xref: /freebsd/sys/kern/kern_cpu.c (revision 8d913481)
173347b07SNate Lawson /*-
273347b07SNate Lawson  * Copyright (c) 2004-2005 Nate Lawson (SDG)
373347b07SNate Lawson  * All rights reserved.
473347b07SNate Lawson  *
573347b07SNate Lawson  * Redistribution and use in source and binary forms, with or without
673347b07SNate Lawson  * modification, are permitted provided that the following conditions
773347b07SNate Lawson  * are met:
873347b07SNate Lawson  * 1. Redistributions of source code must retain the above copyright
973347b07SNate Lawson  *    notice, this list of conditions and the following disclaimer.
1073347b07SNate Lawson  * 2. Redistributions in binary form must reproduce the above copyright
1173347b07SNate Lawson  *    notice, this list of conditions and the following disclaimer in the
1273347b07SNate Lawson  *    documentation and/or other materials provided with the distribution.
1373347b07SNate Lawson  *
1473347b07SNate Lawson  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
1573347b07SNate Lawson  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1673347b07SNate Lawson  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
1773347b07SNate Lawson  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
1873347b07SNate Lawson  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
1973347b07SNate Lawson  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
2073347b07SNate Lawson  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
2173347b07SNate Lawson  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
2273347b07SNate Lawson  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
2373347b07SNate Lawson  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
2473347b07SNate Lawson  * SUCH DAMAGE.
2573347b07SNate Lawson  */
2673347b07SNate Lawson 
2773347b07SNate Lawson #include <sys/cdefs.h>
2873347b07SNate Lawson __FBSDID("$FreeBSD$");
2973347b07SNate Lawson 
3073347b07SNate Lawson #include <sys/param.h>
3173347b07SNate Lawson #include <sys/bus.h>
3273347b07SNate Lawson #include <sys/cpu.h>
3373347b07SNate Lawson #include <sys/eventhandler.h>
3473347b07SNate Lawson #include <sys/kernel.h>
35789f03ceSNate Lawson #include <sys/lock.h>
3673347b07SNate Lawson #include <sys/malloc.h>
3773347b07SNate Lawson #include <sys/module.h>
3873347b07SNate Lawson #include <sys/proc.h>
3973347b07SNate Lawson #include <sys/queue.h>
4073347b07SNate Lawson #include <sys/sched.h>
4173347b07SNate Lawson #include <sys/sysctl.h>
4273347b07SNate Lawson #include <sys/systm.h>
4373347b07SNate Lawson #include <sys/sbuf.h>
44789f03ceSNate Lawson #include <sys/sx.h>
450325089dSNate Lawson #include <sys/timetc.h>
4673347b07SNate Lawson 
4773347b07SNate Lawson #include "cpufreq_if.h"
4873347b07SNate Lawson 
4973347b07SNate Lawson /*
5073347b07SNate Lawson  * Common CPU frequency glue code.  Drivers for specific hardware can
5173347b07SNate Lawson  * attach this interface to allow users to get/set the CPU frequency.
5273347b07SNate Lawson  */
5373347b07SNate Lawson 
5473347b07SNate Lawson /*
5573347b07SNate Lawson  * Number of levels we can handle.  Levels are synthesized from settings
56d269386aSNate Lawson  * so for M settings and N drivers, there may be M*N levels.
5773347b07SNate Lawson  */
58d269386aSNate Lawson #define CF_MAX_LEVELS	64
5973347b07SNate Lawson 
6073347b07SNate Lawson struct cpufreq_softc {
61789f03ceSNate Lawson 	struct sx			lock;
6273347b07SNate Lawson 	struct cf_level			curr_level;
635f0afa04SNate Lawson 	int				curr_priority;
645f0afa04SNate Lawson 	struct cf_level			saved_level;
655f0afa04SNate Lawson 	int				saved_priority;
6673347b07SNate Lawson 	struct cf_level_lst		all_levels;
675f0afa04SNate Lawson 	int				all_count;
6867c8649fSNate Lawson 	int				max_mhz;
6973347b07SNate Lawson 	device_t			dev;
7073347b07SNate Lawson 	struct sysctl_ctx_list		sysctl_ctx;
7173347b07SNate Lawson };
7273347b07SNate Lawson 
7373347b07SNate Lawson struct cf_setting_array {
7473347b07SNate Lawson 	struct cf_setting		sets[MAX_SETTINGS];
7573347b07SNate Lawson 	int				count;
7673347b07SNate Lawson 	TAILQ_ENTRY(cf_setting_array)	link;
7773347b07SNate Lawson };
7873347b07SNate Lawson 
7973347b07SNate Lawson TAILQ_HEAD(cf_setting_lst, cf_setting_array);
8073347b07SNate Lawson 
81789f03ceSNate Lawson #define CF_MTX_INIT(x)		sx_init((x), "cpufreq lock")
82789f03ceSNate Lawson #define CF_MTX_LOCK(x)		sx_xlock((x))
83789f03ceSNate Lawson #define CF_MTX_UNLOCK(x)	sx_xunlock((x))
84789f03ceSNate Lawson #define CF_MTX_ASSERT(x)	sx_assert((x), SX_XLOCKED)
85789f03ceSNate Lawson 
868d913481SNate Lawson #define CF_DEBUG(msg...)	do {		\
878d913481SNate Lawson 	if (cf_verbose)				\
888d913481SNate Lawson 		printf("cpufreq: " msg);	\
898d913481SNate Lawson 	} while (0)
908d913481SNate Lawson 
9173347b07SNate Lawson static int	cpufreq_attach(device_t dev);
9273347b07SNate Lawson static int	cpufreq_detach(device_t dev);
9373347b07SNate Lawson static void	cpufreq_evaluate(void *arg);
9473347b07SNate Lawson static int	cf_set_method(device_t dev, const struct cf_level *level,
9573347b07SNate Lawson 		    int priority);
9673347b07SNate Lawson static int	cf_get_method(device_t dev, struct cf_level *level);
9773347b07SNate Lawson static int	cf_levels_method(device_t dev, struct cf_level *levels,
9873347b07SNate Lawson 		    int *count);
9988c9b54cSNate Lawson static int	cpufreq_insert_abs(struct cpufreq_softc *sc,
10073347b07SNate Lawson 		    struct cf_setting *sets, int count);
10188c9b54cSNate Lawson static int	cpufreq_expand_set(struct cpufreq_softc *sc,
10288c9b54cSNate Lawson 		    struct cf_setting_array *set_arr);
10388c9b54cSNate Lawson static struct cf_level *cpufreq_dup_set(struct cpufreq_softc *sc,
10488c9b54cSNate Lawson 		    struct cf_level *dup, struct cf_setting *set);
10573347b07SNate Lawson static int	cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS);
10673347b07SNate Lawson static int	cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS);
107e959a70bSNate Lawson static int	cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS);
10873347b07SNate Lawson 
10973347b07SNate Lawson static device_method_t cpufreq_methods[] = {
11073347b07SNate Lawson 	DEVMETHOD(device_probe,		bus_generic_probe),
11173347b07SNate Lawson 	DEVMETHOD(device_attach,	cpufreq_attach),
11273347b07SNate Lawson 	DEVMETHOD(device_detach,	cpufreq_detach),
11373347b07SNate Lawson 
