common/os/pool_pset.c

	pool_pset.c revision 7c478bd95313f5f23a4c958a745db2134aa03244
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License, Version 1.0 only
 * (the "License").  You may not use this file except in compliance
 * with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma ident   "%Z%%M% %I% %E% SMI"

#include <sys/pool.h>
#include <sys/pool_impl.h>
#include <sys/pool_pset.h>
#include <sys/cpuvar.h>
#include <sys/cpupart.h>
#include <sys/mutex.h>
#include <sys/errno.h>
#include <sys/systm.h>
#include <sys/cmn_err.h>
#include <sys/fss.h>
#include <sys/exacct.h>
#include <sys/time.h>
#include <sys/policy.h>
#include <sys/class.h>
#include <sys/list.h>
#include <sys/cred.h>
#include <sys/zone.h>

/*
 * Processor set plugin for pools.
 *
 * This file contains various routines used by the common pools layer to create,
 * modify, and destroy processor sets.  All processor sets created by this
 * plug-in are stored in the pool_pset_list doubly-linked list, which is
 * guaranteed to always have an entry for the default processor set,
 * pool_pset_default.
 *
 * Interaction with zones:
 *
 * If pools are enabled, non-global zones only have visibility into the
 * pset of the pool to which they are bound.  This is accomplished by
 * changing the set of processors and processor sets which are visible
 * through both systemcall interfaces and system kstats.
 *
 * To avoid grabbing pool_lock() during cpu change operations, we cache
 * the pset the zone is currently bound to, and can read this value
 * while under cpu_lock.  The special psetid_t token ZONE_PS_INVAL means
 * that pools are disabled, and provides a mechanism for determining if the
 * status of pools without grabbing pool_lock().
 *
 * To avoid grabbing any locks to determine the instantaneous value of
 * the number of configured and online cpus in the zone, we also cache
 * these values in a zone_t.  If these values are zero, the pools
 * facility must be disabled, in which case relevant systemcall
 * interfaces will return the values for the system as a whole.
 *
 * The various kstat interfaces are dealt with as follows: if pools are
 * disabled all cpu-related kstats should be exported to all zones.
 * When pools are enabled we begin maintaining a list of "permitted
 * zones" on a per-kstat basis.  There are various hooks throughout the
 * code to update this list when certain pools- or cpu-related events
 * occur.
 */

static list_t pool_pset_list;   /* doubly-linked list of psets */
pool_pset_t *pool_pset_default; /* default pset */
hrtime_t pool_pset_mod;     /* last modification time for psets */
hrtime_t pool_cpu_mod;      /* last modification time for CPUs */

static pool_pset_t *
pool_lookup_pset_by_id(psetid_t psetid)
{
    pool_pset_t *pset = pool_pset_default;

    ASSERT(pool_lock_held());

    for (pset = list_head(&pool_pset_list); pset;
        pset = list_next(&pool_pset_list, pset)) {
        if (pset->pset_id == psetid)
            return (pset);
    }
    return (NULL);
}

struct setup_arg {
    psetid_t psetid;
    cpu_t *cpu;
    cpu_setup_t what;
};

/*
 * Callback function used to apply a cpu configuration event to a zone.
 */
static int
pool_pset_setup_cb(zone_t *zone, void *arg)
{
    struct setup_arg *sa = arg;

    ASSERT(MUTEX_HELD(&cpu_lock));
    ASSERT(INGLOBALZONE(curproc));
    ASSERT(zone != NULL);

    if (zone == global_zone)
        return (0);
    if (zone_pset_get(zone) != sa->psetid)
        return (0); /* ignore */
    switch (sa->what) {
    case CPU_CONFIG:
        cpu_visibility_configure(sa->cpu, zone);
        break;
    case CPU_UNCONFIG:
        cpu_visibility_unconfigure(sa->cpu, zone);
        break;
    case CPU_ON:
        cpu_visibility_online(sa->cpu, zone);
        break;
    case CPU_OFF:
        cpu_visibility_offline(sa->cpu, zone);
        break;
    case CPU_CPUPART_IN:
        cpu_visibility_add(sa->cpu, zone);
        break;
    case CPU_CPUPART_OUT:
        cpu_visibility_remove(sa->cpu, zone);
        break;
    default:
        cmn_err(CE_PANIC, "invalid cpu_setup_t value %d", sa->what);
    }
    return (0);
}

