kern_cpu.c revision 141945
1141240Snjl/*- 2141240Snjl * Copyright (c) 2004-2005 Nate Lawson (SDG) 3141240Snjl * All rights reserved. 4141240Snjl * 5141240Snjl * Redistribution and use in source and binary forms, with or without 6141240Snjl * modification, are permitted provided that the following conditions 7141240Snjl * are met: 8141240Snjl * 1. Redistributions of source code must retain the above copyright 9141240Snjl * notice, this list of conditions and the following disclaimer. 10141240Snjl * 2. Redistributions in binary form must reproduce the above copyright 11141240Snjl * notice, this list of conditions and the following disclaimer in the 12141240Snjl * documentation and/or other materials provided with the distribution. 13141240Snjl * 14141240Snjl * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15141240Snjl * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16141240Snjl * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17141240Snjl * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18141240Snjl * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19141240Snjl * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20141240Snjl * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21141240Snjl * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22141240Snjl * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23141240Snjl * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24141240Snjl * SUCH DAMAGE. 25141240Snjl */ 26141240Snjl 27141240Snjl#include <sys/cdefs.h> 28141240Snjl__FBSDID("$FreeBSD: head/sys/kern/kern_cpu.c 141945 2005-02-15 07:43:48Z njl $"); 29141240Snjl 30141240Snjl#include <sys/param.h> 31141240Snjl#include <sys/bus.h> 32141240Snjl#include <sys/cpu.h> 33141240Snjl#include <sys/eventhandler.h> 34141240Snjl#include <sys/kernel.h> 35141240Snjl#include <sys/malloc.h> 36141240Snjl#include <sys/module.h> 37141240Snjl#include <sys/proc.h> 38141240Snjl#include <sys/queue.h> 39141240Snjl#include <sys/sched.h> 40141240Snjl#include <sys/sysctl.h> 41141240Snjl#include <sys/systm.h> 42141240Snjl#include <sys/sbuf.h> 43141814Snjl#include <sys/timetc.h> 44141240Snjl 45141240Snjl#include "cpufreq_if.h" 46141240Snjl 47141240Snjl/* 48141240Snjl * Common CPU frequency glue code. Drivers for specific hardware can 49141240Snjl * attach this interface to allow users to get/set the CPU frequency. 50141240Snjl */ 51141240Snjl 52141240Snjl/* 53141240Snjl * Number of levels we can handle. Levels are synthesized from settings 54141240Snjl * so for N settings there may be N^2 levels. 55141240Snjl */ 56141240Snjl#define CF_MAX_LEVELS 32 57141240Snjl 58141240Snjlstruct cpufreq_softc { 59141240Snjl struct cf_level curr_level; 60141923Snjl int curr_priority; 61141923Snjl struct cf_level saved_level; 62141923Snjl int saved_priority; 63141923Snjl struct cf_level_lst all_levels; 64141413Snjl int all_count; 65141945Snjl int max_mhz; 66141240Snjl device_t dev; 67141240Snjl struct sysctl_ctx_list sysctl_ctx; 68141240Snjl}; 69141240Snjl 70141240Snjlstruct cf_setting_array { 71141240Snjl struct cf_setting sets[MAX_SETTINGS]; 72141240Snjl int count; 73141240Snjl TAILQ_ENTRY(cf_setting_array) link; 74141240Snjl}; 75141240Snjl 76141240SnjlTAILQ_HEAD(cf_setting_lst, cf_setting_array); 77141240Snjl 78141240Snjlstatic int cpufreq_attach(device_t dev); 79141240Snjlstatic int cpufreq_detach(device_t dev); 80141240Snjlstatic void cpufreq_evaluate(void *arg); 81141240Snjlstatic int cf_set_method(device_t dev, const struct cf_level *level, 82141240Snjl int priority); 83141240Snjlstatic int cf_get_method(device_t dev, struct cf_level *level); 84141240Snjlstatic int cf_levels_method(device_t dev, struct cf_level *levels, 85141240Snjl int *count); 86141413Snjlstatic int cpufreq_insert_abs(struct cpufreq_softc *sc, 87141240Snjl struct cf_setting *sets, int count); 88141413Snjlstatic int cpufreq_expand_set(struct cpufreq_softc *sc, 89141413Snjl struct cf_setting_array *set_arr); 90141413Snjlstatic struct cf_level *cpufreq_dup_set(struct cpufreq_softc *sc, 91141413Snjl struct cf_level *dup, struct cf_setting *set); 92141240Snjlstatic int cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS); 93141240Snjlstatic int cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS); 94141240Snjl 95141240Snjlstatic device_method_t cpufreq_methods[] = { 96141240Snjl DEVMETHOD(device_probe, bus_generic_probe), 97141240Snjl DEVMETHOD(device_attach, cpufreq_attach), 98141240Snjl DEVMETHOD(device_detach, cpufreq_detach), 99141240Snjl 100141240Snjl DEVMETHOD(cpufreq_set, cf_set_method), 101141240Snjl DEVMETHOD(cpufreq_get, cf_get_method), 102141240Snjl DEVMETHOD(cpufreq_levels, cf_levels_method), 103141240Snjl {0, 0} 104141240Snjl}; 105141240Snjlstatic driver_t cpufreq_driver = { 106141240Snjl "cpufreq", cpufreq_methods, sizeof(struct cpufreq_softc) 107141240Snjl}; 108141240Snjlstatic devclass_t cpufreq_dc; 109141240SnjlDRIVER_MODULE(cpufreq, cpu, cpufreq_driver, cpufreq_dc, 0, 0); 110141240Snjl 111141240Snjlstatic eventhandler_tag cf_ev_tag; 112141240Snjl 113141240Snjlstatic int 114141240Snjlcpufreq_attach(device_t dev) 115141240Snjl{ 116141240Snjl struct cpufreq_softc *sc; 117141240Snjl device_t parent; 118141240Snjl int numdevs; 119141240Snjl 120141240Snjl sc = device_get_softc(dev); 121141240Snjl parent = device_get_parent(dev); 122141240Snjl sc->dev = dev; 123141240Snjl sysctl_ctx_init(&sc->sysctl_ctx); 124141240Snjl TAILQ_INIT(&sc->all_levels); 125141240Snjl sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 126141923Snjl sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 127141945Snjl sc->max_mhz = CPUFREQ_VAL_UNKNOWN; 128141240Snjl 129141240Snjl /* 130141240Snjl * Only initialize one set of sysctls for all CPUs. In the future, 131141240Snjl * if multiple CPUs can have different settings, we can move these 132141240Snjl * sysctls to be under every CPU instead of just the first one. 133141240Snjl */ 134141240Snjl numdevs = devclass_get_count(cpufreq_dc); 135141240Snjl if (numdevs > 1) 136141240Snjl return (0); 137141240Snjl 138141240Snjl SYSCTL_ADD_PROC(&sc->sysctl_ctx, 139141240Snjl SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 140141240Snjl OID_AUTO, "freq", CTLTYPE_INT | CTLFLAG_RW, sc, 0, 141141240Snjl cpufreq_curr_sysctl, "I", "Current CPU frequency"); 142141240Snjl SYSCTL_ADD_PROC(&sc->sysctl_ctx, 143141240Snjl SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 144141240Snjl OID_AUTO, "freq_levels", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, 145141240Snjl cpufreq_levels_sysctl, "A", "CPU frequency levels"); 146141240Snjl cf_ev_tag = EVENTHANDLER_REGISTER(cpufreq_changed, cpufreq_evaluate, 147141240Snjl NULL, EVENTHANDLER_PRI_ANY); 148141240Snjl 149141240Snjl return (0); 150141240Snjl} 151141240Snjl 152141240Snjlstatic int 153141240Snjlcpufreq_detach(device_t dev) 154141240Snjl{ 155141240Snjl struct cpufreq_softc *sc; 156141240Snjl int numdevs; 157141240Snjl 158141240Snjl sc = device_get_softc(dev); 159141240Snjl sysctl_ctx_free(&sc->sysctl_ctx); 160141240Snjl 161141240Snjl /* Only clean up these resources when the last device is detaching. */ 162141240Snjl numdevs = devclass_get_count(cpufreq_dc); 163141240Snjl if (numdevs == 1) 164141240Snjl EVENTHANDLER_DEREGISTER(cpufreq_changed, cf_ev_tag); 165141240Snjl 166141240Snjl return (0); 167141240Snjl} 168141240Snjl 169141240Snjlstatic void 170141240Snjlcpufreq_evaluate(void *arg) 171141240Snjl{ 172141240Snjl /* TODO: Re-evaluate when notified of changes to drivers. */ 173141240Snjl} 174141240Snjl 175141240Snjlstatic int 176141240Snjlcf_set_method(device_t dev, const struct cf_level *level, int priority) 177141240Snjl{ 178141240Snjl struct cpufreq_softc *sc; 179141240Snjl const struct cf_setting *set; 180141814Snjl struct pcpu *pc; 181141814Snjl int cpu_id, error, i; 182141240Snjl 183141240Snjl sc = device_get_softc(dev); 184141240Snjl 185141814Snjl /* 186141814Snjl * Check that the TSC isn't being used as a timecounter. 