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/energy_model.h>
16 #include <linux/export.h>
17 #include <linux/slab.h>
18 #include <linux/pm_opp.h>
19 #include <linux/pm_qos.h>
20 #include <linux/thermal.h>
21 #include <linux/units.h>
22 
23 #include "thermal_trace.h"
24 
25 #define SCALE_ERROR_MITIGATION	100
26 
27 /**
28  * struct devfreq_cooling_device - Devfreq cooling device
29  *		devfreq_cooling_device registered.
30  * @cdev:	Pointer to associated thermal cooling device.
31  * @cooling_ops: devfreq callbacks to thermal cooling device ops
32  * @devfreq:	Pointer to associated devfreq device.
33  * @cooling_state:	Current cooling state.
34  * @freq_table:	Pointer to a table with the frequencies sorted in descending
35  *		order.  You can index the table by cooling device state
36  * @max_state:	It is the last index, that is, one less than the number of the
37  *		OPPs
38  * @power_ops:	Pointer to devfreq_cooling_power, a more precised model.
39  * @res_util:	Resource utilization scaling factor for the power.
40  *		It is multiplied by 100 to minimize the error. It is used
41  *		for estimation of the power budget instead of using
42  *		'utilization' (which is	'busy_time' / 'total_time').
43  *		The 'res_util' range is from 100 to power * 100	for the
44  *		corresponding 'state'.
45  * @capped_state:	index to cooling state with in dynamic power budget
46  * @req_max_freq:	PM QoS request for limiting the maximum frequency
47  *			of the devfreq device.
48  * @em_pd:		Energy Model for the associated Devfreq device
49  */
50 struct devfreq_cooling_device {
51 	struct thermal_cooling_device *cdev;
52 	struct thermal_cooling_device_ops cooling_ops;
53 	struct devfreq *devfreq;
54 	unsigned long cooling_state;
55 	u32 *freq_table;
56 	size_t max_state;
57 	struct devfreq_cooling_power *power_ops;
58 	u32 res_util;
59 	int capped_state;
60 	struct dev_pm_qos_request req_max_freq;
61 	struct em_perf_domain *em_pd;
62 };
63 
devfreq_cooling_get_max_state(struct thermal_cooling_device * cdev,unsigned long * state)64 static int devfreq_cooling_get_max_state(struct thermal_cooling_device *cdev,
65 					 unsigned long *state)
66 {
67 	struct devfreq_cooling_device *dfc = cdev->devdata;
68 
69 	*state = dfc->max_state;
70 
71 	return 0;
72 }
73 
devfreq_cooling_get_cur_state(struct thermal_cooling_device * cdev,unsigned long * state)74 static int devfreq_cooling_get_cur_state(struct thermal_cooling_device *cdev,
75 					 unsigned long *state)
76 {
77 	struct devfreq_cooling_device *dfc = cdev->devdata;
78 
79 	*state = dfc->cooling_state;
80 
81 	return 0;
82 }
83 
devfreq_cooling_set_cur_state(struct thermal_cooling_device * cdev,unsigned long state)84 static int devfreq_cooling_set_cur_state(struct thermal_cooling_device *cdev,
85 					 unsigned long state)
86 {
87 	struct devfreq_cooling_device *dfc = cdev->devdata;
88 	struct devfreq *df = dfc->devfreq;
89 	struct device *dev = df->dev.parent;
90 	unsigned long freq;
91 	int perf_idx;
92 
93 	if (state == dfc->cooling_state)
94 		return 0;
95 
96 	dev_dbg(dev, "Setting cooling state %lu\n", state);
97 
98 	if (state > dfc->max_state)
99 		return -EINVAL;
100 
101 	if (dfc->em_pd) {
102 		perf_idx = dfc->max_state - state;
103 		freq = dfc->em_pd->table[perf_idx].frequency * 1000;
104 	} else {
105 		freq = dfc->freq_table[state];
106 	}
107 
108 	dev_pm_qos_update_request(&dfc->req_max_freq,
109 				  DIV_ROUND_UP(freq, HZ_PER_KHZ));
110 
111 	dfc->cooling_state = state;
112 
113 	return 0;
114 }
115 
116 /**
117  * get_perf_idx() - get the performance index corresponding to a frequency
118  * @em_pd:	Pointer to device's Energy Model
119  * @freq:	frequency in kHz
120  *
121  * Return: the performance index associated with the @freq, or
122  * -EINVAL if it wasn't found.
