Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright 2023 Linaro Limited
4 *
5 * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
6 *
7 * Thermal subsystem debug support
8 */
9#include <linux/debugfs.h>
10#include <linux/ktime.h>
11#include <linux/list.h>
12#include <linux/minmax.h>
13#include <linux/mutex.h>
14#include <linux/thermal.h>
15
16#include "thermal_core.h"
17
18static struct dentry *d_root;
19static struct dentry *d_cdev;
20static struct dentry *d_tz;
21
22/*
23 * Length of the string containing the thermal zone id or the cooling
24 * device id, including the ending nul character. We can reasonably
25 * assume there won't be more than 256 thermal zones as the maximum
26 * observed today is around 32.
27 */
28#define IDSLENGTH 4
29
30/*
31 * The cooling device transition list is stored in a hash table where
32 * the size is CDEVSTATS_HASH_SIZE. The majority of cooling devices
33 * have dozen of states but some can have much more, so a hash table
34 * is more adequate in this case, because the cost of browsing the entire
35 * list when storing the transitions may not be negligible.
36 */
37#define CDEVSTATS_HASH_SIZE 16
38
39/**
40 * struct cdev_debugfs - per cooling device statistics structure
41 * A cooling device can have a high number of states. Showing the
42 * transitions on a matrix based representation can be overkill given
43 * most of the transitions won't happen and we end up with a matrix
44 * filled with zero. Instead, we show the transitions which actually
45 * happened.
46 *
47 * Every transition updates the current_state and the timestamp. The
48 * transitions and the durations are stored in lists.
49 *
50 * @total: the number of transitions for this cooling device
51 * @current_state: the current cooling device state
52 * @timestamp: the state change timestamp
53 * @transitions: an array of lists containing the state transitions
54 * @durations: an array of lists containing the residencies of each state
55 */
56struct cdev_debugfs {
57 u32 total;
58 int current_state;
59 ktime_t timestamp;
60 struct list_head transitions[CDEVSTATS_HASH_SIZE];
61 struct list_head durations[CDEVSTATS_HASH_SIZE];
62};
63
64/**
65 * struct cdev_record - Common structure for cooling device entry
66 *
67 * The following common structure allows to store the information
68 * related to the transitions and to the state residencies. They are
69 * identified with a id which is associated to a value. It is used as
70 * nodes for the "transitions" and "durations" above.
71 *
72 * @node: node to insert the structure in a list
73 * @id: identifier of the value which can be a state or a transition
74 * @residency: a ktime_t representing a state residency duration
75 * @count: a number of occurrences
76 */
77struct cdev_record {
78 struct list_head node;
79 int id;
80 union {
81 ktime_t residency;
82 u64 count;
83 };
84};
85
86/**
87 * struct trip_stats - Thermal trip statistics
88 *
89 * The trip_stats structure has the relevant information to show the
90 * statistics related to temperature going above a trip point.
91 *
92 * @timestamp: the trip crossing timestamp
93 * @duration: total time when the zone temperature was above the trip point
94 * @trip_temp: trip temperature at mitigation start
95 * @trip_hyst: trip hysteresis at mitigation start
96 * @count: the number of times the zone temperature was above the trip point
97 * @min: minimum recorded temperature above the trip point
98 * @avg: average temperature above the trip point
99 */
100struct trip_stats {
101 ktime_t timestamp;
102 ktime_t duration;
103 int trip_temp;
104 int trip_hyst;
105 int count;
106 int min;
107 int avg;
108};
109
110/**
111 * struct tz_episode - A mitigation episode information
112 *
113 * The tz_episode structure describes a mitigation episode. A
114 * mitigation episode begins the trip point with the lower temperature
115 * is crossed the way up and ends when it is crossed the way
116 * down. During this episode we can have multiple trip points crossed
117 * the way up and down if there are multiple trip described in the
118 * firmware after the lowest temperature trip point.
119 *
120 * @timestamp: first trip point crossed the way up
121 * @duration: total duration of the mitigation episode
122 * @node: a list element to be added to the list of tz events
123 * @max_temp: maximum zone temperature during this episode
124 * @trip_stats: per trip point statistics, flexible array
125 */
126struct tz_episode {
127 ktime_t timestamp;
128 ktime_t duration;
129 struct list_head node;
130 int max_temp;
131 struct trip_stats trip_stats[];
132};
133
134/**
135 * struct tz_debugfs - Store all mitigation episodes for a thermal zone
136 *
137 * The tz_debugfs structure contains the list of the mitigation
138 * episodes and has to track which trip point has been crossed in
139 * order to handle correctly nested trip point mitigation episodes.
140 *
141 * We keep the history of the trip point crossed in an array and as we
142 * can go back and forth inside this history, eg. trip 0,1,2,1,2,1,0,
143 * we keep track of the current position in the history array.
144 *
145 * @tz_episodes: a list of thermal mitigation episodes
146 * @tz: thermal zone this object belongs to
147 * @trips_crossed: an array of trip points crossed by id
148 * @nr_trips: the number of trip points currently being crossed
149 */
150struct tz_debugfs {
151 struct list_head tz_episodes;
152 struct thermal_zone_device *tz;
153 int *trips_crossed;
154 int nr_trips;
155};
156
157/**
158 * struct thermal_debugfs - High level structure for a thermal object in debugfs
159 *
160 * The thermal_debugfs structure is the common structure used by the
161 * cooling device or the thermal zone to store the statistics.
162 *
163 * @d_top: top directory of the thermal object directory
164 * @lock: per object lock to protect the internals
165 *
166 * @cdev_dbg: a cooling device debug structure
167 * @tz_dbg: a thermal zone debug structure
168 */
169struct thermal_debugfs {
170 struct dentry *d_top;
171 struct mutex lock;
172 union {
173 struct cdev_debugfs cdev_dbg;
174 struct tz_debugfs tz_dbg;
175 };
176};
177
178void thermal_debug_init(void)
179{
180 d_root = debugfs_create_dir("thermal", NULL);
181 if (IS_ERR(d_root))
182 return;
183
184 d_cdev = debugfs_create_dir("cooling_devices", d_root);
185 if (IS_ERR(d_cdev))
186 return;
187
188 d_tz = debugfs_create_dir("thermal_zones", d_root);
189}
190
191static struct thermal_debugfs *thermal_debugfs_add_id(struct dentry *d, int id)
192{
193 struct thermal_debugfs *thermal_dbg;
194 char ids[IDSLENGTH];
195
196 thermal_dbg = kzalloc(sizeof(*thermal_dbg), GFP_KERNEL);
197 if (!thermal_dbg)
198 return NULL;
199
200 mutex_init(&thermal_dbg->lock);
201
202 snprintf(ids, IDSLENGTH, "%d", id);
203
204 thermal_dbg->d_top = debugfs_create_dir(ids, d);
205 if (IS_ERR(thermal_dbg->d_top)) {
206 kfree(thermal_dbg);
207 return NULL;
208 }
209
210 return thermal_dbg;
211}
212
213static void thermal_debugfs_remove_id(struct thermal_debugfs *thermal_dbg)
