1.. SPDX-License-Identifier: GPL-2.0 2 3=============================================== 4How to Implement a new CPUFreq Processor Driver 5=============================================== 6 7Authors: 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 27 281. What To Do? 29============== 30 31So, you just got a brand-new CPU / chipset with datasheets and want to 32add cpufreq support for this CPU / chipset? Great. Here are some hints 33on what is necessary: 34 35 361.1 Initialization 37------------------ 38 39First of all, in an __initcall level 7 (module_init()) or later 40function check whether this kernel runs on the right CPU and the right 41chipset. If so, register a struct cpufreq_driver with the CPUfreq core 42using cpufreq_register_driver() 43 44What 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 55And optionally 56 57 .flags - Hints for the cpufreq core. 58 59 .driver_data - cpufreq driver specific data. 60 61 .resolve_freq - Returns the most appropriate frequency for a target 62 frequency. Doesn't change the frequency though. 63 64 .get_intermediate and target_intermediate - Used to switch to stable 65 frequency while changing CPU frequency. 66 67 .get - Returns current frequency of the CPU. 68 69 .bios_limit - Returns HW/BIOS max frequency limitations for the CPU. 70 71 .exit - A pointer to a per-policy cleanup function called during 72 CPU_POST_DEAD phase of cpu hotplug process. 73 74 .suspend - A pointer to a per-policy suspend function which is called 75 with interrupts disabled and _after_ the governor is stopped for the 76 policy. 77 78 .resume - A pointer to a per-policy resume function which is called 79 with interrupts disabled and _before_ the governor is started again. 80 81 .ready - A pointer to a per-policy ready function which is called after 82 the policy is fully initialized. 83 84 .attr - A pointer to a NULL-terminated list of "struct freq_attr" which 85 allow to export values to sysfs. 86 87 .boost_enabled - If set, boost frequencies are enabled. 88 89 .set_boost - A pointer to a per-policy function to enable/disable boost 90 frequencies. 91 92 931.2 Per-CPU Initialization 94-------------------------- 95 96Whenever a new CPU is registered with the device model, or after the 97cpufreq driver registers itself, the per-policy initialization function 98cpufreq_driver.init is called if no cpufreq policy existed for the CPU. 99Note that the .init() and .exit() routines are called only once for the 100policy and not for each CPU managed by the policy. It takes a ``struct 101cpufreq_policy *policy`` as argument. What to do now? 102 103If necessary, activate the CPUfreq support on your CPU. 104 105Then, the driver must fill in the following values: 106 107+-----------------------------------+--------------------------------------+ 108|policy->cpuinfo.min_freq _and_ | | 109|policy->cpuinfo.max_freq | the minimum and maximum frequency | 110| | (in kHz) which is supported by | 111| | this CPU | 112+-----------------------------------+--------------------------------------+ 113|policy->cpuinfo.transition_latency | the time it takes on this CPU to | 114| | switch between two frequencies in | 115| | nanoseconds (if appropriate, else | 116| | specify CPUFREQ_ETERNAL) | 117+-----------------------------------+--------------------------------------+ 118|policy->cur | The current operating frequency of | 119| | this CPU (if appropriate) | 120+-----------------------------------+--------------------------------------+ 121|policy->min, | | 122|policy->max, | | 123|policy->policy and, if necessary, | | 124|policy->governor | must contain the "default policy" for| 125| | this CPU. A few moments later, | 126| | cpufreq_driver.verify and either | 127| | cpufreq_driver.setpolicy or | 128| | cpufreq_driver.target/target_index is| 129| | called with these values. | 130+-----------------------------------+--------------------------------------+ 131|policy->cpus | Update this with the masks of the | 132| | (online + offline) CPUs that do DVFS | 133| | along with this CPU (i.e. that share| 134| | clock/voltage rails with it). | 135+-----------------------------------+--------------------------------------+ 136 137For setting some of these values (cpuinfo.min[max]_freq, policy->min[max]), the 138frequency table helpers might be helpful. See the section 2 for more information 139on them. 140 141 1421.3 verify 143---------- 144 145When the user decides a new policy (consisting of 146"policy,governor,min,max") shall be set, this policy must be validated 147so that incompatible values can be corrected. For verifying these 148values cpufreq_verify_within_limits(``struct cpufreq_policy *policy``, 149``unsigned int min_freq``, ``unsigned int max_freq``) function might be helpful. 150See section 2 for details on frequency table helpers. 151 152You need to make sure that at least one valid frequency (or operating 153range) is within policy->min and policy->max. If necessary, increase 154policy->max first, and only if this is no solution, decrease policy->min. 155 156 1571.4 target or target_index or setpolicy or fast_switch? 158------------------------------------------------------- 159 160Most cpufreq drivers or even most cpu frequency scaling algorithms 161only allow the CPU frequency to be set to predefined fixed values. For 162these, you use the ->target(), ->target_index() or ->fast_switch() 163callbacks. 164 165Some cpufreq capable processors switch the frequency between certain 166limits on their own. These shall use the ->setpolicy() callback. 