123  */
get_perf_idx(struct em_perf_domain * em_pd,unsigned long freq)124 static int get_perf_idx(struct em_perf_domain *em_pd, unsigned long freq)
125 {
126 	int i;
127 
128 	for (i = 0; i < em_pd->nr_perf_states; i++) {
129 		if (em_pd->table[i].frequency == freq)
130 			return i;
131 	}
132 
133 	return -EINVAL;
134 }
135 
get_voltage(struct devfreq * df,unsigned long freq)136 static unsigned long get_voltage(struct devfreq *df, unsigned long freq)
137 {
138 	struct device *dev = df->dev.parent;
139 	unsigned long voltage;
140 	struct dev_pm_opp *opp;
141 
142 	opp = dev_pm_opp_find_freq_exact(dev, freq, true);
143 	if (PTR_ERR(opp) == -ERANGE)
144 		opp = dev_pm_opp_find_freq_exact(dev, freq, false);
145 
146 	if (IS_ERR(opp)) {
147 		dev_err_ratelimited(dev, "Failed to find OPP for frequency %lu: %ld\n",
148 				    freq, PTR_ERR(opp));
149 		return 0;
150 	}
151 
152 	voltage = dev_pm_opp_get_voltage(opp) / 1000; /* mV */
153 	dev_pm_opp_put(opp);
154 
155 	if (voltage == 0) {
156 		dev_err_ratelimited(dev,
157 				    "Failed to get voltage for frequency %lu\n",
158 				    freq);
159 	}
160 
161 	return voltage;
162 }
163 
_normalize_load(struct devfreq_dev_status * status)164 static void _normalize_load(struct devfreq_dev_status *status)
165 {
166 	if (status->total_time > 0xfffff) {
167 		status->total_time >>= 10;
168 		status->busy_time >>= 10;
169 	}
170 
171 	status->busy_time <<= 10;
172 	status->busy_time /= status->total_time ? : 1;
173 
174 	status->busy_time = status->busy_time ? : 1;
175 	status->total_time = 1024;
176 }
177 
devfreq_cooling_get_requested_power(struct thermal_cooling_device * cdev,u32 * power)178 static int devfreq_cooling_get_requested_power(struct thermal_cooling_device *cdev,
179 					       u32 *power)
180 {
181 	struct devfreq_cooling_device *dfc = cdev->devdata;
182 	struct devfreq *df = dfc->devfreq;
183 	struct devfreq_dev_status status;
184 	unsigned long state;
185 	unsigned long freq;
186 	unsigned long voltage;
187 	int res, perf_idx;
188 
189 	mutex_lock(&df->lock);
190 	status = df->last_status;
191 	mutex_unlock(&df->lock);
192 
193 	freq = status.current_frequency;
194 
195 	if (dfc->power_ops && dfc->power_ops->get_real_power) {
196 		voltage = get_voltage(df, freq);
197 		if (voltage == 0) {
198 			res = -EINVAL;
199 			goto fail;
200 		}
201 
202 		res = dfc->power_ops->get_real_power(df, power, freq, voltage);
203 		if (!res) {
204 			state = dfc->max_state - dfc->capped_state;
205 
206 			/* Convert EM power into milli-Watts first */
207 			dfc->res_util = dfc->em_pd->table[state].power;
208 			dfc->res_util /= MICROWATT_PER_MILLIWATT;
209 
210 			dfc->res_util *= SCALE_ERROR_MITIGATION;
211 
212 			if (*power > 1)
213 				dfc->res_util /= *power;
214 		} else {
215 			goto fail;
216 		}
217 	} else {
218 		/* Energy Model frequencies are in kHz */
219 		perf_idx = get_perf_idx(dfc->em_pd, freq / 1000);
220 		if (perf_idx < 0) {
221 			res = -EAGAIN;
222 			goto fail;
223 		}
224 
225 		_normalize_load(&status);
226 
227 		/* Convert EM power into milli-Watts first */
228 		*power = dfc->em_pd->table[perf_idx].power;
229 		*power /= MICROWATT_PER_MILLIWATT;
230 		/* Scale power for utilization */
231 		*power *= status.busy_time;
232 		*power >>= 10;
233 	}
234 
235 	trace_thermal_power_devfreq_get_power(cdev, &status, freq, *power);
236 
237 	return 0;
238 fail:
239 	/* It is safe to set max in this case */
240 	dfc->res_util = SCALE_ERROR_MITIGATION;