187141814Snjl * If it is, then return EBUSY and refuse to change the 188141814Snjl * clock speed. 189141814Snjl */ 190141814Snjl if (strcmp(timecounter->tc_name, "TSC") == 0) 191141814Snjl return (EBUSY); 192141814Snjl 193141923Snjl /* 194141923Snjl * If the caller didn't specify a level and one is saved, prepare to 195141923Snjl * restore the saved level. If none has been saved, return an error. 196141923Snjl * If they did specify one, but the requested level has a lower 197141923Snjl * priority, don't allow the new level right now. 198141923Snjl */ 199141923Snjl if (level == NULL) { 200141923Snjl if (sc->saved_level.total_set.freq != CPUFREQ_VAL_UNKNOWN) { 201141923Snjl level = &sc->saved_level; 202141923Snjl priority = sc->saved_priority; 203141923Snjl } else 204141923Snjl return (ENXIO); 205141923Snjl } else if (priority < sc->curr_priority) 206141923Snjl return (EPERM); 207141923Snjl 208141240Snjl /* If already at this level, just return. */ 209141240Snjl if (CPUFREQ_CMP(sc->curr_level.total_set.freq, level->total_set.freq)) 210141240Snjl return (0); 211141240Snjl 212141240Snjl /* First, set the absolute frequency via its driver. */ 213141240Snjl set = &level->abs_set; 214141240Snjl if (set->dev) { 215141240Snjl if (!device_is_attached(set->dev)) { 216141240Snjl error = ENXIO; 217141240Snjl goto out; 218141240Snjl } 219141943Snjl 220141943Snjl /* Bind to the target CPU before switching, if necessary. */ 221141943Snjl cpu_id = PCPU_GET(cpuid); 222141943Snjl pc = cpu_get_pcpu(set->dev); 223141943Snjl if (cpu_id != pc->pc_cpuid) { 224141943Snjl mtx_lock_spin(&sched_lock); 225141943Snjl sched_bind(curthread, pc->pc_cpuid); 226141943Snjl mtx_unlock_spin(&sched_lock); 227141943Snjl } 228141240Snjl error = CPUFREQ_DRV_SET(set->dev, set); 229141943Snjl if (cpu_id != pc->pc_cpuid) { 230141943Snjl mtx_lock_spin(&sched_lock); 231141943Snjl sched_unbind(curthread); 232141943Snjl mtx_unlock_spin(&sched_lock); 233141943Snjl } 234141240Snjl if (error) { 235141240Snjl goto out; 236141240Snjl } 237141240Snjl } 238141240Snjl 239141413Snjl /* Next, set any/all relative frequencies via their drivers. */ 240141413Snjl for (i = 0; i < level->rel_count; i++) { 241141413Snjl set = &level->rel_set[i]; 242141413Snjl if (!device_is_attached(set->dev)) { 243141413Snjl error = ENXIO; 244141413Snjl goto out; 245141413Snjl } 246141943Snjl 247141943Snjl /* Bind to the target CPU before switching, if necessary. */ 248141943Snjl cpu_id = PCPU_GET(cpuid); 249141943Snjl pc = cpu_get_pcpu(set->dev); 250141943Snjl if (cpu_id != pc->pc_cpuid) { 251141943Snjl mtx_lock_spin(&sched_lock); 252141943Snjl sched_bind(curthread, pc->pc_cpuid); 253141943Snjl mtx_unlock_spin(&sched_lock); 254141943Snjl } 255141413Snjl error = CPUFREQ_DRV_SET(set->dev, set); 256141943Snjl if (cpu_id != pc->pc_cpuid) { 257141943Snjl mtx_lock_spin(&sched_lock); 258141943Snjl sched_unbind(curthread); 259141943Snjl mtx_unlock_spin(&sched_lock); 260141943Snjl } 261141413Snjl if (error) { 262141413Snjl /* XXX Back out any successful setting? */ 263141413Snjl goto out; 264141413Snjl } 265141413Snjl } 266141240Snjl 267141923Snjl /* If we were restoring a saved state, reset it to "unused". */ 268141923Snjl if (level == &sc->saved_level) { 269141923Snjl sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 270141923Snjl sc->saved_priority = 0; 271141923Snjl } 272141923Snjl 273141923Snjl /* 274141923Snjl * Before recording the current level, check if we're going to a 275141923Snjl * higher priority and have not saved a level yet. If so, save the 276141923Snjl * previous level and priority. 