the dev at the relevant freq. */ + 12 power = foo_estimate_power(dev, freq); + 13 if (power < 0); + 14 return power; + 15 + 16 /* Return the values to the EM framework */ + 17 *mW = power; + 18 *KHz = freq; + 19 + 20 return 0; + 21 } + 22 + 23 static int foo_cpufreq_init(struct cpufreq_policy *policy) + 24 { + 25 struct em_data_callback em_cb = EM_DATA_CB(est_power); + 26 struct device *cpu_dev; + 27 int nr_opp, ret; + 28 + 29 cpu_dev = get_cpu_device(cpumask_first(policy->cpus)); + 30 + 31 /* Do the actual CPUFreq init work ... */ + 32 ret = do_foo_cpufreq_init(policy); + 33 if (ret) + 34 return ret; + 35 + 36 /* Find the number of OPPs for this policy */ + 37 nr_opp = foo_get_nr_opp(policy); + 38 + 39 /* And register the new performance domain */ + 40 em_register_perf_domain(cpu_dev, nr_opp, &em_cb, policy->cpus); + 41 + 42 return 0; + 43 } diff --git a/Documentation/scheduler/sched-energy.rst b/Documentation/scheduler/sched-energy.rst index 9580c57a52bc..8ba5a581c1d6 100644 --- a/Documentation/scheduler/sched-energy.rst +++ b/Documentation/scheduler/sched-energy.rst @@ -141,7 +141,7 @@ in its previous activation. find_energy_efficient_cpu() uses compute_energy() to estimate what will be the energy consumed by the system if the waking task was migrated. compute_energy() looks at the current utilization landscape of the CPUs and adjusts it to -'simulate' the task migration. The EM framework provides the em_pd_energy() API +'simulate' the task migration. The EM framework provides the em_cpu_energy() API which computes the expected energy consumption of each performance domain for the given utilization landscape.
diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c index 61623e2ff149..10172aa9ce3e 100644 --- a/drivers/cpufreq/scmi-cpufreq.c +++ b/drivers/cpufreq/scmi-cpufreq.c @@ -103,17 +103,12 @@ scmi_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpumask) }
static int __maybe_unused -scmi_get_cpu_power(unsigned long *power, unsigned long *KHz, int cpu) +scmi_get_cpu_power(unsigned long *power, unsigned long *KHz, + struct device *cpu_dev) { - struct device *cpu_dev = get_cpu_device(cpu); unsigned long Hz; int ret, domain;
- if (!cpu_dev) { - pr_err("failed to get cpu%d device\n", cpu); - return -ENODEV; - } - domain = handle->perf_ops->device_domain_id(cpu_dev); if (domain < 0) return domain; @@ -200,7 +195,7 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy)
policy->fast_switch_possible = true;
- em_register_perf_domain(policy->cpus, nr_opp, &em_cb); + em_register_perf_domain(cpu_dev, nr_opp, &em_cb, policy->cpus);
return 0;
diff --git a/drivers/opp/of.c b/drivers/opp/of.c index 9cd8f0adacae..0efd6cf6d023 100644 --- a/drivers/opp/of.c +++ b/drivers/opp/of.c @@ -1047,9 +1047,8 @@ EXPORT_SYMBOL_GPL(dev_pm_opp_get_of_node); * calculation failed because of missing parameters, 0 otherwise. */ static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz, - int cpu) + struct device *cpu_dev) { - struct device *cpu_dev; struct dev_pm_opp *opp; struct device_node *np; unsigned long mV, Hz; @@ -1057,10 +1056,6 @@ static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz, u64 tmp; int ret;
- cpu_dev = get_cpu_device(cpu); - if (!cpu_dev) - return -ENODEV; - np = of_node_get(cpu_dev->of_node); if (!np) return -EINVAL; @@ -1128,6 +1123,6 @@ void dev_pm_opp_of_register_em(struct cpumask *cpus) if (ret || !cap) return;
- em_register_perf_domain(cpus, nr_opp, &em_cb); + em_register_perf_domain(cpu_dev, nr_opp, &em_cb, cpus); } EXPORT_SYMBOL_GPL(dev_pm_opp_of_register_em); diff --git a/drivers/thermal/cpufreq_cooling.c b/drivers/thermal/cpufreq_cooling.c index fe83d7a210d4..fcf2dab1b3b8 100644 --- a/drivers/thermal/cpufreq_cooling.c +++ b/drivers/thermal/cpufreq_cooling.c @@ -333,18 +333,18 @@ static inline bool em_is_sane(struct cpufreq_cooling_device *cpufreq_cdev, return false;
policy = cpufreq_cdev->policy; - if (!cpumask_equal(policy->related_cpus, to_cpumask(em->cpus))) { + if (!cpumask_equal(policy->related_cpus, em_span_cpus(em))) { pr_err("The span of pd %*pbl is misaligned with cpufreq policy %*pbl\n", - cpumask_pr_args(to_cpumask(em->cpus)), + cpumask_pr_args(em_span_cpus(em)), cpumask_pr_args(policy->related_cpus)); return false; }
nr_levels = cpufreq_cdev->max_level + 1; - if (em->nr_cap_states != nr_levels) { + if (em->nr_perf_states != nr_levels) { pr_err("The number of cap states in pd %*pbl (%u) doesn't match the number of cooling levels (%u)\n", - cpumask_pr_args(to_cpumask(em->cpus)), - em->nr_cap_states, nr_levels); + cpumask_pr_args(em_span_cpus(em)), + em->nr_perf_states, nr_levels); return false; }
diff --git a/include/linux/energy_model.h b/include/linux/energy_model.h index d249b88a4d5a..3b38f3bd097d 100644 --- a/include/linux/energy_model.h +++ b/include/linux/energy_model.h @@ -2,6 +2,7 @@ #ifndef _LINUX_ENERGY_MODEL_H #define _LINUX_ENERGY_MODEL_H #include <linux/cpumask.h> +#include <linux/device.h> #include <linux/jump_label.h> #include <linux/kobject.h> #include <linux/rcupdate.h> @@ -10,13 +11,15 @@ #include <linux/types.h>
/** - * em_cap_state - Capacity state of a performance domain - * @frequency: The CPU frequency in KHz, for consistency with CPUFreq - * @power: The power consumed by 1 CPU at this level, in milli-watts + * em_perf_state - Performance state of a performance domain + * @frequency: The frequency in KHz, for consistency with CPUFreq + * @power: The power consumed at this level, in milli-watts (by 1 CPU or + by a registered device). It can be a total power: static and + dynamic. * @cost: The cost coefficient associated with this level, used during * energy calculation. Equal to: power * max_frequency / frequency */ -struct em_cap_state { +struct em_perf_state { unsigned long frequency; unsigned long power; unsigned long cost; @@ -24,63 +27,76 @@ struct em_cap_state {
/** * em_perf_domain - Performance domain - * @table: List of capacity states, in ascending order - * @nr_cap_states: Number of capacity states - * @cpus: Cpumask covering the CPUs of the domain + * @table: List of performance states, in ascending order + * @nr_perf_states: Number of performance states + * @cpus: Cpumask covering the CPUs of the domain. It's here + * for performance reasons to avoid potential cache + * misses during energy calculations in the scheduler * - * A "performance domain" represents a group of CPUs whose performance is - * scaled together. All CPUs of a performance domain must have the same - * micro-architecture. Performance domains often have a 1-to-1 mapping with - * CPUFreq policies. + * In case of CPU device, a "performance domain" represents a group of CPUs + * whose performance is scaled together. All CPUs of a performance domain + * must have the same micro-architecture. Performance domains often have + * a 1-to-1 mapping with CPUFreq policies. In case of other devices the 'cpus' + * field is unused. */ struct em_perf_domain { - struct em_cap_state *table; - int nr_cap_states; - unsigned long cpus[0]; + struct em_perf_state *table; + int nr_perf_states; + unsigned long cpus[]; };
+#define em_span_cpus(em) (to_cpumask((em)->cpus)) + #ifdef CONFIG_ENERGY_MODEL -#define EM_CPU_MAX_POWER 0xFFFF +#define EM_MAX_POWER 0xFFFF
struct em_data_callback { /** - * active_power() - Provide power at the next capacity state of a CPU - * @power : Active power at the capacity state in mW (modified) - * @freq : Frequency at the capacity state in kHz (modified) - * @cpu : CPU for which we do this operation + * active_power() - Provide power at the next performance state of a + * device + * @power : Active power at the performance state in mW (modified) + * @freq : Frequency at the performance state in kHz (modified) + * @dev : Device for which we do this operation (can be a CPU) * - * active_power() must find the lowest capacity state of 'cpu' above + * active_power() must find the lowest performance state of 'dev' above * 'freq' and update 'power' and 'freq' to the matching active power * and frequency. * - * The power is the one of a single CPU in the domain, expressed in - * milli-watts. It is expected to fit in the [0, EM_CPU_MAX_POWER] - * range. + * In case of CPUs, the power is the one of a single CPU in the domain, + * expressed in milli-watts. It is expected to fit in the + * [0, EM_MAX_POWER] range. * * Return 0 on success. */ - int (*active_power)(unsigned long *power, unsigned long *freq, int cpu); + int (*active_power)(unsigned long *power, unsigned long *freq, + struct device *dev); }; #define EM_DATA_CB(_active_power_cb) { .active_power = &_active_power_cb }
struct em_perf_domain *em_cpu_get(int cpu); -int em_register_perf_domain(cpumask_t *span, unsigned int nr_states, - struct em_data_callback *cb); +struct em_perf_domain *em_pd_get(struct device *dev); +int em_register_perf_domain(struct device *dev, unsigned int nr_states, + struct em_data_callback *cb, cpumask_t *span); +void em_unregister_perf_domain(struct device *dev);