241 	return res;
242 }
243 
devfreq_cooling_state2power(struct thermal_cooling_device * cdev,unsigned long state,u32 * power)244 static int devfreq_cooling_state2power(struct thermal_cooling_device *cdev,
245 				       unsigned long state, u32 *power)
246 {
247 	struct devfreq_cooling_device *dfc = cdev->devdata;
248 	int perf_idx;
249 
250 	if (state > dfc->max_state)
251 		return -EINVAL;
252 
253 	perf_idx = dfc->max_state - state;
254 	*power = dfc->em_pd->table[perf_idx].power;
255 	*power /= MICROWATT_PER_MILLIWATT;
256 
257 	return 0;
258 }
259 
devfreq_cooling_power2state(struct thermal_cooling_device * cdev,u32 power,unsigned long * state)260 static int devfreq_cooling_power2state(struct thermal_cooling_device *cdev,
261 				       u32 power, unsigned long *state)
262 {
263 	struct devfreq_cooling_device *dfc = cdev->devdata;
264 	struct devfreq *df = dfc->devfreq;
265 	struct devfreq_dev_status status;
266 	unsigned long freq, em_power_mw;
267 	s32 est_power;
268 	int i;
269 
270 	mutex_lock(&df->lock);
271 	status = df->last_status;
272 	mutex_unlock(&df->lock);
273 
274 	freq = status.current_frequency;
275 
276 	if (dfc->power_ops && dfc->power_ops->get_real_power) {
277 		/* Scale for resource utilization */
278 		est_power = power * dfc->res_util;
279 		est_power /= SCALE_ERROR_MITIGATION;
280 	} else {
281 		/* Scale dynamic power for utilization */
282 		_normalize_load(&status);
283 		est_power = power << 10;
284 		est_power /= status.busy_time;
285 	}
286 
287 	/*
288 	 * Find the first cooling state that is within the power
289 	 * budget. The EM power table is sorted ascending.
290 	 */
291 	for (i = dfc->max_state; i > 0; i--) {
292 		/* Convert EM power to milli-Watts to make safe comparison */
293 		em_power_mw = dfc->em_pd->table[i].power;
294 		em_power_mw /= MICROWATT_PER_MILLIWATT;
295 		if (est_power >= em_power_mw)
296 			break;
297 	}
298 
299 	*state = dfc->max_state - i;
300 	dfc->capped_state = *state;
301 
302 	trace_thermal_power_devfreq_limit(cdev, freq, *state, power);
303 	return 0;
304 }
305 
306 /**
307  * devfreq_cooling_gen_tables() - Generate frequency table.
308  * @dfc:	Pointer to devfreq cooling device.
309  * @num_opps:	Number of OPPs
310  *
311  * Generate frequency table which holds the frequencies in descending
312  * order. That way its indexed by cooling device state. This is for
313  * compatibility with drivers which do not register Energy Model.
314  *
315  * Return: 0 on success, negative error code on failure.
316  */
devfreq_cooling_gen_tables(struct devfreq_cooling_device * dfc,int num_opps)317 static int devfreq_cooling_gen_tables(struct devfreq_cooling_device *dfc,
318 				      int num_opps)
319 {
320 	struct devfreq *df = dfc->devfreq;
321 	struct device *dev = df->dev.parent;
322 	unsigned long freq;
323 	int i;
324 
325 	dfc->freq_table = kcalloc(num_opps, sizeof(*dfc->freq_table),
326 			     GFP_KERNEL);
327 	if (!dfc->freq_table)
328 		return -ENOMEM;
329 
330 	for (i = 0, freq = ULONG_MAX; i < num_opps; i++, freq--) {
331 		struct dev_pm_opp *opp;
332 
333 		opp = dev_pm_opp_find_freq_floor(dev, &freq);
334 		if (IS_ERR(opp)) {
335 			kfree(dfc->freq_table);
336 			return PTR_ERR(opp);
337 		}
338 
339 		dev_pm_opp_put(opp);
340 		dfc->freq_table[i] = freq;
341 	}
342 
343 	return 0;
344 }
345 
346 /**
347  * of_devfreq_cooling_register_power() - Register devfreq cooling device,
348  *                                      with OF and power information.