11473347b07SNate Lawson         DEVMETHOD(cpufreq_set,		cf_set_method),
11573347b07SNate Lawson         DEVMETHOD(cpufreq_get,		cf_get_method),
11673347b07SNate Lawson         DEVMETHOD(cpufreq_levels,	cf_levels_method),
11773347b07SNate Lawson 	{0, 0}
11873347b07SNate Lawson };
11973347b07SNate Lawson static driver_t cpufreq_driver = {
12073347b07SNate Lawson 	"cpufreq", cpufreq_methods, sizeof(struct cpufreq_softc)
12173347b07SNate Lawson };
12273347b07SNate Lawson static devclass_t cpufreq_dc;
12373347b07SNate Lawson DRIVER_MODULE(cpufreq, cpu, cpufreq_driver, cpufreq_dc, 0, 0);
12473347b07SNate Lawson 
12573347b07SNate Lawson static eventhandler_tag	cf_ev_tag;
12673347b07SNate Lawson 
127b070969bSNate Lawson static int		cf_lowest_freq;
1288d913481SNate Lawson static int		cf_verbose;
129b070969bSNate Lawson TUNABLE_INT("debug.cpufreq.lowest", &cf_lowest_freq);
1308d913481SNate Lawson TUNABLE_INT("debug.cpufreq.verbose", &cf_verbose);
131b070969bSNate Lawson SYSCTL_NODE(_debug, OID_AUTO, cpufreq, CTLFLAG_RD, NULL, "cpufreq debugging");
132b070969bSNate Lawson SYSCTL_INT(_debug_cpufreq, OID_AUTO, lowest, CTLFLAG_RW, &cf_lowest_freq, 1,
133b070969bSNate Lawson     "Don't provide levels below this frequency.");
1348d913481SNate Lawson SYSCTL_INT(_debug_cpufreq, OID_AUTO, verbose, CTLFLAG_RW, &cf_verbose, 1,
1358d913481SNate Lawson     "Print verbose debugging messages");
136b070969bSNate Lawson 
13773347b07SNate Lawson static int
13873347b07SNate Lawson cpufreq_attach(device_t dev)
13973347b07SNate Lawson {
14073347b07SNate Lawson 	struct cpufreq_softc *sc;
14173347b07SNate Lawson 	device_t parent;
14273347b07SNate Lawson 	int numdevs;
14373347b07SNate Lawson 
1448d913481SNate Lawson 	CF_DEBUG("initializing %s\n", device_get_nameunit(dev));
14573347b07SNate Lawson 	sc = device_get_softc(dev);
14673347b07SNate Lawson 	parent = device_get_parent(dev);
14773347b07SNate Lawson 	sc->dev = dev;
14873347b07SNate Lawson 	sysctl_ctx_init(&sc->sysctl_ctx);
14973347b07SNate Lawson 	TAILQ_INIT(&sc->all_levels);
150789f03ceSNate Lawson 	CF_MTX_INIT(&sc->lock);
15173347b07SNate Lawson 	sc->curr_level.total_set.freq = CPUFREQ_VAL_UNKNOWN;
1525f0afa04SNate Lawson 	sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN;
15367c8649fSNate Lawson 	sc->max_mhz = CPUFREQ_VAL_UNKNOWN;
15473347b07SNate Lawson 
15573347b07SNate Lawson 	/*
15673347b07SNate Lawson 	 * Only initialize one set of sysctls for all CPUs.  In the future,
15773347b07SNate Lawson 	 * if multiple CPUs can have different settings, we can move these
15873347b07SNate Lawson 	 * sysctls to be under every CPU instead of just the first one.
15973347b07SNate Lawson 	 */
16073347b07SNate Lawson 	numdevs = devclass_get_count(cpufreq_dc);
16173347b07SNate Lawson 	if (numdevs > 1)
16273347b07SNate Lawson 		return (0);
16373347b07SNate Lawson 
1648d913481SNate Lawson 	CF_DEBUG("initializing one-time data for %s\n",
1658d913481SNate Lawson 	    device_get_nameunit(dev));
16673347b07SNate Lawson 	SYSCTL_ADD_PROC(&sc->sysctl_ctx,
16773347b07SNate Lawson 	    SYSCTL_CHILDREN(device_get_sysctl_tree(parent)),
16873347b07SNate Lawson 	    OID_AUTO, "freq", CTLTYPE_INT | CTLFLAG_RW, sc, 0,
16973347b07SNate Lawson 	    cpufreq_curr_sysctl, "I", "Current CPU frequency");
17073347b07SNate Lawson 	SYSCTL_ADD_PROC(&sc->sysctl_ctx,
17173347b07SNate Lawson 	    SYSCTL_CHILDREN(device_get_sysctl_tree(parent)),
17273347b07SNate Lawson 	    OID_AUTO, "freq_levels", CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
17373347b07SNate Lawson 	    cpufreq_levels_sysctl, "A", "CPU frequency levels");
17473347b07SNate Lawson 	cf_ev_tag = EVENTHANDLER_REGISTER(cpufreq_changed, cpufreq_evaluate,
17573347b07SNate Lawson 	    NULL, EVENTHANDLER_PRI_ANY);
17673347b07SNate Lawson 
17773347b07SNate Lawson 	return (0);
17873347b07SNate Lawson }
17973347b07SNate Lawson 
18073347b07SNate Lawson static int
18173347b07SNate Lawson cpufreq_detach(device_t dev)
18273347b07SNate Lawson {
18373347b07SNate Lawson 	struct cpufreq_softc *sc;
18473347b07SNate Lawson 	int numdevs;
18573347b07SNate Lawson 
1868d913481SNate Lawson 	CF_DEBUG("shutdown %s\n", device_get_nameunit(dev));
18773347b07SNate Lawson 	sc = device_get_softc(dev);
18873347b07SNate Lawson 	sysctl_ctx_free(&sc->sysctl_ctx);
18973347b07SNate Lawson 
19073347b07SNate Lawson 	/* Only clean up these resources when the last device is detaching. */
19173347b07SNate Lawson 	numdevs = devclass_get_count(cpufreq_dc);
1928d913481SNate Lawson 	if (numdevs == 1) {
1938d913481SNate Lawson 		CF_DEBUG("final shutdown for %s\n", device_get_nameunit(dev));
19473347b07SNate Lawson 		EVENTHANDLER_DEREGISTER(cpufreq_changed, cf_ev_tag);
1958d913481SNate Lawson 	}
19673347b07SNate Lawson 
19773347b07SNate Lawson 	return (0);
19873347b07SNate Lawson }
19973347b07SNate Lawson 
20073347b07SNate Lawson static void
20173347b07SNate Lawson cpufreq_evaluate(void *arg)
20273347b07SNate Lawson {
20373347b07SNate Lawson 	/* TODO: Re-evaluate when notified of changes to drivers. */
20473347b07SNate Lawson }
20573347b07SNate Lawson 
20673347b07SNate Lawson static int
20773347b07SNate Lawson cf_set_method(device_t dev, const struct cf_level *level, int priority)
20873347b07SNate Lawson {
20973347b07SNate Lawson 	struct cpufreq_softc *sc;
21073347b07SNate Lawson 	const struct cf_setting *set;
2110325089dSNate Lawson 	struct pcpu *pc;
2120325089dSNate Lawson 	int cpu_id, error, i;
21373347b07SNate Lawson 
21473347b07SNate Lawson 	sc = device_get_softc(dev);
215789f03ceSNate Lawson 	error = 0;
216789f03ceSNate Lawson 	set = NULL;
21773347b07SNate Lawson 
2180325089dSNate Lawson 	/*
2190325089dSNate Lawson 	 * Check that the TSC isn't being used as a timecounter.
2200325089dSNate Lawson 	 * If it is, then return EBUSY and refuse to change the
2210325089dSNate Lawson 	 * clock speed.
2220325089dSNate Lawson 	 */
2230325089dSNate Lawson 	if (strcmp(timecounter->tc_name, "TSC") == 0)
2240325089dSNate Lawson 		return (EBUSY);
2250325089dSNate Lawson 
2265f0afa04SNate Lawson 	/*
2275f0afa04SNate Lawson 	 * If the caller didn't specify a level and one is saved, prepare to
2285f0afa04SNate Lawson 	 * restore the saved level.  If none has been saved, return an error.
2295f0afa04SNate Lawson 	 * If they did specify one, but the requested level has a lower
2305f0afa04SNate Lawson 	 * priority, don't allow the new level right now.
