kern_cpu.c revision 141240
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 141240 2005-02-04 05:39:19Z 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 44#include "cpufreq_if.h" 45 46/* 47 * Common CPU frequency glue code. Drivers for specific hardware can 48 * attach this interface to allow users to get/set the CPU frequency. 49 */ 50 51/* 52 * Number of levels we can handle. Levels are synthesized from settings 53 * so for N settings there may be N^2 levels. 54 */ 55#define CF_MAX_LEVELS 32 56 57struct cpufreq_softc { 58 struct cf_level curr_level; 59 int priority; 60 struct cf_level_lst all_levels; 61 device_t dev; 62 struct sysctl_ctx_list sysctl_ctx; 63}; 64 65struct cf_setting_array { 66 struct cf_setting sets[MAX_SETTINGS]; 67 int count; 68 TAILQ_ENTRY(cf_setting_array) link; 69}; 70 71TAILQ_HEAD(cf_setting_lst, cf_setting_array); 72 73static int cpufreq_attach(device_t dev); 74static int cpufreq_detach(device_t dev); 75static void cpufreq_evaluate(void *arg); 76static int cf_set_method(device_t dev, const struct cf_level *level, 77 int priority); 78static int cf_get_method(device_t dev, struct cf_level *level); 79static int cf_levels_method(device_t dev, struct cf_level *levels, 80 int *count); 81static int cpufreq_insert_abs(struct cf_level_lst *list, 82 struct cf_setting *sets, int count); 83static int cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS); 84static int cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS); 85 86static device_method_t cpufreq_methods[] = { 87 DEVMETHOD(device_probe, bus_generic_probe), 88 DEVMETHOD(device_attach, cpufreq_attach), 89 DEVMETHOD(device_detach, cpufreq_detach), 90 91 DEVMETHOD(cpufreq_set, cf_set_method), 92 DEVMETHOD(cpufreq_get, cf_get_method), 93 DEVMETHOD(cpufreq_levels, cf_levels_method), 94 {0, 0} 95}; 96static driver_t cpufreq_driver = { 97 "cpufreq", cpufreq_methods, sizeof(struct cpufreq_softc) 98}; 99static devclass_t cpufreq_dc; 100DRIVER_MODULE(cpufreq, cpu, cpufreq_driver, cpufreq_dc, 0, 0); 101 102static eventhandler_tag cf_ev_tag; 103 104static int 105cpufreq_attach(device_t dev) 106{ 107 struct cpufreq_softc *sc; 108 device_t parent; 109 int numdevs; 110 111 sc = device_get_softc(dev); 112 parent = device_get_parent(dev); 113 sc->dev = dev; 114 sysctl_ctx_init(&sc->sysctl_ctx); 115 TAILQ_INIT(&sc->all_levels); 116 sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN; 117 118 /* 119 * Only initialize one set of sysctls for all CPUs. In the future, 120 * if multiple CPUs can have different settings, we can move these 121 * sysctls to be under every CPU instead of just the first one. 122 */ 123 numdevs = devclass_get_count(cpufreq_dc); 124 if (numdevs > 1) 125 return (0); 126 127 SYSCTL_ADD_PROC(&sc->sysctl_ctx, 128 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 129 OID_AUTO, "freq", CTLTYPE_INT | CTLFLAG_RW, sc, 0, 130 cpufreq_curr_sysctl, "I", "Current CPU frequency"); 131 SYSCTL_ADD_PROC(&sc->sysctl_ctx, 132 SYSCTL_CHILDREN(device_get_sysctl_tree(parent)), 133 OID_AUTO, "freq_levels", CTLTYPE_STRING | CTLFLAG_RD, sc, 0, 134 cpufreq_levels_sysctl, "A", "CPU frequency levels"); 135 cf_ev_tag = EVENTHANDLER_REGISTER(cpufreq_changed, cpufreq_evaluate, 136 NULL, EVENTHANDLER_PRI_ANY); 137 138 return (0); 139} 140 141static int 142cpufreq_detach(device_t dev) 143{ 144 struct cpufreq_softc *sc; 