1 .. SPDX-License-Identifier: GPL-2.0
2
3 ===============================================
4 How to Implement a new CPUFreq Processor Driver
5 ===============================================
6
7 Authors:
8
9
10 - Dominik Brodowski <linux@brodo.de>
11 - Rafael J. Wysocki <rafael.j.wysocki@intel.com>
12 - Viresh Kumar <viresh.kumar@linaro.org>
13
14 .. Contents
15
16 1. What To Do?
17 1.1 Initialization
18 1.2 Per-CPU Initialization
19 1.3 verify
20 1.4 target/target_index or setpolicy?
21 1.5 target/target_index
22 1.6 setpolicy
23 1.7 get_intermediate and target_intermediate
24 2. Frequency Table Helpers
25
26
27
28 1. What To Do?
29 ==============
30
31 So, you just got a brand-new CPU / chipset with datasheets and want to
32 add cpufreq support for this CPU / chipset? Great. Here are some hints
33 on what is necessary:
34
35
36 1.1 Initialization
37 ------------------
38
39 First of all, in an __initcall level 7 (module_init()) or later
40 function check whether this kernel runs on the right CPU and the right
41 chipset. If so, register a struct cpufreq_driver with the CPUfreq core
42 using cpufreq_register_driver()
43
44 What shall this struct cpufreq_driver contain?
45
46 .name - The name of this driver.
47
48 .init - A pointer to the per-policy initialization function.
49
50 .verify - A pointer to a "verification" function.
51
52 .setpolicy _or_ .fast_switch _or_ .target _or_ .target_index - See
53 below on the differences.
54
55 And optionally
56
57 .flags - Hints for the cpufreq core.
58
59 .driver_data - cpufreq driver specific data.
60
61 .get_intermediate and target_intermediate - Used to switch to stable
62 frequency while changing CPU frequency.
63
64 .get - Returns current frequency of the CPU.
65
66 .bios_limit - Returns HW/BIOS max frequency limitations for the CPU.
67
68 .exit - A pointer to a per-policy cleanup function called during
69 CPU_POST_DEAD phase of cpu hotplug process.
70
71 .suspend - A pointer to a per-policy suspend function which is called
72 with interrupts disabled and _after_ the governor is stopped for the
73 policy.
74
75 .resume - A pointer to a per-policy resume function which is called
76 with interrupts disabled and _before_ the governor is started again.
77
78 .ready - A pointer to a per-policy ready function which is called after
79 the policy is fully initialized.
80
81 .attr - A pointer to a NULL-terminated list of "struct freq_attr" which
82 allow to export values to sysfs.
83
84 .boost_enabled - If set, boost frequencies are enabled.
85
86 .set_boost - A pointer to a per-policy function to enable/disable boost
87 frequencies.
88
89
90 1.2 Per-CPU Initialization
91 --------------------------
92
93 Whenever a new CPU is registered with the device model, or after the
94 cpufreq driver registers itself, the per-policy initialization function
95 cpufreq_driver.init is called if no cpufreq policy existed for the CPU.
96 Note that the .init() and .exit() routines are called only once for the
97 policy and not for each CPU managed by the policy. It takes a ``struct
98 cpufreq_policy *policy`` as argument. What to do now?
99
100 If necessary, activate the CPUfreq support on your CPU.
101
102 Then, the driver must fill in the following values:
103
104 +-----------------------------------+--------------------------------------+
105 |policy->cpuinfo.min_freq _and_ | |
106 |policy->cpuinfo.max_freq | the minimum and maximum frequency |
107 | | (in kHz) which is supported by |
108 | | this CPU |
109 +-----------------------------------+--------------------------------------+
110 |policy->cpuinfo.transition_latency | the time it takes on this CPU to |
111 | | switch between two frequencies in |
112 | | nanoseconds (if appropriate, else |
113 | | specify CPUFREQ_ETERNAL) |
114 +-----------------------------------+--------------------------------------+
115 |policy->cur | The current operating frequency of |
116 | | this CPU (if appropriate) |
117 +-----------------------------------+--------------------------------------+
118 |policy->min, | |
119 |policy->max, | |
120 |policy->policy and, if necessary, | |
121 |policy->governor | must contain the "default policy" for|
122 | | this CPU. A few moments later, |
123 | | cpufreq_driver.verify and either |
124 | | cpufreq_driver.setpolicy or |
125 | | cpufreq_driver.target/target_index is|
126 | | called with these values. |
127 +-----------------------------------+--------------------------------------+
128 |policy->cpus | Update this with the masks of the |
129 | | (online + offline) CPUs that do DVFS |
130 | | along with this CPU (i.e. that share|
131 | | clock/voltage rails with it). |
132 +-----------------------------------+--------------------------------------+
133
134 For setting some of these values (cpuinfo.min[max]_freq, policy->min[max]), the
135 frequency table helpers might be helpful. See the section 2 for more information
136 on them.
137
138
139 1.3 verify
140 ----------
141
142 When the user decides a new policy (consisting of
143 "policy,governor,min,max") shall be set, this policy must be validated
144 so that incompatible values can be corrected. For verifying these
145 values cpufreq_verify_within_limits(``struct cpufreq_policy *policy``,
146 ``unsigned int min_freq``, ``unsigned int max_freq``) function might be helpful.
147 See section 2 for details on frequency table helpers.
148
149 You need to make sure that at least one valid frequency (or operating
150 range) is within policy->min and policy->max. If necessary, increase
151 policy->max first, and only if this is no solution, decrease policy->min.
152
153
154 1.4 target or target_index or setpolicy or fast_switch?