214{
215 if (!thermal_dbg)
216 return;
217
218 debugfs_remove(thermal_dbg->d_top);
219
220 kfree(thermal_dbg);
221}
222
223static struct cdev_record *
224thermal_debugfs_cdev_record_alloc(struct thermal_debugfs *thermal_dbg,
225 struct list_head *lists, int id)
226{
227 struct cdev_record *cdev_record;
228
229 cdev_record = kzalloc(sizeof(*cdev_record), GFP_KERNEL);
230 if (!cdev_record)
231 return NULL;
232
233 cdev_record->id = id;
234 INIT_LIST_HEAD(&cdev_record->node);
235 list_add_tail(&cdev_record->node,
236 &lists[cdev_record->id % CDEVSTATS_HASH_SIZE]);
237
238 return cdev_record;
239}
240
241static struct cdev_record *
242thermal_debugfs_cdev_record_find(struct thermal_debugfs *thermal_dbg,
243 struct list_head *lists, int id)
244{
245 struct cdev_record *entry;
246
247 list_for_each_entry(entry, &lists[id % CDEVSTATS_HASH_SIZE], node)
248 if (entry->id == id)
249 return entry;
250
251 return NULL;
252}
253
254static struct cdev_record *
255thermal_debugfs_cdev_record_get(struct thermal_debugfs *thermal_dbg,
256 struct list_head *lists, int id)
257{
258 struct cdev_record *cdev_record;
259
260 cdev_record = thermal_debugfs_cdev_record_find(thermal_dbg, lists, id);
261 if (cdev_record)
262 return cdev_record;
263
264 return thermal_debugfs_cdev_record_alloc(thermal_dbg, lists, id);
265}
266
267static void thermal_debugfs_cdev_clear(struct cdev_debugfs *cdev_dbg)
268{
269 int i;
270 struct cdev_record *entry, *tmp;
271
272 for (i = 0; i < CDEVSTATS_HASH_SIZE; i++) {
273
274 list_for_each_entry_safe(entry, tmp,
275 &cdev_dbg->transitions[i], node) {
276 list_del(&entry->node);
277 kfree(entry);
278 }
279
280 list_for_each_entry_safe(entry, tmp,
281 &cdev_dbg->durations[i], node) {
282 list_del(&entry->node);
283 kfree(entry);
284 }
285 }
286
287 cdev_dbg->total = 0;
288}
289
290static void *cdev_seq_start(struct seq_file *s, loff_t *pos)
291{
292 struct thermal_debugfs *thermal_dbg = s->private;
293
294 mutex_lock(&thermal_dbg->lock);
295
296 return (*pos < CDEVSTATS_HASH_SIZE) ? pos : NULL;
297}
298
299static void *cdev_seq_next(struct seq_file *s, void *v, loff_t *pos)
300{
301 (*pos)++;
302
303 return (*pos < CDEVSTATS_HASH_SIZE) ? pos : NULL;
304}
305
306static void cdev_seq_stop(struct seq_file *s, void *v)
307{
308 struct thermal_debugfs *thermal_dbg = s->private;
309
310 mutex_unlock(&thermal_dbg->lock);
311}
312
313static int cdev_tt_seq_show(struct seq_file *s, void *v)
314{
315 struct thermal_debugfs *thermal_dbg = s->private;
316 struct cdev_debugfs *cdev_dbg = &thermal_dbg->cdev_dbg;
317 struct list_head *transitions = cdev_dbg->transitions;
318 struct cdev_record *entry;
319 int i = *(loff_t *)v;
320
321 if (!i)
322 seq_puts(s, "Transition\tOccurences\n");
323
324 list_for_each_entry(entry, &transitions[i], node) {
325 /*
326 * Assuming maximum cdev states is 1024, the longer
327 * string for a transition would be "1024->1024\0"
328 */
329 char buffer[11];
330
331 snprintf(buffer, ARRAY_SIZE(buffer), "%d->%d",
332 entry->id >> 16, entry->id & 0xFFFF);
333
334 seq_printf(s, "%-10s\t%-10llu\n", buffer, entry->count);
335 }
336
337 return 0;
338}
339
340static const struct seq_operations tt_sops = {
341 .start = cdev_seq_start,
342 .next = cdev_seq_next,
343 .stop = cdev_seq_stop,
344 .show = cdev_tt_seq_show,
345};
346
347DEFINE_SEQ_ATTRIBUTE(tt);
348
349static int cdev_dt_seq_show(struct seq_file *s, void *v)
350{
351 struct thermal_debugfs *thermal_dbg = s->private;
352 struct cdev_debugfs *cdev_dbg = &thermal_dbg->cdev_dbg;
353 struct list_head *durations = cdev_dbg->durations;
354 struct cdev_record *entry;
355 int i = *(loff_t *)v;
356
357 if (!i)
358 seq_puts(s, "State\tResidency\n");
359
360 list_for_each_entry(entry, &durations[i], node) {
361 s64 duration = ktime_to_ms(entry->residency);
362
363 if (entry->id == cdev_dbg->current_state)
364 duration += ktime_ms_delta(ktime_get(),
365 cdev_dbg->timestamp);
366
367 seq_printf(s, "%-5d\t%-10llu\n", entry->id, duration);
368 }
369
370 return 0;
371}
372
373static const struct seq_operations dt_sops = {
374 .start = cdev_seq_start,
375 .next = cdev_seq_next,
376 .stop = cdev_seq_stop,
377 .show = cdev_dt_seq_show,
378};
379
380DEFINE_SEQ_ATTRIBUTE(dt);
381
382static int cdev_clear_set(void *data, u64 val)
383{
384 struct thermal_debugfs *thermal_dbg = data;
385
386 if (!val)
387 return -EINVAL;
388
389 mutex_lock(&thermal_dbg->lock);
390
391 thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
392
393 mutex_unlock(&thermal_dbg->lock);
394
395 return 0;
396}
397
398DEFINE_DEBUGFS_ATTRIBUTE(cdev_clear_fops, NULL, cdev_clear_set, "%llu\n");
399
400/**
401 * thermal_debug_cdev_state_update - Update a cooling device state change
402 *
403 * Computes a transition and the duration of the previous state residency.
404 *
405 * @cdev : a pointer to a cooling device
406 * @new_state: an integer corresponding to the new cooling device state
407 */
408void thermal_debug_cdev_state_update(const struct thermal_cooling_device *cdev,
409 int new_state)
410{
411 struct thermal_debugfs *thermal_dbg = cdev->debugfs;
412 struct cdev_debugfs *cdev_dbg;
413 struct cdev_record *cdev_record;
414 int transition, old_state;
415
416 if (!thermal_dbg || (thermal_dbg->cdev_dbg.current_state == new_state))
417 return;
418
419 mutex_lock(&thermal_dbg->lock);
420
421 cdev_dbg = &thermal_dbg->cdev_dbg;
422
423 old_state = cdev_dbg->current_state;
424
425 /*
426 * Get the old state information in the durations list. If
427 * this one does not exist, a new allocated one will be
428 * returned. Recompute the total duration in the old state and
429 * get a new timestamp for the new state.
430 */
431 cdev_record = thermal_debugfs_cdev_record_get(thermal_dbg,
432 cdev_dbg->durations,
433 old_state);
434 if (cdev_record) {
435 ktime_t now = ktime_get();
436 ktime_t delta = ktime_sub(now, cdev_dbg->timestamp);
437 cdev_record->residency = ktime_add(cdev_record->residency, delta);
438 cdev_dbg->timestamp = now;
439 }
440
441 cdev_dbg->current_state = new_state;
442
443 /*
444 * Create a record for the new state if it is not there, so its
445 * duration will be printed by cdev_dt_seq_show() as expected if it
446 * runs before the next state transition.
447 */
448 thermal_debugfs_cdev_record_get(thermal_dbg, cdev_dbg->durations, new_state);
449
450 transition = (old_state << 16) | new_state;
451
452 /*
453 * Get the transition in the transitions list. If this one
454 * does not exist, a new allocated one will be returned.
455 * Increment the occurrence of this transition which is stored
456 * in the value field.
457 */
458 cdev_record = thermal_debugfs_cdev_record_get(thermal_dbg,
459 cdev_dbg->transitions,
460 transition);
461 if (cdev_record)
462 cdev_record->count++;
463
464 cdev_dbg->total++;
465
466 mutex_unlock(&thermal_dbg->lock);
467}
468
469/**
470 * thermal_debug_cdev_add - Add a cooling device debugfs entry
471 *
472 * Allocates a cooling device object for debug, initializes the
473 * statistics and create the entries in sysfs.
