FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_cpu.c
1 /*-
2 * Copyright (c) 2004-2007 Nate Lawson (SDG)
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD: releng/7.3/sys/kern/kern_cpu.c 202816 2010-01-22 17:08:05Z brueffer $");
29
30 #include <sys/param.h>
31 #include <sys/bus.h>
32 #include <sys/cpu.h>
33 #include <sys/eventhandler.h>
34 #include <sys/kernel.h>
35 #include <sys/lock.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/proc.h>
39 #include <sys/queue.h>
40 #include <sys/sbuf.h>
41 #include <sys/sched.h>
42 #include <sys/smp.h>
43 #include <sys/sysctl.h>
44 #include <sys/systm.h>
45 #include <sys/sx.h>
46 #include <sys/timetc.h>
47 #include <sys/taskqueue.h>
48
49 #include "cpufreq_if.h"
50
51 /*
52 * Common CPU frequency glue code. Drivers for specific hardware can
53 * attach this interface to allow users to get/set the CPU frequency.
54 */
55
56 /*
57 * Number of levels we can handle. Levels are synthesized from settings
58 * so for M settings and N drivers, there may be M*N levels.
59 */
60 #define CF_MAX_LEVELS 64
61
62 struct cf_saved_freq {
63 struct cf_level level;
64 int priority;
65 SLIST_ENTRY(cf_saved_freq) link;
66 };
67
68 struct cpufreq_softc {
69 struct sx lock;
70 struct cf_level curr_level;
71 int curr_priority;
72 SLIST_HEAD(, cf_saved_freq) saved_freq;
73 struct cf_level_lst all_levels;
74 int all_count;
75 int max_mhz;
76 device_t dev;
77 struct sysctl_ctx_list sysctl_ctx;
78 struct task startup_task;
79 };
80
81 struct cf_setting_array {
82 struct cf_setting sets[MAX_SETTINGS];
83 int count;
84 TAILQ_ENTRY(cf_setting_array) link;
85 };
86
87 TAILQ_HEAD(cf_setting_lst, cf_setting_array);
88
89 #define CF_MTX_INIT(x) sx_init((x), "cpufreq lock")
90 #define CF_MTX_LOCK(x) sx_xlock((x))
91 #define CF_MTX_UNLOCK(x) sx_xunlock((x))
92 #define CF_MTX_ASSERT(x) sx_assert((x), SX_XLOCKED)
93
94 #define CF_DEBUG(msg...) do { \
95 if (cf_verbose) \
96 printf("cpufreq: " msg); \
97 } while (0)
98
99 static int cpufreq_attach(device_t dev);
100 static void cpufreq_startup_task(void *ctx, int pending);
101 static int cpufreq_detach(device_t dev);
102 static int cf_set_method(device_t dev, const struct cf_level *level,
103 int priority);
104 static int cf_get_method(device_t dev, struct cf_level *level);
105 static int cf_levels_method(device_t dev, struct cf_level *levels,
106 int *count);
107 static int cpufreq_insert_abs(struct cpufreq_softc *sc,
108 struct cf_setting *sets, int count);
109 static int cpufreq_expand_set(struct cpufreq_softc *sc,
110 struct cf_setting_array *set_arr);
111 static struct cf_level *cpufreq_dup_set(struct cpufreq_softc *sc,
112 struct cf_level *dup, struct cf_setting *set);
113 static int cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS);
114 static int cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS);
115 static int cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS);
116
117 static device_method_t cpufreq_methods[] = {
118 DEVMETHOD(device_probe, bus_generic_probe),
119 DEVMETHOD(device_attach, cpufreq_attach),
120 DEVMETHOD(device_detach, cpufreq_detach),
121
122 DEVMETHOD(cpufreq_set, cf_set_method),
123 DEVMETHOD(cpufreq_get, cf_get_method),
124 DEVMETHOD(cpufreq_levels, cf_levels_method),
125 {0, 0}
126 };
127 static driver_t cpufreq_driver = {
128 "cpufreq", cpufreq_methods, sizeof(struct cpufreq_softc)
129 };
130 static devclass_t cpufreq_dc;
131 DRIVER_MODULE(cpufreq, cpu, cpufreq_driver, cpufreq_dc, 0, 0);
132
133 static int cf_lowest_freq;
134 static int cf_verbose;
135 TUNABLE_INT("debug.cpufreq.lowest", &cf_lowest_freq);
136 TUNABLE_INT("debug.cpufreq.verbose", &cf_verbose);
137 SYSCTL_NODE(_debug, OID_AUTO, cpufreq, CTLFLAG_RD, NULL, "cpufreq debugging");
138 SYSCTL_INT(_debug_cpufreq, OID_AUTO, lowest, CTLFLAG_RW, &cf_lowest_freq, 1,
139 "Don't provide levels below this frequency.");