277141923Snjl */ 278141923Snjl if (sc->curr_level.total_set.freq != CPUFREQ_VAL_UNKNOWN && 279141923Snjl sc->saved_level.total_set.freq == CPUFREQ_VAL_UNKNOWN && 280141923Snjl priority > sc->curr_priority) { 281141923Snjl sc->saved_level = sc->curr_level; 282141923Snjl sc->saved_priority = sc->curr_priority; 283141923Snjl } 284141240Snjl sc->curr_level = *level; 285141923Snjl sc->curr_priority = priority; 286141240Snjl error = 0; 287141240Snjl 288141240Snjlout: 289141240Snjl if (error) 290141240Snjl device_printf(set->dev, "set freq failed, err %d\n", error); 291141240Snjl return (error); 292141240Snjl} 293141240Snjl 294141240Snjlstatic int 295141240Snjlcf_get_method(device_t dev, struct cf_level *level) 296141240Snjl{ 297141240Snjl struct cpufreq_softc *sc; 298141240Snjl struct cf_level *levels; 299141240Snjl struct cf_setting *curr_set, set; 300141240Snjl struct pcpu *pc; 301141240Snjl device_t *devs; 302141240Snjl int count, error, i, numdevs; 303141240Snjl uint64_t rate; 304141240Snjl 305141240Snjl sc = device_get_softc(dev); 306141240Snjl curr_set = &sc->curr_level.total_set; 307141240Snjl levels = NULL; 308141240Snjl 309141240Snjl /* If we already know the current frequency, we're done. */ 310141240Snjl if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 311141240Snjl goto out; 312141240Snjl 313141240Snjl /* 314141240Snjl * We need to figure out the current level. Loop through every 315141240Snjl * driver, getting the current setting. Then, attempt to get a best 316141240Snjl * match of settings against each level. 317141240Snjl */ 318141240Snjl count = CF_MAX_LEVELS; 319141240Snjl levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 320141240Snjl if (levels == NULL) 321141240Snjl return (ENOMEM); 322141240Snjl error = CPUFREQ_LEVELS(sc->dev, levels, &count); 323141240Snjl if (error) 324141240Snjl goto out; 325141240Snjl error = device_get_children(device_get_parent(dev), &devs, &numdevs); 326141240Snjl if (error) 327141240Snjl goto out; 328141240Snjl for (i = 0; i < numdevs && curr_set->freq == CPUFREQ_VAL_UNKNOWN; i++) { 329141240Snjl if (!device_is_attached(devs[i])) 330141240Snjl continue; 331141240Snjl error = CPUFREQ_DRV_GET(devs[i], &set); 332141240Snjl if (error) 333141240Snjl continue; 334141240Snjl for (i = 0; i < count; i++) { 335141413Snjl if (CPUFREQ_CMP(set.freq, levels[i].total_set.freq)) { 336141240Snjl sc->curr_level = levels[i]; 337141240Snjl break; 338141240Snjl } 339141240Snjl } 340141240Snjl } 341141240Snjl free(devs, M_TEMP); 342141240Snjl if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 343141240Snjl goto out; 344141240Snjl 345141240Snjl /* 346141240Snjl * We couldn't find an exact match, so attempt to estimate and then 347141240Snjl * match against a level. 348141240Snjl */ 349141240Snjl pc = cpu_get_pcpu(dev); 350141240Snjl if (pc == NULL) { 351141240Snjl error = ENXIO; 352141240Snjl goto out; 353141240Snjl } 354141240Snjl cpu_est_clockrate(pc->pc_cpuid, &rate); 355141240Snjl rate /= 1000000; 356141240Snjl for (i = 0; i < count; i++) { 357141240Snjl if (CPUFREQ_CMP(rate, levels[i].total_set.freq)) { 358141240Snjl sc->curr_level = levels[i]; 359141240Snjl break; 360141240Snjl } 361141240Snjl } 362141240Snjl 363141240Snjlout: 364141240Snjl if (levels) 365141240Snjl free(levels, M_TEMP); 366141240Snjl *level = sc->curr_level; 367141240Snjl return (0); 368141240Snjl} 369141240Snjl 370141240Snjlstatic int 371141240Snjlcf_levels_method(device_t dev, struct cf_level *levels, int *count) 372141240Snjl{ 373141413Snjl struct cf_setting_array *set_arr; 374141240Snjl struct cf_setting_lst rel_sets; 375141240Snjl struct cpufreq_softc *sc; 376141240Snjl struct cf_level *lev; 377141240Snjl struct cf_setting *sets; 378141240Snjl struct pcpu *pc; 379141240Snjl device_t *devs; 380141413Snjl int error, i, numdevs, set_count, type; 381141240Snjl uint64_t rate; 382141240Snjl 383141240Snjl if (levels == NULL || count == NULL) 384141240Snjl return (EINVAL); 385141240Snjl 386141240Snjl TAILQ_INIT(&rel_sets); 387141240Snjl sc = device_get_softc(dev); 388141240Snjl error = device_get_children(device_get_parent(dev), &devs, &numdevs); 389141240Snjl if (error) 390141240Snjl return (error); 391141240Snjl sets = malloc(MAX_SETTINGS * sizeof(*sets), M_TEMP, M_NOWAIT); 392141240Snjl if (sets == NULL) { 393141240Snjl free(devs, M_TEMP); 394141240Snjl return (ENOMEM); 395141240Snjl } 396141240Snjl 397141240Snjl /* Get settings from all cpufreq drivers. */ 398141240Snjl for (i = 0; i < numdevs; i++) { 399141824Snjl /* Skip devices that aren't ready. */ 400141240Snjl if (!device_is_attached(devs[i])) 401141240Snjl continue; 402141824Snjl 403141824Snjl /* 404141824Snjl * Get settings, skipping drivers that offer no settings or 405141824Snjl * provide settings for informational purposes only. 406141824Snjl */ 407141240Snjl set_count = MAX_SETTINGS; 408141240Snjl error = CPUFREQ_DRV_SETTINGS(devs[i], sets, &set_count, &type); 409141824Snjl if (error || set_count == 0 || (type & CPUFREQ_FLAG_INFO_ONLY)) 410141240Snjl continue; 411141413Snjl 412141824Snjl /* Add the settings to our absolute/relative lists. */ 413141814Snjl switch (type & CPUFREQ_TYPE_MASK) { 414141413Snjl case CPUFREQ_TYPE_ABSOLUTE: 415141413Snjl error = cpufreq_insert_abs(sc, sets, set_count); 416141413Snjl break; 417141413Snjl case CPUFREQ_TYPE_RELATIVE: 418141413Snjl set_arr = malloc(sizeof(*set_arr), M_TEMP, M_NOWAIT); 419141413Snjl if (set_arr == NULL) { 420141413Snjl error = ENOMEM; 421141413Snjl goto out; 422141413Snjl } 423141413Snjl bcopy(sets, set_arr->sets, set_count * sizeof(*sets)); 424141413Snjl set_arr->count = set_count; 425141413Snjl TAILQ_INSERT_TAIL(&rel_sets, set_arr, link); 426141413Snjl break; 427141413Snjl default: 428141413Snjl error = EINVAL; 429141413Snjl break; 430141413Snjl } 431141240Snjl if (error) 432141240Snjl goto out; 433141240Snjl } 434141240Snjl 435141945Snjl /* 436141945Snjl * If there are no absolute levels, create a fake one at 100%. We 437141945Snjl * then cache the clockrate for later use as our base frequency. 438141945Snjl * 439141945Snjl * XXX This assumes that the first time through, if we only have 440141945Snjl * relative drivers, the CPU is currently running at 100%. 441141945Snjl */ 442141240Snjl if (TAILQ_EMPTY(&sc->all_levels)) { 443141945Snjl if (sc->max_mhz == CPUFREQ_VAL_UNKNOWN) { 444141945Snjl pc = cpu_get_pcpu(dev); 445141945Snjl cpu_est_clockrate(pc->pc_cpuid, &rate); 446141945Snjl sc->max_mhz = rate / 1000000; 447141240Snjl } 448141945Snjl memset(&sets[0], CPUFREQ_VAL_UNKNOWN, sizeof(*sets)); 449141945Snjl sets[0].freq = sc->max_mhz; 450141945Snjl sets[0].dev = NULL; 451141413Snjl error = cpufreq_insert_abs(sc, sets, 1); 452141240Snjl if (error) 453141240Snjl goto out; 454141240Snjl } 455141240Snjl 456141413Snjl /* Create a combined list of absolute + relative levels. */ 457141413Snjl TAILQ_FOREACH(set_arr, &rel_sets, link) 458141413Snjl cpufreq_expand_set(sc, set_arr); 459141413Snjl 460141413Snjl /* If the caller doesn't have enough space, return the actual count. */ 461141413Snjl if (sc->all_count > *count) { 462141413Snjl *count = sc->all_count; 463141413Snjl error = E2BIG; 464141413Snjl goto out; 465141413Snjl } 466141413Snjl 467141413Snjl /* Finally, output the list of levels. */ 468141240Snjl i = 0; 469141240Snjl TAILQ_FOREACH(lev, &sc->all_levels, link) { 470141240Snjl levels[i] = *lev; 471141240Snjl i++; 472141240Snjl } 473141413Snjl *count = sc->all_count; 474141240Snjl error = 0; 475141240Snjl 476141240Snjlout: 477141240Snjl /* Clear all levels since we regenerate them each time. */ 478141240Snjl while ((lev = TAILQ_FIRST(&sc->all_levels)) != NULL) { 479141240Snjl TAILQ_REMOVE(&sc->all_levels, lev, link); 480141240Snjl free(lev, M_TEMP); 481141240Snjl } 482141413Snjl while ((set_arr = TAILQ_FIRST(&rel_sets)) != NULL) { 483141413Snjl TAILQ_REMOVE(&rel_sets, set_arr, link); 484141413Snjl free(set_arr, M_TEMP); 485141413Snjl } 