/** - * em_pd_energy() - Estimates the energy consumed by the CPUs of a perf. domain + * em_cpu_energy() - Estimates the energy consumed by the CPUs of a perf. domain * @pd : performance domain for which energy has to be estimated * @max_util : highest utilization among CPUs of the domain * @sum_util : sum of the utilization of all CPUs in the domain * + * This function should be used only for CPU devices. There is no validation, + * i.e. if the EM is a CPU type and has cpumask allocated. It is called from + * the scheduler code quite frequently. + * * Return: the sum of the energy consumed by the CPUs of the domain assuming * a capacity state satisfying the max utilization of the domain. */ -static inline unsigned long em_pd_energy(struct em_perf_domain *pd, +static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, unsigned long max_util, unsigned long sum_util) { unsigned long freq, scale_cpu; - struct em_cap_state *cs; + struct em_perf_state *cs; int i, cpu;
/* @@ -88,16 +104,16 @@ static inline unsigned long em_pd_energy(struct em_perf_domain *pd, * most utilized CPU of the performance domain to a requested frequency, * like schedutil. */ - cpu = cpumask_first(to_cpumask(pd->cpus)); + cpu = cpumask_first(em_span_cpus(pd)); scale_cpu = arch_scale_cpu_capacity(cpu); - cs = &pd->table[pd->nr_cap_states - 1]; + cs = &pd->table[pd->nr_perf_states - 1]; freq = map_util_freq(max_util, cs->frequency, scale_cpu);
/* * Find the lowest capacity state of the Energy Model above the * requested frequency. */ - for (i = 0; i < pd->nr_cap_states; i++) { + for (i = 0; i < pd->nr_perf_states; i++) { cs = &pd->table[i]; if (cs->frequency >= freq) break; @@ -133,7 +149,7 @@ static inline unsigned long em_pd_energy(struct em_perf_domain *pd, * cpu_nrg = ------------------------ * --------- (3) * cs->freq scale_cpu * - * The first term is static, and is stored in the em_cap_state struct + * The first term is static, and is stored in the em_perf_state struct * as 'cs->cost'. * * Since all CPUs of the domain have the same micro-architecture, they @@ -149,35 +165,44 @@ static inline unsigned long em_pd_energy(struct em_perf_domain *pd, }
/** - * em_pd_nr_cap_states() - Get the number of capacity states of a perf. domain + * em_pd_nr_perf_states() - Get the number of performance states of a + performance domain * @pd : performance domain for which this must be done * - * Return: the number of capacity states in the performance domain table + * Return: the number of performance states in the performance domain table */ -static inline int em_pd_nr_cap_states(struct em_perf_domain *pd) +static inline int em_pd_nr_perf_states(struct em_perf_domain *pd) { - return pd->nr_cap_states; + return pd->nr_perf_states; }
#else struct em_data_callback {}; #define EM_DATA_CB(_active_power_cb) { }
-static inline int em_register_perf_domain(cpumask_t *span, - unsigned int nr_states, struct em_data_callback *cb) +static inline int em_register_perf_domain(struct device *dev, + unsigned int nr_states, struct em_data_callback *cb, + cpumask_t *span) { return -EINVAL; } +static inline void em_unregister_perf_domain(struct device *dev) +{ +} static inline struct em_perf_domain *em_cpu_get(int cpu) { return NULL; } -static inline unsigned long em_pd_energy(struct em_perf_domain *pd, +static inline struct em_perf_domain *em_pd_get(struct device *dev) +{ + return NULL; +} +static inline unsigned long em_cpu_energy(struct em_perf_domain *pd, unsigned long max_util, unsigned long sum_util) { return 0; } -static inline int em_pd_nr_cap_states(struct em_perf_domain *pd) +static inline int em_pd_nr_perf_states(struct em_perf_domain *pd) { return 0; } diff --git a/kernel/power/energy_model.c b/kernel/power/energy_model.c index 0a9326f5f421..0f46325eb8de 100644 --- a/kernel/power/energy_model.c +++ b/kernel/power/energy_model.c @@ -2,8 +2,9 @@ /* * Energy Model of CPUs * - * Copyright (c) 2018, Arm ltd. + * Copyright (c) 2018-2020, Arm ltd. * Written by: Quentin Perret, Arm ltd. + * Improvements provided by: Lukasz Luba, Arm ltd. */
#define pr_fmt(fmt) "energy_model: " fmt @@ -12,22 +13,49 @@ #include <linux/cpumask.h> #include <linux/debugfs.h> #include <linux/energy_model.h> +#include <linux/idr.h> #include <linux/sched/topology.h> #include <linux/slab.h>
-/* Mapping of each CPU to the performance domain to which it belongs. */ -static DEFINE_PER_CPU(struct em_perf_domain *, em_data); +/** + * em_device - Performance domain wrapper for device + * @em_pd: Performance domain which carries the energy model + * @dev: Device for which this performance domain is set + * @id: Id of this performance domain + * @em_dev_list: List entry to connect all the devices perf. domain + * @debug_dir: Optional debug directory + * + * Internal structure. It contains a "performance domain" and the corresponding + * device. + */ +struct em_device { + struct em_perf_domain *em_pd; + struct device *dev; + int id; + struct list_head em_dev_list; +#ifdef CONFIG_DEBUG_FS + struct dentry *debug_dir; +#endif +};
+static DEFINE_IDA(em_dev_ida); /* * Mutex serializing the registrations of performance domains and letting * callbacks defined by drivers sleep. */ static DEFINE_MUTEX(em_pd_mutex); +/* List of devices' energy model, protected by 'em_pd_mutex' */ +static LIST_HEAD(em_pd_dev_list); + +static bool _is_cpu_device(struct device *dev) +{ + return (dev->bus == &cpu_subsys); +}
#ifdef CONFIG_DEBUG_FS static struct dentry *rootdir;
-static void em_debug_create_cs(struct em_cap_state *cs, struct dentry *pd) +static void em_debug_create_cs(struct em_perf_state *cs, struct dentry *pd) { struct dentry *d; char name[24]; @@ -43,28 +71,72 @@ static void em_debug_create_cs(struct em_cap_state *cs, struct dentry *pd)
static int em_debug_cpus_show(struct seq_file *s, void *unused) { - seq_printf(s, "%*pbl\n", cpumask_pr_args(to_cpumask(s->private))); + struct em_perf_domain *em_pd = s->private; + struct cpumask *mask = em_span_cpus(em_pd); + + seq_printf(s, "%*pbl\n", cpumask_pr_args(mask));
return 0; } DEFINE_SHOW_ATTRIBUTE(em_debug_cpus);
-static void em_debug_create_pd(struct em_perf_domain *pd, int cpu) +static int em_debug_dev_show(struct seq_file *s, void *unused) +{ + struct em_device *em_dev = s->private; + const char *domain_name; + + domain_name = dev_name(em_dev->dev); + if (!domain_name) + domain_name = "unknown"; + + seq_printf(s, "%s\n", domain_name); + + return 0; +} +DEFINE_SHOW_ATTRIBUTE(em_debug_dev); + +static int em_debug_type_show(struct seq_file *s, void *unused) { + struct em_device *em_dev = s->private; + + if (_is_cpu_device(em_dev->dev)) + seq_puts(s, "EM cpufreq device\n"); + else + seq_puts(s, "EM devfreq device\n"); + + return 0; +} +DEFINE_SHOW_ATTRIBUTE(em_debug_type); + +static void em_debug_create_pd(struct em_device *em_dev) +{ + struct em_perf_domain *em_pd = em_dev->em_pd; struct dentry *d; char name[8]; int i;
- snprintf(name, sizeof(name), "pd%d", cpu); + snprintf(name, sizeof(name), "pd%d", em_dev->id);
/* Create the directory of the performance domain */ d = debugfs_create_dir(name, rootdir);
- debugfs_create_file("cpus", 0444, d, pd->cpus, &em_debug_cpus_fops); + if (_is_cpu_device(em_dev->dev)) + debugfs_create_file("cpus", 0444, d, em_pd, + &em_debug_cpus_fops); + else + debugfs_create_file("dev_name", 0444, d, em_dev, + &em_debug_dev_fops); + + debugfs_create_file("type", 0444, d, em_dev, &em_debug_type_fops); + + /* Create a sub-directory for each performance state */ + for (i = 0; i < em_pd->nr_perf_states; i++) + em_debug_create_cs(&em_pd->table[i], d); +}
- /* Create a sub-directory for each capacity state */ - for (i = 0; i < pd->nr_cap_states; i++) - em_debug_create_cs(&pd->table[i], d); +static void em_debug_remove_pd(struct em_device *em_dev) +{ + debugfs_remove_recursive(em_dev->debug_dir); }
static int __init em_debug_init(void) @@ -76,48 +148,44 @@ static int __init em_debug_init(void) } core_initcall(em_debug_init); #else /* CONFIG_DEBUG_FS */ -static void em_debug_create_pd(struct em_perf_domain *pd, int cpu) {} +static void em_debug_create_pd(struct em_device *em_dev) {} +static void em_debug_remove_pd(struct em_device *em_dev) {} #endif -static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states, - struct em_data_callback *cb) + +static int em_create_perf_table(struct device *dev, struct em_perf_domain *pd, + int nr_states, struct em_data_callback *cb) { unsigned long opp_eff, prev_opp_eff = ULONG_MAX; unsigned long power, freq, prev_freq = 0; - int i, ret, cpu = cpumask_first(span); - struct em_cap_state *table; - struct em_perf_domain *pd; + struct em_perf_state *table; + int i, ret; u64 fmax;
- if (!cb->active_power) - return NULL; - - pd = kzalloc(sizeof(*pd) + cpumask_size(), GFP_KERNEL); - if (!pd) - return NULL; - table = kcalloc(nr_states, sizeof(*table), GFP_KERNEL); if (!table) - goto free_pd; + return -ENOMEM;