140 SYSCTL_INT(_debug_cpufreq, OID_AUTO, verbose, CTLFLAG_RW, &cf_verbose, 1,
141 "Print verbose debugging messages");
142
143 static int
144 cpufreq_attach(device_t dev)
145 {
146 struct cpufreq_softc *sc;
147 struct pcpu *pc;
148 device_t parent;
149 uint64_t rate;
150 int numdevs;
151
152 CF_DEBUG("initializing %s\n", device_get_nameunit(dev));
153 sc = device_get_softc(dev);
154 parent = device_get_parent(dev);
155 sc->dev = dev;
156 sysctl_ctx_init(&sc->sysctl_ctx);
157 TAILQ_INIT(&sc->all_levels);
158 CF_MTX_INIT(&sc->lock);
159 sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN;
160 SLIST_INIT(&sc->saved_freq);
161 /* Try to get current CPU freq to use it as maximum later if needed */
162 pc = cpu_get_pcpu(dev);
163 if (cpu_est_clockrate(pc->pc_cpuid, &rate) == 0)
164 sc->max_mhz = rate / 1000000;
165 else
166 sc->max_mhz = CPUFREQ_VAL_UNKNOWN;
167
168 /*
169 * Only initialize one set of sysctls for all CPUs. In the future,
170 * if multiple CPUs can have different settings, we can move these
171 * sysctls to be under every CPU instead of just the first one.
172 */
173 numdevs = devclass_get_count(cpufreq_dc);
174 if (numdevs > 1)
175 return (0);
176
177 CF_DEBUG("initializing one-time data for %s\n",
178 device_get_nameunit(dev));
179 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
180 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)),
181 OID_AUTO, "freq", CTLTYPE_INT | CTLFLAG_RW, sc, 0,
182 cpufreq_curr_sysctl, "I", "Current CPU frequency");
183 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
184 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)),
185 OID_AUTO, "freq_levels", CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
186 cpufreq_levels_sysctl, "A", "CPU frequency levels");
187
188 /*
189 * Queue a one-shot broadcast that levels have changed.
190 * It will run once the system has completed booting.
191 */
192 TASK_INIT(&sc->startup_task, 0, cpufreq_startup_task, dev);
193 taskqueue_enqueue(taskqueue_thread, &sc->startup_task);
194
195 return (0);
196 }
197
198 /* Handle any work to be done for all drivers that attached during boot. */
199 static void
200 cpufreq_startup_task(void *ctx, int pending)
201 {
202
203 cpufreq_settings_changed((device_t)ctx);
204 }
205
206 static int
207 cpufreq_detach(device_t dev)
208 {
209 struct cpufreq_softc *sc;
210 struct cf_saved_freq *saved_freq;
211 int numdevs;
212
213 CF_DEBUG("shutdown %s\n", device_get_nameunit(dev));
214 sc = device_get_softc(dev);
215 sysctl_ctx_free(&sc->sysctl_ctx);
216
217 while ((saved_freq = SLIST_FIRST(&sc->saved_freq)) != NULL) {
218 SLIST_REMOVE_HEAD(&sc->saved_freq, link);
219 free(saved_freq, M_TEMP);
220 }
221
222 /* Only clean up these resources when the last device is detaching. */
223 numdevs = devclass_get_count(cpufreq_dc);
224 if (numdevs == 1) {
225 CF_DEBUG("final shutdown for %s\n", device_get_nameunit(dev));
226 }
227
228 return (0);
229 }
230
231 static int
232 cf_set_method(device_t dev, const struct cf_level *level, int priority)
233 {
234 struct cpufreq_softc *sc;
235 const struct cf_setting *set;
236 struct cf_saved_freq *saved_freq, *curr_freq;
237 struct pcpu *pc;
238 int error, i;
239
240 sc = device_get_softc(dev);
241 error = 0;
242 set = NULL;
243 saved_freq = NULL;
244
245 /* We are going to change levels so notify the pre-change handler. */
246 EVENTHANDLER_INVOKE(cpufreq_pre_change, level, &error);
247 if (error != 0) {
248 EVENTHANDLER_INVOKE(cpufreq_post_change, level, error);
249 return (error);
250 }
251
252 CF_MTX_LOCK(&sc->lock);
253
254 #ifdef SMP
255 /*
256 * If still booting and secondary CPUs not started yet, don't allow
257 * changing the frequency until they're online. This is because we
258 * can't switch to them using sched_bind() and thus we'd only be
259 * switching the main CPU. XXXTODO: Need to think more about how to
260 * handle having different CPUs at different frequencies.
