kern_cpu.c revision 142590
1/*- 2 * Copyright (c) 2004-2005 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: head/sys/kern/kern_cpu.c 142590 2005-02-26 22:37:49Z njl $"); 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/malloc.h> 36#include <sys/module.h> 37#include <sys/proc.h> 38#include <sys/queue.h> 39#include <sys/sched.h> 40#include <sys/sysctl.h> 41#include <sys/systm.h> 42#include <sys/sbuf.h> 43#include <sys/timetc.h> 44 45#include "cpufreq_if.h" 46 47/* 48 * Common CPU frequency glue code. Drivers for specific hardware can 49 * attach this interface to allow users to get/set the CPU frequency. 50 */ 51 52/* 53 * Number of levels we can handle. Levels are synthesized from settings 54 * so for M settings and N drivers, there may be M*N levels. 55 */ 56#define CF_MAX_LEVELS 64 57 58struct cpufreq_softc { 59 struct cf_level curr_level; 60 int curr_priority; 61 struct cf_level saved_level; 62 int saved_priority; 63 struct cf_level_lst all_levels; 64 int all_count; 65 int max_mhz; 66 device_t dev; 67 struct sysctl_ctx_list sysctl_ctx; 68}; 69 70struct cf_setting_array { 71 struct cf_setting sets[MAX_SETTINGS]; 72 int count; 73 TAILQ_ENTRY(cf_setting_array) link; 74}; 75 76TAILQ_HEAD(cf_setting_lst, cf_setting_array); 77 78static int cpufreq_attach(device_t dev); 79static int cpufreq_detach(device_t dev); 80static void cpufreq_evaluate(void *arg); 81static int cf_set_method(device_t dev, const struct cf_level *level, 82 int priority); 83static int cf_get_method(device_t dev, struct cf_level *level); 84static int cf_levels_method(device_t dev, struct cf_level *levels, 85 int *count); 86static int cpufreq_insert_abs(struct cpufreq_softc *sc, 87 struct cf_setting *sets, int count); 88static int cpufreq_expand_set(struct cpufreq_softc *sc, 89 struct cf_setting_array *set_arr); 90static struct cf_level *cpufreq_dup_set(struct cpufreq_softc *sc, 91 struct cf_level *dup, struct cf_setting *set); 92static int cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS); 93static int cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS); 94static int cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS); 95 96static device_method_t cpufreq_methods[] = { 97 DEVMETHOD(device_probe, bus_generic_probe), 98 DEVMETHOD(device_attach, cpufreq_attach), 99 DEVMETHOD(device_detach, cpufreq_detach), 100 101 DEVMETHOD(cpufreq_set, cf_set_method), 102 DEVMETHOD(cpufreq_get, cf_get_method), 103 DEVMETHOD(cpufreq_levels, cf_levels_method), 104 {0, 0} 105}; 106static driver_t cpufreq_driver = { 107 "cpufreq", cpufreq_methods, sizeof(struct cpufreq_softc) 108}; 109static devclass_t cpufreq_dc; 110DRIVER_MODULE(cpufreq, cpu, cpufreq_driver, cpufreq_dc, 0, 0); 111 112static eventhandler_tag cf_ev_tag; 113 114static int cf_lowest_freq; 115TUNABLE_INT("debug.cpufreq.lowest", &cf_lowest_freq); 116SYSCTL_NODE(_debug, OID_AUTO, cpufreq, CTLFLAG_RD, NULL, "cpufreq debugging"); 117SYSCTL_INT(_debug_cpufreq, OID_AUTO, lowest, CTLFLAG_RW, &cf_lowest_freq, 1, 118 "Don't provide levels below this frequency."); 119 120static int 121cpufreq_attach(device_t dev) 122{ 123 struct cpufreq_softc *sc; 124 device_t parent; 125 int numdevs; 126 127 sc = device_get_softc(dev); 128 parent = device_get_parent(dev); 129 sc->dev = dev; 130 sysctl_ctx_init(&sc->sysctl_ctx); 131 TAILQ_INIT(&sc->all_levels); 132 sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 133 sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 