486141413Snjl sc->all_count = 0; 487141240Snjl free(devs, M_TEMP); 488141240Snjl free(sets, M_TEMP); 489141240Snjl return (error); 490141240Snjl} 491141240Snjl 492141240Snjl/* 493141240Snjl * Create levels for an array of absolute settings and insert them in 494141240Snjl * sorted order in the specified list. 495141240Snjl */ 496141240Snjlstatic int 497141413Snjlcpufreq_insert_abs(struct cpufreq_softc *sc, struct cf_setting *sets, 498141240Snjl int count) 499141240Snjl{ 500141413Snjl struct cf_level_lst *list; 501141240Snjl struct cf_level *level, *search; 502141240Snjl int i; 503141240Snjl 504141413Snjl list = &sc->all_levels; 505141240Snjl for (i = 0; i < count; i++) { 506141240Snjl level = malloc(sizeof(*level), M_TEMP, M_NOWAIT | M_ZERO); 507141240Snjl if (level == NULL) 508141240Snjl return (ENOMEM); 509141240Snjl level->abs_set = sets[i]; 510141413Snjl level->total_set = sets[i]; 511141413Snjl level->total_set.dev = NULL; 512141413Snjl sc->all_count++; 513141240Snjl 514141240Snjl if (TAILQ_EMPTY(list)) { 515141240Snjl TAILQ_INSERT_HEAD(list, level, link); 516141240Snjl continue; 517141240Snjl } 518141240Snjl 519141240Snjl TAILQ_FOREACH_REVERSE(search, list, cf_level_lst, link) { 520141413Snjl if (sets[i].freq <= search->total_set.freq) { 521141240Snjl TAILQ_INSERT_AFTER(list, search, level, link); 522141240Snjl break; 523141240Snjl } 524141240Snjl } 525141240Snjl } 526141240Snjl return (0); 527141240Snjl} 528141240Snjl 529141413Snjl/* 530141413Snjl * Expand a group of relative settings, creating derived levels from them. 531141413Snjl */ 532141240Snjlstatic int 533141413Snjlcpufreq_expand_set(struct cpufreq_softc *sc, struct cf_setting_array *set_arr) 534141413Snjl{ 535141413Snjl struct cf_level *fill, *search; 536141413Snjl struct cf_setting *set; 537141413Snjl int i; 538141413Snjl 539141413Snjl TAILQ_FOREACH(search, &sc->all_levels, link) { 540141413Snjl /* Skip this level if we've already modified it. */ 541141413Snjl for (i = 0; i < search->rel_count; i++) { 542141413Snjl if (search->rel_set[i].dev == set_arr->sets[0].dev) 543141413Snjl break; 544141413Snjl } 545141413Snjl if (i != search->rel_count) 546141413Snjl continue; 547141413Snjl 548141413Snjl /* Add each setting to the level, duplicating if necessary. */ 549141413Snjl for (i = 0; i < set_arr->count; i++) { 550141413Snjl set = &set_arr->sets[i]; 551141413Snjl 552141413Snjl /* 553141413Snjl * If this setting is less than 100%, split the level 554141413Snjl * into two and add this setting to the new level. 555141413Snjl */ 556141413Snjl fill = search; 557141413Snjl if (set->freq < 10000) 558141413Snjl fill = cpufreq_dup_set(sc, search, set); 559141413Snjl 560141413Snjl /* 561141413Snjl * The new level was a duplicate of an existing level 562141413Snjl * so we freed it. Go to the next setting. 563141413Snjl */ 564141413Snjl if (fill == NULL) 565141413Snjl continue; 566141413Snjl 567141413Snjl /* Add this setting to the existing or new level. */ 568141413Snjl KASSERT(fill->rel_count < MAX_SETTINGS, 569141413Snjl ("cpufreq: too many relative drivers (%d)", 570141413Snjl MAX_SETTINGS)); 571141413Snjl fill->rel_set[fill->rel_count] = *set; 572141413Snjl fill->rel_count++; 573141413Snjl } 574141413Snjl } 575141413Snjl 576141413Snjl return (0); 577141413Snjl} 578141413Snjl 579141413Snjlstatic struct cf_level * 580141413Snjlcpufreq_dup_set(struct cpufreq_softc *sc, struct cf_level *dup, 581141413Snjl struct cf_setting *set) 582141413Snjl{ 583141413Snjl struct cf_level_lst *list; 584141413Snjl struct cf_level *fill, *itr; 585141413Snjl struct cf_setting *fill_set, *itr_set; 586141413Snjl int i; 587141413Snjl 588141413Snjl /* 589141413Snjl * Create a new level, copy it from the old one, and update the 590141413Snjl * total frequency and power by the percentage specified in the 591141413Snjl * relative setting. 