261 */
262 if (mp_ncpus > 1 && !smp_active) {
263 device_printf(dev, "rejecting change, SMP not started yet\n");
264 error = ENXIO;
265 goto out;
266 }
267 #endif /* SMP */
268
269 /*
270 * If the requested level has a lower priority, don't allow
271 * the new level right now.
272 */
273 if (priority < sc->curr_priority) {
274 CF_DEBUG("ignoring, curr prio %d less than %d\n", priority,
275 sc->curr_priority);
276 error = EPERM;
277 goto out;
278 }
279
280 /*
281 * If the caller didn't specify a level and one is saved, prepare to
282 * restore the saved level. If none has been saved, return an error.
283 */
284 if (level == NULL) {
285 saved_freq = SLIST_FIRST(&sc->saved_freq);
286 if (saved_freq == NULL) {
287 CF_DEBUG("NULL level, no saved level\n");
288 error = ENXIO;
289 goto out;
290 }
291 level = &saved_freq->level;
292 priority = saved_freq->priority;
293 CF_DEBUG("restoring saved level, freq %d prio %d\n",
294 level->total_set.freq, priority);
295 }
296
297 /* Reject levels that are below our specified threshold. */
298 if (level->total_set.freq < cf_lowest_freq) {
299 CF_DEBUG("rejecting freq %d, less than %d limit\n",
300 level->total_set.freq, cf_lowest_freq);
301 error = EINVAL;
302 goto out;
303 }
304
305 /* If already at this level, just return. */
306 if (CPUFREQ_CMP(sc->curr_level.total_set.freq, level->total_set.freq)) {
307 CF_DEBUG("skipping freq %d, same as current level %d\n",
308 level->total_set.freq, sc->curr_level.total_set.freq);
309 goto skip;
310 }
311
312 /* First, set the absolute frequency via its driver. */
313 set = &level->abs_set;
314 if (set->dev) {
315 if (!device_is_attached(set->dev)) {
316 error = ENXIO;
317 goto out;
318 }
319
320 /* Bind to the target CPU before switching. */
321 pc = cpu_get_pcpu(set->dev);
322 thread_lock(curthread);
323 sched_bind(curthread, pc->pc_cpuid);
324 thread_unlock(curthread);
325 CF_DEBUG("setting abs freq %d on %s (cpu %d)\n", set->freq,
326 device_get_nameunit(set->dev), PCPU_GET(cpuid));
327 error = CPUFREQ_DRV_SET(set->dev, set);
328 thread_lock(curthread);
329 sched_unbind(curthread);
330 thread_unlock(curthread);
331 if (error) {
332 goto out;
333 }
334 }
335
336 /* Next, set any/all relative frequencies via their drivers. */
337 for (i = 0; i < level->rel_count; i++) {
338 set = &level->rel_set[i];
339 if (!device_is_attached(set->dev)) {
340 error = ENXIO;
341 goto out;
342 }
343
344 /* Bind to the target CPU before switching. */
345 pc = cpu_get_pcpu(set->dev);
346 thread_lock(curthread);
347 sched_bind(curthread, pc->pc_cpuid);
348 thread_unlock(curthread);
349 CF_DEBUG("setting rel freq %d on %s (cpu %d)\n", set->freq,
350 device_get_nameunit(set->dev), PCPU_GET(cpuid));
351 error = CPUFREQ_DRV_SET(set->dev, set);
352 thread_lock(curthread);
353 sched_unbind(curthread);
354 thread_unlock(curthread);
355 if (error) {
356 /* XXX Back out any successful setting? */
357 goto out;
358 }
359 }
360
361 skip:
362 /*
363 * Before recording the current level, check if we're going to a
364 * higher priority. If so, save the previous level and priority.
365 */
366 if (sc->curr_level.total_set.freq != CPUFREQ_VAL_UNKNOWN &&
367 priority > sc->curr_priority) {
368 CF_DEBUG("saving level, freq %d prio %d\n",
369 sc->curr_level.total_set.freq, sc->curr_priority);
370 curr_freq = malloc(sizeof(*curr_freq), M_TEMP, M_NOWAIT);
371 if (curr_freq == NULL) {
372 error = ENOMEM;
373 goto out;
374 }
375 curr_freq->level = sc->curr_level;
376 curr_freq->priority = sc->curr_priority;
377 SLIST_INSERT_HEAD(&sc->saved_freq, curr_freq, link);
378 }
379 sc->curr_level = *level;
380 sc->curr_priority = priority;
381
382 /* If we were restoring a saved state, reset it to "unused". */
383 if (saved_freq != NULL) {
384 CF_DEBUG("resetting saved level\n");
385 sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN;
386 SLIST_REMOVE_HEAD(&sc->saved_freq, link);
387 free(saved_freq, M_TEMP);
388 }
389
390 out:
391 CF_MTX_UNLOCK(&sc->lock);
392
393 /*
394 * We changed levels (or attempted to) so notify the post-change
395 * handler of new frequency or error.