2315f0afa04SNate Lawson 	 */
232789f03ceSNate Lawson 	CF_MTX_LOCK(&sc->lock);
2335f0afa04SNate Lawson 	if (level == NULL) {
2345f0afa04SNate Lawson 		if (sc->saved_level.total_set.freq != CPUFREQ_VAL_UNKNOWN) {
2355f0afa04SNate Lawson 			level = &sc->saved_level;
2365f0afa04SNate Lawson 			priority = sc->saved_priority;
2378d913481SNate Lawson 			CF_DEBUG("restoring saved level, freq %d prio %d\n",
2388d913481SNate Lawson 			    level->total_set.freq, priority);
239789f03ceSNate Lawson 		} else {
2408d913481SNate Lawson 			CF_DEBUG("NULL level, no saved level\n");
241789f03ceSNate Lawson 			error = ENXIO;
242789f03ceSNate Lawson 			goto out;
243789f03ceSNate Lawson 		}
244789f03ceSNate Lawson 	} else if (priority < sc->curr_priority) {
2458d913481SNate Lawson 		CF_DEBUG("ignoring, curr prio %d less than %d\n", priority,
2468d913481SNate Lawson 		    sc->curr_priority);
247789f03ceSNate Lawson 		error = EPERM;
248789f03ceSNate Lawson 		goto out;
249789f03ceSNate Lawson 	}
2505f0afa04SNate Lawson 
251b070969bSNate Lawson 	/* Reject levels that are below our specified threshold. */
252789f03ceSNate Lawson 	if (level->total_set.freq <= cf_lowest_freq) {
2538d913481SNate Lawson 		CF_DEBUG("rejecting freq %d, less than %d limit\n",
2548d913481SNate Lawson 		    level->total_set.freq, cf_lowest_freq);
255789f03ceSNate Lawson 		error = EINVAL;
256789f03ceSNate Lawson 		goto out;
257789f03ceSNate Lawson 	}
258b070969bSNate Lawson 
25973347b07SNate Lawson 	/* If already at this level, just return. */
2608d913481SNate Lawson 	if (CPUFREQ_CMP(sc->curr_level.total_set.freq, level->total_set.freq)) {
2618d913481SNate Lawson 		CF_DEBUG("skipping freq %d, same as current level %d\n",
2628d913481SNate Lawson 		    level->total_set.freq, sc->curr_level.total_set.freq);
263789f03ceSNate Lawson 		goto out;
2648d913481SNate Lawson 	}
26573347b07SNate Lawson 
26673347b07SNate Lawson 	/* First, set the absolute frequency via its driver. */
26773347b07SNate Lawson 	set = &level->abs_set;
26873347b07SNate Lawson 	if (set->dev) {
26973347b07SNate Lawson 		if (!device_is_attached(set->dev)) {
27073347b07SNate Lawson 			error = ENXIO;
27173347b07SNate Lawson 			goto out;
27273347b07SNate Lawson 		}
2731196826aSNate Lawson 
2741196826aSNate Lawson 		/* Bind to the target CPU before switching, if necessary. */
2751196826aSNate Lawson 		cpu_id = PCPU_GET(cpuid);
2761196826aSNate Lawson 		pc = cpu_get_pcpu(set->dev);
2771196826aSNate Lawson 		if (cpu_id != pc->pc_cpuid) {
2781196826aSNate Lawson 			mtx_lock_spin(&sched_lock);
2791196826aSNate Lawson 			sched_bind(curthread, pc->pc_cpuid);
2801196826aSNate Lawson 			mtx_unlock_spin(&sched_lock);
2811196826aSNate Lawson 		}
2828d913481SNate Lawson 		CF_DEBUG("setting abs freq %d on %s (cpu %d)\n", set->freq,
2838d913481SNate Lawson 		    device_get_nameunit(set->dev), PCPU_GET(cpuid));
28473347b07SNate Lawson 		error = CPUFREQ_DRV_SET(set->dev, set);
2851196826aSNate Lawson 		if (cpu_id != pc->pc_cpuid) {
2861196826aSNate Lawson 			mtx_lock_spin(&sched_lock);
2871196826aSNate Lawson 			sched_unbind(curthread);
2881196826aSNate Lawson 			mtx_unlock_spin(&sched_lock);
2891196826aSNate Lawson 		}
29073347b07SNate Lawson 		if (error) {
29173347b07SNate Lawson 			goto out;
29273347b07SNate Lawson 		}
29373347b07SNate Lawson 	}
29473347b07SNate Lawson 
29588c9b54cSNate Lawson 	/* Next, set any/all relative frequencies via their drivers. */
29688c9b54cSNate Lawson 	for (i = 0; i < level->rel_count; i++) {
29788c9b54cSNate Lawson 		set = &level->rel_set[i];
29888c9b54cSNate Lawson 		if (!device_is_attached(set->dev)) {
29988c9b54cSNate Lawson 			error = ENXIO;
30088c9b54cSNate Lawson 			goto out;
30188c9b54cSNate Lawson 		}
3021196826aSNate Lawson 
3031196826aSNate Lawson 		/* Bind to the target CPU before switching, if necessary. */
3041196826aSNate Lawson 		cpu_id = PCPU_GET(cpuid);
3051196826aSNate Lawson 		pc = cpu_get_pcpu(set->dev);
3061196826aSNate Lawson 		if (cpu_id != pc->pc_cpuid) {
3071196826aSNate Lawson 			mtx_lock_spin(&sched_lock);
3081196826aSNate Lawson 			sched_bind(curthread, pc->pc_cpuid);
3091196826aSNate Lawson 			mtx_unlock_spin(&sched_lock);
3101196826aSNate Lawson 		}
3118d913481SNate Lawson 		CF_DEBUG("setting rel freq %d on %s (cpu %d)\n", set->freq,
3128d913481SNate Lawson 		    device_get_nameunit(set->dev), PCPU_GET(cpuid));
31388c9b54cSNate Lawson 		error = CPUFREQ_DRV_SET(set->dev, set);
3141196826aSNate Lawson 		if (cpu_id != pc->pc_cpuid) {
3151196826aSNate Lawson 			mtx_lock_spin(&sched_lock);
3161196826aSNate Lawson 			sched_unbind(curthread);
3171196826aSNate Lawson 			mtx_unlock_spin(&sched_lock);
3181196826aSNate Lawson 		}
31988c9b54cSNate Lawson 		if (error) {
32088c9b54cSNate Lawson 			/* XXX Back out any successful setting? */
32188c9b54cSNate Lawson 			goto out;
32288c9b54cSNate Lawson 		}
32388c9b54cSNate Lawson 	}
32473347b07SNate Lawson 
3255f0afa04SNate Lawson 	/* If we were restoring a saved state, reset it to "unused". */
3265f0afa04SNate Lawson 	if (level == &sc->saved_level) {
3278d913481SNate Lawson 		CF_DEBUG("resetting saved level\n");
3285f0afa04SNate Lawson 		sc->saved_level.total_set.freq = CPUFREQ_VAL_UNKNOWN;
3295f0afa04SNate Lawson 		sc->saved_priority = 0;
3305f0afa04SNate Lawson 	}
3315f0afa04SNate Lawson 
3325f0afa04SNate Lawson 	/*
3335f0afa04SNate Lawson 	 * Before recording the current level, check if we're going to a
3345f0afa04SNate Lawson 	 * higher priority and have not saved a level yet.  If so, save the
3355f0afa04SNate Lawson 	 * previous level and priority.
3365f0afa04SNate Lawson 	 */
3375f0afa04SNate Lawson 	if (sc->curr_level.total_set.freq != CPUFREQ_VAL_UNKNOWN &&
3385f0afa04SNate Lawson 	    sc->saved_level.total_set.freq == CPUFREQ_VAL_UNKNOWN &&
3395f0afa04SNate Lawson 	    priority > sc->curr_priority) {
3408d913481SNate Lawson 		CF_DEBUG("saving level, freq %d prio %d\n",
3418d913481SNate Lawson 		    sc->curr_level.total_set.freq, sc->curr_priority);
3425f0afa04SNate Lawson 		sc->saved_level = sc->curr_level;
3435f0afa04SNate Lawson 		sc->saved_priority = sc->curr_priority;
3445f0afa04SNate Lawson 	}
34573347b07SNate Lawson 	sc->curr_level = *level;
3465f0afa04SNate Lawson 	sc->curr_priority = priority;
34773347b07SNate Lawson 	error = 0;
34873347b07SNate Lawson 
34973347b07SNate Lawson out:
350789f03ceSNate Lawson 	CF_MTX_UNLOCK(&sc->lock);
351789f03ceSNate Lawson 	if (error && set)
35273347b07SNate Lawson 		device_printf(set->dev, "set freq failed, err %d\n", error);
35373347b07SNate Lawson 	return (error);
35473347b07SNate Lawson }
35573347b07SNate Lawson 
35673347b07SNate Lawson static int
35773347b07SNate Lawson cf_get_method(device_t dev, struct cf_level *level)
35873347b07SNate Lawson {
35973347b07SNate Lawson 	struct cpufreq_softc *sc;
36073347b07SNate Lawson 	struct cf_level *levels;
36173347b07SNate Lawson 	struct cf_setting *curr_set, set;
36273347b07SNate Lawson 	struct pcpu *pc;
36373347b07SNate Lawson 	device_t *devs;
36473347b07SNate Lawson 	int count, error, i, numdevs;
36573347b07SNate Lawson 	uint64_t rate;
36673347b07SNate Lawson 
36773347b07SNate Lawson 	sc = device_get_softc(dev);
368789f03ceSNate Lawson 	error = 0;
36973347b07SNate Lawson 	levels = NULL;
37073347b07SNate Lawson 
37173347b07SNate Lawson 	/* If we already know the current frequency, we're done. */
372789f03ceSNate Lawson 	CF_MTX_LOCK(&sc->lock);
373789f03ceSNate Lawson 	curr_set = &sc->curr_level.total_set;
3748d913481SNate Lawson 	if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) {
3758d913481SNate Lawson 		CF_DEBUG("get returning known freq %d\n", curr_set->freq);
37673347b07SNate Lawson 		goto out;
3778d913481SNate Lawson 	}
378789f03ceSNate Lawson 	CF_MTX_UNLOCK(&sc->lock);
37973347b07SNate Lawson 
38073347b07SNate Lawson 	/*
38173347b07SNate Lawson 	 * We need to figure out the current level.  Loop through every
38273347b07SNate Lawson 	 * driver, getting the current setting.  Then, attempt to get a best
38373347b07SNate Lawson 	 * match of settings against each level.