145 int numdevs; 146 147 sc = device_get_softc(dev); 148 sysctl_ctx_free(&sc->sysctl_ctx); 149 150 /* Only clean up these resources when the last device is detaching. */ 151 numdevs = devclass_get_count(cpufreq_dc); 152 if (numdevs == 1) 153 EVENTHANDLER_DEREGISTER(cpufreq_changed, cf_ev_tag); 154 155 return (0); 156} 157 158static void 159cpufreq_evaluate(void *arg) 160{ 161 /* TODO: Re-evaluate when notified of changes to drivers. */ 162} 163 164static int 165cf_set_method(device_t dev, const struct cf_level *level, int priority) 166{ 167 struct cpufreq_softc *sc; 168 const struct cf_setting *set; 169 int error; 170 171 sc = device_get_softc(dev); 172 173 /* If already at this level, just return. */ 174 if (CPUFREQ_CMP(sc->curr_level.total_set.freq, level->total_set.freq)) 175 return (0); 176 177 /* First, set the absolute frequency via its driver. */ 178 set = &level->abs_set; 179 if (set->dev) { 180 if (!device_is_attached(set->dev)) { 181 error = ENXIO; 182 goto out; 183 } 184 error = CPUFREQ_DRV_SET(set->dev, set); 185 if (error) { 186 goto out; 187 } 188 } 189 190 /* TODO: Next, set any/all relative frequencies via their drivers. */ 191 192 /* Record the current level. */ 193 sc->curr_level = *level; 194 sc->priority = priority; 195 error = 0; 196 197out: 198 if (error) 199 device_printf(set->dev, "set freq failed, err %d\n", error); 200 return (error); 201} 202 203static int 204cf_get_method(device_t dev, struct cf_level *level) 205{ 206 struct cpufreq_softc *sc; 207 struct cf_level *levels; 208 struct cf_setting *curr_set, set; 209 struct pcpu *pc; 210 device_t *devs; 211 int count, error, i, numdevs; 212 uint64_t rate; 213 214 sc = device_get_softc(dev); 215 curr_set = &sc->curr_level.total_set; 216 levels = NULL; 217 218 /* If we already know the current frequency, we're done. */ 219 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 220 goto out; 221 222 /* 223 * We need to figure out the current level. Loop through every 224 * driver, getting the current setting. Then, attempt to get a best 225 * match of settings against each level. 226 */ 227 count = CF_MAX_LEVELS; 228 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 229 if (levels == NULL) 230 return (ENOMEM); 231 error = CPUFREQ_LEVELS(sc->dev, levels, &count); 232 if (error) 233 goto out; 234 error = device_get_children(device_get_parent(dev), &devs, &numdevs); 235 if (error) 236 goto out; 237 for (i = 0; i < numdevs && curr_set->freq == CPUFREQ_VAL_UNKNOWN; i++) { 238 if (!device_is_attached(devs[i])) 239 continue; 240 error = CPUFREQ_DRV_GET(devs[i], &set); 241 if (error) 242 continue; 243 for (i = 0; i < count; i++) { 244 if (CPUFREQ_CMP(set.freq, levels[i].abs_set.freq)) { 245 sc->curr_level = levels[i]; 246 break; 247 } 248 } 249 } 250 free(devs, M_TEMP); 251 if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) 252 goto out; 253 254 /* 255 * We couldn't find an exact match, so attempt to estimate and then 256 * match against a level. 