474 * @cdev: a pointer to a cooling device
475 * @state: current state of the cooling device
476 */
477void thermal_debug_cdev_add(struct thermal_cooling_device *cdev, int state)
478{
479 struct thermal_debugfs *thermal_dbg;
480 struct cdev_debugfs *cdev_dbg;
481 int i;
482
483 thermal_dbg = thermal_debugfs_add_id(d_cdev, cdev->id);
484 if (!thermal_dbg)
485 return;
486
487 cdev_dbg = &thermal_dbg->cdev_dbg;
488
489 for (i = 0; i < CDEVSTATS_HASH_SIZE; i++) {
490 INIT_LIST_HEAD(&cdev_dbg->transitions[i]);
491 INIT_LIST_HEAD(&cdev_dbg->durations[i]);
492 }
493
494 cdev_dbg->current_state = state;
495 cdev_dbg->timestamp = ktime_get();
496
497 /*
498 * Create a record for the initial cooling device state, so its
499 * duration will be printed by cdev_dt_seq_show() as expected if it
500 * runs before the first state transition.
501 */
502 thermal_debugfs_cdev_record_get(thermal_dbg, cdev_dbg->durations, state);
503
504 debugfs_create_file("trans_table", 0400, thermal_dbg->d_top,
505 thermal_dbg, &tt_fops);
506
507 debugfs_create_file("time_in_state_ms", 0400, thermal_dbg->d_top,
508 thermal_dbg, &dt_fops);
509
510 debugfs_create_file("clear", 0200, thermal_dbg->d_top,
511 thermal_dbg, &cdev_clear_fops);
512
513 debugfs_create_u32("total_trans", 0400, thermal_dbg->d_top,
514 &cdev_dbg->total);
515
516 cdev->debugfs = thermal_dbg;
517}
518
519static struct thermal_debugfs *thermal_debug_cdev_clear(struct thermal_cooling_device *cdev)
520{
521 struct thermal_debugfs *thermal_dbg;
522
523 guard(cooling_dev)(cdev);
524
525 thermal_dbg = cdev->debugfs;
526 if (thermal_dbg)
527 cdev->debugfs = NULL;
528
529 return thermal_dbg;
530}
531
532/**
533 * thermal_debug_cdev_remove - Remove a cooling device debugfs entry
534 *
535 * Frees the statistics memory data and remove the debugfs entry
536 *
537 * @cdev: a pointer to a cooling device
538 */
539void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
540{
541 struct thermal_debugfs *thermal_dbg;
542
543 thermal_dbg = thermal_debug_cdev_clear(cdev);
544 if (!thermal_dbg)
545 return;
546
547 mutex_lock(&thermal_dbg->lock);
548
549 thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
550
551 mutex_unlock(&thermal_dbg->lock);
552
553 thermal_debugfs_remove_id(thermal_dbg);
554}
555
556static struct tz_episode *thermal_debugfs_tz_event_alloc(struct thermal_zone_device *tz,
557 ktime_t now)
558{
559 struct tz_episode *tze;
560 int i;
561
562 tze = kzalloc(struct_size(tze, trip_stats, tz->num_trips), GFP_KERNEL);
563 if (!tze)
564 return NULL;
565
566 INIT_LIST_HEAD(&tze->node);
567 tze->timestamp = now;
568 tze->duration = KTIME_MIN;
569 tze->max_temp = INT_MIN;
570
571 for (i = 0; i < tz->num_trips; i++) {
572 tze->trip_stats[i].trip_temp = THERMAL_TEMP_INVALID;
573 tze->trip_stats[i].min = INT_MAX;
574 }
575
576 return tze;
577}
578
579void thermal_debug_tz_trip_up(struct thermal_zone_device *tz,
580 const struct thermal_trip *trip)
581{
582 struct thermal_debugfs *thermal_dbg = tz->debugfs;
583 int trip_id = thermal_zone_trip_id(tz, trip);
584 ktime_t now = ktime_get();
585 struct trip_stats *trip_stats;
586 struct tz_debugfs *tz_dbg;
587 struct tz_episode *tze;
588
589 if (!thermal_dbg)
590 return;
591
592 tz_dbg = &thermal_dbg->tz_dbg;
593
594 mutex_lock(&thermal_dbg->lock);
595
596 /*
597 * The mitigation is starting. A mitigation can contain
598 * several episodes where each of them is related to a
599 * temperature crossing a trip point. The episodes are
600 * nested. That means when the temperature is crossing the
601 * first trip point, the duration begins to be measured. If
602 * the temperature continues to increase and reaches the
603 * second trip point, the duration of the first trip must be
604 * also accumulated.
605 *
606 * eg.
607 *
608 * temp
609 * ^
610 * | --------
611 * trip 2 / \ ------
612 * | /| |\ /| |\
613 * trip 1 / | | `---- | | \
614 * | /| | | | | |\
615 * trip 0 / | | | | | | \
616 * | /| | | | | | | |\
617 * | / | | | | | | | | `--
618 * | / | | | | | | | |
619 * |----- | | | | | | | |
620 * | | | | | | | | |
621 * --------|-|-|--------|--------|------|-|-|------------------> time
622 * | | |<--t2-->| |<-t2'>| | |
623 * | | | |
624 * | |<------------t1------------>| |
625 * | |
626 * |<-------------t0--------------->|
627 *
628 */
629 if (!tz_dbg->nr_trips) {
630 tze = thermal_debugfs_tz_event_alloc(tz, now);
631 if (!tze)
632 goto unlock;
633
634 list_add(&tze->node, &tz_dbg->tz_episodes);
635 }
636
637 /*
638 * Each time a trip point is crossed the way up, the trip_id
639 * is stored in the trip_crossed array and the nr_trips is
640 * incremented. A nr_trips equal to zero means we are entering
641 * a mitigation episode.
642 *
643 * The trip ids may not be in the ascending order but the
644 * result in the array trips_crossed will be in the ascending
645 * temperature order. The function detecting when a trip point
646 * is crossed the way down will handle the very rare case when
647 * the trip points may have been reordered during this
648 * mitigation episode.
649 */
650 tz_dbg->trips_crossed[tz_dbg->nr_trips++] = trip_id;
651
652 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
653 trip_stats = &tze->trip_stats[trip_id];
654 trip_stats->trip_temp = trip->temperature;
655 trip_stats->trip_hyst = trip->hysteresis;
656 trip_stats->timestamp = now;
657
658unlock:
659 mutex_unlock(&thermal_dbg->lock);
660}
661
662static void tz_episode_close_trip(struct tz_episode *tze, int trip_id, ktime_t now)
663{
664 struct trip_stats *trip_stats = &tze->trip_stats[trip_id];
665 ktime_t delta = ktime_sub(now, trip_stats->timestamp);
666
667 trip_stats->duration = ktime_add(delta, trip_stats->duration);
668 /* Mark the end of mitigation for this trip point. */
669 trip_stats->timestamp = KTIME_MAX;
670}
671
672void thermal_debug_tz_trip_down(struct thermal_zone_device *tz,
673 const struct thermal_trip *trip)
674{
675 struct thermal_debugfs *thermal_dbg = tz->debugfs;
676 int trip_id = thermal_zone_trip_id(tz, trip);
677 ktime_t now = ktime_get();
678 struct tz_episode *tze;
679 struct tz_debugfs *tz_dbg;
680 int i;
681
682 if (!thermal_dbg)
683 return;
684
685 tz_dbg = &thermal_dbg->tz_dbg;
686
687 mutex_lock(&thermal_dbg->lock);
688
689 /*
690 * The temperature crosses the way down but there was not
691 * mitigation detected before. That may happen when the
692 * temperature is greater than a trip point when registering a
693 * thermal zone, which is a common use case as the kernel has
694 * no mitigation mechanism yet at boot time.