396 */
397 EVENTHANDLER_INVOKE(cpufreq_post_change, level, error);
398 if (error && set)
399 device_printf(set->dev, "set freq failed, err %d\n", error);
400
401 return (error);
402 }
403
404 static int
405 cf_get_method(device_t dev, struct cf_level *level)
406 {
407 struct cpufreq_softc *sc;
408 struct cf_level *levels;
409 struct cf_setting *curr_set, set;
410 struct pcpu *pc;
411 device_t *devs;
412 int count, error, i, n, numdevs;
413 uint64_t rate;
414
415 sc = device_get_softc(dev);
416 error = 0;
417 levels = NULL;
418
419 /* If we already know the current frequency, we're done. */
420 CF_MTX_LOCK(&sc->lock);
421 curr_set = &sc->curr_level.total_set;
422 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) {
423 CF_DEBUG("get returning known freq %d\n", curr_set->freq);
424 goto out;
425 }
426 CF_MTX_UNLOCK(&sc->lock);
427
428 /*
429 * We need to figure out the current level. Loop through every
430 * driver, getting the current setting. Then, attempt to get a best
431 * match of settings against each level.
432 */
433 count = CF_MAX_LEVELS;
434 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
435 if (levels == NULL)
436 return (ENOMEM);
437 error = CPUFREQ_LEVELS(sc->dev, levels, &count);
438 if (error) {
439 if (error == E2BIG)
440 printf("cpufreq: need to increase CF_MAX_LEVELS\n");
441 free(levels, M_TEMP);
442 return (error);
443 }
444 error = device_get_children(device_get_parent(dev), &devs, &numdevs);
445 if (error) {
446 free(levels, M_TEMP);
447 return (error);
448 }
449
450 /*
451 * Reacquire the lock and search for the given level.
452 *
453 * XXX Note: this is not quite right since we really need to go
454 * through each level and compare both absolute and relative
455 * settings for each driver in the system before making a match.
456 * The estimation code below catches this case though.
457 */
458 CF_MTX_LOCK(&sc->lock);
459 for (n = 0; n < numdevs && curr_set->freq == CPUFREQ_VAL_UNKNOWN; n++) {
460 if (!device_is_attached(devs[n]))
461 continue;
462 if (CPUFREQ_DRV_GET(devs[n], &set) != 0)
463 continue;
464 for (i = 0; i < count; i++) {
465 if (CPUFREQ_CMP(set.freq, levels[i].total_set.freq)) {
466 sc->curr_level = levels[i];
467 break;
468 }
469 }
470 }
471 free(devs, M_TEMP);
472 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) {
473 CF_DEBUG("get matched freq %d from drivers\n", curr_set->freq);
474 goto out;
475 }
476
477 /*
478 * We couldn't find an exact match, so attempt to estimate and then
479 * match against a level.
480 */
481 pc = cpu_get_pcpu(dev);
482 if (pc == NULL) {
483 error = ENXIO;
484 goto out;
485 }
486 cpu_est_clockrate(pc->pc_cpuid, &rate);
487 rate /= 1000000;
488 for (i = 0; i < count; i++) {
489 if (CPUFREQ_CMP(rate, levels[i].total_set.freq)) {
490 sc->curr_level = levels[i];
491 CF_DEBUG("get estimated freq %d\n", curr_set->freq);
492 goto out;
493 }
494 }
495 error = ENXIO;
496
497 out:
498 if (error == 0)
499 *level = sc->curr_level;
500
501 CF_MTX_UNLOCK(&sc->lock);
502 if (levels)
503 free(levels, M_TEMP);
504 return (error);
505 }
506
507 static int
508 cf_levels_method(device_t dev, struct cf_level *levels, int *count)
509 {
510 struct cf_setting_array *set_arr;
511 struct cf_setting_lst rel_sets;
512 struct cpufreq_softc *sc;
513 struct cf_level *lev;
514 struct cf_setting *sets;
515 struct pcpu *pc;
516 device_t *devs;
517 int error, i, numdevs, set_count, type;
518 uint64_t rate;
519
520 if (levels == NULL || count == NULL)
521 return (EINVAL);
522
523 TAILQ_INIT(&rel_sets);
524 sc = device_get_softc(dev);
525 error = device_get_children(device_get_parent(dev), &devs, &numdevs);
526 if (error)
527 return (error);
528 sets = malloc(MAX_SETTINGS * sizeof(*sets), M_TEMP, M_NOWAIT);
529 if (sets == NULL) {
530 free(devs, M_TEMP);
531 return (ENOMEM);
532 }
533
534 /* Get settings from all cpufreq drivers. */
535 CF_MTX_LOCK(&sc->lock);
536 for (i = 0; i < numdevs; i++) {
537 /* Skip devices that aren't ready. */
538 if (!device_is_attached(devs[i]))
539 continue;
540
541 /*
542 * Get settings, skipping drivers that offer no settings or
543 * provide settings for informational purposes only.