155 -------------------------------------------------------
156
157 Most cpufreq drivers or even most cpu frequency scaling algorithms
158 only allow the CPU frequency to be set to predefined fixed values. For
159 these, you use the ->target(), ->target_index() or ->fast_switch()
160 callbacks.
161
162 Some cpufreq capable processors switch the frequency between certain
163 limits on their own. These shall use the ->setpolicy() callback.
164
165
166 1.5. target/target_index
167 ------------------------
168
169 The target_index call has two arguments: ``struct cpufreq_policy *policy``,
170 and ``unsigned int`` index (into the exposed frequency table).
171
172 The CPUfreq driver must set the new frequency when called here. The
173 actual frequency must be determined by freq_table[index].frequency.
174
175 It should always restore to earlier frequency (i.e. policy->restore_freq) in
176 case of errors, even if we switched to intermediate frequency earlier.
177
178 Deprecated
179 ----------
180 The target call has three arguments: ``struct cpufreq_policy *policy``,
181 unsigned int target_frequency, unsigned int relation.
182
183 The CPUfreq driver must set the new frequency when called here. The
184 actual frequency must be determined using the following rules:
185
186 - keep close to "target_freq"
187 - policy->min <= new_freq <= policy->max (THIS MUST BE VALID!!!)
188 - if relation==CPUFREQ_REL_L, try to select a new_freq higher than or equal
189 target_freq. ("L for lowest, but no lower than")
190 - if relation==CPUFREQ_REL_H, try to select a new_freq lower than or equal
191 target_freq. ("H for highest, but no higher than")
192
193 Here again the frequency table helper might assist you - see section 2
194 for details.
195
196 1.6. fast_switch
197 ----------------
198
199 This function is used for frequency switching from scheduler's context.
200 Not all drivers are expected to implement it, as sleeping from within
201 this callback isn't allowed. This callback must be highly optimized to
202 do switching as fast as possible.
203
204 This function has two arguments: ``struct cpufreq_policy *policy`` and
205 ``unsigned int target_frequency``.
206
207
208 1.7 setpolicy
209 -------------
210
211 The setpolicy call only takes a ``struct cpufreq_policy *policy`` as
212 argument. You need to set the lower limit of the in-processor or
213 in-chipset dynamic frequency switching to policy->min, the upper limit
214 to policy->max, and -if supported- select a performance-oriented
215 setting when policy->policy is CPUFREQ_POLICY_PERFORMANCE, and a
216 powersaving-oriented setting when CPUFREQ_POLICY_POWERSAVE. Also check
217 the reference implementation in drivers/cpufreq/longrun.c
218
219 1.8 get_intermediate and target_intermediate
220 --------------------------------------------
221
222 Only for drivers with target_index() and CPUFREQ_ASYNC_NOTIFICATION unset.
223
224 get_intermediate should return a stable intermediate frequency platform wants to
225 switch to, and target_intermediate() should set CPU to that frequency, before
226 jumping to the frequency corresponding to 'index'. Core will take care of
227 sending notifications and driver doesn't have to handle them in
228 target_intermediate() or target_index().
229
230 Drivers can return '0' from get_intermediate() in case they don't wish to switch
231 to intermediate frequency for some target frequency. In that case core will
232 directly call ->target_index().
233
234 NOTE: ->target_index() should restore to policy->restore_freq in case of
235 failures as core would send notifications for that.
236
237
238 2. Frequency Table Helpers
239 ==========================
240
241 As most cpufreq processors only allow for being set to a few specific
242 frequencies, a "frequency table" with some functions might assist in
243 some work of the processor driver. Such a "frequency table" consists of
244 an array of struct cpufreq_frequency_table entries, with driver specific
245 values in "driver_data", the corresponding frequency in "frequency" and
246 flags set. At the end of the table, you need to add a
247 cpufreq_frequency_table entry with frequency set to CPUFREQ_TABLE_END.
248 And if you want to skip one entry in the table, set the frequency to
249 CPUFREQ_ENTRY_INVALID. The entries don't need to be in sorted in any
250 particular order, but if they are cpufreq core will do DVFS a bit
251 quickly for them as search for best match is faster.
252
253 The cpufreq table is verified automatically by the core if the policy contains a
254 valid pointer in its policy->freq_table field.
255
256 cpufreq_frequency_table_verify() assures that at least one valid
257 frequency is within policy->min and policy->max, and all other criteria
258 are met. This is helpful for the ->verify call.
259
260 cpufreq_frequency_table_target() is the corresponding frequency table
261 helper for the ->target stage. Just pass the values to this function,
262 and this function returns the of the frequency table entry which
263 contains the frequency the CPU shall be set to.
264
265 The following macros can be used as iterators over cpufreq_frequency_table:
266
267 cpufreq_for_each_entry(pos, table) - iterates over all entries of frequency
268 table.
269
270 cpufreq_for_each_valid_entry(pos, table) - iterates over all entries,
271 excluding CPUFREQ_ENTRY_INVALID frequencies.
272 Use arguments "pos" - a ``cpufreq_frequency_table *`` as a loop cursor and
273 "table" - the ``cpufreq_frequency_table *`` you want to iterate over.
274
275 For example::
276
277 struct cpufreq_frequency_table *pos, *driver_freq_table;
278
279 cpufreq_for_each_entry(pos, driver_freq_table) {
280 /* Do something with pos */
281 pos->frequency = ...
282 }
283
284 If you need to work with the position of pos within driver_freq_table,
285 do not subtract the pointers, as it is quite costly. Instead, use the
286 macros cpufreq_for_each_entry_idx() and cpufreq_for_each_valid_entry_idx().
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.