134 sc->max_mhz = CPUFREQ_VAL_UNKNOWN; 135 136 /* 137 * Only initialize one set of sysctls for all CPUs. In the future, 138 * if multiple CPUs can have different settings, we can move these 139 * sysctls to be under every CPU instead of just the first one. 140 */ 141 numdevs = devclass_get_count(cpufreq_dc); 142 if (numdevs > 1) 143 return (0); 144 145 SYSCTL_ADD_PROC(&sc->sysctl_ctx, 146 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 147 OID_AUTO, "freq", CTLTYPE_INT | CTLFLAG_RW, sc, 0, 148 cpufreq_curr_sysctl, "I", "Current CPU frequency"); 149 SYSCTL_ADD_PROC(&sc->sysctl_ctx, 150 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 151 OID_AUTO, "freq_levels", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, 152 cpufreq_levels_sysctl, "A", "CPU frequency levels"); 153 cf_ev_tag = EVENTHANDLER_REGISTER(cpufreq_changed, cpufreq_evaluate, 154 NULL, EVENTHANDLER_PRI_ANY); 155 156 return (0); 157} 158 159static int 160cpufreq_detach(device_t dev) 161{ 162 struct cpufreq_softc *sc; 163 int numdevs; 164 165 sc = device_get_softc(dev); 166 sysctl_ctx_free(&sc->sysctl_ctx); 167 168 /* Only clean up these resources when the last device is detaching. */ 169 numdevs = devclass_get_count(cpufreq_dc); 170 if (numdevs == 1) 171 EVENTHANDLER_DEREGISTER(cpufreq_changed, cf_ev_tag); 172 173 return (0); 174} 175 176static void 177cpufreq_evaluate(void *arg) 178{ 179 /* TODO: Re-evaluate when notified of changes to drivers. */ 180} 181 182static int 183cf_set_method(device_t dev, const struct cf_level *level, int priority) 184{ 185 struct cpufreq_softc *sc; 186 const struct cf_setting *set; 187 struct pcpu *pc; 188 int cpu_id, error, i; 189 190 sc = device_get_softc(dev); 191 192 /* 193 * Check that the TSC isn't being used as a timecounter. 194 * If it is, then return EBUSY and refuse to change the 195 * clock speed. 196 */ 197 if (strcmp(timecounter->tc_name, "TSC") == 0) 198 return (EBUSY); 199 200 /* 201 * If the caller didn't specify a level and one is saved, prepare to 202 * restore the saved level. If none has been saved, return an error. 203 * If they did specify one, but the requested level has a lower 204 * priority, don't allow the new level right now. 205 */ 206 if (level == NULL) { 207 if (sc->saved_level.total_set.freq != CPUFREQ_VAL_UNKNOWN) { 208 level = &sc->saved_level; 209 priority = sc->saved_priority; 210 } else 211 return (ENXIO); 212 } else if (priority < sc->curr_priority) 213 return (EPERM); 214 215 /* Reject levels that are below our specified threshold. */ 216 if (level->total_set.freq <= cf_lowest_freq) 217 return (EINVAL); 218 219 /* If already at this level, just return. */ 220 if (CPUFREQ_CMP(sc->curr_level.total_set.freq, level->total_set.freq)) 221 return (0); 222 223 /* First, set the absolute frequency via its driver. */ 224 set = &level->abs_set; 225 if (set->dev) { 226 if (!device_is_attached(set->dev)) { 227 error = ENXIO; 228 goto out; 229 } 230 231 /* Bind to the target CPU before switching, if necessary. */ 232 cpu_id = PCPU_GET(cpuid); 233 pc = cpu_get_pcpu(set->dev); 234 if (cpu_id != pc->pc_cpuid) { 235 mtx_lock_spin(&sched_lock); 236 sched_bind(curthread, pc->pc_cpuid); 237 mtx_unlock_spin(&sched_lock); 238 } 239 error = CPUFREQ_DRV_SET(set->dev, set); 240 if (cpu_id != pc->pc_cpuid) { 241 mtx_lock_spin(&sched_lock); 242 sched_unbind(curthread); 243 mtx_unlock_spin(&sched_lock); 244 } 245 if (error) { 246 goto out; 