544 */
545 error = CPUFREQ_DRV_TYPE(devs[i], &type);
546 if (error || (type & CPUFREQ_FLAG_INFO_ONLY)) {
547 if (error == 0) {
548 CF_DEBUG("skipping info-only driver %s\n",
549 device_get_nameunit(devs[i]));
550 }
551 continue;
552 }
553 set_count = MAX_SETTINGS;
554 error = CPUFREQ_DRV_SETTINGS(devs[i], sets, &set_count);
555 if (error || set_count == 0)
556 continue;
557
558 /* Add the settings to our absolute/relative lists. */
559 switch (type & CPUFREQ_TYPE_MASK) {
560 case CPUFREQ_TYPE_ABSOLUTE:
561 error = cpufreq_insert_abs(sc, sets, set_count);
562 break;
563 case CPUFREQ_TYPE_RELATIVE:
564 CF_DEBUG("adding %d relative settings\n", set_count);
565 set_arr = malloc(sizeof(*set_arr), M_TEMP, M_NOWAIT);
566 if (set_arr == NULL) {
567 error = ENOMEM;
568 goto out;
569 }
570 bcopy(sets, set_arr->sets, set_count * sizeof(*sets));
571 set_arr->count = set_count;
572 TAILQ_INSERT_TAIL(&rel_sets, set_arr, link);
573 break;
574 default:
575 error = EINVAL;
576 }
577 if (error)
578 goto out;
579 }
580
581 /*
582 * If there are no absolute levels, create a fake one at 100%. We
583 * then cache the clockrate for later use as our base frequency.
584 *
585 * XXX This assumes that the first time through, if we only have
586 * relative drivers, the CPU is currently running at 100%.
587 */
588 if (TAILQ_EMPTY(&sc->all_levels)) {
589 if (sc->max_mhz == CPUFREQ_VAL_UNKNOWN) {
590 pc = cpu_get_pcpu(dev);
591 cpu_est_clockrate(pc->pc_cpuid, &rate);
592 sc->max_mhz = rate / 1000000;
593 }
594 memset(&sets[0], CPUFREQ_VAL_UNKNOWN, sizeof(*sets));
595 sets[0].freq = sc->max_mhz;
596 sets[0].dev = NULL;
597 error = cpufreq_insert_abs(sc, sets, 1);
598 if (error)
599 goto out;
600 }
601
602 /* Create a combined list of absolute + relative levels. */
603 TAILQ_FOREACH(set_arr, &rel_sets, link)
604 cpufreq_expand_set(sc, set_arr);
605
606 /* If the caller doesn't have enough space, return the actual count. */
607 if (sc->all_count > *count) {
608 *count = sc->all_count;
609 error = E2BIG;
610 goto out;
611 }
612
613 /* Finally, output the list of levels. */
614 i = 0;
615 TAILQ_FOREACH(lev, &sc->all_levels, link) {
616 /*
617 * Skip levels that are too close in frequency to the
618 * previous levels. Some systems report bogus duplicate
619 * settings (i.e., for acpi_perf).
620 */
621 if (i > 0 && CPUFREQ_CMP(lev->total_set.freq,
622 levels[i - 1].total_set.freq)) {
623 sc->all_count--;
624 continue;
625 }
626
627 /* Skip levels that have a frequency that is too low. */
628 if (lev->total_set.freq < cf_lowest_freq) {
629 sc->all_count--;
630 continue;
631 }
632
633 levels[i] = *lev;
634 i++;
635 }
636 *count = sc->all_count;
637 error = 0;
638
639 out:
640 /* Clear all levels since we regenerate them each time. */
641 while ((lev = TAILQ_FIRST(&sc->all_levels)) != NULL) {
642 TAILQ_REMOVE(&sc->all_levels, lev, link);
643 free(lev, M_TEMP);
644 }
645 sc->all_count = 0;
646
647 CF_MTX_UNLOCK(&sc->lock);
648 while ((set_arr = TAILQ_FIRST(&rel_sets)) != NULL) {
649 TAILQ_REMOVE(&rel_sets, set_arr, link);
650 free(set_arr, M_TEMP);
651 }
652 free(devs, M_TEMP);
653 free(sets, M_TEMP);
654 return (error);
655 }
656
657 /*
658 * Create levels for an array of absolute settings and insert them in
659 * sorted order in the specified list.