38473347b07SNate Lawson 	 */
38573347b07SNate Lawson 	count = CF_MAX_LEVELS;
38673347b07SNate Lawson 	levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
38773347b07SNate Lawson 	if (levels == NULL)
38873347b07SNate Lawson 		return (ENOMEM);
38973347b07SNate Lawson 	error = CPUFREQ_LEVELS(sc->dev, levels, &count);
390d269386aSNate Lawson 	if (error) {
391d269386aSNate Lawson 		if (error == E2BIG)
392d269386aSNate Lawson 			printf("cpufreq: need to increase CF_MAX_LEVELS\n");
393789f03ceSNate Lawson 		free(levels, M_TEMP);
394789f03ceSNate Lawson 		return (error);
395d269386aSNate Lawson 	}
39673347b07SNate Lawson 	error = device_get_children(device_get_parent(dev), &devs, &numdevs);
397789f03ceSNate Lawson 	if (error) {
398789f03ceSNate Lawson 		free(levels, M_TEMP);
399789f03ceSNate Lawson 		return (error);
400789f03ceSNate Lawson 	}
401789f03ceSNate Lawson 
402789f03ceSNate Lawson 	/*
403789f03ceSNate Lawson 	 * Reacquire the lock and search for the given level.
404789f03ceSNate Lawson 	 *
405789f03ceSNate Lawson 	 * XXX Note: this is not quite right since we really need to go
406789f03ceSNate Lawson 	 * through each level and compare both absolute and relative
407789f03ceSNate Lawson 	 * settings for each driver in the system before making a match.
408789f03ceSNate Lawson 	 * The estimation code below catches this case though.
409789f03ceSNate Lawson 	 */
410789f03ceSNate Lawson 	CF_MTX_LOCK(&sc->lock);
41173347b07SNate Lawson 	for (i = 0; i < numdevs && curr_set->freq == CPUFREQ_VAL_UNKNOWN; i++) {
41273347b07SNate Lawson 		if (!device_is_attached(devs[i]))
41373347b07SNate Lawson 			continue;
41473347b07SNate Lawson 		error = CPUFREQ_DRV_GET(devs[i], &set);
41573347b07SNate Lawson 		if (error)
41673347b07SNate Lawson 			continue;
41773347b07SNate Lawson 		for (i = 0; i < count; i++) {
41888c9b54cSNate Lawson 			if (CPUFREQ_CMP(set.freq, levels[i].total_set.freq)) {
41973347b07SNate Lawson 				sc->curr_level = levels[i];
42073347b07SNate Lawson 				break;
42173347b07SNate Lawson 			}
42273347b07SNate Lawson 		}
42373347b07SNate Lawson 	}
42473347b07SNate Lawson 	free(devs, M_TEMP);
4258d913481SNate Lawson 	if (curr_set->freq != CPUFREQ_VAL_UNKNOWN) {
4268d913481SNate Lawson 		CF_DEBUG("get matched freq %d from drivers\n", curr_set->freq);
42773347b07SNate Lawson 		goto out;
4288d913481SNate Lawson 	}
42973347b07SNate Lawson 
43073347b07SNate Lawson 	/*
43173347b07SNate Lawson 	 * We couldn't find an exact match, so attempt to estimate and then
43273347b07SNate Lawson 	 * match against a level.
43373347b07SNate Lawson 	 */
43473347b07SNate Lawson 	pc = cpu_get_pcpu(dev);
43573347b07SNate Lawson 	if (pc == NULL) {
43673347b07SNate Lawson 		error = ENXIO;
43773347b07SNate Lawson 		goto out;
43873347b07SNate Lawson 	}
43973347b07SNate Lawson 	cpu_est_clockrate(pc->pc_cpuid, &rate);
44073347b07SNate Lawson 	rate /= 1000000;
44173347b07SNate Lawson 	for (i = 0; i < count; i++) {
44273347b07SNate Lawson 		if (CPUFREQ_CMP(rate, levels[i].total_set.freq)) {
44373347b07SNate Lawson 			sc->curr_level = levels[i];
4448d913481SNate Lawson 			CF_DEBUG("get estimated freq %d\n", curr_set->freq);
44573347b07SNate Lawson 			break;
44673347b07SNate Lawson 		}
44773347b07SNate Lawson 	}
44873347b07SNate Lawson 
44973347b07SNate Lawson out:
450789f03ceSNate Lawson 	if (error == 0)
451789f03ceSNate Lawson 		*level = sc->curr_level;
452789f03ceSNate Lawson 
453789f03ceSNate Lawson 	CF_MTX_UNLOCK(&sc->lock);
45473347b07SNate Lawson 	if (levels)
45573347b07SNate Lawson 		free(levels, M_TEMP);
456789f03ceSNate Lawson 	return (error);
45773347b07SNate Lawson }
45873347b07SNate Lawson 
45973347b07SNate Lawson static int
46073347b07SNate Lawson cf_levels_method(device_t dev, struct cf_level *levels, int *count)
46173347b07SNate Lawson {
46288c9b54cSNate Lawson 	struct cf_setting_array *set_arr;
46373347b07SNate Lawson 	struct cf_setting_lst rel_sets;
46473347b07SNate Lawson 	struct cpufreq_softc *sc;
46573347b07SNate Lawson 	struct cf_level *lev;
46673347b07SNate Lawson 	struct cf_setting *sets;
46773347b07SNate Lawson 	struct pcpu *pc;
46873347b07SNate Lawson 	device_t *devs;
46988c9b54cSNate Lawson 	int error, i, numdevs, set_count, type;
47073347b07SNate Lawson 	uint64_t rate;
47173347b07SNate Lawson 
47273347b07SNate Lawson 	if (levels == NULL || count == NULL)
47373347b07SNate Lawson 		return (EINVAL);
47473347b07SNate Lawson 
47573347b07SNate Lawson 	TAILQ_INIT(&rel_sets);
47673347b07SNate Lawson 	sc = device_get_softc(dev);
47773347b07SNate Lawson 	error = device_get_children(device_get_parent(dev), &devs, &numdevs);
47873347b07SNate Lawson 	if (error)
47973347b07SNate Lawson 		return (error);
48073347b07SNate Lawson 	sets = malloc(MAX_SETTINGS * sizeof(*sets), M_TEMP, M_NOWAIT);
48173347b07SNate Lawson 	if (sets == NULL) {
48273347b07SNate Lawson 		free(devs, M_TEMP);
48373347b07SNate Lawson 		return (ENOMEM);
48473347b07SNate Lawson 	}
48573347b07SNate Lawson 
48673347b07SNate Lawson 	/* Get settings from all cpufreq drivers. */
487789f03ceSNate Lawson 	CF_MTX_LOCK(&sc->lock);
48873347b07SNate Lawson 	for (i = 0; i < numdevs; i++) {
489e22cd41cSNate Lawson 		/* Skip devices that aren't ready. */
49073347b07SNate Lawson 		if (!device_is_attached(devs[i]))
49173347b07SNate Lawson 			continue;
492e22cd41cSNate Lawson 
493e22cd41cSNate Lawson 		/*
494e22cd41cSNate Lawson 		 * Get settings, skipping drivers that offer no settings or
495e22cd41cSNate Lawson 		 * provide settings for informational purposes only.
