1// SPDX-License-Identifier: GPL-2.0 2/* 3 * devfreq_cooling: Thermal cooling device implementation for devices using 4 * devfreq 5 * 6 * Copyright (C) 2014-2015 ARM Limited 7 * 8 * TODO: 9 * - If OPPs are added or removed after devfreq cooling has 10 * registered, the devfreq cooling won't react to it. 11 */ 12 13#include <linux/devfreq.h> 14#include <linux/devfreq_cooling.h> 15#include <linux/export.h> 16#include <linux/idr.h> 17#include <linux/slab.h> 18#include <linux/pm_opp.h> 19#include <linux/pm_qos.h> 20#include <linux/thermal.h> 21 22#include <trace/events/thermal.h> 23 24#define HZ_PER_KHZ 1000 25#define SCALE_ERROR_MITIGATION 100 26 27static DEFINE_IDA(devfreq_ida); 28 29/** 30 * struct devfreq_cooling_device - Devfreq cooling device 31 * @id: unique integer value corresponding to each 32 * devfreq_cooling_device registered. 33 * @cdev: Pointer to associated thermal cooling device. 34 * @devfreq: Pointer to associated devfreq device. 35 * @cooling_state: Current cooling state. 36 * @power_table: Pointer to table with maximum power draw for each 37 * cooling state. State is the index into the table, and 38 * the power is in mW. 39 * @freq_table: Pointer to a table with the frequencies sorted in descending 40 * order. You can index the table by cooling device state 41 * @freq_table_size: Size of the @freq_table and @power_table 42 * @power_ops: Pointer to devfreq_cooling_power, used to generate the 43 * @power_table. 44 * @res_util: Resource utilization scaling factor for the power. 45 * It is multiplied by 100 to minimize the error. It is used 46 * for estimation of the power budget instead of using 47 * 'utilization' (which is 'busy_time / 'total_time'). 48 * The 'res_util' range is from 100 to (power_table[state] * 100) 49 * for the corresponding 'state'. 50 * @capped_state: index to cooling state with in dynamic power budget 51 * @req_max_freq: PM QoS request for limiting the maximum frequency 52 * of the devfreq device. 53 */ 54struct devfreq_cooling_device { 55 int id; 56 struct thermal_cooling_device *cdev; 57 struct devfreq *devfreq; 58 unsigned long cooling_state; 59 u32 *power_table; 60 u32 *freq_table; 61 size_t freq_table_size; 62 struct devfreq_cooling_power *power_ops; 63 u32 res_util; 64 int capped_state; 65 struct dev_pm_qos_request req_max_freq; 66}; 67 68static int devfreq_cooling_get_max_state(struct thermal_cooling_device *cdev, 69 unsigned long *state) 70{ 71 struct devfreq_cooling_device *dfc = cdev->devdata; 72 73 *state = dfc->freq_table_size - 1; 74 75 return 0; 76} 77 78static int devfreq_cooling_get_cur_state(struct thermal_cooling_device *cdev, 79 unsigned long *state) 80{ 81 struct devfreq_cooling_device *dfc = cdev->devdata; 82 83 *state = dfc->cooling_state; 84 85 return 0; 86} 87 88static int devfreq_cooling_set_cur_state(struct thermal_cooling_device *cdev, 89 unsigned long state) 90{ 91 struct devfreq_cooling_device *dfc = cdev->devdata; 92 struct devfreq *df = dfc->devfreq; 93 struct device *dev = df->dev.parent; 94 unsigned long freq; 95 96 if (state == dfc->cooling_state) 97 return 0; 98 99 dev_dbg(dev, "Setting cooling state %lu\n", state); 100 101 if (state >= dfc->freq_table_size) 102 return -EINVAL; 103 104 freq = dfc->freq_table[state]; 105 106 dev_pm_qos_update_request(&dfc->req_max_freq, 107 DIV_ROUND_UP(freq, HZ_PER_KHZ)); 108 109 dfc->cooling_state = state; 110 111 return 0; 112} 113 114/** 115 * freq_get_state() - get the cooling state corresponding to a frequency 116 * @dfc: Pointer to devfreq cooling device 117 * @freq: frequency in Hz 118 * 119 * Return: the cooling state associated with the @freq, or 120 * THERMAL_CSTATE_INVALID if it wasn't found. 