592141413Snjl */ 593141413Snjl fill = malloc(sizeof(*fill), M_TEMP, M_NOWAIT); 594141413Snjl if (fill == NULL) 595141413Snjl return (NULL); 596141413Snjl *fill = *dup; 597141413Snjl fill_set = &fill->total_set; 598141413Snjl fill_set->freq = 599141413Snjl ((uint64_t)fill_set->freq * set->freq) / 10000; 600141413Snjl if (fill_set->power != CPUFREQ_VAL_UNKNOWN) { 601141413Snjl fill_set->power = ((uint64_t)fill_set->power * set->freq) 602141413Snjl / 10000; 603141413Snjl } 604141413Snjl if (set->lat != CPUFREQ_VAL_UNKNOWN) { 605141413Snjl if (fill_set->lat != CPUFREQ_VAL_UNKNOWN) 606141413Snjl fill_set->lat += set->lat; 607141413Snjl else 608141413Snjl fill_set->lat = set->lat; 609141413Snjl } 610141413Snjl 611141413Snjl /* 612141413Snjl * If we copied an old level that we already modified (say, at 100%), 613141413Snjl * we need to remove that setting before adding this one. Since we 614141413Snjl * process each setting array in order, we know any settings for this 615141413Snjl * driver will be found at the end. 616141413Snjl */ 617141413Snjl for (i = fill->rel_count; i != 0; i--) { 618141413Snjl if (fill->rel_set[i - 1].dev != set->dev) 619141413Snjl break; 620141413Snjl fill->rel_count--; 621141413Snjl } 622141413Snjl 623141413Snjl /* 624141413Snjl * Insert the new level in sorted order. If we find a duplicate, 625141413Snjl * free the new level. We can do this since any existing level will 626141413Snjl * be guaranteed to have the same or less settings and thus consume 627141413Snjl * less power. For example, a level with one absolute setting of 628141413Snjl * 800 Mhz uses less power than one composed of an absolute setting 629141413Snjl * of 1600 Mhz and a relative setting at 50%. 630141413Snjl */ 631141413Snjl list = &sc->all_levels; 632141413Snjl if (TAILQ_EMPTY(list)) { 633141413Snjl TAILQ_INSERT_HEAD(list, fill, link); 634141413Snjl } else { 635141413Snjl TAILQ_FOREACH_REVERSE(itr, list, cf_level_lst, link) { 636141413Snjl itr_set = &itr->total_set; 637141413Snjl if (CPUFREQ_CMP(fill_set->freq, itr_set->freq)) { 638141413Snjl free(fill, M_TEMP); 639141413Snjl fill = NULL; 640141413Snjl break; 641141413Snjl } else if (fill_set->freq < itr_set->freq) { 642141413Snjl TAILQ_INSERT_AFTER(list, itr, fill, link); 643141413Snjl sc->all_count++; 644141413Snjl break; 645141413Snjl } 646141413Snjl } 647141413Snjl } 648141413Snjl 649141413Snjl return (fill); 650141413Snjl} 651141413Snjl 652141413Snjlstatic int 653141240Snjlcpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS) 654141240Snjl{ 655141240Snjl struct cpufreq_softc *sc; 656141240Snjl struct cf_level *levels; 657141814Snjl int count, devcount, error, freq, i, n; 658141814Snjl device_t *devs; 659141240Snjl 660141814Snjl devs = NULL; 661141240Snjl sc = oidp->oid_arg1; 662141814Snjl levels = malloc(CF_MAX_LEVELS * sizeof(*levels), M_TEMP, M_NOWAIT); 663141240Snjl if (levels == NULL) 664141240Snjl return (ENOMEM); 665141240Snjl 666141240Snjl error = CPUFREQ_GET(sc->dev, &levels[0]); 667141240Snjl if (error) 668141240Snjl goto out; 669141240Snjl freq = levels[0].total_set.freq; 670141240Snjl error = sysctl_handle_int(oidp, &freq, 0, req); 671141240Snjl if (error != 0 || req->newptr == NULL) 672141240Snjl goto out; 673141240Snjl 674141814Snjl /* 675141814Snjl * While we only call cpufreq_get() on one device (assuming all 676141814Snjl * CPUs have equal levels), we call cpufreq_set() on all CPUs. 677141814Snjl * This is needed for some MP systems. 