- /* Build the list of capacity states for this performance domain */ + /* Build the list of performance states for this performance domain */ for (i = 0, freq = 0; i < nr_states; i++, freq++) { /* * active_power() is a driver callback which ceils 'freq' to - * lowest capacity state of 'cpu' above 'freq' and updates + * lowest performance state of 'dev' above 'freq' and updates * 'power' and 'freq' accordingly. */ - ret = cb->active_power(&power, &freq, cpu); + ret = cb->active_power(&power, &freq, dev); if (ret) { - pr_err("pd%d: invalid cap. state: %d\n", cpu, ret); + dev_err(dev, "EM: invalid perf. state: %d\n", + ret); goto free_cs_table; }
/* * We expect the driver callback to increase the frequency for - * higher capacity states. + * higher performance states. */ if (freq <= prev_freq) { - pr_err("pd%d: non-increasing freq: %lu\n", cpu, freq); + dev_err(dev, "EM: non-increasing freq: %lu\n", + freq); goto free_cs_table; }
@@ -125,8 +193,9 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states, * The power returned by active_state() is expected to be * positive, in milli-watts and to fit into 16 bits. */ - if (!power || power > EM_CPU_MAX_POWER) { - pr_err("pd%d: invalid power: %lu\n", cpu, power); + if (!power || power > EM_MAX_POWER) { + dev_err(dev, "EM: invalid power: %lu\n", + power); goto free_cs_table; }
@@ -141,12 +210,12 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states, */ opp_eff = freq / power; if (opp_eff >= prev_opp_eff) - pr_warn("pd%d: hertz/watts ratio non-monotonically decreasing: em_cap_state %d >= em_cap_state%d\n", - cpu, i, i - 1); + dev_dbg(dev, "EM: hertz/watts ratio non-monotonically decreasing: em_perf_state %d >= em_perf_state%d\n", + i, i - 1); prev_opp_eff = opp_eff; }
- /* Compute the cost of each capacity_state. */ + /* Compute the cost of each performance_state. */ fmax = (u64) table[nr_states - 1].frequency; for (i = 0; i < nr_states; i++) { table[i].cost = div64_u64(fmax * table[i].power, @@ -154,39 +223,147 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states, }
pd->table = table; - pd->nr_cap_states = nr_states; - cpumask_copy(to_cpumask(pd->cpus), span); - - em_debug_create_pd(pd, cpu); + pd->nr_perf_states = nr_states;
- return pd; + return 0;
free_cs_table: kfree(table); -free_pd: - kfree(pd); + return -EINVAL; +} + +static struct em_perf_domain *em_create_pd(struct device *dev, int nr_states, + struct em_data_callback *cb, + cpumask_t *cpus) +{ + struct em_perf_domain *pd; + int ret; + + if (_is_cpu_device(dev)) { + pd = kzalloc(sizeof(*pd) + cpumask_size(), GFP_KERNEL); + if (!pd) + return NULL; + + cpumask_copy(em_span_cpus(pd), cpus); + } else { + pd = kzalloc(sizeof(*pd), GFP_KERNEL); + if (!pd) + return NULL; + } + + ret = em_create_perf_table(dev, pd, nr_states, cb); + if (ret) { + kfree(pd); + return NULL; + } + + return pd; +} + +static bool em_cpus_pd_exist(cpumask_t *span) +{ + struct em_device *em_dev; + + /* The em_pd_mutex should be held already. */ + if (list_empty(&em_pd_dev_list)) + return 0; + + list_for_each_entry(em_dev, &em_pd_dev_list, em_dev_list) { + if (!_is_cpu_device(em_dev->dev)) + continue; + + if (cpumask_equal(span, em_span_cpus(em_dev->em_pd))) + return 1; + } + + return 0; +} + +static void em_destroy_pd(struct em_perf_domain *em_pd) +{ + kfree(em_pd->table); + kfree(em_pd); +}
+/** + * em_pd_get() - Return the performance domain for a device + * @dev : Device to find the performance domain for + * + * Returns the performance domain to which 'dev' belongs, or NULL if it doesn't + * exist. + */ +struct em_perf_domain *em_pd_get(struct device *dev) +{ + struct em_device *em_dev; + + if (IS_ERR_OR_NULL(dev)) + return NULL; + + if (_is_cpu_device(dev)) + return em_cpu_get(dev->id); + + mutex_lock(&em_pd_mutex); + + if (list_empty(&em_pd_dev_list)) + goto unlock; + + list_for_each_entry(em_dev, &em_pd_dev_list, em_dev_list) { + if (em_dev->dev == dev) { + mutex_unlock(&em_pd_mutex); + return em_dev->em_pd; + } + } + +unlock: + mutex_unlock(&em_pd_mutex); return NULL; } +EXPORT_SYMBOL_GPL(em_pd_get);
/** * em_cpu_get() - Return the performance domain for a CPU * @cpu : CPU to find the performance domain for * - * Return: the performance domain to which 'cpu' belongs, or NULL if it doesn't + * This function is especially useful for subsystems which operate on CPU id, + * like topology setup code does. It simplifies code in those subsystems. + * + * Returns the performance domain to which 'cpu' belongs, or NULL if it doesn't * exist. */ struct em_perf_domain *em_cpu_get(int cpu) { - return READ_ONCE(per_cpu(em_data, cpu)); + struct em_device *em_dev; + + mutex_lock(&em_pd_mutex); + + if (list_empty(&em_pd_dev_list)) + goto unlock; + + list_for_each_entry(em_dev, &em_pd_dev_list, em_dev_list) { + if (!_is_cpu_device(em_dev->dev)) + continue; + + if (cpumask_test_cpu(cpu, em_span_cpus(em_dev->em_pd))) { + mutex_unlock(&em_pd_mutex); + return em_dev->em_pd; + } + } + +unlock: + mutex_unlock(&em_pd_mutex); + return NULL; } EXPORT_SYMBOL_GPL(em_cpu_get);
/** - * em_register_perf_domain() - Register the Energy Model of a performance domain - * @span : Mask of CPUs in the performance domain - * @nr_states : Number of capacity states to register + * em_register_perf_domain() - Register the Energy Model (EM) of a performance + * domain for the device + * @dev : Device for which the EM is to register + * @nr_states : Number of performance states to register * @cb : Callback functions providing the data of the Energy Model + * @cpus : Pointer to cpumask_t, which in case of a CPU device is + * obligatory. It can be taken from i.e. 'policy->cpus'. For other + * type of devices this should be set to NULL. * * Create Energy Model tables for a performance domain using the callbacks * defined in cb. @@ -196,63 +373,116 @@ EXPORT_SYMBOL_GPL(em_cpu_get); * * Return 0 on success */ -int em_register_perf_domain(cpumask_t *span, unsigned int nr_states, - struct em_data_callback *cb) +int em_register_perf_domain(struct device *dev, unsigned int nr_states, + struct em_data_callback *cb, cpumask_t *cpus) { unsigned long cap, prev_cap = 0; struct em_perf_domain *pd; - int cpu, ret = 0; + struct em_device *em_dev; + int cpu, ret;
- if (!span || !nr_states || !cb) + if (!dev || !nr_states || !cb || !cb->active_power) return -EINVAL;
- /* - * Use a mutex to serialize the registration of performance domains and - * let the driver-defined callback functions sleep. - */ mutex_lock(&em_pd_mutex);
- for_each_cpu(cpu, span) { - /* Make sure we don't register again an existing domain. */ - if (READ_ONCE(per_cpu(em_data, cpu))) { - ret = -EEXIST; - goto unlock; + if (_is_cpu_device(dev)) { + if (!cpus) { + mutex_unlock(&em_pd_mutex); + dev_err(dev, "EM: invalid CPU mask\n"); + return -EINVAL; }
- /* - * All CPUs of a domain must have the same micro-architecture - * since they all share the same table. - */ - cap = arch_scale_cpu_capacity(cpu); - if (prev_cap && prev_cap != cap) { - pr_err("CPUs of %*pbl must have the same capacity\n", - cpumask_pr_args(span)); - ret = -EINVAL; - goto unlock; + /* Make sure we don't register domain for existing CPUs */ + if (em_cpus_pd_exist(cpus)) { + mutex_unlock(&em_pd_mutex); + return -EEXIST; + } + + for_each_cpu(cpu, cpus) { + /* + * All CPUs of a domain must have the same + * micro-architecture since they all share the same + * table. + */ + cap = arch_scale_cpu_capacity(cpu); + if (prev_cap && prev_cap != cap) { + dev_err(dev, "EM: CPUs of %*pbl must have the same capacity\n", + cpumask_pr_args(cpus)); + + mutex_unlock(&em_pd_mutex); + return -EINVAL; + } + prev_cap = cap; } - prev_cap = cap; }
- /* Create the performance domain and add it to the Energy Model. */ - pd = em_create_pd(span, nr_states, cb); + pd = em_create_pd(dev, nr_states, cb, cpus); if (!pd) { - ret = -EINVAL; - goto unlock; + mutex_unlock(&em_pd_mutex); + return -EINVAL; }
- for_each_cpu(cpu, span) { - /* - * The per-cpu array can be read concurrently from em_cpu_get(). - * The barrier enforces the ordering needed to make sure readers - * can only access well formed em_perf_domain structs. - */ - smp_store_release(per_cpu_ptr(&em_data, cpu), pd); + em_dev = kzalloc(sizeof(struct em_device), GFP_KERNEL); + if (!em_dev) { + ret = -ENOMEM; + goto free_pd; }
- pr_debug("Created perf domain %*pbl\n", cpumask_pr_args(span)); -unlock: + em_dev->id = ida_simple_get(&em_dev_ida, 0, 0, GFP_KERNEL); + if (em_dev->id < 0) { + ret = em_dev->id; + goto free_em_dev; + } + + em_dev->em_pd = pd; + em_dev->dev = dev; + + list_add(&em_dev->em_dev_list, &em_pd_dev_list); + + em_debug_create_pd(em_dev); + pr_debug("Created perf domain pd%d\n", em_dev->id); + mutex_unlock(&em_pd_mutex); + return 0;
+free_em_dev: + kfree(em_dev); +free_pd: + kfree(pd); + + mutex_unlock(&em_pd_mutex); return ret; } EXPORT_SYMBOL_GPL(em_register_perf_domain); + +void em_unregister_perf_domain(struct device *dev) +{ + struct em_device *em_dev, *tmp; + + if (IS_ERR_OR_NULL(dev)) + return; + + /* We don't support freeing CPU structures in hotplug */ + if (_is_cpu_device(dev)) + return; + + mutex_lock(&em_pd_mutex); + + list_for_each_entry_safe(em_dev, tmp, &em_pd_dev_list, em_dev_list) { + if (em_dev->dev != dev) + continue; + + list_del(&em_dev->em_dev_list); + + em_debug_remove_pd(em_dev); + + ida_simple_remove(&em_dev_ida, em_dev->id); + + em_destroy_pd(em_dev->em_pd); + kfree(em_dev); + } + + mutex_unlock(&em_pd_mutex); +} +EXPORT_SYMBOL_GPL(em_unregister_perf_domain); diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index f38ff5a335d3..b32d9b591d39 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -6245,7 +6245,7 @@ compute_energy(struct task_struct *p, int dst_cpu, struct perf_domain *pd) max_util = max(max_util, cpu_util); }
