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/idr.h> 25 #include <linux/slab.h> 26 #include <linux/pm_opp.h> 27 #include <linux/thermal.h> 28 29 #include <trace/events/thermal.h> 30 31 #define SCALE_ERROR_MITIGATION 100 32 33 static DEFINE_IDA(devfreq_ida); 34 35 /** 36 * struct devfreq_cooling_device - Devfreq cooling device 37 * @id: unique integer value corresponding to each 38 * devfreq_cooling_device registered. 39 * @cdev: Pointer to associated thermal cooling device. 40 * @devfreq: Pointer to associated devfreq device. 41 * @cooling_state: Current cooling state. 42 * @power_table: Pointer to table with maximum power draw for each 43 * cooling state. State is the index into the table, and 44 * the power is in mW. 45 * @freq_table: Pointer to a table with the frequencies sorted in descending 46 * order. You can index the table by cooling device state 47 * @freq_table_size: Size of the @freq_table and @power_table 48 * @power_ops: Pointer to devfreq_cooling_power, used to generate the 49 * @power_table. 50 * @res_util: Resource utilization scaling factor for the power. 51 * It is multiplied by 100 to minimize the error. It is used 52 * for estimation of the power budget instead of using 53 * 'utilization' (which is 'busy_time / 'total_time'). 54 * The 'res_util' range is from 100 to (power_table[state] * 100) 55 * for the corresponding 'state'. 56 * @capped_state: index to cooling state with in dynamic power budget 57 */ 58 struct devfreq_cooling_device { 59 int id; 60 struct thermal_cooling_device *cdev; 61 struct devfreq *devfreq; 62 unsigned long cooling_state; 63 u32 *power_table; 64 u32 *freq_table; 65 size_t freq_table_size; 66 struct devfreq_cooling_power *power_ops; 67 u32 res_util; 68 int capped_state; 69 }; 70 71 /** 72 * partition_enable_opps() - disable all opps above a given state 73 * @dfc: Pointer to devfreq we are operating on 74 * @cdev_state: cooling device state we're setting 75 * 76 * Go through the OPPs of the device, enabling all OPPs until 77 * @cdev_state and disabling those frequencies above it. 78 */ 79 static int partition_enable_opps(struct devfreq_cooling_device *dfc, 80 unsigned long cdev_state) 81 { 82 int i; 83 struct device *dev = dfc->devfreq->dev.parent; 84 85 for (i = 0; i < dfc->freq_table_size; i++) { 86 struct dev_pm_opp *opp; 87 int ret = 0; 88 unsigned int freq = dfc->freq_table[i]; 89 bool want_enable = i >= cdev_state ? true : false; 90 91 opp = dev_pm_opp_find_freq_exact(dev, freq, !want_enable); 92 93 if (PTR_ERR(opp) == -ERANGE) 94 continue; 95 else if (IS_ERR(opp)) 96 return PTR_ERR(opp); 97 98 dev_pm_opp_put(opp); 99 100 if (want_enable) 101 ret = dev_pm_opp_enable(dev, freq); 102 else 103 ret = dev_pm_opp_disable(dev, freq); 104 105 if (ret) 106 return ret; 107 } 108 109 return 0; 110 } 111 112 static int devfreq_cooling_get_max_state(struct thermal_cooling_device *cdev, 113 unsigned long *state) 114 { 115 struct devfreq_cooling_device *dfc = cdev->devdata; 116 117 *state = dfc->freq_table_size - 1; 118 119 return 0; 120 } 121 122 static int devfreq_cooling_get_cur_state(struct thermal_cooling_device *cdev, 123 unsigned long *state) 124 { 125 struct devfreq_cooling_device *dfc = cdev->devdata; 126 127 *state = dfc->cooling_state; 128 129 return 0; 130 } 131 132 static int devfreq_cooling_set_cur_state(struct