/*
 * Callback function to be executed when a noteworthy cpu event takes
 * place.  Will ensure that the event is reflected by the zones which
 * were affected by it.
 */
/* ARGSUSED */
static int
pool_pset_cpu_setup(cpu_setup_t what, int id, void *arg)
{
    processorid_t cpuid = id;
    struct setup_arg sarg;
    int error;
    cpu_t *c;

    ASSERT(MUTEX_HELD(&cpu_lock));
    ASSERT(INGLOBALZONE(curproc));

    if (!pool_pset_enabled())
        return (0);
    if (what != CPU_CONFIG && what != CPU_UNCONFIG &&
        what != CPU_ON && what != CPU_OFF &&
        what != CPU_CPUPART_IN && what != CPU_CPUPART_OUT)
        return (0);
    c = cpu_get(cpuid);
    ASSERT(c != NULL);
    sarg.psetid = cpupart_query_cpu(c);
    sarg.cpu = c;
    sarg.what = what;

    error = zone_walk(pool_pset_setup_cb, &sarg);
    ASSERT(error == 0);
    return (0);
}

/*
 * Initialize processor set plugin.  Called once at boot time.
 */
void
pool_pset_init(void)
{
    ASSERT(pool_pset_default == NULL);
    pool_pset_default = kmem_zalloc(sizeof (pool_pset_t), KM_SLEEP);
    pool_pset_default->pset_id = PS_NONE;
    pool_pset_default->pset_npools = 1; /* for pool_default */
    pool_default->pool_pset = pool_pset_default;
    list_create(&pool_pset_list, sizeof (pool_pset_t),
        offsetof(pool_pset_t, pset_link));
    list_insert_head(&pool_pset_list, pool_pset_default);
    mutex_enter(&cpu_lock);
    register_cpu_setup_func(pool_pset_cpu_setup, NULL);
    mutex_exit(&cpu_lock);
}

/*
 * Dummy wrapper function that returns 0 to satisfy zone_walk().
 */
static int
pool_pset_zone_pset_set(zone_t *zone, void *arg)
{
    psetid_t psetid = (psetid_t)(uintptr_t)arg;

    ASSERT(MUTEX_HELD(&cpu_lock));
    zone_pset_set(zone, psetid);
    return (0);
}

/*
 * Enable processor set plugin.
 */
int
pool_pset_enable(void)
{
    int error;
    nvlist_t *props;

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));
    /*
     * Can't enable pools if there are existing cpu partitions.
     */
    mutex_enter(&cpu_lock);
    if (cp_numparts > 1) {
        mutex_exit(&cpu_lock);
        return (EEXIST);
    }

    /*
     * We want to switch things such that everything that was tagged with
     * the special ALL_ZONES token now is explicitly visible to all zones:
     * first add individual zones to the visibility list then remove the
     * special "ALL_ZONES" token.  There must only be the default pset
     * (PS_NONE) active if pools are being enabled, so we only need to
     * deal with it.
     *
     * We want to make pool_pset_enabled() start returning B_TRUE before
     * we call any of the visibility update functions.
     */
    global_zone->zone_psetid = PS_NONE;
    /*
     * We need to explicitly handle the global zone since
     * zone_pset_set() won't modify it.
     */
    pool_pset_visibility_add(PS_NONE, global_zone);
    /*
     * A NULL argument means the ALL_ZONES token.
     */
    pool_pset_visibility_remove(PS_NONE, NULL);
    error = zone_walk(pool_pset_zone_pset_set, (void *)PS_NONE);
    ASSERT(error == 0);