660 */
661 static int
662 cpufreq_insert_abs(struct cpufreq_softc *sc, struct cf_setting *sets,
663 int count)
664 {
665 struct cf_level_lst *list;
666 struct cf_level *level, *search;
667 int i;
668
669 CF_MTX_ASSERT(&sc->lock);
670
671 list = &sc->all_levels;
672 for (i = 0; i < count; i++) {
673 level = malloc(sizeof(*level), M_TEMP, M_NOWAIT | M_ZERO);
674 if (level == NULL)
675 return (ENOMEM);
676 level->abs_set = sets[i];
677 level->total_set = sets[i];
678 level->total_set.dev = NULL;
679 sc->all_count++;
680
681 if (TAILQ_EMPTY(list)) {
682 CF_DEBUG("adding abs setting %d at head\n",
683 sets[i].freq);
684 TAILQ_INSERT_HEAD(list, level, link);
685 continue;
686 }
687
688 TAILQ_FOREACH_REVERSE(search, list, cf_level_lst, link) {
689 if (sets[i].freq <= search->total_set.freq) {
690 CF_DEBUG("adding abs setting %d after %d\n",
691 sets[i].freq, search->total_set.freq);
692 TAILQ_INSERT_AFTER(list, search, level, link);
693 break;
694 }
695 }
696 }
697 return (0);
698 }
699
700 /*
701 * Expand a group of relative settings, creating derived levels from them.
702 */
703 static int
704 cpufreq_expand_set(struct cpufreq_softc *sc, struct cf_setting_array *set_arr)
705 {
706 struct cf_level *fill, *search;
707 struct cf_setting *set;
708 int i;
709
710 CF_MTX_ASSERT(&sc->lock);
711
712 /*
713 * Walk the set of all existing levels in reverse. This is so we
714 * create derived states from the lowest absolute settings first
715 * and discard duplicates created from higher absolute settings.
716 * For instance, a level of 50 Mhz derived from 100 Mhz + 50% is
717 * preferable to 200 Mhz + 25% because absolute settings are more
718 * efficient since they often change the voltage as well.
719 */
720 TAILQ_FOREACH_REVERSE(search, &sc->all_levels, cf_level_lst, link) {
721 /* Add each setting to the level, duplicating if necessary. */
722 for (i = 0; i < set_arr->count; i++) {
723 set = &set_arr->sets[i];
724
725 /*
726 * If this setting is less than 100%, split the level
727 * into two and add this setting to the new level.
728 */
729 fill = search;
730 if (set->freq < 10000) {
731 fill = cpufreq_dup_set(sc, search, set);
732
733 /*
734 * The new level was a duplicate of an existing
735 * level or its absolute setting is too high
736 * so we freed it. For example, we discard a
737 * derived level of 1000 MHz/25% if a level
738 * of 500 MHz/100% already exists.
739 */
740 if (fill == NULL)
741 break;
742 }
743
744 /* Add this setting to the existing or new level. */
745 KASSERT(fill->rel_count < MAX_SETTINGS,
746 ("cpufreq: too many relative drivers (%d)",
747 MAX_SETTINGS));
748 fill->rel_set[fill->rel_count] = *set;
749 fill->rel_count++;
750 CF_DEBUG(
751 "expand set added rel setting %d%% to %d level\n",
752 set->freq / 100, fill->total_set.freq);
753 }
754 }
755
756 return (0);
757 }
758
759 static struct cf_level *
760 cpufreq_dup_set(struct cpufreq_softc *sc, struct cf_level *dup,
761 struct cf_setting *set)
762 {
763 struct cf_level_lst *list;
764 struct cf_level *fill, *itr;
765 struct cf_setting *fill_set, *itr_set;
766 int i;
767
768 CF_MTX_ASSERT(&sc->lock);
769
770 /*
771 * Create a new level, copy it from the old one, and update the
772 * total frequency and power by the percentage specified in the
773 * relative setting.
774 */
775 fill = malloc(sizeof(*fill), M_TEMP, M_NOWAIT);
776 if (fill == NULL)
777 return (NULL);
778 *fill = *dup;
779 fill_set = &fill->total_set;
780 fill_set->freq =
781 ((uint64_t)fill_set->freq * set->freq) / 10000;
782 if (fill_set->power != CPUFREQ_VAL_UNKNOWN) {
783 fill_set->power = ((uint64_t)fill_set->power * set->freq)
784 / 10000;
785 }
786 if (set->lat != CPUFREQ_VAL_UNKNOWN) {
787 if (fill_set->lat != CPUFREQ_VAL_UNKNOWN)
788 fill_set->lat += set->lat;
789 else
790 fill_set->lat = set->lat;
791 }
792 CF_DEBUG("dup set considering derived setting %d\n", fill_set->freq);
793
794 /*
795 * If we copied an old level that we already modified (say, at 100%),
796 * we need to remove that setting before adding this one. Since we
797 * process each setting array in order, we know any settings for this
798 * driver will be found at the end.