496e22cd41cSNate Lawson 		 */
497e94a0c1aSNate Lawson 		error = CPUFREQ_DRV_TYPE(devs[i], &type);
4988d913481SNate Lawson 		if (error || (type & CPUFREQ_FLAG_INFO_ONLY)) {
4998d913481SNate Lawson 			if (error == 0) {
5008d913481SNate Lawson 				CF_DEBUG("skipping info-only driver %s\n",
5018d913481SNate Lawson 				    device_get_nameunit(devs[i]));
5028d913481SNate Lawson 			}
503e94a0c1aSNate Lawson 			continue;
5048d913481SNate Lawson 		}
50573347b07SNate Lawson 		set_count = MAX_SETTINGS;
506e94a0c1aSNate Lawson 		error = CPUFREQ_DRV_SETTINGS(devs[i], sets, &set_count);
507e94a0c1aSNate Lawson 		if (error || set_count == 0)
50873347b07SNate Lawson 			continue;
50973347b07SNate Lawson 
510e22cd41cSNate Lawson 		/* Add the settings to our absolute/relative lists. */
5110325089dSNate Lawson 		switch (type & CPUFREQ_TYPE_MASK) {
51288c9b54cSNate Lawson 		case CPUFREQ_TYPE_ABSOLUTE:
51388c9b54cSNate Lawson 			error = cpufreq_insert_abs(sc, sets, set_count);
51488c9b54cSNate Lawson 			break;
51588c9b54cSNate Lawson 		case CPUFREQ_TYPE_RELATIVE:
5168d913481SNate Lawson 			CF_DEBUG("adding %d relative settings\n", set_count);
51788c9b54cSNate Lawson 			set_arr = malloc(sizeof(*set_arr), M_TEMP, M_NOWAIT);
51888c9b54cSNate Lawson 			if (set_arr == NULL) {
51988c9b54cSNate Lawson 				error = ENOMEM;
52088c9b54cSNate Lawson 				goto out;
52188c9b54cSNate Lawson 			}
52288c9b54cSNate Lawson 			bcopy(sets, set_arr->sets, set_count * sizeof(*sets));
52388c9b54cSNate Lawson 			set_arr->count = set_count;
52488c9b54cSNate Lawson 			TAILQ_INSERT_TAIL(&rel_sets, set_arr, link);
52588c9b54cSNate Lawson 			break;
52688c9b54cSNate Lawson 		default:
52788c9b54cSNate Lawson 			error = EINVAL;
52888c9b54cSNate Lawson 		}
52988c9b54cSNate Lawson 		if (error)
53073347b07SNate Lawson 			goto out;
53173347b07SNate Lawson 	}
53273347b07SNate Lawson 
53367c8649fSNate Lawson 	/*
53467c8649fSNate Lawson 	 * If there are no absolute levels, create a fake one at 100%.  We
53567c8649fSNate Lawson 	 * then cache the clockrate for later use as our base frequency.
53667c8649fSNate Lawson 	 *
53767c8649fSNate Lawson 	 * XXX This assumes that the first time through, if we only have
53867c8649fSNate Lawson 	 * relative drivers, the CPU is currently running at 100%.
53967c8649fSNate Lawson 	 */
54073347b07SNate Lawson 	if (TAILQ_EMPTY(&sc->all_levels)) {
54167c8649fSNate Lawson 		if (sc->max_mhz == CPUFREQ_VAL_UNKNOWN) {
54273347b07SNate Lawson 			pc = cpu_get_pcpu(dev);
54373347b07SNate Lawson 			cpu_est_clockrate(pc->pc_cpuid, &rate);
54467c8649fSNate Lawson 			sc->max_mhz = rate / 1000000;
54567c8649fSNate Lawson 		}
54667c8649fSNate Lawson 		memset(&sets[0], CPUFREQ_VAL_UNKNOWN, sizeof(*sets));
54767c8649fSNate Lawson 		sets[0].freq = sc->max_mhz;
54867c8649fSNate Lawson 		sets[0].dev = NULL;
54988c9b54cSNate Lawson 		error = cpufreq_insert_abs(sc, sets, 1);
55073347b07SNate Lawson 		if (error)
55173347b07SNate Lawson 			goto out;
55273347b07SNate Lawson 	}
55373347b07SNate Lawson 
55488c9b54cSNate Lawson 	/* Create a combined list of absolute + relative levels. */
55588c9b54cSNate Lawson 	TAILQ_FOREACH(set_arr, &rel_sets, link)
55688c9b54cSNate Lawson 		cpufreq_expand_set(sc, set_arr);
55788c9b54cSNate Lawson 
55888c9b54cSNate Lawson 	/* If the caller doesn't have enough space, return the actual count. */
55988c9b54cSNate Lawson 	if (sc->all_count > *count) {
56088c9b54cSNate Lawson 		*count = sc->all_count;
56188c9b54cSNate Lawson 		error = E2BIG;
56288c9b54cSNate Lawson 		goto out;
56388c9b54cSNate Lawson 	}
56488c9b54cSNate Lawson 
56588c9b54cSNate Lawson 	/* Finally, output the list of levels. */
56673347b07SNate Lawson 	i = 0;
56773347b07SNate Lawson 	TAILQ_FOREACH(lev, &sc->all_levels, link) {
568b070969bSNate Lawson 		/* Skip levels that have a frequency that is too low. */
569b070969bSNate Lawson 		if (lev->total_set.freq <= cf_lowest_freq) {
570b070969bSNate Lawson 			sc->all_count--;
571b070969bSNate Lawson 			continue;
572b070969bSNate Lawson 		}
573b070969bSNate Lawson 
57473347b07SNate Lawson 		levels[i] = *lev;
57573347b07SNate Lawson 		i++;
57673347b07SNate Lawson 	}
57788c9b54cSNate Lawson 	*count = sc->all_count;
57873347b07SNate Lawson 	error = 0;
57973347b07SNate Lawson 
58073347b07SNate Lawson out:
58173347b07SNate Lawson 	/* Clear all levels since we regenerate them each time. */
58273347b07SNate Lawson 	while ((lev = TAILQ_FIRST(&sc->all_levels)) != NULL) {
58373347b07SNate Lawson 		TAILQ_REMOVE(&sc->all_levels, lev, link);
58473347b07SNate Lawson 		free(lev, M_TEMP);
58573347b07SNate Lawson 	}
586789f03ceSNate Lawson 	sc->all_count = 0;
587789f03ceSNate Lawson 
588789f03ceSNate Lawson 	CF_MTX_UNLOCK(&sc->lock);
58988c9b54cSNate Lawson 	while ((set_arr = TAILQ_FIRST(&rel_sets)) != NULL) {
59088c9b54cSNate Lawson 		TAILQ_REMOVE(&rel_sets, set_arr, link);
59188c9b54cSNate Lawson 		free(set_arr, M_TEMP);
59288c9b54cSNate Lawson 	}
59373347b07SNate Lawson 	free(devs, M_TEMP);
59473347b07SNate Lawson 	free(sets, M_TEMP);
59573347b07SNate Lawson 	return (error);
59673347b07SNate Lawson }
59773347b07SNate Lawson 
59873347b07SNate Lawson /*
59973347b07SNate Lawson  * Create levels for an array of absolute settings and insert them in
60073347b07SNate Lawson  * sorted order in the specified list.
