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