799 */
800 for (i = fill->rel_count; i != 0; i--) {
801 if (fill->rel_set[i - 1].dev != set->dev)
802 break;
803 CF_DEBUG("removed last relative driver: %s\n",
804 device_get_nameunit(set->dev));
805 fill->rel_count--;
806 }
807
808 /*
809 * Insert the new level in sorted order. If it is a duplicate of an
810 * existing level (1) or has an absolute setting higher than the
811 * existing level (2), do not add it. We can do this since any such
812 * level is guaranteed use less power. For example (1), a level with
813 * one absolute setting of 800 Mhz uses less power than one composed
814 * of an absolute setting of 1600 Mhz and a relative setting at 50%.
815 * Also for example (2), a level of 800 Mhz/75% is preferable to
816 * 1600 Mhz/25% even though the latter has a lower total frequency.
817 */
818 list = &sc->all_levels;
819 KASSERT(!TAILQ_EMPTY(list), ("all levels list empty in dup set"));
820 TAILQ_FOREACH_REVERSE(itr, list, cf_level_lst, link) {
821 itr_set = &itr->total_set;
822 if (CPUFREQ_CMP(fill_set->freq, itr_set->freq)) {
823 CF_DEBUG("dup set rejecting %d (dupe)\n",
824 fill_set->freq);
825 itr = NULL;
826 break;
827 } else if (fill_set->freq < itr_set->freq) {
828 if (fill->abs_set.freq <= itr->abs_set.freq) {
829 CF_DEBUG(
830 "dup done, inserting new level %d after %d\n",
831 fill_set->freq, itr_set->freq);
832 TAILQ_INSERT_AFTER(list, itr, fill, link);
833 sc->all_count++;
834 } else {
835 CF_DEBUG("dup set rejecting %d (abs too big)\n",
836 fill_set->freq);
837 itr = NULL;
838 }
839 break;
840 }
841 }
842
843 /* We didn't find a good place for this new level so free it. */
844 if (itr == NULL) {
845 CF_DEBUG("dup set freeing new level %d (not optimal)\n",
846 fill_set->freq);
847 free(fill, M_TEMP);
848 fill = NULL;
849 }
850
851 return (fill);
852 }
853
854 static int
855 cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS)
856 {
857 struct cpufreq_softc *sc;
858 struct cf_level *levels;
859 int count, devcount, error, freq, i, n;
860 device_t *devs;
861
862 devs = NULL;
863 sc = oidp->oid_arg1;
864 levels = malloc(CF_MAX_LEVELS * sizeof(*levels), M_TEMP, M_NOWAIT);
865 if (levels == NULL)
866 return (ENOMEM);
867
868 error = CPUFREQ_GET(sc->dev, &levels[0]);
869 if (error)
870 goto out;
871 freq = levels[0].total_set.freq;
872 error = sysctl_handle_int(oidp, &freq, 0, req);
873 if (error != 0 || req->newptr == NULL)
874 goto out;
875
876 /*
877 * While we only call cpufreq_get() on one device (assuming all
878 * CPUs have equal levels), we call cpufreq_set() on all CPUs.
879 * This is needed for some MP systems.