- return em_pd_energy(pd->em_pd, max_util, sum_util); + return em_cpu_energy(pd->em_pd, max_util, sum_util); }
/* diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index 00911884b7e7..322fbb2d0e50 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -285,7 +285,7 @@ static void perf_domain_debug(const struct cpumask *cpu_map, printk(KERN_CONT " pd%d:{ cpus=%*pbl nr_cstate=%d }", cpumask_first(perf_domain_span(pd)), cpumask_pr_args(perf_domain_span(pd)), - em_pd_nr_cap_states(pd->em_pd)); + em_pd_nr_perf_states(pd->em_pd)); pd = pd->next; }
@@ -390,7 +390,7 @@ static bool build_perf_domains(const struct cpumask *cpu_map) * complexity check. */ nr_pd++; - nr_cs += em_pd_nr_cap_states(pd->em_pd); + nr_cs += em_pd_nr_perf_states(pd->em_pd); }
/* Bail out if the Energy Model complexity is too high. */
Hi, One minor nit. Please see inline:
On 2/21/20 11:47 AM, Lukasz Luba wrote:
Add support of other devices into the Energy Model framework not only the CPUs. Change the interface to be more unified which can handle other devices as well.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com
Documentation/power/energy-model.rst | 133 ++++----
diff --git a/Documentation/power/energy-model.rst b/Documentation/power/energy-model.rst index 90a345d57ae9..7576820664e5 100644 --- a/Documentation/power/energy-model.rst +++ b/Documentation/power/energy-model.rst
@@ -85,13 +89,20 @@ API. 2.3 Accessing performance domains ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+There is two API functions which provide the access to the energy model:
There are two
+em_cpu_get() which takes CPU id as an argument and em_pd_get() with device +pointer as an argument. It depends on the subsystem which interface it is +going to use, but in case of CPU devices both functions return the same +performance domain.
cheers.
Hi Randy,
Thank you for taking the time to look into this patch.
On 2/22/20 12:42 AM, Randy Dunlap wrote:
Hi, One minor nit. Please see inline:
On 2/21/20 11:47 AM, Lukasz Luba wrote:
Add support of other devices into the Energy Model framework not only the CPUs. Change the interface to be more unified which can handle other devices as well.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com
Documentation/power/energy-model.rst | 133 ++++----
diff --git a/Documentation/power/energy-model.rst b/Documentation/power/energy-model.rst index 90a345d57ae9..7576820664e5 100644 --- a/Documentation/power/energy-model.rst +++ b/Documentation/power/energy-model.rst
@@ -85,13 +89,20 @@ API. 2.3 Accessing performance domains ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+There is two API functions which provide the access to the energy model:
There are two
Indeed a mistake, I will fix it.
+em_cpu_get() which takes CPU id as an argument and em_pd_get() with device +pointer as an argument. It depends on the subsystem which interface it is +going to use, but in case of CPU devices both functions return the same +performance domain.
cheers.
Regards, Lukasz
Drop the CPU specific interface with cpumask and switch to struct device. The Energy Model framework supports both: CPUs and devfreq devices. The new interface provides easy way to create a Energy Model (EM), which then might be used in i.e. thermal subsystem.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com --- drivers/cpufreq/cpufreq-dt.c | 2 +- drivers/cpufreq/imx6q-cpufreq.c | 2 +- drivers/cpufreq/mediatek-cpufreq.c | 2 +- drivers/cpufreq/omap-cpufreq.c | 2 +- drivers/cpufreq/qcom-cpufreq-hw.c | 2 +- drivers/cpufreq/scpi-cpufreq.c | 2 +- drivers/cpufreq/vexpress-spc-cpufreq.c | 2 +- drivers/opp/of.c | 40 ++++++++++++++------------ include/linux/pm_opp.h | 14 +++++++-- 9 files changed, 40 insertions(+), 28 deletions(-)
diff --git a/drivers/cpufreq/cpufreq-dt.c b/drivers/cpufreq/cpufreq-dt.c index d2b5f062a07b..bb5d43f8b775 100644 --- a/drivers/cpufreq/cpufreq-dt.c +++ b/drivers/cpufreq/cpufreq-dt.c @@ -275,7 +275,7 @@ static int cpufreq_init(struct cpufreq_policy *policy) policy->cpuinfo.transition_latency = transition_latency; policy->dvfs_possible_from_any_cpu = true;
- dev_pm_opp_of_register_em(policy->cpus); + dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
return 0;
diff --git a/drivers/cpufreq/imx6q-cpufreq.c b/drivers/cpufreq/imx6q-cpufreq.c index 1fcbbd53a48a..af1f51f6f2b8 100644 --- a/drivers/cpufreq/imx6q-cpufreq.c +++ b/drivers/cpufreq/imx6q-cpufreq.c @@ -193,7 +193,7 @@ 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(policy->cpus); + dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
return 0; } diff --git a/drivers/cpufreq/mediatek-cpufreq.c b/drivers/cpufreq/mediatek-cpufreq.c index 0c98dd08273d..7d1212c9b7c8 100644 --- a/drivers/cpufreq/mediatek-cpufreq.c +++ b/drivers/cpufreq/mediatek-cpufreq.c @@ -448,7 +448,7 @@ static int mtk_cpufreq_init(struct cpufreq_policy *policy) policy->driver_data = info; policy->clk = info->cpu_clk;
- dev_pm_opp_of_register_em(policy->cpus); + dev_pm_opp_of_register_em(info->cpu_dev, policy->cpus);
return 0; } diff --git a/drivers/cpufreq/omap-cpufreq.c b/drivers/cpufreq/omap-cpufreq.c index 8d14b42a8c6f..3694bb030df3 100644 --- a/drivers/cpufreq/omap-cpufreq.c +++ b/drivers/cpufreq/omap-cpufreq.c @@ -131,7 +131,7 @@ 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(policy->cpus); + dev_pm_opp_of_register_em(mpu_dev, policy->cpus);
return 0; } diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c index fc92a8842e25..0a04b6f03b9a 100644 --- a/drivers/cpufreq/qcom-cpufreq-hw.c +++ b/drivers/cpufreq/qcom-cpufreq-hw.c @@ -238,7 +238,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy) goto error; }
- dev_pm_opp_of_register_em(policy->cpus); + dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
policy->fast_switch_possible = true;
diff --git a/drivers/cpufreq/scpi-cpufreq.c b/drivers/cpufreq/scpi-cpufreq.c index 20d1f85d5f5a..b0f5388b8854 100644 --- a/drivers/cpufreq/scpi-cpufreq.c +++ b/drivers/cpufreq/scpi-cpufreq.c @@ -167,7 +167,7 @@ static int scpi_cpufreq_init(struct cpufreq_policy *policy)
policy->fast_switch_possible = false;
- dev_pm_opp_of_register_em(policy->cpus); + dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
return 0;
diff --git a/drivers/cpufreq/vexpress-spc-cpufreq.c b/drivers/cpufreq/vexpress-spc-cpufreq.c index 83c85d3d67e3..4e8b1dee7c9a 100644 --- a/drivers/cpufreq/vexpress-spc-cpufreq.c +++ b/drivers/cpufreq/vexpress-spc-cpufreq.c @@ -450,7 +450,7 @@ 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(policy->cpus); + dev_pm_opp_of_register_em(cpu_dev, policy->cpus);
if (is_bL_switching_enabled()) per_cpu(cpu_last_req_freq, policy->cpu) = diff --git a/drivers/opp/of.c b/drivers/opp/of.c index 0efd6cf6d023..5eb43453b0ee 100644 --- a/drivers/opp/of.c +++ b/drivers/opp/of.c @@ -1036,18 +1036,18 @@ EXPORT_SYMBOL_GPL(dev_pm_opp_get_of_node);
/* * Callback function provided to the Energy Model framework upon registration. - * This computes the power estimated by @CPU at @kHz if it is the frequency + * This computes the power estimated by @dev at @kHz if it is the frequency * of an existing OPP, or at the frequency of the first OPP above @kHz otherwise * (see dev_pm_opp_find_freq_ceil()). This function updates @kHz to the ceiled * frequency and @mW to the associated power. The power is estimated as - * P = C * V^2 * f with C being the CPU's capacitance and V and f respectively - * the voltage and frequency of the OPP. + * P = C * V^2 * f with C being the device's capacitance and V and f + * respectively the voltage and frequency of the OPP. * - * Returns -ENODEV if the CPU device cannot be found, -EINVAL if the power - * calculation failed because of missing parameters, 0 otherwise. + * Returns -EINVAL if the power calculation failed because of missing + * parameters, 0 otherwise. */ -static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz, - struct device *cpu_dev) +static int __maybe_unused _get_power(unsigned long *mW, unsigned long *kHz, + struct device *dev) { struct dev_pm_opp *opp; struct device_node *np; @@ -1056,7 +1056,7 @@ static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz, u64 tmp; int ret;
- np = of_node_get(cpu_dev->of_node); + np = of_node_get(dev->of_node); if (!np) return -EINVAL;
@@ -1066,7 +1066,7 @@ static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz, return -EINVAL;
Hz = *kHz * 1000; - opp = dev_pm_opp_find_freq_ceil(cpu_dev, &Hz); + opp = dev_pm_opp_find_freq_ceil(dev, &Hz); if (IS_ERR(opp)) return -EINVAL;
@@ -1086,28 +1086,30 @@ static int __maybe_unused _get_cpu_power(unsigned long *mW, unsigned long *kHz,