    /*
     * It is safe to drop cpu_lock here.  We're still
     * holding pool_lock so no new cpu partitions can
     * be created while we're here.
     */
    mutex_exit(&cpu_lock);
    (void) nvlist_alloc(&pool_pset_default->pset_props,
        NV_UNIQUE_NAME, KM_SLEEP);
    props = pool_pset_default->pset_props;
    (void) nvlist_add_string(props, "pset.name", "pset_default");
    (void) nvlist_add_string(props, "pset.comment", "");
    (void) nvlist_add_int64(props, "pset.sys_id", PS_NONE);
    (void) nvlist_add_string(props, "pset.units", "population");
    (void) nvlist_add_byte(props, "pset.default", 1);
    (void) nvlist_add_uint64(props, "pset.max", 65536);
    (void) nvlist_add_uint64(props, "pset.min", 1);
    pool_pset_mod = pool_cpu_mod = gethrtime();
    return (0);
}

/*
 * Disable processor set plugin.
 */
int
pool_pset_disable(void)
{
    processorid_t cpuid;
    cpu_t *cpu;
    int error;

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));

    mutex_enter(&cpu_lock);
    if (cp_numparts > 1) {  /* make sure only default pset is left */
        mutex_exit(&cpu_lock);
        return (EBUSY);
    }
    /*
     * Remove all non-system CPU and processor set properties
     */
    for (cpuid = 0; cpuid < NCPU; cpuid++) {
        if ((cpu = cpu_get(cpuid)) == NULL)
            continue;
        if (cpu->cpu_props != NULL) {
            (void) nvlist_free(cpu->cpu_props);
            cpu->cpu_props = NULL;
        }
    }

    /*
     * We want to switch things such that everything is now visible
     * to ALL_ZONES: first add the special "ALL_ZONES" token to the
     * visibility list then remove individual zones.  There must
     * only be the default pset active if pools are being disabled,
     * so we only need to deal with it.
     */
    error = zone_walk(pool_pset_zone_pset_set, (void *)ZONE_PS_INVAL);
    ASSERT(error == 0);
    pool_pset_visibility_add(PS_NONE, NULL);
    pool_pset_visibility_remove(PS_NONE, global_zone);
    /*
     * pool_pset_enabled() will henceforth return B_FALSE.
     */
    global_zone->zone_psetid = ZONE_PS_INVAL;
    mutex_exit(&cpu_lock);
    if (pool_pset_default->pset_props != NULL) {
        nvlist_free(pool_pset_default->pset_props);
        pool_pset_default->pset_props = NULL;
    }
    return (0);
}

/*
 * Create new processor set and give it a temporary name.
 */
int
pool_pset_create(psetid_t *id)
{
    char pset_name[40];
    pool_pset_t *pset;
    psetid_t psetid;
    int err;

    ASSERT(pool_lock_held());
    if ((err = cpupart_create(&psetid)) != 0)
        return (err);
    pset = kmem_alloc(sizeof (pool_pset_t), KM_SLEEP);
    pset->pset_id = *id = psetid;
    pset->pset_npools = 0;
    (void) nvlist_alloc(&pset->pset_props, NV_UNIQUE_NAME, KM_SLEEP);
    (void) nvlist_add_int64(pset->pset_props, "pset.sys_id", psetid);
    (void) nvlist_add_byte(pset->pset_props, "pset.default", 0);
    pool_pset_mod = gethrtime();
    (void) snprintf(pset_name, sizeof (pset_name), "pset_%lld",
        pool_pset_mod);
    (void) nvlist_add_string(pset->pset_props, "pset.name", pset_name);
    list_insert_tail(&pool_pset_list, pset);
    return (0);
}

/*
 * Destroy existing processor set.
 */
int
pool_pset_destroy(psetid_t psetid)
{
    pool_pset_t *pset;
    int ret;

