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