247 } 248 } 249 250 /* Next, set any/all relative frequencies via their drivers. */ 251 for (i = 0; i < level->rel_count; i++) { 252 set = &level->rel_set[i]; 253 if (!device_is_attached(set->dev)) { 254 error = ENXIO; 255 goto out; 256 } 257 258 /* Bind to the target CPU before switching, if necessary. */ 259 cpu_id = PCPU_GET(cpuid); 260 pc = cpu_get_pcpu(set->dev); 261 if (cpu_id != pc->pc_cpuid) { 262 mtx_lock_spin(&sched_lock); 263 sched_bind(curthread, pc->pc_cpuid); 264 mtx_unlock_spin(&sched_lock); 265 } 266 error = CPUFREQ_DRV_SET(set->dev, set); 267 if (cpu_id != pc->pc_cpuid) { 268 mtx_lock_spin(&sched_lock); 269 sched_unbind(curthread); 270 mtx_unlock_spin(&sched_lock); 271 } 272 if (error) { 273 /* XXX Back out any successful setting? */ 274 goto out; 275 } 276 } 277 278 /* If we were restoring a saved state, reset it to "unused". */ 279 if (level == &sc->saved_level) { 280 sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 281 sc->saved_priority = 0; 282 } 283 284 /* 285 * Before recording the current level, check if we're going to a 286 * higher priority and have not saved a level yet. If so, save the 287 * previous level and priority. 288 */ 289 if (sc->curr_level.total_set.freq != CPUFREQ_VAL_UNKNOWN && 290 sc->saved_level.total_set.freq == CPUFREQ_VAL_UNKNOWN && 291 priority > sc->curr_priority) { 292 sc->saved_level = sc->curr_level; 293 sc->saved_priority = sc->curr_priority; 294 } 295 sc->curr_level = *level; 296 sc->curr_priority = priority; 297 error = 0; 298 299out: 300 if (error) 301 device_printf(set->dev, "set freq failed, err %d\n", error); 302 return (error); 303} 304 305static int 306cf_get_method(device_t dev, struct cf_level *level) 307{ 308 struct cpufreq_softc *sc; 309 struct cf_level *levels; 310 struct cf_setting *curr_set, set; 311 struct pcpu *pc; 312 device_t *devs; 313 int count, error, i, numdevs; 314 uint64_t rate; 315 316 sc = device_get_softc(dev); 317 curr_set = &sc->curr_level.total_set; 318 levels = NULL; 319 320 /* If we already know the current frequency, we're done. */ 321 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 322 goto out; 323 324 /* 325 * We need to figure out the current level. Loop through every 326 * driver, getting the current setting. Then, attempt to get a best 327 * match of settings against each level. 328 */ 329 count = CF_MAX_LEVELS; 330 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 331 if (levels == NULL) 332 return (ENOMEM); 333 error = CPUFREQ_LEVELS(sc->dev, levels, &count); 334 if (error) { 335 if (error == E2BIG) 336 printf("cpufreq: need to increase CF_MAX_LEVELS\n"); 337 goto out; 338 } 339 error = device_get_children(device_get_parent(dev), &devs, &numdevs); 340 if (error) 341 goto out; 342 for (i = 0; i < numdevs && curr_set->freq == CPUFREQ_VAL_UNKNOWN; i++) { 343 if (!device_is_attached(devs[i])) 344 continue; 345 error = CPUFREQ_DRV_GET(devs[i], &set); 346 if (error) 347 continue; 348 for (i = 0; i < count; i++) { 349 if (CPUFREQ_CMP(set.freq, levels[i].total_set.freq)) { 350 sc->curr_level = levels[i]; 351 break; 352 } 353 } 354 } 355 free(devs, M_TEMP); 356 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 357 goto out; 358 359 /* 360 * We couldn't find an exact match, so attempt to estimate and then 361 * match against a level. 