thermal_cooling_device *cdev, 133 unsigned long state) 134 { 135 struct devfreq_cooling_device *dfc = cdev->devdata; 136 struct devfreq *df = dfc->devfreq; 137 struct device *dev = df->dev.parent; 138 int ret; 139 140 if (state == dfc->cooling_state) 141 return 0; 142 143 dev_dbg(dev, "Setting cooling state %lu\n", state); 144 145 if (state >= dfc->freq_table_size) 146 return -EINVAL; 147 148 ret = partition_enable_opps(dfc, state); 149 if (ret) 150 return ret; 151 152 dfc->cooling_state = state; 153 154 return 0; 155 } 156 157 /** 158 * freq_get_state() - get the cooling state corresponding to a frequency 159 * @dfc: Pointer to devfreq cooling device 160 * @freq: frequency in Hz 161 * 162 * Return: the cooling state associated with the @freq, or 163 * THERMAL_CSTATE_INVALID if it wasn't found. 164 */ 165 static unsigned long 166 freq_get_state(struct devfreq_cooling_device *dfc, unsigned long freq) 167 { 168 int i; 169 170 for (i = 0; i < dfc->freq_table_size; i++) { 171 if (dfc->freq_table[i] == freq) 172 return i; 173 } 174 175 return THERMAL_CSTATE_INVALID; 176 } 177 178 static unsigned long get_voltage(struct devfreq *df, unsigned long freq) 179 { 180 struct device *dev = df->dev.parent; 181 unsigned long voltage; 182 struct dev_pm_opp *opp; 183 184 opp = dev_pm_opp_find_freq_exact(dev, freq, true); 185 if (PTR_ERR(opp) == -ERANGE) 186 opp = dev_pm_opp_find_freq_exact(dev, freq, false); 187 188 if (IS_ERR(opp)) { 189 dev_err_ratelimited(dev, "Failed to find OPP for frequency %lu: %ld\n", 190 freq, PTR_ERR(opp)); 191 return 0; 192 } 193 194 voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */ 195 dev_pm_opp_put(opp); 196 197 if (voltage == 0) { 198 dev_err_ratelimited(dev, 199 "Failed to get voltage for frequency %lu\n", 200 freq); 201 } 202 203 return voltage; 204 } 205 206 /** 207 * get_static_power() - calculate the static power 208 * @dfc: Pointer to devfreq cooling device 209 * @freq: Frequency in Hz 210 * 211 * Calculate the static power in milliwatts using the supplied 212 * get_static_power(). The current voltage is calculated using the 213 * OPP library. If no get_static_power() was supplied, assume the 214 * static power is negligible. 215 */ 216 static unsigned long 217 get_static_power(struct devfreq_cooling_device *dfc, unsigned long freq) 218 { 219 struct devfreq *df = dfc->devfreq; 220 unsigned long voltage; 221 222 if (!dfc->power_ops->get_static_power) 223 return 0; 224 225 voltage = get_voltage(df, freq); 226 227 if (voltage == 0) 228 return 0; 229 230 return dfc->power_ops->get_static_power(df, voltage); 231 } 232 233 /** 234 * get_dynamic_power - calculate the dynamic power 235 * @dfc: Pointer to devfreq cooling device 236 * @freq: Frequency in Hz 237 * @voltage: Voltage in millivolts 238 * 239 * Calculate the dynamic power in milliwatts consumed by the device at 240 * frequency @freq and voltage @voltage. If the get_dynamic_power() 241 * was supplied as part of the devfreq_cooling_power struct, then that 242 * function is used. Otherwise, a simple power model (Pdyn = Coeff * 243 * Voltage^2 * Frequency) is used. 244 */ 245 static unsigned long 246 get_dynamic_power(struct devfreq_cooling_device *dfc, unsigned long freq, 247 unsigned long voltage) 248 { 249 u64 power; 250 u32 freq_mhz; 251 struct devfreq_cooling_power *dfc_power = dfc->power_ops; 252 253 if (dfc_power->get_dynamic_power) 254 return dfc_power->get_dynamic_power(dfc->devfreq, freq, 255 voltage); 256 257 freq_mhz = freq / 1000000; 258 power = (u64)dfc_power->dyn_power_coeff * freq_mhz * voltage * voltage; 259 do_div(power, 1000000000); 260 261 return power; 262 } 263 264 265 static inline unsigned long get_total_power(struct devfreq_cooling_device *dfc, 266 unsigned long freq, 267 unsigned long voltage) 268 { 269 return get_static_power(dfc, freq) + get_dynamic_power(dfc, freq, 270 voltage); 271 } 272 273 274 static int devfreq_cooling_get_requested_power(struct thermal_cooling_device *cdev, 275 struct thermal_zone_device *tz, 276 u32 *power) 277 { 278 struct devfreq_cooling_device *dfc = cdev->devdata; 279 struct devfreq *df = dfc->devfreq; 280 struct devfreq_dev_status *status = &df->last_status; 281 unsigned long state; 282 unsigned long freq = status->current_frequency; 283 unsigned long voltage; 284 u32 dyn_power = 0; 285 u32 static_power = 0; 286 int res; 287 288 state = freq_get_state(dfc, freq); 289 if (state == THERMAL_CSTATE_INVALID) { 290 res = -EAGAIN; 291 goto fail; 292 } 293 294 if (dfc->power_ops->get_real_power) { 295 voltage = get_voltage(df, freq); 296 if (voltage == 0) { 297 res = -EINVAL; 298 goto fail; 299 } 300 301 res = dfc->power_ops->get_real_power(df, power, freq, voltage); 302 if (!res) { 303 state = dfc->capped_state; 304 dfc->res_util = dfc->power_table[state]; 305 dfc->res_util *= SCALE_ERROR_MITIGATION; 306 307 if (*power > 1) 308 dfc->res_util /= *power; 309 } else { 310 goto fail; 311 } 312 } else { 313 dyn_power = dfc->power_table[state]; 314 315 /* Scale dynamic power for utilization */ 316 dyn_power *= status->busy_time; 317 dyn_power /= status->total_time; 318 /* Get static power */ 319 static_power = get_static_power(dfc, freq); 320 321 *power = dyn_power + static_power; 322 } 323 324 trace_thermal_power_devfreq_get_power(cdev, status, freq, dyn_power, 325 static_power, *power); 326 327 return 0; 328 fail: 329 /* It is safe to set max in this case */ 330 dfc->res_util = SCALE_ERROR_MITIGATION; 331 return res; 332 } 333 334 static int devfreq_cooling_state2power(struct thermal_cooling_device *cdev, 335 struct thermal_zone_device *tz, 336 unsigned long state, 337 u32 *power) 338 { 339 struct devfreq_cooling_device *dfc = cdev->devdata; 340 unsigned long freq; 341 u32 static_power; 342 343 if (state >= dfc->freq_table_size) 344 return -EINVAL; 345 346 freq = dfc->freq_table[state]; 347 static_power = get_static_power(dfc, freq); 348 349 *power = dfc->power_table[state] + static_power; 350 return 0; 351 } 352 353 static int devfreq_cooling_power2state(struct thermal_cooling_device *cdev, 354 struct thermal_zone_device *tz, 355 u32 power, unsigned long *state) 356 { 357 struct devfreq_cooling_device *dfc = cdev->devdata; 358 struct devfreq *df = dfc->devfreq; 359 struct devfreq_dev_status *status = &df->last_status; 360 unsigned long freq = status->current_frequency; 361 unsigned long busy_time; 362 s32 dyn_power; 363 u32 static_power; 364 s32 est_power; 365 int i; 366 367 if (dfc->power_ops->get_real_power) { 368 /* Scale for resource utilization */ 369 est_power = power * dfc->res_util; 370 est_power /= SCALE_ERROR_MITIGATION; 371 } else { 372 static_power = get_static_power(dfc, freq); 373 374 dyn_power = power - static_power; 375 dyn_power = dyn_power > 0 ? dyn_power : 0; 376 377 /* Scale dynamic power for utilization */ 378 busy_time = status->busy_time ?: 1; 379 est_power = (dyn_power * status->total_time) / busy_time; 380 } 381 382 /* 383 * Find the first cooling state that is within the power 384 * budget for dynamic power. 