60173347b07SNate Lawson  */
60273347b07SNate Lawson static int
60388c9b54cSNate Lawson cpufreq_insert_abs(struct cpufreq_softc *sc, struct cf_setting *sets,
60473347b07SNate Lawson     int count)
60573347b07SNate Lawson {
60688c9b54cSNate Lawson 	struct cf_level_lst *list;
60773347b07SNate Lawson 	struct cf_level *level, *search;
60873347b07SNate Lawson 	int i;
60973347b07SNate Lawson 
610789f03ceSNate Lawson 	CF_MTX_ASSERT(&sc->lock);
611789f03ceSNate Lawson 
61288c9b54cSNate Lawson 	list = &sc->all_levels;
61373347b07SNate Lawson 	for (i = 0; i < count; i++) {
61473347b07SNate Lawson 		level = malloc(sizeof(*level), M_TEMP, M_NOWAIT | M_ZERO);
61573347b07SNate Lawson 		if (level == NULL)
61673347b07SNate Lawson 			return (ENOMEM);
61773347b07SNate Lawson 		level->abs_set = sets[i];
61888c9b54cSNate Lawson 		level->total_set = sets[i];
61988c9b54cSNate Lawson 		level->total_set.dev = NULL;
62088c9b54cSNate Lawson 		sc->all_count++;
62173347b07SNate Lawson 
62273347b07SNate Lawson 		if (TAILQ_EMPTY(list)) {
6238d913481SNate Lawson 			CF_DEBUG("adding abs setting %d at head\n",
6248d913481SNate Lawson 			    sets[i].freq);
62573347b07SNate Lawson 			TAILQ_INSERT_HEAD(list, level, link);
62673347b07SNate Lawson 			continue;
62773347b07SNate Lawson 		}
62873347b07SNate Lawson 
62973347b07SNate Lawson 		TAILQ_FOREACH_REVERSE(search, list, cf_level_lst, link) {
63088c9b54cSNate Lawson 			if (sets[i].freq <= search->total_set.freq) {
6318d913481SNate Lawson 				CF_DEBUG("adding abs setting %d after %d\n",
6328d913481SNate Lawson 				    sets[i].freq, search->total_set.freq);
63373347b07SNate Lawson 				TAILQ_INSERT_AFTER(list, search, level, link);
63473347b07SNate Lawson 				break;
63573347b07SNate Lawson 			}
63673347b07SNate Lawson 		}
63773347b07SNate Lawson 	}
63873347b07SNate Lawson 	return (0);
63973347b07SNate Lawson }
64073347b07SNate Lawson 
64188c9b54cSNate Lawson /*
64288c9b54cSNate Lawson  * Expand a group of relative settings, creating derived levels from them.
64388c9b54cSNate Lawson  */
64488c9b54cSNate Lawson static int
64588c9b54cSNate Lawson cpufreq_expand_set(struct cpufreq_softc *sc, struct cf_setting_array *set_arr)
64688c9b54cSNate Lawson {
64788c9b54cSNate Lawson 	struct cf_level *fill, *search;
64888c9b54cSNate Lawson 	struct cf_setting *set;
64988c9b54cSNate Lawson 	int i;
65088c9b54cSNate Lawson 
651789f03ceSNate Lawson 	CF_MTX_ASSERT(&sc->lock);
652789f03ceSNate Lawson 
65388c9b54cSNate Lawson 	TAILQ_FOREACH(search, &sc->all_levels, link) {
65488c9b54cSNate Lawson 		/* Skip this level if we've already modified it. */
65588c9b54cSNate Lawson 		for (i = 0; i < search->rel_count; i++) {
65688c9b54cSNate Lawson 			if (search->rel_set[i].dev == set_arr->sets[0].dev)
65788c9b54cSNate Lawson 				break;
65888c9b54cSNate Lawson 		}
6598d913481SNate Lawson 		if (i != search->rel_count) {
6608d913481SNate Lawson 			CF_DEBUG("skipping modified level, freq %d (dev %s)\n",
6618d913481SNate Lawson 			    search->total_set.freq,
6628d913481SNate Lawson 			    device_get_nameunit(search->rel_set[i].dev));
66388c9b54cSNate Lawson 			continue;
6648d913481SNate Lawson 		}
66588c9b54cSNate Lawson 
66688c9b54cSNate Lawson 		/* Add each setting to the level, duplicating if necessary. */
66788c9b54cSNate Lawson 		for (i = 0; i < set_arr->count; i++) {
66888c9b54cSNate Lawson 			set = &set_arr->sets[i];
66988c9b54cSNate Lawson 
67088c9b54cSNate Lawson 			/*
67188c9b54cSNate Lawson 			 * If this setting is less than 100%, split the level
67288c9b54cSNate Lawson 			 * into two and add this setting to the new level.
67388c9b54cSNate Lawson 			 */
67488c9b54cSNate Lawson 			fill = search;
67588c9b54cSNate Lawson 			if (set->freq < 10000)
67688c9b54cSNate Lawson 				fill = cpufreq_dup_set(sc, search, set);
67788c9b54cSNate Lawson 
67888c9b54cSNate Lawson 			/*
67988c9b54cSNate Lawson 			 * The new level was a duplicate of an existing level
68088c9b54cSNate Lawson 			 * so we freed it.  Go to the next setting.
68188c9b54cSNate Lawson 			 */
68288c9b54cSNate Lawson 			if (fill == NULL)
68388c9b54cSNate Lawson 				continue;
68488c9b54cSNate Lawson 
68588c9b54cSNate Lawson 			/* Add this setting to the existing or new level. */
68688c9b54cSNate Lawson 			KASSERT(fill->rel_count < MAX_SETTINGS,
68788c9b54cSNate Lawson 			    ("cpufreq: too many relative drivers (%d)",
68888c9b54cSNate Lawson 			    MAX_SETTINGS));
68988c9b54cSNate Lawson 			fill->rel_set[fill->rel_count] = *set;
69088c9b54cSNate Lawson 			fill->rel_count++;
6918d913481SNate Lawson 			CF_DEBUG(
6928d913481SNate Lawson 			"expand set added rel setting %d%% to %d level\n",
6938d913481SNate Lawson 			    set->freq / 100, fill->total_set.freq);
69488c9b54cSNate Lawson 		}
69588c9b54cSNate Lawson 	}
69688c9b54cSNate Lawson 
69788c9b54cSNate Lawson 	return (0);
69888c9b54cSNate Lawson }
69988c9b54cSNate Lawson 
70088c9b54cSNate Lawson static struct cf_level *
70188c9b54cSNate Lawson cpufreq_dup_set(struct cpufreq_softc *sc, struct cf_level *dup,
70288c9b54cSNate Lawson     struct cf_setting *set)
70388c9b54cSNate Lawson {
70488c9b54cSNate Lawson 	struct cf_level_lst *list;
70588c9b54cSNate Lawson 	struct cf_level *fill, *itr;
70688c9b54cSNate Lawson 	struct cf_setting *fill_set, *itr_set;
70788c9b54cSNate Lawson 	int i;
70888c9b54cSNate Lawson 
709789f03ceSNate Lawson 	CF_MTX_ASSERT(&sc->lock);
710789f03ceSNate Lawson 
71188c9b54cSNate Lawson 	/*
71288c9b54cSNate Lawson 	 * Create a new level, copy it from the old one, and update the
71388c9b54cSNate Lawson 	 * total frequency and power by the percentage specified in the
71488c9b54cSNate Lawson 	 * relative setting.
71588c9b54cSNate Lawson 	 */
71688c9b54cSNate Lawson 	fill = malloc(sizeof(*fill), M_TEMP, M_NOWAIT);
71788c9b54cSNate Lawson 	if (fill == NULL)
71888c9b54cSNate Lawson 		return (NULL);
71988c9b54cSNate Lawson 	*fill = *dup;
72088c9b54cSNate Lawson 	fill_set = &fill->total_set;
72188c9b54cSNate Lawson 	fill_set->freq =
72288c9b54cSNate Lawson 	    ((uint64_t)fill_set->freq * set->freq) / 10000;
72388c9b54cSNate Lawson 	if (fill_set->power != CPUFREQ_VAL_UNKNOWN) {
72488c9b54cSNate Lawson 		fill_set->power = ((uint64_t)fill_set->power * set->freq)
72588c9b54cSNate Lawson 		    / 10000;
72688c9b54cSNate Lawson 	}
72788c9b54cSNate Lawson 	if (set->lat != CPUFREQ_VAL_UNKNOWN) {
72888c9b54cSNate Lawson 		if (fill_set->lat != CPUFREQ_VAL_UNKNOWN)
72988c9b54cSNate Lawson 			fill_set->lat += set->lat;
73088c9b54cSNate Lawson 		else
73188c9b54cSNate Lawson 			fill_set->lat = set->lat;
73288c9b54cSNate Lawson 	}
7338d913481SNate Lawson 	CF_DEBUG("dup set considering derived setting %d\n", fill_set->freq);
73488c9b54cSNate Lawson 
73588c9b54cSNate Lawson 	/*
73688c9b54cSNate Lawson 	 * If we copied an old level that we already modified (say, at 100%),
73788c9b54cSNate Lawson 	 * we need to remove that setting before adding this one.  Since we
73888c9b54cSNate Lawson 	 * process each setting array in order, we know any settings for this
73988c9b54cSNate Lawson 	 * driver will be found at the end.