/** * dev_pm_opp_of_register_em() - Attempt to register an Energy Model - * @cpus : CPUs for which an Energy Model has to be registered + * @dev : Device for which an Energy Model has to be registered + * @cpus : CPUs for which an Energy Model has to be registered. For + * other type of devices it should be set to NULL. * * This checks whether the "dynamic-power-coefficient" devicetree property has * been specified, and tries to register an Energy Model with it if it has. + * Having this property means the voltages are known for OPPs and the EM + * might be calculated. */ -void dev_pm_opp_of_register_em(struct cpumask *cpus) +void dev_pm_opp_of_register_em(struct device *dev, struct cpumask *cpus) { - struct em_data_callback em_cb = EM_DATA_CB(_get_cpu_power); - int ret, nr_opp, cpu = cpumask_first(cpus); - struct device *cpu_dev; + struct em_data_callback em_cb = EM_DATA_CB(_get_power); struct device_node *np; + int ret, nr_opp; u32 cap;
- cpu_dev = get_cpu_device(cpu); - if (!cpu_dev) + if (!dev) return;
- nr_opp = dev_pm_opp_get_opp_count(cpu_dev); + nr_opp = dev_pm_opp_get_opp_count(dev); if (nr_opp <= 0) return;
- np = of_node_get(cpu_dev->of_node); + np = of_node_get(dev->of_node); if (!np) return;
@@ -1123,6 +1125,6 @@ void dev_pm_opp_of_register_em(struct cpumask *cpus) if (ret || !cap) return;
- em_register_perf_domain(cpu_dev, nr_opp, &em_cb, cpus); + em_register_perf_domain(dev, nr_opp, &em_cb, cpus); } EXPORT_SYMBOL_GPL(dev_pm_opp_of_register_em); diff --git a/include/linux/pm_opp.h b/include/linux/pm_opp.h index 747861816f4f..65099de79d0c 100644 --- a/include/linux/pm_opp.h +++ b/include/linux/pm_opp.h @@ -11,6 +11,7 @@ #ifndef __LINUX_OPP_H__ #define __LINUX_OPP_H__
+#include <linux/energy_model.h> #include <linux/err.h> #include <linux/notifier.h>
@@ -360,7 +361,11 @@ int dev_pm_opp_of_get_sharing_cpus(struct device *cpu_dev, struct cpumask *cpuma struct device_node *dev_pm_opp_of_get_opp_desc_node(struct device *dev); struct device_node *dev_pm_opp_get_of_node(struct dev_pm_opp *opp); int of_get_required_opp_performance_state(struct device_node *np, int index); -void dev_pm_opp_of_register_em(struct cpumask *cpus); +void dev_pm_opp_of_register_em(struct device *dev, struct cpumask *cpus); +static inline void dev_pm_opp_of_unregister_em(struct device *dev) +{ + em_unregister_perf_domain(dev); +} #else static inline int dev_pm_opp_of_add_table(struct device *dev) { @@ -400,7 +405,12 @@ static inline struct device_node *dev_pm_opp_get_of_node(struct dev_pm_opp *opp) return NULL; }
-static inline void dev_pm_opp_of_register_em(struct cpumask *cpus) +static inline void dev_pm_opp_of_register_em(struct device *dev, + struct cpumask *cpus) +{ +} + +static inline void dev_pm_opp_of_unregister_em(struct device *dev) { }
The overhauled Energy Model (EM) framework support also devfreq devices. The unified API interface of the EM can be used in the thermal subsystem to not duplicate code. The power table now is taken from EM structure and there is no need to maintain calculation for it locally. In case when the EM is not provided by the device a simple interface for cooling device is used.
There is also an improvement in code related to enabling/disabling OPPs, which prevents from race condition with devfreq governors.
[lkp: Reported the build warning] Reported-by: kbuild test robot lkp@intel.com Reviewed-by: Steven Rostedt (VMware) rostedt@goodmis.org # for tracing code Signed-off-by: Lukasz Luba lukasz.luba@arm.com --- drivers/thermal/devfreq_cooling.c | 425 ++++++++++++++---------------- include/linux/devfreq_cooling.h | 17 -- include/trace/events/thermal.h | 19 +- 3 files changed, 201 insertions(+), 260 deletions(-)
diff --git a/drivers/thermal/devfreq_cooling.c b/drivers/thermal/devfreq_cooling.c index a87d4fa031c8..36cc6ca654e1 100644 --- a/drivers/thermal/devfreq_cooling.c +++ b/drivers/thermal/devfreq_cooling.c @@ -1,17 +1,9 @@ +// SPDX-License-Identifier: GPL-2.0 /* * devfreq_cooling: Thermal cooling device implementation for devices using * devfreq * - * Copyright (C) 2014-2015 ARM Limited - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License version 2 as - * published by the Free Software Foundation. - * - * This program is distributed "as is" WITHOUT ANY WARRANTY of any - * kind, whether express or implied; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * Copyright (C) 2014-2020 ARM Limited * * TODO: * - If OPPs are added or removed after devfreq cooling has @@ -20,6 +12,7 @@
#include <linux/devfreq.h> #include <linux/devfreq_cooling.h> +#include <linux/energy_model.h> #include <linux/export.h> #include <linux/idr.h> #include <linux/slab.h> @@ -39,37 +32,56 @@ static DEFINE_IDA(devfreq_ida); * @cdev: Pointer to associated thermal cooling device. * @devfreq: Pointer to associated devfreq device. * @cooling_state: Current cooling state. - * @power_table: Pointer to table with maximum power draw for each - * cooling state. State is the index into the table, and - * the power is in mW. - * @freq_table: Pointer to a table with the frequencies sorted in descending - * order. You can index the table by cooling device state - * @freq_table_size: Size of the @freq_table and @power_table - * @power_ops: Pointer to devfreq_cooling_power, used to generate the - * @power_table. + * @freq_table: Pointer to a table with the frequencies. + * @max_level: It is the last index, that is, one less than the number of the + * OPPs + * @power_ops: Pointer to devfreq_cooling_power, a more precised model. * @res_util: Resource utilization scaling factor for the power. * It is multiplied by 100 to minimize the error. It is used * for estimation of the power budget instead of using * 'utilization' (which is 'busy_time / 'total_time'). - * The 'res_util' range is from 100 to (power_table[state] * 100) - * for the corresponding 'state'. - * @capped_state: index to cooling state with in dynamic power budget + * The 'res_util' range is from 100 to power * 100 for the + * corresponding 'state'. + * @em: Energy Model which represents the associated Devfreq device */ struct devfreq_cooling_device { int id; struct thermal_cooling_device *cdev; struct devfreq *devfreq; unsigned long cooling_state; - u32 *power_table; u32 *freq_table; - size_t freq_table_size; + size_t max_level; struct devfreq_cooling_power *power_ops; u32 res_util; - int capped_state; + struct em_perf_domain *em; };
+static unsigned long _find_freq_for_state(struct devfreq_cooling_device *dfc, + unsigned long cdev_state) +{ + unsigned long freq; + + if (dfc->em) { + freq = dfc->em->table[dfc->max_level - cdev_state].frequency; + /* Energy Model frequencies are in kHz */ + return freq * 1000; + } else { + return dfc->freq_table[cdev_state]; + } +} + +static void _update_devfreq_max_freq(struct devfreq *df, unsigned long max_freq) +{ + dev_dbg(df->dev.parent, "devfreq_cooling: max_freq=%luHz\n", max_freq); + + mutex_lock(&df->lock); + df->scaling_max_freq = max_freq; + update_devfreq(df); + mutex_unlock(&df->lock); +} + /** - * partition_enable_opps() - disable all opps above a given state + * partition_enable_opps() - disable all OPPs above a given state * @dfc: Pointer to devfreq we are operating on * @cdev_state: cooling device state we're setting * @@ -79,16 +91,33 @@ struct devfreq_cooling_device { static int partition_enable_opps(struct devfreq_cooling_device *dfc, unsigned long cdev_state) { - int i; struct device *dev = dfc->devfreq->dev.parent; + bool want_enable, available; + struct dev_pm_opp *opp; + unsigned long freq; + int i, ret;
- for (i = 0; i < dfc->freq_table_size; i++) { - struct dev_pm_opp *opp; - int ret = 0; - unsigned int freq = dfc->freq_table[i]; - bool want_enable = i >= cdev_state ? true : false; + /* + * Avoid race with devfreq governors trying to use OPPs which are + * going to be disabled. Update devfreq upfront when previous cooling + * state had higher frequency or do it later in opposite case. + */ + freq = _find_freq_for_state(dfc, cdev_state); + + if (dfc->cooling_state < cdev_state) + _update_devfreq_max_freq(dfc->devfreq, freq);
- opp = dev_pm_opp_find_freq_exact(dev, freq, !want_enable); + dev_dbg(dev, "devfreq_cooling: updating OPPs\n"); + for (i = 0; i <= dfc->max_level; i++) { + available = (i >= dfc->cooling_state); + want_enable = (i >= cdev_state); + + if (available == want_enable) + continue; + + freq = _find_freq_for_state(dfc, i); + + opp = dev_pm_opp_find_freq_exact(dev, freq, available);
if (PTR_ERR(opp) == -ERANGE) continue; @@ -106,6 +135,9 @@ static int partition_enable_opps(struct devfreq_cooling_device *dfc, return ret; }
+ if (dfc->cooling_state > cdev_state) + _update_devfreq_max_freq(dfc->devfreq, freq); + return 0; }
@@ -114,7 +146,7 @@ static int devfreq_cooling_get_max_state(struct thermal_cooling_device *cdev, { struct devfreq_cooling_device *dfc = cdev->devdata;
- *state = dfc->freq_table_size - 1; + *state = dfc->max_level;
return 0; } @@ -142,7 +174,7 @@ static int devfreq_cooling_set_cur_state(struct thermal_cooling_device *cdev,