257 */ 258 pc = cpu_get_pcpu(dev); 259 if (pc == NULL) { 260 error = ENXIO; 261 goto out; 262 } 263 cpu_est_clockrate(pc->pc_cpuid, &rate); 264 rate /= 1000000; 265 for (i = 0; i < count; i++) { 266 if (CPUFREQ_CMP(rate, levels[i].total_set.freq)) { 267 sc->curr_level = levels[i]; 268 break; 269 } 270 } 271 272out: 273 if (levels) 274 free(levels, M_TEMP); 275 *level = sc->curr_level; 276 return (0); 277} 278 279static int 280cf_levels_method(device_t dev, struct cf_level *levels, int *count) 281{ 282 struct cf_setting_lst rel_sets; 283 struct cpufreq_softc *sc; 284 struct cf_level *lev; 285 struct cf_setting *sets; 286 struct pcpu *pc; 287 device_t *devs; 288 int error, i, numdevs, numlevels, set_count, type; 289 uint64_t rate; 290 291 if (levels == NULL || count == NULL) 292 return (EINVAL); 293 294 TAILQ_INIT(&rel_sets); 295 sc = device_get_softc(dev); 296 error = device_get_children(device_get_parent(dev), &devs, &numdevs); 297 if (error) 298 return (error); 299 sets = malloc(MAX_SETTINGS * sizeof(*sets), M_TEMP, M_NOWAIT); 300 if (sets == NULL) { 301 free(devs, M_TEMP); 302 return (ENOMEM); 303 } 304 305 /* Get settings from all cpufreq drivers. */ 306 numlevels = 0; 307 for (i = 0; i < numdevs; i++) { 308 if (!device_is_attached(devs[i])) 309 continue; 310 set_count = MAX_SETTINGS; 311 error = CPUFREQ_DRV_SETTINGS(devs[i], sets, &set_count, &type); 312 if (error || set_count == 0) 313 continue; 314 error = cpufreq_insert_abs(&sc->all_levels, sets, set_count); 315 if (error) 316 goto out; 317 numlevels += set_count; 318 } 319 320 /* If the caller doesn't have enough space, return the actual count. */ 321 if (numlevels > *count) { 322 *count = numlevels; 323 error = E2BIG; 324 goto out; 325 } 326 327 /* If there are no absolute levels, create a fake one at 100%. */ 328 if (TAILQ_EMPTY(&sc->all_levels)) { 329 bzero(&sets[0], sizeof(*sets)); 330 pc = cpu_get_pcpu(dev); 331 if (pc == NULL) { 332 error = ENXIO; 333 goto out; 334 } 335 cpu_est_clockrate(pc->pc_cpuid, &rate); 336 sets[0].freq = rate / 1000000; 337 error = cpufreq_insert_abs(&sc->all_levels, sets, 1); 338 if (error) 339 goto out; 340 } 341 342 /* TODO: Create a combined list of absolute + relative levels. */ 343 i = 0; 344 TAILQ_FOREACH(lev, &sc->all_levels, link) { 345 /* For now, just assume total freq equals absolute freq. */ 346 lev->total_set = lev->abs_set; 347 lev->total_set.dev = NULL; 348 levels[i] = *lev; 349 i++; 350 } 351 *count = i; 352 error = 0; 353 354out: 355 /* Clear all levels since we regenerate them each time. */ 356 while ((lev = TAILQ_FIRST(&sc->all_levels)) != NULL) { 357 TAILQ_REMOVE(&sc->all_levels, lev, link); 358 free(lev, M_TEMP); 359 } 360 free(devs, M_TEMP); 361 free(sets, M_TEMP); 362 return (error); 363} 364 365/* 366 * Create levels for an array of absolute settings and insert them in 367 * sorted order in the specified list. 368 */ 369static int 370cpufreq_insert_abs(struct cf_level_lst *list, struct cf_setting *sets, 371 int count) 372{ 373 struct cf_level *level, *search; 374 int i; 375 376 for (i = 0; i < count; i++) { 377 level = malloc(sizeof(*level), M_TEMP, M_NOWAIT | M_ZERO); 378 if (level == NULL) 379 return (ENOMEM); 380 level->abs_set = sets[i]; 381 382 if (TAILQ_EMPTY(list)) { 383 TAILQ_INSERT_HEAD(list, level, link); 384 continue; 385 } 386 387 TAILQ_FOREACH_REVERSE(search, list, cf_level_lst, link) { 388 if (sets[i].freq <= search->abs_set.freq) { 389 TAILQ_INSERT_AFTER(list, search, level, link); 390 break; 391 } 392 } 393 } 394 return (0); 395} 396 397static int 398cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS) 399{ 400 struct cpufreq_softc *sc; 401 struct cf_level *levels; 402 int count, error, freq, i; 403 404 sc = oidp->oid_arg1; 405 count = CF_MAX_LEVELS; 406 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 407 if (levels == NULL) 408 return (ENOMEM); 409 410 error = CPUFREQ_GET(sc->dev, &levels[0]); 411 if (error) 412 goto out; 413 freq = levels[0].total_set.freq; 414 error = sysctl_handle_int(oidp, &freq, 0, req); 415 if (error != 0 || req->newptr == NULL) 416 goto out; 417 418 error = CPUFREQ_LEVELS(sc->dev, levels, &count); 419 if (error) 420 goto out; 421 for (i = 0; i < count; i++) { 422 if (CPUFREQ_CMP(levels[i].total_set.freq, freq)) { 423 error = CPUFREQ_SET(sc->dev, &levels[i], 424 CPUFREQ_PRIO_USER); 425 break; 426 } 427 } 428 if (i == count) 429 error = EINVAL; 430 431out: 432 if (levels) 433 free(levels, M_TEMP); 434 return (error); 435} 436 437static int 438cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS) 439{ 440 struct cpufreq_softc *sc; 441 struct cf_level *levels; 442 struct cf_setting *set; 443 struct sbuf sb; 444 int count, error, i; 445 446 sc = oidp->oid_arg1; 447 sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND); 448 449 /* Get settings from the device and generate the output string. */ 450 count = CF_MAX_LEVELS; 451 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 452 if (levels == NULL) 453 return (ENOMEM); 454 error = CPUFREQ_LEVELS(sc->dev, levels, &count); 455 if (error) 456 goto out; 457 if (count) { 458 for (i = 0; i < count; i++) { 459 set = &levels[i].total_set; 460 sbuf_printf(&sb, "%d/%d ", set->freq, set->power); 461 } 462 } else 463 sbuf_cpy(&sb, "0"); 464 sbuf_trim(&sb); 465 sbuf_finish(&sb); 466 error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req); 467 468out: 469 free(levels, M_TEMP); 470 sbuf_delete(&sb); 471 return (error); 472} 473 474int 475cpufreq_register(device_t dev) 476{ 477 device_t cf_dev, cpu_dev; 478 479 /* 480 * Only add one cpufreq device (on cpu0) for all control. Once 481 * independent multi-cpu control appears, we can assign one cpufreq 482 * device per cpu. 483 */ 484 cf_dev = devclass_get_device(cpufreq_dc, 0); 485 if (cf_dev) { 486 device_printf(dev, 487 "warning: only one cpufreq device at a time supported\n"); 488 return (0); 489 } 490 491 /* Add the child device and sysctls. */ 492 cpu_dev = devclass_get_device(devclass_find("cpu"), 0); 493 cf_dev = BUS_ADD_CHILD(cpu_dev, 0, "cpufreq", 0); 494 if (cf_dev == NULL) 495 return (ENOMEM); 496 device_quiet(cf_dev); 497 498 return (device_probe_and_attach(cf_dev)); 499} 500 501int 502cpufreq_unregister(device_t dev) 503{ 504 device_t cf_dev, *devs; 505 int cfcount, count, devcount, error, i, type; 506 struct cf_setting set; 507 508 /* 509 * If this is the last cpufreq child device, remove the control 510 * device as well. We identify cpufreq children by calling a method 511 * they support. 512 */ 513 error = device_get_children(device_get_parent(dev), &devs, &devcount); 514 if (error) 515 return (error); 516 cf_dev = devclass_get_device(cpufreq_dc, 0); 517 KASSERT(cf_dev != NULL, ("unregister with no cpufreq dev")); 518 cfcount = 0; 519 for (i = 0; i < devcount; i++) { 520 if (!device_is_attached(devs[i])) 521 continue; 522 count = 1; 523 if (CPUFREQ_DRV_SETTINGS(devs[i], &set, &count, &type) == 0) 524 cfcount++; 525 } 526 if (cfcount <= 1) { 527 device_delete_child(device_get_parent(cf_dev), cf_dev); 528 } 529 free(devs, M_TEMP); 530 531 return (0); 532} 533