121 */ 122static unsigned long 123freq_get_state(struct devfreq_cooling_device *dfc, unsigned long freq) 124{ 125 int i; 126 127 for (i = 0; i < dfc->freq_table_size; i++) { 128 if (dfc->freq_table[i] == freq) 129 return i; 130 } 131 132 return THERMAL_CSTATE_INVALID; 133} 134 135static unsigned long get_voltage(struct devfreq *df, unsigned long freq) 136{ 137 struct device *dev = df->dev.parent; 138 unsigned long voltage; 139 struct dev_pm_opp *opp; 140 141 opp = dev_pm_opp_find_freq_exact(dev, freq, true); 142 if (PTR_ERR(opp) == -ERANGE) 143 opp = dev_pm_opp_find_freq_exact(dev, freq, false); 144 145 if (IS_ERR(opp)) { 146 dev_err_ratelimited(dev, "Failed to find OPP for frequency %lu: %ld\n", 147 freq, PTR_ERR(opp)); 148 return 0; 149 } 150 151 voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */ 152 dev_pm_opp_put(opp); 153 154 if (voltage == 0) { 155 dev_err_ratelimited(dev, 156 "Failed to get voltage for frequency %lu\n", 157 freq); 158 } 159 160 return voltage; 161} 162 163/** 164 * get_static_power() - calculate the static power 165 * @dfc: Pointer to devfreq cooling device 166 * @freq: Frequency in Hz 167 * 168 * Calculate the static power in milliwatts using the supplied 169 * get_static_power(). The current voltage is calculated using the 170 * OPP library. If no get_static_power() was supplied, assume the 171 * static power is negligible. 172 */ 173static unsigned long 174get_static_power(struct devfreq_cooling_device *dfc, unsigned long freq) 175{ 176 struct devfreq *df = dfc->devfreq; 177 unsigned long voltage; 178 179 if (!dfc->power_ops->get_static_power) 180 return 0; 181 182 voltage = get_voltage(df, freq); 183 184 if (voltage == 0) 185 return 0; 186 187 return dfc->power_ops->get_static_power(df, voltage); 188} 189 190/** 191 * get_dynamic_power - calculate the dynamic power 192 * @dfc: Pointer to devfreq cooling device 193 * @freq: Frequency in Hz 194 * @voltage: Voltage in millivolts 195 * 196 * Calculate the dynamic power in milliwatts consumed by the device at 197 * frequency @freq and voltage @voltage. If the get_dynamic_power() 198 * was supplied as part of the devfreq_cooling_power struct, then that 199 * function is used. Otherwise, a simple power model (Pdyn = Coeff * 200 * Voltage^2 * Frequency) is used. 201 */ 202static unsigned long 203get_dynamic_power(struct devfreq_cooling_device *dfc, unsigned long freq, 204 unsigned long voltage) 205{ 206 u64 power; 207 u32 freq_mhz; 208 struct devfreq_cooling_power *dfc_power = dfc->power_ops; 209 210 if (dfc_power->get_dynamic_power) 211 return dfc_power->get_dynamic_power(dfc->devfreq, freq, 212 voltage); 213 214 freq_mhz = freq / 1000000; 215 power = (u64)dfc_power->dyn_power_coeff * freq_mhz * voltage * voltage; 216 do_div(power, 1000000000); 217 218 return power; 219} 220 221 222static inline unsigned long get_total_power(struct devfreq_cooling_device *dfc, 223 unsigned long freq, 224 unsigned long voltage) 225{ 226 return get_static_power(dfc, freq) + get_dynamic_power(dfc, freq, 227 voltage); 228} 229 230 231static int devfreq_cooling_get_requested_power(struct thermal_cooling_device *cdev, 232 u32 *power) 233{ 234 struct devfreq_cooling_device *dfc = cdev->devdata; 235 struct devfreq *df = dfc->devfreq; 236 struct devfreq_dev_status *status = &df->last_status; 237 unsigned long state; 238 unsigned long freq = status->current_frequency; 239 unsigned long voltage; 240 u32 dyn_power = 0; 241 u32 static_power = 0; 242 int res; 243 244 state = freq_get_state(dfc, freq); 245 if (state == THERMAL_CSTATE_INVALID) { 246 res = -EAGAIN; 247 goto fail; 248 } 249 250 if (dfc->power_ops->get_real_power) { 251 voltage = get_voltage(df, freq); 252 if (voltage == 0) { 253 res = -EINVAL; 254 goto fail; 255 } 256 257 res = dfc->power_ops->get_real_power(df, power, freq, voltage); 258 if (!res) { 259 state = dfc->capped_state; 260 dfc->res_util = dfc->power_table[state]; 261 dfc->res_util *= SCALE_ERROR_MITIGATION; 262 263 if (*power > 1) 264 dfc->res_util /= *power; 265 } else { 266 goto fail; 267 } 268 } else { 269 dyn_power = dfc->power_table[state]; 270 271 /* Scale dynamic power for utilization */ 272 dyn_power *= status->busy_time; 273 dyn_power /= status->total_time; 274 /* Get static power */ 275 static_power = get_static_power(dfc, freq); 276 277 *power = dyn_power + static_power; 278 } 279 280 trace_thermal_power_devfreq_get_power(cdev, status, freq, dyn_power, 281 static_power, *power); 282 283 return 0; 284fail: 285 /* It is safe to set max in this case */ 286 dfc->res_util = SCALE_ERROR_MITIGATION; 287 return res; 288} 289 290static int devfreq_cooling_state2power(struct thermal_cooling_device *cdev, 291 unsigned long state, 292 u32 *power) 293{ 294 struct devfreq_cooling_device *dfc = cdev->devdata; 295 unsigned long freq; 296 u32 static_power; 297 298 if (state >= dfc->freq_table_size) 299 return -EINVAL; 300 301 freq = dfc->freq_table[state]; 302 static_power = get_static_power(dfc, freq); 303 304 *power = dfc->power_table[state] + static_power; 305 return 0; 306} 307 308static int devfreq_cooling_power2state(struct thermal_cooling_device *cdev, 309 u32 power, unsigned long *state) 310{ 311 struct devfreq_cooling_device *dfc = cdev->devdata; 312 struct devfreq *df = dfc->devfreq; 313 struct devfreq_dev_status *status = &df->last_status; 314 unsigned long freq = status->current_frequency; 315 unsigned long busy_time; 316 s32 dyn_power; 317 u32 static_power; 318 s32 est_power; 319 int i; 320 321 if (dfc->power_ops->get_real_power) { 322 /* Scale for resource utilization */ 323 est_power = power * dfc->res_util; 324 est_power /= SCALE_ERROR_MITIGATION; 325 } else { 326 static_power = get_static_power(dfc, freq); 327 328 dyn_power = power - static_power; 329 dyn_power = dyn_power > 0 ? dyn_power : 0; 330 331 /* Scale dynamic power for utilization */ 332 busy_time = status->busy_time ?: 1; 333 est_power = (dyn_power * status->total_time) / busy_time; 334 } 335 336 /* 337 * Find the first cooling state that is within the power 338 * budget for dynamic power. 