dev_dbg(dev, "Setting cooling state %lu\n", state);
- if (state >= dfc->freq_table_size) + if (state > dfc->max_level) return -EINVAL;
ret = partition_enable_opps(dfc, state); @@ -155,11 +187,11 @@ static int devfreq_cooling_set_cur_state(struct thermal_cooling_device *cdev, }
/** - * freq_get_state() - get the cooling state corresponding to a frequency + * freq_get_state() - get the performance index corresponding to a frequency * @dfc: Pointer to devfreq cooling device - * @freq: frequency in Hz + * @freq: frequency in kHz * - * Return: the cooling state associated with the @freq, or + * Return: the performance index associated with the @freq, or * THERMAL_CSTATE_INVALID if it wasn't found. */ static unsigned long @@ -167,8 +199,8 @@ freq_get_state(struct devfreq_cooling_device *dfc, unsigned long freq) { int i;
- for (i = 0; i < dfc->freq_table_size; i++) { - if (dfc->freq_table[i] == freq) + for (i = 0; i <= dfc->max_level; i++) { + if (dfc->em->table[i].frequency == freq) return i; }
@@ -203,132 +235,79 @@ static unsigned long get_voltage(struct devfreq *df, unsigned long freq) return voltage; }
-/** - * get_static_power() - calculate the static power - * @dfc: Pointer to devfreq cooling device - * @freq: Frequency in Hz - * - * Calculate the static power in milliwatts using the supplied - * get_static_power(). The current voltage is calculated using the - * OPP library. If no get_static_power() was supplied, assume the - * static power is negligible. - */ -static unsigned long -get_static_power(struct devfreq_cooling_device *dfc, unsigned long freq) -{ - struct devfreq *df = dfc->devfreq; - unsigned long voltage; - - if (!dfc->power_ops->get_static_power) - return 0; - - voltage = get_voltage(df, freq); - - if (voltage == 0) - return 0; - - return dfc->power_ops->get_static_power(df, voltage); -} - -/** - * get_dynamic_power - calculate the dynamic power - * @dfc: Pointer to devfreq cooling device - * @freq: Frequency in Hz - * @voltage: Voltage in millivolts - * - * Calculate the dynamic power in milliwatts consumed by the device at - * frequency @freq and voltage @voltage. If the get_dynamic_power() - * was supplied as part of the devfreq_cooling_power struct, then that - * function is used. Otherwise, a simple power model (Pdyn = Coeff * - * Voltage^2 * Frequency) is used. - */ -static unsigned long -get_dynamic_power(struct devfreq_cooling_device *dfc, unsigned long freq, - unsigned long voltage) +static void dfc_em_get_requested_power(struct em_perf_domain *em, + struct devfreq_dev_status *status, + u32 *power, int em_perf_idx) { - u64 power; - u32 freq_mhz; - struct devfreq_cooling_power *dfc_power = dfc->power_ops; - - if (dfc_power->get_dynamic_power) - return dfc_power->get_dynamic_power(dfc->devfreq, freq, - voltage); + unsigned long total_time; + u64 current_power;
- freq_mhz = freq / 1000000; - power = (u64)dfc_power->dyn_power_coeff * freq_mhz * voltage * voltage; - do_div(power, 1000000000); + total_time = status->total_time ?: 1;
- return power; -} + current_power = em->table[em_perf_idx].power;
+ /* Scale power for utilization */ + current_power *= status->busy_time; + current_power = do_div(current_power, total_time);
-static inline unsigned long get_total_power(struct devfreq_cooling_device *dfc, - unsigned long freq, - unsigned long voltage) -{ - return get_static_power(dfc, freq) + get_dynamic_power(dfc, freq, - voltage); + *power = current_power; }
- static int devfreq_cooling_get_requested_power(struct thermal_cooling_device *cdev, struct thermal_zone_device *tz, u32 *power) { struct devfreq_cooling_device *dfc = cdev->devdata; struct devfreq *df = dfc->devfreq; - struct devfreq_dev_status *status = &df->last_status; - unsigned long state; - unsigned long freq = status->current_frequency; - unsigned long voltage; - u32 dyn_power = 0; - u32 static_power = 0; + struct devfreq_dev_status status; + unsigned long voltage, freq; + unsigned long em_perf_idx; int res;
- state = freq_get_state(dfc, freq); - if (state == THERMAL_CSTATE_INVALID) { - res = -EAGAIN; - goto fail; - } + mutex_lock(&df->lock); + res = df->profile->get_dev_status(df->dev.parent, &status); + mutex_unlock(&df->lock); + if (res) + return res;
- if (dfc->power_ops->get_real_power) { + freq = status.current_frequency; + + /* Energy Model frequencies are in kHz */ + em_perf_idx = freq_get_state(dfc, freq / 1000); + if (em_perf_idx == THERMAL_CSTATE_INVALID) + return -EAGAIN; + + /* + * If a more sophisticated cooling device model was not provided by the + * driver, use simple Energy Model power calculation. + */ + if (!dfc->power_ops || !dfc->power_ops->get_real_power) { + dfc_em_get_requested_power(dfc->em, &status, power, + em_perf_idx); + } else { voltage = get_voltage(df, freq); if (voltage == 0) { - res = -EINVAL; - goto fail; + dfc->res_util = SCALE_ERROR_MITIGATION; + return -EINVAL; }
res = dfc->power_ops->get_real_power(df, power, freq, voltage); if (!res) { - state = dfc->capped_state; - dfc->res_util = dfc->power_table[state]; + dfc->res_util = dfc->em->table[em_perf_idx].power; dfc->res_util *= SCALE_ERROR_MITIGATION;
if (*power > 1) dfc->res_util /= *power; } else { - goto fail; + /* It is safe to set max in this case */ + dfc->res_util = SCALE_ERROR_MITIGATION; + return res; } - } else { - dyn_power = dfc->power_table[state]; - - /* Scale dynamic power for utilization */ - dyn_power *= status->busy_time; - dyn_power /= status->total_time; - /* Get static power */ - static_power = get_static_power(dfc, freq); - - *power = dyn_power + static_power; }
- trace_thermal_power_devfreq_get_power(cdev, status, freq, dyn_power, - static_power, *power); + trace_thermal_power_devfreq_get_power(cdev, &status, freq, *power);
return 0; -fail: - /* It is safe to set max in this case */ - dfc->res_util = SCALE_ERROR_MITIGATION; - return res; }
static int devfreq_cooling_state2power(struct thermal_cooling_device *cdev, @@ -337,16 +316,14 @@ static int devfreq_cooling_state2power(struct thermal_cooling_device *cdev, u32 *power) { struct devfreq_cooling_device *dfc = cdev->devdata; - unsigned long freq; - u32 static_power; + int idx;
- if (state >= dfc->freq_table_size) + if (state > dfc->max_level) return -EINVAL;
- freq = dfc->freq_table[state]; - static_power = get_static_power(dfc, freq); + idx = dfc->max_level - state; + *power = dfc->em->table[idx].power;
- *power = dfc->power_table[state] + static_power; return 0; }
@@ -359,36 +336,34 @@ static int devfreq_cooling_power2state(struct thermal_cooling_device *cdev, struct devfreq_dev_status *status = &df->last_status; unsigned long freq = status->current_frequency; unsigned long busy_time; - s32 dyn_power; - u32 static_power; - s32 est_power; + u64 est_power; int i;
- if (dfc->power_ops->get_real_power) { - /* Scale for resource utilization */ - est_power = power * dfc->res_util; - est_power /= SCALE_ERROR_MITIGATION; - } else { - static_power = get_static_power(dfc, freq); - - dyn_power = power - static_power; - dyn_power = dyn_power > 0 ? dyn_power : 0; - - /* Scale dynamic power for utilization */ + /* + * Scale for resource utilization. Use simple Energy Model power + * calculation if a more sophisticated cooling device model does + * not exist. + */ + if (!dfc->power_ops || !dfc->power_ops->get_real_power) { busy_time = status->busy_time ?: 1; - est_power = (dyn_power * status->total_time) / busy_time; + + est_power = power * status->total_time; + est_power = do_div(est_power, busy_time); + } else { + est_power = power * dfc->res_util; + est_power = do_div(est_power, SCALE_ERROR_MITIGATION); }
/* * Find the first cooling state that is within the power - * budget for dynamic power. + * budget. The EM power table is sorted ascending. */ - for (i = 0; i < dfc->freq_table_size - 1; i++) - if (est_power >= dfc->power_table[i]) + for (i = dfc->max_level; i > 0; i--) + if (est_power >= dfc->em->table[i].power) break;
- *state = i; - dfc->capped_state = i; + *state = dfc->max_level - i; + trace_thermal_power_devfreq_limit(cdev, freq, *state, power); return 0; } @@ -400,91 +375,43 @@ static struct thermal_cooling_device_ops devfreq_cooling_ops = { };
/** - * devfreq_cooling_gen_tables() - Generate power and freq tables. - * @dfc: Pointer to devfreq cooling device. - * - * Generate power and frequency tables: the power table hold the - * device's maximum power usage at each cooling state (OPP). The - * static and dynamic power using the appropriate voltage and - * frequency for the state, is acquired from the struct - * devfreq_cooling_power, and summed to make the maximum power draw. - * - * The frequency table holds the frequencies in descending order. - * That way its indexed by cooling device state. + * devfreq_cooling_gen_tables() - Generate frequency table. + * @dfc: Pointer to devfreq cooling device. + * @num_opps: Number of OPPs * - * The tables are malloced, and pointers put in dfc. They must be - * freed when unregistering the devfreq cooling device. + * Generate frequency table which holds the frequencies in descending + * order. That way its indexed by cooling device state. This is for + * compatibility with drivers which do not register Energy Model. * * Return: 0 on success, negative error code on failure. */ -static int devfreq_cooling_gen_tables(struct devfreq_cooling_device *dfc) +static int devfreq_cooling_gen_tables(struct devfreq_cooling_device *dfc, + int num_opps) { struct devfreq *df = dfc->devfreq; struct device *dev = df->dev.parent; - int ret, num_opps; unsigned long freq; - u32 *power_table = NULL; - u32 *freq_table; int i;