    ASSERT(pool_lock_held());

    if (psetid == PS_NONE)
        return (EINVAL);
    if ((pset = pool_lookup_pset_by_id(psetid)) == NULL)
        return (ESRCH);
    if (pset->pset_npools > 0) /* can't destroy associated psets */
        return (EBUSY);
    if ((ret = cpupart_destroy(pset->pset_id)) != 0)
        return (ret);
    (void) nvlist_free(pset->pset_props);
    list_remove(&pool_pset_list, pset);
    pool_pset_mod = gethrtime();
    kmem_free(pset, sizeof (pool_pset_t));
    return (0);
}

/*
 * Change the visibility of a pset (and all contained cpus) in a zone.
 * A NULL zone argument implies the special ALL_ZONES token.
 */
static void
pool_pset_visibility_change(psetid_t psetid, zone_t *zone, boolean_t add)
{
    zoneid_t zoneid = zone ? zone->zone_id : ALL_ZONES;
    cpupart_t *cp;
    cpu_t *c;

    ASSERT(MUTEX_HELD(&cpu_lock));
    ASSERT(psetid != ZONE_PS_INVAL);

    cp = cpupart_find(psetid);
    ASSERT(cp != NULL);
    if (cp->cp_kstat != NULL) {
        if (add)
            kstat_zone_add(cp->cp_kstat, zoneid);
        else
            kstat_zone_remove(cp->cp_kstat, zoneid);
    }

    c = cpu_list;
    do {
        ASSERT(c != NULL);
        if (c->cpu_part == cp && !cpu_is_poweredoff(c)) {
            if (add)
                cpu_visibility_add(c, zone);
            else
                cpu_visibility_remove(c, zone);
        }
    } while ((c = c->cpu_next) != cpu_list);
}

/*
 * Make the processor set visible to the zone.  A NULL value for
 * the zone means that the special ALL_ZONES token should be added to
 * the visibility list.
 */
void
pool_pset_visibility_add(psetid_t psetid, zone_t *zone)
{
    pool_pset_visibility_change(psetid, zone, B_TRUE);
}

/*
 * Remove zone's visibility into the processor set.  A NULL value for
 * the zone means that the special ALL_ZONES token should be removed
 * from the visibility list.
 */
void
pool_pset_visibility_remove(psetid_t psetid, zone_t *zone)
{
    pool_pset_visibility_change(psetid, zone, B_FALSE);
}

/*
 * Quick way of seeing if pools are enabled (as far as processor sets are
 * concerned) without holding pool_lock().
 */
boolean_t
pool_pset_enabled(void)
{
    ASSERT(MUTEX_HELD(&cpu_lock));

    return (zone_pset_get(global_zone) != ZONE_PS_INVAL);
}

struct assoc_zone_arg {
    poolid_t poolid;
    psetid_t newpsetid;
};

/*
 * Callback function to update a zone's processor set visibility when
 * a pool is associated with a processor set.
 */
static int
pool_pset_assoc_zone_cb(zone_t *zone, void *arg)
{
    struct assoc_zone_arg *aza = arg;
    pool_t *pool;
    zoneid_t zoneid = zone->zone_id;

    ASSERT(pool_lock_held());
    ASSERT(MUTEX_HELD(&cpu_lock));

    if (zoneid == GLOBAL_ZONEID)
        return (0);
    pool = zone_pool_get(zone);
    if (pool->pool_id == aza->poolid)
        zone_pset_set(zone, aza->newpsetid);
    return (0);
}

/*
 * Associate pool with new processor set.
 */
int
pool_pset_assoc(poolid_t poolid, psetid_t psetid)
{
    pool_t *pool;
    pool_pset_t *pset, *oldpset;
    int err = 0;

    ASSERT(pool_lock_held());

    if ((pool = pool_lookup_pool_by_id(poolid)) == NULL ||
        (pset = pool_lookup_pset_by_id(psetid)) == NULL) {
        return (ESRCH);
    }
    if (pool->pool_pset->pset_id == psetid) {
        /*
         * Already associated.
         */
        return (0);
    }