695 */
696 if (!tz_dbg->nr_trips)
697 goto out;
698
699 for (i = tz_dbg->nr_trips - 1; i >= 0; i--) {
700 if (tz_dbg->trips_crossed[i] == trip_id)
701 break;
702 }
703
704 if (i < 0)
705 goto out;
706
707 tz_dbg->nr_trips--;
708
709 if (i < tz_dbg->nr_trips)
710 tz_dbg->trips_crossed[i] = tz_dbg->trips_crossed[tz_dbg->nr_trips];
711
712 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
713
714 tz_episode_close_trip(tze, trip_id, now);
715
716 /*
717 * This event closes the mitigation as we are crossing the
718 * last trip point the way down.
719 */
720 if (!tz_dbg->nr_trips)
721 tze->duration = ktime_sub(now, tze->timestamp);
722
723out:
724 mutex_unlock(&thermal_dbg->lock);
725}
726
727void thermal_debug_update_trip_stats(struct thermal_zone_device *tz)
728{
729 struct thermal_debugfs *thermal_dbg = tz->debugfs;
730 struct tz_debugfs *tz_dbg;
731 struct tz_episode *tze;
732 int i;
733
734 if (!thermal_dbg)
735 return;
736
737 tz_dbg = &thermal_dbg->tz_dbg;
738
739 mutex_lock(&thermal_dbg->lock);
740
741 if (!tz_dbg->nr_trips)
742 goto out;
743
744 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
745
746 if (tz->temperature > tze->max_temp)
747 tze->max_temp = tz->temperature;
748
749 for (i = 0; i < tz_dbg->nr_trips; i++) {
750 int trip_id = tz_dbg->trips_crossed[i];
751 struct trip_stats *trip_stats = &tze->trip_stats[trip_id];
752
753 trip_stats->min = min(trip_stats->min, tz->temperature);
754 trip_stats->avg += (tz->temperature - trip_stats->avg) /
755 ++trip_stats->count;
756 }
757out:
758 mutex_unlock(&thermal_dbg->lock);
759}
760
761static void *tze_seq_start(struct seq_file *s, loff_t *pos)
762{
763 struct thermal_debugfs *thermal_dbg = s->private;
764 struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
765
766 mutex_lock(&thermal_dbg->lock);
767
768 return seq_list_start(&tz_dbg->tz_episodes, *pos);
769}
770
771static void *tze_seq_next(struct seq_file *s, void *v, loff_t *pos)
772{
773 struct thermal_debugfs *thermal_dbg = s->private;
774 struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
775
776 return seq_list_next(v, &tz_dbg->tz_episodes, pos);
777}
778
779static void tze_seq_stop(struct seq_file *s, void *v)
780{
781 struct thermal_debugfs *thermal_dbg = s->private;
782
783 mutex_unlock(&thermal_dbg->lock);
784}
785
786static int tze_seq_show(struct seq_file *s, void *v)
787{
788 struct thermal_debugfs *thermal_dbg = s->private;
789 struct thermal_zone_device *tz = thermal_dbg->tz_dbg.tz;
790 struct thermal_trip_desc *td;
791 struct tz_episode *tze;
792 u64 duration_ms;
793 int trip_id;
794 char c;
795
796 tze = list_entry((struct list_head *)v, struct tz_episode, node);
797
798 if (tze->duration == KTIME_MIN) {
799 /* Mitigation in progress. */
800 duration_ms = ktime_to_ms(ktime_sub(ktime_get(), tze->timestamp));
801 c = '>';
802 } else {
803 duration_ms = ktime_to_ms(tze->duration);
804 c = '=';
805 }
806
807 seq_printf(s, ",-Mitigation at %llums, duration%c%llums, max. temp=%dm°C\n",
808 ktime_to_ms(tze->timestamp), c, duration_ms, tze->max_temp);
809
810 seq_printf(s, "| trip | type | temp(m°C) | hyst(m°C) | duration(ms) | avg(m°C) | min(m°C) |\n");
811
812 for_each_trip_desc(tz, td) {
813 const struct thermal_trip *trip = &td->trip;
814 struct trip_stats *trip_stats;
815
816 /*
817 * There is no possible mitigation happening at the
818 * critical trip point, so the stats will be always
819 * zero, skip this trip point
820 */
821 if (trip->type == THERMAL_TRIP_CRITICAL)
822 continue;
823
824 trip_id = thermal_zone_trip_id(tz, trip);
825 trip_stats = &tze->trip_stats[trip_id];
826
827 /* Skip trips without any stats. */
828 if (trip_stats->trip_temp == THERMAL_TEMP_INVALID)
829 continue;
830
831 if (trip_stats->timestamp != KTIME_MAX) {
832 /* Mitigation in progress. */
833 ktime_t delta = ktime_sub(ktime_get(),
834 trip_stats->timestamp);
835
836 delta = ktime_add(delta, trip_stats->duration);
837 duration_ms = ktime_to_ms(delta);
838 c = '>';
839 } else {
840 duration_ms = ktime_to_ms(trip_stats->duration);
841 c = ' ';
842 }
843
844 seq_printf(s, "| %*d | %*s | %*d | %*d | %c%*lld | %*d | %*d |\n",
845 4 , trip_id,
846 8, thermal_trip_type_name(trip->type),
847 9, trip_stats->trip_temp,
848 9, trip_stats->trip_hyst,
849 c, 11, duration_ms,
850 9, trip_stats->avg,
851 9, trip_stats->min);
852 }
853
854 return 0;
855}
856
857static const struct seq_operations tze_sops = {
858 .start = tze_seq_start,
859 .next = tze_seq_next,
860 .stop = tze_seq_stop,
861 .show = tze_seq_show,
862};
863
864DEFINE_SEQ_ATTRIBUTE(tze);
865
866void thermal_debug_tz_add(struct thermal_zone_device *tz)
867{
868 struct thermal_debugfs *thermal_dbg;
869 struct tz_debugfs *tz_dbg;
870
871 thermal_dbg = thermal_debugfs_add_id(d_tz, tz->id);
872 if (!thermal_dbg)
873 return;
874
875 tz_dbg = &thermal_dbg->tz_dbg;
876
877 tz_dbg->tz = tz;
878
879 tz_dbg->trips_crossed = kzalloc(sizeof(int) * tz->num_trips, GFP_KERNEL);
880 if (!tz_dbg->trips_crossed) {
881 thermal_debugfs_remove_id(thermal_dbg);
882 return;
883 }
884
885 INIT_LIST_HEAD(&tz_dbg->tz_episodes);
886
887 debugfs_create_file("mitigations", 0400, thermal_dbg->d_top,
888 thermal_dbg, &tze_fops);
889
890 tz->debugfs = thermal_dbg;
891}
892
893static struct thermal_debugfs *thermal_debug_tz_clear(struct thermal_zone_device *tz)
894{
895 struct thermal_debugfs *thermal_dbg;
896
897 guard(thermal_zone)(tz);
898
899 thermal_dbg = tz->debugfs;
900 if (thermal_dbg)
901 tz->debugfs = NULL;
902
903 return thermal_dbg;
904}
905
906void thermal_debug_tz_remove(struct thermal_zone_device *tz)
907{
908 struct thermal_debugfs *thermal_dbg;
909 struct tz_episode *tze, *tmp;
910 struct tz_debugfs *tz_dbg;
911 int *trips_crossed;
912
913 thermal_dbg = thermal_debug_tz_clear(tz);
914 if (!thermal_dbg)
915 return;
916
917 tz_dbg = &thermal_dbg->tz_dbg;
918
919 mutex_lock(&thermal_dbg->lock);
920
921 trips_crossed = tz_dbg->trips_crossed;
922
923 list_for_each_entry_safe(tze, tmp, &tz_dbg->tz_episodes, node) {
924 list_del(&tze->node);
925 kfree(tze);
926 }
927
928 mutex_unlock(&thermal_dbg->lock);
929
930 thermal_debugfs_remove_id(thermal_dbg);
931 kfree(trips_crossed);
932}
933
934void thermal_debug_tz_resume(struct thermal_zone_device *tz)
935{
936 struct thermal_debugfs *thermal_dbg = tz->debugfs;
937 ktime_t now = ktime_get();
938 struct tz_debugfs *tz_dbg;
939 struct tz_episode *tze;
940 int i;
941
942 if (!thermal_dbg)
943 return;
944
945 mutex_lock(&thermal_dbg->lock);
946
947 tz_dbg = &thermal_dbg->tz_dbg;
948
949 if (!tz_dbg->nr_trips)
950 goto out;
951
952 /*
953 * A mitigation episode was in progress before the preceding system
954 * suspend transition, so close it because the zone handling is starting
955 * over from scratch.