385 */ 386 for (i = 0; i < dfc->freq_table_size - 1; i++) 387 if (est_power >= dfc->power_table[i]) 388 break; 389 390 *state = i; 391 dfc->capped_state = i; 392 trace_thermal_power_devfreq_limit(cdev, freq, *state, power); 393 return 0; 394 } 395 396 static struct thermal_cooling_device_ops devfreq_cooling_ops = { 397 .get_max_state = devfreq_cooling_get_max_state, 398 .get_cur_state = devfreq_cooling_get_cur_state, 399 .set_cur_state = devfreq_cooling_set_cur_state, 400 }; 401 402 /** 403 * devfreq_cooling_gen_tables() - Generate power and freq tables. 404 * @dfc: Pointer to devfreq cooling device. 405 * 406 * Generate power and frequency tables: the power table hold the 407 * device's maximum power usage at each cooling state (OPP). The 408 * static and dynamic power using the appropriate voltage and 409 * frequency for the state, is acquired from the struct 410 * devfreq_cooling_power, and summed to make the maximum power draw. 411 * 412 * The frequency table holds the frequencies in descending order. 413 * That way its indexed by cooling device state. 414 * 415 * The tables are malloced, and pointers put in dfc. They must be 416 * freed when unregistering the devfreq cooling device. 417 * 418 * Return: 0 on success, negative error code on failure. 419 */ 420 static int devfreq_cooling_gen_tables(struct devfreq_cooling_device *dfc) 421 { 422 struct devfreq *df = dfc->devfreq; 423 struct device *dev = df->dev.parent; 424 int ret, num_opps; 425 unsigned long freq; 426 u32 *power_table = NULL; 427 u32 *freq_table; 428 int i; 429 430 num_opps = dev_pm_opp_get_opp_count(dev); 431 432 if (dfc->power_ops) { 433 power_table = kcalloc(num_opps, sizeof(*power_table), 434 GFP_KERNEL); 435 if (!power_table) 436 return -ENOMEM; 437 } 438 439 freq_table = kcalloc(num_opps, sizeof(*freq_table), 440 GFP_KERNEL); 441 if (!freq_table) { 442 ret = -ENOMEM; 443 goto free_power_table; 444 } 445 446 for (i = 0, freq = ULONG_MAX; i < num_opps; i++, freq--) { 447 unsigned long power, voltage; 448 struct dev_pm_opp *opp; 449 450 opp = dev_pm_opp_find_freq_floor(dev, &freq); 451 if (IS_ERR(opp)) { 452 ret = PTR_ERR(opp); 453 goto free_tables; 454 } 455 456 voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */ 457 dev_pm_opp_put(opp); 458 459 if (dfc->power_ops) { 460 if (dfc->power_ops->get_real_power) 461 power = get_total_power(dfc, freq, voltage); 462 else 463 power = get_dynamic_power(dfc, freq, voltage); 464 465 dev_dbg(dev, "Power table: %lu MHz @ %lu mV: %lu = %lu mW\n", 466 freq / 1000000, voltage, power, power); 467 468 power_table[i] = power; 469 } 470 471 freq_table[i] = freq; 472 } 473 474 if (dfc->power_ops) 475 dfc->power_table = power_table; 476 477 dfc->freq_table = freq_table; 478 dfc->freq_table_size = num_opps; 479 480 return 0; 481 482 free_tables: 483 kfree(freq_table); 484 free_power_table: 485 kfree(power_table); 486 487 return ret; 488 } 489 490 /** 491 * of_devfreq_cooling_register_power() - Register devfreq cooling device, 492 * with OF and power information. 