74088c9b54cSNate Lawson 	 */
74188c9b54cSNate Lawson 	for (i = fill->rel_count; i != 0; i--) {
74288c9b54cSNate Lawson 		if (fill->rel_set[i - 1].dev != set->dev)
74388c9b54cSNate Lawson 			break;
7448d913481SNate Lawson 		CF_DEBUG("removed last relative driver: %s\n",
7458d913481SNate Lawson 		    device_get_nameunit(set->dev));
74688c9b54cSNate Lawson 		fill->rel_count--;
74788c9b54cSNate Lawson 	}
74888c9b54cSNate Lawson 
74988c9b54cSNate Lawson 	/*
75088c9b54cSNate Lawson 	 * Insert the new level in sorted order.  If we find a duplicate,
75188c9b54cSNate Lawson 	 * free the new level.  We can do this since any existing level will
75288c9b54cSNate Lawson 	 * be guaranteed to have the same or less settings and thus consume
75388c9b54cSNate Lawson 	 * less power.  For example, a level with one absolute setting of
75488c9b54cSNate Lawson 	 * 800 Mhz uses less power than one composed of an absolute setting
75588c9b54cSNate Lawson 	 * of 1600 Mhz and a relative setting at 50%.
75688c9b54cSNate Lawson 	 */
75788c9b54cSNate Lawson 	list = &sc->all_levels;
75888c9b54cSNate Lawson 	if (TAILQ_EMPTY(list)) {
7598d913481SNate Lawson 		CF_DEBUG("dup done, inserted %d at head\n", fill_set->freq);
76088c9b54cSNate Lawson 		TAILQ_INSERT_HEAD(list, fill, link);
76188c9b54cSNate Lawson 	} else {
76288c9b54cSNate Lawson 		TAILQ_FOREACH_REVERSE(itr, list, cf_level_lst, link) {
76388c9b54cSNate Lawson 			itr_set = &itr->total_set;
76488c9b54cSNate Lawson 			if (CPUFREQ_CMP(fill_set->freq, itr_set->freq)) {
7658d913481SNate Lawson 				CF_DEBUG(
7668d913481SNate Lawson 			"dup done, freeing new level %d, matches %d\n",
7678d913481SNate Lawson 				    fill_set->freq, itr_set->freq);
76888c9b54cSNate Lawson 				free(fill, M_TEMP);
76988c9b54cSNate Lawson 				fill = NULL;
77088c9b54cSNate Lawson 				break;
77188c9b54cSNate Lawson 			} else if (fill_set->freq < itr_set->freq) {
7728d913481SNate Lawson 				CF_DEBUG(
7738d913481SNate Lawson 			"dup done, inserting new level %d after %d\n",
7748d913481SNate Lawson 				    fill_set->freq, itr_set->freq);
77588c9b54cSNate Lawson 				TAILQ_INSERT_AFTER(list, itr, fill, link);
77688c9b54cSNate Lawson 				sc->all_count++;
77788c9b54cSNate Lawson 				break;
77888c9b54cSNate Lawson 			}
77988c9b54cSNate Lawson 		}
78088c9b54cSNate Lawson 	}
78188c9b54cSNate Lawson 
78288c9b54cSNate Lawson 	return (fill);
78388c9b54cSNate Lawson }
78488c9b54cSNate Lawson 
78573347b07SNate Lawson static int
78673347b07SNate Lawson cpufreq_curr_sysctl(SYSCTL_HANDLER_ARGS)
78773347b07SNate Lawson {
78873347b07SNate Lawson 	struct cpufreq_softc *sc;
78973347b07SNate Lawson 	struct cf_level *levels;
7900325089dSNate Lawson 	int count, devcount, error, freq, i, n;
7910325089dSNate Lawson 	device_t *devs;
79273347b07SNate Lawson 
7930325089dSNate Lawson 	devs = NULL;
79473347b07SNate Lawson 	sc = oidp->oid_arg1;
7950325089dSNate Lawson 	levels = malloc(CF_MAX_LEVELS * sizeof(*levels), M_TEMP, M_NOWAIT);
79673347b07SNate Lawson 	if (levels == NULL)
79773347b07SNate Lawson 		return (ENOMEM);
79873347b07SNate Lawson 
79973347b07SNate Lawson 	error = CPUFREQ_GET(sc->dev, &levels[0]);
80073347b07SNate Lawson 	if (error)
80173347b07SNate Lawson 		goto out;
80273347b07SNate Lawson 	freq = levels[0].total_set.freq;
80373347b07SNate Lawson 	error = sysctl_handle_int(oidp, &freq, 0, req);
80473347b07SNate Lawson 	if (error != 0 || req->newptr == NULL)
80573347b07SNate Lawson 		goto out;
80673347b07SNate Lawson 
8070325089dSNate Lawson 	/*
8080325089dSNate Lawson 	 * While we only call cpufreq_get() on one device (assuming all
8090325089dSNate Lawson 	 * CPUs have equal levels), we call cpufreq_set() on all CPUs.
8100325089dSNate Lawson 	 * This is needed for some MP systems.
8110325089dSNate Lawson 	 */
8120325089dSNate Lawson 	error = devclass_get_devices(cpufreq_dc, &devs, &devcount);
81373347b07SNate Lawson 	if (error)
81473347b07SNate Lawson 		goto out;
8150325089dSNate Lawson 	for (n = 0; n < devcount; n++) {
8160325089dSNate Lawson 		count = CF_MAX_LEVELS;
8170325089dSNate Lawson 		error = CPUFREQ_LEVELS(devs[n], levels, &count);
818d269386aSNate Lawson 		if (error) {
819d269386aSNate Lawson 			if (error == E2BIG)
820d269386aSNate Lawson 				printf(
821d269386aSNate Lawson 			"cpufreq: need to increase CF_MAX_LEVELS\n");
8220325089dSNate Lawson 			break;
823d269386aSNate Lawson 		}
82473347b07SNate Lawson 		for (i = 0; i < count; i++) {
82573347b07SNate Lawson 			if (CPUFREQ_CMP(levels[i].total_set.freq, freq)) {
8260325089dSNate Lawson 				error = CPUFREQ_SET(devs[n], &levels[i],
82773347b07SNate Lawson 				    CPUFREQ_PRIO_USER);
82873347b07SNate Lawson 				break;
82973347b07SNate Lawson 			}
83073347b07SNate Lawson 		}
8310325089dSNate Lawson 		if (i == count) {
83273347b07SNate Lawson 			error = EINVAL;
8330325089dSNate Lawson 			break;
8340325089dSNate Lawson 		}
8350325089dSNate Lawson 	}
83673347b07SNate Lawson 
83773347b07SNate Lawson out:
8380325089dSNate Lawson 	if (devs)
8390325089dSNate Lawson 		free(devs, M_TEMP);
84073347b07SNate Lawson 	if (levels)
84173347b07SNate Lawson 		free(levels, M_TEMP);
84273347b07SNate Lawson 	return (error);
84373347b07SNate Lawson }
84473347b07SNate Lawson 
84573347b07SNate Lawson static int
84673347b07SNate Lawson cpufreq_levels_sysctl(SYSCTL_HANDLER_ARGS)
84773347b07SNate Lawson {
84873347b07SNate Lawson 	struct cpufreq_softc *sc;
84973347b07SNate Lawson 	struct cf_level *levels;
85073347b07SNate Lawson 	struct cf_setting *set;
85173347b07SNate Lawson 	struct sbuf sb;
85273347b07SNate Lawson 	int count, error, i;
85373347b07SNate Lawson 
85473347b07SNate Lawson 	sc = oidp->oid_arg1;
85573347b07SNate Lawson 	sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
85673347b07SNate Lawson 
85773347b07SNate Lawson 	/* Get settings from the device and generate the output string. */
85873347b07SNate Lawson 	count = CF_MAX_LEVELS;
85973347b07SNate Lawson 	levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
86073347b07SNate Lawson 	if (levels == NULL)