956 */
957 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
958
959 for (i = 0; i < tz_dbg->nr_trips; i++)
960 tz_episode_close_trip(tze, tz_dbg->trips_crossed[i], now);
961
962 tze->duration = ktime_sub(now, tze->timestamp);
963
964 tz_dbg->nr_trips = 0;
965
966out:
967 mutex_unlock(&thermal_dbg->lock);
968}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright 2023 Linaro Limited
4 *
5 * Author: Daniel Lezcano <daniel.lezcano@linaro.org>
6 *
7 * Thermal subsystem debug support
8 */
9#include <linux/debugfs.h>
10#include <linux/ktime.h>
11#include <linux/list.h>
12#include <linux/minmax.h>
13#include <linux/mutex.h>
14#include <linux/thermal.h>
15
16#include "thermal_core.h"
17
18static struct dentry *d_root;
19static struct dentry *d_cdev;
20static struct dentry *d_tz;
21
22/*
23 * Length of the string containing the thermal zone id or the cooling
24 * device id, including the ending nul character. We can reasonably
25 * assume there won't be more than 256 thermal zones as the maximum
26 * observed today is around 32.
27 */
28#define IDSLENGTH 4
29
30/*
31 * The cooling device transition list is stored in a hash table where
32 * the size is CDEVSTATS_HASH_SIZE. The majority of cooling devices
33 * have dozen of states but some can have much more, so a hash table
34 * is more adequate in this case, because the cost of browsing the entire
35 * list when storing the transitions may not be negligible.
36 */
37#define CDEVSTATS_HASH_SIZE 16
38
39/**
40 * struct cdev_debugfs - per cooling device statistics structure
41 * A cooling device can have a high number of states. Showing the
42 * transitions on a matrix based representation can be overkill given
43 * most of the transitions won't happen and we end up with a matrix
44 * filled with zero. Instead, we show the transitions which actually
45 * happened.
46 *
47 * Every transition updates the current_state and the timestamp. The
48 * transitions and the durations are stored in lists.
49 *
50 * @total: the number of transitions for this cooling device
51 * @current_state: the current cooling device state
52 * @timestamp: the state change timestamp
53 * @transitions: an array of lists containing the state transitions
54 * @durations: an array of lists containing the residencies of each state
55 */
56struct cdev_debugfs {
57 u32 total;
58 int current_state;
59 ktime_t timestamp;
60 struct list_head transitions[CDEVSTATS_HASH_SIZE];
61 struct list_head durations[CDEVSTATS_HASH_SIZE];
62};
63
64/**
65 * struct cdev_record - Common structure for cooling device entry
66 *
67 * The following common structure allows to store the information
68 * related to the transitions and to the state residencies. They are
69 * identified with a id which is associated to a value. It is used as
70 * nodes for the "transitions" and "durations" above.
71 *
72 * @node: node to insert the structure in a list
73 * @id: identifier of the value which can be a state or a transition
74 * @residency: a ktime_t representing a state residency duration
75 * @count: a number of occurrences
76 */
77struct cdev_record {
78 struct list_head node;
79 int id;
80 union {
81 ktime_t residency;
82 u64 count;
83 };
84};
85
86/**
87 * struct trip_stats - Thermal trip statistics
88 *
89 * The trip_stats structure has the relevant information to show the
90 * statistics related to temperature going above a trip point.
91 *
92 * @timestamp: the trip crossing timestamp
93 * @duration: total time when the zone temperature was above the trip point
94 * @count: the number of times the zone temperature was above the trip point
95 * @max: maximum recorded temperature above the trip point
96 * @min: minimum recorded temperature above the trip point
97 * @avg: average temperature above the trip point
98 */
99struct trip_stats {
100 ktime_t timestamp;
101 ktime_t duration;
102 int count;
103 int max;
104 int min;
105 int avg;
106};
107
108/**
109 * struct tz_episode - A mitigation episode information
110 *
111 * The tz_episode structure describes a mitigation episode. A
112 * mitigation episode begins the trip point with the lower temperature
113 * is crossed the way up and ends when it is crossed the way
114 * down. During this episode we can have multiple trip points crossed
115 * the way up and down if there are multiple trip described in the
116 * firmware after the lowest temperature trip point.
117 *
118 * @timestamp: first trip point crossed the way up
119 * @duration: total duration of the mitigation episode
120 * @node: a list element to be added to the list of tz events
121 * @trip_stats: per trip point statistics, flexible array
122 */
123struct tz_episode {
124 ktime_t timestamp;
125 ktime_t duration;
126 struct list_head node;
127 struct trip_stats trip_stats[];
128};
129
130/**
131 * struct tz_debugfs - Store all mitigation episodes for a thermal zone
132 *
133 * The tz_debugfs structure contains the list of the mitigation
134 * episodes and has to track which trip point has been crossed in
135 * order to handle correctly nested trip point mitigation episodes.
136 *
137 * We keep the history of the trip point crossed in an array and as we
138 * can go back and forth inside this history, eg. trip 0,1,2,1,2,1,0,
139 * we keep track of the current position in the history array.
140 *
141 * @tz_episodes: a list of thermal mitigation episodes
142 * @tz: thermal zone this object belongs to
143 * @trips_crossed: an array of trip points crossed by id
144 * @nr_trips: the number of trip points currently being crossed
145 */
146struct tz_debugfs {
147 struct list_head tz_episodes;
148 struct thermal_zone_device *tz;
149 int *trips_crossed;
150 int nr_trips;
151};
152
153/**
154 * struct thermal_debugfs - High level structure for a thermal object in debugfs
155 *
156 * The thermal_debugfs structure is the common structure used by the
157 * cooling device or the thermal zone to store the statistics.