362 */ 363 pc = cpu_get_pcpu(dev); 364 if (pc == NULL) { 365 error = ENXIO; 366 goto out; 367 } 368 cpu_est_clockrate(pc->pc_cpuid, &rate); 369 rate /= 1000000; 370 for (i = 0; i < count; i++) { 371 if (CPUFREQ_CMP(rate, levels[i].total_set.freq)) { 372 sc->curr_level = levels[i]; 373 break; 374 } 375 } 376 377out: 378 if (levels) 379 free(levels, M_TEMP); 380 *level = sc->curr_level; 381 return (0); 382} 383 384static int 385cf_levels_method(device_t dev, struct cf_level *levels, int *count) 386{ 387 struct cf_setting_array *set_arr; 388 struct cf_setting_lst rel_sets; 389 struct cpufreq_softc *sc; 390 struct cf_level *lev; 391 struct cf_setting *sets; 392 struct pcpu *pc; 393 device_t *devs; 394 int error, i, numdevs, set_count, type; 395 uint64_t rate; 396 397 if (levels == NULL || count == NULL) 398 return (EINVAL); 399 400 TAILQ_INIT(&rel_sets); 401 sc = device_get_softc(dev); 402 error = device_get_children(device_get_parent(dev), &devs, &numdevs); 403 if (error) 404 return (error); 405 sets = malloc(MAX_SETTINGS * sizeof(*sets), M_TEMP, M_NOWAIT); 406 if (sets == NULL) { 407 free(devs, M_TEMP); 408 return (ENOMEM); 409 } 410 411 /* Get settings from all cpufreq drivers. */ 412 for (i = 0; i < numdevs; i++) { 413 /* Skip devices that aren't ready. */ 414 if (!device_is_attached(devs[i])) 415 continue; 416 417 /* 418 * Get settings, skipping drivers that offer no settings or 419 * provide settings for informational purposes only. 420 */ 421 error = CPUFREQ_DRV_TYPE(devs[i], &type); 422 if (error || (type & CPUFREQ_FLAG_INFO_ONLY)) 423 continue; 424 set_count = MAX_SETTINGS; 425 error = CPUFREQ_DRV_SETTINGS(devs[i], sets, &set_count); 426 if (error || set_count == 0) 427 continue; 428 429 /* Add the settings to our absolute/relative lists. */ 430 switch (type & CPUFREQ_TYPE_MASK) { 431 case CPUFREQ_TYPE_ABSOLUTE: 432 error = cpufreq_insert_abs(sc, sets, set_count); 433 break; 434 case CPUFREQ_TYPE_RELATIVE: 435 set_arr = malloc(sizeof(*set_arr), M_TEMP, M_NOWAIT); 436 if (set_arr == NULL) { 437 error = ENOMEM; 438 goto out; 439 } 440 bcopy(sets, set_arr->sets, set_count * sizeof(*sets)); 441 set_arr->count = set_count; 442 TAILQ_INSERT_TAIL(&rel_sets, set_arr, link); 443 break; 444 default: 445 error = EINVAL; 446 break; 447 } 448 if (error) 449 goto out; 450 } 451 452 /* 453 * If there are no absolute levels, create a fake one at 100%. We 454 * then cache the clockrate for later use as our base frequency. 455 * 456 * XXX This assumes that the first time through, if we only have 457 * relative drivers, the CPU is currently running at 100%. 458 */ 459 if (TAILQ_EMPTY(&sc->all_levels)) { 460 if (sc->max_mhz == CPUFREQ_VAL_UNKNOWN) { 461 pc = cpu_get_pcpu(dev); 462 cpu_est_clockrate(pc->pc_cpuid, &rate); 463 sc->max_mhz = rate / 1000000; 464 } 465 memset(&sets[0], CPUFREQ_VAL_UNKNOWN, sizeof(*sets)); 466 sets[0].freq = sc->max_mhz; 467 sets[0].dev = NULL; 468 error = cpufreq_insert_abs(sc, sets, 1); 469 if (error) 470 goto out; 471 } 472 473 /* Create a combined list of absolute + relative levels. */ 474 TAILQ_FOREACH(set_arr, &rel_sets, link) 475 cpufreq_expand_set(sc, set_arr); 476 477 /* If the caller doesn't have enough space, return the actual count. */ 478 if (sc->all_count > *count) { 479 *count = sc->all_count; 480 error = E2BIG; 481 goto out; 482 } 483 484 /* Finally, output the list of levels. */ 485 i = 0; 486 TAILQ_FOREACH(lev, &sc->all_levels, link) { 487 /* Skip levels that have a frequency that is too low. */ 488 if (lev->total_set.freq <= cf_lowest_freq) { 489 sc->all_count--; 490 continue; 