880 */
881 error = devclass_get_devices(cpufreq_dc, &devs, &devcount);
882 if (error)
883 goto out;
884 for (n = 0; n < devcount; n++) {
885 count = CF_MAX_LEVELS;
886 error = CPUFREQ_LEVELS(devs[n], levels, &count);
887 if (error) {
888 if (error == E2BIG)
889 printf(
890 "cpufreq: need to increase CF_MAX_LEVELS\n");
891 break;
892 }
893 for (i = 0; i < count; i++) {
894 if (CPUFREQ_CMP(levels[i].total_set.freq, freq)) {
895 error = CPUFREQ_SET(devs[n], &levels[i],
896 CPUFREQ_PRIO_USER);
897 break;
898 }
899 }
900 if (i == count) {
901 error = EINVAL;
902 break;
903 }
904 }
905
906 out:
907 if (devs)
908 free(devs, M_TEMP);
909 if (levels)
910 free(levels, M_TEMP);
911 return (error);
912 }
913
914 static int
915 cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS)
916 {
917 struct cpufreq_softc *sc;
918 struct cf_level *levels;
919 struct cf_setting *set;
920 struct sbuf sb;
921 int count, error, i;
922
923 sc = oidp->oid_arg1;
924 sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
925
926 /* Get settings from the device and generate the output string. */
927 count = CF_MAX_LEVELS;
928 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
929 if (levels == NULL) {
930 sbuf_delete(&sb);
931 return (ENOMEM);
932 }
933 error = CPUFREQ_LEVELS(sc->dev, levels, &count);
934 if (error) {
935 if (error == E2BIG)
936 printf("cpufreq: need to increase CF_MAX_LEVELS\n");
937 goto out;
938 }
939 if (count) {
940 for (i = 0; i < count; i++) {
941 set = &levels[i].total_set;
942 sbuf_printf(&sb, "%d/%d ", set->freq, set->power);
943 }
944 } else
945 sbuf_cpy(&sb, "");
946 sbuf_trim(&sb);
947 sbuf_finish(&sb);
948 error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
949
950 out:
951 free(levels, M_TEMP);
952 sbuf_delete(&sb);
953 return (error);
954 }
955
956 static int
957 cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS)
958 {
959 device_t dev;
960 struct cf_setting *sets;
961 struct sbuf sb;
962 int error, i, set_count;
963
964 dev = oidp->oid_arg1;
965 sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
966
967 /* Get settings from the device and generate the output string. */
968 set_count = MAX_SETTINGS;
969 sets = malloc(set_count * sizeof(*sets), M_TEMP, M_NOWAIT);
970 if (sets == NULL) {
971 sbuf_delete(&sb);
972 return (ENOMEM);
973 }
974 error = CPUFREQ_DRV_SETTINGS(dev, sets, &set_count);
975 if (error)
976 goto out;
977 if (set_count) {
978 for (i = 0; i < set_count; i++)
979 sbuf_printf(&sb, "%d/%d ", sets[i].freq, sets[i].power);
980 } else
981 sbuf_cpy(&sb, "");
982 sbuf_trim(&sb);
983 sbuf_finish(&sb);
984 error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
985
986 out:
987 free(sets, M_TEMP);
988 sbuf_delete(&sb);
989 return (error);
990 }
991
992 int
993 cpufreq_register(device_t dev)
994 {
995 struct cpufreq_softc *sc;
996 device_t cf_dev, cpu_dev;
997
998 /* Add a sysctl to get each driver's settings separately. */
999 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
1000 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
1001 OID_AUTO, "freq_settings", CTLTYPE_STRING | CTLFLAG_RD, dev, 0,
1002 cpufreq_settings_sysctl, "A", "CPU frequency driver settings");
1003
1004 /*
1005 * Add only one cpufreq device to each CPU. Currently, all CPUs
1006 * must offer the same levels and be switched at the same time.
1007 */
1008 cpu_dev = device_get_parent(dev);
1009 if ((cf_dev = device_find_child(cpu_dev, "cpufreq", -1))) {
1010 sc = device_get_softc(cf_dev);
1011 sc->max_mhz = CPUFREQ_VAL_UNKNOWN;
1012 return (0);
1013 }
1014
1015 /* Add the child device and possibly sysctls. */
1016 cf_dev = BUS_ADD_CHILD(cpu_dev, 0, "cpufreq", -1);
1017 if (cf_dev == NULL)
1018 return (ENOMEM);
1019 device_quiet(cf_dev);
1020
1021 return (device_probe_and_attach(cf_dev));
1022 }
1023
1024 int
1025 cpufreq_unregister(device_t dev)
1026 {
1027 device_t cf_dev, *devs;
1028 int cfcount, devcount, error, i, type;
1029
1030 /*
1031 * If this is the last cpufreq child device, remove the control
1032 * device as well. We identify cpufreq children by calling a method
1033 * they support.
1034 */
1035 error = device_get_children(device_get_parent(dev), &devs, &devcount);
1036 if (error)
1037 return (error);
1038 cf_dev = device_find_child(device_get_parent(dev), "cpufreq", -1);
1039 if (cf_dev == NULL) {
1040 device_printf(dev,
1041 "warning: cpufreq_unregister called with no cpufreq device active\n");
1042 return (0);
1043 }
1044 cfcount = 0;
1045 for (i = 0; i < devcount; i++) {
1046 if (!device_is_attached(devs[i]))
1047 continue;
1048 if (CPUFREQ_DRV_TYPE(devs[i], &type) == 0)
1049 cfcount++;
1050 }
1051 if (cfcount <= 1)
1052 device_delete_child(device_get_parent(cf_dev), cf_dev);
1053 free(devs, M_TEMP);
1054
1055 return (0);
1056 }
1057
1058 int
1059 cpufreq_settings_changed(device_t dev)
1060 {
1061
1062 EVENTHANDLER_INVOKE(cpufreq_levels_changed,
1063 device_get_unit(device_get_parent(dev)));
1064 return (0);
1065 }
Cache object: e8cb00b5417d2751dd754b9d56376c8d
|