    /*
     * Hang the new pset off the pool, and rebind all of the pool's
     * processes to it.  If pool_do_bind fails, all processes will remain
     * bound to the old set.
     */
    oldpset = pool->pool_pset;
    pool->pool_pset = pset;
    err = pool_do_bind(pool, P_POOLID, poolid, POOL_BIND_PSET);
    if (err) {
        pool->pool_pset = oldpset;
    } else {
        struct assoc_zone_arg azarg;

        /*
         * Update zones' visibility to reflect changes.
         */
        azarg.poolid = poolid;
        azarg.newpsetid = pset->pset_id;
        mutex_enter(&cpu_lock);
        err = zone_walk(pool_pset_assoc_zone_cb, &azarg);
        ASSERT(err == 0);
        mutex_exit(&cpu_lock);

        oldpset->pset_npools--;
        pset->pset_npools++;
    }
    return (err);
}

/*
 * Transfer specified CPUs between processor sets.
 */
int
pool_pset_xtransfer(psetid_t src, psetid_t dst, size_t size, id_t *ids)
{
    struct cpu *cpu;
    int ret = 0;
    int id;

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));

    if (size == 0 || size > max_ncpus)  /* quick sanity check */
        return (EINVAL);

    mutex_enter(&cpu_lock);
    for (id = 0; id < size; id++) {
        if ((cpu = cpu_get((processorid_t)ids[id])) == NULL ||
            cpupart_query_cpu(cpu) != src) {
            ret = EINVAL;
            break;
        }
        if ((ret = cpupart_attach_cpu(dst, cpu, 1)) != 0)
            break;
    }
    mutex_exit(&cpu_lock);
    if (ret == 0)
        pool_pset_mod = gethrtime();
    return (ret);
}

/*
 * Bind process to processor set.  This should never fail because
 * we should've done all preliminary checks before calling it.
 */
void
pool_pset_bind(proc_t *p, psetid_t psetid, void *projbuf, void *zonebuf)
{
    kthread_t *t;
    int ret;

    ASSERT(pool_lock_held());
    ASSERT(MUTEX_HELD(&cpu_lock));
    ASSERT(MUTEX_HELD(&pidlock));
    ASSERT(MUTEX_HELD(&p->p_lock));

    if ((t = p->p_tlist) == NULL)
        return;
    do {
        ret = cpupart_bind_thread(t, psetid, 0, projbuf, zonebuf);
        ASSERT(ret == 0);
        t->t_bind_pset = psetid;
    } while ((t = t->t_forw) != p->p_tlist);
}

/*
 * See the comment above pool_do_bind() for the semantics of the pset_bind_*()
 * functions.  These must be kept in sync with cpupart_move_thread, and
 * anything else that could fail a pool_pset_bind.
 *
 * Returns non-zero errno on failure and zero on success.
 * Iff successful, cpu_lock is held on return.
 */
int
pset_bind_start(proc_t **procs, pool_t *pool)
{
    cred_t *pcred;
    proc_t *p, **pp;
    kthread_t *t;
    cpupart_t *newpp;
    int ret;

    extern int cpupart_movable_thread(kthread_id_t, cpupart_t *, int);

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));

    mutex_enter(&cpu_lock);
    weakbinding_stop();

    newpp = cpupart_find(pool->pool_pset->pset_id);
    ASSERT(newpp != NULL);
    if (newpp->cp_cpulist == NULL) {
        weakbinding_start();
        mutex_exit(&cpu_lock);
        return (ENOTSUP);
    }

    pcred = crgetcred();