158 *
159 * @d_top: top directory of the thermal object directory
160 * @lock: per object lock to protect the internals
161 *
162 * @cdev_dbg: a cooling device debug structure
163 * @tz_dbg: a thermal zone debug structure
164 */
165struct thermal_debugfs {
166 struct dentry *d_top;
167 struct mutex lock;
168 union {
169 struct cdev_debugfs cdev_dbg;
170 struct tz_debugfs tz_dbg;
171 };
172};
173
174void thermal_debug_init(void)
175{
176 d_root = debugfs_create_dir("thermal", NULL);
177 if (!d_root)
178 return;
179
180 d_cdev = debugfs_create_dir("cooling_devices", d_root);
181 if (!d_cdev)
182 return;
183
184 d_tz = debugfs_create_dir("thermal_zones", d_root);
185}
186
187static struct thermal_debugfs *thermal_debugfs_add_id(struct dentry *d, int id)
188{
189 struct thermal_debugfs *thermal_dbg;
190 char ids[IDSLENGTH];
191
192 thermal_dbg = kzalloc(sizeof(*thermal_dbg), GFP_KERNEL);
193 if (!thermal_dbg)
194 return NULL;
195
196 mutex_init(&thermal_dbg->lock);
197
198 snprintf(ids, IDSLENGTH, "%d", id);
199
200 thermal_dbg->d_top = debugfs_create_dir(ids, d);
201 if (!thermal_dbg->d_top) {
202 kfree(thermal_dbg);
203 return NULL;
204 }
205
206 return thermal_dbg;
207}
208
209static void thermal_debugfs_remove_id(struct thermal_debugfs *thermal_dbg)
210{
211 if (!thermal_dbg)
212 return;
213
214 debugfs_remove(thermal_dbg->d_top);
215
216 kfree(thermal_dbg);
217}
218
219static struct cdev_record *
220thermal_debugfs_cdev_record_alloc(struct thermal_debugfs *thermal_dbg,
221 struct list_head *lists, int id)
222{
223 struct cdev_record *cdev_record;
224
225 cdev_record = kzalloc(sizeof(*cdev_record), GFP_KERNEL);
226 if (!cdev_record)
227 return NULL;
228
229 cdev_record->id = id;
230 INIT_LIST_HEAD(&cdev_record->node);
231 list_add_tail(&cdev_record->node,
232 &lists[cdev_record->id % CDEVSTATS_HASH_SIZE]);
233
234 return cdev_record;
235}
236
237static struct cdev_record *
238thermal_debugfs_cdev_record_find(struct thermal_debugfs *thermal_dbg,
239 struct list_head *lists, int id)
240{
241 struct cdev_record *entry;
242
243 list_for_each_entry(entry, &lists[id % CDEVSTATS_HASH_SIZE], node)
244 if (entry->id == id)
245 return entry;
246
247 return NULL;
248}
249
250static struct cdev_record *
251thermal_debugfs_cdev_record_get(struct thermal_debugfs *thermal_dbg,
252 struct list_head *lists, int id)
253{
254 struct cdev_record *cdev_record;
255
256 cdev_record = thermal_debugfs_cdev_record_find(thermal_dbg, lists, id);
257 if (cdev_record)
258 return cdev_record;
259
260 return thermal_debugfs_cdev_record_alloc(thermal_dbg, lists, id);
261}
262
263static void thermal_debugfs_cdev_clear(struct cdev_debugfs *cdev_dbg)
264{
265 int i;
266 struct cdev_record *entry, *tmp;
267
268 for (i = 0; i < CDEVSTATS_HASH_SIZE; i++) {
269
270 list_for_each_entry_safe(entry, tmp,
271 &cdev_dbg->transitions[i], node) {
272 list_del(&entry->node);
273 kfree(entry);
274 }
275
276 list_for_each_entry_safe(entry, tmp,
277 &cdev_dbg->durations[i], node) {
278 list_del(&entry->node);
279 kfree(entry);
280 }
281 }
282
283 cdev_dbg->total = 0;
284}
285
286static void *cdev_seq_start(struct seq_file *s, loff_t *pos)
287{
288 struct thermal_debugfs *thermal_dbg = s->private;
289
290 mutex_lock(&thermal_dbg->lock);
291
292 return (*pos < CDEVSTATS_HASH_SIZE) ? pos : NULL;
293}
294
295static void *cdev_seq_next(struct seq_file *s, void *v, loff_t *pos)
296{
297 (*pos)++;
298
299 return (*pos < CDEVSTATS_HASH_SIZE) ? pos : NULL;
300}
301
302static void cdev_seq_stop(struct seq_file *s, void *v)
303{
304 struct thermal_debugfs *thermal_dbg = s->private;
305
306 mutex_unlock(&thermal_dbg->lock);
307}
308
309static int cdev_tt_seq_show(struct seq_file *s, void *v)
310{
311 struct thermal_debugfs *thermal_dbg = s->private;
312 struct cdev_debugfs *cdev_dbg = &thermal_dbg->cdev_dbg;
313 struct list_head *transitions = cdev_dbg->transitions;
314 struct cdev_record *entry;
315 int i = *(loff_t *)v;
316
317 if (!i)
318 seq_puts(s, "Transition\tOccurences\n");
319
320 list_for_each_entry(entry, &transitions[i], node) {
321 /*
322 * Assuming maximum cdev states is 1024, the longer
323 * string for a transition would be "1024->1024\0"
324 */
325 char buffer[11];
326
327 snprintf(buffer, ARRAY_SIZE(buffer), "%d->%d",
328 entry->id >> 16, entry->id & 0xFFFF);
329
330 seq_printf(s, "%-10s\t%-10llu\n", buffer, entry->count);
331 }
332
333 return 0;
334}
335
336static const struct seq_operations tt_sops = {
337 .start = cdev_seq_start,
338 .next = cdev_seq_next,
339 .stop = cdev_seq_stop,
340 .show = cdev_tt_seq_show,
341};
342
343DEFINE_SEQ_ATTRIBUTE(tt);
344
345static int cdev_dt_seq_show(struct seq_file *s, void *v)
346{
347 struct thermal_debugfs *thermal_dbg = s->private;
348 struct cdev_debugfs *cdev_dbg = &thermal_dbg->cdev_dbg;
349 struct list_head *durations = cdev_dbg->durations;
350 struct cdev_record *entry;
351 int i = *(loff_t *)v;
352
353 if (!i)
354 seq_puts(s, "State\tResidency\n");
355
356 list_for_each_entry(entry, &durations[i], node) {
357 s64 duration = ktime_to_ms(entry->residency);
358
359 if (entry->id == cdev_dbg->current_state)
360 duration += ktime_ms_delta(ktime_get(),
361 cdev_dbg->timestamp);
362
363 seq_printf(s, "%-5d\t%-10llu\n", entry->id, duration);
364 }
365
366 return 0;
367}
368
369static const struct seq_operations dt_sops = {
370 .start = cdev_seq_start,
371 .next = cdev_seq_next,
372 .stop = cdev_seq_stop,
373 .show = cdev_dt_seq_show,
374};
375
376DEFINE_SEQ_ATTRIBUTE(dt);
377
378static int cdev_clear_set(void *data, u64 val)
379{
380 struct thermal_debugfs *thermal_dbg = data;
381
382 if (!val)
383 return -EINVAL;
384
385 mutex_lock(&thermal_dbg->lock);
386
387 thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
388
389 mutex_unlock(&thermal_dbg->lock);
390
391 return 0;
392}
393
394DEFINE_DEBUGFS_ATTRIBUTE(cdev_clear_fops, NULL, cdev_clear_set, "%llu\n");
395
396/**
397 * thermal_debug_cdev_state_update - Update a cooling device state change
398 *
399 * Computes a transition and the duration of the previous state residency.
400 *
401 * @cdev : a pointer to a cooling device
402 * @new_state: an integer corresponding to the new cooling device state
403 */
404void thermal_debug_cdev_state_update(const struct thermal_cooling_device *cdev,
405 int new_state)
406{
407 struct thermal_debugfs *thermal_dbg = cdev->debugfs;
408 struct cdev_debugfs *cdev_dbg;
409 struct cdev_record *cdev_record;
410 int transition, old_state;
411
412 if (!thermal_dbg || (thermal_dbg->cdev_dbg.current_state == new_state))
413 return;
414
415 mutex_lock(&thermal_dbg->lock);
416
417 cdev_dbg = &thermal_dbg->cdev_dbg;
418
419 old_state = cdev_dbg->current_state;
420
421 /*
422 * Get the old state information in the durations list. If
423 * this one does not exist, a new allocated one will be
424 * returned. Recompute the total duration in the old state and
425 * get a new timestamp for the new state.