349  * @np:	Pointer to OF device_node.
350  * @df:	Pointer to devfreq device.
351  * @dfc_power:	Pointer to devfreq_cooling_power.
352  *
353  * Register a devfreq cooling device.  The available OPPs must be
354  * registered on the device.
355  *
356  * If @dfc_power is provided, the cooling device is registered with the
357  * power extensions.  For the power extensions to work correctly,
358  * devfreq should use the simple_ondemand governor, other governors
359  * are not currently supported.
360  */
361 struct thermal_cooling_device *
of_devfreq_cooling_register_power(struct device_node * np,struct devfreq * df,struct devfreq_cooling_power * dfc_power)362 of_devfreq_cooling_register_power(struct device_node *np, struct devfreq *df,
363 				  struct devfreq_cooling_power *dfc_power)
364 {
365 	struct thermal_cooling_device *cdev;
366 	struct device *dev = df->dev.parent;
367 	struct devfreq_cooling_device *dfc;
368 	struct em_perf_domain *em;
369 	struct thermal_cooling_device_ops *ops;
370 	char *name;
371 	int err, num_opps;
372 
373 
374 	dfc = kzalloc(sizeof(*dfc), GFP_KERNEL);
375 	if (!dfc)
376 		return ERR_PTR(-ENOMEM);
377 
378 	dfc->devfreq = df;
379 
380 	ops = &dfc->cooling_ops;
381 	ops->get_max_state = devfreq_cooling_get_max_state;
382 	ops->get_cur_state = devfreq_cooling_get_cur_state;
383 	ops->set_cur_state = devfreq_cooling_set_cur_state;
384 
385 	em = em_pd_get(dev);
386 	if (em && !em_is_artificial(em)) {
387 		dfc->em_pd = em;
388 		ops->get_requested_power =
389 			devfreq_cooling_get_requested_power;
390 		ops->state2power = devfreq_cooling_state2power;
391 		ops->power2state = devfreq_cooling_power2state;
392 
393 		dfc->power_ops = dfc_power;
394 
395 		num_opps = em_pd_nr_perf_states(dfc->em_pd);
396 	} else {
397 		/* Backward compatibility for drivers which do not use IPA */
398 		dev_dbg(dev, "missing proper EM for cooling device\n");
399 
400 		num_opps = dev_pm_opp_get_opp_count(dev);
401 
402 		err = devfreq_cooling_gen_tables(dfc, num_opps);
403 		if (err)
404 			goto free_dfc;
405 	}
406 
407 	if (num_opps <= 0) {
408 		err = -EINVAL;
409 		goto free_dfc;
410 	}
411 
412 	/* max_state is an index, not a counter */
413 	dfc->max_state = num_opps - 1;
414 
415 	err = dev_pm_qos_add_request(dev, &dfc->req_max_freq,
416 				     DEV_PM_QOS_MAX_FREQUENCY,
417 				     PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE);
418 	if (err < 0)
419 		goto free_table;
420 
421 	err = -ENOMEM;
422 	name = kasprintf(GFP_KERNEL, "devfreq-%s", dev_name(dev));
423 	if (!name)
424 		goto remove_qos_req;
425 
426 	cdev = thermal_of_cooling_device_register(np, name, dfc, ops);
427 	kfree(name);
428 
429 	if (IS_ERR(cdev)) {
430 		err = PTR_ERR(cdev);
431 		dev_err(dev,
432 			"Failed to register devfreq cooling device (%d)\n",
433 			err);
434 		goto remove_qos_req;
435 	}
436 
437 	dfc->cdev = cdev;
438 
439 	return cdev;
440 
441 remove_qos_req:
442 	dev_pm_qos_remove_request(&dfc->req_max_freq);
443 free_table:
444 	kfree(dfc->freq_table);
445 free_dfc:
446 	kfree(dfc);
447 
448 	return ERR_PTR(err);
449 }
450 EXPORT_SYMBOL_GPL(of_devfreq_cooling_register_power);
451 
452 /**
453  * of_devfreq_cooling_register() - Register devfreq cooling device,
454  *                                with OF information.