    /*
     * Check for the PRIV_PROC_PRIOCNTL privilege that is required
     * to enter and exit scheduling classes.  If other privileges
     * are required by CL_ENTERCLASS/CL_CANEXIT types of routines
     * in the future, this code will have to be updated.
     */
    if (secpolicy_setpriority(pcred) != 0) {
        weakbinding_start();
        mutex_exit(&cpu_lock);
        crfree(pcred);
        return (EPERM);
    }

    for (pp = procs; (p = *pp) != NULL; pp++) {
        mutex_enter(&p->p_lock);
        if ((t = p->p_tlist) == NULL) {
            mutex_exit(&p->p_lock);
            continue;
        }
        /*
         * Check our basic permissions to control this process.
         */
        if (!prochasprocperm(p, curproc, pcred)) {
            mutex_exit(&p->p_lock);
            weakbinding_start();
            mutex_exit(&cpu_lock);
            crfree(pcred);
            return (EPERM);
        }
        do {
            /*
             * Check that all threads can be moved to
             * a new processor set.
             */
            thread_lock(t);
            ret = cpupart_movable_thread(t, newpp, 0);
            thread_unlock(t);
            if (ret != 0) {
                mutex_exit(&p->p_lock);
                weakbinding_start();
                mutex_exit(&cpu_lock);
                crfree(pcred);
                return (ret);
            }
        } while ((t = t->t_forw) != p->p_tlist);
        mutex_exit(&p->p_lock);
    }
    crfree(pcred);
    return (0); /* with cpu_lock held and weakbinding stopped */
}

/*ARGSUSED*/
void
pset_bind_abort(proc_t **procs, pool_t *pool)
{
    mutex_exit(&cpu_lock);
}

void
pset_bind_finish(void)
{
    weakbinding_start();
    mutex_exit(&cpu_lock);
}

static pool_property_t pool_pset_props[] = {
    { "pset.name",          DATA_TYPE_STRING,   PP_RDWR },
    { "pset.comment",       DATA_TYPE_STRING,   PP_RDWR },
    { "pset.sys_id",        DATA_TYPE_UINT64,   PP_READ },
    { "pset.units",         DATA_TYPE_STRING,   PP_RDWR },
    { "pset.default",       DATA_TYPE_BYTE,     PP_READ },
    { "pset.min",           DATA_TYPE_UINT64,   PP_RDWR },
    { "pset.max",           DATA_TYPE_UINT64,   PP_RDWR },
    { "pset.size",          DATA_TYPE_UINT64,   PP_READ },
    { "pset.load",          DATA_TYPE_UINT64,   PP_READ },
    { "pset.poold.objectives",  DATA_TYPE_STRING,
        PP_RDWR | PP_OPTIONAL },
    { NULL,             0,          0 }
};

static pool_property_t pool_cpu_props[] = {
    { "cpu.sys_id",         DATA_TYPE_UINT64,   PP_READ },
    { "cpu.comment",        DATA_TYPE_STRING,   PP_RDWR },
    { "cpu.status",         DATA_TYPE_STRING,   PP_RDWR },
    { "cpu.pinned",         DATA_TYPE_BYTE,
        PP_RDWR | PP_OPTIONAL },
    { NULL,             0,          0 }
};

/*
 * Put property on the specified processor set.
 */
int
pool_pset_propput(psetid_t psetid, nvpair_t *pair)
{
    pool_pset_t *pset;
    int ret;

    ASSERT(pool_lock_held());

    if ((pset = pool_lookup_pset_by_id(psetid)) == NULL)
        return (ESRCH);
    ret = pool_propput_common(pset->pset_props, pair, pool_pset_props);
    if (ret == 0)
        pool_pset_mod = gethrtime();
    return (ret);
}

/*
 * Remove existing processor set property.
 */
int
pool_pset_proprm(psetid_t psetid, char *name)
{
    pool_pset_t *pset;
    int ret;

    ASSERT(pool_lock_held());

    if ((pset = pool_lookup_pset_by_id(psetid)) == NULL)
        return (EINVAL);
    ret = pool_proprm_common(pset->pset_props, name, pool_pset_props);
    if (ret == 0)
        pool_pset_mod = gethrtime();
    return (ret);
}

/*
 * Put new CPU property.
 * Handle special case of "cpu.status".
 */
int
pool_cpu_propput(processorid_t cpuid, nvpair_t *pair)
{
    int ret = 0;
    cpu_t *cpu;