167 168 1691.5. target/target_index 170------------------------ 171 172The target_index call has two arguments: ``struct cpufreq_policy *policy``, 173and ``unsigned int`` index (into the exposed frequency table). 174 175The CPUfreq driver must set the new frequency when called here. The 176actual frequency must be determined by freq_table[index].frequency. 177 178It should always restore to earlier frequency (i.e. policy->restore_freq) in 179case of errors, even if we switched to intermediate frequency earlier. 180 181Deprecated 182---------- 183The target call has three arguments: ``struct cpufreq_policy *policy``, 184unsigned int target_frequency, unsigned int relation. 185 186The CPUfreq driver must set the new frequency when called here. The 187actual frequency must be determined using the following rules: 188 189- keep close to "target_freq" 190- policy->min <= new_freq <= policy->max (THIS MUST BE VALID!!!) 191- if relation==CPUFREQ_REL_L, try to select a new_freq higher than or equal 192 target_freq. ("L for lowest, but no lower than") 193- if relation==CPUFREQ_REL_H, try to select a new_freq lower than or equal 194 target_freq. ("H for highest, but no higher than") 195 196Here again the frequency table helper might assist you - see section 2 197for details. 198 1991.6. fast_switch 200---------------- 201 202This function is used for frequency switching from scheduler's context. 203Not all drivers are expected to implement it, as sleeping from within 204this callback isn't allowed. This callback must be highly optimized to 205do switching as fast as possible. 206 207This function has two arguments: ``struct cpufreq_policy *policy`` and 208``unsigned int target_frequency``. 209 210 2111.7 setpolicy 212------------- 213 214The setpolicy call only takes a ``struct cpufreq_policy *policy`` as 215argument. You need to set the lower limit of the in-processor or 216in-chipset dynamic frequency switching to policy->min, the upper limit 217to policy->max, and -if supported- select a performance-oriented 218setting when policy->policy is CPUFREQ_POLICY_PERFORMANCE, and a 219powersaving-oriented setting when CPUFREQ_POLICY_POWERSAVE. Also check 220the reference implementation in drivers/cpufreq/longrun.c 221 2221.8 get_intermediate and target_intermediate 223-------------------------------------------- 224 225Only for drivers with target_index() and CPUFREQ_ASYNC_NOTIFICATION unset. 226 227get_intermediate should return a stable intermediate frequency platform wants to 228switch to, and target_intermediate() should set CPU to that frequency, before 229jumping to the frequency corresponding to 'index'. Core will take care of 230sending notifications and driver doesn't have to handle them in 231target_intermediate() or target_index(). 232 233Drivers can return '0' from get_intermediate() in case they don't wish to switch 234to intermediate frequency for some target frequency. In that case core will 235directly call ->target_index(). 236 237NOTE: ->target_index() should restore to policy->restore_freq in case of 238failures as core would send notifications for that. 239 240 2412. Frequency Table Helpers 242========================== 243 244As most cpufreq processors only allow for being set to a few specific 245frequencies, a "frequency table" with some functions might assist in 246some work of the processor driver. Such a "frequency table" consists of 247an array of struct cpufreq_frequency_table entries, with driver specific 248values in "driver_data", the corresponding frequency in "frequency" and 249flags set. At the end of the table, you need to add a 250cpufreq_frequency_table entry with frequency set to CPUFREQ_TABLE_END. 251And if you want to skip one entry in the table, set the frequency to 252CPUFREQ_ENTRY_INVALID. The entries don't need to be in sorted in any 253particular order, but if they are cpufreq core will do DVFS a bit 254quickly for them as search for best match is faster. 255 256The cpufreq table is verified automatically by the core if the policy contains a 257valid pointer in its policy->freq_table field. 258 259cpufreq_frequency_table_verify() assures that at least one valid 260frequency is within policy->min and policy->max, and all other criteria 261are met. This is helpful for the ->verify call. 262 263cpufreq_frequency_table_target() is the corresponding frequency table 264helper for the ->target stage. Just pass the values to this function, 265and this function returns the of the frequency table entry which 266contains the frequency the CPU shall be set to. 267 268The following macros can be used as iterators over cpufreq_frequency_table: 269 270cpufreq_for_each_entry(pos, table) - iterates over all entries of frequency 271table. 272 273cpufreq_for_each_valid_entry(pos, table) - iterates over all entries, 274excluding CPUFREQ_ENTRY_INVALID frequencies. 275Use arguments "pos" - a ``cpufreq_frequency_table *`` as a loop cursor and 276"table" - the ``cpufreq_frequency_table *`` you want to iterate over. 277 278For example:: 279 280 struct cpufreq_frequency_table *pos, *driver_freq_table; 281 282 cpufreq_for_each_entry(pos, driver_freq_table) { 283 /* Do something with pos */ 284 pos->frequency = ... 285 } 286 287If you need to work with the position of pos within driver_freq_table, 288do not subtract the pointers, as it is quite costly. Instead, use the 289macros cpufreq_for_each_entry_idx() and cpufreq_for_each_valid_entry_idx(). 290