491 } 492 493 levels[i] = *lev; 494 i++; 495 } 496 *count = sc->all_count; 497 error = 0; 498 499out: 500 /* Clear all levels since we regenerate them each time. */ 501 while ((lev = TAILQ_FIRST(&sc->all_levels)) != NULL) { 502 TAILQ_REMOVE(&sc->all_levels, lev, link); 503 free(lev, M_TEMP); 504 } 505 while ((set_arr = TAILQ_FIRST(&rel_sets)) != NULL) { 506 TAILQ_REMOVE(&rel_sets, set_arr, link); 507 free(set_arr, M_TEMP); 508 } 509 sc->all_count = 0; 510 free(devs, M_TEMP); 511 free(sets, M_TEMP); 512 return (error); 513} 514 515/* 516 * Create levels for an array of absolute settings and insert them in 517 * sorted order in the specified list. 518 */ 519static int 520cpufreq_insert_abs(struct cpufreq_softc *sc, struct cf_setting *sets, 521 int count) 522{ 523 struct cf_level_lst *list; 524 struct cf_level *level, *search; 525 int i; 526 527 list = &sc->all_levels; 528 for (i = 0; i < count; i++) { 529 level = malloc(sizeof(*level), M_TEMP, M_NOWAIT | M_ZERO); 530 if (level == NULL) 531 return (ENOMEM); 532 level->abs_set = sets[i]; 533 level->total_set = sets[i]; 534 level->total_set.dev = NULL; 535 sc->all_count++; 536 537 if (TAILQ_EMPTY(list)) { 538 TAILQ_INSERT_HEAD(list, level, link); 539 continue; 540 } 541 542 TAILQ_FOREACH_REVERSE(search, list, cf_level_lst, link) { 543 if (sets[i].freq <= search->total_set.freq) { 544 TAILQ_INSERT_AFTER(list, search, level, link); 545 break; 546 } 547 } 548 } 549 return (0); 550} 551 552/* 553 * Expand a group of relative settings, creating derived levels from them. 554 */ 555static int 556cpufreq_expand_set(struct cpufreq_softc *sc, struct cf_setting_array *set_arr) 557{ 558 struct cf_level *fill, *search; 559 struct cf_setting *set; 560 int i; 561 562 TAILQ_FOREACH(search, &sc->all_levels, link) { 563 /* Skip this level if we've already modified it. */ 564 for (i = 0; i < search->rel_count; i++) { 565 if (search->rel_set[i].dev == set_arr->sets[0].dev) 566 break; 567 } 568 if (i != search->rel_count) 569 continue; 570 571 /* Add each setting to the level, duplicating if necessary. */ 572 for (i = 0; i < set_arr->count; i++) { 573 set = &set_arr->sets[i]; 574 575 /* 576 * If this setting is less than 100%, split the level 577 * into two and add this setting to the new level. 578 */ 579 fill = search; 580 if (set->freq < 10000) 581 fill = cpufreq_dup_set(sc, search, set); 582 583 /* 584 * The new level was a duplicate of an existing level 585 * so we freed it. Go to the next setting. 586 */ 587 if (fill == NULL) 588 continue; 589 590 /* Add this setting to the existing or new level. */ 591 KASSERT(fill->rel_count < MAX_SETTINGS, 592 ("cpufreq: too many relative drivers (%d)", 593 MAX_SETTINGS)); 594 fill->rel_set[fill->rel_count] = *set; 595 fill->rel_count++; 596 } 597 } 598 599 return (0); 600} 601 602static struct cf_level * 603cpufreq_dup_set(struct cpufreq_softc *sc, struct cf_level *dup, 604 struct cf_setting *set) 605{ 606 struct cf_level_lst *list; 607 struct cf_level *fill, *itr; 608 struct cf_setting *fill_set, *itr_set; 609 int i; 610 611 /* 612 * Create a new level, copy it from the old one, and update the 613 * total frequency and power by the percentage specified in the 614 * relative setting. 