455  * @np: Pointer to OF device_node.
456  * @df: Pointer to devfreq device.
457  */
458 struct thermal_cooling_device *
of_devfreq_cooling_register(struct device_node * np,struct devfreq * df)459 of_devfreq_cooling_register(struct device_node *np, struct devfreq *df)
460 {
461 	return of_devfreq_cooling_register_power(np, df, NULL);
462 }
463 EXPORT_SYMBOL_GPL(of_devfreq_cooling_register);
464 
465 /**
466  * devfreq_cooling_register() - Register devfreq cooling device.
467  * @df: Pointer to devfreq device.
468  */
devfreq_cooling_register(struct devfreq * df)469 struct thermal_cooling_device *devfreq_cooling_register(struct devfreq *df)
470 {
471 	return of_devfreq_cooling_register(NULL, df);
472 }
473 EXPORT_SYMBOL_GPL(devfreq_cooling_register);
474 
475 /**
476  * devfreq_cooling_em_register() - Register devfreq cooling device with
477  *		power information and automatically register Energy Model (EM)
478  * @df:		Pointer to devfreq device.
479  * @dfc_power:	Pointer to devfreq_cooling_power.
480  *
481  * Register a devfreq cooling device and automatically register EM. The
482  * available OPPs must be registered for the device.
483  *
484  * If @dfc_power is provided, the cooling device is registered with the
485  * power extensions. It is using the simple Energy Model which requires
486  * "dynamic-power-coefficient" a devicetree property. To not break drivers
487  * which miss that DT property, the function won't bail out when the EM
488  * registration failed. The cooling device will be registered if everything
489  * else is OK.
490  */
491 struct thermal_cooling_device *
devfreq_cooling_em_register(struct devfreq * df,struct devfreq_cooling_power * dfc_power)492 devfreq_cooling_em_register(struct devfreq *df,
493 			    struct devfreq_cooling_power *dfc_power)
494 {
495 	struct thermal_cooling_device *cdev;
496 	struct device *dev;
497 	int ret;
498 
499 	if (IS_ERR_OR_NULL(df))
500 		return ERR_PTR(-EINVAL);
501 
502 	dev = df->dev.parent;
503 
504 	ret = dev_pm_opp_of_register_em(dev, NULL);
505 	if (ret)
506 		dev_dbg(dev, "Unable to register EM for devfreq cooling device (%d)\n",
507 			ret);
508 
509 	cdev = of_devfreq_cooling_register_power(dev->of_node, df, dfc_power);
510 
511 	if (IS_ERR_OR_NULL(cdev))
512 		em_dev_unregister_perf_domain(dev);
513 
514 	return cdev;
515 }
516 EXPORT_SYMBOL_GPL(devfreq_cooling_em_register);
517 
518 /**
519  * devfreq_cooling_unregister() - Unregister devfreq cooling device.
520  * @cdev: Pointer to devfreq cooling device to unregister.
521  *
522  * Unregisters devfreq cooling device and related Energy Model if it was
523  * present.
524  */
devfreq_cooling_unregister(struct thermal_cooling_device * cdev)525 void devfreq_cooling_unregister(struct thermal_cooling_device *cdev)
526 {
527 	struct devfreq_cooling_device *dfc;
528 	struct device *dev;
529 
530 	if (IS_ERR_OR_NULL(cdev))
531 		return;
532 
533 	dfc = cdev->devdata;
534 	dev = dfc->devfreq->dev.parent;
535 
536 	thermal_cooling_device_unregister(dfc->cdev);
537 	dev_pm_qos_remove_request(&dfc->req_max_freq);
538 
539 	em_dev_unregister_perf_domain(dev);
540 
541 	kfree(dfc->freq_table);
542 	kfree(dfc);
543 }
544 EXPORT_SYMBOL_GPL(devfreq_cooling_unregister);
545