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));

    if (nvpair_type(pair) == DATA_TYPE_STRING &&
        strcmp(nvpair_name(pair), "cpu.status") == 0) {
        char *val;
        int status;
        int old_status;
        (void) nvpair_value_string(pair, &val);
        if (strcmp(val, PS_OFFLINE) == 0)
            status = P_OFFLINE;
        else if (strcmp(val, PS_ONLINE) == 0)
            status = P_ONLINE;
        else if (strcmp(val, PS_NOINTR) == 0)
            status = P_NOINTR;
        else if (strcmp(val, PS_FAULTED) == 0)
            status = P_FAULTED;
        else if (strcmp(val, PS_SPARE) == 0)
            status = P_SPARE;
        else
            return (EINVAL);
        ret = p_online_internal(cpuid, status, &old_status);
    } else {
        mutex_enter(&cpu_lock);
        if ((cpu = cpu_get(cpuid)) == NULL)
            ret = EINVAL;
        if (cpu->cpu_props == NULL) {
            (void) nvlist_alloc(&cpu->cpu_props,
                NV_UNIQUE_NAME, KM_SLEEP);
            (void) nvlist_add_string(cpu->cpu_props,
                "cpu.comment", "");
        }
        ret = pool_propput_common(cpu->cpu_props, pair, pool_cpu_props);
        if (ret == 0)
            pool_cpu_mod = gethrtime();
        mutex_exit(&cpu_lock);
    }
    return (ret);
}

/*
 * Remove existing CPU property.
 */
int
pool_cpu_proprm(processorid_t cpuid, char *name)
{
    int ret;
    cpu_t *cpu;

    ASSERT(pool_lock_held());
    ASSERT(INGLOBALZONE(curproc));

    mutex_enter(&cpu_lock);
    if ((cpu = cpu_get(cpuid)) == NULL || cpu_is_poweredoff(cpu)) {
        ret = EINVAL;
    } else {
        if (cpu->cpu_props == NULL)
            ret = EINVAL;
        else
            ret = pool_proprm_common(cpu->cpu_props, name,
                pool_cpu_props);
    }
    if (ret == 0)
        pool_cpu_mod = gethrtime();
    mutex_exit(&cpu_lock);
    return (ret);
}

/*
 * This macro returns load average multiplied by 1000 w/o losing precision
 */
#define PSET_LOAD(f)    (((f >> 16) * 1000) + (((f & 0xffff) * 1000) / 0xffff))

/*
 * Take a snapshot of the current state of processor sets and CPUs,
 * pack it in the exacct format, and attach it to specified exacct record.
 */
int
pool_pset_pack(ea_object_t *eo_system)
{
    ea_object_t *eo_pset, *eo_cpu;
    cpupart_t *cpupart;
    psetid_t mypsetid;
    pool_pset_t *pset;
    nvlist_t *nvl;
    size_t bufsz;
    cpu_t *cpu;
    char *buf;
    int ncpu;

    ASSERT(pool_lock_held());

    mutex_enter(&cpu_lock);
    mypsetid = zone_pset_get(curproc->p_zone);
    for (pset = list_head(&pool_pset_list); pset;
        pset = list_next(&pool_pset_list, pset)) {
        psetid_t psetid = pset->pset_id;