615 */ 616 fill = malloc(sizeof(*fill), M_TEMP, M_NOWAIT); 617 if (fill == NULL) 618 return (NULL); 619 *fill = *dup; 620 fill_set = &fill->total_set; 621 fill_set->freq = 622 ((uint64_t)fill_set->freq * set->freq) / 10000; 623 if (fill_set->power != CPUFREQ_VAL_UNKNOWN) { 624 fill_set->power = ((uint64_t)fill_set->power * set->freq) 625 / 10000; 626 } 627 if (set->lat != CPUFREQ_VAL_UNKNOWN) { 628 if (fill_set->lat != CPUFREQ_VAL_UNKNOWN) 629 fill_set->lat += set->lat; 630 else 631 fill_set->lat = set->lat; 632 } 633 634 /* 635 * If we copied an old level that we already modified (say, at 100%), 636 * we need to remove that setting before adding this one. Since we 637 * process each setting array in order, we know any settings for this 638 * driver will be found at the end. 639 */ 640 for (i = fill->rel_count; i != 0; i--) { 641 if (fill->rel_set[i - 1].dev != set->dev) 642 break; 643 fill->rel_count--; 644 } 645 646 /* 647 * Insert the new level in sorted order. If we find a duplicate, 648 * free the new level. We can do this since any existing level will 649 * be guaranteed to have the same or less settings and thus consume 650 * less power. For example, a level with one absolute setting of 651 * 800 Mhz uses less power than one composed of an absolute setting 652 * of 1600 Mhz and a relative setting at 50%. 653 */ 654 list = &sc->all_levels; 655 if (TAILQ_EMPTY(list)) { 656 TAILQ_INSERT_HEAD(list, fill, link); 657 } else { 658 TAILQ_FOREACH_REVERSE(itr, list, cf_level_lst, link) { 659 itr_set = &itr->total_set; 660 if (CPUFREQ_CMP(fill_set->freq, itr_set->freq)) { 661 free(fill, M_TEMP); 662 fill = NULL; 663 break; 664 } else if (fill_set->freq < itr_set->freq) { 665 TAILQ_INSERT_AFTER(list, itr, fill, link); 666 sc->all_count++; 667 break; 668 } 669 } 670 } 671 672 return (fill); 673} 674 675static int 676cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS) 677{ 678 struct cpufreq_softc *sc; 679 struct cf_level *levels; 680 int count, devcount, error, freq, i, n; 681 device_t *devs; 682 683 devs = NULL; 684 sc = oidp->oid_arg1; 685 levels = malloc(CF_MAX_LEVELS * sizeof(*levels), M_TEMP, M_NOWAIT); 686 if (levels == NULL) 687 return (ENOMEM); 688 689 error = CPUFREQ_GET(sc->dev, &levels[0]); 690 if (error) 691 goto out; 692 freq = levels[0].total_set.freq; 693 error = sysctl_handle_int(oidp, &freq, 0, req); 694 if (error != 0 || req->newptr == NULL) 695 goto out; 696 697 /* 698 * While we only call cpufreq_get() on one device (assuming all 699 * CPUs have equal levels), we call cpufreq_set() on all CPUs. 700 * This is needed for some MP systems. 701 */ 702 error = devclass_get_devices(cpufreq_dc, &devs, &devcount); 703 if (error) 704 goto out; 705 for (n = 0; n < devcount; n++) { 706 count = CF_MAX_LEVELS; 707 error = CPUFREQ_LEVELS(devs[n], levels, &count); 708 if (error) { 709 if (error == E2BIG) 710 printf( 711 "cpufreq: need to increase CF_MAX_LEVELS\n"); 712 break; 713 } 714 for (i = 0; i < count; i++) { 715 if (CPUFREQ_CMP(levels[i].total_set.freq, freq)) { 716 error = CPUFREQ_SET(devs[n], &levels[i], 717 CPUFREQ_PRIO_USER); 718 break; 719 } 720 } 721 if (i == count) { 722 error = EINVAL; 723 break; 724 } 725 } 726 727out: 728 if (devs) 729 free(devs, M_TEMP); 730 if (levels) 731 free(levels, M_TEMP); 732 return (error); 733} 734 735static int 736cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS) 737{ 738 struct cpufreq_softc *sc; 739 struct cf_level *levels; 740 struct cf_setting *set; 741 struct sbuf sb; 742 int count, error, i; 743 744 sc = oidp->oid_arg1; 745 sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND); 746 747 /* Get settings from the device and