- num_opps = dev_pm_opp_get_opp_count(dev); - - if (dfc->power_ops) { - power_table = kcalloc(num_opps, sizeof(*power_table), - GFP_KERNEL); - if (!power_table) - return -ENOMEM; - } - - freq_table = kcalloc(num_opps, sizeof(*freq_table), + dfc->freq_table = kcalloc(num_opps, sizeof(*dfc->freq_table), GFP_KERNEL); - if (!freq_table) { - ret = -ENOMEM; - goto free_power_table; - } + if (!dfc->freq_table) + return -ENOMEM;
for (i = 0, freq = ULONG_MAX; i < num_opps; i++, freq--) { - unsigned long power, voltage; struct dev_pm_opp *opp;
opp = dev_pm_opp_find_freq_floor(dev, &freq); if (IS_ERR(opp)) { - ret = PTR_ERR(opp); - goto free_tables; + kfree(dfc->freq_table); + return PTR_ERR(opp); }
- voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */ dev_pm_opp_put(opp); - - if (dfc->power_ops) { - if (dfc->power_ops->get_real_power) - power = get_total_power(dfc, freq, voltage); - else - power = get_dynamic_power(dfc, freq, voltage); - - dev_dbg(dev, "Power table: %lu MHz @ %lu mV: %lu = %lu mW\n", - freq / 1000000, voltage, power, power); - - power_table[i] = power; - } - - freq_table[i] = freq; + dfc->freq_table[i] = freq; }
- if (dfc->power_ops) - dfc->power_table = power_table; - - dfc->freq_table = freq_table; - dfc->freq_table_size = num_opps; - return 0; - -free_tables: - kfree(freq_table); -free_power_table: - kfree(power_table); - - return ret; }
/** @@ -509,7 +436,7 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df, struct thermal_cooling_device *cdev; struct devfreq_cooling_device *dfc; char dev_name[THERMAL_NAME_LENGTH]; - int err; + int err, num_opps;
dfc = kzalloc(sizeof(*dfc), GFP_KERNEL); if (!dfc) @@ -517,22 +444,38 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
dfc->devfreq = df;
- if (dfc_power) { - dfc->power_ops = dfc_power; - + dfc->em = em_pd_get(df->dev.parent); + if (dfc->em) { devfreq_cooling_ops.get_requested_power = devfreq_cooling_get_requested_power; devfreq_cooling_ops.state2power = devfreq_cooling_state2power; devfreq_cooling_ops.power2state = devfreq_cooling_power2state; + + dfc->power_ops = dfc_power; + + num_opps = em_pd_nr_perf_states(dfc->em); + } else { + /* Backward compatibility for drivers which do not use IPA */ + dev_dbg(df->dev.parent, "missing EM for cooling device\n"); + + num_opps = dev_pm_opp_get_opp_count(df->dev.parent); + + err = devfreq_cooling_gen_tables(dfc, num_opps); + if (err) + goto free_dfc; }
- err = devfreq_cooling_gen_tables(dfc); - if (err) + if (num_opps <= 0) { + err = -EINVAL; goto free_dfc; + } + + /* max_level is an index, not a counter */ + dfc->max_level = num_opps - 1;
err = ida_simple_get(&devfreq_ida, 0, 0, GFP_KERNEL); if (err < 0) - goto free_tables; + goto free_table; dfc->id = err;
snprintf(dev_name, sizeof(dev_name), "thermal-devfreq-%d", dfc->id); @@ -553,8 +496,7 @@ of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
release_ida: ida_simple_remove(&devfreq_ida, dfc->id); -free_tables: - kfree(dfc->power_table); +free_table: kfree(dfc->freq_table); free_dfc: kfree(dfc); @@ -582,7 +524,24 @@ EXPORT_SYMBOL_GPL(of_devfreq_cooling_register); */ struct thermal_cooling_device *devfreq_cooling_register(struct devfreq *df) { - return of_devfreq_cooling_register(NULL, df); + struct thermal_cooling_device *dfc; + struct device_node *np = NULL; + struct device *dev; + + if (IS_ERR_OR_NULL(df)) + return ERR_PTR(-EINVAL); + + dev = df->dev.parent; + + if (dev && dev->of_node) + np = of_node_get(dev->of_node); + + dfc = of_devfreq_cooling_register(np, df); + + if (np) + of_node_put(np); + + return dfc; } EXPORT_SYMBOL_GPL(devfreq_cooling_register);
@@ -594,14 +553,14 @@ void devfreq_cooling_unregister(struct thermal_cooling_device *cdev) { struct devfreq_cooling_device *dfc;
- if (!cdev) + if (IS_ERR_OR_NULL(cdev)) return;
dfc = cdev->devdata;
thermal_cooling_device_unregister(dfc->cdev); ida_simple_remove(&devfreq_ida, dfc->id); - kfree(dfc->power_table); + kfree(dfc->freq_table);
kfree(dfc); diff --git a/include/linux/devfreq_cooling.h b/include/linux/devfreq_cooling.h index 4635f95000a4..5e11bbc3ec58 100644 --- a/include/linux/devfreq_cooling.h +++ b/include/linux/devfreq_cooling.h @@ -23,17 +23,6 @@
/** * struct devfreq_cooling_power - Devfreq cooling power ops - * @get_static_power: Take voltage, in mV, and return the static power - * in mW. If NULL, the static power is assumed - * to be 0. - * @get_dynamic_power: Take voltage, in mV, and frequency, in HZ, and - * return the dynamic power draw in mW. If NULL, - * a simple power model is used. - * @dyn_power_coeff: Coefficient for the simple dynamic power model in - * mW/(MHz mV mV). - * If get_dynamic_power() is NULL, then the - * dynamic power is calculated as - * @dyn_power_coeff * frequency * voltage^2 * @get_real_power: When this is set, the framework uses it to ask the * device driver for the actual power. * Some devices have more sophisticated methods @@ -53,14 +42,8 @@ * max total (static + dynamic) power value for each OPP. */ struct devfreq_cooling_power { - unsigned long (*get_static_power)(struct devfreq *devfreq, - unsigned long voltage); - unsigned long (*get_dynamic_power)(struct devfreq *devfreq, - unsigned long freq, - unsigned long voltage); int (*get_real_power)(struct devfreq *df, u32 *power, unsigned long freq, unsigned long voltage); - unsigned long dyn_power_coeff; };
#ifdef CONFIG_DEVFREQ_THERMAL diff --git a/include/trace/events/thermal.h b/include/trace/events/thermal.h index 135e5421f003..8a5f04888abd 100644 --- a/include/trace/events/thermal.h +++ b/include/trace/events/thermal.h @@ -153,31 +153,30 @@ TRACE_EVENT(thermal_power_cpu_limit, TRACE_EVENT(thermal_power_devfreq_get_power, TP_PROTO(struct thermal_cooling_device *cdev, struct devfreq_dev_status *status, unsigned long freq, - u32 dynamic_power, u32 static_power, u32 power), + u32 power),
- TP_ARGS(cdev, status, freq, dynamic_power, static_power, power), + TP_ARGS(cdev, status, freq, power),
TP_STRUCT__entry( __string(type, cdev->type ) __field(unsigned long, freq ) - __field(u32, load ) - __field(u32, dynamic_power ) - __field(u32, static_power ) + __field(u32, busy_time) + __field(u32, total_time) __field(u32, power) ),
TP_fast_assign( __assign_str(type, cdev->type); __entry->freq = freq; - __entry->load = (100 * status->busy_time) / status->total_time; - __entry->dynamic_power = dynamic_power; - __entry->static_power = static_power; + __entry->busy_time = status->busy_time; + __entry->total_time = status->total_time; __entry->power = power; ),
- TP_printk("type=%s freq=%lu load=%u dynamic_power=%u static_power=%u power=%u", + TP_printk("type=%s freq=%lu load=%u power=%u", __get_str(type), __entry->freq, - __entry->load, __entry->dynamic_power, __entry->static_power, + __entry->total_time == 0 ? 0 : + (100 * __entry->busy_time) / __entry->total_time, __entry->power) );
Add device to the Energy Model framework. It will create a dedicated and unified data structures used i.e. in the thermal framework. The power model used in dev_pm_opp subsystem is simplified and created based on DT 'dynamic-power-coefficient', volatage and frequency. It is similar to the CPU model used in Energy Aware Scheduler.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com --- drivers/gpu/drm/panfrost/panfrost_devfreq.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/drivers/gpu/drm/panfrost/panfrost_devfreq.c b/drivers/gpu/drm/panfrost/panfrost_devfreq.c index 413987038fbf..d527a5113950 100644 --- a/drivers/gpu/drm/panfrost/panfrost_devfreq.c +++ b/drivers/gpu/drm/panfrost/panfrost_devfreq.c @@ -105,6 +105,8 @@ int panfrost_devfreq_init(struct panfrost_device *pfdev) } pfdev->devfreq.devfreq = devfreq;
+ dev_pm_opp_of_register_em(dev, NULL); + cooling = of_devfreq_cooling_register(dev->of_node, devfreq); if (IS_ERR(cooling)) DRM_DEV_INFO(dev, "Failed to register cooling device\n"); @@ -118,6 +120,7 @@ void panfrost_devfreq_fini(struct panfrost_device *pfdev) { if (pfdev->devfreq.cooling) devfreq_cooling_unregister(pfdev->devfreq.cooling); + dev_pm_opp_of_unregister_em(&pfdev->pdev->dev); dev_pm_opp_of_remove_table(&pfdev->pdev->dev); }
On Fri, Feb 21, 2020 at 1:48 PM Lukasz Luba lukasz.luba@arm.com wrote:
Add device to the Energy Model framework. It will create a dedicated and unified data structures used i.e. in the thermal framework. The power model used in dev_pm_opp subsystem is simplified and created based on DT 'dynamic-power-coefficient', volatage and frequency. It is
typo.