493 * @np: Pointer to OF device_node. 494 * @df: Pointer to devfreq device. 495 * @dfc_power: Pointer to devfreq_cooling_power. 496 * 497 * Register a devfreq cooling device. The available OPPs must be 498 * registered on the device. 499 * 500 * If @dfc_power is provided, the cooling device is registered with the 501 * power extensions. For the power extensions to work correctly, 502 * devfreq should use the simple_ondemand governor, other governors 503 * are not currently supported. 504 */ 505 struct thermal_cooling_device * 506 of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df, 507 struct devfreq_cooling_power *dfc_power) 508 { 509 struct thermal_cooling_device *cdev; 510 struct devfreq_cooling_device *dfc; 511 char dev_name[THERMAL_NAME_LENGTH]; 512 int err; 513 514 dfc = kzalloc(sizeof(*dfc), GFP_KERNEL); 515 if (!dfc) 516 return ERR_PTR(-ENOMEM); 517 518 dfc->devfreq = df; 519 520 if (dfc_power) { 521 dfc->power_ops = dfc_power; 522 523 devfreq_cooling_ops.get_requested_power = 524 devfreq_cooling_get_requested_power; 525 devfreq_cooling_ops.state2power = devfreq_cooling_state2power; 526 devfreq_cooling_ops.power2state = devfreq_cooling_power2state; 527 } 528 529 err = devfreq_cooling_gen_tables(dfc); 530 if (err) 531 goto free_dfc; 532 533 err = ida_simple_get(&devfreq_ida, 0, 0, GFP_KERNEL); 534 if (err < 0) 535 goto free_tables; 536 dfc->id = err; 537 538 snprintf(dev_name, sizeof(dev_name), "thermal-devfreq-%d", dfc->id); 539 540 cdev = thermal_of_cooling_device_register(np, dev_name, dfc, 541 &devfreq_cooling_ops); 542 if (IS_ERR(cdev)) { 543 err = PTR_ERR(cdev); 544 dev_err(df->dev.parent, 545 "Failed to register devfreq cooling device (%d)\n", 546 err); 547 goto release_ida; 548 } 549 550 dfc->cdev = cdev; 551 552 return cdev; 553 554 release_ida: 555 ida_simple_remove(&devfreq_ida, dfc->id); 556 free_tables: 557 kfree(dfc->power_table); 558 kfree(dfc->freq_table); 559 free_dfc: 560 kfree(dfc); 561 562 return ERR_PTR(err); 563 } 564 EXPORT_SYMBOL_GPL(of_devfreq_cooling_register_power); 565 566 /** 567 * of_devfreq_cooling_register() - Register devfreq cooling device, 568 * with OF information. 569 * @np: Pointer to OF device_node. 570 * @df: Pointer to devfreq device. 571 */ 572 struct thermal_cooling_device * 573 of_devfreq_cooling_register(struct device_node *np, struct devfreq *df) 574 { 575 return of_devfreq_cooling_register_power(np, df, NULL); 576 } 577 EXPORT_SYMBOL_GPL(of_devfreq_cooling_register); 578 579 /** 580 * devfreq_cooling_register() - Register devfreq cooling device. 581 * @df: Pointer to devfreq device. 582 */ 583 struct thermal_cooling_device *devfreq_cooling_register(struct devfreq *df) 584 { 585 return of_devfreq_cooling_register(NULL, df); 586 } 587 EXPORT_SYMBOL_GPL(devfreq_cooling_register); 588 589 /** 590 * devfreq_cooling_unregister() - Unregister devfreq cooling device. 591 * @cdev: Pointer to devfreq cooling device to unregister. 592 */ 593 void devfreq_cooling_unregister(struct thermal_cooling_device *cdev) 594 { 595 struct devfreq_cooling_device *dfc; 596 597 if (!cdev) 598 return; 599 600 dfc = cdev->devdata; 601 602 thermal_cooling_device_unregister(dfc->cdev); 603 ida_simple_remove(&devfreq_ida, dfc->id); 604 kfree(dfc->power_table); 605 kfree(dfc->freq_table); 606 607 kfree(dfc); 608 } 609 EXPORT_SYMBOL_GPL(devfreq_cooling_unregister); 610