generate the output string. */ 748 count = CF_MAX_LEVELS; 749 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 750 if (levels == NULL) 751 return (ENOMEM); 752 error = CPUFREQ_LEVELS(sc->dev, levels, &count); 753 if (error) { 754 if (error == E2BIG) 755 printf("cpufreq: need to increase CF_MAX_LEVELS\n"); 756 goto out; 757 } 758 if (count) { 759 for (i = 0; i < count; i++) { 760 set = &levels[i].total_set; 761 sbuf_printf(&sb, "%d/%d ", set->freq, set->power); 762 } 763 } else 764 sbuf_cpy(&sb, "0"); 765 sbuf_trim(&sb); 766 sbuf_finish(&sb); 767 error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req); 768 769out: 770 free(levels, M_TEMP); 771 sbuf_delete(&sb); 772 return (error); 773} 774 775static int 776cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS) 777{ 778 device_t dev; 779 struct cf_setting *sets; 780 struct sbuf sb; 781 int error, i, set_count; 782 783 dev = oidp->oid_arg1; 784 sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND); 785 786 /* Get settings from the device and generate the output string. */ 787 set_count = MAX_SETTINGS; 788 sets = malloc(set_count * sizeof(*sets), M_TEMP, M_NOWAIT); 789 if (sets == NULL) 790 return (ENOMEM); 791 error = CPUFREQ_DRV_SETTINGS(dev, sets, &set_count); 792 if (error) 793 goto out; 794 if (set_count) { 795 for (i = 0; i < set_count; i++) 796 sbuf_printf(&sb, "%d/%d ", sets[i].freq, sets[i].power); 797 } else 798 sbuf_cpy(&sb, "0"); 799 sbuf_trim(&sb); 800 sbuf_finish(&sb); 801 error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req); 802 803out: 804 free(sets, M_TEMP); 805 sbuf_delete(&sb); 806 return (error); 807} 808 809int 810cpufreq_register(device_t dev) 811{ 812 struct cpufreq_softc *sc; 813 device_t cf_dev, cpu_dev; 814 815 /* Add a sysctl to get each driver's settings separately. */ 816 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), 817 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), 818 OID_AUTO, "freq_settings", CTLTYPE_STRING | CTLFLAG_RD, dev, 0, 819 cpufreq_settings_sysctl, "A", "CPU frequency driver settings"); 820 821 /* 822 * Add only one cpufreq device to each CPU. Currently, all CPUs 823 * must offer the same levels and be switched at the same time. 824 */ 825 cpu_dev = device_get_parent(dev); 826 if ((cf_dev = device_find_child(cpu_dev, "cpufreq", -1))) { 827 sc = device_get_softc(cf_dev); 828 sc->max_mhz = CPUFREQ_VAL_UNKNOWN; 829 return (0); 830 } 831 832 /* Add the child device and possibly sysctls. */ 833 cf_dev = BUS_ADD_CHILD(cpu_dev, 0, "cpufreq", -1); 834 if (cf_dev == NULL) 835 return (ENOMEM); 836 device_quiet(cf_dev); 837 838 return (device_probe_and_attach(cf_dev)); 839} 840 841int 842cpufreq_unregister(device_t dev) 843{ 844 device_t cf_dev, *devs; 845 int cfcount, devcount, error, i, type; 846 847 /* 848 * If this is the last cpufreq child device, remove the control 849 * device as well. We identify cpufreq children by calling a method 850 * they support. 851 */ 852 error = device_get_children(device_get_parent(dev), &devs, &devcount); 853 if (error) 854 return (error); 855 cf_dev = device_find_child(device_get_parent(dev), "cpufreq", -1); 856 cfcount = 0; 857 for (i = 0; i < devcount; i++) { 858 if (!device_is_attached(devs[i])) 859 continue; 860 if (CPUFREQ_DRV_TYPE(devs[i], &type) == 0) 861 cfcount++; 862 } 863 if (cfcount <= 1) 864 device_delete_child(device_get_parent(cf_dev), cf_dev); 865 free(devs, M_TEMP); 866 867 return (0); 868} 869