        if (!INGLOBALZONE(curproc) && mypsetid != psetid)
            continue;
        cpupart = cpupart_find(psetid);
        ASSERT(cpupart != NULL);
        eo_pset = ea_alloc_group(EXT_GROUP |
            EXC_LOCAL | EXD_GROUP_PSET);
        (void) ea_attach_item(eo_pset, &psetid, sizeof (id_t),
            EXC_LOCAL | EXD_PSET_PSETID | EXT_UINT32);
        /*
         * Pack info for all CPUs in this processor set.
         */
        ncpu = 0;
        cpu = cpu_list;
        do {
            if (cpu->cpu_part != cpupart)   /* not our pset */
                continue;
            ncpu++;
            eo_cpu = ea_alloc_group(EXT_GROUP
                | EXC_LOCAL | EXD_GROUP_CPU);
            (void) ea_attach_item(eo_cpu, &cpu->cpu_id,
                sizeof (processorid_t),
                EXC_LOCAL | EXD_CPU_CPUID | EXT_UINT32);
            if (cpu->cpu_props == NULL) {
                (void) nvlist_alloc(&cpu->cpu_props,
                    NV_UNIQUE_NAME, KM_SLEEP);
                (void) nvlist_add_string(cpu->cpu_props,
                    "cpu.comment", "");
            }
            (void) nvlist_dup(cpu->cpu_props, &nvl, KM_SLEEP);
            (void) nvlist_add_int64(nvl, "cpu.sys_id", cpu->cpu_id);
            (void) nvlist_add_string(nvl, "cpu.status",
                (char *)cpu_get_state_str(cpu));
            buf = NULL;
            bufsz = 0;
            (void) nvlist_pack(nvl, &buf, &bufsz,
                NV_ENCODE_NATIVE, 0);
            (void) ea_attach_item(eo_cpu, buf, bufsz,
                EXC_LOCAL | EXD_CPU_PROP | EXT_RAW);
            (void) nvlist_free(nvl);
            kmem_free(buf, bufsz);
            (void) ea_attach_to_group(eo_pset, eo_cpu);
        } while ((cpu = cpu->cpu_next) != cpu_list);

        (void) nvlist_dup(pset->pset_props, &nvl, KM_SLEEP);
        (void) nvlist_add_uint64(nvl, "pset.size", ncpu);
        (void) nvlist_add_uint64(nvl, "pset.load",
            (uint64_t)PSET_LOAD(cpupart->cp_hp_avenrun[0]));
        buf = NULL;
        bufsz = 0;
        (void) nvlist_pack(nvl, &buf, &bufsz, NV_ENCODE_NATIVE, 0);
        (void) ea_attach_item(eo_pset, buf, bufsz,
            EXC_LOCAL | EXD_PSET_PROP | EXT_RAW);
        (void) nvlist_free(nvl);
        kmem_free(buf, bufsz);

        (void) ea_attach_to_group(eo_system, eo_pset);
    }
    mutex_exit(&cpu_lock);
    return (0);
}

/*
 * Get dynamic property for processor sets.
 * The only dynamic property currently implemented is "pset.load".
 */
int
pool_pset_propget(psetid_t psetid, char *name, nvlist_t *nvl)
{
    cpupart_t *cpupart;
    pool_pset_t *pset;
    int ret = ESRCH;

    ASSERT(pool_lock_held());

    mutex_enter(&cpu_lock);
    pset = pool_lookup_pset_by_id(psetid);
    cpupart = cpupart_find(psetid);
    if (cpupart == NULL || pset == NULL) {
        mutex_exit(&cpu_lock);
        return (EINVAL);
    }
    if (strcmp(name, "pset.load") == 0)
        ret = nvlist_add_uint64(nvl, "pset.load",
            (uint64_t)PSET_LOAD(cpupart->cp_hp_avenrun[0]));
    else
        ret = EINVAL;
    mutex_exit(&cpu_lock);
    return (ret);
}

/*
 * Get dynamic property for CPUs.
 * The only dynamic property currently implemented is "cpu.status".
 */
int
pool_cpu_propget(processorid_t cpuid, char *name, nvlist_t *nvl)
{
    int ret = ESRCH;
    cpu_t *cpu;

    ASSERT(pool_lock_held());

    mutex_enter(&cpu_lock);
    if ((cpu = cpu_get(cpuid)) == NULL) {
        mutex_exit(&cpu_lock);
        return (ESRCH);
    }
    if (strcmp(name, "cpu.status") == 0) {
        ret = nvlist_add_string(nvl, "cpu.status",
            (char *)cpu_get_state_str(cpu));
    } else {
        ret = EINVAL;
    }
    mutex_exit(&cpu_lock);
    return (ret);
}