339 */ 340 for (i = 0; i < dfc->freq_table_size - 1; i++) 341 if (est_power >= dfc->power_table[i]) 342 break; 343 344 *state = i; 345 dfc->capped_state = i; 346 trace_thermal_power_devfreq_limit(cdev, freq, *state, power); 347 return 0; 348} 349 350static struct thermal_cooling_device_ops devfreq_cooling_ops = { 351 .get_max_state = devfreq_cooling_get_max_state, 352 .get_cur_state = devfreq_cooling_get_cur_state, 353 .set_cur_state = devfreq_cooling_set_cur_state, 354}; 355 356/** 357 * devfreq_cooling_gen_tables() - Generate power and freq tables. 358 * @dfc: Pointer to devfreq cooling device. 359 * 360 * Generate power and frequency tables: the power table hold the 361 * device's maximum power usage at each cooling state (OPP). The 362 * static and dynamic power using the appropriate voltage and 363 * frequency for the state, is acquired from the struct 364 * devfreq_cooling_power, and summed to make the maximum power draw. 365 * 366 * The frequency table holds the frequencies in descending order. 367 * That way its indexed by cooling device state. 368 * 369 * The tables are malloced, and pointers put in dfc. They must be 370 * freed when unregistering the devfreq cooling device. 371 * 372 * Return: 0 on success, negative error code on failure. 373 */ 374static int devfreq_cooling_gen_tables(struct devfreq_cooling_device *dfc) 375{ 376 struct devfreq *df = dfc->devfreq; 377 struct device *dev = df->dev.parent; 378 int ret, num_opps; 379 unsigned long freq; 380 u32 *power_table = NULL; 381 u32 *freq_table; 382 int i; 383 384 num_opps = dev_pm_opp_get_opp_count(dev); 385 386 if (dfc->power_ops) { 387 power_table = kcalloc(num_opps, sizeof(*power_table), 388 GFP_KERNEL); 389 if (!power_table) 390 return -ENOMEM; 391 } 392 393 freq_table = kcalloc(num_opps, sizeof(*freq_table), 394 GFP_KERNEL); 395 if (!freq_table) { 396 ret = -ENOMEM; 397 goto free_power_table; 398 } 399 400 for (i = 0, freq = ULONG_MAX; i < num_opps; i++, freq--) { 401 unsigned long power, voltage; 402 struct dev_pm_opp *opp; 403 404 opp = dev_pm_opp_find_freq_floor(dev, &freq); 405 if (IS_ERR(opp)) { 406 ret = PTR_ERR(opp); 407 goto free_tables; 408 } 409 410 voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */ 411 dev_pm_opp_put(opp); 412 413 if (dfc->power_ops) { 414 if (dfc->power_ops->get_real_power) 415 power = get_total_power(dfc, freq, voltage); 416 else 417 power = get_dynamic_power(dfc, freq, voltage); 418 419 dev_dbg(dev, "Power table: %lu MHz @ %lu mV: %lu = %lu mW\n", 420 freq / 1000000, voltage, power, power); 421 422 power_table[i] = power; 423 } 424 425 freq_table[i] = freq; 426 } 427 428 if (dfc->power_ops) 429 dfc->power_table = power_table; 430 431 dfc->freq_table = freq_table; 432 dfc->freq_table_size = num_opps; 433 434 return 0; 435 436free_tables: 437 kfree(freq_table); 438free_power_table: 439 kfree(power_table); 440 441 return ret; 442} 443 444/** 445 * of_devfreq_cooling_register_power() - Register devfreq cooling device, 446 * with OF and power information. 447 * @np: Pointer to OF device_node. 448 * @df: Pointer to devfreq device. 449 * @dfc_power: Pointer to devfreq_cooling_power. 450 * 451 * Register a devfreq cooling device. The available OPPs must be 452 * registered on the device. 453 * 454 * If @dfc_power is provided, the cooling device is registered with the 455 * power extensions. For the power extensions to work correctly, 456 * devfreq should use the simple_ondemand governor, other governors 457 * are not currently supported. 