similar to the CPU model used in Energy Aware Scheduler.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com
drivers/gpu/drm/panfrost/panfrost_devfreq.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/drivers/gpu/drm/panfrost/panfrost_devfreq.c b/drivers/gpu/drm/panfrost/panfrost_devfreq.c index 413987038fbf..d527a5113950 100644 --- a/drivers/gpu/drm/panfrost/panfrost_devfreq.c +++ b/drivers/gpu/drm/panfrost/panfrost_devfreq.c @@ -105,6 +105,8 @@ int panfrost_devfreq_init(struct panfrost_device *pfdev) } pfdev->devfreq.devfreq = devfreq;
dev_pm_opp_of_register_em(dev, NULL);
Can't fail?
cooling = of_devfreq_cooling_register(dev->of_node, devfreq); if (IS_ERR(cooling)) DRM_DEV_INFO(dev, "Failed to register cooling device\n");@@ -118,6 +120,7 @@ void panfrost_devfreq_fini(struct panfrost_device *pfdev) { if (pfdev->devfreq.cooling) devfreq_cooling_unregister(pfdev->devfreq.cooling);
dev_pm_opp_of_unregister_em(&pfdev->pdev->dev); dev_pm_opp_of_remove_table(&pfdev->pdev->dev);
Does it make sense to keep this (and the registration side) as separate calls? Perhaps there's some ordering requirement with everything between dev_pm_opp_of_add_table() and dev_pm_opp_of_register_em()?
While you're just adding 2 lines, it seems there's a lot of complexity exposed to the driver just to initialize devfreq/opp.
Rob
Hi Rob,
On 2/25/20 8:57 PM, Rob Herring wrote:
On Fri, Feb 21, 2020 at 1:48 PM Lukasz Luba lukasz.luba@arm.com wrote:
Add device to the Energy Model framework. It will create a dedicated and unified data structures used i.e. in the thermal framework. The power model used in dev_pm_opp subsystem is simplified and created based on DT 'dynamic-power-coefficient', volatage and frequency. It is
typo.
I'll fix it.
similar to the CPU model used in Energy Aware Scheduler.
Signed-off-by: Lukasz Luba lukasz.luba@arm.com
drivers/gpu/drm/panfrost/panfrost_devfreq.c | 3 +++ 1 file changed, 3 insertions(+)
diff --git a/drivers/gpu/drm/panfrost/panfrost_devfreq.c b/drivers/gpu/drm/panfrost/panfrost_devfreq.c index 413987038fbf..d527a5113950 100644 --- a/drivers/gpu/drm/panfrost/panfrost_devfreq.c +++ b/drivers/gpu/drm/panfrost/panfrost_devfreq.c @@ -105,6 +105,8 @@ int panfrost_devfreq_init(struct panfrost_device *pfdev) } pfdev->devfreq.devfreq = devfreq;
dev_pm_opp_of_register_em(dev, NULL);Can't fail?
Yes, it can fail but the function does not return anything. It can easily fail, it's looking for "dynamic-power-coefficient" in the device node. The DT binding for the devfreq devices would also be good to add..
I would have to probably change it into returning 'int' and modify all old cpufreq drivers.
cooling = of_devfreq_cooling_register(dev->of_node, devfreq); if (IS_ERR(cooling)) DRM_DEV_INFO(dev, "Failed to register cooling device\n");@@ -118,6 +120,7 @@ void panfrost_devfreq_fini(struct panfrost_device *pfdev) { if (pfdev->devfreq.cooling) devfreq_cooling_unregister(pfdev->devfreq.cooling);
dev_pm_opp_of_unregister_em(&pfdev->pdev->dev); dev_pm_opp_of_remove_table(&pfdev->pdev->dev);Does it make sense to keep this (and the registration side) as separate calls? Perhaps there's some ordering requirement with everything between dev_pm_opp_of_add_table() and dev_pm_opp_of_register_em()?
Yes, dev_pm_opp_of_register_em() uses em_data_callback which operates on OPPs to calculate power values and costs, so the the OPP table should be already there.
While you're just adding 2 lines, it seems there's a lot of complexity exposed to the driver just to initialize devfreq/opp.
It depends, for example devfreq devices like buses would likely never use the energy model. Potential clients would be GPUs, DSPs, ISPs.
Could you help me with defining a DT binding for this "dynamic-power-coefficient" entry? It could be used in different types of devices. Should it be placed in each of these devices documentation file, or in some one common file?
Thank you for your comments.
Regards, Lukasz
On 26/02/2020 10:06 am, Lukasz Luba wrote: [...]
@@ -118,6 +120,7 @@ void panfrost_devfreq_fini(struct panfrost_device *pfdev) { if (pfdev->devfreq.cooling) devfreq_cooling_unregister(pfdev->devfreq.cooling); + dev_pm_opp_of_unregister_em(&pfdev->pdev->dev); dev_pm_opp_of_remove_table(&pfdev->pdev->dev);
Does it make sense to keep this (and the registration side) as separate calls? Perhaps there's some ordering requirement with everything between dev_pm_opp_of_add_table() and dev_pm_opp_of_register_em()?
Yes, dev_pm_opp_of_register_em() uses em_data_callback which operates on OPPs to calculate power values and costs, so the the OPP table should be already there.
While you're just adding 2 lines, it seems there's a lot of complexity exposed to the driver just to initialize devfreq/opp.
It depends, for example devfreq devices like buses would likely never use the energy model. Potential clients would be GPUs, DSPs, ISPs.
Still, it seems less than ideal for every client to have to remember to make all these individual calls, all in the right order (especially when it comes to undoing them in failure paths).
I haven't quite grasped whether the energy model is conceptually "owned" by the OPP table or by the cooling device, but either way it would seem to be a much nicer API if there were simply an additional "with energy model" variant of the registration call, and the standard removal call just automatically cleaned up an energy model as well if one was present.
Robin.
On 2/26/20 1:55 PM, Robin Murphy wrote:
On 26/02/2020 10:06 am, Lukasz Luba wrote: [...]
@@ -118,6 +120,7 @@ void panfrost_devfreq_fini(struct panfrost_device *pfdev) { if (pfdev->devfreq.cooling) devfreq_cooling_unregister(pfdev->devfreq.cooling); + dev_pm_opp_of_unregister_em(&pfdev->pdev->dev); dev_pm_opp_of_remove_table(&pfdev->pdev->dev);
Does it make sense to keep this (and the registration side) as separate calls? Perhaps there's some ordering requirement with everything between dev_pm_opp_of_add_table() and dev_pm_opp_of_register_em()?
Yes, dev_pm_opp_of_register_em() uses em_data_callback which operates on OPPs to calculate power values and costs, so the the OPP table should be already there.
While you're just adding 2 lines, it seems there's a lot of complexity exposed to the driver just to initialize devfreq/opp.
It depends, for example devfreq devices like buses would likely never use the energy model. Potential clients would be GPUs, DSPs, ISPs.
Still, it seems less than ideal for every client to have to remember to make all these individual calls, all in the right order (especially when it comes to undoing them in failure paths).
There are 3 things that register and unregister process must take into account: a) EM struct is populated based on OPPs of the device (OPPs must be there before EM tries to calculate per-OPP-cost) b) EM is returned to subsystems like: scheduler or thermal (unregister must be done after removing cooling device) c) EM might be created with driver specific callback function, called for each OPP by the EM during setup (do not put default simple EM model into framework function)
So a) and b) shouldn't be hard to code, but I agree, it would be easier for driver developer to not think about them. Let me try to experiment and address this.
I haven't quite grasped whether the energy model is conceptually "owned" by the OPP table or by the cooling device, but either way it would seem to be a much nicer API if there were simply an additional "with energy model" variant of the registration call, and the standard removal call just automatically cleaned up an energy model as well if one was present.
This dev_pm_opp_of_* could potentially be used directly inside devfreq_cooling_register() and then 'unregister' call done internally. Just limiting devfreq_cooling to a call to: dev_pm_opp_of_register_em() would limit the EM for devfreq devices to use only this simple DT model (which is based on "dynamic-power-coefficient") which fails in i.e. GPU SCMI where OPP power comes from firmware.
That would require something like devfreq_cooling_with_em_register(<old_args_here>, struct em_data_callback *em_cb)
and then if the *em_cb is set to null, it could call dev_pm_opp_of_register_em() trying to use it's default em_cb function, which seeks for "dynamic-power-coefficient".
This is doable inside devfreq_cooling, register and unregister of EM would be avoided in drivers code.
Thank you Robin for your comments clarifying these things.
Regards, Lukasz
dri-devel@lists.freedesktop.org
-
Lukasz Luba -
Randy Dunlap -
Rob Herring -
Robin Murphy