426 */
427 cdev_record = thermal_debugfs_cdev_record_get(thermal_dbg,
428 cdev_dbg->durations,
429 old_state);
430 if (cdev_record) {
431 ktime_t now = ktime_get();
432 ktime_t delta = ktime_sub(now, cdev_dbg->timestamp);
433 cdev_record->residency = ktime_add(cdev_record->residency, delta);
434 cdev_dbg->timestamp = now;
435 }
436
437 cdev_dbg->current_state = new_state;
438
439 /*
440 * Create a record for the new state if it is not there, so its
441 * duration will be printed by cdev_dt_seq_show() as expected if it
442 * runs before the next state transition.
443 */
444 thermal_debugfs_cdev_record_get(thermal_dbg, cdev_dbg->durations, new_state);
445
446 transition = (old_state << 16) | new_state;
447
448 /*
449 * Get the transition in the transitions list. If this one
450 * does not exist, a new allocated one will be returned.
451 * Increment the occurrence of this transition which is stored
452 * in the value field.
453 */
454 cdev_record = thermal_debugfs_cdev_record_get(thermal_dbg,
455 cdev_dbg->transitions,
456 transition);
457 if (cdev_record)
458 cdev_record->count++;
459
460 cdev_dbg->total++;
461
462 mutex_unlock(&thermal_dbg->lock);
463}
464
465/**
466 * thermal_debug_cdev_add - Add a cooling device debugfs entry
467 *
468 * Allocates a cooling device object for debug, initializes the
469 * statistics and create the entries in sysfs.
470 * @cdev: a pointer to a cooling device
471 * @state: current state of the cooling device
472 */
473void thermal_debug_cdev_add(struct thermal_cooling_device *cdev, int state)
474{
475 struct thermal_debugfs *thermal_dbg;
476 struct cdev_debugfs *cdev_dbg;
477 int i;
478
479 thermal_dbg = thermal_debugfs_add_id(d_cdev, cdev->id);
480 if (!thermal_dbg)
481 return;
482
483 cdev_dbg = &thermal_dbg->cdev_dbg;
484
485 for (i = 0; i < CDEVSTATS_HASH_SIZE; i++) {
486 INIT_LIST_HEAD(&cdev_dbg->transitions[i]);
487 INIT_LIST_HEAD(&cdev_dbg->durations[i]);
488 }
489
490 cdev_dbg->current_state = state;
491 cdev_dbg->timestamp = ktime_get();
492
493 /*
494 * Create a record for the initial cooling device state, so its
495 * duration will be printed by cdev_dt_seq_show() as expected if it
496 * runs before the first state transition.
497 */
498 thermal_debugfs_cdev_record_get(thermal_dbg, cdev_dbg->durations, state);
499
500 debugfs_create_file("trans_table", 0400, thermal_dbg->d_top,
501 thermal_dbg, &tt_fops);
502
503 debugfs_create_file("time_in_state_ms", 0400, thermal_dbg->d_top,
504 thermal_dbg, &dt_fops);
505
506 debugfs_create_file("clear", 0200, thermal_dbg->d_top,
507 thermal_dbg, &cdev_clear_fops);
508
509 debugfs_create_u32("total_trans", 0400, thermal_dbg->d_top,
510 &cdev_dbg->total);
511
512 cdev->debugfs = thermal_dbg;
513}
514
515/**
516 * thermal_debug_cdev_remove - Remove a cooling device debugfs entry
517 *
518 * Frees the statistics memory data and remove the debugfs entry
519 *
520 * @cdev: a pointer to a cooling device
521 */
522void thermal_debug_cdev_remove(struct thermal_cooling_device *cdev)
523{
524 struct thermal_debugfs *thermal_dbg;
525
526 mutex_lock(&cdev->lock);
527
528 thermal_dbg = cdev->debugfs;
529 if (!thermal_dbg) {
530 mutex_unlock(&cdev->lock);
531 return;
532 }
533
534 cdev->debugfs = NULL;
535
536 mutex_unlock(&cdev->lock);
537
538 mutex_lock(&thermal_dbg->lock);
539
540 thermal_debugfs_cdev_clear(&thermal_dbg->cdev_dbg);
541
542 mutex_unlock(&thermal_dbg->lock);
543
544 thermal_debugfs_remove_id(thermal_dbg);
545}
546
547static struct tz_episode *thermal_debugfs_tz_event_alloc(struct thermal_zone_device *tz,
548 ktime_t now)
549{
550 struct tz_episode *tze;
551 int i;
552
553 tze = kzalloc(struct_size(tze, trip_stats, tz->num_trips), GFP_KERNEL);
554 if (!tze)
555 return NULL;
556
557 INIT_LIST_HEAD(&tze->node);
558 tze->timestamp = now;
559
560 for (i = 0; i < tz->num_trips; i++) {
561 tze->trip_stats[i].min = INT_MAX;
562 tze->trip_stats[i].max = INT_MIN;
563 }
564
565 return tze;
566}
567
568void thermal_debug_tz_trip_up(struct thermal_zone_device *tz,
569 const struct thermal_trip *trip)
570{
571 struct tz_episode *tze;
572 struct tz_debugfs *tz_dbg;
573 struct thermal_debugfs *thermal_dbg = tz->debugfs;
574 int trip_id = thermal_zone_trip_id(tz, trip);
575 ktime_t now = ktime_get();
576
577 if (!thermal_dbg)
578 return;
579
580 mutex_lock(&thermal_dbg->lock);
581
582 tz_dbg = &thermal_dbg->tz_dbg;
583
584 /*
585 * The mitigation is starting. A mitigation can contain
586 * several episodes where each of them is related to a
587 * temperature crossing a trip point. The episodes are
588 * nested. That means when the temperature is crossing the
589 * first trip point, the duration begins to be measured. If
590 * the temperature continues to increase and reaches the
591 * second trip point, the duration of the first trip must be
592 * also accumulated.
593 *
594 * eg.
595 *
596 * temp
597 * ^
598 * | --------
599 * trip 2 / \ ------
600 * | /| |\ /| |\
601 * trip 1 / | | `---- | | \
602 * | /| | | | | |\
603 * trip 0 / | | | | | | \
604 * | /| | | | | | | |\
605 * | / | | | | | | | | `--
606 * | / | | | | | | | |
607 * |----- | | | | | | | |
608 * | | | | | | | | |
609 * --------|-|-|--------|--------|------|-|-|------------------> time
610 * | | |<--t2-->| |<-t2'>| | |
611 * | | | |
612 * | |<------------t1------------>| |
613 * | |
614 * |<-------------t0--------------->|
615 *
616 */
617 if (!tz_dbg->nr_trips) {
618 tze = thermal_debugfs_tz_event_alloc(tz, now);
619 if (!tze)
620 goto unlock;
621
622 list_add(&tze->node, &tz_dbg->tz_episodes);
623 }
624
625 /*
626 * Each time a trip point is crossed the way up, the trip_id
627 * is stored in the trip_crossed array and the nr_trips is
628 * incremented. A nr_trips equal to zero means we are entering
629 * a mitigation episode.
630 *
631 * The trip ids may not be in the ascending order but the
632 * result in the array trips_crossed will be in the ascending
633 * temperature order. The function detecting when a trip point
634 * is crossed the way down will handle the very rare case when
635 * the trip points may have been reordered during this
636 * mitigation episode.
637 */
638 tz_dbg->trips_crossed[tz_dbg->nr_trips++] = trip_id;
639
640 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
641 tze->trip_stats[trip_id].timestamp = now;
642
643unlock:
644 mutex_unlock(&thermal_dbg->lock);
645}
646
647void thermal_debug_tz_trip_down(struct thermal_zone_device *tz,
648 const struct thermal_trip *trip)
649{
650 struct thermal_debugfs *thermal_dbg = tz->debugfs;
651 struct tz_episode *tze;
652 struct tz_debugfs *tz_dbg;
653 ktime_t delta, now = ktime_get();
654 int trip_id = thermal_zone_trip_id(tz, trip);
655 int i;
656
657 if (!thermal_dbg)
658 return;
659
660 mutex_lock(&thermal_dbg->lock);
661
662 tz_dbg = &thermal_dbg->tz_dbg;
663
664 /*
665 * The temperature crosses the way down but there was not
666 * mitigation detected before. That may happen when the
667 * temperature is greater than a trip point when registering a
668 * thermal zone, which is a common use case as the kernel has
669 * no mitigation mechanism yet at boot time.