458 */ 459struct thermal_cooling_device * 460of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df, 461 struct devfreq_cooling_power *dfc_power) 462{ 463 struct thermal_cooling_device *cdev; 464 struct devfreq_cooling_device *dfc; 465 char dev_name[THERMAL_NAME_LENGTH]; 466 int err; 467 468 dfc = kzalloc(sizeof(*dfc), GFP_KERNEL); 469 if (!dfc) 470 return ERR_PTR(-ENOMEM); 471 472 dfc->devfreq = df; 473 474 if (dfc_power) { 475 dfc->power_ops = dfc_power; 476 477 devfreq_cooling_ops.get_requested_power = 478 devfreq_cooling_get_requested_power; 479 devfreq_cooling_ops.state2power = devfreq_cooling_state2power; 480 devfreq_cooling_ops.power2state = devfreq_cooling_power2state; 481 } 482 483 err = devfreq_cooling_gen_tables(dfc); 484 if (err) 485 goto free_dfc; 486 487 err = dev_pm_qos_add_request(df->dev.parent, &dfc->req_max_freq, 488 DEV_PM_QOS_MAX_FREQUENCY, 489 PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE); 490 if (err < 0) 491 goto free_tables; 492 493 err = ida_simple_get(&devfreq_ida, 0, 0, GFP_KERNEL); 494 if (err < 0) 495 goto remove_qos_req; 496 dfc->id = err; 497 498 snprintf(dev_name, sizeof(dev_name), "thermal-devfreq-%d", dfc->id); 499 500 cdev = thermal_of_cooling_device_register(np, dev_name, dfc, 501 &devfreq_cooling_ops); 502 if (IS_ERR(cdev)) { 503 err = PTR_ERR(cdev); 504 dev_err(df->dev.parent, 505 "Failed to register devfreq cooling device (%d)\n", 506 err); 507 goto release_ida; 508 } 509 510 dfc->cdev = cdev; 511 512 return cdev; 513 514release_ida: 515 ida_simple_remove(&devfreq_ida, dfc->id); 516 517remove_qos_req: 518 dev_pm_qos_remove_request(&dfc->req_max_freq); 519 520free_tables: 521 kfree(dfc->power_table); 522 kfree(dfc->freq_table); 523free_dfc: 524 kfree(dfc); 525 526 return ERR_PTR(err); 527} 528EXPORT_SYMBOL_GPL(of_devfreq_cooling_register_power); 529 530/** 531 * of_devfreq_cooling_register() - Register devfreq cooling device, 532 * with OF information. 533 * @np: Pointer to OF device_node. 534 * @df: Pointer to devfreq device. 535 */ 536struct thermal_cooling_device * 537of_devfreq_cooling_register(struct device_node *np, struct devfreq *df) 538{ 539 return of_devfreq_cooling_register_power(np, df, NULL); 540} 541EXPORT_SYMBOL_GPL(of_devfreq_cooling_register); 542 543/** 544 * devfreq_cooling_register() - Register devfreq cooling device. 545 * @df: Pointer to devfreq device. 546 */ 547struct thermal_cooling_device *devfreq_cooling_register(struct devfreq *df) 548{ 549 return of_devfreq_cooling_register(NULL, df); 550} 551EXPORT_SYMBOL_GPL(devfreq_cooling_register); 552 553/** 554 * devfreq_cooling_unregister() - Unregister devfreq cooling device. 555 * @cdev: Pointer to devfreq cooling device to unregister. 556 */ 557void devfreq_cooling_unregister(struct thermal_cooling_device *cdev) 558{ 559 struct devfreq_cooling_device *dfc; 560 561 if (!cdev) 562 return; 563 564 dfc = cdev->devdata; 565 566 thermal_cooling_device_unregister(dfc->cdev); 567 ida_simple_remove(&devfreq_ida, dfc->id); 568 dev_pm_qos_remove_request(&dfc->req_max_freq); 569 kfree(dfc->power_table); 570 kfree(dfc->freq_table); 571 572 kfree(dfc); 573} 574EXPORT_SYMBOL_GPL(devfreq_cooling_unregister); 575