1.. SPDX-License-Identifier: GPL-2.0
2
3=======================
4Energy Model of devices
5=======================
6
71. Overview
8-----------
9
10The Energy Model (EM) framework serves as an interface between drivers knowing
11the power consumed by devices at various performance levels, and the kernel
12subsystems willing to use that information to make energy-aware decisions.
13
14The source of the information about the power consumed by devices can vary greatly
15from one platform to another. These power costs can be estimated using
16devicetree data in some cases. In others, the firmware will know better.
17Alternatively, userspace might be best positioned. And so on. In order to avoid
18each and every client subsystem to re-implement support for each and every
19possible source of information on its own, the EM framework intervenes as an
20abstraction layer which standardizes the format of power cost tables in the
21kernel, hence enabling to avoid redundant work.
22
23The power values might be expressed in milli-Watts or in an 'abstract scale'.
24Multiple subsystems might use the EM and it is up to the system integrator to
25check that the requirements for the power value scale types are met. An example
26can be found in the Energy-Aware Scheduler documentation
27Documentation/scheduler/sched-energy.rst. For some subsystems like thermal or
28powercap power values expressed in an 'abstract scale' might cause issues.
29These subsystems are more interested in estimation of power used in the past,
30thus the real milli-Watts might be needed. An example of these requirements can
31be found in the Intelligent Power Allocation in
32Documentation/driver-api/thermal/power_allocator.rst.
33Important thing to keep in mind is that when the power values are expressed in
34an 'abstract scale' deriving real energy in milli-Joules would not be possible.
35
36The figure below depicts an example of drivers (Arm-specific here, but the
37approach is applicable to any architecture) providing power costs to the EM
38framework, and interested clients reading the data from it::
39
40       +---------------+  +-----------------+  +---------------+
41       | Thermal (IPA) |  | Scheduler (EAS) |  |     Other     |
42       +---------------+  +-----------------+  +---------------+
43               |                   | em_cpu_energy()   |
44               |                   | em_cpu_get()      |
45               +---------+         |         +---------+
46                         |         |         |
47                         v         v         v
48                        +---------------------+
49                        |    Energy Model     |
50                        |     Framework       |
51                        +---------------------+
52                           ^       ^       ^
53                           |       |       | em_dev_register_perf_domain()
54                +----------+       |       +---------+
55                |                  |                 |
56        +---------------+  +---------------+  +--------------+
57        |  cpufreq-dt   |  |   arm_scmi    |  |    Other     |
58        +---------------+  +---------------+  +--------------+
59                ^                  ^                 ^
60                |                  |                 |
61        +--------------+   +---------------+  +--------------+
62        | Device Tree  |   |   Firmware    |  |      ?       |
63        +--------------+   +---------------+  +--------------+
64
65In case of CPU devices the EM framework manages power cost tables per
66'performance domain' in the system. A performance domain is a group of CPUs
67whose performance is scaled together. Performance domains generally have a
681-to-1 mapping with CPUFreq policies. All CPUs in a performance domain are
69required to have the same micro-architecture. CPUs in different performance
70domains can have different micro-architectures.
71
72
732. Core APIs
74------------
75
762.1 Config options
77^^^^^^^^^^^^^^^^^^
78
79CONFIG_ENERGY_MODEL must be enabled to use the EM framework.
80
81
822.2 Registration of performance domains
83^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
84
85Drivers are expected to register performance domains into the EM framework by
86calling the following API::
87
88  int em_dev_register_perf_domain(struct device *dev, unsigned int nr_states,
89		struct em_data_callback *cb, cpumask_t *cpus);
90
91Drivers must provide a callback function returning <frequency, power> tuples
92for each performance state. The callback function provided by the driver is free
93to fetch data from any relevant location (DT, firmware, ...), and by any mean
94deemed necessary. Only for CPU devices, drivers must specify the CPUs of the
95performance domains using cpumask. For other devices than CPUs the last
96argument must be set to NULL.
97See Section 3. for an example of driver implementing this
98callback, and kernel/power/energy_model.c for further documentation on this
99API.
100
101
1022.3 Accessing performance domains
103^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
104
105There are two API functions which provide the access to the energy model:
106em_cpu_get() which takes CPU id as an argument and em_pd_get() with device
107pointer as an argument. It depends on the subsystem which interface it is
108going to use, but in case of CPU devices both functions return the same
109performance domain.
110
111Subsystems interested in the energy model of a CPU can retrieve it using the
112em_cpu_get() API. The energy model tables are allocated once upon creation of
113the performance domains, and kept in memory untouched.
114
115The energy consumed by a performance domain can be estimated using the
116em_cpu_energy() API. The estimation is performed assuming that the schedutil
117CPUfreq governor is in use in case of CPU device. Currently this calculation is
118not provided for other type of devices.
119
120More details about the above APIs can be found in include/linux/energy_model.h.
121
122
1233. Example driver
124-----------------
125
126This section provides a simple example of a CPUFreq driver registering a
127performance domain in the Energy Model framework using the (fake) 'foo'
128protocol. The driver implements an est_power() function to be provided to the
129EM framework::
130
131  -> drivers/cpufreq/foo_cpufreq.c
132
133  01	static int est_power(unsigned long *mW, unsigned long *KHz,
134  02			struct device *dev)
135  03	{
136  04		long freq, power;
137  05
138  06		/* Use the 'foo' protocol to ceil the frequency */
139  07		freq = foo_get_freq_ceil(dev, *KHz);
140  08		if (freq < 0);
141  09			return freq;
142  10
143  11		/* Estimate the power cost for the dev at the relevant freq. */
144  12		power = foo_estimate_power(dev, freq);
145  13		if (power < 0);
146  14			return power;
147  15
148  16		/* Return the values to the EM framework */
149  17		*mW = power;
150  18		*KHz = freq;
151  19
152  20		return 0;
153  21	}
154  22
155  23	static int foo_cpufreq_init(struct cpufreq_policy *policy)
156  24	{
157  25		struct em_data_callback em_cb = EM_DATA_CB(est_power);
158  26		struct device *cpu_dev;
159  27		int nr_opp, ret;
160  28
161  29		cpu_dev = get_cpu_device(cpumask_first(policy->cpus));
162  30
163  31     	/* Do the actual CPUFreq init work ... */
164  32     	ret = do_foo_cpufreq_init(policy);
165  33     	if (ret)
166  34     		return ret;
167  35
168  36     	/* Find the number of OPPs for this policy */
169  37     	nr_opp = foo_get_nr_opp(policy);
170  38
171  39     	/* And register the new performance domain */
172  40     	em_dev_register_perf_domain(cpu_dev, nr_opp, &em_cb, policy->cpus);
173  41
174  42	        return 0;
175  43	}
176