86173347b07SNate Lawson 		return (ENOMEM);
86273347b07SNate Lawson 	error = CPUFREQ_LEVELS(sc->dev, levels, &count);
863d269386aSNate Lawson 	if (error) {
864d269386aSNate Lawson 		if (error == E2BIG)
865d269386aSNate Lawson 			printf("cpufreq: need to increase CF_MAX_LEVELS\n");
86673347b07SNate Lawson 		goto out;
867d269386aSNate Lawson 	}
86873347b07SNate Lawson 	if (count) {
86973347b07SNate Lawson 		for (i = 0; i < count; i++) {
87073347b07SNate Lawson 			set = &levels[i].total_set;
87173347b07SNate Lawson 			sbuf_printf(&sb, "%d/%d ", set->freq, set->power);
87273347b07SNate Lawson 		}
87373347b07SNate Lawson 	} else
87473347b07SNate Lawson 		sbuf_cpy(&sb, "0");
87573347b07SNate Lawson 	sbuf_trim(&sb);
87673347b07SNate Lawson 	sbuf_finish(&sb);
87773347b07SNate Lawson 	error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
87873347b07SNate Lawson 
87973347b07SNate Lawson out:
88073347b07SNate Lawson 	free(levels, M_TEMP);
88173347b07SNate Lawson 	sbuf_delete(&sb);
88273347b07SNate Lawson 	return (error);
88373347b07SNate Lawson }
88473347b07SNate Lawson 
885e959a70bSNate Lawson static int
886e959a70bSNate Lawson cpufreq_settings_sysctl(SYSCTL_HANDLER_ARGS)
887e959a70bSNate Lawson {
888e959a70bSNate Lawson 	device_t dev;
889e959a70bSNate Lawson 	struct cf_setting *sets;
890e959a70bSNate Lawson 	struct sbuf sb;
891e959a70bSNate Lawson 	int error, i, set_count;
892e959a70bSNate Lawson 
893e959a70bSNate Lawson 	dev = oidp->oid_arg1;
894e959a70bSNate Lawson 	sbuf_new(&sb, NULL, 128, SBUF_AUTOEXTEND);
895e959a70bSNate Lawson 
896e959a70bSNate Lawson 	/* Get settings from the device and generate the output string. */
897e959a70bSNate Lawson 	set_count = MAX_SETTINGS;
898e959a70bSNate Lawson 	sets = malloc(set_count * sizeof(*sets), M_TEMP, M_NOWAIT);
899e959a70bSNate Lawson 	if (sets == NULL)
900e959a70bSNate Lawson 		return (ENOMEM);
901e959a70bSNate Lawson 	error = CPUFREQ_DRV_SETTINGS(dev, sets, &set_count);
902e959a70bSNate Lawson 	if (error)
903e959a70bSNate Lawson 		goto out;
904e959a70bSNate Lawson 	if (set_count) {
905e959a70bSNate Lawson 		for (i = 0; i < set_count; i++)
906e959a70bSNate Lawson 			sbuf_printf(&sb, "%d/%d ", sets[i].freq, sets[i].power);
907e959a70bSNate Lawson 	} else
908e959a70bSNate Lawson 		sbuf_cpy(&sb, "0");
909e959a70bSNate Lawson 	sbuf_trim(&sb);
910e959a70bSNate Lawson 	sbuf_finish(&sb);
911e959a70bSNate Lawson 	error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
912e959a70bSNate Lawson 
913e959a70bSNate Lawson out:
914e959a70bSNate Lawson 	free(sets, M_TEMP);
915e959a70bSNate Lawson 	sbuf_delete(&sb);
916e959a70bSNate Lawson 	return (error);
917e959a70bSNate Lawson }
918e959a70bSNate Lawson 
91973347b07SNate Lawson int
92073347b07SNate Lawson cpufreq_register(device_t dev)
92173347b07SNate Lawson {
92267c8649fSNate Lawson 	struct cpufreq_softc *sc;
92373347b07SNate Lawson 	device_t cf_dev, cpu_dev;
92473347b07SNate Lawson 
925e959a70bSNate Lawson 	/* Add a sysctl to get each driver's settings separately. */
926e959a70bSNate Lawson 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
927e959a70bSNate Lawson 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
928e959a70bSNate Lawson 	    OID_AUTO, "freq_settings", CTLTYPE_STRING | CTLFLAG_RD, dev, 0,
929e959a70bSNate Lawson 	    cpufreq_settings_sysctl, "A", "CPU frequency driver settings");
930e959a70bSNate Lawson 
93173347b07SNate Lawson 	/*
9320325089dSNate Lawson 	 * Add only one cpufreq device to each CPU.  Currently, all CPUs
9330325089dSNate Lawson 	 * must offer the same levels and be switched at the same time.
93473347b07SNate Lawson 	 */
9350325089dSNate Lawson 	cpu_dev = device_get_parent(dev);
93667c8649fSNate Lawson 	if ((cf_dev = device_find_child(cpu_dev, "cpufreq", -1))) {
93767c8649fSNate Lawson 		sc = device_get_softc(cf_dev);
93867c8649fSNate Lawson 		sc->max_mhz = CPUFREQ_VAL_UNKNOWN;
93973347b07SNate Lawson 		return (0);
94067c8649fSNate Lawson 	}
94173347b07SNate Lawson 
9420325089dSNate Lawson 	/* Add the child device and possibly sysctls. */
9430325089dSNate Lawson 	cf_dev = BUS_ADD_CHILD(cpu_dev, 0, "cpufreq", -1);
94473347b07SNate Lawson 	if (cf_dev == NULL)
94573347b07SNate Lawson 		return (ENOMEM);
94673347b07SNate Lawson 	device_quiet(cf_dev);
94773347b07SNate Lawson 
94873347b07SNate Lawson 	return (device_probe_and_attach(cf_dev));
94973347b07SNate Lawson }
95073347b07SNate Lawson 
95173347b07SNate Lawson int
95273347b07SNate Lawson cpufreq_unregister(device_t dev)
95373347b07SNate Lawson {
95473347b07SNate Lawson 	device_t cf_dev, *devs;
955e94a0c1aSNate Lawson 	int cfcount, devcount, error, i, type;
95673347b07SNate Lawson 
95773347b07SNate Lawson 	/*
95873347b07SNate Lawson 	 * If this is the last cpufreq child device, remove the control
95973347b07SNate Lawson 	 * device as well.  We identify cpufreq children by calling a method
96073347b07SNate Lawson 	 * they support.
96173347b07SNate Lawson 	 */
96273347b07SNate Lawson 	error = device_get_children(device_get_parent(dev), &devs, &devcount);
96373347b07SNate Lawson 	if (error)
96473347b07SNate Lawson 		return (error);
96567c8649fSNate Lawson 	cf_dev = device_find_child(device_get_parent(dev), "cpufreq", -1);
96671ab130cSNate Lawson 	if (cf_dev == NULL) {
96771ab130cSNate Lawson 		device_printf(dev,
96871ab130cSNate Lawson 	"warning: cpufreq_unregister called with no cpufreq device active\n");
96971ab130cSNate Lawson 		return (0);
97071ab130cSNate Lawson 	}
97173347b07SNate Lawson 	cfcount = 0;
97273347b07SNate Lawson 	for (i = 0; i < devcount; i++) {
97373347b07SNate Lawson 		if (!device_is_attached(devs[i]))
97473347b07SNate Lawson 			continue;
975e94a0c1aSNate Lawson 		if (CPUFREQ_DRV_TYPE(devs[i], &type) == 0)
97673347b07SNate Lawson 			cfcount++;
97773347b07SNate Lawson 	}
9780325089dSNate Lawson 	if (cfcount <= 1)
97973347b07SNate Lawson 		device_delete_child(device_get_parent(cf_dev), cf_dev);
98073347b07SNate Lawson 	free(devs, M_TEMP);
98173347b07SNate Lawson 
98273347b07SNate Lawson 	return (0);
98373347b07SNate Lawson }
984