670 */
671 if (!tz_dbg->nr_trips)
672 goto out;
673
674 for (i = tz_dbg->nr_trips - 1; i >= 0; i--) {
675 if (tz_dbg->trips_crossed[i] == trip_id)
676 break;
677 }
678
679 if (i < 0)
680 goto out;
681
682 tz_dbg->nr_trips--;
683
684 if (i < tz_dbg->nr_trips)
685 tz_dbg->trips_crossed[i] = tz_dbg->trips_crossed[tz_dbg->nr_trips];
686
687 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
688
689 delta = ktime_sub(now, tze->trip_stats[trip_id].timestamp);
690
691 tze->trip_stats[trip_id].duration =
692 ktime_add(delta, tze->trip_stats[trip_id].duration);
693
694 /*
695 * This event closes the mitigation as we are crossing the
696 * last trip point the way down.
697 */
698 if (!tz_dbg->nr_trips)
699 tze->duration = ktime_sub(now, tze->timestamp);
700
701out:
702 mutex_unlock(&thermal_dbg->lock);
703}
704
705void thermal_debug_update_temp(struct thermal_zone_device *tz)
706{
707 struct thermal_debugfs *thermal_dbg = tz->debugfs;
708 struct tz_episode *tze;
709 struct tz_debugfs *tz_dbg;
710 int trip_id, i;
711
712 if (!thermal_dbg)
713 return;
714
715 mutex_lock(&thermal_dbg->lock);
716
717 tz_dbg = &thermal_dbg->tz_dbg;
718
719 if (!tz_dbg->nr_trips)
720 goto out;
721
722 for (i = 0; i < tz_dbg->nr_trips; i++) {
723 trip_id = tz_dbg->trips_crossed[i];
724 tze = list_first_entry(&tz_dbg->tz_episodes, struct tz_episode, node);
725 tze->trip_stats[trip_id].count++;
726 tze->trip_stats[trip_id].max = max(tze->trip_stats[trip_id].max, tz->temperature);
727 tze->trip_stats[trip_id].min = min(tze->trip_stats[trip_id].min, tz->temperature);
728 tze->trip_stats[trip_id].avg = tze->trip_stats[trip_id].avg +
729 (tz->temperature - tze->trip_stats[trip_id].avg) /
730 tze->trip_stats[trip_id].count;
731 }
732out:
733 mutex_unlock(&thermal_dbg->lock);
734}
735
736static void *tze_seq_start(struct seq_file *s, loff_t *pos)
737{
738 struct thermal_debugfs *thermal_dbg = s->private;
739 struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
740
741 mutex_lock(&thermal_dbg->lock);
742
743 return seq_list_start(&tz_dbg->tz_episodes, *pos);
744}
745
746static void *tze_seq_next(struct seq_file *s, void *v, loff_t *pos)
747{
748 struct thermal_debugfs *thermal_dbg = s->private;
749 struct tz_debugfs *tz_dbg = &thermal_dbg->tz_dbg;
750
751 return seq_list_next(v, &tz_dbg->tz_episodes, pos);
752}
753
754static void tze_seq_stop(struct seq_file *s, void *v)
755{
756 struct thermal_debugfs *thermal_dbg = s->private;
757
758 mutex_unlock(&thermal_dbg->lock);
759}
760
761static int tze_seq_show(struct seq_file *s, void *v)
762{
763 struct thermal_debugfs *thermal_dbg = s->private;
764 struct thermal_zone_device *tz = thermal_dbg->tz_dbg.tz;
765 struct thermal_trip *trip;
766 struct tz_episode *tze;
767 const char *type;
768 int trip_id;
769
770 tze = list_entry((struct list_head *)v, struct tz_episode, node);
771
772 seq_printf(s, ",-Mitigation at %lluus, duration=%llums\n",
773 ktime_to_us(tze->timestamp),
774 ktime_to_ms(tze->duration));
775
776 seq_printf(s, "| trip | type | temp(°mC) | hyst(°mC) | duration | avg(°mC) | min(°mC) | max(°mC) |\n");
777
778 for_each_trip(tz, trip) {
779 /*
780 * There is no possible mitigation happening at the
781 * critical trip point, so the stats will be always
782 * zero, skip this trip point
783 */
784 if (trip->type == THERMAL_TRIP_CRITICAL)
785 continue;
786
787 if (trip->type == THERMAL_TRIP_PASSIVE)
788 type = "passive";
789 else if (trip->type == THERMAL_TRIP_ACTIVE)
790 type = "active";
791 else
792 type = "hot";
793
794 trip_id = thermal_zone_trip_id(tz, trip);
795
796 seq_printf(s, "| %*d | %*s | %*d | %*d | %*lld | %*d | %*d | %*d |\n",
797 4 , trip_id,
798 8, type,
799 9, trip->temperature,
800 9, trip->hysteresis,
801 10, ktime_to_ms(tze->trip_stats[trip_id].duration),
802 9, tze->trip_stats[trip_id].avg,
803 9, tze->trip_stats[trip_id].min,
804 9, tze->trip_stats[trip_id].max);
805 }
806
807 return 0;
808}
809
810static const struct seq_operations tze_sops = {
811 .start = tze_seq_start,
812 .next = tze_seq_next,
813 .stop = tze_seq_stop,
814 .show = tze_seq_show,
815};
816
817DEFINE_SEQ_ATTRIBUTE(tze);
818
819void thermal_debug_tz_add(struct thermal_zone_device *tz)
820{
821 struct thermal_debugfs *thermal_dbg;
822 struct tz_debugfs *tz_dbg;
823
824 thermal_dbg = thermal_debugfs_add_id(d_tz, tz->id);
825 if (!thermal_dbg)
826 return;
827
828 tz_dbg = &thermal_dbg->tz_dbg;
829
830 tz_dbg->tz = tz;
831
832 tz_dbg->trips_crossed = kzalloc(sizeof(int) * tz->num_trips, GFP_KERNEL);
833 if (!tz_dbg->trips_crossed) {
834 thermal_debugfs_remove_id(thermal_dbg);
835 return;
836 }
837
838 INIT_LIST_HEAD(&tz_dbg->tz_episodes);
839
840 debugfs_create_file("mitigations", 0400, thermal_dbg->d_top,
841 thermal_dbg, &tze_fops);
842
843 tz->debugfs = thermal_dbg;
844}
845
846void thermal_debug_tz_remove(struct thermal_zone_device *tz)
847{
848 struct thermal_debugfs *thermal_dbg;
849 struct tz_episode *tze, *tmp;
850 struct tz_debugfs *tz_dbg;
851 int *trips_crossed;
852
853 mutex_lock(&tz->lock);
854
855 thermal_dbg = tz->debugfs;
856 if (!thermal_dbg) {
857 mutex_unlock(&tz->lock);
858 return;
859 }
860
861 tz->debugfs = NULL;
862
863 mutex_unlock(&tz->lock);
864
865 tz_dbg = &thermal_dbg->tz_dbg;
866
867 mutex_lock(&thermal_dbg->lock);
868
869 trips_crossed = tz_dbg->trips_crossed;
870
871 list_for_each_entry_safe(tze, tmp, &tz_dbg->tz_episodes, node) {
872 list_del(&tze->node);
873 kfree(tze);
874 }
875
876 mutex_unlock(&thermal_dbg->lock);
877
878 thermal_debugfs_remove_id(thermal_dbg);
879 kfree(trips_crossed);
880}