678141814Snjl */ 679141814Snjl error = devclass_get_devices(cpufreq_dc, &devs, &devcount); 680141240Snjl if (error) 681141240Snjl goto out; 682141814Snjl for (n = 0; n < devcount; n++) { 683141814Snjl count = CF_MAX_LEVELS; 684141814Snjl error = CPUFREQ_LEVELS(devs[n], levels, &count); 685141814Snjl if (error) 686141240Snjl break; 687141814Snjl for (i = 0; i < count; i++) { 688141814Snjl if (CPUFREQ_CMP(levels[i].total_set.freq, freq)) { 689141814Snjl error = CPUFREQ_SET(devs[n], &levels[i], 690141814Snjl CPUFREQ_PRIO_USER); 691141814Snjl break; 692141814Snjl } 693141240Snjl } 694141814Snjl if (i == count) { 695141814Snjl error = EINVAL; 696141814Snjl break; 697141814Snjl } 698141240Snjl } 699141240Snjl 700141240Snjlout: 701141814Snjl if (devs) 702141814Snjl free(devs, M_TEMP); 703141240Snjl if (levels) 704141240Snjl free(levels, M_TEMP); 705141240Snjl return (error); 706141240Snjl} 707141240Snjl 708141240Snjlstatic int 709141240Snjlcpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS) 710141240Snjl{ 711141240Snjl struct cpufreq_softc *sc; 712141240Snjl struct cf_level *levels; 713141240Snjl struct cf_setting *set; 714141240Snjl struct sbuf sb; 715141240Snjl int count, error, i; 716141240Snjl 717141240Snjl sc = oidp->oid_arg1; 718141240Snjl sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND); 719141240Snjl 720141240Snjl /* Get settings from the device and generate the output string. */ 721141240Snjl count = CF_MAX_LEVELS; 722141240Snjl levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 723141240Snjl if (levels == NULL) 724141240Snjl return (ENOMEM); 725141240Snjl error = CPUFREQ_LEVELS(sc->dev, levels, &count); 726141240Snjl if (error) 727141240Snjl goto out; 728141240Snjl if (count) { 729141240Snjl for (i = 0; i < count; i++) { 730141240Snjl set = &levels[i].total_set; 731141240Snjl sbuf_printf(&sb, "%d/%d ", set->freq, set->power); 732141240Snjl } 733141240Snjl } else 734141240Snjl sbuf_cpy(&sb, "0"); 735141240Snjl sbuf_trim(&sb); 736141240Snjl sbuf_finish(&sb); 737141240Snjl error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req); 738141240Snjl 739141240Snjlout: 740141240Snjl free(levels, M_TEMP); 741141240Snjl sbuf_delete(&sb); 742141240Snjl return (error); 743141240Snjl} 744141240Snjl 745141240Snjlint 746141240Snjlcpufreq_register(device_t dev) 747141240Snjl{ 748141945Snjl struct cpufreq_softc *sc; 749141240Snjl device_t cf_dev, cpu_dev; 750141240Snjl 751141240Snjl /* 752141814Snjl * Add only one cpufreq device to each CPU. Currently, all CPUs 753141814Snjl * must offer the same levels and be switched at the same time. 754141240Snjl */ 755141814Snjl cpu_dev = device_get_parent(dev); 756141945Snjl if ((cf_dev = device_find_child(cpu_dev, "cpufreq", -1))) { 757141945Snjl sc = device_get_softc(cf_dev); 758141945Snjl sc->max_mhz = CPUFREQ_VAL_UNKNOWN; 759141240Snjl return (0); 760141945Snjl } 761141240Snjl 762141814Snjl /* Add the child device and possibly sysctls. */ 763141814Snjl cf_dev = BUS_ADD_CHILD(cpu_dev, 0, "cpufreq", -1); 764141240Snjl if (cf_dev == NULL) 765141240Snjl return (ENOMEM); 766141240Snjl device_quiet(cf_dev); 767141240Snjl 768141240Snjl return (device_probe_and_attach(cf_dev)); 769141240Snjl} 770141240Snjl 771141240Snjlint 772141240Snjlcpufreq_unregister(device_t dev) 773141240Snjl{ 774141240Snjl device_t cf_dev, *devs; 775141240Snjl int cfcount, count, devcount, error, i, type; 776141240Snjl struct cf_setting set; 777141240Snjl 778141240Snjl /* 779141240Snjl * If this is the last cpufreq child device, remove the control 780141240Snjl * device as well. We identify cpufreq children by calling a method 781141240Snjl * they support. 782141240Snjl */ 783141240Snjl error = device_get_children(device_get_parent(dev), &devs, &devcount); 784141240Snjl if (error) 785141240Snjl return (error); 786141945Snjl cf_dev = device_find_child(device_get_parent(dev), "cpufreq", -1); 787141240Snjl cfcount = 0; 788141240Snjl for (i = 0; i < devcount; i++) { 789141240Snjl if (!device_is_attached(devs[i])) 790141240Snjl continue; 791141240Snjl count = 1; 792141240Snjl if (CPUFREQ_DRV_SETTINGS(devs[i], &set, &count, &type) == 0) 793141240Snjl cfcount++; 794141240Snjl } 795141814Snjl if (cfcount <= 1) 796141240Snjl device_delete_child(device_get_parent(cf_dev), cf_dev); 797141240Snjl free(devs, M_TEMP); 798141240Snjl 799141240Snjl return (0); 800141240Snjl} 801